Algorithmic Trading Problems

My main concern, four years ago, when algorithmic trading became popular was that it is a feed-forward system. The algorithms respond, in near real-time, to changes in electronic signals (bid, offer, trades) hoping to move the market in a direction favorable to the algorithm's goal(s). The problems with feed-forward systems specifically and controller theory in general are well documented and one of the primary problems is the inability of these systems to respond to novel situations like what was experienced this week in the markets.

I have worked in the financial markets the past seven years, and before that spent nearly ten years in the high energy physics research environment of Fermilab. I was more than a bit concerned when I started seeing algorithmic trading specifications, typically using the FIX protocol, show up on my desk about two years ago. My concerns center around the general lack of training most programmers receive regarding loop control and controller theory.

Therefore, this week when the markets started to fluctuate, in novel ways, I was not surprised when technical snafus at NYSE, Dow Jones and Arca started the electronic trading version of a microphone placed too close to a speaker. I have to wonder, but will probably never know, what role algorithmic trading engines contributed to the inflated tick volumes observed this week?