However, the more complex our portfolio selection method is, the more we risk bringing in a data mining bias. I must confess that when we were doing the’ original BJS study, we tried things that do not appear in the published article. Moreover, we were reacting to prior work suggesting a relatively flat slope for the line relating average return to beta. Thus our article had elements of data mining too. To minimize the data mining problem, BJS used a very simple portfolio strategy. We chose securi- ties using historical estimates of beta, and we used many securities to diversifjr out the factors not related to beta. But t h s method does have flaws. For example, beta is hghly correlated with both total risk and resid- ual risk across stocks. So what we call the “beta factor” might better be called the “total risk factor” or the “residual risk factor.” I can’t think of any reliable way to distinguish among these. When doing the BJS study, we considered esti- mating the entire covariance matrix for our population of stocks, and using that to improve the efficiency of our test. We realized that this would require us to deal with uncertainty in our estimated covariances. We decided that the potential for improved efficiency was small, while the potential for error in our econometric methods was large. So we did not pursue that route. Others have used different methods to update our study. My view is that in the presence of data mining and estimate error and changing risk premi- ums, none of these methods adds enough accuracy to warrant its complexity. I view most of these methods as our method expressed in different language. For example, Fama and MacBeth  start with cross-sectional regressions of return on beta, and FALL 1993 THE JOURNAL O F PORTFOLIO MANAGEMENT 11 The Journal of Portfolio Management 1993.20.1:8-18. Downloaded from by NEW YORK UNIVERSITY on 02/20/15.It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.
Subscribe to view the full document.
The Journal of Portfolio Management 1993.20.1:8-18. Downloaded from by NEW YORK UNIVERSITY on 02/20/15.It is illegal to make unauthorized copies of this article, forward to an unauthorized user or to post electronically without Publisher permission.