MoU proposal to all state and/or non-state actors interested in the demise of the dollar and the advancement of humanity and equality for the masses. USD HEGEMONY TERMINATION IS GUARANTEED AND OUR ALLIANCE SIMPLY ACCELERATES THE PROCESS FROM A DECADE OR MORE TO THE NEXT EIGHTEEN MONTHS TO TOTAL GLOBAL DESERTION OF US DOLLAR AS RESERVE CURRENCY  

Abstract

Traditionally, the autoregressive integrated moving average (ARIMA) model has been one of the most widely used linear models in time series forecasting. However, the ARIMA model cannot easily capture the nonlinear patterns. Support vector machines (SVMs), a novel neural network technique, have been successfully applied in solving nonlinear regression estimation problems. Therefore, this investigation proposes a hybrid methodology that exploits the unique strength of the ARIMA model and the SVMs model in forecasting stock prices problems. Real data sets of stock prices were used to examine the forecasting accuracy of the proposed model. The results of computational tests are very promising.

A hybrid ARIMA and support vector machines model in stock price forecasting
Download Citation | A hybrid ARIMA and support vector machines model in stock price forecasting | Traditionally, the autoregressive integrated moving average (ARIMA) model has been one of the most widely used linear models in time series... | Find, read and cite all the research you need on Research…
World Affairs
We believe that the solutions to the world’s most challenging problems are found when the private, philanthropic and public sectors work together. Every day, we convene thought leaders, change makers and engaged citizens to share ideas, learn from each other and effect change. Connecting people in t…
Real Vision Finance
Real Vision™ is where you can gain an understanding of the complex world of finance, business and the global economy with real in-depth analysis from real experts. For even more access and more content, become a Real Vision member for $1 for 7 Days: https://rvtv.io/membership
Tie Finance
In this channel you’ll learn to analyze any company’s financial performance through financial analysis. Videos will help you to use analytical tools to understand the numbers.You will find as well Market insights and investment strategies.Go forward and boost your financial knowledge. Main focus …

investment strategic partnership and bond yield curve supremacy via algorithmic powered machine learning models that dominate any trading equilibrium always guaranteeing profit at the expense of other traders directly siphoning hedge fund and other institutional fund positions for maximum value transfer from their portfolio(s) to ours.

US8140416B2 - Algorithmic trading system and method - Google Patents
A system and method for allowing market participants to evaluate the likelihood of finding hidden volume. The model can predict hidden volume and assess the probability that a market order will be executed within the spread and better than the mid-quote. The cost per immediate execution can be…
US8571967B1 - System and method for algorithmic trading strategies - Google Patents
Various embodiments of the present invention are directed to systems and methods for algorithmic trading strategies and/or systems and methods for use in executing an order directed to a security traded in a market. More particularly, one embodiment of the present invention relates to a method…
Patents on Algorithmic Trading / High-Frequency Trading – FMeasure

Staff Report on Algorithmic Trading in U.S. Capital Markets

As Required by Section 502 of the

Economic Growth, Regulatory Relief, and Consumer Protection Act of 2018

This is a report by the Staff of the U.S. Securities and Exchange Commission. The  Commission has expressed no view regarding the analysis, findings, or conclusions  contained herein.

August 5, 2020

1

Table of Contents

I. Introduction...................................................................................................................................................3 A. Congressional Mandate .........................................................................................................................3 B. Overview .....................................................................................................................................................4 C. Algorithmic Trading and Markets .....................................................................................................5

II. Overview of Equity Market Structure..................................................................................................7 A. Trading Centers ........................................................................................................................................9 B. Market Data............................................................................................................................................. 19

III. Overview of Debt Market Structure .................................................................................................. 23 A. Types of Debt Securities or Instruments ..................................................................................... 23 B. Data and Communications ................................................................................................................ 28

IV. Benefits and Risks of Algorithmic Trading in Equities .............................................................. 30 A. Investors................................................................................................................................................... 30 B. Brokers...................................................................................................................................................... 36 C. Principal Trading .................................................................................................................................. 37 D. Operational Risks to Firms and the Market................................................................................ 42 E. Studies of Effects on Market Quality and Provision of Liquidity........................................ 44 F. Effects of the COVID-19 Pandemic ................................................................................................. 47

V. Benefits and Risks of Algorithmic Trading in Corporate and Municipal Bonds............... 51 A. Liquidity Search and Trade Execution.......................................................................................... 51 B. ETF Market Making and Arbitrage................................................................................................. 53 C. Studies of Effects on Market Quality and Provision of Liquidity........................................ 53

VI. Regulatory Responses to Market Complexity, Volatility, and Instability ........................... 55 A. Improving Market Transparency.................................................................................................... 55 B. Mitigating Price Volatility.................................................................................................................. 60 C. Facilitating Market Stability and Security................................................................................... 63 D. Additional Ongoing and Potential Commission and Staff Actions ..................................... 67

VII. Summary of Studies on Algorithmic Trading ................................................................................ 69 A. Equities ..................................................................................................................................................... 69 B. Debt Securities ....................................................................................................................................... 82

VIII. Conclusion................................................................................................................................................... 83 IX. Bibliography to Summary of Academic Studies ........................................................................... 85 X. Appendix: Market Participants, Roles, and Obligations ............................................................ 92

2

I. Introduction

A. Congressional Mandate

The Economic Growth, Regulatory Relief, and Consumer Protection Act of 2018 requires  the staff of the U.S. Securities and Exchange Commission (the “SEC” or “Commission”) to  submit to Congress a report on the risks and benefits of algorithmic trading in the U.S.  capital markets.1 Specifically, § 502 provides:

(a) In General. Not later than 18 months after the date of enactment of this Act, the staff of  the Securities and Exchange Commission shall submit to the Committee on Banking,  Housing, and Urban Affairs of the Senate and the Committee on Financial Services of  the House of Representatives a report on the risks and benefits of algorithmic trading  in capital markets in the United States.

(b) Matters Required To Be Included. The matters covered by the report required by  subsection (a) shall include the following:

(1) An assessment of the effect of algorithmic trading in equity and debt markets in  the United States on the provision of liquidity in stressed and normal market  conditions.

(2) An assessment of the benefits and risks to equity and debt markets in the  United States by algorithmic trading.

(3) An analysis of whether the activity of algorithmic trading and entities that  engage in algorithmic trading are subject to appropriate Federal supervision  and regulation.

(4) A recommendation of whether

(A) based on the analysis described in paragraphs (1), (2), and (3), any  changes should be made to regulations; and

(B) the Securities and Exchange Commission needs additional legal

authorities or resources to effect the changes described in subparagraph  (A).

1 Economic Growth, Regulatory Relief, and Consumer Protection Act, Pub. L. No. 115-174,  § 502, 132 Stat. 1296, 1361-62 (2018).

3

B. Overview

As required by § 502 of the Economic Growth, Regulatory Relief, and Consumer Protection  Act of 2018, this staff report describes the benefits and risks of algorithmic trading in the  U.S. equity and debt markets.

Broadly speaking, and as more fully discussed below, algorithmic trading in the equities— and to a lesser extent—in the debt market, has improved many measures of market quality  and liquidity provision during normal market conditions, though studies have also shown  that some types of algorithmic trading may exacerbate periods of unusual market stress or  volatility. Advances in technology and communications have enabled many market  participants to more efficiently provide liquidity, more efficiently access market liquidity,  implement new trading services, and more effectively manage risk across a range of  markets.

Furthermore, commenters have observed that the increasing complexity of multiple  interconnected markets may have increased the risk that operational or systems failures at  trading firms, platforms, or infrastructure may have broad, potentially unexpected,  detrimental effects on the markets and investors. A number of observers have noted that even as some uses of algorithms may contribute to market complexity, algorithms  generally help market participants navigate market complexity. A common theme echoed  by nearly all market professionals, academic researchers, and other students of the securities markets is that that algorithmic trading, in one form or another, is an integral  and permanent part of our modern capital markets.

Several variations of algorithmic trading strategies have developed and expanded over the  last several decades. These developments have been driven, in pertinent part, by the  growth in available market data generated by and consumed by market professionals,  major advances in computational power and the speed of data transmission, and the  exponential increase in the speed of securities trading. Enhancements in algorithmic  trading strategies have also been driven by investor demands for execution quality, the  search for alpha and trading profits, and the application of sophisticated quantitative  analytics. The Commission and other regulators have responded with a range of tools  intended to mitigate risks to investors and to help ensure fair, efficient, and orderly  markets. Commission staff will continue to monitor technological change and its influence  on investment, trading, and the capital markets, and will continue to assess the need for  additional regulation, resources, or legal authority.2

2 The significant and rapid economic impact precipitated by the COVID-19 pandemic was  reflected in extraordinary trading in the U.S. secondary markets for equity and debt during  the spring of 2020. While this report briefly discusses recent market events, including  certain significant impacts on trading as market participants reacted to the effects of  COVID-19, the report is focused on the broader questions raised in § 502. In April 2020,

4

C. Algorithmic Trading and Markets

The use of algorithms in trading is pervasive in today’s markets. Any analysis of the role  that algorithmic trading plays in the US equity and debt markets requires an understanding  of equity and debt market structure,3 the role played by different participants in those  markets, and the extent to which algorithmic trading is used by market professionals. 4

In describing the uses of algorithms in trading, it is useful to first define an algorithm. At its  most general level, an algorithm is a finite, deterministic, and effective problem-solving

the Commission announced the formation of an internal, cross-divisional COVID-19 Market  Monitoring Group to assist with Commission and staff actions and analysis related to the  effects of COVID-19 on markets, issuers, and investors, and with responding to requests for  information, analysis and assistance from fellow regulators and other public sector  partners. See “SEC Forms Cross-Divisional COVID-19 Market Monitoring Group,” Press  Release 2020-95 (Apr. 24, 2020); see also “COVID-19 Market Monitoring Group — Update  and Current Efforts,” Statement of Chairman Jay Clayton and S.P. Kothari (May 13, 2020),  available at: https://www.sec.gov/news/public-statement/statement-clayton-kothari covid-19-2020-05-13 (describing some of the initial work of the COVID-19 Market  Monitoring Group); COVID-19 Market Monitoring Group, “Credit Ratings, Procyclicality and  Related Financial Stability Issues: Select Observations” (Jul. 15, 2020), available at: https://www.sec.gov/news/public-statement/covid-19-monitoring-group-2020-07-15.

3 The section of this staff report on equity market structure echoes aspects of the  Commission’s 2010 Concept Release on Equity Market Structure. See Concept Release on  Equity Market Structure, Exch. Act. Rel. No. 61358, 75 Fed. Reg. 3594 (Jan. 21, 2010)  (“Concept Release”). That Concept Release described the transition of modern equity  trading markets away from a largely centralized, manual structure to the dispersed,  automated structure that exists today. The Concept Release provided many useful  institutional details; this report updates some of these details, and describes important  developments that have occurred since 2010. When discussing debt markets, this report  focuses on corporate and municipal bonds. While the markets for U.S. Treasury securities  are described briefly, they are not a focus of this report.

4 The main body of this staff report presumes familiarity with core concepts in securities  market structure, such as the distinction between acting as a broker and trading as  principal, key differences between types of trading venues such as national securities  exchanges and alternative trading systems, the difference between providing and  demanding liquidity, and legal obligations such as best execution. Background on these  concepts may be found in the appendix to this report, which provides a more general orientation to market participants, roles, and obligations.

5

method suitable for implementation as a computer program.5 In modern equity and debt  markets, many problems are solved and decisions made in this computational, algorithmic  manner. Today, algorithms address many of the problems and decisions that have long  been central to the business of trading. What instrument(s) should be invested in or  traded? What price should be bid or offered? What order size is optimal? What should be  the response to a request for a quotation? What risk will be taken on by facilitating a  trade? How does that risk change with the size of the trade? Is the risk of a trade  appropriate to a firm’s available capital? What is the relationship between the price of  different but related securities or financial products? To what market should an order be  sent? Is it more effective to provide liquidity or demand liquidity? Should an order be  displayed or non-displayed? To which broker should an order be sent? When should an  order be submitted to a trading center? In general, algorithms utilize a rich array of market  information to very quickly assess the state of the market, to determine when, where, and  how to trade, and then to implement the resulting trading decision(s) in the marketplace.6

As described in more detail below, algorithms are broadly used in contemporary securities  markets, and the range of uses differs across asset classes and across the roles and  investment objectives of market participants. In light of the wide diversity of algorithms in  modern trading, it is not a goal of this report to define a single type of trading or activity as  uniquely algorithmic. Rather, this staff report attempts to describe many dimensions of the  contemporary secondary markets for equity and debt securities that operate  algorithmically. The types of trading described in more detail below each fundamentally  depend upon computerized algorithms, and the data and technological infrastructure  through which they operate, to address the types of problems and tasks described above.

The staff’s approach differs from the more narrow approaches taken in much of the  literature on algorithmic trading, which generally seek to examine a specific type of  algorithmic activity. For example, one study defines algorithmic trading as “a tool for  professional traders that may observe market parameters or other information in real-time  and automatically generates/carries out trading decisions without human intervention.”7  Other approaches, for example, characterize algorithmic trading as the use of programmed

5 See, e.g., Robert Sedgewick & Kevin Wayne, Algorithms, 4 (4th Ed. 2011) (“The term  algorithm is used in computer science to describe a finite, deterministic, and effective  problem-solving method suitable for implementation as a computer program”).

6 These are just a few of the questions and decisions that algorithms address in today’s  markets and the scope as well as the granularity of issues that algorithms address is  virtually unbounded.

7 Peter Gomber, Björn Arndt, Marco Lutat, Tim Uhle, High Frequency Trading, 14 (Goethe  Univ. Frankfurt Am. Main, Working Paper, 2011) (available at

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1858626).

6

trading instructions to execute small portions of larger orders over time.8 While activity  meeting these definitions is encompassed in the approach taken here, this staff report’s  coverage is broader, reaching areas where algorithms are used and may be important, but  in some cases may not be used as exclusively or extensively as in the activities described in  these examples.9

II. Overview of Equity Market Structure

Today’s equity market structure is highly fragmented, consisting of fifteen national  securities exchanges, over thirty alternative trading systems, multiple single-dealer  platforms within broker-dealers, and other forms of order matching. The equity markets  are also highly complex, with dozens of different order types, a multitude of market  connectivity options, and a rich array of market information products providing data in  speeds often measured in microseconds. This data is the key input into the wide variety of  algorithmic trading strategies that rapidly submit orders across venues, creating and  moving the prices of securities, which, in turn, generate more data that drives the next set  of algorithmic trading decisions.

In Section 11A of the Exchange Act,10 Congress directed the Commission to facilitate the  establishment of a national market system. The Commission is required to do so in  accordance with the findings and objectives Congress outlined in the Exchange Act:

• The securities markets are an important national asset which must be preserved and  strengthened;

• New data processing and communications techniques create the opportunity for more  efficient and effective market operations;

• It is in the public interest and appropriate for the protection of investors and the  maintenance of fair and orderly markets to assure–

– Economically efficient execution of securities transactions;

– Fair competition among brokers and dealers, among exchange markets, and  between exchange markets and markets other than exchange markets;

8 See Katie Kolchin, Electronic Trading Market Structure Primer, SIFMA Insights, pp. 15-16  (Oct. 10, 2019), available at: https://www.sifma.org/resources/research/electronic trading-market-structure-primer/.

9 For a more detailed discussion of some of the methodological issues involved with trying  to precisely define algorithmic trading and its subsets, see Staff of Division of Trading and  Markets, U.S. Securities & Exchange Commission, Equity Market Structure Literature Review  Part II: High Frequency Trading, 4-11 (Mar. 18, 2014) (“HFT Literature Review”).

10 15 U.S.C. 78k-1.

7

– The availability to brokers, dealers, and investors of information with respect  to quotations for and transactions in securities;

– The practicability of brokers executing investors’ orders in the best market;  and

– An opportunity, consistent with economically efficient execution and the ability  to execute orders in the best market, for investors’ orders to be executed  without the participation of a dealer; and

• The linking of all markets for qualified securities through communication and data  processing facilities will foster efficiency, enhance competition, increase the  information available to brokers, dealers, and investors, facilitate the offsetting of  investors’ orders, and contribute to best execution of such orders.

These findings and objectives give a paramount place to the interests of investors, and  conclude that the interests of investors are best served by a market structure that is  designed to promote and maintain both (1) an opportunity for interaction of all buying and  selling interest and (2) fair competition among all types of market centers.11 As the  Commission has noted, these objectives can be difficult to reconcile.12 For example,  maximizing order interaction in individual securities may be in tension with market center  competition for order flow, and market center competition for order flow may lead to  fragmentation in the order flow for individual securities.13 As the Commission has stated,  its “task has been to facilitate an appropriately balanced market structure that promotes  competition among markets, while minimizing the potentially adverse effects of  fragmentation on efficiency, price transparency, best execution of investor orders, and  order interaction.”14

The secondary market for U.S.-listed equity securities that has developed within this  structure is now primarily automated.15 The process of trading has changed dramatically

11 Notice of Filing of Proposed Rule Change by the New York Stock Exchange, Inc. to Rescind  Exchange Rule 390; Commission Request for Comment on Issues Relating to Market  Fragmentation, Exch. Act Rel. No. 42450, 65 Fed. Reg. 10577, 10580 (Feb. 28, 2000)  (“Fragmentation Release”).

12 See id. (“although the objectives of vigorous competition on price and fair market center  competition may not always be entirely congruous, they both serve to further the interests  of investors and therefore must be reconciled in the structure of the national market  system”); see also Concept Release at 3597.

13 See, e.g., Concept Release at 3597.

14 Id.

15 See, e.g., id. at 3594.

8

primarily as a result of developments in technologies for generating, routing, and executing  orders, as well as by the requirements imposed by law and regulation.16 Today, equity  trading volume generally is dispersed among many automated trading centers that  compete for order flow in the same stocks, principally by offering execution services designed to fill the needs of the wide variety of market participants.17 Maintaining fair, efficient, and orderly markets requires an understanding of the dependence of modern  markets on algorithms used, among other things, for order routing, handling, and  execution.

The following overview summarizes elements of the market structure most salient to  algorithmic trading, including the various types of equity trading centers and the market  data that facilitates communication among trading centers and participants.

A. Trading Centers

A reasonable place to start in describing current equity market structure is an overview of  the major types of trading centers and their share of volume in NMS stocks.18 Broadly  speaking, the market can be divided into registered national securities exchanges and off exchange trading venues, which include alternative trading systems (ATSs) and several  types of broker-dealer internalization platforms.19 Nearly all of these trading centers  depend on automated systems and algorithms to perform their important role in the  market structure for U.S. equities.

16 Id.

17 Id.

18 See, e.g., id. at 3597-3600. “NMS stock” means any security or class of securities, other  than an option, for which transaction reports are collected, processed, and made available  pursuant to an effective transaction reporting plan. See 17 CFR 242.600(b)(48) (defining  “NMS stock” as “any NMS security other than an option”), 17 CFR 242.600(b)(47) (defining  “NMS security” as “any security or class of securities for which transaction reports are  collected, processed, and made available pursuant to an effective transaction reporting  plan, or an effective national market system plan for reporting transactions in listed  options”). In general, NMS stocks are those listed on a national securities exchange. See Concept Release at 3597 n.20.

19 A broker-dealer internalizes an order when it executes the order out of its own inventory  of securities, rather than routing it to an exchange or other platform, or matches buyers  and sellers together outside of an ATS or exchange. See, e.g., U.S. Securities & Exchange  Commission Investor Publications, Trade Execution: What Every Investor Should Know (Jan. 16, 2013), available at https://www.sec.gov/reportspubs/investor publications/investorpubstradexechtm.html; Concept Release at 3599-3600.

9

Table 1 summarizes, for all NMS stocks in 2019, the percentage of trades, share volume,  and dollar volume executed on each registered exchange or reported to each trade  reporting facility.20 As summarized in Table 2, approximately 78% of all trades were  executed on registered exchanges, and 22% off-exchange; 63% of all shares traded were  executed on-exchange, and 37% off-exchange; and 65% of dollar-volume was executed on exchange, and 35% off-exchange.

Table 1: Percentage of All Trades, Shares, and Dollar Volume in 2019 at  National Securities Exchanges or Reported to Trade Reporting Facilities (TRFs)

Venue/TRF Trades Shares $ Vol.

Cboe BYX 6.2% 3.8% 3.0%

Cboe BZX 8.7% 5.5% 6.4%

Cboe EDGA 4.3% 2.2% 2.1%

Cboe EDGX 6.4% 4.8% 4.7%

IEX 3.8% 2.7% 2.9%

Nasdaq 24.1% 17.2% 19.7%

Nasdaq BX 3.1% 1.8% 1.8%

Nasdaq PSX 0.9% 0.7% 0.9%

NYSE 8.5% 13.5% 12.4%

NYSE American 0.4% 0.3% 0.2%

NYSE Arca 9.4% 8.4% 9.3%

NYSE Chicago <0.01% 0.4% 0.8%

NYSE National 2.1% 1.4% 0.8%

TRF Nasdaq Carteret 18.6% 29.7% 29.3%

TRF Nasdaq Chicago 0.1% 0.1% 0.1%

TRF NYSE 3.5% 7.5% 5.6%

Source: NYSE TAQ

20 Trades executed otherwise than on a national securities exchange must be reported in a  timely manner to a trade-reporting facility. See, e.g., FINRA Rules 6300A - 6380B, 7200A - 7280B. Currently there are three Trade Reporting Facilities.

10

Table 2: Percentage of All NMS Stock Trades, Shares, and Dollar Volume in  2018 at All Registered Exchanges or Reported to TRFs

Venue Trades Shares $ Vol.

