April 18, 2000

Jonathan G. Katz
U.S. Securities & Exchange Commission
450 Fifth Street, NW
Washington, DC 20549-0609

Re: Release No. 34-42450; File No. SR-NYSE-99-48; Market Fragmentation

Dear Mr. Katz:

I am Robert Wood, Distinguished Professor of Finance at the Fogelman College of Business and Economics, The University of Memphis. I served on the Brady Commission that studied the market crash of October 1987 and on the NASD Economic Advisory Board. I have studied the functioning of securities markets for the past 20 years. I am delighted to have the opportunity to contribute to the discussion of market fragmentation.

The ongoing debate regarding market structure, which has accelerated during the past six years, reflects the incredible pressure brought to bear on legacy market structures by the relentless onslaught of technology. If we were to design a trading system from scratch today clearly it would be entirely electronic in nature. All trading systems that have been created in the past 20 years are electronic based. Yet we have well-established legacy systems that employ massive amounts of capital and many employees to support intermediation with human intervention - dealers and specialists. Hence, there is great resistance to change.

Furthermore, most of our traders have grown up with legacy systems and are expert in using them, causing further resistance to change. Typically, regulatory agencies are captured by the industry they regulate, so the SEC will have many internal pressures to resist change. Finally, the ever-accelerating rate of technological advancement pressures the SEC to make changes with greater frequency and within ever-shorter time frames - a process fraught with danger. This is a difficult time to be a regulator.

Fragmentation and Competition: Fragmentation is a synonym for competition. The key thrust of the NYSE Market 2000 submission was that order flow should be consolidated in one location providing the deepest possible pool of liquidity - at the NYSE for listed stocks. While this sounds appealing, it creates a quasi-monopoly. Although the regional exchanges never were able to provide price competition to the NYSE consistent with the Dominant Firm Model1, prior to ECNs all technological innovation in market structure came from the regional exchanges. For example, the Pacific Coast Scorex system proceed the NYSE DOT system. According to Garvey (1944)2, it has ever been thus. Monopolists resist change. Only by avoiding monopolies or quasi-monopolies, that is by fostering competition, will the SEC create a climate that will capture the efficiencies afforded by technological development.

Evidence of the benefit of fragmentation is provided by McInish and Wood (1996).3 In this study 980 NYSE stocks are grouped into five portfolios that are closely matched for all known determinants of bid/ask spreads, yet where the difference in fragmentation across the regional exchanges was maximized. The paper finds that the greater the level of fragmentation, the tighter the bid/ask spread, while volatility does not increase with fragmentation. This result matches our intuition that competition improves the quality of markets.

Some practitioners argue that it is harder to find liquidity with fragmentation across ECNs. Note that the regionals and the NYSE were always informationally linked, so while order flow was fragmented in one sense it was consolidated for dissemination. The consolidation of information from disparate trading venues is precisely where computers and networks excel. Since there is a demand for that consolidation, it will naturally occur unless regulatory or monopolistic barriers prevent the process.

Markets naturally fragment. Block trades separate from smaller trade sizes. The Paris Bourse, when reforming their market in response to the competitive threat from London's Big Bang in 1986, initially attempted to force block trades through the same trading venue as retail trades. The block business fled to London. Only when separate block trading was permitted did that order flow return to Paris.

Further evidence of the natural fragmentation of order flow is the purchase of retail order flow pioneered by Bernie and Peter Maddoff. As Benston and Wood (1999)4 show, the inventory costs for trades under 499 shares on average are negative. In other words, these trades tend to be highly uninformed. So purchasing this order flow to be traded separately can be viewed as the invisible hand bidding away the excess profits that are created in forcing those orders to be traded in the same venue with institutional orders. We see the revenue generated by the purchase of retail order flow resulting in lower commissions for retail traders. If small orders were forced to trade with institutional orders, retail investors would be subsidizing institutions - not a desirable result.

Finally, evidence of natural fragmentation is reflected in the growth of Posit, to mention only one of a variety of trading venues developed in recent years. Passive traders are able to buy at the spread midpoint in Posit if they can find matching liquidity on the contra side. Further, passive traders can buy at the bid on an ECN or the AZX auction. In other words, order flow is naturally diverse and forcing all orders into the same venue will result in cross-subsidies, which is economically inefficient.

