Paper: arXiv 2307.04045
Authors: Alexander Nikiporenko
Abstract
A financial portfolio contains assets that offer a return with a certain level of risk. To maximise returns or minimise risk, the portfolio must be optimised - the ideal combination of optimal quantities of assets must be found. The number of possible combinations is vast. Furthermore, to make the problem realistic, constraints can be imposed on the number of assets held in the portfolio and the maximum proportion of the portfolio that can be allocated to an asset. This problem is unsolvable using quadratic programming, which means that the optimal solution cannot be calculated. A group of algorithms, called metaheuristics, can find near-optimal solutions in a practical computing time. These algorithms have been successfully used in constrained portfolio optimisation. However, in past studies the computation time of metaheuristics is not limited, which means that the results differ in both quality and computation time, and cannot be easily compared. This study proposes a different way of testing metaheuristics, limiting their computation time to a certain duration, yielding results that differ only in quality. Given that in some use cases the priority is the quality of the solution and in others the speed, time limits of 1, 5 and 25 seconds were tested. Three metaheuristics - simulated annealing, tabu search, and genetic algorithm - were evaluated on five sets of historical market data with different numbers of assets. Although the metaheuristics could not find a competitive solution in 1 second, simulated annealing found a near-optimal solution in 5 seconds in all but one dataset. The lowest quality solutions were obtained by genetic algorithm.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 4.5/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper involves advanced mathematical concepts like NP-hardness and quadratic mixed-integer programming, but its empirical rigor is limited to small-scale historical data tests without live backtesting or trading infrastructure, placing it in the theoretical experimentation domain.
Research Flowchart
flowchart TD
A["Research Goal<br>Limit Metaheuristic Computation Time<br>for Portfolio Optimisation Comparison"] --> B{"Key Methodology<br>Three Time Limits: 1s, 5s, 25s"}
B --> C["Input Data<br>Five historical market datasets<br>varying asset counts"]
C --> D["Computational Process<br>Execute three metaheuristics on datasets<br>Simulated Annealing, Tabu Search, Genetic Algorithm"]
D --> E["Key Findings"]
subgraph E ["Outcomes"]
E1["1 Second<br>No competitive solutions found"]
E2["5 Seconds<br>SA achieved near-optimal solutions<br>GA yielded lowest quality"]
E3["25 Seconds<br>Extended testing confirms time-quality tradeoff"]
end
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 evaluating metaheuristics under fixed time limits to compare solution quality directly.
Source sentence
This study proposes a different way of testing metaheuristics, limiting their computation time to a certain duration, yielding results that differ only in quality.
Assumes 5
- Execution time for market orders is less than one second, making sub‑second solutions potentially too late for real‑world use. execution
Source sentence
Given that the execution time for market orders is less than a second[6], a solution found in 5 seconds may already be too late.
- Backtesting on historical data can approximate real trading performance despite risks of overfitting. model
Source sentence
Backtesting involves a number of risks, the most significant of which is overfitting, which results in the algorithms only performing well on a specific dataset[37].
- Using multiple datasets with different numbers of assets mitigates overfitting risk. data
Source sentence
To prevent this, the metaheuristics are tested on five datasets containing different numbers of assets.
- Solution quality is measured by deviation from the unconstrained efficient frontier (UEF). statistical
Source sentence
These values have been calculated using Numerical Algorithms Group library routine E04NAF[44] for evaluating the quality of solution portfolios by measuring their deviation from the UEF[12].
Show 1 more
- Stochastic metaheuristics require multiple runs and averaging to obtain reliable quality estimates. model
Source sentence
Furthermore, given that all the metaheuristics chosen for this study make stochastic decisions, the quality of their solutions is not consistent[22]. Therefore, each algorithm is tested several times and the results are averaged.
Data
No dataset named in the text.
Costs
Unclear
Finds 5
- Simulated annealing (SA) found near‑optimal solutions within 5 seconds for all but one dataset, outperforming the other metaheuristics. performance
Source sentence
Although the metaheuristics could not find a competitive solution in 1 second, simulated annealing found a near-optimal solution in 5 seconds in all but one dataset.
- Genetic algorithm (GA) produced the lowest‑quality solutions and failed to find competitive results even at 5 seconds. performance
Source sentence
The lowest quality solutions were obtained by genetic algorithm.
- SA showed the smallest percentage improvement between 5‑25 seconds across all datasets, indicating faster convergence. performance
Source sentence
SA offers the lowest percentage of improvement between 5-25 seconds for all datasets.
- MPEs obtained under the 25‑second limit were smaller than those reported in prior studies for all but one case (GA on Nikkei‑225). performance
Source sentence
... MPEs obtained under the 25-second time limits in this study are smaller in all but one case, the GA on Nikkei‑225.
- Compared to a previous study, SA achieved a lower average MPE (1.2507 vs 1.4391), demonstrating improved solution quality. performance
Source sentence
Compared to the results obtained in [13], only the SA implemented in this study shows better results: in [13] the average MPE is 1.4391 compared to 1.2507 in this study.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 24 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.