Using structural break inference for forecasting time series

Abstract Rather than relying on a potentially poor point estimate of coefficient break date when forecasting, this paper proposes averaging forecasts over sub-samples indicated by confidence interval or set for the date. Further, we examine whether explicit consideration possible variance and use two-step methodology improves forecast accuracy compared with using heteroskedasticity robust inference. Our Monte Carlo results empirical application to US productivity growth show that likelihood ratio-based typically performs well in comparison other methods, while inference is particularly useful occurs concurrently after any break.