By far the most important part of the correlation between stock price movements is explained by the effects of broad market movements. On the other hand, a model based only on overall market betas would predict a higher correlation between GM and AT&T than between GM and Ford.

The author develops five clusters of stocks corresponding to major extra-market factors, monitoring the explanatory power of each cluster statistically to determine the point in the clustering process at which the value of a cluster as a descriptor of extra-market stock price movement peaks. The clustering process terminates when clusters begin to be diluted by irrelevant stocks — for example, at the point where food companies join the utilities cluster. The explanatory power of the resulting clusters for the evaluation of extra-market risk represents better than a 30 per cent improvement over the single-index model.

The clusters embody important fundamental characteristics. The utilities cluster is dominated by interest rate sensitivity, while the cyclical cluster is strongly influenced by the economic outlook. Thus, although factor betas change over time, they are actually more stable than market betas.

The multiple factor risk model can be applied to stock classification, portfolio optimization and performance measurement, where, for example, its use simply eliminates the pesky benchmark selection problem.