Paper: arXiv 2501.12074

Authors: Keon Vin Park

Abstract

Optimizing portfolio performance is a fundamental challenge in financial modeling, requiring the integration of advanced clustering techniques and data-driven optimization strategies. This paper introduces a comparative backtesting approach that combines clustering-based portfolio segmentation and Sharpe ratio-based optimization to enhance investment decision-making. First, we segment a diverse set of financial assets into clusters based on their historical log-returns using K-Means clustering. This segmentation enables the grouping of assets with similar return characteristics, facilitating targeted portfolio construction. Next, for each cluster, we apply a Sharpe ratio-based optimization model to derive optimal weights that maximize risk-adjusted returns. Unlike traditional mean-variance optimization, this approach directly incorporates the trade-off between returns and volatility, resulting in a more balanced allocation of resources within each cluster. The proposed framework is evaluated through a backtesting study using historical data spanning multiple asset classes. Optimized portfolios for each cluster are constructed and their cumulative returns are compared over time against a traditional equal-weighted benchmark portfolio.

Complexity vs Empirical Score

  • Math Complexity: 5.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)

Why this score: The paper employs standard but non-trivial mathematical optimization (maximizing the Sharpe ratio with SLSQP) alongside clustering, which is moderately complex. It demonstrates high empirical rigor through a detailed backtesting framework using real historical data, explicit code implementation, and specific performance metrics (e.g., total return, Sharpe ratio) comparing optimized portfolios to a benchmark.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Optimize Portfolio Performance<br>(Risk-Adjusted Returns)"] --> B["Input Data:<br>Historical Multi-Asset Log-Returns"]
  B --> C["Methodology 1:<br>K-Means Clustering<br>(Group assets by return similarity)"]
  C --> D["Methodology 2:<br>Sharpe Ratio Optimization<br>(Calculate optimal weights per cluster)"]
  D --> E["Computational Process:<br>Backtesting<br>(Construct portfolios over time)"]
  E --> F["Outcome 1:<br>Cluster-Specific<br>Optimized Portfolios"]
  E --> G["Outcome 2:<br>Performance Comparison<br>(Optimized vs. Equal-Weight Benchmark)"]

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 comparative backtesting framework that combines K‑Means clustering of assets with Sharpe ratio‑based portfolio optimization.

Source sentence
This paper introduces a comparative backtesting approach that combines clustering-based portfolio segmentation and Sharpe ratio-based optimization to enhance investment decision-making.

Assumes 5

  • The sum of portfolio weights must equal 1. model
    Source sentence
    The sum of portfolio weights must equal 1: N X i=1 wi = 1, where wi is the weight of the i-th asset.
  • No short selling is allowed (weights are non‑negative). execution
    Source sentence
    No short selling was allowed, ensuring: wi ≥0 ∀i.
  • Historical log‑returns are used as the basis for clustering. data
    Source sentence
    First, we segment a diverse set of financial assets into clusters based on their historical log-returns using K-Means clustering.
  • Training period (2010‑2019) is used for clustering and optimization; testing period (2020‑2024) is used for backtesting. data
    Source sentence
    Training Period: January 1, 2010, to December 31, 2019, used for clustering and portfolio optimization. Testing Period: January 1, 2020, to January 1, 2024, used for backtesting and performance evaluation.
Show 1 more
  • Daily closing prices from Yahoo Finance are accurate and sufficient for return calculation. data
    Source sentence
    Historical stock prices were sourced from Yahoo Finance using Python’s yfinance library.

Data

Backtest sample: 2010 – 2024 · S&P 500 stocks · equities · daily

  • Daily closing prices of ten S&P 500 stocks · Yahoo Finance market-prices public
    2010-01-01 to 2024-01-01 · AAPL, MSFT, GOOGL, AMZN, TSLA, NVDA, META, JPM, V, UNH
    Source sentence
    ... Historical stock prices were sourced from Yahoo Finance using Python’s yfinance library. The dataset covers the period from January 1, 2010, to January 1, 2024.

Recognised sources: Yahoo Finance

Costs

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

Finds 2

Reported performance (Clustering and Sharpe ratio-based optimization): Sharpe 0.84 (benchmark 0.73) · annual return 24.7% · total return 141.0%

  • The selected cluster portfolio (Cluster 2) achieved a total return of 140.98% versus 107.59% for the equal‑weighted benchmark, with higher annualized return (24.67% vs 20.09%) and Sharpe ratio (0.84 vs 0.73). performance
    Source sentence
    The Selected Cluster Portfolio (Cluster 2) achieved a total return of 140.98%, outperforming the benchmark portfolio’s total return of 107.59%.
  • Volatility of the selected cluster portfolio was slightly higher (30.60%) than the benchmark (30.09%). performance
    Source sentence
    The selected cluster portfolio exhibited slightly higher volatility (30.60%) compared to the benchmark portfolio (30.09%).

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 14 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 (7)

  1. Practical Variational Inference for Neural Networks (2011)
  2. Weight Uncertainty in Neural Networks (2015)
  3. Pattern Recognition and Machine Learning (2006)
  4. Time Series Analysis (1994)
  5. Long Short-Term Memory (1997)
  6. Gated Recurrent Unit with Bayesian Regularization (2018)
  7. Bayesian Learning for Neural Networks (1996)

0 of 7 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.