Exchanges 78% 63% 65%

Off-Exchange 22% 37% 35%

Source: NYSE TAQ

Currently, only national securities exchanges display quotations in the consolidated  quotation data widely distributed to the public.21 Trades executed off-exchange (i.e., about  35% of equity dollar volume, as shown in Table 2) take place on ATSs and dealer platforms  where quotes are not publicly displayed. Because they do not publicly display quotes,  these venues are commonly referred to as “dark pools” of liquidity.

1. National Securities Exchanges

In 2019, national securities exchanges together executed approximately 78% of trades,  63% of share volume, and 65% of dollar volume in NMS stocks. In 2019, no single  exchange accounted for more than 24% of all NMS stock trades, 17% of all NMS stock share  volume and 20% of NMS stock dollar volume. Figure 1 compares the percentages of trades,  share volume, and dollar volume across all registered exchanges in 2019.

21 These consolidated market data plans are discussed more fully below. See infra Section  III.B.

11

Figure 1: % of Trades, Shares, and Dollar Volume in 2019

While there are now fifteen registered national securities exchanges for equities, and  thirteen equities exchanges operating,22 twelve are owned by three corporate entities,  commonly known as “exchange families.”23 Figure 2 shows the percentage of trades, share  volume, and dollar volume executed at each exchange family during all of 2019.24

22 As of the date of publication of this staff report, Long-Term Stock Exchange, Inc. and  MEMX LLC have not begun trading operations.

23 The exchange families are (1) CBOE Global Markets, Inc., which owns CBOE BYX  Exchange, Inc., CBOE BZX Exchange, Inc., CBOE EDGA Exchange, Inc., and CBOE EDGX  Exchange, Inc.; (2) Nasdaq, Inc., which owns Nasdaq BX, Inc., Nasdaq PHLX LLC, and The  Nasdaq Stock Market LLC; and (3) Intercontinental Exchange, Inc., which owns New York  Stock Exchange LLC, NYSE Arca, Inc., NYSE American LLC, NYSE Chicago, Inc., and NYSE  National, Inc.

24 Long-Term Stock Exchange is not reflected in the 2019 data because it was not yet  executing trades as a national securities exchange.

12

Figure 2: % of Trades, Shares, and Dollar Volume in 2019, by Exchange Family

Trading and communication at national securities exchanges are now almost entirely  automated. Order entry, message acknowledgement, matching algorithms, trade  confirmations, and market data systems all operate at microsecond or nanosecond  timescales.

To reduce time delay, or “latency,” between exchange systems and market participants, and  to otherwise facilitate order entry and trade execution, exchanges offer data and  connectivity services to market participants, including for example, allowing participants to  place their servers close to exchange matching engines and data feeds to minimize data  transmission time. Exchanges also offer market participants a variety of services for (1) receiving and processing data, and (2) moving data between data centers around the  country (such as fiber-optic cables, millimeter waves, and microwaves). Put simply,  computers running sophisticated algorithms consume and analyze this data to help market  participants respond to market developments.

In addition to offering various data services, national securities exchanges generally offer  an extensive range of order types that facilitate automated trading. These order types  provide market participants with a multitude of options for interacting with other market  participants, including, for example, (1) providing liquidity by posting orders to a central

13

limit order book, (2) removing liquidity by matching with an order already resting on the  book, (3) displaying quotes to the market, (4) providing non-displayed liquidity, (5)  accessing liquidity within the quoted spread, (6) accessing non-displayed liquidity, or (7)  repricing orders based on changing market conditions or to meet certain regulatory  obligations. Market participants often use algorithms to pursue more than one of these or  other order options simultaneously. Because most exchange matching algorithms use a  system based upon price-time priority, many order types are oriented towards helping  participants achieve or retain priority in an order book queue.25 Two exchanges also offer  order types that automatically reprice orders based on predicted changes in prices derived  from activity at other markets.26 One registered exchange offers a “speedbump,” or  intentionally-implemented delays in executions, intended to mitigate the advantages that  some market participants may have in receiving and processing market data and rapidly  taking liquidity.27

25 Generally, under a system of price-time priority, better priced orders are at the top of the  order queue, with ties at the same price resolved in favor of the order first to arrive in time.  Ties in arrival time (rare given the granularity of current timestamps) are sometimes  resolved in favor of the order with the largest size. The New York Stock Exchange has a  “parity” model, in which each floor broker, a stock’s designated market maker, and the  central limit order book receive parity in the execution of orders, with allocations for  smaller orders determined through a “parity wheel.” Nasdaq PSX operates on a “price setter pro rata” model, under which resting orders that set the PSX BBO are guaranteed a  certain proportion of an execution against incoming marketable orders, with other resting  orders at the same price filled on a size pro-rata basis out of the remaining shares. Cboe  EDGX uses price-retail-time priority, in which displayed limit orders from or on behalf of  individual retail investors are given priority over other orders at the same price.

26 See IEX Rule 11.190(g); NYSE American Rule 7.31E(h)(3)(D).

27 See IEX Rule 11.510(a). NYSE American recently eliminated its delay mechanism. See NYSE American LLC; Notice of Filing and Immediate Effectiveness of Proposed Rule Change To  Amend Exchange Rules 1.1E and 7.29E To Eliminate the Delay Mechanism and Amend  Exchange Rule 7.31E and Related Exchange Rules To Re-Introduce Previously-Approved Order  Types and Modifiers, Exch. Act Rel. No. 87550, 84 Fed. Reg. 64359 (Nov. 21, 2019).  Recently, several exchanges have proposed “asymmetrical” speedbumps that would  intentionally delay incoming marketable orders, but allowing resting orders to be cancelled  or modified without delay. One was withdrawn, and one was disapproved. See Exch. Act  Rel. No. 84337, 83 Fed. Reg. 50720 (Oct. 9, 2018) (setting aside approval order under  delegated authority after filing was withdrawn); “Order Disapproving Proposed Rule  Change to Introduce a Liquidity Provider Protection Delay Mechanism on EDGA,” Exch. Act  Rel. No. 88261, 85 Fed. Reg. 11426 (Feb. 27, 2020).

14

In addition to providing continuous trading through their limit order books throughout the  day, national securities exchanges may perform opening and closing auctions in their listed  securities.28 These auctions have increased in importance in recent years. This increase is  correlated with, and to at least some meaningful extent has been driven by, the increase in

popularity of investment products that incorporate exchange closing prices in their  operations, including index mutual funds.29 The listing exchanges vary in the percentage of  their volume executed in auctions. Table 3 shows, for example, the average daily  percentage of share volume in auctions for the listing exchanges in 2019.

Table 3: Average Daily Percentage of Share Volume in Auctions Per Listing Exchange in 2019

VENUE Auc. %

Cboe BZX 0.4%

Nasdaq 12.3%

NYSE 33.8%

NYSE American 14.6%

NYSE Arca 8.5%

Source: NYSE TAQ

Most exchanges have adopted fee schedules that differentiate between the providers of  liquidity and the takers of liquidity.30 Most exchanges use a “maker-taker” model, paying  rebates to providers of liquidity, charging a fee to takers of liquidity, with the exchanges keeping any difference between (1) the amount paid to the exchange by takers of liquidity  and (2) the amount paid by the exchange to providers of liquidity, as revenue on each

28 As needed, the listing exchanges also perform intraday re-opening auctions. See, e.g., Plan  to Address Extraordinary Market Volatility, Section VII(B). Some exchanges may also  perform auctions in securities that they do not list, see, e.g., NYSE Arca Rule 7.35-E(a)(1),  and some may match orders at the listing market closing auction price, see Order Setting  Aside Action by Delegated Authority and Approving a Proposed Rule Change, as Modified by  Amendments No. 1 and 2, to Introduce Cboe Market Close, a Closing Match Process for Non BZX Listed Securities under New Exchange Rule 11.28, Exch. Act Rel. No. 88008 (Jan. 21,  2020).

29 See, e.g., Robin Wigglesworth, “The 30 minutes that have an outsized role in US stock  trading,” Financial Times (Apr. 24, 2018); Corrie Driebusch, Alexander Osipovich and  Gregory Zuckerman, “What’s the Biggest Trade on the New York Stock Exchange? The Last  One,” The Wall Street Journal (Mar. 14, 2018).

30 Rule 610(c) of Regulation NMS caps the access fee for executions against the best  displayed prices of a national securities exchange at $0.0003 per share for stocks price at  or above $1.00. See 17 CFR § 610(c)(1).

15

trade. A smaller number of exchanges employ a taker-maker model, paying rebates to the  consumers of liquidity, charging the providers, and again keeping the difference as  revenue. An even smaller number of exchanges charge a flat fee for all orders. Because of  the substantial amount of money rebated back to trading participants, some order types  may be oriented towards helping participants capture these rebates.31

2. Alternative Trading Systems

In 2019, thirty-three ATSs executed trades in NMS stocks.32 That year, these ATSs executed  approximately 10.2% of share volume in NMS stocks. The top two ATSs each executed  approximately 1-2% of share volume in NMS stocks, with most ATSs executing under 1% of  share volume.

Table 4: 2019 Top Ten ATSs by Share Volume: Percentage of ATS Volume, Off Exchange Volume, and NMS Stock Volume

ATS % ATS % Off-Exch. % NMS UBS ATS 19.3% 5.3% 2.0% CROSSFINDER 9.8% 2.7% 1.0% JPM-X 7.1% 1.9% 0.7% MS POOL (ATS-4) 7.1% 1.9% 0.7% SIGMA X2 6.7% 1.8% 0.7% LEVEL ATS 6.6% 1.8% 0.7% THE BARCLAYS ATS 6.0% 1.6% 0.6% BIDS ATS 5.1% 1.4% 0.5% SUPERX ATS 3.9% 1.1% 0.4% MS TRAJECTORY CROSS (ATS-1) 3.1% 0.8% 0.3% Source: FINRA OTC/ATS Transparency

31 Such order types may include, for example, non-routable post-only orders that seek only  to provide liquidity (on venues providing rebates for the provision of liquidity) and will  not, upon order entry, execute against a resting order on the other side of the market  (either by be re-priced automatically or cancelled) or be routed to a different trading  center.

32 As reflected in the FINRA OTC/ATS Transparency data. FINRA OTC/ATS Transparency  Data is provided via http://www.finra.org/industry/OTC-Transparency and is copyrighted  by FINRA.

16

Currently, no NMS stock ATS publishes quotation data in the consolidated data feed. 33 In  other words, at the moment, all NMS stock ATSs are operating as “dark pools.” ATSs could  publish their quotation data to operate as electronic communications networks, or  “ECNs.”34 One ATS, IntelligentCross, currently publishes its own data feed of quotations  and executions on the ATS.35

ATSs offer many different types of services and cater to different trading objectives. Some  offer block order36 crossing networks, others match smaller customer orders with other  customers and/or with broker-dealer or bank inventory, and some allow for order  matching to be segmented by specific categories of market participants. In many cases, the  same market makers that provide liquidity on exchanges also provide liquidity on ATSs.  Frequently ATS functionality is intended to mitigate the effect of trading on subsequent  prices for an instrument.37 Some offer unique order types not available on exchanges, such  as conditional orders, intended to facilitate the search for larger blocks of liquidity. Many  ATSs are operated by multi-service broker-dealers, while some are operated by  independent firms or consortiums. A number of ATSs have developed trading models that  are alternatives to the more prevalent price-time priority matching engines: for example,

33 As noted above, supra n. 20, however, trades in these off-exchange venues are publicly  reported.

34 See 17 CFR § 600(b)(24); see also Concept Release at 3599 (“The key characteristic of an  ECN is that it provides its best-priced orders for inclusion in the consolidated quotation  data, whether voluntarily or as required by Rule 301(b)(3) of Regulation ATS”). In the past,  some ECNs with displayed quotations have, at times, represented a significant amount of  market share, and some eventually evolved into registered national securities exchanges  (including all four CBOE equities exchanges, IEX, NYSE Arca, Island, Instinet, and BRUT).

35 See, e.g., IntelligentCross Form ATS-N, Item 15 (filing of Jan. 16, 2020) (available at: https://www.sec.gov/divisions/marketreg/form-ats-n-filings.htm).

36 Generally speaking, a block order is a particularly large order for a given market, and its  precise meaning changes in different contexts. For example, Regulation NMS defines an  order of block size as an order of at least 10,000 shares or for a quantity of stock having a  market value of at least $200,000. 17 CFR § 242.600(b)(10).

37 The importance of managing the informational and price impact of order entry and  trading activity is described in more detail in the Appendix. See infra n. 320 and  accompanying text.

17

one ATS uses machine learning to execute orders at a time intended to minimize  subsequent impacts on prices,38 and others allow for periodic auctions.39

3. Broker-Dealer Internalizers

A third major category of venue for equity trading is broker-dealer internalization. Like  national securities exchanges and ATSs, internalizers are heavily dependent on  sophisticated algorithms to conduct their core functions. As described in more detail below, there are several types of internalization, including wholesale market makers,  single-dealer platforms, and central risk book trading and block positioning. Internalized  trades of broker-dealers reflect liquidity that is not included in public quotation data.40 In  2019, approximately 27% of NMS stock share volume was executed by broker-dealer  internalizers.41

While hundreds of broker-dealers internalize trades, much of this volume is handled by a  relatively small number of large firms. Table 5 illustrates this concentration.42 In 2019,

38 See Form ATS-N filings and information for IntelligentCross ATS, available at  https://www.sec.gov/divisions/marketreg/form-ats-n-filings.htm.

39 See Form ATS-N filings and information for CODA ATS, available at  https://www.sec.gov/divisions/marketreg/form-ats-n-filings.htm.

40 See, e.g., Concept Release at 3599.

41 This figure was calculated by subtracting the percentage of share volume in NMS stocks  executed on ATSs from the total percentage of share volume in NMS stocks executed off exchange, as reflected in NYSE TAQ data and FINRA OTC/ATS Transparency data. Cf. also Concept Release at 3599, noting that in September 2009 approximately 17.5% of NMS  stock share volume was executed by broker-dealer internalization.

42 This list should be read as a rough illustration of the distribution of share volume in the  non-ATS off-exchange market. This list is derived from the OTC Transparency data FINRA  makes available to the public on its website. FINRA’s public data reflects firms with the  obligation to report each off-exchange trade to a TRF (generally the executing broker dealer), so only shows the firm involved on one side of each trade. Moreover, trading by  broker-dealers that are not registered with FINRA (but that are registered with another  SRO) is not reflected in this data, regardless of the volume of such off-exchange trading,  because non-FINRA-members never have obligations to act as the reporting party under  FINRA’s trade reporting rules. See FINRA Rule 6110(b). Currently, a list of OATS Reporting  Non-FINRA Member firms is available at https://www.finra.org/industry/oats/oats reporting-non-finra-member-firm-list. Finally, FINRA’s data for the period covered here  groups all firms executing a number of trades below certain thresholds into a single de  minimis category (though more recent FINRA data no longer includes a de minimis

18

each of the two largest internalizers executed more share volume off-exchange than was  executed in any individual ATS, and more share volume than was executed on each of eight national securities exchanges.43 Internalizers can be significant sources of liquidity in  today’s markets.

Table 5: 2019 Top Ten Internalizing Broker-Dealers by Share Volume: Percentage of  Internalized Volume, Off-Exchange Volume, and NMS Stock Volume

Firm % Intern. % Off-Exch. % NMS CITADEL SECURITIES LLC 24.3% 16.7% 6.2% VIRTU AMERICAS LLC 13.2% 9.0% 3.4% G1 EXECUTION SERVICES, LLC 6.9% 4.8% 1.8% TWO SIGMA SECURITIES, LLC 1.5% 1.0% 0.4% WOLVERINE SECURITIES, LLC 0.7% 0.5% 0.2% JANE STREET CAPITAL LLC 0.5% 0.4% 0.1% UBS SECURITIES LLC 0.4% 0.3% 0.1% VIRTU FINANCIAL BD LLC 0.3% 0.2% 0.1% GOLDMAN SACHS & CO. LLC 0.3% 0.2% 0.1% ACS EXECUTION SERVICES, LLC 0.1% 0.0% 0.0% Source: FINRA OTC/ATS Transparency

B. Market Data

Modern equity markets are connected in part by the data flowing between market centers.  An enormous volume of data is available to market participants. In recent years, there has  been an exponential growth in the amount of market data that is available, the speed with

which it is disseminated, and the computer power used to analyze and react to price  movements. As discussed below, for different types of investors and market professionals,  the speed with which information can be acquired, analyzed, and acted upon is valued to

category). See FINRA Rule 6110(b); Order Approving Proposed Rule Change To Expand OTC  Equity Trading Volume Data Published on FINRA's Website, Exch. Act Rel. No. 86706, 84 Fed.  Reg. 44341 (Aug. 23, 2019) (change to FINRA rules that would, among other things,  eliminate the de minimis exception to public disclosures as of December 2, 2019); FINRA  Regulatory Notice 19-29 (Sept. 13, 2019). In other words, it is likely that the attributions of  volume in this data source understate the volume of some market participants, and omit  entirely other market participants.

43 Both of these broker-dealers offer a range of internalization services, including for both  retail and institutional investors, and, as reflected in retail broker order routing public  disclosures, are particularly significant internalizers of retail investor trades.

19

varying degrees. Most equity market participants use, in some respect, quotation data44 and last-sale transaction data.45 Market centers and securities information processors  (SIPs), described below, also communicate important regulatory and administrative  messages, such as when trading in a particular security is unexpectedly halted or paused.  Some data feeds carry additional information about market dynamics, such as messages  providing updates on order imbalances at regular intervals leading up to the daily closing  auction.46

This wide range of market data provided to market participants by exchanges is distributed through two broad categories of data feeds: (1) consolidated data feeds, and (2)  proprietary data feeds.

The consolidated data feeds are operated by the self-regulatory organizations (SROs) via  National Market System (NMS) plans pursuant to Commission regulation and oversight.47 The consolidated data feeds include top of book quotations, last sale information, and

44 Quotation data can include information about both the best available prices at a given  market (often called the “top of book”) and quotes resting in the order book at prices  higher (for sell orders) or lower (for buy orders) (often called “depth of book” data). For  some market participants information about the cancellation or modification of individual  quotations is also important.

45 As noted above, the Exchange Act includes a Congressional finding that it is in the public  interest and appropriate for the protection of investors and maintenance of fair and  orderly markets to assure the availability to brokers, dealers, and investors of information  with respect to quotations for and transactions in securities. See 15 U.S.C. 78k 1(a)(1)(C)(iii).

46 Ahead of each auction, each listing exchange gathers market and limit orders to execute  in the closing auction. The number of buy and number of sell orders may not align. In  order to attract liquidity and potentially improve the quality of the closing auction, listing  exchanges disseminate messages providing the side and size of an order imbalance as the  auction approaches.

47 This section describes the currently prevailing structure for the provision of  consolidated equity data feeds. In February 2020, the Commission proposed rules that  would update and expand the content of NMS market data and that would introduce a  decentralized consolidation model under which competing consolidators, rather than the  existing exclusive securities information processors, would collect, consolidate, and  disseminate certain NMS information. See Market Data Infrastructure, Exch. Act Rel. No.  88216, 85 Fed. Reg. 16726 (Mar. 24, 2020) (“Market Data Infrastructure Proposal”).

20

important regulatory messages from exchanges.48 Currently, there are three equity market  data plans: the CQ plan (for quotations in securities not listed on Nasdaq), the CTA plan (for  transaction reports in securities not listed on Nasdaq), and the UTP Plan (for both  quotation and transaction reports in Nasdaq-listed equities).49 These plans together are  often referred to as the “SIPs.”50

Proprietary data feeds offer additional and different market information from the SIPs. For  example, some proprietary data feeds provide all displayed order messages at an exchange,  including individual odd-lot orders, as well as order modifications and cancellations; others

may not provide message-by-message data, but summarize the total displayed shares  available at each level in the order book; others provide only the top-of-book across an  exchange family’s related markets; and some offer detailed auction imbalance information.  For various reasons, including because they do not need to go through a consolidation  process at a separate geographic location, proprietary data feeds often reach market  participants sooner than the SIPs.51

48 Best-priced quotation and last sale data is often referred to as “core data.” See, e.g., Order  Setting Aside Action by Delegated Authority and Approving Proposed Rule Change Relating to  NYSE Arca Data, Exch. Act. Rel. No. 59039, 73 Fed. Reg. 74770, 74779 (Dec. 9, 2008) (“Core

data is the best-priced quotations and comprehensive last sale reports of all markets that  the Commission, pursuant to Rule 603(b), requires a central processor to consolidate and  distribute to the public pursuant to joint-SRO plans”); Order Directing the Exchanges and  the Financial Industry Regulatory Authority to Submit a New National Market System Plan  Regarding Consolidated Equity Market Data, Exch. Act Rel. No. 88827, 85 Fed. Reg. 28702,  28703 (May 13, 2020) (“SIP Governance Order”) (noting that “core data” consists of “(1)  The price, size, and exchange of the last sale; (2) each exchange’s current highest bid and

lowest offer, and the shares available at those prices; and (3) the national best bid and offer  (‘‘NBBO’’) (i.e., the highest bid and lowest offer currently available on any exchange)”).

49 In May 2020, the Commission issued an order requiring the national securities exchanges  and FINRA to consolidate the three current equity market data plans into a new single  equity market data plan and to implement specific governance provisions within that plan.  See SIP Governance Order. Certain SROs have petitioned for review of this order in the D.C.  Circuit.

