We examine forecasting performance of the recent fractionally cointegrated vector autoregressive (FCVAR) model. The model is applied to daily polling data of political support in the United Kingdom for 2010 - 2015. We compare with popular competing models and at various forecast horizons. Our findings show that the precision of fore- casts generated by the FCVAR model is better than all multivariate and univariate models in the portfolio, and the four variants of the FCVAR model considered are generally ranked as the top four models in terms of forecast accuracy. Furthermore, the FCVAR model significantly outperforms the standard cointegrated VAR (CVAR) model at all forecast horizons and the relative forecast improvement is highest at longer forecast horizons, where the root mean squared forecast error of the FCVAR model is up to 20% lower than that of the CVAR benchmark model. In an empirical application to the prediction of vote shares in the 2015 UK general election, forecasts generated by the FCVAR model leading into the election appear to provide a more informative assessment of the current state of public opinion on electoral support than that suggested by the hung government prediction of the opinion poll. Specifically, the FCVAR model projects the correct direction for the realized vote shares in the election for both the Conservative and Labour parties.
forecasting fractional cointegration opinion poll data vector autoregressive model