Abstract:Financial forecasting has been an important and active area of machine learning research because of the challenges it presents and the potential rewards that even minor improvements in prediction accuracy or forecasting may entail. Traditionally, financial forecasting has heavily relied on quantitative indicators and metrics derived from structured financial statements. Earnings conference call data, including text and audio, is an important source of unstructured data that has been used for various prediction tasks using deep earning and related approaches. However, current deep learning-based methods are limited in the way that they deal with numeric data; numbers are typically treated as plain-text tokens without taking advantage of their underlying numeric structure. This paper describes a numeric-oriented hierarchical transformer model to predict stock returns, and financial risk using multi-modal aligned earnings calls data by taking advantage of the different categories of numbers (monetary, temporal, percentages etc.) and their magnitude. We present the results of a comprehensive evaluation of NumHTML against several state-of-the-art baselines using a real-world publicly available dataset. The results indicate that NumHTML significantly outperforms the current state-of-the-art across a variety of evaluation metrics and that it has the potential to offer significant financial gains in a practical trading context.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve several key problems in financial prediction, especially in stock market prediction and financial risk management. Specifically, the author attempts to improve the accuracy of stock price movement prediction based on multi - modal data from earnings calls by improving the processing of numerical data. The following are the main objectives of this research: 1. **Improve numerical data processing**: - Traditional methods usually treat numbers as ordinary text tokens when processing numerical data, without fully utilizing their inherent numerical structures. This paper proposes a new numerical - oriented hierarchical Transformer model (NumHTML), which can better understand different categories of numerical values (such as currency, time, percentage, etc.) and their magnitudes. 2. **Use multi - modal data for prediction**: - Earnings calls contain unstructured data in both text and audio forms. These data contain important information about the company's future performance. In this paper, a multi - modal model is constructed by combining text and audio features to more accurately predict stock returns and financial risks. 3. **Introduce auxiliary tasks to improve the performance of the main task**: - In addition to the main task of stock price movement prediction, this paper also introduces volatility prediction as an auxiliary task. By using the Pareto multi - task learning algorithm, the model can significantly improve the accuracy of stock price movement prediction without sacrificing the performance of volatility prediction. 4. **Verify the effectiveness of the model**: - The author has carried out a comprehensive evaluation on a real - world public data set, proving that the NumHTML model performs well under multiple evaluation metrics compared with the existing state - of - the - art baseline models, and has significant potential for financial gains in the actual trading environment. ### Summary The main contribution of this paper is to propose a numerical - oriented hierarchical Transformer model (NumHTML). By improving the representation of numerical data and using multi - modal data, it significantly improves the accuracy of stock price movement prediction based on earnings calls. In addition, by introducing volatility prediction as an auxiliary task, the predictive ability of the model is further enhanced.

NumHTML: Numeric-Oriented Hierarchical Transformer Model for Multi-task Financial Forecasting

Hidformer: Hierarchical Dual-Tower Transformer Using Multi-Scale Mergence for Long-Term Time Series Forecasting

Numeral Understanding in Financial Tweets for Fine-Grained Crowd-Based Forecasting

Accurate Stock Price Forecasting Based on Deep Learning and Hierarchical Frequency Decomposition

Stock Price Forecasting with Empirical Mode Decomposition Based Ensemble \(\nu \)-Support Vector Regression Model

A Novel Hybrid Model (EMD-TI-LSTM) for Enhanced Financial Forecasting with Machine Learning

HDML: hybrid data-driven multi-task learning for China's stock price forecast

Enhancing Time Series Forecasting: A Hierarchical Transformer with Probabilistic Decomposition Representation

Background-aware Multi-source Fusion Financial Trend Forecasting Mechanism

F-HMTC: Detecting Financial Events for Investment Decisions Based on Neural Hierarchical Multi-Label Text Classification

Multi-task Learning for Financial Forecasting

Modality-aware Transformer for Financial Time series Forecasting

Ensemble of temporal Transformers for financial time series

Multi-source Transfer Learning with Ensemble for Financial Time Series Forecasting

Research and Application of a Novel Hybrid Model Based on a Deep Neural Network Combined with Fuzzy Time Series for Energy Forecasting

NumLLM: Numeric-Sensitive Large Language Model for Chinese Finance

Revolutionizing Bitcoin price forecasts: A comparative study of advanced hybrid deep learning architectures

A novel hierarchical feature selection with local shuffling and models reweighting for stock price forecasting

Forecasting price in a new hybrid neural network model with machine learning

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

A novel hybrid model for stock price forecasting integrating Encoder Forest and Informer