Blame the Instructions, Not the Machines
Originally published on Blogger
Following the dramatic flash crash on May 6, there has been a lot of attention paid to the mechanics of trading (automation, frequency, scale and speed) but not enough to the kinds of strategies that are being implemented using these mechanisms. Trading algorithms do whatever they are instructed to do, and market movements result from the distribution of instructions and not the technology used to implement them. Technology certainly matters, but in an indirect way. Just as changes in climate can alter the distribution of species in an ecosystem, driving some to extinction and allowing others to proliferate, new technologies can alter the distribution of strategies among the population of traders. Major changes of this kind can affect systemic stability, in the case of markets and ecosystems alike.
The variety of trading strategies in use is vast, but I find it useful to partition them into two broad categories: those that are information augmenting and those that are information extracting. The first group of strategies are based on some form of fundamental analysis: examination of balance sheets, growth potential, and risk, for instance, and trading based on departures of prices from estimated valuations. Such strategies require the investment of resources in information gathering, and end up feeding information to the market. The other class of strategies use market data itself to direct trades. These could be non-directional and arbitrage-based, or directional strategies based on such factors as momentum. This latter class of strategies use volume, price, and other market data as a basis for entering and exiting positions.
A market dominated by information augmenting strategies will tend to be stable and to track information as it arises in the economy. But information extracting strategies can be very profitable in stable markets as long as they react quickly and forcefully to new market data. Changes in technology have made rapid responses to market data feasible on a large scale, resulting in an increase in total market wealth that is invested on the basis of such strategies. The problem is that if too many people are using such strategies, there isn't enough information getting into prices systematically, and certain technical strategies can start generating mutually amplifying responses to noise.
The SEC-CFTC preliminary report on the crash contains a wealth of information and some interesting clues about the kinds of strategies that may have been implicated. First, "approximately 200 securities traded, at their lows, almost 100% below their previous day’s values." These trades, "occurred at extraordinarily low prices – five cents or less – which indicates an execution against a “stub” quote of a market maker." The overwhelming majority of these trades, it turns out, were short sales:
During the period of peak market volatility, 2:45 p.m. to 2:55 p.m., the broken trades executed at five cents or less were primarily short sales. Short sales account for approximately 70.1% of executions against “stub” quotes between 2:45 p.m. and 2:50 p.m., and approximately 90.1% of executions against “stub” quotes between 2:50 p.m. and 2:55 p.m.
In other words, the trades at the most extreme prices were not generated by retail investors whose stop loss orders were converted to market sell orders as prices fell: they were generated by short selling in a falling market.
Also interesting is the case of securities that displayed “aberrant behavior” on the upside:
Sotheby’s (BID) is actively traded and has a narrow bid-ask spread from 2:44 p.m. through 2:49 p.m. after which volume is low but bid and ask quotes remain stable. However, after about 2:57 p.m. volume spikes dramatically and trades are executed at a high (presumably stub) quote of approximately $100,000... BID trades through the national best offer multiple times between 2:57:05 p.m. and 2:57:12 p.m. This includes trades at approximately $100,000 which is presumably a top-end stub quote.
A single round lot of shares in Sotheby's would have cost ten million dollars at this price. Given that the orders were executed, it seems inconceivable to me that they came from retail investors.
What kinds of strategies could have been responsible for these trades? In January of this year the SEC published a Concept Release on Equity Market Structure that explicitly discussed the destabilizing consequences of certain strategies used by proprietary trading firms. Of special concern were strategies based on order anticipation and momentum ignition:
One example of an order anticipation strategy is when a proprietary firm seeks to ascertain the existence of one or more large buyers (sellers) in the market and to buy (sell) ahead of the large orders with the goal of capturing a price movement in the direction of the large trading interest... The type of order anticipation strategy referred to in this release involves any means to ascertain the existence of a large buyer (seller) that does not involve violation of a duty, misappropriation of information, or other misconduct. Examples include the employment of sophisticated pattern recognition software to ascertain from publicly available information the existence of a large buyer (seller), or the sophisticated use of orders to “ping” different market centers in an attempt to locate and trade in front of large buyers and sellers... An important issue for purposes of this release is whether the current market structure and the availability of sophisticated, high-speed trading tools enable proprietary firms to engage in order anticipation strategies on a greater scale than in the past.
A very different type of potentially destabilizing strategy seeks to engineer and exploit momentum in prices:
Another type of directional strategy that may raise concerns in the current market structure is momentum ignition. With this strategy, the proprietary firm may initiate a series of orders and trades... in an attempt to ignite a rapid price move either up or down. For example, the trader may intend that the rapid submission and cancellation of many orders, along with the execution of some trades, will “spoof” the algorithms of other traders into action and cause them to buy (sell) more aggressively. Or the trader may intend to trigger standing stop loss orders that would help cause a price decline. By establishing a position early, the proprietary firm will attempt to profit by subsequently liquidating the position if successful in igniting a price movement.
Order anticipation and momentum ignition are just extreme cases of a broad range of directional strategies that are either information extracting or seek to trigger information extracting algorithms. If too great a share of total market activity is driven by such strategies, major departures of prices from fundamentals will arise sooner or later. It is important, therefore, to allow such strategies to take heavy losses when they do eventually misfire. Macroeconomic Resilience has an excellent analytical post on the crash that makes a similar point:
Policy measures that aim to stabilise the system by countering the impact of positive feedback processes select against and weed out negative feedback processes – Stabilisation reduces system resilience. The decision to cancel errant trades is an example of such a measure. It is critical that all market participants who implement positive feedback strategies... suffer losses and those who step in to buy in times of chaos i.e. the negative-feedback providers are not denied of the profits that would accrue to them if markets recover. This is the real damage done by policy paradigms such as the “Greenspan/Bernanke Put” that implicitly protect asset markets. They leave us with a fragile market prone to collapse even with a “normal storm”, unless there is further intervention as we saw from the EU/ECB. Of course, every subsequent intervention that aims to stabilise the system only further reduces its resilience.
By canceling trades, the exchanges reversed a redistribution of wealth that would have altered the composition of strategies in the trading population. I'm sure that many retail investors whose stop loss orders were executed at prices far below anticipated levels were relieved. But the preponderance of short sales among trades at the lowest prices and the fact that aberrant price behavior also occurred on the upside suggests to me that the largest beneficiaries of the cancellation were proprietary trading firms making directional bets based on rapid responses to incoming market data. The widespread cancellation of trades following the crash served as an implicit subsidy to such strategies and, from the perspective of market stability, is likely to prove counter-productive.