Measuring Intertemporal Substitution: The Importance of Method Choices and Selective Reporting

Abstract

I examine 2,735 estimates of the elasticity of intertemporal substitution in consumption (EIS) reported in 169 published studies. The literature shows strong selective reporting: researchers discard negative and insignificant estimates too often, which pulls the mean estimate up by about 0.5. The reporting bias dwarfs the effects of methods, with the exception of the choice between micro and macro data. When I correct the mean for the bias, for macro estimates I get zero, even though the reported t‐statistics are on average two. The corrected mean of micro estimates of the EIS for asset holders is around 0.3–0.4. Calibrations greater than 0.8 are inconsistent with the bulk of the empirical evidence.

Subjective Intertemporal Substitution

“…In general, these forecast errors will be correlated with rr t+1 , requiring the use of instruments to estimate the EIS, and hence generating a host of econometric problems that the literature has not fully been able to address. Partly as a result of these problems, as well as of the limitations in the available consumption data, the available estimates of this important parameter range from close to 0 to well above 1, as nicely illustrated in the meta study of Havranek (2015).…”

“…One of the challenges with this GMM approach is that it assumes that agents' expectations make rational use of the same conditioning information available to the econometrician, so as to produce forecast errors that are orthogonal to that information. Therefore, traditional tests of the Euler equation cannot be disentangled from auxiliary assumptions on the expectation formation process, resulting in estimates of the EIS that are quite sensitive to the instruments used to build the expectations, as well as to other details of the specification (see Havranek (2015) for a comprehensive meta study).…”

Natural Resources and Economic Growth: A Meta-Analysis

“…Suppose, for example, that a specific method choice made by the authors of primary studies affects both the standard error and the reported point estimate in the same direction. Then the standard error variable will be correlated with the error term, and we obtain a biased estimate of β 1 (Havranek, 2015). A solution is to use an instrument for the standard error that is correlated with the standard error but not with method choices.…”

Habit formation in consumption: A meta-analysis

“…For example, Havranek (2015) finds strong publication bias in the literature that uses consumption Euler equations to estimate the elasticity of intertemporal substitution (often the same specification used to estimate habit formation). Most economists believe that the elasticity of substitution should be positive because negative elasticity implies a convex utility function.…”