Paper: arXiv 2509.24151

Authors: Mingshu Li, Dhruv Desai, Jerinsh Jeyapaulraj, Philip Sommer, Riya Jain, Peter Chu, Dhagash Mehta

Abstract

Accurately measuring portfolio similarity is critical for a wide range of financial applications, including Exchange-traded Fund (ETF) recommendation, portfolio trading, and risk alignment. Existing similarity measures often rely on exact asset overlap or static distance metrics, which fail to capture similarities among the constituents (e.g., securities within the portfolio) as well as nuanced relationships between partially overlapping portfolios with heterogeneous weights. We introduce STRAPSim (Semantic, Two-level, Residual-Aware Portfolio Similarity), a novel method that computes portfolio similarity by matching constituents based on semantic similarity, weighting them according to their portfolio share, and aggregating results via residual-aware greedy alignment. We benchmark our approach against Jaccard, weighted Jaccard, as well as BERTScore-inspired variants across public classification, regression, and recommendation tasks, as well as on corporate bond ETF datasets. Empirical results show that our method consistently outperforms baselines in predictive accuracy and ranking alignment, achieving the highest Spearman correlation with return-based similarity. By leveraging constituent-aware matching and dynamic reweighting, portfolio similarity offers a scalable, interpretable framework for comparing structured asset baskets, demonstrating its utility in ETF benchmarking, portfolio construction, and systematic execution.

Complexity vs Empirical Score

  • Math Complexity: 3.5/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper introduces a novel similarity metric (STRAPSim) that uses dynamic matching and residual updates, which involves some algorithmic design but does not rely on advanced stochastic calculus or heavy theoretical proofs, resulting in moderate math complexity. It is highly empirical, with extensive benchmarking against multiple baselines on real-world financial datasets (including corporate bond ETFs) and uses standard predictive metrics like Spearman correlation, indicating strong backtest readiness and data/implementation focus.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Develop STRAPSim"] --> B["Data Sources<br>ETF Datasets"]
  B --> C["Methodology<br>Three-Step Process"]
  C --> D1["Semantic Matching<br>Constituent Similarity"]
  C --> D2["Weight Alignment<br>Portfolio Share"]
  C --> D3["Residual-Aware<br>Greedy Aggregation"]
  D1 & D2 & D3 --> E["Comparison<br>Baselines vs STRAPSim"]
  E --> F["Outcomes<br>Superior Performance"]
  F --> G["Spearman Correlation<br>Ranking Alignment"]
  F --> H["ETF Recommendation<br>Portfolio Trading"]

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 STRAPSim, a novel portfolio similarity metric that matches constituents by semantic similarity, weights them by portfolio share, and aggregates via residual‑aware greedy alignment.

Source sentence
We introduce STRAPSim (Semantic, Two-level, Residual-Aware Portfolio Similarity), a novel method that computes portfolio similarity by matching constituents based on semantic similarity, weighting them according to their portfolio share, and aggregating results via residual-aware greedy alignment.

Assumes 2

  • ETFs with greater constituent‑level similarity should exhibit higher return correlation. market
    Source sentence
    The underlying assumption is that ETFs with greater constituent-level similarity should exhibit higher return correlation.
  • Portfolio trades are executed as all‑or‑none baskets without additional market impact beyond the reported similarity metric. execution
    Source sentence
    Portfolio trading involves executing a basket of bonds as a single, all-or-none transaction with one or more market-makers, rather than trading each bond individually.

Data

  • Toy public datasets (Iris, Breast Cancer, Big Mac, Movie Rating) · UCI/Kaggle synthetic public
    Source sentence
    Table 1 provides a brief summary of the toy datasets employed in this study, including datasets like Big Mac, breast cancer, and others.
  • Corporate Bond ETFs market-prices public
    March 2024 portfolio data; monthly retur · 20 corporate bond ETFs comprising 6,870 bonds (average 511 bonds per ETF)
    Source sentence
    The experiment was conducted on 20 corporate bond ETFs, which consists of 6870 different bonds in total, with an average value of 511 bonds per ETF. The similarity was computed using March 2024 portfolio data and return correlations based on monthly returns from February 2022 to March 2024.

Costs

Not applicable The paper reports no trading or execution result.

Finds 1

  • STRAPSim achieves the highest average Spearman rank correlation (0.6783) with monthly return correlations, outperforming Jaccard, weighted Jaccard, and BertScore. performance
    Source sentence
    STRAPSim outperforms the other similarity metrics, achieving higher correlation coefficients and lower p-values, which demonstrates a stronger monotonic relationship with the correlation of monthly returns.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 7 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.