Asset allocation is critically dependent on the ability to forecast the equity risk premium (ERP) out-of-sample. Constrained forecasting models have been recently introduced in order to improve the poor out-of-sample forecasting ability of macroeconomic variables like the dividend-price ratio. This paper critically investigates the nature of these constraints and their implications for dynamic asset allocation. Consistent with the existing evidence, such models improve the economic benefit for a mean-variance investor over a very long sample period (1947-2013). However, seen from a conditional viewpoint, we show that constrained models generate significant economic relative losses in periods of high volatility and market drawdowns, when it matters the most for asset allocators to retain assets and client base. Additionally, we find that risk-averse investors that face investment constraints –either by mandate or regulation– like short-selling or leverage constraints, can find little benefit in constrained ERP forecasting models, even across the business cycle. Our findings pose a significant challenge on the practical application of constrained ERP forecasting models and call for new model designs that actively incorporate some form of regime dependency.

Related Documents

Back to Top