Paper: arXiv 2502.13461
Authors: Cheng Yu, Zhoufan Zhu, Ke Zhu
Abstract
Style investing creates asset classes (or the so-called “styles”) with low correlations, aligning well with the principle of “Holy Grail of investing” in terms of portfolio selection. The returns of styles naturally form a tensor-valued time series, which requires new tools for studying the dynamics of the conditional correlation matrix to facilitate the aforementioned principle. Towards this goal, we introduce a new tensor dynamic conditional correlation (TDCC) model, which is based on two novel treatments: trace-normalization and dimension-normalization. These two normalizations adapt to the tensor nature of the data, and they are necessary except when the tensor data reduce to vector data. Moreover, we provide an easy-to-implement estimation procedure for the TDCC model, and examine its finite sample performance by simulations. Finally, we assess the usefulness of the TDCC model in international portfolio selection across ten global markets and in large portfolio selection for 1800 stocks from the Chinese stock market.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)
Why this score: The paper introduces a novel tensor GARCH model with advanced multi-way conditional heteroskedasticity specifications and trace-normalization, requiring deep mathematical treatment. It is heavily implementation-focused, with simulations, a large-scale empirical application (1800 stocks), and clear portfolio selection backtests.
Research Flowchart
flowchart TD A["Research Goal: Pursue 'Holy Grail of investing'<br>via Style Investing & Correlation Dynamics"] --> B["Data Input: Tensor-Valued<br>Style Return Time Series"] B --> C["Methodology: TDCC Model<br>with Trace & Dimension Normalization"] C --> D["Computation: Estimation &<br>Simulation Validation"] D --> E["Application 1: International<br>Portfolio Selection (10 Markets)"] D --> F["Application 2: Large Portfolio<br>Selection (1800 Stocks)"] E --> G["Key Outcomes: Effective<br>Dynamic Correlation Modeling"] F --> G
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 tensor dynamic conditional correlation (TDCC) model for tensor-valued time series.
Source sentence
we introduce a new tensor dynamic conditional correlation (TDCC) model, which is based on two novel treatments: trace-normalization and dimension-normalization.
Assumes 3
- The model requires trace-normalization to solve identification issues. model
Source sentence
the TDCC model needs a trace-normalization to solve the issue of model identification
- Each tensor entry may have its own GARCH specification for conditional variance. model
Source sentence
the TDCC model allows each σ2 i1i2,t (within yt) to have its own GARCH specification
- Conditional correlations among styles are typically very small. data
Source sentence
the conditional correlations among styles are typically very small
Data
multi-asset style returns · equities · weekly
- International style returns dataset market-prices public
Source sentence
Our first application is related to Example 1, where the 440 styles are constructed from ten global markets, eleven sectors, with four capitalization levels.
- Chinese stock style returns dataset market-prices commercial
Source sentence
Our second application is relevant to Example 2, based on 125 styles constructed from 1800 stocks according to their levels of three characteristics: “Size”, “OP”, and “Inv”.
Costs
Not modeled
Finds 3
Sharpe 1.84
- TDCC-based minimum-variance portfolios outperform competing methods that break the tensor structure. performance
Source sentence
According to the out-of-sample results, the minimum-variance portfolios selected by the TDCC-based method outperform those by other competing methods, which break the tensor structure of returns into matrix or vector counterpart during the modeling process.
- TDCC-NLS method yields substantially higher information ratios than MDCC and VDCC methods. performance
Source sentence
the value of IR from the TDCC-NLS method is 22.5% and 336.2% higher than that from the MDCC1-NLC and VDCC-NLS methods, respectively.
