Monday, May 10, 2010

Reflections on the Flash Crash

Index Universe observes that much of the unusual trading activity last Thursday involved exchange traded funds and notes:
Nasdaq has released a list of 281 securities that saw unusual activity during yesterday’s “flash crash” on the market... In all, 193 of the 281 securities (68.7 percent) on the NASDAQ list were exchange-traded funds or exchange-traded notes... The New York Stock Exchange has published a similar list, detailing 173 different securities whose trades will be cancelled. In all, 111 of those securities (64.2 percent) were ETFs or ETNs...
It was not immediately clear why ETFs dominate the lists.
Izabella Kaminska follows up on FT Alphaville:
ETF and ETN trading is closely related to high-frequency trading... Constant market-making and arbitrage opportunities are provided to authorised participants (often high frequency trading firms) by the ETF model’s dependence on converging to the net asset value on a daily basis. A typical fund has about five authorised participants.

The so-called creation and redemption mechanism allows authorised participants to lock-in profits when the shares of ETFs over-price or under-price the NAV, since only they are allowed to redeem or create shares at the official NAV price of the funds.
Dynamic hedging is needed to protect the arbitrage until the moment the creation or redemption process can take place... A significant change in any constituent stock in the interim can can hence fuel frantic fine-tuning of positions ahead of NAV publication time.
Can the algorithmic strategies used by authorized participants making markets in exchange traded funds help account for the crash? Not really. Index arbitrage of this kind simply brings the prices of exchange traded funds in line with the prices of their constituent securities, and is non-directional. This activity could explain a spike in volume as a result of sharp movements in prices, but this is a symptom rather than a cause of the crash. Something else caused the prices of the funds and/or the constituent securities to drop, and index arbitrage activity picked up as a result. What was this cause?

Some have pointed to the fact that liquidity vanished from the market during the crash, or that stop loss orders were triggered as prices fell. While these effects certainly accelerated and amplified the decline, there must have been an independent source of massive selling pressure that ran through the available bids, triggered stop orders, and caused electronic market makers to shut down. Again, where did this overwhelming selling pressure come from?

The best explanation that I have seen is contained in a message by an anonymous analyst that Yves Smith posted earlier today. The hypothesis is that the initial trigger came from algorithms implementing volume-sensitive technical strategies:
Volume was gigantic yesterday before we really went into freefall. As of 2 p.m., some 40 minutes before Armageddon, we were tracking for a massive 15.6 billion share day (we ended up doing 19.3 billion – the second largest day ever after the October 10th, 2008 whitewash). Half an hour later, at 2:30 p.m. – still ten minutes before the bottom fell out – volume had surged and we were tracking for a 17.2 billion share day. The period between 2 p.m. and 2:40 p.m. saw immense selling pressure in both the cash market and the futures market, and that occurred with the E-minis still north of 1120...

In other words, it was not a sudden, random surge of volume from a fat finger that overwhelmed the market. It was a steady onslaught of selling that pressured the market lower in order to catch up with the carnage taking place in the credit markets and the currency markets...
So what happened here? Three things:
  1. Sellers probably had orders in algorithms – percentage-of-volume strategies most likely, maybe VWAP – and could not cancel, could not “get an out.” These sellers could be really “quanty” types, or high freqs, or they could be vanilla buy side accounts. It really doesn’t matter. The issue here is that the trader did not anticipate such a sharp price move and did not put a limit on the order...
  2. Sell stop orders were triggered which forced market sell orders into an already well offered market. 
  3. While the market was well offered, it was not well bid. Liquidity disappeared... Bids disappeared, spreads blew out, and no one was trading except a handful of orphaned algo orders, stop sell orders, and maybe a few opportunists who had loaded up the order book with low ball bids (“just in case”). High frequency accounts and electronic market makers were, by all accounts, nowhere to be found.
It boils down to this: this episode exposed structural flaws in how a trade is implemented (think orphaned algo orders) and it exposed the danger of leaving market making up to a network of entities with no mandate to ensure the smooth and orderly functioning of the market (think of the electronic market makers and high freqs who can pull bids instantaneously as opposed to a specialist on the floor who has a clearly defined mandate to provide liquidity).
This rings true to me. Accounting for the crash requires us to go beyond the mechanics of the trading process (automation, scale, speed) and to examine the kinds of strategies that were being implemented by the algorithms. A market dominated by technical analysis is always going to be vulnerable to this kind of instability. The fact that the prices of some securities and funds crashed to absurd levels that were clearly out of line with fundamentals made this obvious and resulted in a quick recovery. But what if the trading strategies had given rise to upward rather than downward instability? It would have been more difficult to establish conclusively that assets were overpriced, and accordingly more risky to enter positions to bring them back in line with fundamentals. This, presumably, is how asset price bubbles get started. 


  1. Is there any economic benefit to automatic algorithmic trading? I can see where it might be profitable in some instances, but these traders seem to be trying to win a zero-sum game.

    Why not make this type of automatic trading illegal and make a real person authorize each bid and ask? It would certainly slow down the market and perhaps reduce volume, but would that be a bad thing? Would prices be any farther away from their "fundamental" value?

    I admit I don't know much about these markets, but if the Dow can lose 1000 points with no technical error, it seems like something needs to be fixed.

  2. Liam, it would be impossible to ban algorithmic trading - for one thing, it is not generally possible to tell whether an order submitted electronically is coming from a human being or a program. Also, exchange traded funds rely on algorithmic trading to keep share prices in line with net asset values. There's clearly a problem here that needs to be addressed but banning trading mechanisms is neither feasible nor necessarily desirable.

  3. Well, technically it should be possible to prevent algorithmic trading similar to how facebook prevents automated registrations using a text image. Stock exchanges have been around for much longer than computers or the internet so it is hard to understand why these robots are essential. On the other hand the harm this kind of automation can cause is very easy to see. We ban steroids in sports for a similar reason because otherwise, very quickly it becomes the wrong kind of race for the detriment of everybody. Formula 1 race engines are limited to 18000 rpm for a similar reason.

  4. The thing to ban may not be algorithmic trading but instant implementation of trades received.

    Technical strategy day trading is ultimately about swiftly appropriating information contained in market price information for yourself. People who are trading on real world information have little opportunity to benefit from their news because it is so quickly shared, reducing their incentive to trade on information instead of technical strategies.

    Delaying trade implementation would have the effect of a very short copyright, which would make technical strategies much less profitable.

    Delayed implementation regimes are also not nearly so old school as they seem. Billions of dollars of negotiable instrument transactions, for example, are closed at the end of every business day, rather than in real time. Mutual funds routine do the same thing - closing transactions at prices determined after the fact. Even wire transfers are much further from real time than action movies featuring payoffs to Swiss bank accounts would suggest; if you don't start in the morning, your odds of getting the deal done by the end of the day dim. Yet, somehow, business survives.