Paper: arXiv 2406.03652

Authors: Duy Khanh Lam

Abstract

This paper investigates the problem of ensembling multiple strategies for sequential portfolios to outperform individual strategies in terms of long-term wealth. Due to the uncertainty of strategies’ performances in the future market, which are often based on specific models and statistical assumptions, investors often mitigate risk and enhance robustness by combining multiple strategies, akin to common approaches in collective learning prediction. However, the absence of a distribution-free and consistent preference framework complicates decisions of combination due to the ambiguous objective. To address this gap, we introduce a novel framework for decision-making in combining strategies, irrespective of market conditions, by establishing the investor’s preference between decisions and then forming a clear objective. Through this framework, we propose a combinatorial strategy construction, free from statistical assumptions, for any scale of component strategies, even infinite, such that it meets the determined criterion. Finally, we test the proposed strategy along with its accelerated variant and some other multi-strategies. The numerical experiments show results in favor of the proposed strategies, albeit with small tradeoffs in their Sharpe ratios, in which their cumulative wealths eventually exceed those of the best component strategies while the accelerated strategy significantly improves performance.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • 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 presents a formal, distribution-free framework using online learning theory and combinatorial analysis, indicating high mathematical sophistication, while the experiments utilize a 27-year real market dataset (CRSP) with specific performance metrics like Sharpe ratios and cumulative wealth, demonstrating substantial empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Ensemble multiple portfolio strategies<br>for long-term investment robustness"] --> B{"Problem:"}
  B --> C["No distribution-free framework<br>for combining strategies"]
  C --> D["Methodology:<br>Novel Preference Framework"]
  D --> E["Key Process:<br>Define investor preference<br>between decisions"]
  E --> F["Construction:<br>Combinatorial strategy<br>free from statistical assumptions"]
  F --> G["Experiments:<br>Test on financial data<br>vs. baselines"]
  G --> H["Outcome:<br>Proposed strategy exceeds<br>best component wealth"]
  H --> I["Outcome:<br>Accelerated variant shows<br>significant performance boost"]

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 distribution‑free preference framework for ensembling strategies and a combinatorial strategy construction that works for any number of component strategies.

Source sentence
we introduce a novel framework for decision-making in combining strategies, irrespective of market conditions, by establishing the investor’s preference between decisions and then forming a clear objective. Through this framework, we propose a combinatorial strategy construction, free from statistical assumptions, for

Assumes 5

  • No commission fees are charged on trades. execution
    Source sentence
    assuming there are no commission fees
  • The infinite sequence of market data is deterministic but unknown, not random. statistical
    Source sentence
    the infinite sequence x∞1 is treated as deterministic but unknown data, rather than as random, throughout this paper.
  • Component strategies are no‑short portfolios. market
    Source sentence
    sequentially decides combinations of various observable component strategies of no-short portfolios over time.
  • Initial capital is set to one unit. other
    Source sentence
    Given the initial capital S0b0:= S0 = 1
Show 1 more
  • The investor’s preference is defined on cumulative wealth of a strategy. model
    Source sentence
    we introduce a general framework of distribution-free and time-consistent preference based on the cumulative wealth of a strategy, which constitutes the investor’s goal.

Data

Backtest sample: six blue‑chip stocks (CRSP data) · equities

  • CRSP daily stock file · The Center for Research in Security Prices, LLC market-prices public
    27 years · six blue‑chip stock symbols (AMD, BA, HON, KO, MMM, XOM)
    Source sentence
    The dataset utilized for these experiments is sourced from The Center for Research in Security Prices, LLC, spanning a comprehensive range of 27 years, which includes many historically significant market events and comprises six blue-chip stock symbols.

Recognised sources: CRSP

Costs

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

Source sentence
assuming there are no commission fees

Finds 3

  • The proposed strategies eventually exceed the cumulative wealth of the best component strategies, with the accelerated variant improving performance further, despite small Sharpe‑ratio tradeoffs. performance
    Source sentence
    their cumulative wealths eventually exceed those of the best component strategies while the accelerated strategy significantly improves performance.
  • The UC (universal combinatorial) strategy outperforms the best mean‑variance component strategy in cumulative wealth across all partition experiments. performance
    Source sentence
    Overall, all partitions lead to the outperformance of the UC strategy over the best M-V strategy in terms of cumulative wealth.
  • The WAE strategy performs poorly compared to the UC strategy across all partitions, even though it attains a higher Sharpe ratio. performance
    Source sentence
    The results demonstrate that the WAE strategy performs poorly in comparison to the UC strategy across all partitions, despite its higher Sharpe ratio.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 16 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 (23)

  1. MONEY: Ensemble learning for stock price movement prediction via a convolutional network with adversarial hypergraph model (2023) — link
  2. An integrated approach of ensemble learning methods for stock index prediction using investor sentiments (2023) — link
  3. An Intelligent Fusion Model with Portfolio Selection and Machine Learning for Stock Market Prediction (2022) — link
  4. Dynamic stock-decision ensemble strategy based on deep reinforcement learning (2022) — link
  5. Open Problem: Fast and Optimal Online Portfolio Selection (2020)
  6. A comprehensive evaluation of ensemble learning for stock-market prediction (2020) — link
  7. Online Portfolio Selection (2018) — link
  8. OSQP: an operator splitting solver for quadratic programs (2017) — link
and 15 more
  1. A Rewriting System for Convex Optimization Problems (2017) — link
  2. CVXPY: A Python-Embedded Modeling Language for Convex Optimization (2016) — link
  3. Prediction, learning, and games (2006) — link
  4. On prediction of individual sequences (1999) — link
  5. Regret in the On-Line Decision Problem (1999) — link
  6. On‐Line Portfolio Selection Using Multiplicative Updates (1998) — link
  7. Universal portfolio selection (1998) — link
  8. Universal Portfolios With and Without Transaction Costs (1997) — link
  9. Universal portfolios with side information (1996) — link
  10. How to use expert advice (1993) — link
  11. Complexity of strings in the class of Markov sources (1986) — link
  12. The performance of universal encoding (1981) — link
  13. ‘Elements of Information Theory, 2nd edition’ (2006)
  14. Universal Portfolios (1996) — link
  15. I am thankful to my advisor, Giulio Bottazzi, and Daniele Giachini at Scuola Superiore Sant’Anna for providing the dataset for the experiments and their helpful comments on this paper

0 of 23 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.