AUTHOR=Lv Xueling , Xiong Xiong , Geng Baojun 

TITLE=Increasing the prediction performance of temporal convolution network using multimodal combination input: Evidence from the study on exchange rates

JOURNAL=Frontiers in Physics

VOLUME=Volume 10 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2022.1008445

DOI=10.3389/fphy.2022.1008445

ISSN=2296-424X

ABSTRACT=The exchange rate market is one of the most important financial markets in the world. The direction of exchange rate influences foreign trade and capital flows. This study presents a multimodal combination-based exchange rate market trend forecasting method to more accurately identify the volatility link between exchange rates, unlike the conventional exchange rate forecasting, in which only information related to itself is used as input. We select multiple exchange rate characteristics of other countries as input data and use the Pearson correlation coefficient and random forest model to filter these exchange rate characteristics. For trend forecasting, we combine a temporal convolutional network model with data with higher correlation as additional characteristics. For the experimental samples, we use the historical exchange rates of the Euro, Ruble, Australian dollar, and British pound corresponding to the Renminbi over a nine-year period. The experimental results show a more stable forecasting effect with the method proposed in this study than that with traditional models.