AbstractThis paper explores the cross-sectional variation of expected returns for a large cross section of industry and size/book-to-market portfolios. We employ mixed data sampling (MIDAS) to estimate a portfolio’s conditional beta with the market and with alternative risk factors and innovations to well-known macroeconomic variables. The market risk premium is positive and significant, and the result is robust to alternative asset pricing specifications and model misspecification. However, the traditional 2-pass ordinary least squares (OLS) cross-sectional regressions produce an estimate of the market risk premium that is negative, and significantly different from 0. Using alternative procedures, we compare both beta estimators. We conclude that beta estimates under MIDAS present lower mean absolute forecasting errors and generate better out-of-sample performance of the optimized portfolios relative to OLS betas.
Forecasting the Market Risk Premium with Artificial Neural Networks