Option Strategies and Neural Networks by Thomas B. Rubino Jr.
Confidence intervals are a statistical tool that describes the accuracy of a prediction, while neural networks provide predictions. This article combines the two to define the risk in a trade.
In recent years, neural networks have gained acceptance in solving problems related to portfolio management and market timing. Traders trust these models because they are accurate; nevertheless, many traders still view neural nets as a black box where data goes in one end and a prediction comes out the other. Now it's time to shine some light into that box, gain additional information from your neural nets, and use that information to build successful trading strategies.
Statistical tools that describe the accuracy of a prediction are particularly useful to traders. Confidence intervals, like the outermost ring on a marksman's target, place boundaries on a prediction's error. While it would be nice to hit the bull's-eye, there's a far greater probability of hitting the target somewhere inside the outermost ring. Confidence intervals are even more sophisticated, because it is possible to assign a probability to the actual value falling within the confidence interva .
Statistics literature often employs 95% confidence intervals, meaning 95 out of 100 actual observations should fall within the interval. These tools have long been applied to linear techniques in order to evaluate forecasts. The theory behind confidence intervals for nonlinear regression is also well known. Since neural networks are a class of nonlinear regression models, it is possible to create confidence intervals for neural nets.