Abstract
The Feldstein–Horioka puzzle has triggered a broad range of econometric specifications to investigate saving-investment (SI) relations. In this article, we attempt to determine a family of econometric models that is most suitable in explaining actual ratios of domestic investment to GDP via cross-validation techniques. Comparing between, pooled, time and country dependent specifications of the SI relation, the country dependent model is best performing. Moreover, error correction models formalizing adjustment dynamics of domestic investment ratios are markedly outperformed by static panel models. Supporting evidence for a cointegration relation between domestic saving and investment ratios is not found.
Notes
1 Apart from model comparison by means of absolute forecast errors we also consider CV criteria derived from squared forecast errors. The results are similar as those from (3).