50 “SIP” is an acronym for “securities information processor,” which is defined in Exchange  Act Section 3(a)(22)(A), 15 U.S.C. 78c(a)(22)(A); see also Exchange Act Section 11A(b), 15  U.S.C. 78k-1(b).

51 However, data distributed over the consolidated feeds cannot be transmitted by  proprietary feed to a vendor or user any sooner than it is transmitted to a consolidated  data processor. See 17 CFR 603(a); Regulation NMS, Exch. Act Rel. No. 51808, 70 Fed. Reg.  37495, 37567 (June 25, 2005) (“Reg NMS Adopting Release”) (“adopted rule 603(a)

21

Equity market data is physically disseminated and moved between data centers in a range  of ways, including by fiber-optic cable and wirelessly via microwave towers. Moreover, the  data can be physically accessed in a variety of ways, including through servers co-located in  an exchange’s data center, ports and wires with various capacities and bandwidths, as well  as through hardware that can process data directly, rather than relying on the often slower  process of using software to process the data following receipt. The methods used to  physically access and process data affect the speed and efficiency with which market  participants are able to transact in the markets.52

The availability of different levels of market data and different access speeds to both  markets and market data can advantage some market participants over others. For  example, by accessing more granular data from proprietary market data feeds at higher  speed, some users may be able to react to market events more strategically and more  quickly than participants relying only on SIP data.53 Similarly, brokers and other market  participants using advanced connectivity tools, such as microwave data transmissions and  high-bandwidth connections, can process data and enter orders more rapidly than other  market participants, particularly during periods of high volume when message traffic, and  therefore network latency, may be at its highest.

These market data considerations—including differing levels of both content and speed of  access—extend beyond the cash equity markets. Many market participants make trading  decisions and control risk using information from trading venues for other types of  instruments. Accordingly, market data from those trading venues can be very important to

prohibits an SRO or broker-dealer from transmitting data to a vendor or user any sooner  than it transmits the data to a Network processor”). For additional discussion of  differences between the current consolidated data feeds and proprietary feeds, see, e.g.,  Market Data Infrastructure Proposal at 20-25.

52 For example, “co-location is a service that enables exchange customers to place their  servers in close proximity to an exchange’s matching engine in order to help minimize  network and other types of latencies between the matching engine of the exchange and the  servers of market participants.” Notice of Proposed Order Directing the Exchanges and the  Financial Industry Regulatory Authority to Submit a New National Market System Plan  Regarding Consolidated Equity Market Data, Exch. Act Rel. No. 87906, 85 Fed. Reg. 2164,  2169 n.55 (Jan. 14, 2020) (“SIP Governance Proposed Order”). Similarly, “[d]ata  connections that use fiber optic cable transmit data more slowly than data connections that  use wireless microwave transmissions, though microwave connections are susceptible to  interruption by weather conditions and are therefore less reliable than fiber connections.”  Id.

53 See, e.g., SIP Governance Proposed Order at 2169-70 (discussing potential implications  for competitiveness between the consolidated data feeds and exchange proprietary feeds).

22

the algorithmic trading strategies of equity market participants. Common examples of this  cross-market data use include access to options market data and futures market data.54

III. Overview of Debt Market Structure

Cash debt markets have historically operated as over-the-counter principal markets, in  which participants trade for their own accounts. Historically, the market operated largely by “voice,” with trades negotiated and effected bilaterally between counterparties. In the  last several decades, a range of dealers and platforms have implemented several models of  electronic trading. Some of these models have automated aspects of the bilateral  communication and trading process, while others have provided alternative trading models  such as central limit order books. The development of automated tools and platforms has  differed across the Treasury, corporate, and municipal bond markets, reflecting significant underlying differences in the terms of these instruments and the structure of the markets  in which they trade.

A. Types of Debt Securities or Instruments

Unlike equity securities, debt securities are not standardized, even by issuer. A given  issuer might have tens, hundreds, or more than a thousand different types or “series” of debt securities outstanding, each with a different notional value, maturity date, and interest  rate. When a particular issuer has multiple series of bonds outstanding, secondary market  liquidity is generally concentrated in the more recently-issued bonds (and, among the more  recently-issued bonds, in the series with the greater amount outstanding).

1. U.S. Treasury Securities

The market for U.S. Treasury securities is the deepest and most liquid government  securities market in the world.55 The U.S. Treasury issues bills, nominal fixed-rate coupon  securities, nominal floating rate securities, and inflation-indexed securities (TIPS). Most  secondary trading in Treasuries occurs across the most-recently issued (or, “on-the-run”)  nominal coupon securities.56

54 For example, in the futures market, the E-Mini S&P 500 Futures contract traded on the  CME is often regarded as a central focal point for price formation in the equities market.  See, e.g., Joel Hasbrouck, “Intraday Price Formation in U.S. Equity Index Markets,” 58  Journal of Finance 58: 2375-2399 (Dec. 2003).

55 See, e.g., Joint Staff Report: The U.S. Treasury Market on October 15, 2014, at 1 (July 13,  2015) (“Treasury Market Report”).

56 Id. at 11.

23

Treasury securities are traded on multiple venues.57 Interdealer trading of on-the-run Treasury securities occurs mainly on centralized electronic trading platforms using a  central limit order book. These venues are the primary points of price discovery for the on the-run securities. This market has evolved considerably over the past decade. Where  participation on interdealer platforms was once limited to primary dealers, principal  trading firms now account for more than half of the trading activity in the electronic  interdealer markets.58 Dealer-to-customer trading, in both on-the-run and off-the-run  securities, is usually bilateral, either through voice or through an electronic platform, using  for example, a request-for-quote (RFQ) process or streaming quotes.59 Some dealers and  electronic market makers now provide their customers with a direct stream of continuous  prices and sizes at which they are willing to trade in a range of issues.60 With price  discovery increasingly occurring in electronic order books, trading in off-the-run securities  occurs primarily through voice channels.61

57 Id.

58 See, e.g., id. at 36. More specifically, the Report notes that 56% of the volume in the on the-run 10-year note is handled by principal trading firms, with about 35% handled by  bank-dealers, and the remaining 9% being split among non-bank dealers, hedge funds, and  asset managers. More recent data show that, as of 2019, “PTFs account for around 60  percent of electronic IDB volumes.” See Remarks of Deputy Secretary Justin Muzinich at the  2019 U.S. Treasury Market Structure Conference, Sept. 23, 2019 (available at: https://home.treasury.gov/news/press-releases/sm782).

59 Id. at 11.

60 See, e.g., Kevin McPartland, U.S. Treasury Trading No Longer a Divided Market, Greenwich  Associates (Dec. 12, 2018); Kevin McPartland, How Bilateral Streams for U.S. Treasuries  Really Work, And What They Mean for the Market, Greenwich Associates (June 18, 2019),  available at: https://www.greenwich.com/blog/how-bilateral-streams-us-treasuries really-work.

61 Treasury Market Report at 35. Recently, non-central limit order book electronic  platforms, such as OpenDoor Trading, have begun providing electronic venues for trading  in off-the-run Treasuries and TIPS. See, e.g., John McCrank, OpenDoor opens up trading in  illiquid U.S. Treasuries, Reuters (June 14, 2017), available at:

https://www.reuters.com/article/us-usa-treasuries-opendoor-idUSKBN1952DN.

24

2. Corporate Bonds

Corporate bond trading predominantly occurs over voice. Much of this voice activity is  concentrated in the largest dealers—for example, one recent study estimates that 56% of  buy-side volume in U.S. investment-grade corporate bonds is with the five largest dealers.62

In the past decade, a number of venues have developed to trade corporate bonds  electronically. A recent estimate is that 26% of corporate bond volume was traded  electronically in the third quarter of 2018.63 As described more fully below, much of this  activity can be described as the automation of bilateral voice trading or RFQ trading. All-to all trading, in which any participant can provide quotes to and trade with any other  participant, is increasingly a meaningful proportion of the electronic corporate bond  market; for example, approximately 8% of volume in investment grade corporate bond  may be executed on all-to-all platforms.64 Roughly 70% of corporate bond trades are for  fewer than one hundred bonds, and 90% of these small trades are effected electronically;  however, these small trades represent approximately 3-4% of the total value of corporate  bonds traded on any given day.65 The electronic trading venues are regulated in a range of  ways, with some as ATSs, some as broker-dealers, and others outside of either of those two  regulatory structures.66 One national securities exchange, NYSE, provides the capability to  trade corporate bonds.

62 Kevin McPartland, Corporate Bond Trading in 2019: Competition is Good, Complexity is  Not, Greenwich Associates, p. 5 (Jan. 8, 2019).

63 Id. at 2, 4.

64 Id. at 8

65 Kevin McPartland, “The Challenge of Trading Corporate Bonds Electronically,” 1,  Greenwich Associates (May 13, 2019) available at:

https://www.greenwich.com/blog/challenge-trading-corporate-bonds-electronically.

66 See, e.g. FIMSAC Electronic Trading Subcommittee Recommendation on Oversight of  Electronic Trading Platforms for Corporate and Municipal Bonds (July 16, 2018). For a  more detailed discussion of quoting and trading on a recent sample of corporate bond

activity on ATSs, see Staff of the Division of Economic and Risk Analysis of the U.S.  Securities and Trading Commission, Access to Capital and Market Liquidity, pp. 178-90  (Aug. 8, 2017) (“DERA Bond Study”), available at: https://www.sec.gov/files/access-to capital-and-market-liquidity-study-2017.pdf

25

Average daily volume traded in corporate bonds has grown from a recent low of $14.3  billion in 2008 to $31.2 billion in 2018.67 Over the same period, the amount of corporate  debt issued annually grew from approximately $717 billion in 2008 (from 946 new issues)  to more than $1.3 trillion in 2018 (from 1,270 new issues) with a high of more than $1.65 trillion in 2017 (in 1,671 new issues).68 As discussed above, most corporate bond trading  is concentrated in new issues, so the increase in ADV may be related to the increase in  corporate bond issuance.69

Generally, corporate bonds are held by investors, with trades facilitated by broker-dealers  and other intermediaries. Demand for intermediation is driven by various factors,  including the general difficulty of finding natural counterparties for specific bonds due to  the idiosyncratic nature of different series of bonds.70 This intermediation requires capital,  which, at least at the dealer level, has been relatively more limited since the credit crisis ten  years ago.71 Technology has helped mitigate the drop in dealer liquidity, and other types of liquidity providers have entered the corporate debt markets. The trading of corporate  bond ETFs, in particular, has introduced new liquidity suppliers, including those who may  not be able to meet the capital requirements imposed on large dealers, including principal  trading firms and quantitative hedge funds.72

67 Capital Markets Fact Book 2019, 20, SIFMA, available at:

https://www.sifma.org/resources/research/fact-book/.

68 Id. at 7-8.

69 See McPartland, Corporate Bond Trading in 2019 at 3.

70 McPartland, The Challenge of Trading Corporate Bonds Electronically, at p. 2.

71 See, e.g., Tobias Adrian, Nina Boyarchenko, & Or Shachar, “Dealer Balance Sheets and  Bond Liquidity Provision,” Federal Reserve Bank of New York Staff Report No. 803 (Mar.  2017). For a more detailed discussion of research on recent trends in corporate bond  market liquidity and potential relationships to regulatory changes, see DERA Bond Study at  109-119.

72 McPartland, Corporate Bond Trading in 2019 at 5. For a more detailed discussion of  studies on the evolution of electronic trading in corporate bond markets, see DERA Bond  Study at 119-123.

26

3. Municipal Bonds

Municipal bond investors predominantly hold municipal securities for the long-term,73 and  a significant percentage of municipal bonds are held by retail investors.74 Trading in  municipal bonds is concentrated in the period after issuance, and becomes infrequent  afterwards.75 Of the approximately one million outstanding series of municipal securities,  on average slightly more than one percent trade on any given day.76

Municipal bonds are predominantly traded over-the-counter by voice, either between dealers and customers or between dealers.77 The market is fragmented, given, among  other things, the number of unique municipal securities, the number of issuers, and low  trading volume for most bonds.78 Largely because of their tax treatment, shorting of  municipal bonds is difficult and rare.79 In recent years, several platforms have developed  that facilitate the electronic trading of municipal bonds. Several of these are ATSs. One

73 Simon Z. Wu, John Bagley & Marcelo Vieira, Staff of the Municipal Securities Rulemaking  Board, Analysis of Municipal Securities Pre-Trade Data from Alternative Trading Systems, at  4 (Oct. 2018), available at: http://www.msrb.org/~/media/Files/Resources/Analysis-of Municipal-Securities-Pre-Trade-Data.ashx?la=en (“MSRB ATS Study”).

74 See, e.g., Municipal Securities Rulemaking Board, “Muni Facts,” (Mar. 2019), available at: http://www.msrb.org/msrb1/pdfs/MSRB-Muni-Facts.pdf (noting that nearly two-thirds of  municipal securities are held by individual investors either directly or through mutual  funds).

75 Id.

76 Id.; 2018 Fact Book, 34, Municipal Securities Rulemaking Board, available at: http://www.msrb.org/Market-Transparency/Market-Data-Publications/MSRB-Fact Book.aspx (average of 15,588 unique securities traded per day in 2018).

77 MSRB ATS Study at 4.

78 Id.

79 The interest on most municipal securities is exempt from federal income tax and, in some  cases, state and local taxes. The Internal Revenue Service does not allow both the  borrower and lender of a municipal security to claim a tax exemption, so in effect the  lender of a municipal security would be trading tax-exempt interest for taxable interest.  See Exch. Act Rel. No. 34-33743 (Mar. 9, 1994), 59 FR 12767, 12769 n.24 (Mar. 17, 1994)  citing Internal Revenue Code, Sec. 6045(d); see also FINRA Notice 15-27 (July 2015).

27

recent study estimates that 12-15% of municipal bond trading is electronic.80 While some  municipal bond ETFs exist, they are still a relatively small, but growing, proportion of fixed  income ETFs.81

Among fixed income securities, municipal bonds have a uniquely high percentage of direct  retail investors. For example, one recent study estimates that retail investors directly hold  42% of municipal bonds by value, compared to 13% of Treasuries and 8% of corporate  bonds.82 Retail investors may purchase municipal bonds through broker-dealers or  investment advisers, who in turn source liquidity from banks or nonbank dealers and other liquidity providers.83 The high volume of retail participation results in a large number of  small trades, which may make aspects of the municipal bond market suited to electronic  trading.84

B. Data and Communications

In debt markets, market data collection and distribution is uneven and fragmented. Pre trade transparency information on quotes or pricing generally can only be purchased from  individual platforms or arranged through bilateral relationships. Post-trade transparency,  in the form of transaction reports, generally is available for corporate and municipal bonds.

1. Transaction Reports in Corporate Bonds: TRACE

Transactions in corporate bonds must be reported to the Trade Reporting and Compliance  Engine (TRACE) operated by FINRA.85 TRACE data is disseminated by FINRA immediately

80 Kevin McPartland, The Modernization of Municipal Bond Trading, 2, Greenwich Associates  (May 6, 2019).

81 See id. at 5, estimating that municipal bonds make up approximately 6% of fixed income  ETF assets under management; see also Simon Z. Wu and Meghan Burns, Staff of the  Municipal Securities Rulemaking Board, Municipal Bond ETFs: Impact on the Municipal  Bond Market, at 5-6 (Apr. 2018), available at:

http://www.msrb.org/~/media/Files/Resources/MSRB-Municipal-Bond-ETFs Report.ashx.

82 McPartland, The Modernization of Municipal Bond Trading, at 3.

83 Id. at 5.

84 Id. at 7.

85 See FINRA Rule 6730.

28

upon receipt.86 Each FINRA member that is party to a transaction in a TRACE-eligible  security must report the trade as soon as practicable, but generally no later than within  fifteen minutes of the execution of the trade.87 Detailed data is available through subscription data feeds, and the FINRA website makes available for free aggregate statistics  and details about individual trades, bonds, and issuers. The transaction data disseminated  by FINRA includes, among other things, the issuer, CUSIP number for the bond, the entity  type of the reporting and contra parties,88 execution time, quantity, price, side (buy or sell),  size,89 and whether the trade was executed on an ATS.

FINRA has proposed publishing aggregate trade count and volume statistics for each  corporate bond ATS, by CUSIP.90 The stated purpose of this proposal is to provide the  market with more readily available information about potential sources of liquidity. FINRA currently makes similar data available for equity ATSs.

2. Transaction Reports in Municipal Bonds: EMMA

Transactions in municipal bonds must be reported to the Municipal Securities Rulemaking  Board’s (MSRB) Real-time Transaction Reporting System (RTRS).91 The MSRB  disseminates the data upon receipt through subscription data feeds, and the MSRB’s

86 Following a recommendation by the Commission’s Fixed Income Market Structure  Advisory Committee, FINRA has proposed a pilot program to study potential changes to  corporate block trade dissemination. See FINRA Notice 19-12 (Apr. 12, 2019). Under the  proposed pilot, the cap for disclosing the exact size of a trade would be raised to $10  million for investment-grade corporate bonds and to $5 million for non-investment grade,  and delay by 48 hours dissemination of reports for trades above those caps.

87 See FINRA Rule 6730(a).

88 These entity types can be broker-dealer, customer, non-member affiliate, and alternative  trading system. Six months after a trade, the FINRA-member parties to each trade are  identified.

89 For trades in investment grade bonds up to $5 million, the exact size of the trade is  disseminated, but for trades above $5 million, the trade size is listed as “5MM+”; the  analogous cap for non-investment-grade corporate bonds is $1 million. After six months,  the exact size of capped reports is revealed. These rules mean that, for example, for several  months, a $6 million dollar trade and a $100 million trade in an investment grade bond  would both be reported as “5MM+”.

90 See FINRA Notice 19-22 (July 9, 2019).

91 Reports can be sent either directly to the MSRB or through the NSCC. See MSRB Rule G 14 RTRS Procedures(a)(i). Most trades must be reported within fifteen minutes. See MSRB  Rule G-14 RTS Procedures(a).

29

Electronic Municipal Market Access (EMMA) website makes available the trade data  together with free aggregate statistics about individual trades, bonds, and issuers. The  transaction data disseminated includes, among other things, the issuer, CUSIP number of  the bond, trade date and time, price, size, trade type (i.e., inter-dealer or dealer-customer),  and whether the trade was executed on an ATS.

IV. Benefits and Risks of Algorithmic Trading in Equities

As described above, the current markets for secondary trading in NMS stocks are  predominantly electronic. While some pockets of activity can be described as manual, most  of the lifecycle of trading is automated. Algorithms now facilitate the provision of and  search for liquidity by a broad range of participants in the equities market across a diverse  set of trading venues. This pervasive automation has also created new operational risks for  firms and the market infrastructure more generally. Broadly speaking, studies have shown  that algorithmic trading in equities has improved many measures of market quality and  liquidity provision during normal market conditions, though other studies have also shown  that some types of algorithmic trading may exacerbate periods of unusual market stress or  volatility.

A. Investors

For many investors, both retail and institutional, algorithms play a significant role in the  investment and trading process. While investor orders are generally routed and executed  through broker algorithms of various types, the ability to use routing and execution  algorithms directly is also becoming more accessible to investors, as are the data feeds and  processing tools that are essential to the use of algorithms. Some investors, both retail and  institutional, also use algorithms to actively make investment and trading decisions  through the rapid analysis of potentially voluminous amounts of market data. In other  cases, investors track an outside reference, such as an index, and investment and trading  decisions may be informed by algorithmically-determined decisions about composition implicit in a benchmark index or other standard or set of rules.92

92 This may be the case for some index mutual funds and ETFs, and investors holding ETNs  may have a similar investment experience.

30

1. Retail Investors

As has been the case for many years,93 retail brokers largely route marketable retail  investor orders (i.e., orders that can be executed immediately) to internalizing broker dealers known as “wholesale” market makers for execution.94 Wholesale market-makers  can frequently provide retail orders with some degree of price improvement, meaning they  can execute the orders inside the spread of the national best bid and national best offer.95 Wholesale market-makers may also be able to provide retail orders with size improvement  (i.e., more shares at a single price point than may be available at the national best bid or  national best offer quoted on national securities exchanges). Some retail brokers receive  payment for order flow in exchange for routing orders to these market-makers.96  Wholesale market makers are willing to provide price improvement to retail investors and  purchase order flow from brokers because access to the orders provides wholesalers, who

93 See, e.g., Concept Release at 3600 (“OTC market makers, for example, appear to handle a  very large percentage of marketable (immediately executable) order flow of individual  investors that is routed by retail brokerage firms”).

94 See, e.g., CFA Institute, Dark Pools, Internalization, and Equity Market Quality, at 16  (2012) (“Internalization is also thought to account for almost 100% of retail marketable  order flow”).

95 In recent years, several firms, coordinated by the Financial Information Forum, have  voluntarily disclosed retail execution quality statistics that include information on the  average percentage of retail orders given price improvement and the average price  improvement on each order. See, e.g. https://fif.com/tools/retail-execution-quality statistics. These summary statistics are distinct from the disclosures required by SEC Rule  605. A staff review of these voluntary statistics from several retail brokers and wholesale  market-makers for Q1 of 2019 indicates that some wholesale market-makers provide price  improvement to more than approximately 85% of retail orders from a range of order sizes.  In some cases, such as for smaller orders in more frequently-traded stocks, on average  more than 95% of retail orders received price improvement. The monthly disclosures  required by SEC Rule 605 also include information about each market center’s price  improvement, but do not specifically break out statistics for retail orders. Because  wholesale market-makers do not engage in quoting activity as part of their internalization  activities, they can execute trades at prices more granular than the one-penny increment  required for quotes in most equity securities.

