Multiscale SUR Estimation of Systematic Risk

We propose a multiscale version of the seemingly unrelated regressions model, based on wavelet transform-based time series observations. Each regression equation refers to a different time scale, which enables the use of across-scale error covariances in the feasible GLS estimation procedure for efficiency gains. We demonstrate the advantages of the proposed method over OLS with two studies: an empirical study using stock market returns for the main US industrial sectors and a detailed Monte Carlo simulation study with alternative wavelet filters. We also provide explanations for the suitability of the proposed method for estimating long-term systematic risk.