Puzzling Out Feldstein-Horioka: an Extensive Analysis using Time Varying Parameter Models

The puzzle of Feldstein and Horioka (1980) stems from the early eighties: the authors identified the correlation between a country's savings-to-GDP and investment-to-GDP ratio as a measure of international capital mobility. By using a cross-section analysis they concluded that the long-awaited capital mobility was not significantly present. The puzzle rose from the fact that the seminal article inspired a vast amount of literature with different research methods and data samples, but without a real wide consensus. We argue that the last 30 years brought a dramatic change in capital mobility as trade liberalization and information technology leaped forward, and it's this evolution which may have caused the mixed research results in the literature. We therefore suggest using time varying parameter (TVP) models which allow us to control on the structural change which possibly took place during the second half of the last century.In the literature there are more suggested ways for identifying a TVP model on the savings-investment puzzle - we try to estimate all of them and also have a new own idea to try out. When taking the time dimension of savings and investment data into account, the Feldstein-Horioka question can be interpreted simply as a cointegration problem between the two variables. One TVP method naturally arises: let's make directly the elements of the cointegrating equation (CE) time-varying (as used in Arisoy, 2010). Another, maybe more sophisticated solution is to have a constant CE but make the error correction term (ECT) time varying (as in Papapetrou, 2006). Both share the feature that they do not connect the estimations on different countries. Our idea is to build a model which identifies a common component in the degree of capital mobility among the analysed countries, and thus it will be estimated on all countries at once. In all cases we use latent variable state-space model setups and estimate them with the Kalman Filter and Maximum Likelihood which also gives us confidence intervals to the estimated sequences.Our hypothesis is that the degree of capital mobility has risen significantly in the last 50 years, possibly with an acceleration around the turn of the century. However, because of a high dispersion among specific countries, we expect this result to hold only on averages.