Imperfect alpha forecasts?

Think this is a new topic on the exam. CFAI Volume 6, Pages 483-485 “Suppose an analyst assigned to a security provides you with a forecast of alpha = 20%…” You then look at the quality of the analyst’s previous forecasts and apply a discount to alpha. [ok so far] The discount is worked out by looking at the ratio of explained (forecast) to total variance. [what the *&%^] Now I’m probably wrong but I have a problem with this: Rather like Sharpe Vs Sortino this seems to only look at overall variation rather than downside variation in forecasts. Hence if you had an outstanding analyst who always picked cheap stocks but took a conservative view of their alpha you would end up discounting their analysis, and underweighting these stocks in the portfolio. I would suggest that any quality factor should be able to exceed 1 if, say, an analyst’s track record is always to underestimate? Otherwise surely it would be better to look at the downside only? Ah…now I’m looking at this from a long only stand point. Is it more justifyable if you can go short? I don’t think so because analysts might well still be conservative with their negative alpha estimations couldn’t they? Is my understanding right, am I missing something and/or is this just something we need to learn and apply while disengaging brain?

DNAdiver, you have to use linear regression to estimate alpha and beta of a security. a measure of how well linear regression explains behavior of a security (or how close security prices can be approximated with a line alpha + beta*t) is called R^2 and equal the ratio of explained variance to total variance. For example, if your R^2 = 1, it means linear regression explains all variation of the security (linear regression can potentially be used well for alpha forecast), if R^2 = 0 (linear regression doesn’t help at all). now if you construct a confidence interval for alpha from its estimate, its going to be relatively tighter for higher R^2. http://en.wikipedia.org/wiki/Linear_regression

^ That’s right. Suppose that you had an analyst who consistently underestimated alpha for the stocks he followed so that if the analyst said the stocks alpha was rf + 5%, the expectation of alpha was > rf + 5%. What would you do? 1) Publish a paper disproving EMH - I can use this analyst’s ability to show that security prices are not unbiased. 2) Keep quiet, find all his high alpha estimates, borrow money like crazy and invest in all of them (note: I can hedge out all my systematic risk). 3) Disengage brain and talk about how this is like a Sortino ratio.