IMF Working Papers

Under near-singularity conditions typically generated by persistence in current account data the predictions of present value models become extremely sensitive to small sample estimation error. Moreover, traditional Wald tests will distort the likelihood that the model is true. Using OECD data we find that: (i) the Wald test often leads to the wrong inference compared to a valid test; (ii) in all cases posterior distributions of the predicted series and associated correlation coefficients and variance ratios are very wide. In particular, one cannot draw any firm conclusion regarding excess current account volatility.