Some more thoughts on the flash crash

I have written a number of times about the flash crash during the last five months (here, here, here, here, here, and here.) This post tries to put together the key issues as I see them.

• The flash crash happened towards the end of the day in the US when Asia and Europe were both closed and therefore nothing happened in those markets. The rest of the world has therefore been able to think of this as America’s problem. Had this happened when other markets were open, it would have been every body else’s problem.

  Quite apart from that, I think the flash crash is a serious issue for regulators in all countries. Large market orders interacting with a thin order book can cause a flash crash in any market anywhere in the world.

• The crash happened on a relatively calm and benign day. Yes, there were some protests in Greece, but that hardly counts as a crisis situation. Things would have been much worse if the crash had happened when a big bank was tottering or when a serious terrorist attack or a grave natural calamity was in progress.

  In this sense, the flash crash episode can even be regarded as a good thing to have happened. Without causing much damage, it provided a wake up call to regulators to analyze market designs and make them more robust. If we do not take corrective action, we will probably end up facing a much bigger market disturbance on a bad day.

• It is a mistake to think that the flash crash can happen only with automated high frequency trading. With a sufficiently thin order book, a flash crash can be triggered by a cascade of stop loss orders (at one level, a stop loss order is also an algorithm). Indeed, even a cascade of market orders is sufficient.

• It is time to reconsider the whole issue of internalization that allows brokers to execute proprietary trades against customer order flow without routing everything to a transparent execution venue.

  The SEC report discusses internalization at length, but its cognitive capture is so complete that it does not see anything wrong in what happened. What are the fiduciary responsibility of a market maker who thinks that the market is too risky to trade in on its own account but merrily routes customer orders into that same market? Is it acceptable to argue that this is fine because the trades would be cancelled anyway?

• The CFTC/SEC report has not provided confidence to the retail investors about the integrity of the market structure. It has failed to provide a convincing and conclusive answer to whether any market abuses happened during the flash crash.

• There is need to examine whether the SEC’s implementation of the National Market System is fundamentally flawed.

  The technology underlying the dissemination of the National Best Bid and Offer (NBBO) is so antiquated compared to the rest of the modern market infrastructure that the professionals seem to be relying only on proprietary price feeds.

  At the same time, as trading gets fragmented across multiple execution venues, the need for best execution is only becoming more and more pressing.

• As trading has moved to ever shorter latencies, the surveillance capability of the regulators has fallen far behind. Regulators do not seem to have the capabilities to analyze high frequency data at all and I think this is a serious problem.

• There is a major problem with the accuracy of time stamps of major market infrastructure providers. Today using GPS time synchronization and PTP instead of NTP, it is possible to achieve accuracy of a few microseconds, and regulators need to mandate this for systemically important entities.

• We should consider eliminating the ability of exchanges to cancel wrong trades completely.

  The SEC has moved to make the cancellation process a little more objective and transparent than before. But today’s information technology makes it possible to get rid of trade cancellation completely. If an objective algorithm exists to determine the trades to be cancelled, the same algorithm can be used to prevent those trades from happening at all. Unlike in a manual system, the reaction time of a computer can be fast enough to “cancel” the trade in real time by not matching the trade at all.

• The evidence seems to suggest that the market structure has become so fragile that it is an accident waiting to happen. I am reminded of the old nursery rhyme:

  For want of a nail the shoe was lost.
  For want of a shoe the horse was lost.
  For want of a horse the rider was lost.
  For want of a rider the battle was lost.
  For want of a battle the kingdom was lost.
  And all for the want of a horseshoe nail.

  When this kind of a thing happens, the solution is not to go in search of the blacksmith who lost the nail (or the mutual fund that did a large futures trade), but to build more buffers and make the system more robust so that a few nails and horses can be lost without catastrophic consequences. Just as we stress test individual banks, it is also necessary to stress test the entire market structure.

  The best way to do that is to build simulated computer models of the entire market structure including the most popular trading algorithms and then stress test the whole edifice.

Posted at 8:23 pm IST on Wed, 20 Oct 2010         permanent link

Comments