Dhaka Stock Market analysis with ARIMA-LSTM Hybrid Model

Abstract

This paper proposes the forecasting of correlation coe cients of Dhaka Stock Ex- change market assets required for portfolio optimization using an ARIMA-LSTM hybrid model. We have developed a robust model that encompasses both linearity and non-linearity within the datasets of the Dhaka stock market with a hybrid com- bining ARIMA model and a Recurrent Neural Network called LSTM. Our hybrid model tries to utilize the unique properties of both the ARIMA model and the LSTM model. We have ltered the linear components in the datasets using the ARIMA model and passed the residuals obtained onto the LSTM model which deals with the nonlinear components and random errors. We have compared the empirical results of this model with several other traditional statistical models used in portfolio man- agement namely the Single Index model, Constant Correlation model and Historical Model. We have also predicted the correlation coe cients using the ARIMA model to see how one of the model in our hybrid performs individually. The test results show that the hybrid model excels the other models in accuracy and indicates that the ARIMA-LSTM hybrid model can be an e ective way of predicting correlation coe cients required for portfolio optimization.