We've all been duped by The Terminator movies. The vision of a hellish future with murderous and murderously efficient robots after the computer networks became "self-aware" and taking over the world seem to be on the same shelf as the Flying Car, personal jet packs and the cure for the common cold. We've been had. Why? Because robots are dumb. Look no further than Knight Capital. Last week, the respected and tech-savvy market maker lost $440 million in a single day and had to sell itself to other firms in order to trade another day. The culprit was a newly minted trading algorithm that the firm deployed and soon went haywire. We all know what happened next but looking back, it was the Perfect Storm of failure among markets, technology and the people who should have known better.
Advanced Trading has covered much of Knight Capital because high frequency trading and algorithmic trading is our sweet spot. And we realize that once you create inter-connected networks that allow millions of messages to travel around the globe in seconds and often less, that things are never going to slow down.
We Humans rarely slows down and will only do so grudgingly if safety is concerned. We can buy cars that go 150 miles per hour but no one would drive that way with our loved ones in tow. (Besides the roads and highways are too crowded.) The age of supersonic flight has gone to rest with the Concorde even though space travel entrepreneurs like Virgin's Richard Branson and others are intrigued by public space flight via private companies. And even the most ardent fans of military fighter jets has to wonder when the next air show will feature the US Air Force Thunderbirds in their pilot-less drones instead of F-16 Falcons.
But what can be done with algorithmic trading when the trades are happening so fast that the machines may be the only entities that can stop the trade before it does too much damage? Science writer and former software engineer Ellen Ullman has an idea:
Credit card companies offer us a model, albeit a flawed one. Cardholder liability is limited by law: if a charge is fraudulent, issuers are responsible for most of the cost. They therefore have an incentive to detect events that are out of the ordinary. These companies use artificial intelligence programs that get a "sense" of your normal purchases: where you use the card, the range of the amounts involved, the sort of vendors you deal with. The programs can sometimes be too vigilant (freezing your card on your first trip to Paris), but their reaction is often correct. They put the questionable charge in abeyance, and the company contacts you, asking, Is this valid? Creepy as it is to have the issuer probing into your habits, by protecting themselves, credit card companies also work for you.
Algorithmic stock traders can learn from this example. Each company should create artificial intelligence programs that recognize unusual patterns. The S.E.C. and other regulatory bodies should independently deploy their own systems. The role of these "watcher" programs would be to slow things down and inform a human being that something seems strange: artificial intelligence meets human intelligence, and the human gets to sort things out.
The FAA and NORAD have systems that can see every plane and helicopter that is in the air, why not a system that sees every trade?
Or perhaps the solutions comes down to a matter of time. Before we build this super surveillance system, we could partition algorithmic trading to a one-hour window in the trading day. Bear with me. If we keep all algorithmic trading within a certain window then the damage from a rogue algo would be contained because the entire market would be watching.
But that would just create greater volatility, you say? Perhaps but the average trading day has it's own pattern: tons of activity at the very start and end of the day with a near flat lull from 11am to 3:30pm. Place the Algo Hour in the middle of the day, say noon eastern time, and everyone in the mainland US can trade along with some folks over in Europe before they head home.
When we are approaching the speed to light in our trades maybe the only safe response is to limit time and space.
Would this work? Tell me at firstname.lastname@example.org.