Streetwise Professor

October 3, 2010

A Flash in the Pan

Filed under: Derivatives,Economics,Exchanges,Politics — The Professor @ 8:28 pm

The CFTC-SEC joint report on the May 6 Flash Crash.  I am underwhelmed.  It provides some new details, but (a) leaves the most important questions unanswered, and (b) fails to perform some rather straightforward analyses that would go a long way to providing a much stronger basis for its conclusions–or disprove them.

The report supports most of the conjectures I (and others) made in the immediate aftermath of the crash.  In the chaos of the time, with the profound uncertainty surrounding the Greek crisis, liquidity on the major markets went down.  In particular, the depth on the buy side of the S&P Emini contract on the CME declined dramatically.  Into that declining liquidity came a sequence of large sell orders entered as part of a hedge implemented by a firm not named in the report, but widely reported to be Waddell and Reed.  Given the diminished liquidity, the sell orders pushed down the market, which triggered arbitrage trades that dragged down other markets as well.

Whenever you see a big price move in a short period, you look for positive feedbacks.  The positive feedback I conjectured in this case at the time was a dynamic put replication strategy, in which the hedger sells more as price declines in order to match the delta of a put position.  Although the report indicates that the W&R strategy was a hedge, it does not discuss in detail what triggered the initiation of the strategy: why did the hedge kick in just then?  Key question, but an unanswered one.  It is plausible that the earlier selloff had triggered the initiation of the W&R program.

But the report does suggest that once the original order was submitted, there was a positive feedback mechanism, but a truly novel one, and something different than dynamic put replication.  In particular, the algorithm that W&R employed based its sell orders on market volume.  It had a given number of contracts to sell (75K), and the timing of those sales at any time was conditioned on trading volume in the immediately prior minutes.  According to the report, the initial big price drop triggered a substantial amount of HFT trading activity.  This was mainly offsetting: HFT buys and sells were almost exactly equal in magnitude.  But this generated volume; this volume caused the W&R program to submit more sell orders; and this reinforced the price decline.

This is a plausible story, but the Commissions could have done more to support it.  In particular, the Report asserts a causal connection: W&R sell orders; price decline; big uptick in HFT trades; more W&R orders; more price declines.  These causal hypotheses are readily tested in a standard vector autoregression framework.  The elements of the vector could include price changes (on a second or five second time frequency, for example); (signed) W&R orders; HFT buys; HFT sells; and the buys and sells of other trader types.  It would also be worthwhile to include buy side and sell side depth changes in the vector to see whether liquidity was responding dynamically to any of these other variables, and vice versa.

This framework would permit a more definitive attribution of the price move to the various factors identified in the report, and permit a better understanding of the actual dynamics and feedbacks.  If, indeed, the mechanism posited in the Report is correct, the VAR analysis would demonstrate this.

A better understanding of the various feedback mechanisms would also facilitate the development of better market surveillance tools.  Most importantly, it would provide information that would permit more discriminating regulation of HFT–or demonstrate that no such regulation is necessary.

But the most disappointing part of the report is that although it documents the decline in EMini liquidity, and discusses in general terms the potential causes of this (uncertainty surrounding Greece, rising volatility as indicated by the VIX), it does not undertake a thorough empirical investigation.  With normal liquidity, the W&R orders likely would not have had any appreciable price impact.  With the emaciated liquidity of the afternoon of 6 May, the sell program plausibly had the price impact the Commissions attributed to it.  So understanding why liquidity declined is essential to an understanding of why the Flash Crash occurred.

In other words, if the sell program was the match, the lack of liquidity was the dry tinder.  That’s what needs a more thorough investigation.

Again, the VAR framework would be very useful in this regard.  One could investigate the dynamic relations between buy and sell side depth changes, volatility changes, price changes, customer order flows, and other factors that could affect liquidity.  This would permit a more discriminating analysis of what triggered the decline in liquidity that made the Flash Crash possible.

The analysis could be extended to include a larger sample period, in order to get a more complete understanding of what drives liquidity in these markets.  Without understanding this, it is impossible to understand the role of HFT.  A more thorough understanding of the dynamics of liquidity supply would also help exchanges monitor markets for potentially destabilizing developments.  This could permit the development of smarter circuit breakers.

The report makes clear, as should have been evident soon after the Crash happened, that the CME’s stop order logic played a crucial role in stopping the cascading prices.  This was a pretty smart circuit breaker, but a better understanding of the dynamics of liquidity supply would permit the development of even smarter ones.

And this shouldn’t be a one time, fire-and-forget exercise.  The dynamics of liquidity supply will almost certainly evolve over time.  Frequent updating of the analysis would help identify these changes, and permit a similar updating of market monitoring and circuit breaker mechanisms.

In brief, we know a little more about the causes of the Crash, but not much more than we did on the day and the days immediately following. The Commissions’ staffs obviously trolled through massive amounts of data, for which they should be congratulated.  But it was possible to do so much more with that data.  Maybe the staffs did, but it didn’t make it into the Report.  Standard analytical tools applied to that data could provide a far more detailed understanding of the complex feedback mechanisms that contributed to that Crash, and which could recur in future ones.  Let’s hope that analysis gets done, either by the Commissions, the exchanges, or independent scholars.  Armed with such an analysis, it will be possible to craft better policy responses.  Without it, policy mistakes are more likely.

Print Friendly, PDF & Email

No Comments »

No comments yet.

RSS feed for comments on this post. TrackBack URI

Leave a comment

Powered by WordPress