|
|
|
|
|
|
|
could explain the market's actions. If I encounter a good predictive variable, then I optimize it initially and I don't bother with it any further. |
|
|
|
|
|
|
|
|
Neal: So what is your opinion of systems that optimize periodically? |
|
|
|
|
|
|
|
|
Committee: Generally, they are low on my list of systems I want to work with. It reminds me of going for medical treatment and only having your symptoms treated. The root cause of the problem is not addressed, and new symptoms show up that need treating again. I believe a system that needs periodic optimization does not contain good explanatory variables. The optimization treats a surface phenomenon but does not get to the root cause of what makes the market move. |
|
|
|
|
|
|
|
|
Neal: What other statistical methods do you use? |
|
|
|
|
|
|
|
|
Committee: One of my favorites is factor analysis. Factor analysis is defined as an analysis technique to reduce a large number of variables to a more manageable set of dimensions or underlying constructs which explain a large portion of the variability among the various measures. In plain English, factor analysis is used to group commodities together by profits. Each single group contains a number of highly correlated commodities. I am a great believer in portfolio diversification and I use factor analysis to check how well I've diversified. Here is an example of how I check my portfolio's diversification. Let's say I want to develop a portfolio using 90 different commodities. First, I run all the commodities through my system, using a ten-year period. The results I use are the daily change in equity for each commodity. The result is 90 columns, each containing the daily equity change. The rows all line up according to date. Now I run the factor analysis on the data. If the result is 30 groups, then I am very happy. This means there are 30 groupings that are not correlated to each other. Also, one group has a very small effect |
|
|
|
|
|