Improving exchange rate forecasting via a new deep multimodal fusion model

Area of Science:

• Computational finance
• Machine learning for time series analysis
• Financial econometrics

Background:

• Previous exchange rate prediction models relying solely on market indicators yielded unsatisfactory results.
• The complex interplay of diverse information types and their couplings significantly influences exchange rates.
• Existing models often fail to adequately capture the relationship between market data and investor sentiment.

Purpose of the Study:

• To develop an innovative multimodal fusion-based long short-term memory (MF-LSTM) model for forecasting the USD/CNY exchange rate.
• To enhance prediction accuracy by integrating both market indicators and investor sentiment.
• To demonstrate the effectiveness of multimodal fusion in financial time series forecasting.

Main Methods:

• A novel MF-LSTM model was developed, featuring parallel LSTM modules for feature extraction from distinct data modalities.
• A shared representation layer was employed to effectively fuse these extracted features.
• Sentiment analysis on social media microblogs was performed using bidirectional encoder representations from transformers (BERT) for the text modality.

Main Results:

• The proposed MF-LSTM model demonstrated superior performance compared to nine baseline algorithms over a 15-month experimental period.
• Experimental results confirmed the effectiveness of treating market indicators and investor sentiments distinctly based on their unique characteristics.
• The deep coupled multimodal fusion approach proved more effective than shallow models in capturing inter-modal couplings.

Conclusions:

• Incorporating multimodal fusion into financial time series forecasting is both practicable and effective.
• The MF-LSTM model offers a significant advancement in exchange rate prediction accuracy by leveraging diverse data sources.
• This research highlights the importance of considering investor sentiment alongside traditional market indicators for more robust financial forecasting.