Cyclicality in the prices of risk: what more can we learn from explainable AI?

Abstract

We uncover the temporal patterns of the prices of risk through industry portfolios with varying sensitivities to the economic and financial cycles. Conditioning on the highs and lows of the cycles is key for statistical significance of the intertemporal component. Unlike market risk, its price decreases during an economic downturn but increases under tight funding conditions. Predictive machine learning models and their SHAP values suggest that a limited number of firm characteristics convey the most informative signals about asset risk premia. Valuation ratios are more important determinants for Cyclical relative to Defensive industries, whereas Return characteristics become crucial during recessions. Valuation Insight

The prices of risk that affect discount factors and present values are found to vary substantially over time depending separately on industry sensitivity to economic and financial cycles. Based on predictive machine learning models, the firm characteristics are uncovered that provide the most information about discount factors at the industry level. Valuation ratios are more important indicators of discount factors for cyclical industries than for defensive industries.