96 For a more thorough description of payment for order flow practices, their historical  development and regulation, and some potential concerns about the practice, see Staff of  the Division of Trading and Markets, U.S. Securities & Exchange Commission, Memorandum  to Equity Market Structure Advisory Committee on Certain Issues Affecting Customers in the  Current Equity Market Structure, 5-11 (Jan. 26, 2016).

31

generally have more information and processing power than retail traders and brokers, with a relatively low risk of adverse selection97 and a potential opportunity to profit from  the informational content in aggregate retail order flow. For example, because wholesale  market-makers are allowed to choose whether to execute any given order based on their  own preferences and views of market conditions, they may, for example, trade against incoming retail orders or route relatively risky or unfavorable orders to exchanges or other  market centers.98

This discussion demonstrates the significant extent to which fast, effective processing of  market data is central to the business of wholesale market-making in the cash equity  markets: the market maker algorithm’s data-driven assessment of the market is not only  central to its obligations with respect to best execution and compliance with the order  protection rule, but allows it to make order-handling decisions, and provides the standards  for evaluating price improvement. Wholesale market-makers also, over the long term,  acquire voluminous amounts of market data, including information about retail order  flows, which can be used in future modeling for order-handling, trading, and risk decisions.

Some specialized retail brokers allow individual retail customers to use more sophisticated  broker algorithms that operate in a manner that is generally otherwise available only to  institutional investors as described below, or allow retail customers to use their own  algorithms. These specialized retail brokers typically operate on a smaller scale than the  retail brokers described above, and may not rely as heavily on wholesaler market-makers  as more traditional retail brokers.

2. Institutional Investors

The broad category of “institutional investors” encompasses a diverse range of market  participants.99 This category includes, among others, registered investment companies,  pension funds, insurance companies, endowments, and private investment funds such as hedge funds, all of which employ a wide variety of trading strategies. While their needs and  approaches to trading vary, they generally share a common focus on achieving high  execution quality, which requires them to effectively manage the explicit and implicit costs  of trading. The diversity of approaches to trading among institutions is reflected in the

97 Id. at 6; see also Concept Release at 3612 (“Liquidity providers generally consider the  orders of individual investors very attractive to trade with because such investors are  presumed on average to not be as informed about short-term price movements as are  professional traders”).

98 A broker serving retail customers may also operate an affiliated alternative trading  system to which the retail orders may be routed.

99 See infra Section X.A for additional detail on the composition of this category of investors.

32

range of ways that they may use algorithms to (1) decide what to trade, (2) manage trade  execution and (3) assess trading performance.

A number of institutional entities, such as mutual funds, track indices, and others may invest in products, such as ETFs, that do so, and, as a result, they generally trade in  response to market movements and other factors with the objective of keeping their  underlying investments in line with a benchmark index.100 Similarly, some institutions  may have targeted or fixed-weight proportions of equities within their investment  portfolios, and so trade into or out of positions when market changes cause a portfolio to  deviate too far from this target. For these institutions and investments, the decision  algorithms built into their products or strategies affect which instruments to trade and  when.

Index-oriented trading is not the only form of algorithmic trading that is driven by market  movements. For example, some systematic institutional equity trading is algorithmically  connected to other asset classes or indicators. This type of linkage is present in, for  example, index option delta and gamma101 hedging strategies. These strategies generally drive increased selling in equities markets when measures of volatility (such as the VIX)  increase. Systematic volatility targeting and risk-parity strategies may similarly adjust  their portfolio holdings depending on movements in some measure of volatility.102 Volatility-oriented strategies may have a momentum effect on stock prices: because  volatility often rises with declining prices, strategies that drive increased selling of equities

100 Such trading is pronounced and apparent on days when commonly-used benchmark  indices are rebalanced.

101 In options trading, “delta” measures the amount the cost of an option is expected to  change given a change in the cost of the underlying asset. Delta is a proportion between 0  and 1, or 0 and -1, depending on whether the option is a long or short put or call. For  example, the cost of an option with a delta of .50 would be expected to move $0.50 for  every $1 price change in the underlying stock. “Gamma” is an estimate of how much an  option’s delta is expected to change given a change in the cost of the underlying asset.  Gamma is a proportion between 0 and 1 for long options, or, for short options, 0 and -1. To  extend the previous example, if an option has a delta of .50 and a gamma of .15, with a one  dollar increase in the cost of the underlying stock, the cost of the option will increase by  $0.50, and the option’s delta will increase from .50 to .65. For a more thorough summary  introduction, see, e.g., the “Advanced Concepts” section at

https://www.optionseducation.org/.

102 Generally speaking, volatility targeting strategies increase or decrease leverage in a  portfolio as volatility moves above or below a target level. In risk-parity strategies, a  portfolio is determined by the proportion of risk contributed to the portfolio by each asset,  rather than by the proportion of capital allocated to each asset.

33

as volatility rises may, at least in the near term, further contribute to or exacerbate price  declines.103

Institutional investors use a variety of approaches when executing trades. Some firms have  their own trader or traders who implement investment decisions made by portfolio  managers. At long-only mutual funds, for example, institutional traders typically utilize  algorithms provided to them by brokers, although some create their own algorithms that  determine when, where and how to execute an order, and then use brokers to execute  orders in the marketplace. Other institutional firms may not have a dedicated trading staff  but may employ professionals who create, analyze and execute algorithmic trading models.  At firms where trading is highly automated, trading staff may perform a more monitor-like  function, ensuring that systems are operating properly, and that trading is occurring as intended and within risk limits. Some institutional investors may design and operate their  own trading algorithms, while others may purchase firm-specific algorithmic trading  services from third parties. The technical expertise, infrastructure, and resources required  to design and manage algorithmic trading systems directly may be outside the abilities of  many institutional investors, or they may prefer to outsource trade execution (as well as  other aspects of their investment strategy) to other market participants. Notably, some  technology providers are beginning to offer predictive analytics products that operate on  real-time market data and incorporate machine learning. These types of products  potentially could provide a more generally available tool that is analogous to and  competitive with the low-latency data access, processing, and execution tools used by some  of the fastest market participants.104

Institutions that do not create their own algorithms generally use algorithms provided to  them by institutional brokers. Over the past decade, the “manual handling of institutional  orders is increasingly rare, and has been replaced by sophisticated institutional order  execution algorithms and smart order routing systems.”105 Institutional firms may send a  single large “parent” order to a broker that will generally divide it into many smaller “child”  orders to be executed in the market. Institutions may also send several larger orders to  multiple brokers for similar treatment. In some cases, institutional investors may also send  orders directly to specific broker algorithms or suites of algorithms.106 An institutional  firm may have a core group of brokers used in most securities or market conditions, but

103 See, e.g., Campbell R. Harvey et al., The Impact of Volatility Targeting, J. Portfolio Mgmt,  14, 30-31 (Vol. 45, Fall 2018).

104 See, e.g., the “Signum” product offered by Exegy (https://www.exegy-signum.com).

105 Disclosure of Order Handling Information, Exch. Act Rel. No. 78309, 81 Fed. Reg. 49432,  49436 (July 27, 2016) (“Order Handling Proposing Release”).

106 Institutional broker algorithms are more fully described below.

34

generally will be able to call on a larger pool of specialized brokers. However, use of broker  algorithms appears to be highly concentrated; for example, one recent study estimates that  approximately two-thirds of institutional order flow is sent through the three largest  broker algorithm suites.107

Institutional firms are focused on costs of trading, with such costs being characterized as “explicit” and “implicit.”108 Because institutions generally trade in substantial size, both  explicit transaction costs—e.g., commissions paid to brokers—and implicit costs—e.g.,  information leakage and resulting adverse market impact—can be significant. However,  the broker routing and execution process can often be opaque to institutional investors.109 For example, an institutional investor may not know the number or identity of venues to  which its orders have been routed, whether and how extensively a broker employs  actionable indications of interest, or whether there are compensation arrangements  between brokers and market centers that may affect broker routing decisions. Institutional  investors generally conduct, either on their own or through a third-party provider,  transaction cost analysis in order to assess the quality of executions received from different  brokers, different algorithms, or in different market centers. The availability of data is  central to this cost assessment process, as is the capacity to effectively analyze it and incorporate the results into future execution decisions. As described in more detail below,  the Commission recently took steps to improve the scope and consistency of data available  to investors about broker order handling.110

Increasingly, institutional firms route to brokers and assess their performance using a tool  called an “algo wheel.”111 An algo wheel, which can be operated by an investor or provided  by a third-party, connects investors into multiple broker algorithm offerings, and chooses  brokers and individual algorithms based on specified constraints or preferences. An algo  wheel allows a firm to closely track the performance of broker algorithms under different  market conditions, and can enable a firm to switch between different brokers without input  from a human trader. One recent study estimates that about a quarter of institutional “buy-

107 Richard Johnson, Trends in Global Equity Electronic Execution, 7, Greenwich Associates  (Apr. 23, 2019).

108 See, e.g., 81 Fed. Reg. at 49436 (“Institutional customers have long focused on the  execution quality of their large orders, and the potential impacts from information leakage  and conflicts of interest faced by their broker-dealers”).

109 Id.

110 See Disclosure of Order Handling Information, 83 Fed. Reg. 58338 (Nov. 19, 2018)  (“Order Handling Adopting Release”). See also Order Handling Proposing Release for  additional background on institutional order routing practices.

111 See, e.g., Johnson, Trends in Global Equity Execution, at 5-7.

35

side” investors use algo-wheels,112 while another recent study estimates about 14% of such  investors use algo wheels.113 Much of the performance measurement built into algo wheels  is associated with transaction cost analysis, but algo wheels can also inform the evaluation  of best execution, since they can facilitate assessments of which brokers route most effectively under different circumstances.114

B. Brokers

Brokers are tasked by their customers with finding liquidity in a complex, fragmented  market, achieving best execution, and minimizing information leakage and other implicit  costs. To try to meet these goals, brokers use, and offer to their customers, a wide range of  execution algorithms.

While brokers tend to offer a large suite of algorithms, many of the core types of algorithms  are more or less similar across brokers. For example, many brokers offer algorithms that provide their customers with a volume-weighted average price (VWAP), a time-weighted  average price (TWAP), a minimum implementation shortfall (i.e., minimizing the total costs  of trading relative to the market price at the time a trading decision is made), or trading at  a specified percentage of market volume (PVOL or POV). Broker algorithms may seek to  take liquidity resting at trading venues, to provide liquidity at venues, or some combination  of the two.115 Broker algorithms take into consideration the diversity of venues available,  including exchanges, ATSs, single dealer platforms, and central risk books. Moreover,  algorithms account for the increasingly wide and complex range of order types available at  venues such as exchanges and ATSs, which may have different effects on how orders are  handled.

Brokers generally allow some degree of customization for their algorithms to suit customer  needs. However, at some firms this process can be highly manual, and so may be available

112 Id. at 6.

113 Campbell Peters, “Technology and the Buy-Side Liquidity Chase”, TabbFORUM (July 19,  2019).

114 Id. at 7; see also Larry Tabb, Algo Wheels: Best Execution, Workflow Solution, or Both?,  TabbFORUM (Oct. 15, 2019).

115 See, e.g, Order Handling Proposing Release at 49436.

36

only to large customers.116 Improvements in software used in the customization process  may make precise customization more widely available.117

In searching for liquidity, many brokers turn initially to off-exchange venues. In some  cases, this may include in-house sources of liquidity, such as alternative trading systems,  single-dealer platforms, or central risk books operated by the broker or an affiliate.118  Some studies have raised questions, however, about whether execution quality may suffer  when brokers prefer their own or an affiliated ATS.119

An increasingly common tool used across multiple dark pools, particularly for large block  size orders, is the so-called “conditional order.” Conditional orders allow investors to algorithmically search for liquidity in multiple venues simultaneously, but require the user  of a conditional order to affirmatively execute the order or begin negotiations for a trade  when a response is received.120 This additional step can allow a search for liquidity to  minimize risks of being executed in a size or on other terms that are different than  anticipated or desired, but also may create risks of information leakage.121

C. Principal Trading

Equities markets have long included firms trading their own principal acting as market  intermediaries, transferring risk between other market participants and attempting to  profit directly from this intermediation. On exchanges, historically such participants  included, for example, specialists, registered market makers, and floor traders;122 off

116 Larry Tabb, Fragmentation vs. Liquidity: Can Technology Resolve the Debate?,  TabbFORUM (Aug. 4, 2019), available at: https://tabbforum.com/opinions/fragmentation vs-liquidity-can-technology-resolve-the-debate/.

117 Id.

118 Single dealer platforms and central risk books are described more fully below.

119 Amber Anand, Mehrdad Samadi, Jonathan Sokobin, & Kumar Ventkataraman, FINRA  Office of the Chief Economist, Institutional Order Handling and Broker-Affiliated Trading  Venues (Feb. 22, 2019), available at:

https://www.finra.org/sites/default/files/OCE_WP_jan2019.pdf.

120 Campbell Peters, Conditional Orders: The Great Liquidity Aggregator, TabbFORUM (May  30, 2019).

121 Tabb, “Fragmentation vs. Liquidity: Can Technology Resolve the Debate?”

122 See, e.g., Stock Exchange Practices, Senate Report No. 1455, 73rd Congress 2d Session  (June 16, 1934).

37

exchanges, OTC market making and block positioning were often performed by large investment banks and the broker-dealer affiliates of large banks.123 In modern electronic  equity markets, much of this activity is handled by a range of technologically and  quantitatively sophisticated firms trading as principal, who rely on algorithms in many  aspects of their business in order to trade competitively.

1. High Frequency Trading

For more than ten years, most activity in the U.S. equity markets has been conducted by  professional traders using short-term strategies that place a high number of orders, and  generate a large number of trades, on a daily basis.124 This activity has commonly been

called high frequency trading, or “HFT,” though there is no statutory or regulatory  definition of this term.125 The 2010 Concept Release noted five characteristics typical of  principal trading firms engaged in HFT:

1. The use of extraordinarily high-speed and sophisticated computer programs for  generating, routing, and executing orders;

2. Use of co-location services and individual data feeds offered by exchanges and others  to minimize network and other types of latencies;

3. Very short time-frames for establishing and liquidating positions; 4. The submission of numerous orders that are cancelled shortly after submission; and 5. Ending the trading day in as close to a flat position as possible (that is, not carrying  significant, unhedged positions overnight).126

Not all of these characteristics must be present for a firm to properly be described as  engaging in HFT.127 Broadly speaking, it is essential for firms engaged in these strategies to  have the information technology infrastructure and computational sophistication to  quickly and accurately process massive volumes of data from a wide range of sources,  implement trading and risk decisions based on that data, and quickly enter orders based on  those decisions before identified trading opportunities pass. To engage in high frequency  trading, nearly all aspects of trading must be implemented algorithmically.

123 See, e.g. Larry Tabb, The Future of Liquidity: Risk Transformation, TabbFORUM (July 22,  2019).

124 See, e.g., Concept Release at 3606.

125 Traders using these strategies may also be organized in a variety of ways, including as  market-making desks within multi-service broker-dealers or as hedge funds. Id.

126 Id.

127 See HFT Literature Review at 4.

38

High frequency trading is not a singular, monolithic type of activity.128 It includes a range  of strategies, which, as described more fully below, may have different effects on market  quality, particularly under different types of market conditions. The distinction between  strategies that primarily provide liquidity and those that primarily demand liquidity may

be particularly important when analyzing potential effects on market quality.129 The 2010  Equity Market Structure Concept Release described four broad types of short-term high  frequency trading strategies: passive market-making, arbitrage, structural, and  directional.130

a. Passive Market-Making

Passive market-making involves submitting non-marketable orders on both sides (buy or  “bid,” and sell or “offer”) of the marketplace. Profits are earned primarily by earning the  spread between bids and offers, supplemented by liquidity rebates offered by many  exchanges for offering resting liquidity.131 Passive market makers may trade aggressively,  sometimes rapidly demanding liquidity, in order to quickly liquidate positions accumulated  through providing liquidity. Passive orders are generally not executed immediately and  rest on an order book, and so must be updated as conditions change. Passive market makers are vulnerable to “adverse selection,” or prices moving quickly in one direction  against their bids or offers, which can make it difficult to profitably trade out of a  position.132 As part of managing this risk, these strategies can produce enormous volumes  of modification and cancellation messages.

128 See, e.g. HFT Literature Review at 9.

129 See, e.g. Donald MacKenzie, ‘Making’, ‘taking’ and the material political economy of  algorithmic trading, Economy and Society, 47:4, 501-23 (2018) at 502 (describing the  distinction between liquidity providing and liquidity demanding algorithms as “the single  most important divide within HFT” (emphasis in original)); id. at 511 (noting that while  HFT strategies at times necessarily blend passive and aggressive activity, it is more  common to predominantly specialize in one or the other); HFT Literature Review at 9-10.

130 See Concept Release at 3607-10. The summary here largely follows the more detailed  discussion in the Concept Release.

131 In order to trade as a market maker, the market participant must be able to consistently  move their orders to the order book in each venue in which it trades. This effort requires  fast, high-quality market data, as well as technology capable of quickly processing it. Many  exchanges also offer order types that may assist resting orders in achieving or maintaining  queue priority as conditions change.

132 Some exchanges, such as IEX and NYSE American, offer order types that will  automatically reprice certain resting orders based on algorithmically-generated  predictions of when market-wide prices may be moving against the order. For a more

39

b. Arbitrage

Arbitrage strategies generally seek to capture pricing discrepancies between related  products or markets, such as between an ETF and its underlying basket of stocks, or  between futures contracts on the S&P 500 index and ETFs on that index.133 Arbitrage  strategies are likely to demand liquidity and involve substantial hedging of positions across  products and markets. These strategies do not depend on directional price moves in a  single product, but on the divergence and convergence of prices between products.134

c. Structural

Structural strategies attempt to exploit structural vulnerabilities in the market or in certain  market participants. For example, traders with the lowest-latency market data and  processing tools may be able to profit by trading with market participants who receive and  process data more slowly and, as a result, have not yet updated their prices to reflect the  most recent events.135

d. Directional

Directional strategies generally involve establishing a short-term long or short position in  anticipation of a price move up or down. These strategies generally require demanding  liquidity to build such a position.136 Some directional strategies may focus on predicting  price movements faster than other market participants, which requires sophisticated  analytics and rapid processing abilities. For example, order anticipation strategies may  attempt to predict or infer the existence of a large buyer or seller in the market, in order to

general discussion of adverse selection and its role in market-making and quote-setting, see Merritt B. Fox, Lawrence R. Glosten, & Gabriel V. Rauterberg, The New Stock Market: Law,  Economics, and Policy, ch. 3 (2019) (“The New Stock Market”).

133 Some ETF market-makers also act as Authorized Participants in ETFs, though the roles  are distinct. Managing the ETF create-redeem process requires technological  sophistication additional to that required for market-making.

134 See HFT Literature Review at 8.

135 See, e.g., MacKenzie supra note 129 at 512-13 (noting that the response times for this  type of strategy, as reported in interviews with high frequency traders, has declined from  about 5 microseconds in 2011 to 300 nanoseconds in 2018).

136 See, e.g., id. at 512.

40

buy or sell ahead of the large order.137 Trading on such predictions may often contribute to  the process of price discovery in a stock.138

2. Single Dealer Platforms

On single dealer platforms, an individual dealer stands ready to trade with other market  participants, generally offering some mode of indicative pricing.139 Some single dealer  platforms may execute block trades, and others may focus on smaller trades. Some single  dealer platforms account for meaningful percentages of consolidated volume. For example,  recent estimates indicate that some single dealer platforms may account for as much as 1%  of consolidated average daily share volume.140

Like other participants in equities markets, single dealer platforms must rapidly process  large volumes of market data in order to make trading and risk decisions, as well as  effectively handle the routing and communications necessary to managing a large number  of electronic orders.

3. Central Risk Books

Central risk books have become important sources of block liquidity for many market  participants.141 Central risk books, generally offered as a type of principal trading by large,

137 As the Concept Release notes, such order anticipation is an old strategy to which  investors have long been vulnerable. See Concept Release at 3609. This vulnerability  makes information leakage a central concern for institutional investors.

138 The Equity Market Structure Concept Release also noted that another type of directional  strategy, momentum ignition, “may raise concerns.” Id. Momentum ignition strategies  involve initiating a series of orders and trades in order to attempt to ignite a rapid price  move up or down. As the Commission noted in the Concept Release, any market  participant that manipulates the market has engaged in prohibited conduct. See id.

139 In a recent request for comment on a proposed disclosure rule, FINRA proposed a  definition of “single dealer platform”: “an electronic trading platform owned and operated  by a member on which the member trades solely for its own account when executing  orders routed to the [single dealer platform] and represents either the buy or sell side of  each trade on a proprietary basis.” See FINRA Regulatory Notice 18-28.

140 See Rosenblatt Securities, Let There Be Light: Rosenblatt’s Monthly US Dark Liquidity  Tracker (Dec. 18, 2019).

141 See, e.g., Campbell Peters, Technology and the Buy-side Liquidity Chase, TabbFORUM (July  19, 2019); Valerie Bogard, Justin Schack, & Anish Puaar, Central Risk Books: What the Buy  Side Needs to Know, Rosenblatt Securities Trading Talk (Oct. 11, 2018) (“Rosenblatt CRBs”).

