Stable estimation of Granger-causal factors of country-level innovation

Abstract

Increasing the innovativeness of a country is a known way to uplift its people, but how to increase the innovativeness of a country is not well understood. In this paper, we develop such an understanding by analyzing time series of global competitiveness index data and of development indicators data. Specifically, we estimate causative factors of innovation that countries can invest in or otherwise pursue through policy decisions. We take a sparse multivariate regression approach to find Granger-causal development indicators for an innovation index, which is based on group orthogonal matching pursuit. Due to high correlation between various indicators, small perturbations can cause the support of the sparse regression solution to change drastically with nonzeros shifting among correlated sets of indicators. Such behavior does not detract from predictive accuracy, but is undesirable for interpretation and decision making. To address this issue, we use randomization and stability selection techniques. We show favorable empirical results.