Paper: arXiv 2404.00187

Authors: Bahar Arslan, Vanni Noferini, Spyridon Vrontos

Abstract

We investigate an application of network centrality measures to portfolio optimization, by generalizing the method in [“Pozzi, Di Matteo and Aste, \emph{“Spread of risks across financial markets: better to invest in the peripheries”}, Scientific Reports 3:1665, 2013”], that however had significant limitations with respect to the state of the art in network theory. In this paper, we systematically compare many possible variants of the originally proposed method on S&P 500 stocks. We use daily data from twenty-seven years as training set and their following year as test set. We thus select the best network-based methods according to different viewpoints including for instance the highest Sharpe Ratio and the highest expected return. We give emphasis in new centrality measures and we also conduct a thorough analysis, which reveals significantly stronger results compared to those with more traditional methods. According to our analysis, this graph-theoretical approach to investment can be used successfully by investors with different investment profiles leading to high risk-adjusted returns.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.5)

Why this score: The paper presents advanced network theory concepts and mathematical derivations for various centrality measures, indicating high mathematical complexity. It is empirically rigorous with a large-scale backtest on S&P 500 data over 28 years, using Sharpe ratio and expected returns for model selection, though it lacks explicit code or live implementation details.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Optimize portfolio using graph centralities"] --> B["Data Input<br>S&P 500 stocks (27 years daily data)"]
  B --> C["Methodology<br>Compute centrality variants & compare"]
  C --> D["Computational Process<br>Backtest on training set & validate on test set"]
  D --> E{"Evaluation Metrics"}
  E --> F["Max Sharpe Ratio<br>Best Risk-Adjusted Returns"]
  E --> G["Max Expected Return<br>Highest Profitability"]
  E --> H["New Centrality Measures<br>Superior to traditional methods"]
  F --> I["Outcome<br>Successful strategy for diverse investor profiles"]

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 systematic comparison of many graph‑centrality based portfolio selection methods on S&P 500 stocks, introducing a wide range of centrality measures, subgraph centralities and a simple thresholding approach for adjacency matrices.

Source sentence
We propose the use of a much wider range of centrality measures, including the fairly new NBTW and the use of subgraph centralities (Exponential, Katz and NBTW) which perform excellently on our tests.

Assumes 3

  • The investor can hold at most a fixed number m of stocks from the S&P 500 constituent list. market
    Source sentence
    We consider an investor who invest in a maximum of m stocks from the constituent list of the S&P 500.
  • Daily returns are used for training (1990‑2016) and testing (1991‑2017) without accounting for transaction costs. execution
    Source sentence
    We use the data in a daily frequency. The train set is equal to 1990 : 2016 and the test set is 1991 : 2017.
  • A target annual return rtarget is set in the optimization, lowered only if not achievable. model
    Source sentence
    We set an annual target return, rtarget, in the optimization schemes used. The target return is not always achieved by the optimization procedure; in these cases, the target is lowered to the highest possible positive value.

Data

  • S&P 500 daily returns market-prices public
    1990-01 to 2017-12 · S&P 500 constituents
    Source sentence
    The data employed cover the period from January 1990 to December 2017. We use the data in a daily frequency.

Costs

Not modeled A trading or execution result is reported without costs.

Finds 5

  • Portfolios built from peripheral stocks using centrality measures such as Exponential, Katz Min, NBTW Subgraph and Exponential Subgraph achieve the highest Sharpe Ratios. performance
    Source sentence
    Our findings suggest that the following factors contribute to achieving high Sharpe Ratios (SR) in the portfolios: • Centrality measures such as Exponential, Katz Min, NBTW Subgraph and Exponential Subgraph. • Constructing portfolios predominantly with peripheral stocks.
  • A simple threshold (0.5 or 0.6) for constructing adjacency matrices outperforms more sophisticated filtering techniques. performance
    Source sentence
    We have proposed the use of threshold instead of other more sophisticated filtering techniques. We have found that in spite of its simplicity threshold actually does better and have depicted the most used threshold values.
  • A much wider range of centrality measures, including the new NBTW variants, perform very well compared to traditional methods. performance
    Source sentence
    We have tested a much wider range of centrality measures, including the fairly new NBTW ones which perform very well.
  • Equally weighted portfolios, despite their simplicity, outperform minimum‑variance or mean‑variance portfolios in most cases. performance
    Source sentence
    Equally weighted portfolios, in spite of their simplicity, outperform minimum variance or mean variance in most o…
  • When the performance metric is expected return, portfolios built from central stocks using subgraph centralities (Exponential Subgraph, NBTW Subgraph, Katz Subgraph) and a threshold of 0.7 achieve the highest returns. performance
    Source sentence
    • Employing subgraph centrality measures such as Exponential Subgraph, NBTW Sub‑graph, and Katz Subgraph. • Constructing portfolios predominantly with central stocks. •

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 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 (21)

  1. Weighted Enumeration of Nonbacktracking Walks on Weighted Graphs (2022) — link
  2. On the exponential generating function for non-backtracking walks (2018) — link
  3. The Deformed Graph Laplacian and Its Applications to Network Centrality Analysis (2018) — link
  4. Non-backtracking walk centrality for directed networks (2018) — link
  5. Matching exponential-based and resolvent-based centrality measures (2016) — link
  6. Spread of risk across financial markets: better to invest in the peripheries (2013) — link
  7. Network Properties Revealed through Matrix Functions (2010) — link
  8. Networks: An Introduction (2010) — link
and 13 more
  1. Comparison of Prim and Kruskal on Shanghai and Shenzhen 300 Index Hierarchical Structure Tree (2009) — link
  2. Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy? (2009) — link
  3. Correlation based networks of equity returns sampled at different time horizons (2006) — link
  4. Improved estimation of the covariance matrix of stock returns with an application to portfolio selection (2003) — link
  5. Geometry of minimum spanning trees on scale-free networks (2003) — link
  6. Modern Portfolio Theory, 1950 to Date (1997) — link
  7. Portfolio Selection (1971) — link
  8. Structure of complex networks (2005)
  9. Risk and asset allocation (2005) — link
  10. Finland. Supported by an Academy of Finland grant (Suomen Akatemian p¨a¨at¨os 331240)
  11. The methods proposed outperform other previously proposed methods such as the MST
  12. Using the weights calculated in step 3, we invest in these portfolios in year t + 1, and calculate the portfolios realised returns
  13. We calculate the weights of equally-weighted and optimal portfolios for these m stocks based on their returns in year t

0 of 21 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.