One story circulating to explain today’s selloff is a large, mistaken order entry by Citi. Certainly not outside the realm of possibility. It’s happened before. There’s the famous story of the MATIF trader who swung around in his chair to speak to a colleague, and unwittingly rested his elbow on the “Sell 100” key on his trading keyboard, triggering a deluge of sell orders that caused the market to tank before he realized his mistake and started buying furiously. Or the case of the guy who thought he was in the training mode of the Eurex system, when in fact he was in the live trading mode; he was goofing around, submitting big orders into the “simulated” market that were in fact going into the real market and causing the price to go crazy. As I recall, this cost his employer (a German bank) $150 million. Or the error in Japan where somebody submitted an order at a price that was off by several orders of magnitude.
So yes, stuff like this can happen in computerized markets, although systems are being made more robust to these kind of errors. More robust, not completely so.
And, perhaps, once the original mistaken order went in, and affected prices of Dow stocks, that triggered other programs that caused additional selling.
But I am highly confident that the near immediate snap back was also computer driven, as other algos’ signals indicated that the new prices were too low and submitted buy orders.
The post mortem will be interesting.
The terms “program trading” and “computerized trading” are used so much it’s worthwhile distinguishing the problematic kinds from the beneficial ones.
Computerized trading was widely believed to have exacerbated the 1987 Crash. The order execution then wasn’t computerized, but portfolio insurance strategies made order submissions contingent on price movements according to a computer algorithm, so the trading was computerized in some relevant sense.
This is the kind of automated trading that is problematic because it can create destabilizing positive feedback effects. Synthesizing index puts as portfolio insurance through a dynamic trading strategy means that big price declines triggered more sell orders that arguably exacerbated the price declines which caused additional sales, and so on. Option hedging strategies (e.g., dealers use dynamic hedges to manage the risk of options they’ve traded with customers) can have the same effect.
Stop orders (which probably contributed to what transpired today) can have a similar effect; price declines (rises) trigger sell (buy) orders that can exacerbate price moves. These kinds of orders are as old as the market.
Margin-driven trades can do the same: those suffering losses on a price decline (increase) sell (buy) and who cannot come up with the necessary margin sell (buy) to close positions–again, as old as the market.
Some computerized trading–and I would argue that most algo trading–is very different. It is a negative feedback strategy. That’s what market makers do: they sell into purchases and buy into sales. Much algo trading is effectively automated market making. It is, in effect, the realization of what the great Fisher Black imagined in 1971, when he wrote an article in the Financial Analyst’s Journal titled “Towards a Fully Automated Stock Exchange.” Black envisioned a fully automated specialist that made markets, and thereby stabilized prices. Computerized/algo market making programs based on negative feedback would have bought on today’s decline as Black described. The rapid snap-back is perfectly consistent with that.
Moral of the story: ignore categorical condemnations of computerized or quantitative trading in the aftermath of today’s events. There’s good, bad, and ugly. Be careful, and try to distinguish them.