Conclusions

---

This paper proposes and illustrates some innovations in the combination of forecasts. A new basic approach to consensus of individual forecasts was based on casting the problem as estimation of the mode of the density of expert estimates. This approach relaxes the assumption of normally distributed forecast errors and permits nonsymmetric loss functions. For the needed mode estimation, the Markov chain Monte Carlo technique was applied. This is a Bayesian technique necessitated by the typically small sample of individual method forecasts available for the mode estimation step. The effectiveness of the method was illustrated in comparison to popular existing combination methods. The results show that the proposed approach is clearly competitive with respect to RSME error measures for the example basic time series.

Figure 2: Weibull probability plot for the arrival of U.S. citizens from foreign travel in 1978

---