41

global firms, aggregate firm-wide risk across desks, regions, and asset classes, seeking to  maximize firms’ capital while staying within overall risk limits.142 This process can reduce  hedging costs and optimize resources across a firm. This data aggregation and analysis  requires significant quantitative and technological sophistication, including the ability to  reconcile the cross-market risk profiles of different instruments with potentially very  different types of market and product data.143 This process can allow central risk books to  provide liquidity for large orders, often with favorable pricing, and generally depends upon  having sufficient capital to take on positions and hedge risk.144

Central risk book liquidity can be accessed through a variety of channels, including through broker trading desks and algorithms, and may be reflected in ATSs.145 Some firms generate  streaming indications of interest, including actionable indications of interest, available  through information services such as Bloomberg. One survey notes that the liquidity  offered by central risk books is “most likely smaller blocks of blue-chip stocks, rather than  large blocks of small- or mid-cap stocks.”146

D. Operational Risks to Firms and the Market

The electronic, automated, and interconnected nature of modern equity markets has  created operational risks for both individual firms and the markets as a whole. As  illustrated by the types of events described below, operational failures can have  detrimental effects throughout the market system. As multiple regulators have now  emphasized, it is essential for a range of market participants to have in place policies,  procedures, and practices to ensure the robust operation and resilience of technological  systems.147

142 Rosenblatt CRBs at 1. Firms other than large, global banks may also offer central risk  books or employ central risk management structures. See, e.g., Larry Tabb, The Central Risk  Book: Rethink Risk, Rethink Trading (Dec. 5, 2017).

143 Rosenblatt CRBs at 2; Tabb, The Central Risk Book.

144 Rosenblatt CRBs at 2.

145 Id. at 3-4.

146 Peters, Technology and the Buy-Side Liquidity Chase, supra note 141.

147 See, e.g., Risk Management Controls for Brokers or Dealers With Market Access, Exch. Act  Rel. No. 63241, 75 Fed. Reg. 69791 (Nov. 15, 2010) (“Market Access Rule Adopting  Release”); Regulation Systems Compliance and Integrity, Exch. Act Rel. No. 73639, 79 Fed.  Reg. 72252 (Dec. 5, 2014) (“SCI Adopting Release”); FINRA Regulatory Notice 15-09  (Mar. 2015).

42

Errors from improper technology development, testing, and implementation at individual  firms can have severe effects on those firms. For example, in 2012, a broker-dealer  experienced a significant error in its equity order routing system following a systems  update, erroneously sending millions of orders into the market over a forty-five minute  period, and ultimately costing the firm more than $460 million in losses.148 Similarly,  improper controls at individual firms can negatively impact markets. For example, between 2011 and 2013, a firm improperly allowed essentially anonymous non-U.S.  traders to enter billions of orders into U.S. markets, and did so without implementing risk  management controls reasonably designed to ensure compliance with applicable  regulatory requirements, which also resulted in the firm violating other regulatory  requirements.149 Another firm, as a result of a coding change and series of changes to  routing logic, and a failure to impose adequate post-trade surveillance, between 2010 and  2014 erroneously allowed millions of orders with a notional value of approximately $116  billion to be sent in violation of Rule 611 of Regulation NMS.150

It also is important for algorithmic trading platforms and other core infrastructure systems  to maintain proper controls and data integrity.151 During the last decade, for example,  inadequate policies and procedures and systems errors at exchanges resulted in violations  of the securities laws as well as trading disruptions;152 systems failures have interrupted  initial public offerings;153 capacity failures at one of the equity consolidated data feeds  caused one SIP provider to fail, leading to a trading halt in all securities listed on one

148 See Knight Capital Americas LLC, Exch. Act Rel. No. 70694 (Oct. 16, 2013) (settled  matter).

149 Wedbush Securities Inc., Jeffrey Bell, and Christina Fillhart, Exch. Act Rel. No. 73652 (Nov.  20, 2014) (settled matter).

150 Latour Trading LLC, Exch. Act Rel. No. 76029 (Sept. 30, 2015) (settled matter).

151 For more comprehensive discussions of systems events throughout the securities  markets, see Regulation Systems Compliance and Integrity, Exch. Act Rel. No. 69077, 78 Fed.  Reg. 18084 (Mar. 25, 2013) (“SCI Proposing Release”) at 18089-90; SCI Adopting Release at  72254-56.

152 EDGX Exchange, Inc., EDGA Exchange, Inc., and Direct Edge ECN LLC, Exch. Act Rel. No.  65556 (Oct. 13, 2011) (settled matter); New York Stock Exchange LLC, NYSE Arca, Inc., NYSE  MKT LLC f/k/a NYSE Amex LLC, and Archipelago Securities, L.L.C., Exch. Act Rel. No. 72065  (May 1, 2014) (settled matter).

153 SCI Proposing Release at 18089; The Nasdaq Stock Market, LLC and Nasdaq Execution  Services, LLC, Exch. Act Rel. No. 69655 (May 29, 2013) (settled matter).

43

exchange;154 and opening auctions have been delayed as a result of high volumes and  unusual volatility.155 A previous Commission staff report concluded that the interaction  between automated execution programs and algorithmic trading strategies can quickly  erode liquidity and result in disorderly markets, and that concerns about data integrity,  especially those that involve the publication of trades and quotes to the consolidated tape (SIP), can contribute to pauses or halts in many automated trading systems and in turn lead  to a reduction in general market liquidity.156

E. Studies of Effects on Market Quality and Provision of Liquidity

As illustrated by much of the preceding discussion, algorithms are used in a diverse range  of trading activities and, across the various activities and market participants in the cash  equity market, are virtually ubiquitous. This ubiquity and diversity has, understandably,

meant that studies on “algorithmic trading” are not always focused on the same activity.  For example, much of the literature on algorithmic trading focuses on high frequency  trading, either through proxy measurements of HFT activity, or using datasets that  specifically identify high frequency trading firms or accounts. The methodology used in  any given case for identifying the relevant firms or accounts can shape the results found, as  can decisions about whether to focus on metrics relevant to primarily passive liquidity providing activity, aggressive liquidity-taking activity, or all trading without a distinction  between liquidity providing and taking.157 It is unsurprising that academic studies  generally are narrowly focused, as the amount of data, computing power and sophistication  necessary to engage in broader study are daunting and costly, and relevant data may not be widely available or easily accessible. As a result, Commission staff notes that using a single  or just a few studies as a basis for broad market conclusions entails risk and that it is likely  that greater insight will be provided by viewing academic literature as a whole, recognizing

154 SCI Adopting Release at 72255

155 Staff of the Office of Analytics and Research, Division and Trading and Markets, Equity  Market Volatility on August 24, 2015 (Dec. 2015) (available at:

https://www.sec.gov/marketstructure/research/equity_market_volatility.pdf) (“Aug. 24,  2015 Report”).

156 Findings Regarding the Market Events of May 6, 2010: Report of the Staffs of the CFTC and  SEC to the Joint Advisory Committee on Emerging Regulatory Issues (Sept. 30, 2010),  available at: https://www.sec.gov/news/studies/2010/marketevents-report.pdf (“Flash  Crash Report”), at p. 8; SCI Proposing Release at 18089.

157 For a more fulsome discussion of these methodological issues, see HFT Literature  Review at 4-11.

44

the general and individual limitations of the work, as opposed to viewing studies in  isolation.

The following summary attempts to distill a range of studies from the academic literature,  many of which have focused on high frequency trading. Generally, studies on this type of  algorithmic trading indicate that some dimensions and activities can have positive effects

on market quality and efficiency, while others may impose costs on other market  participants or pose risks during periods of unusual market stress. More detailed  discussions of individual academic studies are available in a later section of this staff  report, as well as in literature reviews previously published by Commission staff.158

Studies have generally concluded that high frequency trading may have improved standard  measures of market quality during normal market periods.159 For example, passive  market-making activity is generally viewed as reducing spreads, through competition to both narrow spreads and achieve queue priority,160 and through the improved risk  management that is possible with automated systems.161 In addition, liquidity-demanding  strategies may help to improve price efficiency. Some studies have concluded that HFT  activity can reduce intraday volatility,162 though results are mixed on this point.163

Some studies have concluded that high frequency trading activity may also contribute to  increased costs for other market participants.164 For example, the ability of HFT algorithms

158 See HFT Literature Review.

159 For example, “primarily passive HFT strategies appear to have beneficial effects on  market quality, such as by reducing spreads and reducing intraday volatility on average”  (See HFT Literature Review at 9) and “aggressive HFT strategies can improve certain  dimensions of price discovery, at least across very short time-frames” (HFT Literature  Review at 10).

160 J. Hasbrouck, High frequency quoting: Short-term volatility in bids and offers, 53 J. Fin.and  Quantitative Analysis 613 (2018); J. Brogaard & C. Garriott, High-Frequency Trading  Competition, 54 J. Fin. and Quantitative Analysis 1469 (2019).

161 T. Hendershott, C.M. Jones & A.J. Menkveld, Does Algorithmic Trading Improve Liquidity?,  66 J. Finance 1 (2011).

162 J. Hasbrouck & G. Saar, Low-Latency Trading, 16 J. Financ. Mark. 646 (2013). 163 HFT Literature Review at 10.

164 Id. at 10-11; A. Shkilko & K. Sokolov, Every Cloud Has a Silver Lining: Fast Trading,  Microwave Connectivity and Trading Costs (Working Paper 2016), available at: https://ssrn.com/abstract=2848562 (arguing that bad weather affecting microwave  communications improves outcomes for other market participants).

45

to achieve queue priority can make it difficult for even marginally slower firms to  successfully provide liquidity.165 Some strategies may also involve trading against stale  orders that have not been updated to incorporate information available to participants  with the fast data processing technology.166 Studies have also found evidence that HFT  firms engage in order anticipation and momentum ignition strategies. 167 HFT firms can  exploit information asymmetries derived from the speed with which they can access and  process trading information as compared to other market participants.

Several studies have concluded that improvements in speed are valuable primarily on a  relative basis, and that they do not necessarily provide more fundamental value.168 Some  have argued that the technological “arms race” may therefore be socially wasteful.169

Various studies conclude that during periods of unusually high volatility or market stress  the use of algorithms may exacerbate price movements. There is evidence that during  periods of market stress, market participants self-impose trading halts or otherwise slow  their activity in order to minimize their market risk.170 The withdrawal of liquidity caused  by such a pause may cause prices to move further and more rapidly than they otherwise  would due to a sudden absence of countervailing trading pressure. There is also evidence  that some algorithmic trading firms aggressively trade into rapid price movements,

165 C. Yao & M. Ye, Why Trading Speed Matters: A Tale of Queue Rationing under Price  Controls, 31 Rev. Fin. Stud. 2157 (2018).

166 E. Budish, P. Cramton &J. Shim, The High-Frequency Trading Arms Race: Frequent batch  Auctions as a Market Design Response, 130 Q. J. of Econ. 1547 (2015); J. Brogaard, T.  Hendershott & R. Riordan, High Frequency Trading and the 2008 Short-Sale Ban, 124 J. Fin. Econ. 22 (2016); M. Aquilina, E. Budish, and P. O’Neill, Quantifying the High-Frequency Trading “Arms Race”: A Simple New Methodology and Estimates, UK Financial Conduct  Authority Occasional Working Paper No. 50 (2020) (available at:

https://www.fca.org.uk/publications/occasional-papers/occasional-paper-no-50- quantifying-high-frequency-trading-arms-race-new-methodology).

167 HFT Literature Review at 10.

168 M. Gai, C. Yao & J. Ye, The Externalities of High Frequency Trading (Working Paper 2013),  available at: https://ssrn.com/abstract=2066839; Budish et al., supra note 167; M. Baron et  al., Risk and Return in High-Frequency Trading, 54 J. Fin. And Quantitative Analysis  993(2019).

169 Budish et al., supra note 166.

170 Flash Crash Report at 6; HFT Literature Review at 34.

46

exacerbating price movements by quickly exhausting available resting liquidity.171 Some  researchers have argued, however, that quantitative investors may ultimately act as shock  absorbers, since quantitative models will be able to signal when prices are so low that  profit potential is worth the risk of trading during periods of extreme stress.172

F. Effects of the COVID-19 Pandemic

Beginning in late February 2020, world markets came under severe stress as a result of the  global COVID-19 pandemic. Volatility, trading volumes, and message traffic increased  significantly above their recent averages, and remained at elevated levels for several  weeks. For example, the VIX, a widely-used measure of market volatility, peaked at an  intraday value of 83.56 on March 16th, which is about five times higher than its average  value for 2019.173

At the same time, many market participants and SROs were forced to alter their  operational, supervisory, and compliance protocols to accommodate their trading and  support personnel working from home or back-up facilities. Despite these challenges, U.S.  equity markets functioned without significant technical, or logistical, disruption.

Research on market activity and the actions of market participants, as well as the role of  algorithmic trading during the initial stages of the pandemic, is ongoing and developing. It  is beyond the scope of this report to provide a comprehensive overview of COVID-19’s  impact on U.S. capital markets. However, the following initial observations may be relevant  to this report.

1. NYSE Floor Closure

The New York Stock Exchange (NYSE) operates a hybrid market model unique to U.S. equity markets. It combines electronic trading with human presence and participation in  the matching process on a physical trading floor. Designated Market Makers (DMMs,  formerly known as “specialists”) have obligations to provide fair and orderly markets in  their designated securities. In addition to DMMs, a number of Floor Brokers maintain

171 See, e.g., Flash Crash Report at 48.

172 Kevin McPartland, Benefits and Future of Quantitative Investing, Greenwich Associates,  pp. 7-8 (May 17, 2018).

173 See also, e.g., SIFMA, COVID-19 Related Market Turmoil Recap: Part I - Equities, ETFs,  Listed Options & Capital Formation, p. 6 (Jun. 2020) (noting a record closing VIX value of  82.69 on Mar. 16, 2020).

47

physical presence on the floor of NYSE.174 While most intraday trading on NYSE happens  electronically, DMMs and Floor Brokers typically play a significant role during the opening  and closing auctions. Specifically, they fulfill three main functions in the NYSE hybrid  market model:

• Provide access to “D Orders,” a unique order type only available through floor  personnel.175 While D Orders are available throughout the trading session, most  executions occur in the closing auction.

• Provide access to NYSE’s “Parity and Priority” structure, which allows orders entered  via DMMs and Floor Brokers to have the same priority on the NYSE book as electronic  orders which arrived earlier.176 As noted above, this structure differentiates NYSE’s  model from the price-time priority available on most of the other exchanges.

• Serve as a source of information for off-floor traders, especially around auctions and  significant market events.

On March 23, 2020, in response to the spread of COVID-19 in the New York metropolitan  area and in the interests of its employees’ safety, NYSE moved to fully-electronic trading  and temporarily closed its main physical trading floor.177 According to NYSE’s filing with  the Commission of March 20, 2020:

Because the Trading Floor facilities will be closed, Floor brokers will not be  able to enter orders on the Trading Floor. As a result, there will not be any  Floor Broker Participants in allocations and there will not be any order types  unique to Floor brokers, such as D Orders. In addition, because DMMs will not  be on the Trading Floor, DMMs will not engage in any manual actions, such as

174 See, e.g., 2020 NYSE Trading Floor Broker Directory, available at:

https://www.nyse.com/publicdocs/nyse/NYSE_Trading_Floor_Broker_Directory.pdf.

175 See, e.g., NYSE, “The Floor Broker’s Modern Trading Tool,” available at:  https://www.nyse.com/article/trading/d-order.

176 See, e.g., NYSE, “Parity & Priority Fact Sheet,” available at:

https://www.nyse.com/publicdocs/nyse/markets/nyse/Parity_and_Priority_Fact_Sheet.pd f.

177 See Notice of Filing and Immediate Effectiveness of Proposed Rule Change To Amend Rules 7.35A, 7.35B, and 7.35C for a Temporary Period, Exch. Act Rel. No. 88,444 (Mar. 20, 2020),  85 Fed. Reg. 17141 (Mar. 26, 2020). During this period, several options exchanges also  closed their trading floors. These closures are not discussed here because options trading  is not the focus of this report.

48

facilitating an Auction manually or publishing pre-opening indications before  a Core Open or Trading Halt Auction.178

On May 26, 2020, NYSE commenced “Phase I” of the reopening process, allowing a number  of Floor Brokers to return to the floor of the exchange.179 It was followed on June 17th by “Phase II,” in which certain DMMs were allowed to be physically present on the floor and  resume their main functions in a limited number of stocks.180

The move by NYSE to fully electronic trading did not result in any significant market  interruption or system-related issues. However, there are differing anecdotal views and  limited research on how the floor closure affected certain metrics of market quality, such as  liquidity and price formation during the opening and closing auctions, as well as effective  spreads and displayed liquidity during the trading day. For example, Brogaard,  Ringgenberg, and Roesch (2020) compared changes in intraday market quality metrics on  NYSE with a control group comprising NASDAQ stocks.181 They show a relatively larger  increase in effective spreads in NYSE stocks as compared to the control group following the  floor closure, as well as more significant degradation in other metrics, such as volatility and  “pricing errors.” The authors subsequently conclude that the NYSE hybrid model benefits  overall intraday trading quality, with most of the benefit concentrated around opening and  closing, when volatility is at its highest.

2. Volatility Controls

A later section of this report describes various measures the Commission and other  regulators have implemented to modulate extreme price swings in individual securities

178 Id. at 17142 (internal citations omitted).

179 See, e.g., Notice of Filing and Immediate Effectiveness of Proposed Rule Change To Extend  the Temporary Period for Specified Commentaries to Rules 7.35, 7.35A, 7.35B, and 7.35C,  Exch. Act Rel. No. 88,933 (May 22, 2020), 85 Fed. Reg. 32059 (May 28, 2020).

180 See, e.g., Notice of Filing and Immediate Effectiveness of Proposed Rule Change To Add, for  a Temporary Period That Begins on June 17, 2020, Commentary .06 to Rule 7.35A;  Commentary .03 to Rule 7.35B; Supplementary Material .20 to Rule 76; and an Amendment to  Supplementary Material .30 to Rule 36 To Support the Partial Return of Designated Market  Makers to the Trading Floor, Exch. Act Rel. No. 89,086 (Jun. 17, 2020), 85 Fed. Reg. 37712  (Jun. 23, 2020).

181 Jonathan Brogaard, Matthew Ringgenberg, & Dominick Rösch, Does Floor Trading  Matter? (Jun. 2020), available at:

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3609007.

49

and equity markets as a whole.182 Their implementation and usefulness have been  extensively tested during the period of market volatility caused by the COVID-19 pandemic.  For example, Market Wide Circuit Breakers (MWCB) were triggered four times during  March 2020. Notably, on March 9th, the MWCB was triggered for the first time since both  the reference index for the MWCB (from the DJIA to S&P500) and the thresholds for  different CB levels were changed in 2012. All four MWCBs were Level 1 circuit breakers,  i.e., triggered by a 7% drop in the index. The MWCBs appeared to have operated as  intended, with generally issue-free pausing and re-opening processes, as well as orderly trading following re-opening of the market. The incidence of individual stock volatility  halts also significantly increased in March. For example, on March 18, 2020, there were  1,475 limit-up limit-down volatility halts in 643 unique symbols, compared to a typical  daily median count of approximately ten halts in approximately seven unique symbols.183

3. Liquidity and Spreads

Periods of heightened volatility normally lead to a degradation in market quality and  increased implicit execution costs for investors. The period of severe market volatility  caused by COVID-19 has resulted in increased effective spread measures and market  impact costs across the board. Mittal, Saraiya, and Berkow (2020)184 compare various  market characteristics during the period of heightened volatility with the period of relative  calm in January 2020. They find that the normalized spread costs during the crisis period  increased by 7.2 times for S&P 500 stocks and 4.1 times for Russel 2000 stocks. They also  find that the realized market impact (in addition to spread costs) of trading a number of  shares equivalent to 2% of the daily volume for an S&P 500 stock during the crisis is  comparable to that of 10% of the daily volume during the “normal” period.

4. General Observations on Initial Months of COVID-19

During the initial months of the COVID-19 pandemic, the fully electronic, complex, and  interconnected U.S. equity markets operated without significant disruption. Notably, this  continuity was accomplished with most brokers, buy-side traders, exchange personnel, and regulators working from home. Even on days when markets were paused due to sharp  drops, re-openings resulted in orderly resumption of trading. Impact cost and spreads  measures have responded, as they always do, to heightened uncertainty, as liquidity  providers re-priced their risk and investors’ demand for liquidity has increased. And while

182 See infra Section VI.B.

183 Source: NYSE TAQ. To find typical daily values, the distribution of limit-up limit-down  halts was calculated from January 1, 2019 through June 30, 2020.

184 Hitesh Mittal, Nigam Saraiya, and Kathryn Berkow, US Equity Liquidity in the COVID-19  Crisis (Mar. 31, 2020), available at: https://bestexresearch.com/wp

content/uploads/2020/04/BestEx-Research-Market-Impact-Analysis-20200331.pdf

50

the markets came under severe stress, with unprecedented volatility and sharp daily  fluctuations, they have proved to be resilient, efficient, and transparent.

V. Benefits and Risks of Algorithmic Trading in Corporate and  Municipal Bonds

In the secondary markets for corporate and municipal debt securities, algorithms have  begun to address a range of long-identified information issues, including the distribution  and gathering of quotations, pricing, and trade matching and execution. These changes  have been accompanied by the growth of liquidity provision by participants other than  traditional dealers and an expansion of portfolio trading and bond exchange-traded  products.

A. Liquidity Search and Trade Execution

In some portions of the debt markets, algorithms are reshaping the problems of finding and  providing liquidity. The most notable developments are the automation of the request-for quote process and streaming quotations directly between counterparties.185 The relative  openness of many RFQ platforms and streaming quotation tools has allowed  technologically-sophisticated non-dealer liquidity providers to move into the corporate  and municipal bond markets.

