Most recent empirical work implies that the presence of low-dimensional deterministic chaos increases the complexity of the financial time series behavior. In this study we propose the Generalized Multilayer Perceptron (GMLP), and the Bayesian inference via Markov Chain Monte Carlo (MCMC) method for parameter estimation and one-step-ahead prediction. By out-of-sample prediction approach, these proposed methods are compared to autoregressive integrated moving average (ARIMA) models which have been used as a benchmark. The deterministic Mackey-Glass equation with errors that follow an ARCH (p) process (MG-ARCH (p)) is applied to generate the data set used in this study. It turns out that GMLP outperforms the other two forecasting methods using RMSE, MAPE, and MAE criteria of forecasting accuracy.
M. Shahwan, T.
"A Comparison of Bayesian Methods and Artificial Neural Networks for Forecasting Chaotic Financial Time Series,"
Journal of Statistics Applications & Probability: Vol. 1:
2, Article 2.
Available at: https://digitalcommons.aaru.edu.jo/jsap/vol1/iss2/2