Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stock Volume Forecasting with Advanced Information by Conditional Variational Auto-Encoder

Parley R Yang, Alexander Y Shestopaloff

TL;DR

The paper addresses stock-volume forecasting under advanced information such as rebalancing dates. It develops a conditional variational auto-encoder (CVAE) framework to capture non-linear interactions and generate forecast paths incorporating both advanced and ordinary information. Empirically, CVAE-based models (univariate and multivariate) outperform traditional linear baselines in long-term and short-term forecasts, while enabling scenario generation through the decoder to interpret feature effects. The work also discusses correlations in non-stationary time series and presents extensions and counterfactual analyses to illustrate practical interpretability and applicability to derivative pricing and risk management.

Abstract

We demonstrate the use of Conditional Variational Encoder (CVAE) to improve the forecasts of daily stock volume time series in both short and long term forecasting tasks, with the use of advanced information of input variables such as rebalancing dates. CVAE generates non-linear time series as out-of-sample forecasts, which have better accuracy and closer fit of correlation to the actual data, compared to traditional linear models. These generative forecasts can also be used for scenario generation, which aids interpretation. We further discuss correlations in non-stationary time series and other potential extensions from the CVAE forecasts.

Stock Volume Forecasting with Advanced Information by Conditional Variational Auto-Encoder

TL;DR

The paper addresses stock-volume forecasting under advanced information such as rebalancing dates. It develops a conditional variational auto-encoder (CVAE) framework to capture non-linear interactions and generate forecast paths incorporating both advanced and ordinary information. Empirically, CVAE-based models (univariate and multivariate) outperform traditional linear baselines in long-term and short-term forecasts, while enabling scenario generation through the decoder to interpret feature effects. The work also discusses correlations in non-stationary time series and presents extensions and counterfactual analyses to illustrate practical interpretability and applicability to derivative pricing and risk management.

Abstract

We demonstrate the use of Conditional Variational Encoder (CVAE) to improve the forecasts of daily stock volume time series in both short and long term forecasting tasks, with the use of advanced information of input variables such as rebalancing dates. CVAE generates non-linear time series as out-of-sample forecasts, which have better accuracy and closer fit of correlation to the actual data, compared to traditional linear models. These generative forecasts can also be used for scenario generation, which aids interpretation. We further discuss correlations in non-stationary time series and other potential extensions from the CVAE forecasts.
Paper Structure (26 sections, 17 equations, 10 figures, 2 tables, 2 algorithms)

This paper contains 26 sections, 17 equations, 10 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Motivation
  3. Literature Review
  4. Contributions
  5. Notations
  6. Methodology: From Non-Linear Modelling To Generative Algorithm For Forecasting
  7. Time series in a CVAE context
  8. Generative scheme from CVAE
  9. A special forecasting scenario: advanced information
  10. Forecasting algorithms using advanced information and CVAE
  11. Empirical Application: Daily Stock Volume Forecasting
  12. Data Availability and Processing
  13. Forecasting Tasks
  14. Summary of Models
  15. Evaluation Metrics
  16. ...and 11 more sections

Figures (10)

  • Figure 1: Illustrations of raw data and processed data for training and testing
  • Figure 2: Short Term Rolling Forecasts: M-CVAE and VAR(1) Illustrations
  • Figure 3: Long Term Forecasts: U-CVAE and ARMA(1,1) Illustrations
  • Figure 4: Long Term Forecasts: A zoomed-in plot for all models in March - April 2023, for ticker ASML.AS
  • Figure 5: RB interpretation: ASML illustration (generated by U-CVAE)
  • ...and 5 more figures