Paper: arXiv 2412.09517
Authors: Andrea Bucci, Michele Palma, Chao Zhang
Abstract
Traditional methods employed in matrix volatility forecasting often overlook the inherent Riemannian manifold structure of symmetric positive definite matrices, treating them as elements of Euclidean space, which can lead to suboptimal predictive performance. Moreover, they often struggle to handle high-dimensional matrices. In this paper, we propose a novel approach for forecasting realized covariance matrices of asset returns using a Riemannian-geometry-aware deep learning framework. In this way, we account for the geometric properties of the covariance matrices, including possible non-linear dynamics and efficient handling of high-dimensionality. Moreover, building upon a Fréchet sample mean of realized covariance matrices, we are able to extend the HAR model to the matrix-variate. We demonstrate the efficacy of our approach using daily realized covariance matrices for the 50 most capitalized companies in the S&P 500 index, showing that our method outperforms traditional approaches in terms of predictive accuracy.
Complexity vs Empirical Score
- Math Complexity: 9.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced Riemannian geometry, manifold theory, and custom geometric deep learning architectures (SPDNet), indicating very high mathematical density. It also validates the method on a substantial dataset (50 stocks from S&P 500) with specific predictive accuracy benchmarks and a portfolio optimization application, demonstrating strong empirical backing.
Research Flowchart
flowchart TD Start["Research Goal: Forecast Realized Covariance Matrices<br>using Geometric Deep Learning"] --> Input["Data: Daily Realized Covariance Matrices<br>50 largest S&P 500 Equities"] Input --> Method["Methodology: Riemannian-Geometry-Aware<br>Deep Learning Framework"] Method --> Process["Process: Fréchet Sample Mean +<br>Matrix-Variate HAR Model Extension"] Process --> Manifold["Key Innovation: Accounts for Riemannian<br>Manifold Structure & High-Dimensionality"] Manifold --> Outcome["Outcome: Outperforms Traditional Methods<br>in Predictive Accuracy"]
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 Riemannian‑geometry‑aware deep learning framework to forecast realized covariance matrices, including a diagonal‑block input construction and a geometric HAR extension.
Source sentence
we propose a novel approach for forecasting realized covariance matrices of asset returns using a Riemannian-geometry-aware deep learning framework.
Assumes 6
- Covariance matrices are symmetric positive‑definite and lie on a Riemannian manifold. model
Source sentence
An additional fundamental limitation of all these methods is that they treat SPD matrices as if they were elements of the Euclidean space, overlooking the fact that SPD matrices inherently rely on a Riemannian manifold
- The log‑Euclidean distance is an appropriate metric for the Fréchet mean and loss. model
Source sentence
In our approach, we first specify d(·) as a log-Euclidean distance (Arsigny et al., 2007) defined as follows
- Daily realized covariance series are stationary. statistical
Source sentence
The covariance stationarity of each time series was validated through unit root tests, including the Augmented Dickey-Fuller (ADF), ADF-GLS, Phillips-Perron (PP), and KPSS tests.
- Constructing a block‑diagonal matrix of SPD lagged covariances yields another SPD matrix. model
Source sentence
It must be noticed that the construction of a block diagonal matrix using SPD matrices is still an SPD matrix.
Show 2 more
- Portfolio evaluation assumes no short‑selling constraints unless otherwise stated. execution
Source sentence
GMV indicates a portfolio without no-short-selling constraints.
- No transaction costs are incurred in the portfolio simulations. execution
Source sentence
1No transaction costs are assumed.
Data
- Daily realized covariance matrices market-prices commercial
Source sentence
The empirical application of our forecasting method utilizes time series data comprising intradaily close spot prices for the 50 largest U.S. stocks by market capitalization, as of November 2021, all of which are part of the S&P 500 index. This data obtained from Lobster spans the period from June 27th, 2007 to July 1s
Costs
Not modeled
Source sentence
1No transaction costs are assumed.
Finds 5
- The proposed geometric deep‑learning approach outperforms traditional methods in predictive accuracy for realized covariance matrices. predictive-accuracy
Source sentence
showing that our method outperforms traditional approaches in terms of predictive accuracy.
- The GeoHAR model with a log‑Euclidean Fréchet mean achieves the best forecasting performance among all tested models. performance
Source sentence
In particular, a model with ten lags and trained using the logarithmic Euclidean distance as a loss function and the GeoHAR using a log-Euclidean-based
- ReSPDNet trained with log‑Euclidean loss and 10 lags (ReSPDNet‑LE10) is the best‑performing model in the statistical selection matrix. performance
Source sentence
with a log-Euclidean loss and 10 lags, which is also the only model in the SSM.
