In this paper, we investigate the implications of non-separable preferences over durable and nondurable consumption for asset pricing tests when adjusting durable consumption is costly. In an economy without adjustment costs, in which a frictionless rental market exists for the durable good, the standard Euler equation with respect to nondurable consumption will hold for each individual agent as well as for aggregate data. If the adjustment of the durable good is costly, however, aggregation generally fails. We use aggregate data to find substantial deviations from the frictionless model, consistent with the presence of non-convex adjustment costs for the durable good. We also show how empirical asset pricing tests that use aggregate data can be affected by these deviations. We then propose and implement asset pricing tests that are robust to the presence of adjustment costs by relying on microeconomic data. Using household-level observations of nondurable food and durable housing consumption, our estimation results suggest that preferences are indeed non-separable in the two consumption goods and that reasonable structural parameters characterize agents' intertemporal utility optimizations.
IntroductionConsumption-based asset pricing models, despite their many empirical failures, have been at the heart of asset pricing research for over two decades. Naturally, the starting point of these models is a particular assumption on what consumption is and here researchers have, for the most part, restricted themselves to measuring consumption as consumption of nondurable goods and services. Recently though, there has been renewed interest in an old topic, namely the role of durable consumption in asset pricing tests.
1Introducing durable consumption into the agent's utility function brings new modelling issues to the asset pricing literature. One such issue is whether changing durable consumption is subject to adjustment costs and how these adjustment costs potentially affect asset prices. This question is particularly important if one uses aggregate data to test asset pricing implications.On the one hand, standard aggregation arguments will generally not hold if individual agents are subject to adjustment costs. Asset pricing tests using aggregate data therefore risk being misspecified if they do not account for consumption heterogeneity across individual agents.On the other hand, however, aggregation introduces smoothness that could make infrequent adjustments at the micro level irrelevant for understanding aggregate phenomena. In other words, macroeconomic data might behave as if generated by a representative agent in a frictionless economy. In this paper we therefore investigate whether asset pricing tests that use aggregate data can safely ignore adjustment costs, either because they are irrelevant at the individual level or, more likely, because frictions at the micro level do not affect the relationship between aggregate data and asset returns.Our tests rely on the fact that in a frictionless economy wit...