Semiparametric Portfolios : Improving Portfolio Performance by Exploiting Non-Linearities in Firm Characteristics

We present a semiparametric portfolio optimization method in which portfolio weights are parameterized as a non-linear function of firm characteristics. This approach generalizes the linear parametric portfolio policy of Brandt et al. (2009) and can be applied to high-dimensional problems at a relatively low computational cost. An empirical implementation exploiting the size, value, and momentum anomalies in the universe of CRSP stocks reveals that non-linearities as well as interaction effects are both important and complementary for the portfolio construction problem. Moreover, an out-of-sample evaluation indicates that the semiparametric strategies perform well in terms of returns, risk, and risk-adjusted returns both in the absence and in the presence of transaction costs. Our evidence suggests that allowing for a more flexible relation between portfolio weights and firm characteristics can provide a more accurate description of the empirical patterns seen in data