Application of self-exciting threshold autoregressive model on exchange rate in Nigeria: a comparative approach

Abstract

Exchange rates and many other financial time series data exhibit structural breaks and volatility. Nonlinearity test and a structural break test were used to detect the nonlinearity and the break date in NGN/EUR. The study revealed that nonlinearity and threshold nonlinearity exist in the exchange rate series. The results showed that the SETAR model can explain abrupt changes in NGN/EUR. The identified structural break date coincides with identifiable economic and political shocks. Given the evidence of structural break in the series, we applied unit root test and find that NGN/EUR is stationary which indicate that the ADF unit root tests are bias towards non-rejection of non-stationarity. In modelling the exchange rates data set, two SETAR models were generated, that is SETAR(2;5,2) model without dummy variable was used as a benchmark, while, dummy variable was added so as to address the identified structural break which generated SETAR(3;5,3). The diagnostic tests revealed that, the SETAR model is adequate for forecasting (i.e. both models are free from serial correlation and heteroscedasticity). The forecast results showed that the SETAR(3;5,3) model with the inclusion of dummy variable performs better than that of the SETAR(2;5,2) model without dummy variable.