Investment-Cash Flow Sensitivity and Proxy Quality Thresholds

Abstract

We revisit the problem of error-laden proxies being used for 'q' in regressions of investment on cash flow and q. We give a menu of different prior information sets that can identify the sign of the coefficient on cash flow or the sign of the difference in cash-flow coefficients between groups of firms. The prior informations sets are arguably weaker than those necessary to identify the precise values of the coefficients. An element of each set is a prior lower bound on a correlation between the proxy and q. Like most researchers, we obtain positive OLS estimates of the coefficient on cash flow. We show, however, that the partial correlation between observed and true q must be believed to very large (from an objective viewpoint, unrealistically large) before one can legitimately infer that there is a positive coefficient on cash flow in the regression of investment on cash flow and true q.

Keywords: external finance constraints, investment, errors-in-variables, coefficient sign, prior information

JEL Classification: G31, E22, C40