Time-Series Tests of a Non-Expected-Utility Model of Asset Pricing

This paper provides two alternative estimation and testing procedures of a representative-agent model of asset pricing which relies on a particular parametrization of non-expected-utility preferences. The first is based on maximum-likelihood estimates, supplemented with an explicit model of time varying first and second moments (where the time-variation of second moments in modelled with an ARCH-Autoregressive Conditionally Heteroskedastic-process); the second is based on generalized-method-of moments estimates. We perform our tests on a data set that includes monthly observations of rates of return on US stock prices and US consumption of nondurables and services. Our results are directly comparable to a test of the dynamic capital asset pricing model performed by Hansen and Singleton (1983), and to a recent test of the model studied here performed by Epstein and Zin (1989).