Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On Multivariate Financial Time Series Classification

Grégory Bournassenko

TL;DR

The paper addresses the challenge of classifying multivariate financial time series across small- and large-data regimes. It contrasts traditional approaches like SVMs and RNN/LSTM methods on small datasets with cutting-edge big-data techniques such as ConvTimeNet and graph-based ideas on large, high-dimensional financial series, including EUR/USD. It demonstrates that small data benefit from regularization and multivariate enrichment, while big data require scalable architectures and parallel computing, with ConvTimeNet showing strong performance and 20% cumulative returns on EUR/USD in one scenario. The work underscores infrastructure considerations (Spark, GPU) and points to non-Euclidean extensions (GNNs) as promising directions for future gains in finance analytics.

Abstract

This article investigates the use of Machine Learning and Deep Learning models in multivariate time series analysis within financial markets. It compares small and big data approaches, focusing on their distinct challenges and the benefits of scaling. Traditional methods such as SVMs are contrasted with modern architectures like ConvTimeNet. The results show the importance of using and understanding Big Data in depth in the analysis and prediction of financial time series.

On Multivariate Financial Time Series Classification

TL;DR

The paper addresses the challenge of classifying multivariate financial time series across small- and large-data regimes. It contrasts traditional approaches like SVMs and RNN/LSTM methods on small datasets with cutting-edge big-data techniques such as ConvTimeNet and graph-based ideas on large, high-dimensional financial series, including EUR/USD. It demonstrates that small data benefit from regularization and multivariate enrichment, while big data require scalable architectures and parallel computing, with ConvTimeNet showing strong performance and 20% cumulative returns on EUR/USD in one scenario. The work underscores infrastructure considerations (Spark, GPU) and points to non-Euclidean extensions (GNNs) as promising directions for future gains in finance analytics.

Abstract

This article investigates the use of Machine Learning and Deep Learning models in multivariate time series analysis within financial markets. It compares small and big data approaches, focusing on their distinct challenges and the benefits of scaling. Traditional methods such as SVMs are contrasted with modern architectures like ConvTimeNet. The results show the importance of using and understanding Big Data in depth in the analysis and prediction of financial time series.

Paper Structure

This paper contains 47 sections, 69 equations, 38 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Context and Relevance
  3. Introduction to time series classification in finance
  4. Importance of accurate predictions in the financial market
  5. Overview of multivariate time series and its significance
  6. Research Objectives and Questions
  7. Objective: To explore and compare different methodologies for classifying financial market data using MTS
  8. Key questions: How do small data and big data approaches differ? What are the challenges in predicting financial markets with MTS?
  9. Multivariate Time Series Classification with Small Data
  10. Overview of Small Data in Financial Time Series
  11. Characteristics of Small Data
  12. Importance of Multivariate vs. Univariate Analysis
  13. Small Data Example
  14. Machine Learning Approaches for Small Data
  15. Traditional Machine Learning Techniques
  16. ...and 32 more sections

Figures (38)

  • Figure 1: Example of windowing susto2018time
  • Figure 2: A unified deep learning framework for time series classification ismail2019deep
  • Figure 3: Architecture of a standard LSTM (NLP) wang2016attention
  • Figure 4: Early stopping based on cross-validation genccay2001pricing
  • Figure 5: Strike daily closing price (2024-06-04 to 2024-09-18)
  • ...and 33 more figures