Page 310 - DCOM504_SECURITY_ANALYSIS_AND_PORTFOLIO_MANAGEMENT
P. 310

Unit 12: Models




               We repeated this calculation by substituting a historical beta (calculated using trailing  Notes
               daily returns) for the predicted beta; we used trailing periods of 60, 120, 180, 240, 300, and
               360 business days to calculate six different values of trailing historical betas. We then
               repeated all of these calculations using 60 business days’ returns for the calculation of the
               realized beta.
          2.   Interpreting the Results: A perfect predictor would have regression results of a = 0, b = 1,
               correlation = 1, and MAE = 0. While these results are far from perfect, it is important to
               remember that they are for individual securities; predictions for portfolios can be expected
               to be far more reliable.

               It is more useful to look at the results on a comparative basis. For each line, the shaded
               values of b, correlation, and MAE are the closest to ideal. We can see that all of the shaded
               numbers are associated with either the daily historical beta or the average of the predicted
               and historical beta. While we cannot conclude from this that daily historical betas are
               significantly better predictors of realized beta than Barra-predicted  betas, it certainly
               raises the question of whether the Barra-betas (or any other multi-factor betas) are the best
               predictors.

               !         There are a few other interesting results worth noting:
             Caution

             (a)  The “b” in the regression results for the predicted betas are greater than 1. This is not
                 necessarily good  or bad, but simply indicates that the predicted  betas have less
                 dispersion than the realized betas. This makes intuitive sense, since the predicted
                 betas are based on longer-term factor relationships.
             (b)  The “b” in the regression results for the historical betas increases as the length of the
                 trailing period increases. This indicates that the dispersion of historical betas decreases
                 as the trailing period increases, which also makes intuitive sense.

             (c)  All of the prediction results are better for the 60-day realized betas than for the 20-
                 day realized betas.
             (d)  The historical beta appears to have the largest relative advantage for trailing periods
                 of 240-300 days (for both the 20-day and the 60-day realized betas).
          3.   Implications: As mentioned previously, we should not rush to draw any hard conclusions
               from these results. A brief study such as this has its limitations, not the least of which is the
               fact that it uses less than four years worth of data. However, the evidence presented above
               supports the following claim: In recent years, a simple daily historical beta has been at least as
               good a predictor of short-term security betas as the predicted betas generated by a sophisticated
               multi-factor equity model.

               Since beta is such a primary feature of any equity factor model, this has implications for
               our investment process. It raises the question of how much we should rely on the numbers
               generated by multi-factor models for our risk controls. While these numbers are useful
               and should not be ignored, we can no longer claim that they are the best numbers available
               for this purpose. For risk-control purposes, the daily historical beta appears to be at least
               as important a measure as the multi-factor predicted beta.












                                            LOVELY PROFESSIONAL UNIVERSITY                                  305
   305   306   307   308   309   310   311   312   313   314   315