Novel hybrid model based on echo state neural network applied to the prediction of stock price return volatility

The prediction of stock price return volatilities is important for financial companies and investors to help measure managing market risk support decision-making. literature points out alternative models - such as the widely used heterogeneous autoregressive (HAR) specification which attempt forecast realized accurately. However, recent variants artificial neural networks, echo state network (ESN), a recurrent based on reservoir computing paradigm, have potential improving time series prediction. This paper proposes novel hybrid model that combines HAR specification, ESN, particle swarm optimization (PSO) metaheuristic, named HAR-PSO-ESN, feature design with power consistent PSO metaheuristic approach hyperparameters tuning. proposed benchmarked against existing specifications, integrated moving average (ARIMA), HAR, multilayer perceptron (MLP), in forecasting daily three Nasdaq (National Association Securities Dealers Automated Quotations) stocks, considering 1-day, 5-days, 21-days ahead horizons. predictions are evaluated terms r-squared mean squared error performance metrics, statistical comparison made through Friedman test followed by post-hoc Nemenyi test. Results show HAR-PSO-ESN produces more accurate most cases, an R2 (coefficient determination) 0.635, 0.510, 0.298, 5.78 × 10?8, 1.16 10?7, 1, 5, 21 days set, respectively. improvement statistically significant rank 1.44 different datasets