Nonconvex Adjustment Cost in Lumpy Investment Models: Mean versus Variance

Abstract

This paper revisits the canonical assumption of nonconvex capital adjustment costs in lumpy investment models as in Khan and Thomas (2008), which are assumed to follow a uniform distribution from zero to an upper bound, without distinguishing between the mean and the variance of the distribution. Unlike the usual claim that the upper bound stands for the size (represented by the mean) of a nonconvex cost, I show that in order to generate an empirically consistent interest elasticity of aggregate investment, both a sizable mean and a sizable variance are necessary. The mean governs the importance of the extensive margin in aggregate investment dynamics, while the variance governs how sensitive the extensive margin is to changes in the real interest rate. As a result, both the mean and the variance are quantitatively important for aggregate investment dynamics.