Paper: arXiv 2306.05667
Authors: Andrés García-Medina, Benito Rodriguéz-Camejo
Abstract
This work aims to deal with the optimal allocation instability problem of Markowitz’s modern portfolio theory in high dimensionality. We propose a combined strategy that considers covariance matrix estimators from Random Matrix Theory~(RMT) and the machine learning allocation methodology known as Nested Clustered Optimization~(NCO). The latter methodology is modified and reformulated in terms of the spectral clustering algorithm and Minimum Spanning Tree~(MST) to solve internal problems inherent to the original proposal. Markowitz’s classical mean-variance allocation and the modified NCO machine learning approach are tested on financial instruments listed on the Mexican Stock Exchange~(BMV) in a moving window analysis from 2018 to 2022. The modified NCO algorithm achieves stable allocations by incorporating RMT covariance estimators. In particular, the allocation weights are positive, and their absolute value adds up to the total capital without considering explicit restrictions in the formulation. Our results suggest that can be avoided the risky \emph{“short position”} investment strategy by means of RMT inference and statistical learning techniques.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematics including Random Matrix Theory, Wishart distributions, Marchenko-Pastur law, Tracy-Widom distributions, and Painlevé differential equations, which is dense and highly advanced. Empirically, it applies a moving window backtest from 2018 to 2022 on Mexican Stock Exchange data with specific covariance estimators and portfolio optimization methods, demonstrating implementation-heavy testing.
Research Flowchart
flowchart TD
A["Research Goal: Solve Markowitz's allocation instability<br>in Mexican markets (high dimensionality)"] --> B{"Key Methodology"}
subgraph B ["Proposed Combined Strategy"]
B1["Random Matrix Theory<br>RMT Covariance Estimators"]
B2["Modified Nested Clustered Optimization<br>NCO: Spectral Clustering + MST"]
end
C["Data: BMV Equities<br>Moving Window Analysis (2018–2022)"] --> D["Computational Process"]
subgraph D ["Implementation Steps"]
D1["Estimate Covariance Matrix via RMT"]
D2["Apply Modified NCO Algorithm"]
D3["Compare vs. Classical Markowitz MV"]
end
D --> E{"Key Findings"}
subgraph E ["Outcomes"]
E1["Stable Allocations Achieved"]
E2["Positive Weights (No Explicit Constraints)"]
E3["Short Position Avoidance via RMT/ML"]
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 a combined strategy that uses Random Matrix Theory covariance estimators together with a modified Nested Clustered Optimization allocation method.
Source sentence
We propose a combined strategy that considers covariance matrix estimators from Random Matrix Theory (RMT) and the machine learning allocation methodology known as Nested Clustered Optimization (NCO).
Assumes 5
- The data matrix X has i.i.d. Gaussian entries with zero mean and covariance Σ. statistical
Source sentence
To start with, consider a data matrix X of dimension p × n matrix with i.i.d. Gaussian vector entries Xi ∼ Np(0, Σ), i = 1, . . . , n
- The covariance matrix Σ is positive definite and the number of observations n is at least the number of assets p. model
Source sentence
and is assumed Σ to be positive definite with n ≥p.
- Missing price days (<10% missing) are imputed by linear interpolation. data
Source sentence
Only markets with less than 10% missing days were included. Thus, we study a set of p = 28 return times series of length T = 232 transaction weeks imputed by linear interpolation.
- Returns are standardized by removing the mean, normalizing by cross‑sectional weekly volatility, and then by sample volatility. data
Source sentence
First, remove the sample mean µi of each asset. Second, normalize the returns by the cross-sectional weekly volatility ˆσi,t =
Show 1 more
- The portfolio analysis is performed under a minimum‑variance scenario with risk aversion g = 1, target risk level G = 1 and dimensional factor q = 1/2. execution
Source sentence
In all cases is considered a minimum variance portfolio scenario (g = 1) at a fixed level G = 1 and dimensional factor q = 1/2.
Data
- Weekly closing prices of Mexican Stock Exchange listed companies market-prices public
Source sentence
We consider the close prices of companies listed on the Mexican Stock Exchange from 2017-12-29 to 2022-07-01 at a weekly frequency. The data were collected from https://finance.yahoo.com/, and the returns are computed through the logarithm differences of the closing prices.
Costs
Not modeled
Finds 5
- The linear RMT estimator yields the lowest mean‑square error and mean‑absolute error for both Markowitz and NCO methodologies. performance
Source sentence
M(Ξlinear) 0.000238 0.013284 1.049836
- Using the linear estimator, the gap between out‑of‑sample and in‑sample risks is significantly minimized. statistical
Source sentence
It can be seen that this gap is significantly minimized when we use the linear estimator
- In‑sample risk decreases when the NCO methodology is applied to the raw empirical matrix. performance
Source sentence
while it decreases with NCO (see Fig. 4a)
- The absolute sum of portfolio weights is minimized (≈1) when NCO is combined with the linear estimator. performance
Source sentence
the absolute sum is minimized with NCO under the linear estimator during the study period, obtaining an average value practically equal to one.
- Both Markowitz and NCO methodologies reduce the overestimation of out‑sample risk under the linear and TW estimators, with the combination of NCO and the linear estimator being the best strategy. performance
Source sentence
Markowitz and NCO methodologies reduce the overestimation of the out‑sample risk under the linear and TW estimators (see Fig. 4b). Nevertheless, the combination of NCO with the linear estimator is the best strategy to deal with the out‑sample risk overestimation.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 25 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.