Neural Networks to Detect Nonlinearities in Time Series: Analysis of Business Cycle in France and the United Kingdom,

In this research we investigate possible existence of nonlinearities in business cycle fluctuations in France and United Kingdom (U.K.) real gross domestic product (GDP). We model the relationship between the real GDP in these countries using neural network linearity tests via in-sample as well as jackknife out-of-sample testing. Our results based on neural network linearity tests for possible existence of nonlinearities due to Terasvirta el al. (1993) using in-sample forecasts from neural nets in France and U.K. show statistically significant evidence of nonlinearities in both the series. Similarly, the results on linearity tests based on jackknife out-of-sample forecast also show statistically significant evidence of nonlinearities in both France and U.K. series. Moreover, the results based on neural network test for neglected nonlinearities that was proposed by Lee el al. (1993) also show statistically significant evidence of nonlinearities in both the countries. Therefore, policymakers are not able to evaluate the impact of monetary policy or any other shocks on output in these countries that are based on predictions from linear models.