Wavelet Denoising and Attention-based RNN- ARIMA Model to Predict Forex Price

Abstract

Every change of trend in the forex market presents a great opportunity as well as a risk for investors. Accurate forecasting of forex prices is a crucial element in any effective hedging or speculation strategy. However, the complex nature of the forex market makes the predicting problem challenging, which has prompted extensive research from various academic disciplines. In this paper, a novel approach that integrates the wavelet denoising, Attention-based Recurrent Neural Network (ARNN), and Autoregressive Integrated Moving Average (ARIMA) are proposed. Wavelet transform removes the noise from the time series to stabilize the data structure. ARNN model captures the robust and non-linear relationships in the sequence and ARIMA can well fit the linear correlation of the sequential information. By hybridization of the three models, the methodology is capable of modelling dynamic systems such as the forex market. Our experiments on USD/JPY five-minute data outperforms the baseline methods. Root-Mean-Squared-Error (RMSE) of the hybrid approach was found to be 1.65 with a directional accuracy of ~76%.

A Survey of Forex and Stock Price Prediction Using Deep Learning

“…[27,35] It was clear that papers [24,57] achieved the best performance using the DNN model, having RMSEs smaller than 0.001. Papers [11,29,31,32,36,38,71,73] had great performance, while papers [27,35,37] did not have low RMSEs. Among all the papers, 33% of the CNN papers had an RMSE below 0.001, 33% of the CNN papers were in the range of 0.01-0.1, and 33% of the CNN papers were in the range of 0.1-1.…”

“…We found that paper [22] had the highest accuracy compared to any other papers, paper [16] had the highest return rate followed by papers [14,18,21]. Paper [24,31] had the lowest RMSE. Paper [19] had a higher F-measure than that of papers [14,18], but paper [18] achieved a higher Sharpe ratio than that of paper [14].…”

“…15 shows the details of each paper's RMSE based on the numerical range. In this table, papers [11,12,24] used the CNN-based model; papers [32,[35][36][37][38][39][45][46][47]57,65] used the LSTM-based model; papers [71][72][73] used the DNN-based model; papers [27,29,31] used the RNN-based model; and paper [86] used another deep learning method. [27,35] It was clear that papers [24,57] achieved the best performance using the DNN model, having RMSEs smaller than 0.001.…”

“…Z proposed a novel Attention-based RNN (ARNN), where wavelet denoised input was passed to ARNN. The forecast was calculated using the Autoregressive integrated moving average (ARIMA) and the output of the ARNN model [31].…”

“…Table 17 shows the details of the papers' MAE. In this table, papers [12,24] used the CNN-based model; papers [34,37,39,40,[42][43][44] used the LSTM-based model; papers [71,72] used the DNN-based model; and papers [29,31] used the RNN-based model. It could be found that paper [24] performed best, which used the CNN model.…”