Evaluating the forecasting performance of linear and nonlinear monetary policy rules for South Africa

This paper compares forecast performance of linear and nonlinear monetary policy rules using South African data. Recursive forecasts values are computed for 1- to 12-steps ahead for the out-of-sample period 2006:01 to 2010:12. For the nonlinear models we use bootstrap method for multi-step ahead forecasts as opposed to point forecasts approach used for linear models. The aim of the paper is to evaluate the performance of three competing models in an out-of-sample forecasting exercise. Overall ranking reveals the superiority of the nonlinear model that distinguishes between downward and upward movements in the business cycles in closely matching the historical record. As such, forecasting performance tests reveal that the South African Reserve bank pays particular attention to business cycles movements when setting its policy rate.