In the corporate bond market, algorithms are central to the process of automating the RFQ  process.186 An automated RFQ process may look something like the following stylized  example.187 A platform may allow parties to identify or restrict the specific types of  counterparties with whom they may communicate, using factors such as whether a party  underwrote a new issue, has traded recently in a particular security, or has expressed interest in trading a similar bond. A platform may also allow the party posting an RFQ to

185 While not widely adopted across all types of debt securities, limit order books on  platforms such as Brokertec and Nasdaq Fixed Income have become a key locus of trading  in the interdealer market for benchmark, on-the-run U.S. Treasury securities. Some of the  risks associated with electronic central limit order book trading in the U.S. Treasury market  are described in the Treasury Market Report on the events of October 15, 2014. See Treasury Market Report. Beyond Treasury securities, and even outside of the interdealer  market in on-the-run Treasury securities, central limit order books have not seen  widespread adoption in the debt markets. See, e.g., Kevin McPartland, Treasury Traders Shy  Away from Order Books, Greenwich Associates (Jan. 30, 2018).

186 McPartland, Corporate Bond Trading in 2019, at 7.

187 See id.

51

set parameters, including, among other things, limiting pricing responses to a specified  deviation from a pricing estimate, defining the time during which pricing responses must  be received, and identifying a minimum number of pricing responses needed. Then, with  these parameters set, when a party posts an RFQ, it is automatically sent to a dealer or  counterparty list. Counterparties then respond automatically, generally based on their own  pricing logic and algorithms. The platform then automatically confirms the best price,  consistent with parameters set by the party posting the RFQ. Some platforms may allow the party posting to an RFQ to review RFQ results and affirmatively confirm a trade rather than  executing the trade automatically.

In the corporate bond market, $1 million is currently a rough upper bound for trades that  can consistently be executed in an automated manner.188 The use of automated trading  declines above that size, with very little adoption for block trades.

In many bonds, particularly bonds issued in smaller sizes and bonds that have been  outstanding, pricing is a difficult task because each instrument trades relatively  infrequently, and can do so at inconsistent sizes and under different market conditions. A  variety of algorithms are designed to address this issue. Many models use so-called “matrix  pricing” to estimate the price of a particular bond by looking at data for similar bonds, with,  for example, comparable issuers, maturities, coupons, or credit ratings. A number of  platforms now use machine learning algorithms to generate a price or spread for specific  instruments.189

For an increasing number of bonds, market participants now stream to counterparties  continuous prices or quote continuous prices on a platform. Dealers, principal traders, and  customers are able to stream prices. These streams are generally bilateral, allowing a user  to tailor liquidity sources across the market. This data can be supplemented by data from  RFQ platforms, providing seekers of liquidity with an increasingly broader view of the  market.190 Streaming or quoting continuous prices has become significantly more common  in municipal bonds, where the typically small trade sizes may be amenable to electronic  trading.

188 Id. at 7.

189 See, e.g., MarketAxess Research, Composite+: Algorithmic Pricing in the Corporate Bond  Market, available at https://content.marketaxess.com/sites/default/files/2018- 08/MKTX_Composite%2B_whitepaper.pdf.

190 See, e.g., McPartland, How Bilateral Streams for U.S. Treasuries Really Work, And What  They Mean for the Market (discussing use of streaming quotes in U.S. Treasury market);  SIFMA, SIFMA Electronic Bond Trading Report: US Corporate & Municipal Securities, p. 8  (Feb. 2016) (noting growing adoption of streaming price protocols from dealers in  corporate and municipal securities).

52

B. ETF Market Making and Arbitrage

The share of global assets under management in fixed income ETFs has increased  significantly over the past decade.191 A recent estimate is that approximately 1.6% of  global fixed income assets under management are in fixed-income ETFs.192 ETFs have  made it possible for investors to indirectly take positions on cash bond markets by trading  intraday in the more liquid and transparent equity market. ETFs also provide investors  with a generally more efficient means of accessing a diversified exposure to bonds as  compared with directly assembling a bond portfolio.

The growth in ETFs has presented arbitrage opportunities to firms willing to trade  between the ETF market and the underlying bonds. A market maker that is an Authorized  Participant of an ETF can use the create-and-redeem process to manage the risk of taking  on positions in either the equity market (for the ETF) or cash bond market (for the  underlying bonds). This cross-market trading and risk-management activity depends on  the effective and rapid processing of data on potentially hundreds of individual securities.  Like the expansion of RFQ platforms, this arbitrage opportunity has attracted non-dealer  liquidity providers to be active in the corporate and municipal cash bond markets.193 Developments in the technological infrastructure to conduct ETF arbitrage and market making have also facilitated the expansion of portfolio trading, where investors can request  a single price for a list of bonds, as opposed to trading them individually.194

C. Studies of Effects on Market Quality and Provision of Liquidity

Academic research on the effects of algorithmic trading on secondary debt markets is  relatively limited. Lack of available data is an important constraint. Order level data is  usually only available in the most liquid and “electronified” markets, such as on-the-run  Treasuries. 195 Order level data is usually not available in less liquid debt markets.

191 FIMSAC Subcommittee on ETFs and Bond Funds, “Report”, at 6 (Apr. 10, 2019). 192 Id.

193 McPartland, The Challenge of Trading Corporate Bonds Electronically, at 5-6. (“the profit  opportunity presented by fixed-income ETF arbitrage strategies has brought a number of  principal trading firms and some quantitative hedge funds into the corporate bond  market”).

194 Id.

195 Most of the academic research on algorithmic trading in the fixed income markets  comes from studying the on-the-run Treasuries market. Fleming (2016) estimates that  trading in on-the-run securities accounts for roughly 85% of total trading volume across  nominal Treasury securities. The majority of trading in on-the-run Treasuries occurs

53

Although transaction data is available through TRACE, there is no attribute on TRACE disseminated reports that identifies trades executed by algorithms.

Overall, research shows that algorithmic trading is prevalent in already liquid debt markets  (e.g., on-the-run Treasuries). Studies of these markets generally find an overall positive  effect of algorithmic trading on liquidity and price discovery during “normal” times. They  also find that “electronification” lowers trading costs, since less intermediation is required  for transactions to be executed. There is, however, some evidence of algorithmic trading  being associated with increased volatility, but such evidence is not prevalent, and it  generally is present during special market conditions, such as periods of unusually high  volatility.196 However, there are very few studies focusing on algorithmic trading in the  corporate and municipal bond markets.197

The Treasury Market Report on the extraordinary volatility of October 15, 2014  highlighted several areas of risk related to the use of automated trading.198 These risks are  similar to those that others have described with respect to automated trading in equities  markets, including: operational risks from malfunctioning or incorrectly deployed  algorithms; market liquidity risks from abrupt changes in trading strategies; market  integrity risks from acts of manipulation; transmission risks from interconnected markets  with closely related instruments; clearing and settlement risks from firms clearing outside  a central counterparty structure; and risks to the effectiveness of risk management from  the speed at which markets and risk positions can change.

electronically on the BrokerTec platform. Fleming, Mizrach, and Nguyen (2018) analyze  the microstructure of the BrokerTec electronic platform, and report that it accounts for  60% of electronic interdealer trading for each of the on-the-run 2-, 5- and 10-year notes.  See also the discussion of potential limitations of use of academic research, supra Section  IV.E.

196 For a more detailed discussion of studies analyzing electronic markets for U.S. Treasury  securities, see infra Section VII.B.

197 See, e.g., Bank of International Settlement, Electronic trading in fixed income markets,  at 23 (Jan. 2016) (noting that “[e]mpirical works on the impact of AT and particularly HFT  on market quality are numerous, but unfortunately relatively few studies focus specifically  on bond markets due to a lack of data”).

198 See Treasury Market Report Appendix C.

54

VI. Regulatory Responses to Market Complexity, Volatility, and  Instability

Algorithmic trading, including trading that relies on rapid access to and processing of large  amounts of market data, is ubiquitous in our equity markets and is increasingly important  in our debt markets. Algorithmic trading has brought secondary market participants  important benefits such as increased liquidity, cost reductions, and improvements in other  measures of market quality. But advances in technology, and related developments in the  provision of and access to market data, have also contributed to the growth of complexity  in markets, arguably have contributed to episodes of volatility and dislocation, and have  changed (and in some cases increased) the firm-level and market risks stemming from system errors and operational failures.

Over the last decade, the Commission and self-regulatory organizations have taken various steps to address these developments, including the evolving firm-level and market risks.  Many of these steps are outlined below and Commission staff will continue to monitor  these developments and, as may be necessary or appropriate, provide advice and make  recommendations to the Commission, including whether the Commission may or may not  need additional statutory authority to address market developments and emerging risks.

A. Improving Market Transparency

To promote a better understanding of the operation of our algorithm-driven and  increasingly complex equity markets as well as our evolving debt markets, recently the  Commission, SROs, and Commission staff have sought to expand transparency into several  aspects of modern markets with an eye toward various regulatory objectives, including  facilitating further analysis of market efficiency and integrity and fostering competition.199

1. Large Trader Reporting

In 2011, the Commission adopted rules to assist the Commission in identifying and  obtaining trading information on market participants that conduct a substantial amount of

199 A number of these initiatives require the Commission or Commission staff to collect,  store, or access sensitive market and participant data and information. See, e.g., Chairman  Jay Clayton, “Statement on Cybersecurity” (Sept. 20, 2017) (available at: https://www.sec.gov/news/public-statement/statement-clayton-2017-09-20). The  Commission and Commission staff review these various data sets with the perspective that  data should only be collected and accessed to the extent that it is necessary to further the  agency’s mission and that it can reasonably be protected. Id.

55

trading activity in U.S. securities markets.200 This rule improves the Commission’s ability  to identify large market participants, and collect and analyze information on their trading  activity.201 Firms began reporting required information in 2012.

2. Consolidated Audit Trail

In July 2012, the Commission approved Rule 613 of Regulation NMS, which requires the  self-regulatory organizations to submit and implement a national market system plan to  create a consolidated audit trail (CAT) that would allow regulators to efficiently and  accurately track virtually all activity in U.S. equity and options markets.202 The Commission  approved a National Market System Plan for implementing the CAT in November 2016.203 In September 2019, the Commission proposed amendments to the NMS Plan designed to  improve transparency and financial accountability of the development of the CAT.204 In  March 2020, the Commission granted conditional exemptive relief to, among other things,  reduce the amount personally identifiable information in the CAT database.205 Additional  details on Plan implementation and proposed timelines are available on the Plan  website.206

200 See 17 CFR § 240.13h-1, 249.327; Large Trader Reporting, Exch. Act Rel. No. 64976, 76  Fed. Reg. 46959 (Aug. 3, 2011).

201 76 Fed. Reg. at 46961, 46963 (the system of large trader reporting “represents an  important enhancement to the Commission’s capabilities to uniformly identify large  traders and quickly obtain information on their trading activity in a manner that can be  implemented expeditiously by leveraging an existing reporting system”).

202 See 17 CFR § 242.613; Consolidated Audit Trail, Exch. Act Rel. No. 67457, 77 Fed. Reg.  45721 (Aug. 1, 2012).

203 See Order Approving the National Market System Plan Governing the Consolidated Audit  Trail, Exch. Act. Rel. No. 79318, 81 Fed. Reg. 84696 (Nov. 23, 2016).

204 See Proposed Amendments to the National Market System Plan Governing the  Consolidated Audit Trail, Exch. Act Rel. No. 86901, 84 Fed. Reg. 48458 (Sep. 13, 2019).

205 See Order Granting Conditional Exemptive Relief, Pursuant to Section 36 and Rule 608(e)  of the Securities Exchange Act of 1934, from Section 6.4(d)(ii)(C) and Appendix D Sections  4.1.6, 6.2, 8.1.1, 8.2, 9.1, 9.2, 9.4, 10.1, and 10.3 of the National Market System Plan Governing  the Consolidated Audit Trail, Exch. Act Rel. No. 88393 (Mar. 17, 2020).

206 See Consolidated Audit Trail, LLC, https://www.catnmsplan.com/.

56

3. FINRA ATS and OTC Transparency

As discussed above, FINRA makes publicly available statistics on off-exchange equity  executions, both in alternative trading systems and at non-ATS OTC venues.207 FINRA began collecting statistics on volume and number of transactions from ATSs in 2014, and then made the statistics publicly available on an aggregated basis.208 In 2015, FINRA  expanded the scope of publicly disseminated data to include non-ATS equity volume  executed over-the-counter.209 FINRA also expanded its public disclosures to include  information about equity block-size transactions on ATSs210 and in non-ATS over-the counter transactions.211 FINRA has also proposed including in its public disclosures data  on ATS transactions in corporate and agency debt securities.212 While FINRA at one point  charged for professional or vendor access to the data discussed above, this data is now  widely and freely available for public use.213

4. MSRB ATS Trade Indicator

In 2016, the MSRB began requiring a specific indicator on trade reports for trades executed  on ATSs.214 This indicator is included both on trades where an ATS takes a principal  position between buyer and seller, and where an ATS connects a buyer and seller but does  not take a principal or agency position between the parties. The ATS indicator is included  on transaction data disseminated publicly.

207 See, e.g., FINRA Rule 6110(b); see also File No. SR-FINRA-2013-042, 79 Fed. Reg. 4213  (Jan. 17, 2014) (approving FINRA’s collection and public dissemination of ATS statistics);  File No. SR-FINRA-2015-020, 80 Fed. Reg. 61246 (Oct. 9, 2015) (approving expansion to  OTC data generally).

208 See, e.g. FINRA Regulatory Notice 14-07 (Feb. 2014).

209 See, e.g., FINRA Regulatory Notice 15-48 (Nov. 2015).

210 See FINRA Regulatory Notice 16-14 (Apr. 2016).

211 See FINRA Regulatory Notice 18-28 (Sept. 11, 2018).

212 See FINRA Regulatory Notice 19-22 (July 9, 2019).

213 See File No. SR-FINRA-2015-023, 80 Fed. Reg. 39811 (July 10, 2015).

214 See MSRB Regulatory Notice 2015-07 (May 26, 2015); MSRB Rule G-14 RTRS  Procedures(b).

57

5. TRACE for U.S. Treasury Securities

Beginning in July 2017, FINRA began requiring member firms to report to TRACE  transactions executed in U.S. Treasury securities.215 This requirement was, in part, a  response to the unusual market volatility of October 15, 2014, which highlighted, among  other things, the need for official-sector access to data regarding the cash market for  Treasury securities.216 Treasury securities are traded by broker-dealers that are FINRA  members, as well as by market participants who are not registered broker-dealers, such as  commercial banks and principal trading firms.217 To expand the scope of its data collection  in Treasury securities, in 2019 FINRA began requiring certain large ATSs to report to  TRACE the identities of non-FINRA member counterparties.218 Currently, the data  submitted to TRACE is available only to regulators, including the Department of the  Treasury. However, in 2020 FINRA will begin publishing weekly aggregated transaction  information and statistics on U.S. Treasury Securities.219

6. Rule ATS-N

In August 2018, the Commission expanded the disclosure requirements for NMS Stock  ATSs and required ATSs to implement safeguards to protect subscribers’ confidential  trading information.220 On new Form ATS-N, ATSs must disclose key information about  their manner of operations and the ATS-related activities of their broker-dealer operators and affiliates. These disclosures are intended to allow market participants to better

215 See FINRA Regulatory Notice 16-39 (Oct. 2016); SR-FINRA-2016-027, 81 Fed. Reg.  73167 (Oct. 24, 2016) (SEC approval of FINRA rules requiring reporting). The reporting  requirement was effective July 10, 2017.

216 See FINRA Notice 16-39 at 2.

217 Id.

218 See FINRA Regulatory Notice 18-34 (Oct. 4, 2018); File No. SR-FINRA-2018-023, 83 Fed.  Reg. 40601 (Aug. 15, 2018). The requirement was effective April 1, 2019. The requirement  will not include trades between two non-FINRA member firms.

219 See, e.g., Financial Industry Regulatory Authority, Inc.; Order Approving Proposed Rule  Change To Allow FINRA To Publish or Distribute Aggregated Transaction Information and  Statistics on U.S. Treasury Securities, Exch. Act Rel. No. 87837, 84 FR 71986 (Dec. 30, 2019);  see also U.S. Department of the Treasury, Quarterly Refunding Statement (Feb. 5, 2020)  (noting that “the public report of weekly aggregated transactions will provide the most  comprehensive account of how much, in what security types, and in what segments of the  market Treasury securities are traded”).

220 See ATS-N Adopting Release, 83 Fed. Reg. 38768 (Aug. 7, 2018).

58

understand how their orders will interact and be executed inside each ATS, and to help  market participants understand differences, if any, in the treatment in the ATS between  subscribers, on the one hand, and the broker-dealer operator and its affiliates, on the other  hand. The disclosure is also intended to facilitate analysis of potential conflicts of interest more generally as well as risks of information leakage. In addition, the disclosures are  intended to make NMS Stock ATSs more comparable with one another, and to help market  participants compare these venues with other market centers in the national market  system.

7. Disclosure of Order Handling Information

In November 2018, the Commission amended its requirements with respect to order  handling and routing disclosures.221 These amendments enhanced the quarterly public  reports that broker-dealers were already required to publish, by mandating disclosure of,  among other things, payment for order flow arrangements and profit-sharing relationships.  The amendments also require broker-dealers, upon request by a customer who places a  “not held” order,222 to provide a customer with a standardized set of individualized  disclosures about the firm’s handling of the customer’s orders, including average rebates  received from (or fees paid to) trading venues, and information about orders that provided  or removed liquidity.223

8. Staff Reports on Episodes of Extreme Volatility

To facilitate market understanding of the dynamics of complex markets during periods of  extreme volatility, Commission staff, in some cases working alongside the staff of other  financial regulators, have published reports describing and analyzing market events.  Specifically, reports were published following the Flash Crash of May 6, 2010,224 the  unusual volatility in the U.S. Treasury market on October 15, 2014,225 and the equity  market volatility of August 24, 2015.226 These reports discuss in detail the market

221 See 17 CFR § 242.605-606; Order Handling Adopting Release.

222 A not-held order generally gives a broker-dealer price and time discretion in the  handling of that order.

223 See 17 CFR § 242.606(b)(3).

224 See Flash Crash Report.

225 See Treasury Market Report.

226 August 24, 2015 Report; Austin Gerig and Keegan Murphy, The Determinants of ETF  Trading Pauses on August 24th, 2015 (Feb. 2016), available at:

https://www.sec.gov/marketstructure/research/determinants_eft_trading_pauses.pdf.

59

dynamics during these periods of unusual volatility. They also provide insight into the  complexity and data-driven nature of markets as well as some of the limits of regulatory  oversight and analysis as a result of data limitations. Some of these limitations have been  or are expected to be addressed, including when the CAT is more fully operational.

9. Market Structure Statistics and Research

The SEC website publishes market data statistics and research on market structure issues,  and makes available tools for the public to visualize changes in market structure data.227 The statistics available on this website are derived from the Commission’s Market  Information Data Analytics System (MIDAS), which provides Commission staff with market  data comparable to that used by some of the more sophisticated market participants,  including the equity and options SIPs, equity exchange proprietary data, fixed-income data, futures market data, and cryptocurrency data.

B. Mitigating Price Volatility

As algorithmic and electronic trading have become more prevalent in today’s markets,  several notable events and other considerations have lent support to the concern that  algorithmic markets may be increasingly prone to quick, large market moves unrelated to  fundamental economic information about the underlying companies or the broader  economy. To help mitigate the negative effects of algorithmic price swings that may occur  too rapidly for human detection and engagement and may unduly destabilize markets, the  Commission and other regulators have implemented several controls to modulate large,  rapid price moves in individual equity securities and the equity markets more generally.

1. Regulation SHO (Short Selling) Circuit Breaker

In 2010, the Commission approved rules requiring trading centers to have in place policies  and procedures to restrict short selling in NMS stocks when a stock has declined 10% or  more relative to the previous day’s closing price.228 Once this short-sale circuit breaker has  been triggered, for the remainder of the day and the following day, short sale orders may  generally, subject to certain exemptions, not be executed or displayed at a price that is less  than or equal to the current national best bid.229 This rule is intended to prevent short

227 See Market Structure, https://www.sec.gov/marketstructure/. Members of the public  may also email Commission staff about this website and market structure issues at  marketstructure@sec.gov.

228 See 17 CFR § 242.201; Amendments to Regulation SHO, Exch. Act Rel. No. 61595, 75 Fed.  Reg. 11231 (Mar. 10, 2010).

229 17 CFR § 242.201(b).

60

selling, including potentially manipulative or abusive short selling, from driving down  further the price of a security that has already experienced a significant intra-day price  decline, and to facilitate the ability of long sellers to sell first upon such a decline.230

2. Single-Stock Circuit Breakers

One response to the volatility in equity markets on May 6, 2010 was the introduction of a  single-stock circuit breaker pilot program.231 This program was implemented through  three stages of rule filings by the exchanges and FINRA, beginning in June 2010.232 In the  first stage, the Commission approved rules to pause trading during periods of  extraordinary market volatility in stocks included in the S&P 500.233 The second stage  added to the pilot securities in the Russell 1000 index, as well as specified exchange traded  products.234 The third stage added all remaining NMS stocks to the pilot.235 All rights and  warrants were later exempted from the pilot.236 The single-stock circuit breaker pilot  expired at the end of July 31, 2012, and was replaced by the “limit-up, limit-down” plan,  described below.

3. Limit-Up, Limit-Down Plan

To replace the expiring single-stock circuit breaker pilot, in 2012 the SEC approved on a  pilot basis, and in 2013 the SROs implemented, the Plan to Address Extraordinary Market  Volatility, more frequently called the “limit-up, limit-down” plan.237 The Plan has since  been amended eighteen times, and has been made permanent.238 Under the Plan, the SIPs

230 75 Fed. Reg. at 11231.

