Approaching reverse causality with a Two-Stages Panel VAR: an application to growth and institutions

Sammanfattning: A number of florid academic debates in the field of empirical economics are characterized by vivid discussions around the issue of reverse causality. One of these is the discourse around the relationship between institutional quality and growth. This thesis attempts to propose a flexible two-stages econometric procedure for the incorporation of reverse causality in the empirical analysis and for the empirical study of path dependence through IRA, and it presents an application to the institutions-growth relationship. Whereas the first stage is aimed at demeaning the variables of interest of measurable time-and-space variation and of both time and country fixed effects, the second estimates a Panel Vector Autoregressive model on the demeaned variables by simple OLS: thanks to the presence of the first stage, results are robust to the inclusion of a pooled intercept, which remains statistically insignificant, and to the misspecification of causal priority when making identifying restrictions. It is also found that previous attempts to estimate the institutions-growth relationship by means of a PVAR - without the inclusion of a first stage greatly overestimate both the magnitude of parameters and the persistence of path dependence, as represented by Impulse Response Functions, due to omitted variable bias. Finally, it is recognized that the current computational possibilities of statistical softwares with respect to PVARs pose quite strict limitations to the application of the proposed procedure, and a future project aimed at programming more flexible packages is envisaged.