Abstract:
|
Financial risk premia have been heavily studied in the academic literature. Recently, Gagliardini et al. (2016) have developed an econometric framework to infer the path of risk premia from large unbalanced panels of stock return. Following on the recent interest in selecting relevant factor exposures and characteristics, we extend the above methodology in the framework of penalized two-pass regression. Building on the framework of Jacob et al. (2009), we use an overlapping group-LASSO technique in the first- pass regression to select the relevant factors. Therefore, our method allows to select, at the stock level, common and asset specific instruments while at the same time ensuring that the select models fulfil the no-arbitrage condition(s). Then, in the second-pass regression, we apply the adaptive LASSO technique of Zou (2006) to provide a sparse estimator of the risk premia. We showed that the proposed approach selects the relevant instruments and risk premia with a probability going to one asymptotically. Finally, we illustrate the benefits of the proposed methodology to a data-set of US stocks from July 1964 to December 2019.
|