Dear Quantitative Finance Members,

I was wondering if you can clarify me the following issue. I am trying to estimate the cost of equity following "Industry costs of equity" (Fama and French, 1997). I am not sure if I correctly understood the steps that I need to follow. Here they are:

  1. Obtain firm returns (from CRSP database) and SMB, HML, Rm-rf from Fama and French website.
  2. Run $$R_i -r_f = \beta_0 + \beta_1(R_m-r_f) + \beta_2 SMB + \beta_3 HML$$ over each out of 48 industries, whole sample.
  3. Save estimated coefficients of $\beta_0, \beta_1, \beta_2, \beta_3$
  4. Estimate fitted Cost of Equity ($CE$) as $$CE = \beta_0 + \beta_1 \times (R_m-r_f) + \beta_2 \times (SMB) + \beta_3 \times (HML)$$

Please, correct me if I am wrong


1 Answer 1


First of all, the cost of equity is the expected returns on equity of a stock. This means that you could take any estimate of $\mathbb{E}(R_i)$ as cost of equity.

  • a first version would be the empirical average ${1\over D} \sum_{d=1}^D R_i(d)$ where $d$ are available days in your database.
  • to obtain a more robust version of the cost of equity you can rely on proxies of stable sources of returns.
  • if you believe in reliability of FF factors, you can do it this way:

    1. estimate the beta to the factors (like you propose in your question; you forgot the epsilon --i.e. residuals--): $$R_i -r_f = \beta_0 + \beta_1(R_m-r_f) + \beta_2 SMB + \beta_3 HML+\epsilon$$

    2. plug the empirical averages in place of the daily (or weekly) version of the data you used to estimate your beta (no more epsilon if you used any non-biased version of regression to obtain the betas): $$CE := \mathbb{E}(R_i) = r_f + \beta_0 + \beta_1(\mathbb{E}(R_m)-r_f) + \beta_2 \mathbb{E}(SMB) + \beta_3 \mathbb{E}(HML)$$

  • [EDIT] If you want to have one cost of equity by industry / sector, you can simply perform a regression within each sector, thus you will obtain $\mathbb{E}(\beta_i\vert Sector)$ in place of $\beta_i$ (for $i\in\{0,\ldots, 3\}$). As a consequence, you will replace $\mathbb{E}(R_i)$ by $\mathbb{E}(R_i\vert Sector)$. And it is what you want.

  • $\begingroup$ Thank you for the answer. One clarification, should I run main FF eq. within industry-year? Or just within industry? $\endgroup$ Jun 24, 2018 at 12:27
  • $\begingroup$ just within industry, otherwise your estimate of beta will be very noisy (250 points in a year). Betas should be stable over years. $\endgroup$
    – lehalle
    Jun 24, 2018 at 12:35
  • $\begingroup$ Thanks a lot. The very last question, please. SMB and HML (in the FF website) are constructed taking the whole sample of firm. Do I have to take these values or shall I reconstuct SMB and HML within industry? $\endgroup$ Jun 24, 2018 at 12:46
  • $\begingroup$ if you do one regression per industry, no real need to rebuild the portfolio, you can take theirs (I will update my answer). By the way: if you like my answer, please vote +1 on it ;{)} $\endgroup$
    – lehalle
    Jun 24, 2018 at 12:53
  • $\begingroup$ Sorry for asking late question...If I use monthly observations of returns, do I need to use monthly observations of FF factors? And if I estimate 5-years trailing Betas do I calculate cost of equity as beta_{t-1}*Market_premim_{t}, am I right? $\endgroup$ Sep 21, 2018 at 12:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.