Passive traders, hedge fund traders, value fund traders, growth fund traders, arbitrageurs, VWAP traders, long/short traders, small-cap traders, indexers, retail traders, etc., all have different needs. While all traders need to be informationally linked, forcing them into the same venue, or even demanding that no trade-throughs occur system-wide (i.e. not permitting separate price discovery in various venues), is economically suboptimal. Institutional traders should even be permitted separate venues, if they wish, where they can display size to one another (perhaps anonymously if they choose), as historically occurred in Instinet - this is just another form of block trading, a venue that is already denied to retail traders. Permitting various market mechanisms to compete for order flow will provide the greatest reduction of trading costs.

CLOBs and Competition: As noted above, forcing all trades through a CLOB will likely impair market efficiency by forcing cross-subsidization. As experience has repeatedly taught us, institutions with monopolistic powers end up behaving like monopolists-they avoid innovation and create costly fiefdoms. For example, the number and nature of anti-competitive rules in legacy market mechanisms, such as NYSE rules 3905 and 500, are astounding for an activity that lies at the heart of our free-enterprise system.

If a CLOB is controlled by the NASD, the NYSE or a combination of the two, how could it avoid monopolistic conduct that has characterized their past behavior? If a separate entity is created by a group of sell-side and/or buy-side firms - a Super-NMS - that is given monopolistic powers, how will its structure and behavior different from that of the NASD/NYSE? In my opinion, only the discipline of competition will avoid monopolistic behavior. The crucial aspect of competition in securities markets is not competition for a particular order, but rather competition for order flow. Only when competing market mechanisms are permitted to bid for order flow will we obtain the desired result. Having many bidders on one inefficient, monopolistic mechanism will not produce nearly the reduction in trading cost as having fragmented bidders across highly efficient, competing mechanisms that are informationally linked. Forcing system-wide time/price priority (no trade-throughs) may not be feasible with competing mechanisms6 -- it is certainly not desirable if it produces a monopoly.

It is the competitive structure of market mechanisms that determine the bid/ask spread that retail traders pay7, not how many bidders are competing for a particular order on a given structure.8 An examination of ECNs reveals that they are lean and highly technically competent-very low cost and quick to adopt new technology to meet customer needs. Contrast this effectiveness to that of either legacy mechanism, which provide very high cost systems that perform very poorly. The significant reduction in trading costs that will result from the competition for order flow will result in even further increases in trading volume/liquidity, which in turn will result in ongoing tightening of bid/ask spreads.

Perhaps a role exists for a Super-NMS as proposed by MGM9 to combine and disseminate all trades and quotes from all trading venues such that available liquidity would be readily apparent and immediately accessible. Yet both the NYSE ITS and the Nasdaq Selectnet systems perform the trade- execution function so poorly that competition should prevail in this area as well. With rapidly increasing network capacities and computer power, duplication of broadcast facilities is not onerous. ECNs should be permitted to become exchanges with the opportunity to develop their own trade execution/quote dissemination networks. The market will rapidly supply the software to consolidate all such networks on the trader's desktop. The resulting competition will either force the ITS and SelectNet systems to improve rapidly or force them out of business.

Attempting to negotiate an extension of ITS to embrace all ECNs seems particularly fruitless, given their anti-competitive unanimous vote requirement and its archaic structure by design effectively limits competition. ITS has demonstrated its willingness to go to great lengths to hinder competition, as was demonstrated in their treatment of Optimark.

Market Structures in Transition: The transition from human intermediation to electronic intermediation is particularly difficult for regulators since what is appropriate for legacy systems may not be appropriate for electronic systems. For example, in a competing dealer system each dealer will strive to have the "ax" - the majority of the order flow, thus the majority of the information, thus even more order flow. So they must show size. But showing size in a mixed electronic/dealer system can be dangerous.

When the Nasdaq Small Order Entry System (SOES) was changed from an order indication system to an automatic execution system in June 1988, the mother of all unintended securities regulatory consequences, stale quotes suddenly became very expensive options. Rogue broker/dealers began picking off their compatriots' stale quotes. Benston and Wood (1999)10 show that this change resulted in an increase in trading cost on the Nasdaq market of about $13 billion through 1996. Ironically, this abuse of SOES, a system that was intended to benefit retail customers, resulted in retail trading costs increasing at a rate four times greater than that experienced by institutional orders.

