Forecasting Stock Indices: Stochastic and Artificial Neural Network Models

In recent years, there has been a bloom in the stock investors due to availability of various platforms that have provided an opportunity even for small scale investors to earn profits from the market. However, due to very high uncertainty, bad investments can lead to large financial losses and hence need for tools that can predict stock behaviour, arises. The main objective of this article is to provide a comparative empirical analysis of stochastic models with artificial neural networks in the prediction of stock indices across different markets. We consider three types of models, namely the time series models: autoregressive integrated moving average and autoregressive fractionally integrated moving average; jump diffusion models: Merton jump diffusion and Kou jump diffusion; the artificial neural network models: feed-forward network and the long short term memory. These models are used to forecast 10, 20 and 30 days ahead prices of major stock indices across different markets which include both developed and emerging economies. It is shown that the long short-term memory performs better than other considered models on most of the considered indices over all the time horizons. The results also indicate the forecasts provided by the LSTM model are significant from both statistical point of view and can possibly be used for profitable investments.