Postmortem analysis is nearly a cultural obsession. Witness the popularity of ESPN’s “SportsCenter,” which may be trumped only by daily postmortems relating to the U.S. presidential campaign. A viewer cannot turn on the television without a review of how a candidate might have improved her chances, or how he “misspoke” and certainly how each event had a cumulative effect on the (still forecast) outcome of the election.
The market dislocation beginning in August of last year offers numerous possibilities for postmortems’ combination of hindsight and soothsaying. Titles such as “What Happened to the Quants in August 2007” and “Deciphering the 2007/2008 Liquidity and Credit Crunch” now appear regularly on conference agendas. No surprise here: Anything out of the ordinary offers an opportunity to check a favorite theory against events.
Questions on Costs
Transaction research is no exception, and the queries have been coming in. Are transaction costs rising with volatility? What about those spreads? And we see a shift toward VWAP trading in the second half of 2007-tell us why.
VWAP aside, there is an interesting lesson for trading cost analysis from that period. Intuitive and familiar influences may have failed to predict trading costs in the short term. Yet shifts in trader behavior, as predicted, and even guided, by transaction research, resulted in a rapid return to average performance, despite the continuation of a tumultuous environment.
As volatility trends go, the period from August 2007 to the first of this year is an excellent illustration. Depending on your favorite measure, volatility rose between 50 and 70 percent. The levels of, say, the VIX, are not unprecedented historically. On the other hand, transaction research only began to “come of age” in 2002, and retrospective analyses often focused on the drop in equity transaction costs during the 2002-2005 period, as volatility declined.
It is difficult, however, to link transaction costs to changes in volatility in today’s environment. Over the last three years or so, the correlation between transaction costs in our database and 60-day averages of forward-looking volatility is less than 3 percent, for example. That figure moves around by a few basis points, depending on whether one looks at buys, sells or capitalization differences and the like, but the relationship between the level of transaction costs and volatility is tenuous, at best. In contrast, the correlation between bid-ask spreads and volatility during the period is more than 30 percent. It seems almost a shame that, post-decimalization, spreads lost predictive power with respect to transaction costs.
Just as the bottom line in the presidential campaign must be the election result, the important thing here is trading performance as measured by trading costs, since the shortfall chisels away investor returns. So, what happened to transaction costs following August 2007? They climbed-all the way to their average value since August of 2004. Equity trading costs looked about the same in January 2008 as they did in June 2007 or December 2006.
The real news is not in the levels of cost, nor the lack of relationship with two-sided price movements, but rather the path by which costs reached their January level.
Costs rose sharply in the aggregate following August, tracking volatility for a short period. Trading strategies then changed. The decline in costs from the peak was almost 40 percent, and began within a couple of months after the initial volatility spike.
Amongst other things, our data suggest that algorithmic trading usage increased by more than 20 percent, moving from an average to a high volatility period as characterized by the VIX. Although this might seem counterintuitive to some, previous research has documented the efficacy of algorithmic trading relative to alternatives, as order size falls and volatility rises, which was the case during this period.
Crossing networks tend to expect less flow in high volatility periods, and usage of algorithms to access multiple dark pools did appear to decline initially. Despite information leakage from algorithms shopping an order to multiple pools, as opposed to the confidentiality of direct crossing network access, algorithmic access to dark pools increased after the initial reaction to draw back.
The performance of algorithmic access to dark pools tends to degrade more slowly with order size, relative to other algorithmic trading alternatives, as volatility goes up, explaining part of this behavior.
The structure of the dark pool environment and the nature of access, coupled with smaller execution sizes given the multiplicity of venues, also make order execution look somewhat like interval VWAP.
Did I say VWAP? Anecdotally, the use of VWAP strategies increased sharply in the second half of 2007. VWAP has known problems as a suitable “benchmark” against which to judge trading performance.
On the other hand, one of the results in transaction research is a demonstration of its usefulness as a “strategy” in achieving the more legitimate benchmark of implementation shortfall under certain conditions.
For order sizes approximating those observed in algorithmic trading applications and in a high volatility environment, VWAP essentially lies on the efficient frontier linking trading costs to opportunity costs of slow execution, expressed in terms of risk.
It provides much the same cost improvement as a passive cost-minimization strategy, but with even less risk as volatility rises. Another example lies in a comparison between VWAP and volume participation strategies, which can look quite similar. The performance of volume participation degrades sharply, however, relative to VWAP as market uncertainty increases, making VWAP the better strategy in the face of high volatility.
These stories suggest one way to view the low correlation between transaction costs and volatility: As volatility changes, strategies adapt, and costs fall back in line with frictions inherent in today’s market structure.
The same view applies to transaction research. Trading cost analysis is not about the predictability of costs for any particular order in a given name over a single interval of time, any more than portfolio decisions are made on the basis of the statistical predictability of excess returns; the former is admittedly low, and the latter typically is much lower.
Transaction research is fundamentally about insulating trading performance from volatility, or other changes in market conditions, by providing information leading to smart strategy choices and the ability to change them.
It once was true that objective analysis of a successful political campaign after the vote was of the most use to the candidates in the next election. The day-by-day postmortems and resulting shifts in real-time strategy now rule campaigns. Similarly, transaction-cost analysis is no longer simply a post-trade event, but rather an input to trading decisions made not only at the beginning of the day, but continuously through the life cycle of an order.
Ian Domowitz is Managing Director of networking and analytical and research products at Investment Technology Group.
(c) 2008 Traders Magazine and SourceMedia, Inc. All Rights Reserved.