231 77 Fed. Reg. 33499-500 (June 6, 2012).

232 Id.

233 See Exch. Act Rel. Nos. 62252 (June 10, 2010), 75 Fed. Reg. 34186 (June 16, 2010);  62251 (June 10, 2010), 75 Fed. Reg. 34183 (June 16, 2010).

234 See Exch. Act Rel. Nos. 62884 (Sept. 10, 2010), 75 Fed. Reg. 56618 (Sept. 16, 2010)); and  Securities Exch. Act Rel. No. 62883 (Sept. 10, 2010), 75 Fed. Reg. 56608 (Sept. 16, 2010).

235 See Exch. Act Rel. No. 64735 (June 23, 2011), 76 Fed. Reg. 38243 (June 29, 2011).

236 See, e.g., Exch. Act Rel. No. 65810 (Nov. 23, 2011) 76 Fed. Reg. 74080 (Nov. 30, 2011)  (SR-NYSE-2011-57).

237 See Exch. Act Rel. No. 67091, 77 Fed. Reg. 33498 (June 6, 2012).

238 See, e.g., Exch. Act Rel. No. 85623, 84 Fed. Reg. 16086 (Apr. 4, 2019) (approving the  eighteenth amendment).

61

distributing consolidated data for individual stocks calculate price bands for each stock  above and below a reference price. If the national best bid or national best offer of a stock  equals or falls outside the upper or lower limits of one of these bands, the stock enters a  “limit state.” If a stock remains in a limit state for fifteen seconds, trading in the stock is  paused for five minutes. Trading in the stock then reopens with an auction at the stock’s  primary listing exchange. The Plan is intended to pause trading when rapid price moves  result from, for example, erroneous trades or gaps in liquidity, while not inappropriately  restricting more fundamental price moves.239

Securities are divided into tiers, with each tier having a different threshold for the  applicable price bands. Generally more liquid stocks have tighter price bands, and less liquid stocks have wider price bands. The price bands are also wider for stocks in the more  liquid tier in the minutes leading up to the closing auction, to avoid entering a pause during  the price movement that may accompany the close of each trading day.240

All trading centers are required to establish, maintain, and enforce written policies and  procedures designed to comply with the Plan.

4. Market-Wide Circuit Breakers

Also following the extraordinary volatility on May 6, 2010, the national securities  exchanges and FINRA, in 2012, updated their rules providing for market-wide circuit  breakers in the event of severe, market-wide downturns.241 The market-wide circuit  breaker is intended to pause, and, if needed, halt all trading in the event that the broad  market is declining rapidly.

Generally, if the S&P 500 index declines 7% since the end of the previous day’s close (Level  1), trading in all equity stocks and options is paused for fifteen minutes. If it declines to  13% from the prior day’s close (Level 2), the market pauses again for fifteen minutes. If the  index declines 20% from the prior day’s close (Level 3), then all trading is paused until the  next trading day. After 3:25pm, pauses will not occur at the 7% and 13% level, though a  halt will occur for the remainder of the day if a decline reaches the 20% level.

A market-wide circuit breaker has been triggered four times. On both March 9, 2020 and  March 12, 2020, several minutes after the market opened, the S&P 500 index declined 7%  from the prior trading day’s closing price, triggering fifteen-minute Level 1 halts in all

239 77 Fed. Reg. at 33500.

240 See 84 Fed. Reg. at 16092 (approving proposal to eliminate the doubling of price bands  for Tier 2 stocks at the end of the trading day).

241 See Exch. Act Rel. No. 67090, 77 Fed. Reg. 33531 (June 6, 2012) (approving, on a pilot  basis, the SRO market-wide circuit breaker rules adjusting limits and using the S&P 500  index rather than the Dow Jones Industrial Average).

62

equities and options trading. On March 16, 2020, the Level 1 halt was triggered nearly  immediately after the market opening, when the S&P 500 index rapidly declined more than  7% compared with the prior trading day’s close. On March 18, 2020, the Level 1 was  triggered early in the afternoon, several hours after the opening of regular continuous  trading. In each case, trading resumed in an orderly fashion following the halts. Level 2  and Level 3 market-wide circuit breakers have not been triggered to date.

C. Facilitating Market Stability and Security

Due to the complexity and interconnection of modern markets, algorithmic trading  presents significant operational and related risks to market participants, investors and our  economy more broadly, and the Commission, SROs, and Commission staff have focused on  matters related to risk management, operational controls, resilience, and security.

1. Market Access Rule

In response to operational risks posed by the growth and expansion of algorithmic trading,  and the risks posed by sponsored and direct access specifically, in 2010, the Commission  adopted a rule requiring broker-dealers with direct access, or who provide sponsored market access to others, to adopt a system of risk management controls and supervisory  procedures reasonably designed to manage financial, regulatory, and other risks of that  access.242 These requirements apply to broker-dealers with access to trading directly on  exchanges or ATSs, including broker-dealers providing sponsored or direct access.243 They  also apply to broker-dealer operators of ATSs that provide access to trading on their ATSs  to a person other than a broker-dealer.244

The required financial risk management controls and supervisory procedures must be  reasonably designed to prevent the entry of orders that exceed appropriate pre-set credit  or capital thresholds, or that appear to be erroneous.245 The regulatory risk management  controls and supervisory procedures must also be reasonably designed to prevent the  entry of orders unless there has been compliance with all regulatory requirements that (1)  must be satisfied on a pre-order entry basis, (2) are designed to prevent the entry of orders  that the broker or dealer or customer is restricted from trading, (3) restrict market access

242 See 17 CFR § 240.15c3-5; Market Access Rule Adopting Release.

243 17 CFR §240.15c3-5(a)(1).

244 Id.

245 17 CFR § 240.15c3-5(c)(1).

63

technology and systems to authorized persons, and (4) assure appropriate surveillance  personnel receive immediate post-trade execution reports.246

The risk management controls and supervisory procedures required by the Market Access  Rule must be reviewed for effectiveness at least annually, and the broker-dealer’s chief  executive officer must certify annually that the broker-dealer’s controls and procedures  comply with the requirements.247 FINRA and the Commission’s examination staff inspect  broker-dealer compliance with the Market Access Rule.

2. Regulation SCI

To help manage and mitigate operational risks in the markets,248 the Commission in 2014  adopted Regulation Systems Compliance and Integrity (“Regulation SCI”), which requires  classes of important market participants (“SCI entities”) to implement comprehensive  policies and procedures to help ensure the resilience and robustness of their information  technology systems, and that those systems operate in compliance with the federal  securities laws and applicable (e.g., SRO) rules. Regulation SCI also requires SCI entities to report to the Commission on certain events to facilitate Commission oversight of market  infrastructure.249 Covered entities include most SROs, high-volume ATSs, NMS plan  processors, and certain clearing agencies.

SCI entities must mandate participation by members or participants in scheduled testing of  business continuity and disaster recovery plans, and coordinate with each other on an  industry- or sector-wide basis.250 In addition to requiring notification of certain events to  the Commission, the rules also require SCI entities to provide information about events to  affected members or participants, or, for major events, to all members or participants of

246 17 CFR § 240.15c3-5(c)(2).

247 17 CFR § 240.15c3-5(e).

248 See, e.g., 79 Fed. Reg. at 72253 (“At the same time, these technological advances have  generated an increased risk of operational problems with automated systems, including  failures, disruptions, delays, and intrusions. Given the speed and interconnected nature of  the U.S. securities markets, a seemingly minor systems problem at a single entity can  quickly create losses and liability for market participants, and spread rapidly across the  national market system, potentially creating widespread damage and harm to market  participants, including investors”). Commission staff continues to analyze and assess  changes in market operational and cybersecurity risks, and whether to recommend to the  Commission related regulatory action.

249 See 17 CFR § 242.1000-1007; SCI Adopting Release.

250 17 CFR § 242.1004.

64

the entity.251 SCI entities must also review their systems annually, submit quarterly reports  on material systems changes to the Commission, and maintain appropriate books and  records.252 The Commission’s examination staff inspects SCI entities for compliance with  Regulation SCI, generally on an annual basis.

3. FINRA Guidance on the Supervision of Algorithmic Trading

Recognizing the potential for algorithmic trading strategies to adversely impact market and  firm stability, FINRA in 2015 provided guidance to its broker-dealer members on effective  supervision and control practices for member firms and market participants that use  algorithmic strategies.253 FINRA’s guidance is intended to complement Regulation SCI, and  to emphasize to broker-dealers the importance of robust policies and procedures designed  to protect against some of the risks addressed by Regulation SCI for SCI entities.254

At a general level, FINRA’s guidance suggests that firms: undertake a holistic review of  their trading activity and consider implementing a cross-disciplinary committee to assess  and react to the evolving risks associated with algorithmic strategies; focus efforts on the  development of algorithmic strategies and on how those strategies are tested and  implemented; test algorithmic strategies prior to being put into production; develop their  policies and procedures to include review of trading activity after an algorithmic strategy is  in place or has been changed; and ensure that there is effective communication between  compliance staff and the staff responsible for algorithmic strategy development.255

4. FINRA Registration Requirement for Developers of Algorithms

In 2016, FINRA implemented a rule requiring registration as a Securities Trader by each  associated person who is primarily responsible for the design, development, or significant  modification of an algorithmic trading strategy or the day-to-day supervision or direction

251 17 CFR § 242.1002.

252 17 CFR § 242.1003.

253 See FINRA Regulatory Notice 15-09 (Mar. 2015).

254 Id. at 4.

255 Id. at 5-7.

65

of these activities.256 To register as a Securities Trader, associated persons must pass  qualification exams.257

5. Inspections and Examinations for Controls on Algorithmic Trading

In its 2020 Examination Priorities, the staff of the Commission’s Office of Compliance  Inspections and Examinations (OCIE) stated that it will focus on, among other things,  controls around the use of automated trading algorithms by broker-dealers. 258 Noting that  “[p]oorly designed trading algorithms have the potential to adversely impact market and  broker-dealer stability,” staff stated that OCIE may “examine how broker-dealers supervise  algorithmic trading activities, including the development, testing, implementation,  maintenance, and modification of the computer programs that support their automated  trading activities and controls around access to computer code.” 259

6. Participation in Financial Stability Oversight Council

The Chairman of the Commission is one of the voting members of the Financial Stability  Oversight Council (FSOC).260 FSOC has also taken notice of the important recent  technological and structural changes in financial markets. For example, in its 2019 annual  report, FSOC noted that:

The evolution of financial markets has been driven by technological advances  and regulatory developments. While new technologies have reduced

256 See FINRA Regulatory Notice 16-21 (June 2016); Exch. Act Rel. No. 77551, 81 Fed. Reg.  21914 (Apr. 13, 2016) (Order Approving File No. SR-FINRA-2016-007).

257 FINRA Rule 1220(b)(4) (requiring passage of the Securities Industry Essentials exam  and the Series 57 Securities Trader Representative Exam).

258 See Staff of the Office of Compliance Inspections and Examinations, 2020 Examination  Priorities, p. 17 (Jan. 7, 2020) (“2020 Exam Priorities”) (available at:

https://www.sec.gov/about/offices/ocie/national-examination-program-priorities 2020.pdf).

259 Id.

260 The other voting members of the Council are the Secretary of the Treasury (who chairs  the Council), the Chairman of the Board of Governors of the Federal Reserve System, the  Comptroller of the Currency, the Director of the Consumer Financial Protection Bureau, the  Chairperson of the Federal Deposit Insurance Corporation, the Chairperson of the  Commodity Futures Trading Commission, the Director of the Federal Housing Finance  Agency, the Chairman of the National Credit Union Administration, and a presidentially appointed independent member with insurance expertise.

66

transaction costs and made financial data more widely available, the

increased use of technology and the entry of new types of market

participants have created new types of risks. The increased use of automated  trading systems and the ability to quote and execute transactions at higher  speeds increase the potential for severe market disruptions from operational  events at market makers or other participants. In some markets, economies  of scale associated with new technologies have led to higher concentration  and greater dependency for liquidity on a small number of participants. The  emergence of new trading venues has fragmented trading and required the  implementation of technological solutions to connect markets. The Council  recommends that regulators continue to evaluate structural changes in  financial markets and consider their impact on the efficiency and stability of  the financial system. Regulators should also assess the complex linkages  among markets, examine factors that could cause stress to propagate across  markets, and consider potential ways to mitigate these risks. 261

FSOC is also responsible for, among other things, designating key financial market utilities  as “systemically important” (“SIFMUs”). These utilities perform a variety of functions in the  market, including the clearance and settlement of cash, securities, and derivatives  transactions; many of them are central counterparties and are responsible for clearing a  large majority of trades in their respective markets.262

D. Additional Ongoing and Potential Commission and Staff Actions

In addition to the actions discussed above that are focused on improving transparency,  mitigating volatility and enhancing the stability and security of our trading infrastructure,  on an ongoing basis, Commission staff monitors and assesses market integrity, efficiency, and resiliency. In particular, Commission staff currently is monitoring and assessing changes driven by advances in technology, the virtual ubiquity of algorithmic trading in  listed equities, the increasing reliance on algorithmic trading in debt securities, and the  firm-specific and general dependence on electronic systems as well as the risks created by  these developments. The Commission has taken various actions at the recommendation of  the staff in response to these developments. In addition, as indicated in the Commission’s  published rulemaking agenda, the Commission and staff are contemplating further relevant  measures. Ongoing and potential actions include:

261 Financial Stability Oversight Council, 2019 Annual Report, pp. 5-6 (Dec. 4, 2019)  (available at: https://home.treasury.gov/system/files/261/FSOC2019AnnualReport.pdf).

262 For a list of the designated SIFMUs, as well as the bases for the Council’s designations,  see https://home.treasury.gov/policy-issues/financial-markets-financial-institutions-and fiscal-service/fsoc/designations.

67

• In January 2020, the Commission proposed an order with respect to the governance  of the consolidated equity market data plans,263 and in May 2020, the Commission  approved an order directing the SROs to submit a new National Market System Plan  for consolidated equity market data.264 Certain SROs have petitioned for review of  this order in the D.C. Circuit.

• In February 2020, the Commission proposed rules related to equity market data  infrastructure.265 Commission staff will continue to consider the issues raised in, as  well as public feedback on, these proposals.

• The Limit-Up, Limit-Down Plan provides that the Operating Committee of the Plan  will annually provide the Commission and make publicly available a report  concerning the Plan’s performance during the preceding year, which will include an  update on Plan operations, an analysis of any amendments implemented during the  period covered by the report, and analysis of potential material emerging issues that  may directly impact the operation of the Plan.266 The Division of Trading and  Markets and the Division of Economic and Risk Analysis will continue to analyze this  information and make recommendations as appropriate.

• OCIE staff will examine firms with respect to their controls around automated trading algorithms.267

• OCIE staff will continue to evaluate whether SCI entities have established,  maintained, and enforced SCI policies and procedures as required, and will continue  to perform examinations to review whether SCI entities have taken appropriate  action in response to past examinations.268

• The Division of Trading and Markets is considering recommending that the  Commission propose amendments to the National Market System Plan Governing  the Consolidated Audit Trail regarding data security.269

263 See SIP Governance Proposed Order.

264 See SIP Governance Order.

265 See Market Data Infrastructure Proposal.

266 See Plan to Address Extraordinary Market Volatility, Appendix B.

267 2020 Exam Priorities, at 16-17.

268 Id. at 20.

269 See Unified Agenda of Regulatory and Deregulatory Actions, RIN 3235-AM62 (Fall 2019)  (available at:

68

VII. Summary of Studies on Algorithmic Trading

A. Equities

This section summarizes some of the analysis and conclusions from the academic literature that focuses on the market effects of algorithmic trading and high-frequency trading,  including effects on liquidity, price efficiency, and volatility. This section directly  references some of the relevant academic studies.270 In addition, it also discusses the  analysis and conclusions from regulatory studies as well as the academic literature that  focuses on the market effects of some of the market and regulatory initiatives discussed  above.

A number of the academic studies discussed below examine the effects of algorithmic  trading, which encompasses the activity of a broad set of market participants, including high-frequency traders (“HFTs”).271 Since HFTs, at least historically, have accounted for a  large portion of algorithmic trading activity in the U.S. equity markets, many studies  specifically focus on examining the effects of HFTs. As discussed above, high-frequency  trading is classified as a subcategory of algorithmic trading and generally refers to  professional traders who use extremely fast data access and processing capabilities to  execute short-term strategies. HFTs generally trade frequently intra-daily and avoid  carrying a position overnight.272

The literature on algorithmic trading by HFTs in the secondary markets for U.S. equities is  extensive. Commission staff previously published literature reviews on the related topics

https://www.reginfo.gov/public/do/eAgendaViewRule?pubId=201910&RIN=3235- AM62).

270 The studies discussed in this section use data from both the US and foreign markets  (mainly Canada and Europe) to examine the effects of algorithmic trading and high frequency trading. While studies of foreign markets do not directly examine US markets,  they can provide insight into the effects of algorithmic trading and high-frequency trading  that could be applicable to US markets. Some of the studies of foreign markets discussed  here use detailed data or market structure changes as identification in order to study some  effects of algorithmic trading and high-frequency trading that might be difficult to examine  using the available data on US markets.

271 See supra Section IV.

272 See supra Section IV.C.1.

69

of market fragmentation273 and high-frequency trading.274 In addition to summarizing the  economic literature available at the time, these reviews discuss some of the methodological  issues associated with studying algorithmic trading and HFTs, such as the difficulty of  identifying relevant activity, particularly in publicly available datasets that do not explicitly  flag algorithmic trading or high frequency trading.275 Additionally, many articles survey  the academic literature related to algorithmic trading and HFTs.276

Overall, most academic studies find that algorithmic trading and HFTs have improved  market quality and helped reduce transaction costs.277 There is ample evidence suggesting  that, under normal market conditions, algorithmic trading and HFTs improve liquidity and  price efficiency and reduce short term volatility. However, there is some evidence, mostly  from the Flash Crash, that in certain instances algorithmic trading and HFTs may  exacerbate price movements during periods of high volatility or market stress.

1. Liquidity

The academic literature has provided some important insight into questions associated  with algorithmic trading and high-frequency trading. Although the results are not without  exceptions or limitations, the literature has generally established that algorithmic trading  and high-frequency trading improve liquidity, at least under normal market conditions.  Some academic studies that examine different types of HFTs find that the results vary

273 See Staff of the Division of Trading and Markets, Equity Market Structure Literature  Review Part I: Market Fragmentation (Oct. 7, 2013) (available at

https://www.sec.gov/marketstructure/research/fragmentation-lit-review-100713.pdf). 274 See HFT Literature Review.

275 See, e.g. HFT Literature Review, pp. 4-11.

276 See Biais and Woolley (2011), Chordia, Goyal, Lehmann and Saar (2013), Easley, L´opez  de Prado and O’Hara (2013), Gomber, Arndt, Lutat and Uhle (2011), Goldstein, Kumar and Graves (2014), Jones (2013), Kirilenko and Lo (2013), Biais and Foucault (2014), O’Hara  (2015), Chung and Lee (2016), Menkveld (2016), Davies and Sirri (2018), and Gomber and  Zimmerman (2018).

277 It also should be recognized that, both in discrete market segments and more generally,  sophisticated and well-resourced market participants, including exchanges, dealers and  proprietary trading firms, have data access and computing capabilities that significantly  exceed those of most academics. Further, it also should be recognized that because the  market has been subject to rapid change as a result of technological, regulatory and other  developments, the efficacy of period-to-period comparisons may be limited. See also the  discussion above with respect to potential limitations on the use of academic studies, supra Section IV.E.

70

based on the type of HFT. Most studies that examine HFT market makers find that they  improve liquidity and reduce spreads. Other studies find that HFTs who “pick off” stale  orders can worsen liquidity by increasing “adverse selection” costs.278 The rest of this  section discusses in more detail the different effects that studies believe algorithmic  trading and high-frequency trading have or may have on liquidity, including the effects of  liquidity supply by algorithmic trading and HFTs, the effects of HFTs activities that may  increase adverse selection, the effects of HFTs competition, and the effects of changes in  HFTs speed.

a. Algorithmic Trading and HFT Liquidity Supply

A number of academic studies find that algorithmic trading and high-frequency trading  reduces bid-ask spreads. Most of these studies argue that faster speeds or an improved  ability to monitor the market allow algorithmic traders and HFT liquidity suppliers to  reduce their adverse selection costs, which allows them to quote tighter spreads. For  example, Hendershott, Jones, and Menkveld (2011) examine the introduction of  algorithmic trading on the NYSE and find that it reduces the bid-ask spread, which they  attribute to algorithmic trader price quotes experiencing lower adverse-selection cost.  Additionally, Brogaard, Hagströmer, Nordén, and Riordan (2015) examine an increase in  the speed of HFT market makers and find it reduces their adverse selection costs, which  allows them to quote tighter markets. 279

Other academic studies find that algorithmic trading and HFTs improve liquidity by  smoothing it over time. For example, Hendershott and Riordan (2013) find that  algorithmic traders inter-temporally smooth liquidity by providing liquidity when bid-ask  spreads are wide and taking liquidity when spreads are sufficiently narrow. Carrion  (2013) finds a similar result for HFTs.280

278 In this context, “adverse selection” refers to the ability of HFTs to react faster than other  participants, such as passive market makers, and trade against resting orders that have not  been updated for market movements. One strategy for passive market makers to avoid or  minimize “adverse selection” costs resulting from rapid market movements, would be to  widen their bid-ask spread. The effectiveness of such a strategy would depend on, among  other factors, whether other market makers are willing to quote a narrower spread.

