Paper: arXiv 2310.14536

Authors: Xin Du, Kai Moriyama, Kumiko Tanaka-Ishii

Abstract

This paper shows a novel machine learning model for realized volatility (RV) prediction using a normalizing flow, an invertible neural network. Since RV is known to be skewed and have a fat tail, previous methods transform RV into values that follow a latent distribution with an explicit shape and then apply a prediction model. However, knowing that shape is non-trivial, and the transformation result influences the prediction model. This paper proposes to jointly train the transformation and the prediction model. The training process follows a maximum-likelihood objective function that is derived from the assumption that the prediction residuals on the transformed RV time series are homogeneously Gaussian. The objective function is further approximated using an expectation-maximum algorithm. On a dataset of 100 stocks, our method significantly outperforms other methods using analytical or naive neural-network transformations.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical concepts such as normalizing flows, maximum-likelihood estimation with EM approximation, and distribution theory, indicating high mathematical complexity. It also demonstrates empirical rigor by testing on a substantial dataset (100 stocks), comparing against multiple baselines, and detailing the implementation and training process for the neural network and prediction model.

Research Flowchart

  flowchart TD
  A["Research Goal: Realized Volatility Prediction<br>with Skewed/Fat-Tail Data"] --> B["Data: 100 Stocks"]
  B --> C["Method: Co-Training Flow & Model"]
  C --> D["Normalization: Invertible Neural Network<br>Transforms RV to Latent Gaussian"]
  D --> E["Forecasting: Neural Network<br>Predicts Transformed RV"]
  E --> F["Training: Max Likelihood via EM Algorithm"]
  F --> G["Outcome: Jointly Optimized Model"]
  G --> H["Result: Significantly Outperforms<br>Analytical/Naive Transformations"]

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 proposes a new machine‑learning model that jointly trains a normalizing‑flow transformation and a linear prediction model for realized volatility.

Source sentence
This paper proposes to jointly train the transformation and the prediction model.

Assumes 2

  • The prediction residuals on the transformed RV time series are homogeneously Gaussian. statistical
    Source sentence
    The training process follows a maximum-likelihood objective function that is derived from the assumption that the prediction residuals on the transformed RV time series are homogeneously Gaussian.
  • Residuals are homogeneous across time, i.e., follow the same Gaussian distribution for every timestep. statistical
    Source sentence
    In addition, we assume the residuals to be homogeneous across time (i.e., 𝜀ₛ follows the same Gaussian distribution for every timestep

Data

  • 100 major NYSE stocks realized volatility series · New York Stock Exchange market-prices unknown
    December 1st, 2015 to October 25th, 2017 · 100 major stocks listed on the NYSE
    Source sentence
    The proposed method is evaluated on a dataset of 100 major stocks listed on the New York Stock Exchange. The dataset covers a continuous time period of 480 trading days from December 1st, 2015, to October 25th, 2017.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The NODE transformation with τ=0.25 achieved the lowest RMSE of 0.5620, a 0.0075 improvement over the identity transformation, and was best on 46 of the 100 stocks. performance
    Source sentence
    ... achieved the lowest RMSE score at 0.5620, a clear improvement of 0.0075 compared with the identity transformation that scored 0.5695.
  • Using a transformation greatly improved Gaussianity of residuals, raising average R² from 93.22% to 99.14% and reducing average skewness from 1.788 to 0.2141. statistical
    Source sentence
    was improved to 99.14%, and skewness was reduced to 0.2141 with a NODE having τ=5.00.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 9 related papers). Coverage is partial, so this is neither replication nor refutation.

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.