Figure 1 plots the average relative effective spreads (RES) for a large sample of NYSE and Nasdaq stocks from 1987 through September 1999.11 All NYSE RES values are multiplied by 1.9 to scale them up to the Nasdaq level in the period from January 1987 to May 1988, before mandatory SOES.12 This plot shows the dramatic increase in Nasdaq trading costs following the onset of mandatory SOES, and a return to parity with the NYSE in September 1999 following the introduction of the new order handling rules and other changes. But an army of thousands of self-proclaimed "SOES bandits" was spawned in the 1990's that today present, in my opinion, a significant threat to markets. This situation resulted from dealers operating in a hybrid human/electronic intermediation environment (as well as some foolish market rules and regulations).

A member of my current MBA class, a former day trader, described a manipulative trading tactic facilitated by our present hybrid environment. He stated that a technique frequently employed at his former day trading firm was for a group of day traders to attack a stock, usually with a small market cap and perhaps only a few market makers. Ideally the stock would be thinly traded and might be a potential takeover target, such as a regional bank, or high-tech, bio-tech or pharmaceutical stocks where news shocks are common. The day traders would launch a swift buying campaign with (generally) 1000-share trades. Market makers would find themselves going short in a rapidly rising market. As the share price started rising rapidly with strong volume, other traders would be attracted to the fray through chat rooms. Soon those who launched the attack would offer liquidity back at higher prices, and dealers, anxious by this point to cover their shorts in a rising market, would do so.

My experience in recent months on trading desks reveals a trading pattern similar to that experienced with portfolio insurance trading prior to the 1987 market crash - intensive buying/selling that punches through the liquidity barrier provided by the book with such force that natural buyers/sellers step aside. We are seeing these patterns not only in dot-coms, but in members of the S&P500 as well. To the extent that these episodes result from manipulation as described above, regulators should be aggressively pursuing the perpetrators. Markets are built on trust. Such manipulation undermines trust to the detriment of us all.

To the extent that erratic trading behavior occurs from those who have not yet mastered trading in an electronic world, it will eventually be self-limiting. In any event it would seem prudent for the SEC/NYSE/NASDR to carefully monitor the situation.

Many traders steeped in legacy systems are having difficulty adjusting to the new trading venues. This is to be expected - change is daunting for all of us. Yet it is senseless to consider halting progress to prevent heartburn for those who are not keeping up. Some astute traders have adapted to the electronic environment with outstanding results. Instead of buying/selling a block of 500,000 shares in the upstairs market, a trader may post 100 successive 5000-share limit orders to buy at the bid, or perhaps inside the spread. In this manner perhaps even negative trading costs can be obtained - that is, earning revenues from trading by supplying liquidity with limit orders rather than demanding liquidity with market orders. Of course individual traders will make many mistakes in the process of learning to use electronic markets.

Time/Price Priority and Decimalization: Time/price priority is a desirable attribute of a trading system since it encourages the placement of limit orders. With highly efficient, competing ECNs that had their own connectivity (avoiding SelectNet/ITS), such a rule may be possible. System-wide time/price priority seems unfeasible at present given the miserable performance of SelectNet and ITS, particularly at the open or in heavy trading. In any event, the penalty of such a rule should not be a monopoly. But there are other reasons why time/price priority is of decreasing importance discussed below.

We should not fear decimalization. Trade increments of 1/256ths are presently being used on ECNs. Many exchanges around the world have increments of less than one penny. Decimalization will lead to lower trading costs as I show below.

There are dealers and pseudo-dealers. A dealer is a sell-side firm (an intermediary) that buys/sells to/from inventory, hoping to make money on the bid/ask spread (and perhaps commissions). A dealer does not like decimalization and would prefer that the minimum trading increment had remained at 1/8. A pseudo-dealer loves decimalization. Why the difference?

The wider the tick increment the more the dealer tends to make on each trade. For true dealers, it is that simple. A pseudo-dealer is a "natural" - a trader not buying for inventory but rather, for example, rebalancing a portfolio - who wishes to buy at the bid and sell at the ask, thereby earning the spread instead of paying it. Even better, with decimalization the pseudo-dealer can jump the queue for a penny. Dealers hate having their limit orders jumped-in fact they are less inclined to provide limit orders if pseudo-dealers are permitted to jump the queue by a penny. In other words, decimalization incents pseudo-dealers to provide liquidity at the same time it discourages dealers from doing so. But pseudo-dealer liquidity is inherently more valuable to markets -- trading costs are lower when the naturals trade directly without intermediaries. This important but subtle point is expanded in a footnote.13

Note that a pseudo-dealer's view of a limit order is vastly different than that of a true dealer. Risks to a dealer include trading against someone with greater knowledge (adverse selection), or having the market turn against his/her existing inventory position. The pseudo-dealer has no fear of adverse selection, but faces the risk of the market moving away before a limit is executed. The pseudo-dealer's cost structure is much lower than a dealer's - they have no regulatory oversight, little of the support structure required by a dealer, etc. So pseudo-dealers can afford to provide liquidity at a lower cost than dealers.