- Geometric approaches (e.g., GeoHAR‑LELE) achieve low portfolio turnover while maintaining low risk, outperforming Cholesky‑based strategies that have high turnover. performance
Source sentence
some geometric approaches, such as the GeoHAR-LELE, manage to maintain relatively low turnover while delivering near‑optimal risk metrics.
- Using a Euclidean loss for training SPD networks leads to poorer predictive results compared to geometric‑aware loss functions. negative
Source sentence
Overall, our findings demonstrate that using an Euclidean loss function to train a neural network for SPD matrices leads to poor predictive results.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 40 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.
Cited References (53)
- The distribution of realized exchange rate volatility (2001)
- Modeling and Forecasting Realized Volatility (2003)
- The distribution of exchange rate volatility (2001)
- A New Parametrization of Correlation Matrices (2021) — link
- Geometric Means in a Novel Vector Space Structure on Symmetric Positive‐Definite Matrices (2007) — link
- Forecasting co-volatilities via factor models with asymmetry and long memory in realized covariance (2015) — link
- Realized matrix-exponential stochastic volatility with asymmetry, long memory and higher-moment spillovers (2022) — link
- Fractionally integrated generalized autoregressive conditional heteroskedasticity (1996) — link
and 45 more
- Estimating Quadratic Variation Using Realised Variance (2002)
- Econometric Analysis of Realized Volatility and Its Use in Estimating Stochastic Volatility Models (2002)
- Forecasting multivariate realized stock market volatility (2011)
- Modeling and forecasting (un)reliable realized covariances for more reliable financial decisions (2018) — link
- Riemannian batch normalization for SPD neural networks (2019)
- Realized Volatility Forecasting with Neural Networks (2020)
- Comparing unconstrained parametrization methods for return covariance matrix prediction (2022)
- A statistical recurrent model on the manifold of symmetric positive definite matrices (2018)
- Modeling and Forecasting Large Realized Covariance Matrices and Portfolio Choice (2017) — link
- Modelling and Forecasting Multivariate Realized Volatility (2011)
- A Simple Approximate Long-Memory Model of Realized Volatility (2009)
- Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy? (2009) — link
- Predicting the Daily Covariance Matrix for S$&$P 100 Stocks Using Intraday Data - But Which Frequency to Use? (2008)
- Non-Euclidean Statistics for Covariance Matrices, with Applications to Diffusion Tensor Imaging (2009) — link
- Procrustes analysis (2016) — link
- Dynamic conditional correlation: a simple class of multivariate GARCH models (2002)
- Multivariate simultaneous generalized ARCH (1995)
- Volatility is rough (2018) — link
- Generalized Procrustes analysis. (1975)
- Factor state–space models for high-dimensional realized covariance matrices of asset returns (2020) — link
- The Model Confidence Set (2011)
- A geometric framework for covariance dynamics (2022) — link
- From manifold to manifold: Geometry-aware dimensionality reduction for SPD matrices (2014)
- Computing a nearest symmetric positive semidefinite matrix (1988) — link
- A Riemannian network for SPD matrix learning (2017)
- Estimating Long Memory in Volatility (2005) — link
- Bayesian semiparametric modeling of realized covariance matrices (2016) — link
- A Multivariate Generalized Orthogonal Factor GARCH Model (2007) — link
- On loss functions and ranking forecasting performances of multivariate volatility models (2013)
- Riemannian-geometric regime-switching covariance hedging (2024) — link
- Electroencephalogram signals classification for sleep-state decision–a Riemannian geometry approach (2012)
- Portfolio selection (1952)
- Object Oriented Data Analysis (2022)
- Multivariate high-frequency-based volatility ( HEAVY ) models (2011)
- The Conditional Autoregressive F-Riesz Model for Realized Covariance Matrices (2024) — link
- Increasing the information content of realized volatility forecasts* (2023) — link
- Inference on the maximal rank of time-varying covariance matrices using high-frequency data (2021)
- Why does stock market volatility change over time? (1989)
- Pedestrian detection via classification on Riemannian manifolds (2008)
- GO-GARCH: A Multivariate Generalized Orthogonal GARCH Model (2002) — link
- A full-factor multivariate GARCH model (2003) — link
- Dreamnet: A deep Riemannian manifold network for spd matrix learning (2022)
- The Generalized Conditional Autoregressive Wishart Model for Multivariate Realized Volatility (2017) — link
- Graph-Based Methods for Forecasting Realized Covariances (2024) — link
- Machine Learning Classification of Riemannian Tangent Spaces Based on MI-BCI (2023)
0 of 53 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.