Finding moving-band statistical arbitrages via convex–concave optimization

Kasper Johansson, Thomas M. Schmelzer, Stephen Boyd

Abstract

We propose a new method for finding statistical arbitrages that can contain more assets than just the traditional pair. We formulate the problem as seeking a portfolio with the highest volatility, subject to its price remaining in a band and a leverage limit. This optimization problem is not convex, but can be approximately solved using the convex-concave procedure, a specific sequential convex programming method. We show how the method generalizes to finding moving-band statistical arbitrages, where the price band midpoint varies over time.

Introduction

Statistical arbitrage (stat-arb) is a cornerstone of quantitative finance, traditionally focused on identifying pairs of assets that exhibit mean-reverting behavior. However, finding these opportunities in a universe of thousands of stocks is computationally difficult, and standard statistical methods like co-integration tests often fail to scale to portfolios involving multiple assets.

This paper introduces a novel optimization framework that moves beyond simple pairs trading. Instead of asking if a portfolio is statistically mean-reverting, we seek to identify the most volatile portfolio that remains within a predictable price range. By directly optimizing for the characteristics a trader desires—high "wiggle" within a stable band—we can identify more robust and profitable trading opportunities.

Methodology

The Optimization Framework

We formulate the search for a statistical arbitrage as a mathematical optimization problem with three primary components:

  1. The Objective: We maximize the portfolio's price variation (the sum of squared differences). High volatility is essential; without it, there are no price swings to exploit for profit.
  2. The Band Constraint: The portfolio price must stay within a specific range (e.g., +/- 1 unit from a midpoint).
  3. Leverage and Sparsity: We impose a limit on total position size. By using an $\ell_1$ norm constraint, the optimizer naturally selects a small, manageable number of assets (typically 5–11) rather than hundreds, ensuring the resulting portfolio is practical to trade.

The Convex-Concave Procedure (CCP)

Maximizing volatility—a convex function—is a non-convex problem, which is typically difficult to solve. To address this, we utilize the Convex-Concave Procedure (CCP). This iterative method linearizes the objective at each step, transforming a hard non-convex problem into a sequence of simple Linear Programs (LPs) that can be solved in seconds.

Moving-Band Generalization

A key innovation of this research is the generalization to moving-band statistical arbitrages. Unlike fixed-band strategies, which can be broken by long-term market drift, the moving-band approach allows the center of the band to follow a moving average of the portfolio's own price. This allows the strategy to adapt to structural market changes while still profiting from short-term mean reversion.

Empirical Results

Using the CRSP US Stock Databases (covering over 15,000 assets from 2010 to 2023), we compared the performance of fixed-band and moving-band strategies. The results indicate that the moving-band approach is significantly more robust.

Metric Fixed-Band Stat-Arb Moving-Band Stat-Arb
Profitability 68% of portfolios 79% of portfolios
Avg. Annualized Return 2% 16%
Avg. Annualized Risk 35% 20%
Sharpe Ratio 0.81 0.84
Max Drawdown 17% 12%

Moving-band portfolios stayed active twice as long as fixed-band ones and were far less likely to hit stop-loss conditions, demonstrating superior resilience to market drift.

Conclusion and Future Work

Direct optimization via the Convex-Concave Procedure provides a scalable and effective way to discover multi-asset statistical arbitrages. The transition from fixed to moving bands represents a significant step forward in making these strategies robust to real-world market conditions. Future research will focus on the optimal weighting of "meta-portfolios" consisting of multiple stat-arbs and the integration of more sophisticated execution policies to further minimize transaction costs.

Summary updated

BibTeX
@inproceedings{johansson2024finding,
  author = {Kasper Johansson and Thomas M. Schmelzer and Stephen Boyd},
  title = {{Finding moving-band statistical arbitrages via convex–concave optimization}},
  booktitle = {Research},
  year = {2024},
  doi = {10.21203/rs.3.rs-4196380/v1},
}
Figure