I am reading the paper and get the following question. I think here is how the regression is constructed:
- First step:
$R_t^i = \alpha^i + \beta^i \cdot MarketBeta_t + \gamma_i\cdot Size_t + \nu \cdot Value_t$
$t = 1....T, i = 1....N$
The first step, for each portfolio, regress the portfolio return $R_t$ on three factors, and get $\beta, \gamma, \nu$.
However, this portfolio is formed based on Size, Value. So, you would expect the loading to factors are pretty high.
- Second step: $E[R^i] = \beta^i \cdot RiskPremium1 + \gamma^i \cdot RiskPremium2 + \nu^i \cdot RiskPremium3$
with the second regression, you get the risk premium for each factor loading.
Here is what confuse me:
Using Value as an example, you form the portfolios based on Value, and find higher value has higher expected return. Then you regress portfolio returns on Value, of course the higher return has higher factor beta, and the higher factor beta has higher expected return.
Isn't the regression redundant? Isn't comparing the portfolio returns made the point?