The stock market crash last week exposed me to the world of algorithmic trading. Silly me, I thought free market economics had triumphed and that the Chicago school of efficient markets implied that there is no strategy better than a buy and hold. Perhaps, efficient markets only exists in academics and regulatory agencies and not in financial markets - which must be true since according to Rajiv Sethi (emphasis mine):
they have become increasingly common recently, and now account for three-fifths of total volume in US equities:
Algorithms have become a common feature of trading, not only in shares but in derivatives such as options and futures. Essentially software programs, they decide when, how and where to trade certain financial instruments without the need for any human intervention... markets have come to be dominated by “high-frequency traders” who rely on the perfect marriage of technology and speed. They use algorithms to trade at ultra-fast speeds, seeking to profit from fleeting opportunities presented by minute price changes in markets. According to Tabb Group, a consultancy, algorithmic and high-frequency trading accounts for more than 60 per cent of activity in US equity markets.
This is a recipe for disaster:
[In] a market dominated by technical analysis, changes in prices and other market data will be less reliable indicators of changes in information regarding underlying asset values. The possibility then arises of market instability, as individuals respond to price changes as if they were informative when in fact they arise from mutually amplifying responses to noise.
Under such conditions, algorithmic strategies can suffer heavy losses. They do so not because of "computer error" but because of the faithful execution of programs that are responding mechanically to market data.
But the fact that the market gained in the next fifteen minutes is a triumph for free market economics - that the market was mispriced and participants acknowledged it.
So, how could something like this happen? An insight from Kid Dynamite (HT: MR):
... one possible explanation is that an algorithm gets into a "sell loop," where it is sending out a sell order which, for some reason or another doesn't get immediately acknowledged, or doesn't get acknowledged within the threshold for the algorithm. So, the algo, thinking its order didn't make it, spits out another order, which doesn't get acknowledged, and it spits out another, etc etc etc. Why would the algo be coded to send out a new order before it had an "out" on the prior order? I don't know - that seems odd to me, (any algo writers out there? please share your knowledge in the comments section) but I remember a great story that my boss used to tell on our trading desk about how in the old days, when they sent a program trading order to sell a basket of stocks, the old dot-matrix printer behind them would loudly whir to life and print up a confirmation that the basket went. One day, trying to sell a basket, they got no confirmation. Silence. So they hit the button again. And again... and again... Until finally someone shouted - "THE PRINTER IS OUT OF PAPER," as they watched the market fall under the cascade of sell orders that had just been sent.
The recent news is now that this may in fact be the case (no, not that the printer was out of paper):
Regulators examining the causes of the brief stock market free fall last Thursday are looking closely at heavy selling in the market for stock-index futures by a single trader, beginning 10 minutes before stock prices began to plummet.
Gary Gensler, the chairman of the Commodity Futures Trading Commission (CFTC), said at a congressional hearing on Tuesday, May 11, that during that crucial time period, the futures trader, whom he would not identify, accounted for about 9% of trading volume in the most actively traded stock-index derivative contract, known as the 500 e-mini futures contract.
All of the trader’s orders were to sell, Gensler said, while most of the other 250 traders who were active in the same market that day were both buying and selling securities.
As the trader’s orders went through, the futures index on the Chicago Mercantile Exchange (CME) began to plummet.
What surprised me was how long it was taking the SEC to reconstruct the trades until I read this (emphasis mine):
We now have more than 50 market centers, which has brought added competition. Today, algorithmic trading interests are wired across markets -- equity, fixed income, futures and options; the market is the network -- and yet our regulators work in silos. Responsibilities are divided between the SEC and CFTC. Within equities markets, we have multiple Self Regulatory Organizations setting rules -- more silos: New York Stock Exchange, NASDAQ, FINRA, National Stock Exchange, and more. All too often, those rules have been watered down and eliminated in the absence of the SEC establishing these and other regulatory controls across equity markets. We created a "National Market System" but we forgot to create a "National Regulatory and Surveillance System" to go along with it. ... The Commission does not yet collect by rule the data it needs efficiently to reconstruct unusual market activity.
Yet, Jim Hamilton has the most insightful comment:
Let's take a look at one of the amusing examples of trading on Thursday. Exelon is a perfectly sound utility. The stock started and ended the day at about $42, yet at one point allegedly changed hands for a penny a share. If you are the owner of 100 shares of this stock on Wednesday May 12, the company will send you a dividend check for $55. They'll send you another $55 (and likely eventually even more) every 3 months thereafter, as long as you own the stock. What rational person would possibly prefer to have $1 on Thursday May 6 in preference to owning 100 shares of the stock, say, for at least another week?
No sane person ever would. I wonder how many of the sell orders came not from actual humans, but instead from instructions programmed to execute automatically by computers. Anybody following such a strategy is obviously not a subscriber to my theory of fundamental value, but instead seems to be playing Keynes' beauty contest game, having persuaded themselves, based on the historical correlations, that when the crowd starts to sell, if you can instruct your computer to sell at the "market price" faster than a human can even think, you'll come out ahead.
Economists and financial engineers seem to approach stock markets differently. The financial engineers often see the game as figuring out whatever the historical predictability has been and trying to get ahead of it. Economists tend to ask what the equilibrium in the market would look like if everybody played the game the way the financial engineers do. The answer to that second question is, if momentum-chasing algorithms come to rule the financial world, those who try to follow them will be the biggest losers.
Another notion that's popular with many financial gurus these days is the claim that you can eliminate certain risks to your portfolio with the right strategy of automatic trading and stop-loss sell orders. Again that claim invites an economic question-- if you are getting an insurance policy, who is selling it to you? I believe the implicit answer is, you are counting on the market-maker to insure you by taking the other side of your escape transactions. But the curious thing about such an insurance policy is that the market-maker gets to decide what premium to charge you after you ask to collect on the policy. You just might find that the state of the world when you and your buddies all most desperately want to cash in on your insurance is exactly the time when the premium proves to be ruinously expensive.
Or if I haven't persuaded you with these arguments, let me try a more modest suggestion-- those of you whose computer programs sent an order to sell Exelon even if you only get a penny a share might want to consider tweaking the code just a bit.