Abstract:Many deep learning works on financial time-series forecasting focus on predicting future prices/returns of individual assets with numerical price-related information for trading, and hence propose models designed for univariate, single task and/or unimodal settings. Forecasting for investment and risk management involves multiple tasks in multivariate settings: forecasts of expected returns and risks of assets in portfolios, and correlations between these assets. As different sources/types of time-series influence future returns, risks and correlations of assets in different ways, it is also important to capture time-series from different modalities. Hence, this paper addresses financial time-series forecasting for investment and risk management in a multivariate, multitask and multimodal setting. Financial time-series forecasting is however challenging due to the low signal-to-noise ratios typical in financial time-series, and as intra-series and inter-series relationships of assets evolve across time. To address these challenges, our proposed Temporal Implicit Multimodal Network (TIME) model learns implicit inter-series relationship networks between assets from multimodal financial time-series at multiple time-steps adaptively. TIME then uses dynamic network and temporal encoding modules to jointly capture such evolving relationships, multimodal financial time-series and temporal representations. Our experiments show that TIME outperforms other state-of-the-art models on multiple forecasting tasks and investment and risk management applications.

What problem does this paper attempt to address?

The paper aims to address the issues of multivariate, multi-task, and multimodal settings in financial time series forecasting to support investment and risk management decisions. Specifically, the study focuses on the following points: 1. **Multivariate, Multi-task, and Multimodal Settings**: Traditional financial time series forecasting work usually focuses on predicting the price or return of a single asset and is based on a single modality (typically numerical price-related information) to support trading decisions. In contrast, investment and risk management require predicting the expected returns and risks (volatility) of multiple assets, as well as the correlations between these assets, which involves multivariate, multi-task, and multimodal settings. 2. **Challenges of Non-stationarity**: Financial time series forecasting faces challenges including low signal-to-noise ratio and the non-stationarity of relationships within and between sequences over time. For example, the correlation between different assets may increase with high volatility and declining returns, and decrease with low volatility and stable returns. 3. **Learning Dynamic Implicit Networks**: To address the above challenges, the paper proposes a new model framework called "Temporal Implicit Multimodal Network (TIME)." This model can adaptively discover dynamic implicit relationship networks from multimodal financial time series and use these networks for prediction. 4. **Key Contributions**: - Proposed the first adaptive implicit relationship network discovery method for multiple financial forecasting tasks, supporting investment and risk management; - Designed an attention mechanism-based module that can adaptively discover implicit relationship networks at multiple time points and predict correlations between assets; - Proposed a temporal vectorization module with various functional forms that can adaptively capture temporal patterns for time-sensitive dot-product attention sequence encoding; - Trained the model on multiple related tasks, utilizing complementary information to improve overall prediction performance and reduce the risk of overfitting; - Designed a general model that can be applied to different types and quantities of modality data and generate representations in a shared embedding space. Through the above methods, the paper aims to provide a solution capable of effectively handling the complexity and diversity in financial time series forecasting to support practical investment and risk management applications.

Temporal Implicit Multimodal Networks for Investment and Risk Management

Temporal Multimodal Multivariate Learning

Investment and Risk Management with Online News and Heterogeneous Networks

Financial time series forecasting with multi-modality graph neural network

Bayesian dynamic financial networks with time-varying predictors

Multimodal deep learning for finance: integrating and forecasting international stock markets

Multi-task Learning for Financial Forecasting

Spatiotemporal Adaptive Neural Network for Long-term Forecasting of Financial Time Series

Dynamic dependence networks: Financial time series forecasting and portfolio decisions (with discussion)

Temporal-spatial dependencies enhanced deep learning model for time series forecast

Low-Rank Temporal Attention-Augmented Bilinear Network for financial time-series forecasting

Multi-Modal Financial Time-Series Retrieval Through Latent Space Projections

Text2TimeSeries: Enhancing Financial Forecasting through Time Series Prediction Updates with Event-Driven Insights from Large Language Models

Masked Multi-Step Multivariate Time Series Forecasting with Future Information

Learning Dynamic Multimodal Network Slot Concepts from the Web for Forecasting Environmental, Social and Governance Ratings

MMFNet: Multi-Scale Frequency Masking Neural Network for Multivariate Time Series Forecasting

Rating Multi-Modal Time-Series Forecasting Models (MM-TSFM) for Robustness Through a Causal Lens

"The Squawk Bot": Joint Learning of Time Series and Text Data Modalities for Automated Financial Information Filtering

Agree to Disagree: Personalized Temporal Embedding and Routing for Stock Forecast

TV-Net: Temporal-Variable feature harmonizing Network for multivariate time series classification and interpretation

Deep Video Prediction for Time Series Forecasting