A stock market crash alarm system based on a Hurst Index Neural Network

Stock market events are commonly modelled using a nor- mal distribution. This approach leaves out big heavy tails, and non- parametric methods such as neural networks can be an appropriate al- ternative approach to the study of non-normal behavior. However, learn- ing devices which are too generic in use may lead to non-ideal decisions, which result in nancial losses. To solve this problem, and de ne a de- fensive investment strategy, we use core method ideas for encoding two functions, moving average and Hurst exponent, which work as indicators over a nancial time series. The multilayer perceptron can then act as a warning system for stock market crashes, avoiding negative asset re- turns, and decreasing the investment risk. Simulation is performed using EURO STOXX 50 nancial index to illustrate the potential of such a strategy.