Estimating and Testing Investment-Based Asset Pricing Models

The standard investment-based asset pricing model with homogeneous of degree one operating profit and adjustment cost functions predicts that a firm's stock return should be equal to physical capital investment return, state-by-state. Yet, previous work testing the model typically examines the weaker prediction that stock returns and investment returns should be equal on average, and focuses on simple specifications with one physical capital input and quadratic adjustment costs. We document that by following this approach, the implied time series R2 of the standard model is negative. We show how to incorporate both the model-implied time series and cross sectional restrictions in the estimation and testing of investment-based models using the generalized method of moments. Our approach uncovers a novel tradeoff between cross sectional and time series fit in the data: the baseline one-capital input investment-based model with quadratic adjustment costs cannot fit both sets of moments simultaneously. Perhaps surprisingly, even when only the time series moments are used in the estimation of the model to maximize its time series fit, the implied time series R2 of the model remains negative. Our approach can be extended to estimate and test non-homogeneous of degree one models in which the state-by-state equality between investment and stock returns does not hold. By incorporating a larger set of model predictions in the estimation and testing of investment-based models, our methodology can be useful to guide improvements in the specification of this class of models in future research