279 See, e.g., Boehmer, Fong, and Wu (2018) , Carrion (2013), Hendershott and Riordan  (2011), Hendershott and Riordan (2013), Korajczyk and Murphy (2019), Malinova, Park  and Riordan (2018), and Riordan and Storkenmaier (2012)

280 See also Hagstroomer, Norden and Zhang (2014), Hendershott and Riordan (2011), and  Jarnecic and Snape (2014).

71

Most academic studies conclude that the tighter quotes supplied by algorithmic trading and  HFTs improve effective spreads and reduce the trading costs of retail investors.281  However, academic studies find mixed results on the impact HFTs have on institutional  trading costs.282 Brogaard et al. (2014b) finds no clear evidence that HFTs impact  institutional execution costs, with institutional traders’ costs remaining unchanged when  HFT activity increases. Korajczyk and Murphy (2019) find evidence that HFTs only  increase the costs of large institutional trades, i.e. above $2 million, and that they decrease  the transaction costs of smaller institutional trades.283 Tong (2015) finds that an increase  in HFT activity causes an increase in the price impact of institutional orders, which  increases their implementation shortfall costs.284

Menkveld (2016) argues that an additional benefit of HFTs is they helped the market as a  whole migrate quickly to electronic trading, which, in turn, yielded lower transaction costs and more volume. He argues that there is a symbiotic relationship between new electronic  venues and HFTs. New venues need HFTs to insert aggressively priced bid and ask quotes,  and HFTs need the new venues to satisfy their requirements in terms of automation, speed,  and low fees.

Although HFT market makers are the primary liquidity suppliers in equity markets, they  usually do not have obligations to provide liquidity. One concern is that this could cause  them to scale back from supplying liquidity when market conditions are uncertain or

281 See, e.g., Conrad, Wahal, and Xiang (2015), Korajczyk and Murphy (2019), Malinova, K.,  A. Park, and R. Riordan (2018), Riordan and Storkenmaier (2012), and Bershova and  Rakhlin (2013).

282 As discussed above, in order to reduce their price impact, large institutional “parent”  orders are divided by algorithms into many smaller “child” orders to be executed in the  market. See supra Section IV.A.2. Beason and Wahal (2019) provide a detailed examination  of the child orders produced by four widely used standardized institutional trading  algorithms.

283 They find the increased price impact caused by HFTs increases the costs of large trades,  but the lower effective spread caused by HFTs lowers the costs of small trades. Van Kervel  and Menkveld (2019) also find that HFTs increase the execution costs of large institutional  trades. Malinova, Park and Riordan (2018) examine the average overall effects on

institutional trades in the same setting as Korajczyk and Murphy (2019) and find that  increased HFT activity causes them to decline.

284 Tong (2015) also finds that an increase in HFT activity causes the effective spread of  institutional orders to decline, but the increase in the price impact dominates the decrease  in the effective spread and causes the implementation shortfall costs of institutional  investors to increase.

72

otherwise unfavorable. Boehmer, Li and Saar (2018) find that the strategies of HFTs are  correlated and Malceniece et al. (2019) find that HFTs cause significant increases in co movement in the returns and liquidity of stocks. This correlation in price movements and  the supply of liquidity could be the result of HFT market makers withdrawing from (and/or  cause HFT market makers to withdraw from) the market at the same time. Anand and  Venkataraman (2016) find evidence that market makers scale back in unison when market  conditions are unfavorable, which contributes to covariation in liquidity supply, both  within and across stocks.285

Market making quoting obligations could improve liquidity in these circumstances. Anand  and Venkataraman (2016) compare HFTs without quoting obligations to designated  market makers (DMMs), i.e. HFTs who have quoting obligations. They find that DMMs  continued to participate at times when the other HFTs scale back, which reduces execution  uncertainty. Clark-Joseph, Ye, and Zi (2017) also look at how DMM obligations on NYSE  affect liquidity and find evidence that is consistent with the idea that DMMs cause  meaningful improvements in liquidity.

b. HFT Activities and Increased Adverse Selection

This subsection discusses certain HFT activities that could increase the adverse selection  costs of some traders, including: “stale quote arbitrage,” “order anticipation,” “quote  stuffing,” and “spoofing.”

Stale Quote Arbitrage

Some HFTs can use their speed advantage to pick off stale limit orders. This can raise the  adverse selection costs of market makers and lead to them quoting wider spreads.  Academic studies have found evidence of HFTs being able to trade against stale quotes. For  example, Brogaard, Hendershott and Riordan (2016) find that HFTs adversely select limit  orders and this affects liquidity negatively.286 Aquilina, Budish, and O’Neill (2020) attempt  to empirically estimate the costs they believe these arbitrage opportunities impose on  investors and other market participants.

Even with their speed advantage, HFT market makers cannot completely avoid being  adversely selected on their limit orders. Menkveld (2013) and Brogaard, Hendershott and Riordan (2014) find that HFT market makers are adversely selected on their quotes.287  Budish, Cramton and Shim (2015) argue that this creates incentives for HFTs to overinvest

285 Anand and Venkataraman (2016) find that HFT liquidity suppliers scale back when  there are large order imbalances. They also find that HFTs earn higher profits and supply  more liquidity during periods of higher stock volatility.

286 See also Brogaard, Hendershott and Riordan (2014) and Van Kervel (2015). 287 See also Biais, DeClerck and Moinas (2016).

73

in speed to be able to react the fastest. This competitive dynamic, which the authors refer  to as a technological “arms race,” may not benefit market participants or market efficiency.

Order Anticipation

A number of academic studies show that HFTs are able to predict the order flow of other  traders. For example, Hirschey (2018) finds that the aggressive buying and selling of HFTs  is correlated with the aggressive buying and selling of non-HFTs in the next 30 seconds. He  interprets this finding as anticipatory trading by HFTs. Similarly, Clark-Joseph (2013) also  suggests that HFTs employ order anticipation strategies in the E-mini S&P 500 futures  market by submitting small aggressive marketable orders and observing the responses.288

Yang and Zhu (2019) provide a model of “back-running,” where strategic traders use order  anticipation strategies based on past order flows to predict the order flow of institutional  investors. Empirical evidence suggests that HFTs are able to back-run on the long-lasting  informed orders of institutional investors, which may increase institutional investors’  transaction costs. For example, Van Kervel and Menkveld (2019) find that HFTs initially  trade in the opposite direction of large institutional orders, but eventually change direction  and take positions in the same direction for the most informed institutional orders, which  increases the execution costs for these orders.289

Quote Stuffing

One harmful market strategy HFTs may engage in is “quote stuffing,” which refers to an  HFT strategy in which a very large number of orders to buy or sell securities are placed in  quick succession and then canceled almost immediately.290 Davies and Sirri (2018) discuss that this type of activity can be used to take advantage of orders that are based on the best  bid and offer and can also impact market integrity by clogging message traffic and delaying  other traders with slower connections from updating or submitting their orders.291

288 See also Breckenfelder (2019) and Raman, Robe and Yadav (2014).

289 Korajczyk and Murphy (2019) find that during larger institutional trade executions,  HFTs submit more same-direction orders. They find this leads to higher transaction costs  for large, informed trades and lower transaction costs for small, uninformed trades.

290 The Concept Release discusses other forms of HFT directional strategies that may  adversely affect some market participants, including order anticipation and momentum  ignition strategies.

291 Quote stuffing can be difficult to detect. Periods of excessive quoting could be a part of a  manipulative trading strategy. Alternatively, Hasbrouck (2018) also discusses that periods  of excessive quoting could also be the result of HFTs competing with each other to  undercut prices. Gai, Yao, and Ye (2013) identify potential quote stuffing in NASDAQ by

74

Egginton, Van Ness, and Van Ness (2016) look at intense episodic spikes in quoting activity  and find that they have a negative impact on market quality, with targeted stocks suffering  decreased liquidity, higher trading costs, and increased short-term volatility.

Spoofing

Another harmful strategy HFTs may engage in is “spoofing,” which involves the submission  and cancellation of buy and sell orders without the intention to trade in order to  manipulate other traders.292 Lee, Eom, and Park (2013) examine spoofing in the Korea  Exchange. They find that investors strategically placed spoofing orders to create the  impression of substantial order book imbalances in order to manipulate subsequent prices.  They find that the stocks targeted for spoofing had higher return volatility, lower market  capitalization, lower price level, and lower managerial transparency.

c. HFT Competition

Academic studies provide mixed evidence about how competition among HFTs affects  liquidity. A number of these studies generally find that more competition between HFTs  seems to decrease spreads and improve price resiliency, i.e., a quicker narrowing of the  spread after it widens. For example, Brogaard and Garriott (2019) examine how the entry  of new HFTs affects competition among HFTs and find that bid–ask spreads decrease and  effective and realized spreads for non-HFTs narrow when new HFTs enter the market.293 In contrast, Yao and Ye (2018) find that competition among HFTs can increase the costs of

examining abnormally high levels of co-movement of message flows for stocks in the same  data channel.

292 It should be noted that to the extent that any trading activity, including order activity, is  manipulative it is subject to legal and regulatory sanction. The Commission has brought  several actions based on alleged “spoofing” and the inspections and enforcement staff of  the Commission review trading activity for possible violations of anti-manipulation laws  and regulations. See, e.g., “SEC Charges Firms Involved in Layering, Manipulation Schemes”  (Mar. 10, 2017) (noting filing of charges against individuals and a securities firm for  engaging in and facilitating layering and other market manipulation) (available at:  https://www.sec.gov/news/pressrelease/2017-63.html); “SEC Charges 18 Traders in $31  Million Stock Manipulation Scheme” (Oct. 16, 2019) (noting an emergency action and asset  freeze against defendants allegedly engaged in a market manipulation scheme creating the  false appearance of trading interest and activity) (available at:

https://www.sec.gov/news/press-release/2019-216).

293 See also Hasbrouck (2018).

75

non-HFTs.294 They find that when the relative tick size, i.e. the minimum tick size  increment divided by the stock price, is larger, HFTs compete more intensely to be the first  one to the front of the limit order book queue in order to supply liquidity. This increases  the difficulty of establishing time priority for non-HFT limit orders and compels them to  submit more market orders as the relative tick size increases, which increases their trading  costs.295

d. HFT Speed

Academic studies provide mixed evidence regarding how an increase in the speed of HFTs  affects liquidity.296 Brogaard, Hagströmer, Nordén, and Riordan (2015) find that HFT  market makers were most likely to take advantage of an optional speed upgrade offered by  an exchange. When they did, it improved liquidity because it allowed them to reduce their  adverse selection costs, which allowed them to quote tighter markets. In contrast, Shkilko  and Sokolov (2016) examine instances of bad weather disrupting microwave trading  networks between Chicago and New York and reducing the speed advantages of certain HFTs who rely on these networks. When this occurs, they find that adverse selection risk  imposed by HFTs falls, which causes trading costs to decline and liquidity to improve.297

More speed heterogeneity among HFTs can lead to intermediation chains that improve  liquidity. Menkveld (2016) discusses how differences in the trading speed and inventory  holding periods of market makers can lead to the creation of intermediation chains, with a  series of financial intermediaries passing along shares between end users. He argues that  intermediation chains can be useful in completing trades between end users, either by  forcing intermediaries to line up in a productive sequence, or by having intermediaries

294 Breckenfelder (2019) also examines HFT competition and finds that when HFTs  compete, their speculative, i.e. directional, trading increases, which causes market liquidity  to deteriorate.

295 They find that a smaller relative tick size makes it easier for non-HFTs to execute their  limit orders.

296 Baron et al. (2019) also investigate how differences in latency effect competition  between HFTs. They find that there are differences in relative speed across HFT firms and  that the fastest firms tend to earn the largest trading revenues and remain in the market  longer. New HFT entrants tend to be slower, underperform, and more likely to exit the  market.

297 Gai, Yao and Ye (2013) examine the impact of two speed upgrades on NASDAQ and find  that the speed improvements led to substantial increases in the number of cancelled orders  but did not lead to improvements in quoted spreads, effective spreads, trading volume or  price efficiency. The authors argue that these results indicate that only relative speed  matters.

76

effectively share the burden of an information asymmetry, or by having faster  intermediaries trading more aggressively, thereby revealing information early and thus  reducing information asymmetry later. Weller (2013) presents direct evidence on  intermediation chains in commodity futures trading, where high-frequency market makers  provide rapid execution to investors and effectively consume inventory risk-bearing  services from slower market makers. Brogaard, Hagstroomer, Nordoen and Riordan (2015) show that after an exchange offered a richer menu of colocation services, HFTs  sorted themselves across the various options (not all bought the fastest service). Bid-ask  spreads declined and depth improved after the event, consistent with intermediation  chains benefiting liquidity.

2. Price Efficiency

Academic papers also examine how algorithmic trading and high frequency trading affect  the price discovery process and price efficiency.298 The majority of these studies find that  algorithmic trading and high-frequency trading improve price efficiency and decrease the  time it takes for prices to incorporate new information. For example, Brogaard,

Hendershott and Riordan (2014) find that, overall, HFTs facilitate price efficiency by  trading in the direction of permanent price changes and in the opposite direction of  transitory pricing errors.299

On the other hand, Weller (2018) finds evidence that algorithmic trading may reduce price efficiency by reducing the incentives for market participants to engage in costly research to  learn more about companies. He finds that increased algorithmic trading leads to greater  price jumps on earnings announcements. This implies increased algorithmic trading  decreases the research market participants conduct to predict company earnings, which  results in more surprise when earnings announcements are released.

Academic papers also generally find that algorithmic trading and high frequency trading cause quotes to contribute more to the price discovery process, as opposed to trades. For  example, Hendershott, Jones, and Menkveld (2011) find that an increase in algorithmic  trading caused quotes to become more informative and contribute more to price discovery  as opposed to trades.300 Brogaard, Hendershott and Riordan (2019) more closely examine

298 Price efficiency refers to the degree to which the price of a security incorporates all  available information about the security. Price discovery refers to the process through  which new information is incorporated into the price of a security.

299 See also Benos, Brugler, Hjalmarsson and Zikes (2017), Boehmer, Fong, and Wu (2018),  Brogaard, Hendershott and Riordan (2019), Carrion (2013), Conrad, Wahal, and Xiang  (2015), Hendershott, Jones, and Menkveld (2011), Hendershott and Riordan (2011), and  Riordan and Storkenmaier (2012).

300 See also Hendershott and Riordan (2011) and Riordan and Storkenmaier (2012).

77

the contribution of HFT and non-HFT limit and market orders to price discovery and find  that the majority of price discovery occurs predominately through limit orders submitted  by HFTs. They also find that the contribution to price discovery from limit orders  decreases and the contribution from market orders increases when volatility increases and  that this change is correlated with and may be due to changes in HFT behavior.

A number of academic papers look at the speed at which HFTs process public information  and how quickly they incorporate it into stock prices. Hu, Pan and Wang (2017) examine a  small group of fee-paying HFTs who receive the Michigan Index of Consumer Sentiment  two seconds before its broader release and find that most of the index-futures price  discovery happens within 0.2 seconds after HFTs had their early peek. Chordia, Green, and  Kottimukkalur (2018) examine trading around macroeconomic announcement surprises  and find that prices respond to the surprises within 5 milliseconds, but that profits from  trading quickly around the announcements are relatively small.

3. Volatility

Drawing connections between algorithmic trading and high-frequency trading and rapid  changes in prices would seem to be a straightforward argument given the immense speed  at which they trade. It would also appear a small step further to conclude that rapid and  significant changes (i.e., increased price volatility) would be driven by algorithmic trading.  However, academic research has cast some doubt on this conclusion. Although some  studies argue otherwise, a number of academic papers study the effects of algorithmic  trading and high-frequency trading on volatility in equity markets and find evidence that, under normal market conditions, they reduce short term volatility. However, there is some  evidence, mostly from the Flash Crash, that in certain instances algorithmic trading and  high-frequency trading may exacerbate price movements during periods of uncertainty or  market stress.

a. Short Term Volatility

Most academic studies find that algorithmic trading and high-frequency trading reduce short term volatility. For example, Brogaard, Hendershott and Riordan (2014) find that  HFTs trade against transitory price movements, which can reduce volatility.301  Additionally, Boehmer, Li and Saar (2018) find evidence that increased competition  between HFT market makers contributes to lower volatility.

However, in contrast to the previous studies, Boehmer, Fong, and Wu (2018) find that  algorithmic trading increases short-term volatility and that the effects are stronger in small

301 See also Groth (2011) Hagströmer and Nordén (2013), Hasbrouck and Saar (2013)  Hendershott and Riordan (2011), and Hendershott and Riordan (2013).

78

stocks.302 The authors note that the increase in volatility cannot be attributed to faster  price discovery or to the penchant of algorithmic traders for entering volatile markets.

Some academic studies also find that algorithmic trading and high-frequency trading  continue to reduce volatility during periods of heightened volatility. For example,  Hendershott and Riordan (2013) find that algorithmic trading contributes to volatility  dampening in turbulent market phases because algorithmic traders do not retreat from or  attenuate trading during these times.303 Brogaard, Carrion, Moyaert, Riordan, Shkilko and Sokolov (2018) also find that HFTs trade against extreme price movements and thus  stabilize prices during periods of heightened volatility. However, as discussed below in the  section on the Flash Crash, there is evidence that in certain periods of market stress HFTs  can exacerbate volatility, including because they may withdraw from the market en masse.

b. The Flash Crash

Generally, it appears algorithmic trading and HFTs improve market quality and reduce  volatility during “normal” market periods. However, it is possible that such strategies and  market participants may impact the market differently during periods of uncertainty and  market stress. One area of concern has been whether high-frequency trading promotes sudden and unexpected price dislocations. Some researchers suggest that the ability of  HFTs to leave the market rapidly has made the markets more fragile and could exacerbate  periods of market stress. Kirilenko and Lo (2013) suggest that, under certain market  conditions, the automated execution of large orders can create a “feedback-loop” or vicious  cycle effects. These could in turn generate systemic destabilizing market events, such as  the May 2010 “Flash Crash.”

The “Flash Crash” occurred on May 6, 2010, when an algorithm rapidly sold 75,000 S&P500  e-mini futures contracts. Major equity indices in both the futures and securities markets  were already down over 4% from their prior-day close by the time the large sell order hit  at 2:30 PM. Indices moved down a further 5-6% in a matter of minutes before rebounding  almost as quickly. The CFTC and SEC (2010) joint report finds that many of the almost  8,000 individual equity securities and ETFs traded that day suffered similar price declines  and reversals within a short period of time, falling 5%, 10% or even 15% before recovering  most, if not all, of their losses. However, some equities experienced even more severe price  moves, both up and down. Over 20,000 trades across more than 300 securities were  executed at prices more than 60% away from their values just moments before.

The CFTC and SEC (2010) joint report attributes the price declines to a sequence of events, including the exhaustion of the liquidity supply by HFTs, traditional buyers, and cross  market arbitrageurs who spread the price pressure to other markets. Eventually a “hot  potato” effect developed where blocks of futures contracts rapidly moved among the same

302 See also Bershova and Rakhlin (2013) and Malceniece et al. (2019). 303 See also Groth (2011).

79

set of HFTs. Equity market prices declined and some algorithms withdrew their orders on  the bid side in the equity market. These same algorithms, along with other algorithms, also drove demand for short positions in futures contracts and similarly withdrew from the  futures market. This withdrawal from the equity market and of short demand in the  futures market created a negative feedback loop that caused the bid side to be exhausted  and virtually fall away so that sell orders were executed at distressed prices during a period of several minutes until a sufficient number of market participants recalibrated their algorithms or otherwise. When a five-second pause was triggered on the CME, prices  began to recover and within minutes they had risen to almost their previous levels.

A number of academic studies examine the “Flash Crash” and the role that algorithmic  trading and HFTs played in it. Most studies are consistent with the CFTC and SEC joint  report and conclude that HFTs did not cause the 2010 Flash Crash, but their withdrawal  from the market may have exacerbated the rapid price declines. For example, Easley,  López de Prado, and O’Hara (2012) attribute the 2010 Flash Crash to the combination of  automated market makers and increased order flow “toxicity,” which combined to cause  market makers to withdraw their quotes and liquidate positions.304

Aldrich, Grundfest, and Laughlin (2017) find that instability of the market data  infrastructure also contributed to the May 2010 Flash Crash. A lag in the data feed caused  stale prices for the SPY ETF to be disseminated to the market. The authors argue this  caused uncertainty among algorithmic traders, and that uncertainty rationally caused them  to withdraw liquidity, which contributed to the price collapses.

4. Regulatory Studies and Speed Bumps

a. Single-Stock Circuit Breaker Pilot and Limit-Up Limit-Down Plan

A number of studies look at the effects of the single-stock circuit breaker pilot and “limit up, limit-down” plan.305 Brogaard and Roshak (2016) find that the introduction of single stock circuit breakers enhances market stability by reducing extreme events; however, this  comes at the cost of reduced price efficiency in the market. Hautsch and Horvath (2019)  examine trading pauses during the single-stock circuit breaker pilot and “limit-up, limit down” plan and find that, on average, trading pauses enhance price discovery during the  break but increase volatility and widen bid-ask spreads after the break. They argue there is  a trade-off between the benefits of trading pauses in terms of their function as safeguards  and protectors from extreme price movements and their adverse effects on volatility, price  stability, and transaction costs.

304 See also Kirilenko, Kyle, Samadi, and Tuzun (2017), McInish, Upson and Wood (2014),  and Menkveld and Yueshen (2018).

305 See supra Section VI.B.2 and Section VI.B.3 for discussions of the single-stock circuit  breaker pilot and “limit-up, limit-down” plan.

80