Paper: arXiv 2407.04500
Authors: Philipp Wirth, Francesca Medda, Thomas Schröder
Abstract
In this paper, we introduce the Dynamic Modularity-Spectral Algorithm (DynMSA), a novel approach to identify clusters of stocks with high intra-cluster correlations and low inter-cluster correlations by combining Random Matrix Theory with modularity optimisation and spectral clustering. The primary objective is to uncover hidden market structures and find diversifiers based on return correlations, thereby achieving a more effective risk-reducing portfolio allocation. We applied DynMSA to constituents of the S&P 500 and compared the results to sector- and market-based benchmarks. Besides the conception of this algorithm, our contributions further include implementing a sector-based calibration for modularity optimisation and a correlation-based distance function for spectral clustering. Testing revealed that DynMSA outperforms baseline models in intra- and inter-cluster correlation differences, particularly over medium-term correlation look-backs. It also identifies stable clusters and detects regime changes due to exogenous shocks, such as the COVID-19 pandemic. Portfolios constructed using our clusters showed higher Sortino and Sharpe ratios, lower downside volatility, reduced maximum drawdown and higher annualised returns compared to an equally weighted market benchmark.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.5)
Why this score: The paper employs advanced mathematical techniques including Random Matrix Theory, modularity optimization, and spectral clustering, with detailed derivations in the methodology section. It demonstrates high empirical rigor through extensive backtesting on S&P 500 constituents, showing outperformance against benchmarks with specific metrics (Sharpe/Sortino ratios, drawdowns) and sensitivity to regime changes like COVID-19.
Research Flowchart
flowchart TD
A["Research Goal: Identify Stable Stock Clusters<br>for Risk-Reduced Portfolio Allocation"] --> B["Methodology: DynMSA Algorithm"]
B --> C["Random Matrix Theory Filtering<br>Remove Market Noise"]
C --> D["Modularity Optimisation<br>Sector-based Calibration"]
D --> E["Spectral Clustering<br>Correlation-based Distance"]
E --> F["Apply to S&P 500 Data"]
F --> G{"Key Outcomes"}
G --> H["Outperformed Benchmarks<br>Higher Sharpe/Sortino Ratios"]
G --> I["Detected Regime Shifts<br>e.g., COVID-19 Pandemic"]
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 the Dynamic Modularity-Spectral Algorithm (DynMSA), a novel method combining Random Matrix Theory, modularity optimisation and spectral clustering to identify stock clusters with high intra‑cluster and low inter‑cluster correlations.
Source sentence
In this paper, we introduce the Dynamic Modularity-Spectral Algorithm (DynMSA), a novel approach to identify clusters of stocks with high intra-cluster correlations and low inter-cluster correlations by combining Random Matrix Theory with modularity optimisation and spectral clustering.
Assumes 2
- The Leiden algorithm may assign every stock to its own cluster. model
Source sentence
If we assume, that the Leiden algorithm assigns every stock to its own cluster, we ultimately remain at the starting point of our problem.
- Correlation matrix is built from daily closing prices, implying no intraday price effects. data
Source sentence
The data we use to compute the correlation matrix are daily closing prices for S&P 500
Data
S&P 500 stocks · equities
- S&P 500 daily closing prices market-prices public
Source sentence
The data we use to compute the correlation matrix are daily closing prices for S&P 500
- Equally weighted performance benchmark benchmark public
Source sentence
Additionally we compute an equally weighted performance benchmark which is also rebalanced monthly.
Costs
Unclear - sources disagree
Finds 4
Sharpe 0.92 (benchmark 0.70) · annual return 17.6% · max drawdown 17.2%
- DynMSA outperforms baseline models in intra‑ and inter‑cluster correlation differences, especially over medium‑term look‑backs. statistical
Source sentence
Testing revealed that DynMSA outperforms baseline models in intra- and inter-cluster correlation differences, particularly over medium-term correlation look-backs.