Should the SEC favor dealers or pseudo-dealers in this argument? In my opinion pseudo-dealers should be favored since doing so reduces execution costs not only by tightening spreads but, of much greater importance, by retaining trading costs to a greater degree on the buy-side, thereby increasing returns to pension and mutual fund holders. With growing network connectivity and computer power, finding the other side of a trade becomes increasingly easier. Hence, intermediaries intuitively ought to be paid relatively less, yet this result only occurs to the extent that naturals are able to supply liquidity to one another. Decimal trading enhances that process. Similarly, if institutional traders are permitted an electronic block-trading venue where they can display size to one another anonymously, where intermediaries may participate as well (but not retail interests or frontrunners), trading costs will be significantly reduced. As noted above, retail traders are already denied access to block trading.

Natural traders steeped in legacy systems who have not yet attempted to master electronic trading will not appreciate the intricacies of these arguments, and will favor wider spreads with greater intermediary volume readily available. But note that as spreads have tightened with tick-size reduction and trading volume has soared, quoted size available at the bid and ask are lower. It is not that the liquidity has been reduced - in fact it has been increased - but it is there in a different form, thereby requiring a shift in trading strategy.

Foreign Competition: We should not fear foreign competition for two reasons. First, competition is the salvation of markets. The Paris Bourse remained essentially unchanged since it was founded under Napoleon until 1986, with the exception of the introduction of electric lights. Everything changed with the Big Bang in London in 1986. Suddenly the market was beyond the control of French politicians and regulators. They had to improve their market or it would die. With several modifications over the years they succeeded in recapturing their order flow. France and the rest of Europe are blessed with vastly improved equity markets and a significantly lower cost of capital (and thus greater job formation, etc.) as a result of foreign competition. We wish that Toronto could mount effective competition to our markets.

Second, order flow naturally returns to home base, baring trading inefficiencies. An interesting example is Daimler-Chrysler, where the same stock (not an ADR) is traded both in New York and Frankfurt. While the majority of the trading started in New York, 95% of the order flow shifted to Frankfurt within a year, where the majority of the information is generated.14

Summary: The history of markets in the U.S. is one of consolidation. Order flow tends to consolidate (within natural groupings as discussed above). In other words, liquidity begets liquidity. Thus, it is very difficult to start a new trading mechanism to compete with existing mechanisms. The fact that a number of new trading venues have successfully started in recent years is indicative of the fact that the legacy systems are very far from where they should be in terms of cost and efficiency of operation. This is understandable, since, while they compete for listings, for many years they enjoyed virtual monopolies for the order flow of the securities listed by themselves. Monopolists resist change.

The SEC is presented with a rare opportunity - that of facilitating a sea change in securities trading to the benefit of our entire society. We are vitally dependent upon the functioning of our securities markets, as a study of the history of civilization reveals. In my opinion the key criteria for the SEC in determining how to proceed is avoiding regulation that protects or enhances the position of legacy systems, but rather to provide a competitive environment wherein the electronic systems can flourish. Certainly there will be a shakeout, but the survivors will produce far better results for our securities markets than the legacy systems.

Summarizing my comments in the SEC framework:

a. Require Greater Disclosure by Market Centers and Brokers Concerning Trade Executions and Order Routing

With respect to the abuses that may be occurring as outlined above, this would be desirable. Although directly investigating suspected abusers may be more efficient.

b. Restrict Internalization and Payment for Order Flow

Restricting payment for order flow is not wise since retail trades would then subsidize institutional trades. The current emphasis on best execution helps with internalization concerns for mainstream firms. But greater reporting should be required - firms should be required to report for each internalized order, the current BBO, the customer, order size and execution size, and all proprietary trades executed in the 30 minute interval prior to the internalized execution.15 These data would facilitate automated detection of abuses perpetrated (for example) by the Long Island "chop shops."

c. Require Exposure of Market Orders to Price Competition

Not if this would stifle competition between trading venues.

d. Adopt an Intermarket Prohibition Against Market Makers Trading Ahead of Previously Displayed and Accessible Investor Limit Orders

