Working Paper: NBER ID: w14543
Authors: George M. Constantinides; Anisha Ghosh
Abstract: A novel methodology in testing the long-run risks model of Bansal and Yaron (2004) is presented based on the observation that, under the null, the potentially latent state variables, "long-run risk" and the conditional variance of its innovation, are known a¢ ne functions of the observable market-wide price-dividend ratio and risk free rate. In linear forecasting regressions of consumption growth and returns by the price-dividend ratio and risk free rate, the model implies much higher forecastability than what is observed in the data over 1931 -2009. The co-integrated variant of the model by Bansal, Gallant, and Tauchen (2007), also implies much higher forecastability of returns than what is observed in the data. Finally, we reject the models' implications in jointly pricing the cross-section of returns and fitting the unconditional time series moments of consumption and dividend growth. The results suggest that either some important state variable is missing or that the models should be generalized in a way that the lagged price-dividend ratio and risk free enter the regressions in a non-linear fashion.
Keywords: Asset Pricing; Long Run Risks; Consumption Growth
JEL Codes: G12
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Long-run risk model (G17) | higher forecastability of consumption growth (E27) |
| Long-run risk model (G17) | higher forecastability of market returns (G17) |
| price-dividend ratio and risk-free rate (G12) | expected market return (G17) |
| price-dividend ratio and risk-free rate (G12) | equity premium (G12) |
| price-dividend ratio and risk-free rate (G12) | dividend growth (G35) |
| price-dividend ratio and risk-free rate (G12) | consumption growth (E20) |
| Long-run risk model (G17) | rejection of original LRR model (C52) |
| Long-run risk model (G17) | rejection of cointegrated variant (C32) |
| missing state variable or model generalization (C20) | rejection of models (C52) |