Paper: arXiv 2507.05994
Authors: Duy Khanh Lam
Abstract
In the online portfolio optimization framework, existing learning algorithms generate strategies that yield significantly poorer cumulative wealth compared to the best constant rebalancing portfolio in hindsight, despite being consistent in asymptotic growth rate. While this unappealing performance can be improved by incorporating more side information, it raises difficulties in feature selection and high-dimensional settings. Instead, the inherent serial dependence of assets’ returns, such as day-of-the-week and other calendar effects, can be leveraged. Although latent serial dependence patterns are commonly detected using large training datasets, this paper proposes an algorithm that learns such dependence using only gradually revealed data, without any assumption on their distribution, to form a strategy that eventually exceeds the cumulative wealth of the best constant rebalancing portfolio. Moreover, the classical Kelly criterion, which requires independent assets’ returns, is generalized to accommodate serial dependence in a market modeled as an independent and identically distributed process of random matrices. In such a stochastic market, where existing learning algorithms designed for stationary processes fail to apply, the proposed learning algorithm still generates a strategy that asymptotically grows to the highest rate among all strategies, matching that of the optimal strategy constructed under the generalized Kelly criterion. The experimental results with real market data demonstrate the theoretical guarantees of the algorithm and its performance as expected, as long as serial dependence is significant, regardless of the validity of the generalized Kelly criterion in the experimental market. This further affirms the broad applicability of the algorithm in general contexts.
Complexity vs Empirical Score
- Math Complexity: 9.0/10
- Empirical Rigor: 4.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper is highly theoretical, featuring dense mathematical derivations, a generalized Kelly criterion, and proofs of asymptotic optimality, placing it in the Lab Rats quadrant. While it uses real market data for experiments, the empirical section is relatively brief and lacks the comprehensive backtesting and implementation details typical of higher-rigor studies.
Research Flowchart
flowchart TD A["Research Goal: Beat Best Constant Rebalancing Portfolio<br>with Serial Dependence"] --> B["Method: Generalized Kelly Criterion<br>and Universal Learning Algorithm"] B --> C["Input: Real Market Data<br>with Significant Serial Dependence"] C --> D["Process: Learn Latent Patterns<br>Gradually & Distribution-Free"] D --> E["Process: Generalize Kelly for<br>Stochastic Market/Random Matrices"] E --> F["Outcome: Asymptotically Highest<br>Wealth Growth Rate"] F --> G["Outcome: Cumulative Wealth<br>Exceeds Best CRP in Hindsight"]
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 learning algorithm that leverages serial dependence in asset returns, without distributional assumptions, to construct a dynamic strategy that eventually exceeds the cumulative wealth of the best constant rebalancing portfolio and matches the optimal growth rate of the generaliz
Source sentence
this paper proposes an algorithm that learns such dependence using only gradually revealed data, without any assumption on their distribution, to form a strategy that eventually exceeds the cumulative wealth of the best constant rebalancing portfolio.
Assumes 6
- No commission fees are charged on trades. execution
Source sentence
assuming no commission fees, the cumulative wealth and the growth rate after n time periods of investment yielded by a generic strategy
- Asset returns are positive and the strategy is self‑financed. market
Source sentence
Noting that, since the assets’ returns are positive and the strategy is self‑financed, the resulting Sn
- The return process can be modeled as a block‑wise i.i.d. (k‑cyclic) stochastic process. model
Source sentence
the block‑wise i.i.d. stochastic process X_{kt+1},...,X_{kt+k}
- Serial dependence in returns is significant enough to be exploitable. statistical
Source sentence
as long as serial dependence is significant, regardless of the validity of the generalized Kelly criterion in the experimental market
Show 2 more
- The true cycle length k lies within a known small set of possibilities. model
Source sentence
As long as the true k lies within a known small set of possibilities k1,...,kz, the constructed strategy
- Trading is performed at daily closing prices without slippage. execution
Source sentence
daily adjusted closing prices in USD over a long time horizon of 27 years
Data
US stocks · equities · daily
- CRSP daily stock file market-prices public
Source sentence
the dataset used in the experiments covers the daily adjusted closing prices in USD over a long time horizon of 27 years, from December 31, 1992, to December 31, 2019, i.e., 6798 trading days, provided by the Center for Research in Security Prices, LLC (CRSP).
Costs
Not modeled
Finds 4
- The k‑PUP strategies (e.g., 2‑PUP and 6‑PUP) eventually outperform the best constant strategy in cumulative wealth, with 2‑PUP significantly outperforming 6‑PUP. performance
Source sentence
... is exceeded by those of the 2-PUP and 6-PUP strategies, with the former significantly outperforming the latter
- The Universal Portfolio strategy does not achieve the highest asymptotic growth rate, invalidating the Kelly criterion in the experimental market. theoretical
Source sentence
the asymptotic growth rate of the Universal Portfolio strategy is not the highest, thereby invalidating the Kelly criterion in the experimental market
- The best k‑cyclic constant strategy yields higher cumulative wealth than the best constant strategy, with a large gap as k increases. performance
Source sentence
the best k‑cyclic constant strategy yields higher cumulative wealth than that yielded by the best constant strategy, with a large gap as k increases
- The proposed algorithm asymptotically attains the optimal growth rate among all dynamic strategies, matching the rate of the optimal strategy under the generalized Kelly criterion. theoretical
Source sentence
... generates a strategy that asymptotically grows to the highest rate among all strategies, matching that of the optimal strategy constructed under the generalized Kelly criterion
Checked by others
Supported by
- Betting Around the Clock: Time Change and Long Term Model Risk
Compare the findings
This paper: The Universal Portfolio strategy does not achieve the highest asymptotic growth rate, invalidating the Kelly criterion in the experimental market.
That paper: In a general semi‑martingale setting the Kelly rule does not maximize the average growth rate unless log‑returns are normally distributed.
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.