The Problem of Estimating Causal Relations Using Models Involving Accounting Semi-Identities

Researches will not draw the correct inferences about the values of coefficients estimated in a linear regression when both the dependent and the independent variable are part of an accounting identity (or accounting semi-identity close to a full one). The coefficients cannot tell us about a causal relation as the two variables will change together to satisfy the identity. A good example of a model that suffers from this problem is the Fazzari, Hubbard and Petersen (1988) investment-cash flow sensitivity model. The econometric problems in estimating accounting identities are not new, and have been highlighted in the debate about the Cobb-Douglas production function estimation. The causal influence on the dependent variable when variables do not form part of the accounting semi-identity can be estimated correctly if the independent variable in the accounting semi-identity leaves a large part of the dependent variable unexplained. A feasible way to solve the problem is to disaggregate the identity by replacing either the dependent or the independent variable by an appropriate substitute that is not part of the identity