Tuesday, February 9, 2016

I did a lot of work on things like this...

I did a lot of work on things like this ... back in the early 2000s, when I worked for an electronic stock and options company.  “Algorithmic trading” was just coming into its own back then, and some companies were making a lot of money with some relatively simple algorithms.  For instance, “pairs trading” was a common strategy.  The notion behind it was very simple: by examining historical stock pricing data, someone would identify a pair of stocks where the price on one of the pair (call it "AAA") change before the price on the other (call it "BBB") changed, and in a predictable way.  In that case, you could write a program that watched the price of "AAA", and automatically bought or sold "BBB" on the assumption that it's price would change sometime later.

We built a pairs trading platform, and sold it as a service to hedge fund traders.  It didn't take us long to observe a problem: soon after someone started using a particular pair of stocks, the correlation between the pair's prices would decline and disappear.  Why?  With some investigation, it was obvious: other people would soon start using the same pair, and after that, the market erased any benefit (when lots of people see such a correlation, the price automatically gets adjusted by market forces to eliminate the benefit – it gets “priced into” the trailing stock).  How did these other people know to trade in the pair?  That's also easy: with a simple algorithm like pairs trading, you can figure it out from the trades that someone makes – and those trades are public information.

So our CEO decided that what we needed was an algorithmic trading platform that used an algorithm so complex that nobody could reverse-engineer it.  Personally, I thought that was crazy – with all the trading data available, it seemed really unlikely that such an algorithm existed.  I thought a more sustainable route was an algorithmic trading platform that depended on something more reliable: that computers are always going to be faster than humans.  We had a series of contentious management meetings on the subject, with the result that we sent off to build that complex algorithm platform.  Shortly afterwards I was laid off.  I have no idea if my skepticism was even part of why I was laid off. 

I note with some belated satisfaction, however, that the high-speed trading is in fact the way the industry has gone – albeit with far more fanaticism than I'd ever have thought.  For instance, there are now point-to-point radio links in place between traders and the markets in New York (and similarly in other parts of the world).  These links exploit the tiny advantage in the speed of electronic communications that a straight-line route has over the traditional wired route.  That tiny advantage justifies many millions of dollars in cost to set up those radio links.  Amazing!

No comments:

Post a Comment