Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Global Neural Networks and The Data Scaling Effect in Financial Time Series Forecasting

Chen Liu, Minh-Ngoc Tran, Chao Wang, Richard Gerlach, Robert Kohn

TL;DR

This paper investigates volatility forecasting in finance using global training of neural networks, arguing that local, per-series training fails to exploit cross-sectional information and data diversity. By pooling thousands of stock time series, global NNs—especially an LSTM-based universal volatility model—achieve superior out-of-sample accuracy and robust performance in zero-shot predictions, even with limited historical data (as little as 12 months). The study demonstrates data size and diversity drive gains, shows global NNs can learn key volatility stylized facts and are resilient to outliers, and provides a comprehensive evaluation framework spanning statistical and economic metrics. Practically, the work advocates for a data-centric shift toward global estimation, offering scalable, interpretable, and reliable volatility and risk forecasts for portfolios and risk management applications.

Abstract

Neural networks have revolutionized many empirical fields, yet their application to financial time series forecasting remains controversial. In this study, we demonstrate that the conventional practice of estimating models locally in data-scarce environments may underlie the mixed empirical performance observed in prior work. By focusing on volatility forecasting, we employ a dataset comprising over 10,000 global stocks and implement a global estimation strategy that pools information across cross-sections. Our econometric analysis reveals that forecasting accuracy improves markedly as the training dataset becomes larger and more heterogeneous. Notably, even with as little as 12 months of data, globally trained networks deliver robust predictions for individual stocks and portfolios that are not even in the training dataset. Furthermore, our interpretation of the model dynamics shows that these networks not only capture key stylized facts of volatility but also exhibit resilience to outliers and rapid adaptation to market regime changes. These findings underscore the importance of leveraging extensive and diverse datasets in financial forecasting and advocate for a shift from traditional local training approaches to integrated global estimation methods.

Global Neural Networks and The Data Scaling Effect in Financial Time Series Forecasting

TL;DR

This paper investigates volatility forecasting in finance using global training of neural networks, arguing that local, per-series training fails to exploit cross-sectional information and data diversity. By pooling thousands of stock time series, global NNs—especially an LSTM-based universal volatility model—achieve superior out-of-sample accuracy and robust performance in zero-shot predictions, even with limited historical data (as little as 12 months). The study demonstrates data size and diversity drive gains, shows global NNs can learn key volatility stylized facts and are resilient to outliers, and provides a comprehensive evaluation framework spanning statistical and economic metrics. Practically, the work advocates for a data-centric shift toward global estimation, offering scalable, interpretable, and reliable volatility and risk forecasts for portfolios and risk management applications.

Abstract

Neural networks have revolutionized many empirical fields, yet their application to financial time series forecasting remains controversial. In this study, we demonstrate that the conventional practice of estimating models locally in data-scarce environments may underlie the mixed empirical performance observed in prior work. By focusing on volatility forecasting, we employ a dataset comprising over 10,000 global stocks and implement a global estimation strategy that pools information across cross-sections. Our econometric analysis reveals that forecasting accuracy improves markedly as the training dataset becomes larger and more heterogeneous. Notably, even with as little as 12 months of data, globally trained networks deliver robust predictions for individual stocks and portfolios that are not even in the training dataset. Furthermore, our interpretation of the model dynamics shows that these networks not only capture key stylized facts of volatility but also exhibit resilience to outliers and rapid adaptation to market regime changes. These findings underscore the importance of leveraging extensive and diverse datasets in financial forecasting and advocate for a shift from traditional local training approaches to integrated global estimation methods.
Paper Structure (50 sections, 11 equations, 8 figures, 9 tables)

This paper contains 50 sections, 11 equations, 8 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Ready-to-go Python Notebook
  3. Methodology
  4. Training schemes
  5. Local training.
  6. Global training.
  7. Neural Networks
  8. Dataset
  9. Experiment setup for data scaling effect study
  10. Evaluation metrics
  11. Statistical metrics:
  12. Economic utility:
  13. Statistical significance:
  14. Data Scaling Effect
  15. Local GARCH versus Local NNs
  16. ...and 35 more sections

Figures (8)

  • Figure 1: Graphical representation of the basic RNN.
  • Figure 2: GARCH data scaling effect: For supervised forecasts, the models are evaluated on 10 training stocks. For zero-shot forecasts, the models are evaluated on 1531 unseen stocks.
  • Figure 3: NNs data scaling effect: For supervised forecasts, the models are evaluated on 10 training stocks. For zero-shot forecasts, the models are evaluated on 1531 unseen stocks.
  • Figure 4: Temporal importance for the universal LSTM model.
  • Figure 5: News impact curve of the universal LSTM model.
  • ...and 3 more figures