Paper: arXiv 2310.14881
Authors: Yuanrong Wang, Antonio Briola, Tomaso Aste
Abstract
Modern portfolio optimization is centered around creating a low-risk portfolio with extensive asset diversification. Following the seminal work of Markowitz, optimal asset allocation can be computed using a constrained optimization model based on empirical covariance. However, covariance is typically estimated from historical lookback observations, and it is prone to noise and may inadequately represent future market behavior. As a remedy, information filtering networks from network science can be used to mitigate the noise in empirical covariance estimation, and therefore, can bring added value to the portfolio construction process. In this paper, we propose the use of the Statistically Robust Information Filtering Network (SR-IFN) which leverages the bootstrapping techniques to eliminate unnecessary edges during the network formation and enhances the network’s noise reduction capability further. We apply SR-IFN to index component stock pools in the US, UK, and China to assess its effectiveness. The SR-IFN network is partially disconnected with isolated nodes representing lesser-correlated assets, facilitating the selection of peripheral, diversified and higher-performing portfolios. Further optimization of performance can be achieved by inversely proportioning asset weights to their centrality based on the resultant network.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper introduces advanced network science concepts (IFN, SR-IFN with bootstrapping) and detailed mathematical derivations for covariance filtering, aligning with high mathematical complexity. It includes rigorous empirical testing across multiple markets (US, UK, China), incorporates transaction costs, and presents backtest-ready portfolio selection/optimization methods, demonstrating strong data and implementation focus.
Research Flowchart
flowchart TD
A["Research Goal:<br>Enhance Portfolio Optimization<br>using Network Science"] --> B["Data: US, UK, China<br>Index Component Stocks"]
B --> C["Compute Empirical Covariance<br>Matrix from Historical Returns"]
C --> D["Construct Statistically Robust<br>Information Filtering Network<br>SR-IFN using Bootstrapping"]
D --> E{"Network Analysis"}
E --> F["Identify Peripheral Nodes<br>Low Centrality /<br>Isolated Assets"]
E --> G["Select Diversified Portfolio<br>Assets with Low Correlation"]
F --> H["Optimization:<br>Weight Assets Inversely<br>Proportional to Centrality"]
G --> H
H --> I["Key Outcomes:<br>Reduced Noise / Enhanced Stability /<br>Superior Risk-Adjusted Returns"]
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 the Statistically Robust Information Filtering Network (SR-IFN), a bootstrapped edge‑elimination method that improves noise reduction in empirical covariance estimation for portfolio construction.
Source sentence
we propose the use of the Statistically Robust Information Filtering Network (SR-IFN) which leverages the bootstrapping techniques to eliminate unnecessary edges during the network formation and enhances the network’s noise reduction capability further.
Assumes 5
- Portfolio weights are set equal to 1/N for simplicity and controlled comparison. execution
Source sentence
Consequently, the weights of the portfolio are maintained at 1/𝑁, where 𝑁represents the total number of assets in the chosen portfolio, for the sake of simplicity and controlled comparison.
- Rebalancing occurs every t_rebalance days with a look‑back window of t_lookback days for estimating correlations. execution
Source sentence
The portfolio undergoes rebalancing every 𝑡rebalance days, with a historical look-back period of 𝑡lookback days for the measurement of empirical correlation and other historical statistical properties.
- Transaction costs are modeled as a flat 20 basis‑point charge per trade. execution
Source sentence
Experiments are included both with and without transaction costs of 20 basis points (bps) to simulate commission and bid-ask spread costs.
- Portfolio size is restricted to between 50 and 100 assets to reduce tail variance. model
Source sentence
we restrict our analysis to portfolios comprising more than 50 and less than 100 assets, in order to mitigate the high variance at the tail of the performance distribution.
Show 1 more
- Temporal sequential dependence is reduced by randomly bootstrapping observations in each repetition. statistical
Source sentence
Temporal sequential dependence is reduced by randomly boot‑strapping the observations in each repetition, which also results
Data
- US, UK, and China index component stock pools (NASDAQ, FTSE, HS300) market-prices unknown
Source sentence
Our experiments spanned a decade-long history across three distinct markets, utilizing the first 70% of the data to select parameters such as rebalancing frequency and look-back window size. The results reported are based on the remaining three years of out-of-sample data.
Costs
Modeled
20 basis points (bps) flat fee per trade, used in both in‑sample grid search and out‑of‑sample performance reporting.
Source sentence
Experiments are included both with and without transaction costs of 20 basis points (bps) to simulate commission and bid-ask spread costs.
Finds 5
- SR‑IFN improves Sharpe Ratios by roughly 1% for NASDAQ, 367% for FTSE, and 32% for HS300 when 20 bps transaction costs are applied. performance
Source sentence
Our findings indicate that the deployment of such an innovative approach results in a Sharpe Ratio improvement of 1%, 367%, and 32% with 20bps transaction costs for market indices.
- Peripheral portfolios (PTP) consistently outperform benchmark portfolios (PBP, RBP, CTP) across all three markets, both with and without transaction costs. performance
Source sentence
In all three markets, the two peripheral portfolios, PTP and PBP, both yield superior performance compared to RBP and CTP, suggesting a clear advantage in adopting the peripheral portfolio.
- Out‑of‑sample average Sharpe Ratio for NASDAQ is 1.10 without transaction fees and declines only marginally when 20 bps fees are applied. performance
Source sentence
When examining the out-of-sample period without transaction fees, the average Sharpe Ratio for NASDAQ is 1.10, while with the inclusion of 20bps transaction fees, it marginally dec…
- Optimizing peripheral portfolio weights by centrality measures raises NASDAQ Sharpe from 1.08 to 1.12. performance
Source sentence
For the NASDAQ index, the average Sharpe Ratio is improved from 1.08 to 1.12, representing an
- Risk metrics improve: annualized standard deviation drops by 1.4% and maximum drawdown by 5.1% under SR‑IFN. performance
Source sentence
the risk metrics are notably reduced in both annualized standard deviation of 1.4% and maximum drawdown of 5.1%.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 45 related papers). Coverage is partial, so this is neither replication nor refutation.
4 papers in the archive : Systematic comparison of deep generative models applied to …
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.