Non-parametric estimation of marginal propensities to consume: The case of regression splines

We investigate a non-parametric method to estimating marginal propensities to consume (MPC) using regression splines. This approach complements existing methods by relaxing a number of strong requirements on the part of the theory, otherwise necessary to acquire identification. Specifically, we show that a regression spline of consumption on income and assets yields a surprisingly precise estimate of the MPC at the household-level using only cross-sectional data. As a tradeoff on the part of the regression theory, the proposed method is somewhat data intensive. We use our proposed method to estimate the marginal propensity to consume for each household in Denmark, using detailed tax records on income and wealth. We estimate an aggregate MPC out of liquid assets of 49%, but with considerable heterogeneity over the wealth distribution. For example, we find the difference between the estimated MPC for the average household in the 2nd income decile compared to the 8th income decile to be more than 30 percentage points. Results indicate that it may be important to take the heterogeneity of MPC across households into account when assessing the impacts of proposed fiscal policies or changes in long interest rates.