What happened? Once again, today's mini-crash raised the question, "How is it possible for computers to react before humans do?"
It's a good question, because the old sequence of events (news comes out, markets react) has been subverted: now, it's markets drop in seconds, and we all stand around trying to figure what the heck just happened.
High frequency trading programs can react to news quickly in several ways. Here's two simple methods:
1) Key word searches. Much is made of the idea that computers can be programmed to respond to key word searches that would, for example, respond to a headline that had the words "White House" and "explosion" in it and immediately institute a sell program.
It's true you can do this, but it would be dangerous. And stupid. It takes more than the ability to dumbly read the news. You have to have a database of historical events to know what that information is worth. It's very easy for a computer to misread the wording, or for the information to be false, or for the person posting the news to have it wrong. The ideal system would be to scan keywords and then brings them to the attention of traders.
This is not to say it doesn't happen: Reuters and Dow Jones have feeds that you can build trading programs around keyword searches, and you can do it for free with Twitter. And I have no doubt some might have done this. But you'd feel pretty stupid if you did, after what happened today, wouldn't you?
2) Trend following. This is where you can make money. The object is to buy low and sell high (duh). It doesn't care what the news is. If a program can sense a sudden outsize move in the market to the downside, it starts selling...then buys when the market bottoms and begins to turn around.
Easy enough, if you have a program clever enough to catch the bottom.
I spoke to one high frequency trader who told me he made a lot of money selling into the drop today, but then lost 70 percent of that when the program did not correctly call the exact bottom. So these programs aren't perfect. All this, in about a minute, by the way. But hey, 30 percent ain't bad, no?
One other point: while there was no doubt a lot of selling and buying in those few minutes, the main thing I saw standing on the floor, watching the designated market makers (formerly specialists) was that all the bids got cancelled.
Now, this makes sense to me. If you're posting bids and all of a sudden message traffic (bids, offers, modifications, cancels) goes crazy, your first reaction is to cancel the bids...and here it would be perfectly reasonable to have a program that would do that.
One final point: the most important takeaway from today, from the trader's perspective, is that the credibility of Twitter as a news source for business has taken a big hit. This wasn't the first time: Twitter accounts for CBS and NPR had already been hacked.
"CNBC didn't have it, Google News didn't have it," one high frequency trader said to me. "Twitter is the first place that news breaks, but what we found out today is that a lot of it is fake."