Forecasting GDP Growth Using Genetic Programming

Abstract

Monetary policy affects the economy with long and variable lags, and for this reason policy-makers require reliable forecasts of economic activity. Hence, forecasts of real GDP growth have become more and more necessary. Haiming Guo (2006) proposed a new modified ARIMA model and used it to forecast the GDP growth of China from 1978 to 2004. Their experimental data show that the modified ARIMA model could provide more accurate forecasts than conventional ARIMA. However, all these models are linear. In this paper, we propose a new genetic programming method to forecast the GDP time series of China, United States and Japan from 1980 to 2006. Experimental results show that genetic programming yield statistically lower forecast errors for the year- over-year GDP data relative to modified linear ARIMA models. Moreover, we use the proposed method to forecast the future GDP growth of China, United States and Japan from 2007 to 2020, and we surprisingly find that the GDP of Japan fluctuates periodically, however the GDP of China and United States increases stably in the near future. According to the predicted data we can see that the GDP of China will exceed the GDP of Japan for the first time in 2020 or so.