Recurrent neural networks for hierarchical time series forecasting: An application to the S&P 500 market value

This paper investigates the use of Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures, for hierarchical time series forecasting in financial markets. Using market value data from the top 70 companies in the S&P 500 index, we...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Physica A Jg. 678; S. 130869
Hauptverfasser: Munyao, Jackson Ndoto, Oluoch, Lillian Achola, Iftikhar, Hasnain, Rodrigues, Paulo Canas
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 15.11.2025
Schlagworte:
ARIMA
Forecasting
Hierarchical time series
Hierarchical time series forecasting
LSTM and GRU
Recurrent neural networks (RNNs)
S&P 500
S&P 500
ARIMA
LSTM and GRU
Hierarchical time series forecasting
Recurrent neural networks (RNNs)
Forecasting
Hierarchical time series
ISSN:0378-4371
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper investigates the use of Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures, for hierarchical time series forecasting in financial markets. Using market value data from the top 70 companies in the S&P 500 index, we evaluate forecasts across three hierarchical levels: company, sector, and market total, applying various reconciliation strategies to ensure coherence. The proposed framework is compared with traditional models (Autoregressive Integrated Moving Average and Exponential Smoothing) under multiple reconciliation methods, including Middle-Out with forecast proportions. Results show that RNN-based models outperform statistical benchmarks in terms of accuracy across levels, particularly when combined with Middle-Out reconciliation. We also discuss practical aspects such as computational cost and implementation trade-offs, highlighting the relevance of deep learning methods for structured financial forecasting tasks.
ISSN:0378-4371
DOI:10.1016/j.physa.2025.130869