OUT-OF-SAMPLE FORECAST PERFORMANCE AS A TEST FOR NONLINEARITY IN TIME-SERIES

This article uses a local-information, near-neighbor forecasting metho dology as a prediction test for evidence of a noisy, chaotic data-gene rating process underlying the Divisia monetary-aggregate series. Using a nonparametric method known to perform well with low-dimensional cha otic processes infected by noise, accompanied by a robust test of fore cast performance evaluation, we compare out-of-sample forecasting accu racy from the local-information method to forecasting accuracy from th e best fitting global linear model. Our results fail to substantiate p revious claims for determinism in the Divisia monetary-aggregate serie s because the degree of forecast improvement obtained by the local-inf ormation method is not consistent with the hypothesis of a low-dimensi onal attractor underlying the Divisia data.