Fame-French alpha for a single stock

Question

I want to study the impact of corporate culture on risk-adjusted stock returns. After quantifying corporate culture I wanted to use panel methodology (I have a sample of 100 S&P500 companies over 10 years) to study the question. As for the risk-adjusted returns I use Jensen's alpha from CAPM and Treynor measure and estimate those using rolling regressions of the previous 36-months returns for each month. After estimating monthly alphas for each stock, I calculate cumulative quarterly alphas ((1+r_m1)(1+r_m2)(1+r_m3)-1) and regress quarterly alphas on one-period lagged corporate culture measures. However, I also want to calculate alphas using Fama-French model. Can I do it for single stocks and proceed in the same fashion? I read journal articles and I found no paper where an author would calculate alpha for a single stock... everywhere portfolios are constructed. Can anyone explain to me the reason why constructing portfolios are favourable and assess my model? Does it make sense to calculate alphas for single stocks?

Thank you!

Answer 1

The reason for using Fama French for portfolios is generally that you try to quantify whether your anomaly/strategy etc. is actually capable of providing returns in excess of what could be achieved by passive exposure to the known risk factors included in the model. CAPM essentially does the same but only looks at the passive exposure to market index.

I don't know if there is necessarily any methodological problems with using Fama French model to estimate the alpha for a single stock, but I think it makes a lot of sense to consider if it is the right tool in your case.

I would carefully consider what you want to measure - I would imagine, that certain measures of corporate culture would be significantly correlated with some of the Fama French factors. Without knowledge of the list of variables you plan on investigating, I would still assume that smaller companies probably have very different cultural characteristics than larger companies, and same most likely applies to value factor (think e.g. Facebook vs. Morgan Stanley). Hence it might not make a lot of sense to capture this variation using FF model - you could, however, run the same analysis with subsamples using similar criteria as is used in construction Fama French factors if that interests you.