Neural net multidimensional nonlinear prediction technologies are applied to forecasting financial return distributions. This paper reports on neural net prediction technologies applied to data on bilateral gamma distributional parameters obtained from time series and option data. The exercises conducted consider predicting longer maturity risk neutral distributions from their shorter maturity counterparts, predicting tomorrow’s option surface from today’s as synthesized by the Sato process based on the bilateral gamma density at unit time, and physical return distribution of a stock return from the physical distribution for the exchange traded fund for the stock’s sector. It is observed that considerable improvements are offered in predicting longer maturity risk neutral distributions at the lower spectrum of the set of maturities. Improvements are not as pronounced for tomorrow’s surface from today’s. There are improvements for index component physical return density prediction from the distribution of the index.
Madan et al. (Wed,) studied this question.