- Portfolios built from DynMSA clusters achieve higher Sortino and Sharpe ratios, lower downside volatility, reduced maximum drawdown and higher annualised returns versus an equally weighted market benchmark. performance
Source sentence
Portfolios constructed using our clusters showed higher Sortino and Sharpe ratios, lower downside volatility, reduced maximum drawdown and higher annualised returns compared to an equally weighted market benchmark.
- Spectral clustering significantly improves inter‑cluster separation and the objective function for 3‑, 6‑ and 12‑month look‑backs, but shows no change for 24‑month look‑backs. statistical
Source sentence
For the 3-months period, the reduction in inter-cluster correlation is highly significant
- The method identifies stable clusters over time and detects regime changes caused by exogenous shocks such as the COVID‑19 pandemic. other
Source sentence
Our method identified stable clusters over time and detected regime changes due to exogenous shocks such as the COVID-19 pandemic.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 21 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 (51)
- Hierarchical structure in financial markets (1999)
- Fast unfolding of communities in large networks (2008)
- Community Detection for Correlation Matrices (2015)
- Community detection in graphs (2010)
- Building diversified portfolios that outperform out of sample (2016)
- Finding and evaluating community structure in networks (2004)
- Consensus clustering in complex networks (2012)
- Portofolio selection (1952)
and 43 more
- Spread of risk across financial markets: better to invest in the peripheries (2013)
- A random matrix theory approach to financial cross-correlations (2000)
- Safety first and the holding of assets (1952)
- A tutorial on spectral clustering (2007)
- From Louvain to Leiden: guaranteeing well-connected communities (2019)
- Resolution limit in community detection (2007)
- Multiscale correlation networks analysis of the US stock market: a wavelet analysis (2017)
- Sentiment correlation in financial news networks and associated market movements (2021)
- Stock network stability in times of crisis (2014)
- A statistical method for evaluating systematic relationships (1958)
- Normalized cuts and image segmentation (2000) — link
- Asymmetric correlations of equity portfolios (2002)
- Extreme correlation of international equity markets (2001)
- Noise dressing of financial correlation matrices (1999)
- Noisy covariance matrices and portfolio optimization II (2003)
- Random matrix approach to cross correlations in financial data (2002)
- Cleaning large correlation matrices: tools from random matrix theory (2017)
- Distribution of eigenvalues for some sets of random matrices (1967)
- Clustering and information in correlation based financial networks (2004)
- Maximal spanning trees, asset graphs and random matrix denoising in the analysis of dynamics of financial networks (2009)
- Comparing partitions (1985)
- Objective criteria for the evaluation of clustering methods (1971)
- Clustering financial time series: How long is enough? (2016) — link
- The myth of diversification (2009)
- The Myth of Diversification Reconsidered (2021)
- Equity market connectedness across regimes of geopolitical risks: Historical evidence and theory (2023)
- Contagious investor sentiment and international markets (2017)
- Stability and similarity in financial networks—How do they change in times of turbulence? (2021)
- A tool for filtering information in complex systems (2005)
- Understanding changes in the topology and geometry of financial market correlations during a market crash (2021)
- Dynamics of cluster structures in a financial market network (2014)
- What can cluster analysis offer in investing?-Measuring structural changes in the investment universe (2021)
- Interconnectedness in the global financial market (2021)
- Measuring financial asset return and volatility spillovers, with application to global equity markets (2009)
- International stock return comovements (2009)
- Financial contagion during COVID–19 crisis (2021)
- Equity portfolio diversification (2008)
- Risk parity and diversification (2011)
- Honey, I Shrunk the Sample Covariance Matrix (2004)
- Common risk factors in the returns on stocks and bonds (1993)
- A five-factor asset pricing model (2015)
- Industry classification schemes: An analysis and review (2016)
- Industry classifications and return comovement (2007)
0 of 51 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.