Paper: arXiv 2407.10175
Authors: Parley R Yang, Alexander Y Shestopaloff
Abstract
We employ a Bayesian modelling technique for high dimensional cointegration estimation to construct low volatility portfolios from a large number of stocks. The proposed Bayesian framework effectively identifies sparse and important cointegration relationships amongst large baskets of stocks across various asset spaces, resulting in portfolios with reduced volatility. Such cointegration relationships persist well over the out-of-sample testing time, providing practical benefits in portfolio construction and optimization. Further studies on drawdown and volatility minimization also highlight the benefits of including cointegrated portfolios as risk management instruments.
Complexity vs Empirical Score
- Math Complexity: 9.5/10
- 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 uses advanced high-dimensional Bayesian statistics and complex matrix decompositions (Spike-and-Slab Lasso, VECM) but also demonstrates out-of-sample testing on large stock universes (400-1000 stocks) with specific performance metrics like volatility reduction and Sharpe ratio.
Research Flowchart
flowchart TD A["Research Goal: Construct Low Volatility Portfolio<br>from High-Dimensional Stock Data"] --> B["Data Inputs: Time-series<br>of N Stocks (High-Dimensional)"] B --> C["Methodology: Bayesian Sparse<br>Cointegration Estimation"] C --> D["Computational Process:<br>Posterior Sampling & Model Fitting"] D --> E["Key Outcome 1: Identified<br>Persistent Cointegration Relationships"] E --> F["Key Outcome 2: Constructed<br>Low Volatility Portfolio"] F --> G["Outcome: Reduced Volatility &<br>Effective Risk Management (Drawdown)"]
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 Bayesian modelling technique for high‑dimensional cointegration estimation to build low‑volatility stock portfolios.
Source sentence
We employ a Bayesian modelling technique for high dimensional cointegration estimation to construct low volatility portfolios from a large number of stocks.
Assumes 4
- Stock prices are non‑stationary I(1) while their lag‑differences are stationary I(0). model
Source sentence
Stock prices typically do not satisfy this assumption, though their lag-difference or log lag-difference does (such as in models assuming Brownian or Geometric Brownian motion).
- A sparse Bayesian prior allows reliable estimation with only about two years of data despite high dimensionality. statistical
Source sentence
This brings down the number of observations required to 2 years, while maintaining the capability to detect the underlining cointegration structure amongst circa 400 to 1000 stocks.
- Re‑estimation and portfolio rebalancing are performed on a monthly basis, limiting over‑trading costs. execution
Source sentence
re-estimation hence rebalancing are done on a monthly basis, the risk and cost of over-trading is mitigated from the start.
- High‑market‑cap stocks and indices have only minor component adjustments each quarter. market
Source sentence
and that the indices only have minor adjustments of components per quarter.
Data
US stocks (SPY) · equities · daily
- US stock universe market-prices unknown
Source sentence
the rolling in-sample training period is July 2021 to June 2023, with testing being July 2023.
- European stock universe market-prices unknown
Source sentence
We now enlarge the asset space to include European markets.
- First set of 9 portfolios market-prices unknown
Source sentence
the first set of 9 portfolios were trained based on data up to December 2022 (first column of Table 1) — these are then run in the testing period in January 2023.
Costs
Unclear - sources disagree
Source sentence (the authors' words)
... re-estimation hence rebalancing are done on a monthly basis, the risk and cost of over‑trading is mitigated
Finds 4
Sharpe 2.26 (benchmark 1.70) · max drawdown 0.3%
- Low‑volatility property of the cointegrated portfolios persists throughout out‑of‑sample testing. performance
Source sentence
As reported, the low volatility is persistent throughout the out-of-sample testing period, which means the cointegrated portfolio tends to be robust in its low-volatility property in the testing period proceeding the in-sample training dataset.
- Out‑of‑sample Sharpe ratio of the optimised portfolio is 1.96 versus 1.70 for the benchmark. performance
Source sentence
Out of sample Sharpe ratio of the optimised portfolio for the period is reported at 1.96, while the benchmark is at 1.70.
- Maximum drawdown of the optimised portfolio is 4.5% compared to 10.7% for the benchmark. performance
Source sentence
Maximum drawdown of the portfolio is at 4.5% while the benchmark is at 10.7%.
- Joint US‑European optimised portfolios achieve Sharpe ratios of 1.76‑1.80, modestly above benchmark levels (1.7 US, 1.42 European). performance
Source sentence
Out of sample Sharpe ratio of the portfolios are 1.76 and 1.80 for I and II respectively, comparing to a benchmark of 1.7 and 1.42 for the US and European indices.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 34 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 (13)
- Co-Integration and Error Correction: Representation, Estimation, and Testing (1987)
- Bayesian Analysis of High Dimensional Vector Error Correction Model — → 2312.17061
- Determination of vector error correction models in high dimensions (2019)
- Spike-and-slab meets lasso: A review of the spike-and-slab lasso (2021)
- Pairs trading (2005)
- Sequential asset ranking in nonstationary time series (2022)
- Portfolio Selection: A Statistical Learning Approach (2022)
- Theoretically Motivated Data Augmentation and Regularization for Portfolio Construction (2022)
and 5 more
- Graph Denoising Networks: A Deep Learning Framework for Equity Portfolio Construction (2023)
- Sparse and stable Markowitz portfolios (2009)
- The Spike-and-Slab LASSO (2018)
- Bayesian inference with the l 1-ball prior: solving combinatorial problems with exact zeros (2023)
- Portfolio Selection (1952)
1 of 13 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.