Paper: arXiv 2308.11202

Authors: Kapil Panda

Abstract

Portfolio optimization is a task that investors use to determine the best allocations for their investments, and fund managers implement computational models to help guide their decisions. While one of the most common portfolio optimization models in the industry is the Markowitz Model, practitioners recognize limitations in its framework that lead to suboptimal out-of-sample performance and unrealistic allocations. In this study, I refine the Markowitz Model by incorporating machine learning to improve portfolio performance. By using a hierarchical clustering-based approach, I am able to enhance portfolio performance on a risk-adjusted basis compared to the Markowitz Model, across various market factors.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • 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 employs advanced mathematical techniques like hierarchical clustering, recursive bisection, and quasi-diagonalization of covariance matrices, indicating moderate-to-high complexity. Empirical rigor is strong due to the use of long-term CRSP and Ken French data (1965-2022), multiple look-back windows, and comparative out-of-sample performance metrics against the Markowitz model.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Enhance Markowitz Model<br>using Machine Learning"] --> B["Input: Financial Asset Data<br>Historical Returns & Volatility"]
  
  B --> C["Key Methodology<br>Hierarchical Clustering<br>to group correlated assets"]
  
  C --> D["Computational Process<br>Apply Markowitz Optimization<br>on Clusters vs. Individual Assets"]
  
  D --> E{"Evaluation"}
  E -->|"Out-of-Sample"| F["Key Finding: Hierarchical Model<br>Superior Risk-Adjusted Returns"]
  E -->|"Constraints"| G["Key Finding: More Realistic<br>Allocations vs. Standard Markowitz"]
  
  F & G --> H["Outcome<br>Refined Portfolio Optimization Framework"]

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 hierarchical clustering‑based portfolio optimization model that refines the Markowitz framework using machine learning to improve risk‑adjusted performance.

Source sentence
I refine the Markowitz Model by incorporating machine learning to improve portfolio performance. By using a hierarchical clustering-based approach, I am able to enhance portfolio performance on a risk-adjusted basis compared to the Markowitz Model

Assumes 5

  • All investors are assumed to have the same credit and the model does not account for brokerage and tax charges. model
    Source sentence
    assuming that all investors have the same credit and failing to account for additional charges (i.e., brokerage and other taxes that can sway investors’ decisions)
  • Excess returns are computed by subtracting the risk‑free rate from raw stock returns. data
    Source sentence
    I start by taking each company’s monthly return values from the CRSP data and subtracting them from the risk-free rates I received from the Ken French data library to get the data’s excess return value
  • A long‑short strategy is used for portfolio construction. execution
    Source sentence
    Finally, I allocate and build the portfolio by using a long-short strategy which allows me to take advantage of both upward and downward market movements
  • Stocks are grouped into clusters based on correlation‑derived distance metrics. model
    Source sentence
    I devise a clustering algorithm that groups assets based on their correlations and similarities to form clusters
Show 1 more
  • The analysis uses a 3‑month buy‑and‑hold holding period. execution
    Source sentence
    portfolios were analyzed using a 3-month buy-and-hold, long-short strategy

Data

  • CRSP monthly stock returns · Center for Research in Security Prices (CRSP) market-prices public
    1965-2022 · common stocks listed on NYSE, AMEX, and NASDAQ
    Source sentence
    In this research, I used the monthly returns of common stocks listed on the three major U.S. stock exchanges: NYSE, AMEX, and NASDAQ, as reported by the Center of Research in Security Prices (CRSP) from 1965-2022
  • Ken French risk‑free rate data · Ken French data library benchmark public
    Source sentence
    Using these filters, I specifically obtained data concerned with price, return, shares out, CFACPR, and weighted value returns, for thousands of companies in various sectors. Additionally, I also obtained monthly risk-free data from Ken French’s data

Recognised sources: Kenneth French Data Library CRSP

Costs

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

Finds 3

  • The clustered portfolio yields higher average monthly excess returns and Sharpe‑ratio improvements (≈0.34% and 0.08% respectively) while maintaining comparable standard deviation to the Markowitz portfolio. performance
    Source sentence
    the clustered portfolio can produce higher risk-adjusted returns and Sharpe-ratio measures by roughly 0.34% and 0.08%, respectively, while maintaining comparable standard deviation measures
  • The Markowitz portfolio exhibits lower average monthly excess returns and Sharpe‑ratios (differences of 0.15% and 0.32%) and a significantly higher standard deviation (5.29%) compared to the clustered portfolio. performance
    Source sentence
    the Markowitz portfolio, which has a lower average monthly excess return rate and lower Sharpe-ratios, with a difference of 0.15% and 0.32%, respectively, while having significantly higher standard deviation rates of 5.29%, when compared to the clustered portfolio
  • Across 12‑month, 60‑month, and 120‑month look‑back periods, the clustered portfolio consistently outperforms the Markowitz model on all three metrics. performance
    Source sentence
    With the 60-month and 120-month look-back windows, we can see that the trend continues where the clustered portfolio produces average monthly excess returns, standard deviation, and Sharpe-ratios that outperform the market value while notably exceeding the values produced by the Markowitz portfolio

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 8 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.