Missing Aggregate Dynamics: On the Slow Convergence of Lumpy Adjustment Models

Abstract

The dynamic response of aggregate variables to shocks is one of the central concerns of applied macroeconomics. The main measurement procedure for these dynamics consists of estimating an ARMA or VAR (VARs, for short). In non- or semi-structural approaches, the characterization of dynamics stops there. In other, more structural approaches, researcher try to uncover underlying adjustment cost parameters from the estimated VARs. Yet, in others, such as in RBC models, these estimates are used as the benchmark over which the success of the calibration exercise, and the need for further theorizing, is assessed. The main point of this paper is that when the microeconomic adjustment underlying the corresponding aggregates is lumpy, conventional VARs procedures are often inadequate for all of the above practices. In particular, the researcher will conclude that there is less persistence in the response of aggregate variables to aggregate shocks than there really is. Paradoxically, while idiosyncratic productivity and demand shocks smooth away microeconomic non-convexities and are often used as a justification for approximating aggregate dynamics with linear models, their presence exacerbate the bias. Since in practice idiosyncratic uncertainty is many times larger than aggregate uncertainty, we conclude that the problem of missing aggregate dynamics is prevalent in empirical and quantitative macroeconomic research.

Keywords: Speed of adjustment, Discrete adjustment, Lumpy adjustment, Aggregation, Calvo model, ARMA process, Partial adjustment, Expected response time, Monetary policy, Investment, Labor demand, Sticky prices, Idiosyncratic shocks, Impulse response function, Time-to-build

JEL Classification: C22, C43, D2, E2, E5