Forecasting Non-Causal Processes

This research explores the forecast quality of a non-causal autoregressive model compared to existing causal models. The forecasted data consists of electricity, gas and oil prices. Different degrees of freedom in the error term of the non-causal model and different forecast horizons, up to 12 months, are tested. Overall the non-causal forecasts perform comparable to causal and vector autoregressive model forecasts according to Diebold Mariano tests for electricity and gas data. For oil data the non-causal forecasts perform worse. Lower degrees of freedom are favored in the non-causal models. Parameter estimations converge in the non-causal model, therefore the results are robust.