Rhody Trader

I’ve been involved in what you could rightly call a long-running project. It involves a neural net based trading system which a professor friend of mine has been very slowly bringing up to being something useful. Needless to say, if I were running the show things would have moved much more quickly. It being under the control of an academic, however, snail’s pace doesn’t even do the progress justice.

Anyway, he finally got something producing visible trading indications a bit over a year ago. The results were fantastic. It focused on stocks and was generating monthly portfolio gains in the 10% area and better. That didn’t account for what would be meaningful transaction fees given the trade frequency, but even factoring that in the gains were quite exciting.

Now, my big issue with all this was that we needed to produce a much longer and deeper back test to see how the system would do in alternate volatility environments. Meanwhile, the professor was fixated on reducing the loss %, which ran about 20%. I keep trying to tell him that in trading 80% is a good win rate, especially if your winners generally exceed your losers. He’s kept tweaking, though, eventually to the point where he broke something – not surprisingly, as far as I’m concerned.

Aside from the researcher’s need to keep improving things - which I understand – he’s got a metric that he wants to meet or exceed. To provide a quote:

Back to AT, the 5-6 papers / symposiums we attended were reporting error rates under 5%.  I can believe that because their models were superior to what we have created.  No surprise there becasue we just didn’t get back to the model building stage.  What they lack (and what we have to my surprise) is extreme creativity on how we (and other American firms) have built a self-managed AT system – in our case using basic micro parts.

AT means “auto trader”, in case you’re wondering. When he says “they” he’s referring mainly to European researchers, who apparently have an edge on us Yanks in the more quantitative aspects of finance. Error rates means incorrect directional calls – losing trades. What I don’t know is how the winning trades compare to the losing ones.

I keep trying to explain that it’s a question of expectancy and frequency (with transaction costs accounte for), not win %, that determines trading success. The effort to increase win % almost invariably reduces the ratio of size of winning trades to losing ones. At some point it usually means tipping over the edge into declining performance.

Hopefully one day we’ll actually have something we can put into productive use. In the meantime he wants to help a student do some parallel work on the forex side.

At least I’m not dealing with one of the professors who believes his knowledge of financial theory guarantees him of trading success. Those kind of arrogant academics often fine themselves getting bitch slapped by the markets.

Trackback URI |