- Conditional correlations among the 125 Chinese styles are larger (around 0.4 on average) than those among international styles. statistical
Source sentence
Compared with the findings of the first application, the conditional correlations among these 125 styles are larger (around 0.4 on average)
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 23 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 (78)
- Projected estimation for large-dimensional matrix factor models (2022)
- Multivariate variance targeting in the BEKK-GARCH model (2014)
- Quasi-maximum likelihood estimation and inference in dynamic models with time-varying covariance (1992)
- Estimation, inference and specification analysis (1994)
- Time series models for realized covariance matrices based on the matrix- F distribution (2022)
- Large dynamic covariance matrices (2019)
- Factor models for matrix-valued high-dimensional time series (2019)
- Autoregressive models for matrix-valued time series (2021)
and 70 more
- Multivariate GARCH models: A survey (2006)
- Dynamic factor multivariate GARCH model (2014)
- A multivariate generalized orthogonal factor GARCH model (2007)
- Statistical inference for high-dimensional matrix-variate factor models (2021)
- High dimensional minimum variance portfolio estimation under statistical factor models (2021)
- Autoregressive conditional heteroscedasticity with estimates of the variance of U nited K ingdom inflation (1982)
- A capital asset pricing model with time-varying covariances (1988)
- Multivariate simultaneous generalized ARCH (1995)
- Common persistence in conditional variances (1993)
- Modelling the coherence in short-run nominal exchange rates: A multivariate generalized ARCH model (1990)
- Generalized autoregressive conditional heteroskedasticity (1986)
- Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models (2002)
- Modelling matrix time series via a tensor CP -decomposition (2023)
- Reduced rank autoregressive models for matrix time series (2022)
- Modeling high-dimensional matrix-variate observations by tensor factorization (2022) — link
- GARCH Models: S tructure, S tatistical I nference and F inancial A pplications (2019)
- On asymptotic theory for multivariate GARCH models (2009)
- Factor models for high-dimensional tensor time series (2022)
- Asymptotic theory for a vector ARMA-GARCH model (2003)
- Advanced E conometrics (1985)
- Diagnostic checking of nonlinear multivariate time series with multivariate ARCH errors (1997)
- A multivariate generalized autoregressive conditional heteroscedasticity model with time-varying correlations (2002)
- Stationarity and geometric ergodicity of BEKK multivariate GARCH models (2011)
- Asymptotic theory for multivariate GARCH processes (2003)
- The varimax criterion for analytic rotation in factor analysis (1958)
- Comparing predictive accuracy (2002)
- Modeling dynamic transport network with matrix factor models: An application to international trade flow (2022)
- Testing and support recovery of correlation structures for matrix-valued observations with an application to stock market data (2023)
- GO-GARCH : A multivariate generalized orthogonal GARCH model (2002)
- A well-conditioned estimator for large-dimensional covariance matrices (2004)
- Nonlinear shrinkage estimation of large-dimensional covariance matrices (2012)
- High-dimensional low-rank tensor autoregressive time series modeling (2024)
- Constrained factor models for high-dimensional matrix-variate time series (2020)
- The riskmetrics 2006 methodology (2007)
- Residual-based diagnostics for conditional heteroscedasticity models (2002)
- Automatic specification testing for vector autoregressions and multivariate nonlinear time series models (2013)
- Lecture notes for ``Time series analysis for forecasting and model building’’ (2008)
- Matrix V ariate D istributions (1999)
- Dynamic conditional correlation: O n properties and estimation (2013)
- Fitting vast dimensional time-varying covariance models (2021)
- The power of (non-) linear shrinking: A review and guide to covariance matrix estimation (2022)
- A five–factor asset pricing model (2015)
- Digesting anomalies: An investment approach (2015)
- Multi-linear tensor autoregressive models (2021) — link
- Rank determination in tensor factor model (2022)
- CP factor model for dynamic tensors (2024)
- Statistical inference for large-dimensional tensor factor model by weighted/unweighted projection (2022) — link
- Tucker tensor factor models for high-dimensional higher-order tensor observations (2022) — link
- Robust estimation of large factor models for tensor-valued time series (2023) — link
- Tensor decompositions and applications (2009)
- A generalized dynamic conditional correlation model: S imulation and application to many assets (2009)
- Dynamic equicorrelation (2012)
- Dynamic portfolio selection with sector-specific regularization (2024)
- Matrix GARCH model: Inference and application (2024)
- High-dimensional covariance matrix estimation (2020)
- Principles: Life and Work (1st Edition) (2017)
- Quantifying and interpreting collective behavior in financial markets (2001)
- On financial market correlation structures and diversification benefits across and within equity sectors (2022)
- Dynamic correlation of market connectivity, risk spillover and abnormal volatility in stock price (2022)
- Style investing (2003)
- Style Investing and Institutional Investors (2008)
- Style investing, comovement and return predictability (2013)
- Portfolio selection (1952)
- Systemic risk and international portfolio choice (2004)
- International diversification benefits with foreign exchange investment styles (2014)
- Thinking through categories (2002)
- GARCH for groups (1996)
- Tensor factor model estimation by iterative projection (2024)
- Semi-parametric tensor factor analysis by iteratively projected singular value decomposition (2024)
- Rank and factor loadings estimation in time series tensor factor model by pre-averaging (2024)
0 of 78 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.