MMA More on Factor Analysis
Previous  Top  Next

These paragraphs, offered by Bob Pelletier, were derived from a conversation with Steven Davis to stimulate additional thought --

Another view of the Factor Analysis table might be that lower order factors 1, 2, 3, & 4 identify markets with high cell scores that are correlated, hence markets that qualify for outright long or short positions, and the higher order factors, say 7, 8, 9 and 10 represent spread opportunities. The cell scores of larger positive factors identify candidate long positions in the historical past and the large negative cell score factors identify short positions.

Under these rules, factor 10 in our example suggests profits could have been captured by trading two corn contracts against one contract each of soybeans and wheat. All other cell scores for factor 10 appear to suggest nominal effects and, for that reason, might be ignored. The suggested inclusion to spread corn against wheat and soybeans has to do with the magnitude of the respective cell factors of these three markets versus the seven other nominal cell factors of the balance under factor 10. Please review the Unstretched Index results to determine the long short legs of the spread.

Kachigan's Multivariate Statistical Analysis, p. 236 - 270, suggests that it might "be somewhat illogical to retain a factor that contains even less information than an original input variable." This would be the case for factor 10, which explains less than one percent of the variances in the data. However, we believe it would be OK to review and act upon low variance items, given the promising objectivity of the price data (as compared with personality traits) and the relative magnitude differences between the factor 10 loadings.

The intermarket movement of correlated market products is imprecise because all of the information that affects them cannot be totally known. Weather, marketing, supply and demand and many other factors cannot be completely quantified. Fortunately, however, thanks to Unfair Advantage's enormous data repository of related markets, many elements of the equation are known through economically related products that coexist in a related environment.