Statistical model predicts a McCain victory?

There has been a lot of buzz lately about a simple statistical model proposed by astrophysicists Prof. Gott and Dr. Colley that uses the median polls of each state to predict the November electoral vote. (For our un-American readers, the electoral vote is what determines the outcome of a general election, not the popular vote, in case the nightmarish 2000 election has not already drilled this fact into the world's collective consciousness.)

Dr. Colley has set up a website to track daily such polls to gauge the mood of the states. The authors have tested this method on the 2004 election, as well as numerous sporting events outcomes, and found it to be highly predictive.

Right now, they are betting on a McCain victory.

But there is one caveat that many bloggers have pointed out, and it is the same caveat that I have previously applied to the predictive accuracy of political futures market such as intrade.com. The caveat is this: polls (and futures market) change with time. And at different times, they predict different election outcomes. So for example, at this point (June 2008), the polls predict a McCain victory, while the futures market at intrade.com predicts an Obama victory. Who is right?

The answer is: neither. As Dr. Colley has explained to me, no backtest as far back as the June of an election year has been conducted. (Their research was based on polls from September onwards.) So we do not know if the June polling prediction has any accuracy. Similarly, as I pointed out before, the futures market can swing violently even on Election Day, even in the last hours of an election.

One advantage of the Gott and Colley method though, is that the predictions resulting from median poll statistics are remarkably stable over time. In 2004, there was very little movement in the electoral tally from September through election day. Extrapolating this result, we can be somewhat more confident of their prediction vs. Intrade.com's, even at this early date.

And in any case, I have observed that the political futures markets are highly mean-reverting, implying that the current large 20 points spread between the Obama and McCain futures is destined to decrease in the coming months.

As an arbitrage trader, I have therefore proceeded to short the Obama future.

Pre-order my book from Amazon.com

A reader told me that he can now pre-order my book "Quantitative Trading: How to Build Your Own Algorithmic Trading Business" from Amazon.com. The scheduled publication date is November 24, 2008.

Parameterless trading models

A portfolio manager that I used to work for like to pronounce that his trading models have "no free parameters". As is customary in our secretive industry, he would not elaborate further on his technique.

Lately, I begin to understand what a trading model with no free parameter means. It doesn't mean that it does not contain any lookback period for calculating trends, or thresholds for entry or exit. I think that would be impossible. It just means that all such parameters are dynamically optimized in a moving lookback window. This way, if you ask: "Does the model have a fixed profit cap?", the trader can honestly reply: "No, profit cap is not an input parameter. It is determined by the model itself."

The advantage of a parameterless trading model is that it minimizes the danger of overfitting the model to multiple input parameters. (The so-called "data-snooping bias".) So the backtest performance should be much closer to the actual forward performance.

Now, it is quite computationally challenging to optimize all these parameters just-in-time for your next order, but it is often even more difficult to do that in a backtest, given that a multidimensional optimization need to be performed for each historical bar. As a result, I personally have seldom traded parameterless models, until I get to research my regime-switching model. That model is almost parameterless (I left out a few parameters from optimization because of a lack of time, not because of any technical difficulties).

The reason backtest optimization can now be done within a few minutes is due to my use of Alphacet Discovery's server-based optimization engine. There may be other optimization software out there that performs similar functions efficiently -- I welcome comments from the reader.