Paper: arXiv 2410.09062

Authors: Alex Li

Abstract

Predicting volatility in financial markets, including stocks, index ETFs, foreign exchange, and cryptocurrencies, remains a challenging task due to the inherent complexity and non-linear dynamics of these time series. In this study, I apply TimeMixer, a state-of-the-art time series forecasting model, to predict the volatility of global financial assets. TimeMixer utilizes a multiscale-mixing approach that effectively captures both short-term and long-term temporal patterns by analyzing data across different scales. My empirical results reveal that while TimeMixer performs exceptionally well in short-term volatility forecasting, its accuracy diminishes for longer-term predictions, particularly in highly volatile markets. These findings highlight TimeMixer’s strength in capturing short-term volatility, making it highly suitable for practical applications in financial risk management, where precise short-term forecasts are critical. However, the model’s limitations in long-term forecasting point to potential areas for further refinement.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical concepts including multiscale decomposition and specific neural network blocks (PDM/FMM), indicative of high complexity. It is backed by empirical results across multiple asset classes (stocks, ETFs, FX, crypto) and discusses performance metrics, showing significant data and implementation efforts.

Research Flowchart

  flowchart TD
  A["Research Goal: Predict Volatility<br>in Global Financial Markets"] --> B["Data Collection<br>Multi-asset Time Series"]
  B --> C["Methodology: Apply TimeMixer<br>Multiscale-Mixing Approach"]
  C --> D["Computational Process<br>Capture Short & Long-Term Patterns"]
  D --> E{"Key Findings / Outcomes"}
  E --> F["Strong Performance in<br>Short-Term Forecasting"]
  E --> G["Limitations in<br>Long-Term Forecasting"]
  F --> H["Application: Financial<br>Risk Management"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper applies the TimeMixer multiscale‑mixing model to forecast volatility of global financial assets.

Source sentence
In this study, I apply TimeMixer, a state-of-the-art time series forecasting model, to predict the volatility of global financial assets.

Assumes 3

  • Volatility is measured as annualized standard deviation of daily log returns over a 21‑day rolling window. data
    Source sentence
    σ = √252 × std(log Pt Pt−1, window = 21) Where: log Pt Pt−1 represents the daily log returns. std() denotes the standard deviation. window = 21 refers to the rolling window of 21 days (approximately one trading month). √252 annualizes the volatility, assuming 252 trading days per year.
  • OHLCV data from Yahoo Finance is accurate and representative of the assets studied. data
    Source sentence
    (Open, High, Low, Close, Volume) data obtained from Yahoo Finance to predict volatility.
  • The validation set comprises 10% of the training set size. model
    Source sentence
    The validation set comprises 10% of the training set size.

Data

  • Stock OHLCV dataset · Yahoo Finance market-prices public
    2010-2023 · AAPL, GOOGL, MSFT, AMZN, TSLA, BRK-B, ...
    Source sentence
    Ticker Symbol Date Range Company AAPL 2010 - 2023 Apple Inc. GOOGL 2010 - 2023 Alphabet Inc. MSFT 2010 - 2023 Microsoft Corporation AMZN 2010 - 2023 Amazon.com, Inc. TSLA 2010 - 2023 Tesla, Inc.
  • Cryptocurrency OHLCV dataset · Yahoo Finance crypto-onchain public
    2014-2023 · BTCUSD, ETHUSD, LTCUSD, BCHUSD, DOGEUSD, XRPUSD, ADAUSD, DOTUSD, BNBUSD, SOLUSD
    Source sentence
    BTCUSD 2014 - 2023 Bitcoin/USD ETHUSD 2017 - 2023 Ethereum/USD LTCUSD 2014 - 2023 Litecoin/USD BCHUSD 2017 - 2023 Bitcoin Cash/USD DOGEUSD 2017 - 2023 Dogecoin/USD XRPUSD 2017 - 2023 XRP/USD ADAUSD 2017 - 2023 Cardano/USD DOTUSD 2020 - 2023 Polkadot/USD BNBUSD 2017 - 2023 BNB/USD SOLUSD 2020 - 2023 Solana/USD

Recognised sources: Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • TimeMixer excels in short‑term volatility forecasting, achieving very low errors for 12‑day horizons. predictive-accuracy
    Source sentence
    TimeMixer excels in predicting short-term volatility, outperforming its long-term forecasts.
  • For AAPL, the model attains MAE = 0.0037 and RMSE = 0.0059 over a 12‑day period. performance
    Source sentence
    AAPL shows a low MAE of 0.0037 and RMSE of 0.0059 over a 12-day period
  • Error metrics increase substantially for longer horizons, e.g., NVDA MAE rises to 0.2883 at 336 days. negative
    Source sentence
    NVDA’s MAE increases to 0.2883 at 336 days
  • Highly volatile assets such as DOGE/USD and SOL/USD exhibit notably higher errors, especially over longer horizons. predictive-accuracy
    Source sentence
    the model struggles with highly volatile assets such as DOGE/USD, SOL/USD, and TSLA, where error metrics increase notably
  • More stable assets like BRK‑B maintain relatively low errors even for long‑term forecasts (e.g., MAE = 0.0543 at 720 days). performance
    Source sentence
    more stable stocks like BRK-B and JNJ show comparatively lower errors over longer periods, with BRK-B having an MAE of

Checked by others

Supported by

  • A Dynamic Approach to Stock Price Prediction: Comparing RNN and Mixture of Experts Models Across Different Volatility Profiles (2024)
    On the question: Does high volatility increase prediction error for RNN and TimeMixer models?
    Compare the findings

    This paper: Highly volatile assets such as DOGE/USD and SOL/USD exhibit notably higher errors, especially over longer horizons.

    That paper: The RNN model on volatile companies yields MSE = 2.4265 and MAE = 1.4678, indicating high error on extreme variations.

    A reports MSE = 2.4265, MAE = 1.4678 for volatile firms; B notes DOGE/USD and SOL/USD have notably higher errors, especially at longer horizons.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (13)

  1. How noise trading affects markets: An experimental analysis (2009)
  2. Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation (1982)
  3. Generalized autoregressive conditional heteroskedasticity (1986)
  4. Long Short-term Memory (1997)
  5. Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014) — link
  6. Timemixer: Decomposable multiscale mixing for time series forecasting (2024) — link
  7. Time series analysis: forecasting and control (2015)
  8. Macroeconomics and reality (1980)
and 5 more
  1. Gradient-based learning applied to document recognition (1998)
  2. Attention is all you need (2017)
  3. Factoring wavelet transforms into lifting steps (1998)
  4. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis (1998)
  5. Learning representations by back-propagating errors (1986)

0 of 13 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.