Forecasting with the Fokker-Planck model: Bayesian setting of parameter

Using a closed solution to a Fokker-Planck model of a time series, a probability distribution for the next point in the time series is developed. This probability distribution has one free parameter. Various Bayesian approaches to setting this parameter are tested by forecasting some real world time series. Results show a more than 25% reduction in the '95% point' of the probability distribution (the safety stock required in these real world situations), versus the conventional ARMA approach, without a significant increase in actuals exceeding this level.