This is advisable.

e. Provide Intermarket Time Priority for Limit Orders or Quotations that Improve the NBBO

No, since (1) given the poor performance of SelectNet and IT'S the availability of an order at any instant is not really knowable, and (2) this system-wide rule is likely to stifle inter-venue competition in favor of a monopoly. If ECNs are able to bypass SelectNet/ITS with extremely fast direct linkages, such a rule might be feasible.

f. Establish Price/Time Priority for All Displayed Trading Interest

As noted above, if ECNs are able to bypass SelectNet/ITS with extremely fast direct linkages, such a rule might be feasible. With smaller trading increments, time/price priority becomes less important. However, if institutional buy-side firms wish to establish a trading venue where they could display size to one another privately (and anonymously if they chose), they should be permitted to do so.

Thank you for the opportunity to comment on these important issues.


Robert A. Wood
Distinguished Professor of Finance

1 Garbade, K., and W, Silber. 1979. Dominant and satellite markets: A study of dually traded securities. Review of Economics and Statistics 61:455-60.
2 Garvey, George. 1944. Rivals and interlopers in the history of the New York security market. Journal of Political Economy 52:128-43.
3 McInish, T. and R. Wood, 1996. Competition, fragmentation and market quality, The Industrial Organization and Regulation of the Securities Industry, National Bureau of Economic Research, Andrew Lo, ed.: Chicago, University of Chicago Press.
4 Benston, G. and R. Wood, Day trading on nasdaq's automatic small order execution system (SOES): Adverse selection and spreads, working paper, October 5, 1999.
5 Rule 390 was recently rescinded.
6 In trade-throughs are permitted at present on Nasdaq - theoretically Optimark price discovery could be outside the Nasdaq NBBO.
7 Institutional traders often have sufficient market power to negotiate inside the bid/ask spread. Thus, they are somewhat less concerned about the spread. Yet many institutional trades, particularly those executed in haste, such as index arbitrage and long/short trades, will generally execute at the spread.
8 We see this clearly in options markets where if order flow for a particular option is fragmented between two exchanges spreads are tighter than if all order flow is concentrated at one exchange.
9 MGM is short for Merrill Lynch, Goldman Sachs, Morgan Stanley Dean Witter.
10 Benston, G. and R. Wood, Day trading on Nasdaq's automatic small order execution system (SOES): Adverse selection and spreads, working paper, October 5, 1999.
11 Relative effective spreads are measured as the distance between the spread midpoint and the price of a trade divided by the spread midpoint. See the Benston and Wood paper referenced previously for further details.
12 This scaling does not imply that Nasdaq trading was more expensive than NYSE trading in the 1987-88 period since institutions trade net on Nasdaq and were paying 7-9 cents per share to trade on the NYSE.
13 On the surface, it seems obvious that when naturals, let's say money funds, supply liquidity to one another trading costs are reduced and fund returns increased - if, collectively, funds give less money to intermediaries by supplying more liquidity to one another and those retained funds flow into fund returns, fund returns are increased. But many have difficulty grasping this concept. Perhaps an admittedly cooked example will help.
Suppose that a Nasdaq 100 futures contract will expire in a call auction with 1 million arbitraged shares long and 1,000,100 arb shares short (one share of each stock). The arbs are indifferent to price - they only want the contract closing price. So they have no view of price. To address the imbalance and at the same time set the price, a dealer resolves the imbalance by entering an order to sell the 100 shares at a price that reflects the current market ask. If the auction were repeated on the following day with the imbalance on the other side, a dealer would set the price at the bid. We can see that there is an inherent bid/ask spread in the auction process, but essential all of the spread remains with the naturals. Depending on the relationship of each natural's position to the imbalance, they either earn or pay the bid/ask spread. In repeated samples with random positioning, the naturals are essentially trading with no spread. Extending this logic to the continuous auction market, to the extent that naturals supply liquidity to one another they greatly reduce trading costs and increase pension and mutual fund returns.
14 Although the competing hypothesis, that trading costs are lower in Frankfurt, needs to be examined.
15 The appropriate rule should be based on a thorough study-perhaps the simplest approach is just to report all proprietary trades in a stock for a day where an internalizing trade is executed. While these data are available elsewhere, the intent of this proposal is to create a database where regulators could easily implement an automated problem detection system.


Figure 1: NASDAQ and NYSE Relative Effective Spreads