Macro Variables Predictions using three Classifiers: KNN, Sigmoid and Softmax

In this paper, we have predicted five macro variables, namely the equity market, the foreign exchange rate, the central bank’s policy rate, the GDP growth rate, and the inflation rate, using three Classifiers, K-Nearest Neighbor, Sigmoid and Softmax, for nine different markets, from different geographical regions, in order to generalize and present results that are not specific to a country, nor region, nor model. Our aim was to identify common trends in macro variables behaviors, from the perspective of the predictive power of the variables. We have applied as well a Weighted KNN approach and compared its prediction accuracy to the simple KNN, in addition, we have performed a behavior analysis of K in KNN. Finally, we have performed a sensitivity analysis on the Sigmoid hidden nodes versus the error term. We found out that the yield curve on its own had valuable information about long term equity markets, and macro variables seemed to hold predictive power over the future long-term behavior of the FX. In addition, the Policy Rate seemed to be affected by macro variables in the short term, however, in the long term it was more affected by the yield curve. Moreover, the inflation rate and GDP seemed to be dominated by their own past values, the Autoregressive process. The best long-term predictions accuracies were achieved by the equity indices and inflation rates. On the other hand, the best short-term prediction accuracy was achieved by the Policy Rate. Finally, the FX rate was the most unpredictable value, as its average prediction accuracy was the worst for all forecasted horizons.