What Is Statistical Arbitrage? Types & Strategies

2026.01.30 2026.01.30
What Is Statistical Arbitrage? How Traders Use Stat Arb

Oleg Tkachenkohttps://www.litefinance.org/blog/authors/oleg-tkachenko/

Statistical arbitrage in trading is an alternative to traditional strategies based on technical and fundamental analysis. The method is based on a mathematical model that uses statistical correlations between the prices of highly correlated assets. If the spread between their quotes widens, it is highly likely to return to its average value.

This review explores the statistical arbitrage strategy and explains how to implement it using algorithmic software.

Major Takeaways

• Statistical arbitrage, or stat arb, is a trading strategy based on mathematical modeling and searching for price divergences between highly correlated assets.
• The strategy assumes that, over a long time horizon, a stable average spread can be identified for a pair of highly correlated assets (spread trading). If the spread deviates from this level and widens, positions can be opened in the short term and closed when the spread returns to its average value.
• Statistical arbitrage requires specialized software and automated trading systems (robots). Manual trading is rarely used due to the complexity of calculations and the inability to compete with high-frequency trading (HFT) algorithms, which identify and eliminate price divergences in fractions of a second.
• Statistical arbitrage can be applied across various markets, including stocks, currency pairs, ETFs, and cryptocurrencies. However, each market requires a separate mathematical model that accounts for its specific characteristics and other risk factors.

Statistical Arbitrage Definition and Core Principles

Statistical arbitrage is a quantitative trading strategy based on identifying temporary price discrepancies between financial instruments that historically exhibit high correlation. Unlike classical arbitrage, the stat arb approach is based not on guaranteed risk-free profits, but on the mathematical expectation that asset prices will return to their average over time. At the same time, technical analysis indicators are not usually employed.

Statistical and classical arbitrage strategies at a glance:

So, how does statistical arbitrage work? There are two assets whose prices usually move in the same direction, maintaining a relatively stable distance between them. This distance is called a spread.

For example, one asset may begin to rise sharply due to a fundamental factor. At the same time, the spread widens abnormally over a short period. However, the statistical model shows that the spread has a stable mean that it returns to over time.

If this happens, the trader does the following: once the spread reaches an extreme level, they open opposite positions simultaneously—selling the asset that has appreciated more and buying the one that has lagged. When the spread narrows as prices converge, the trader locks in profits.

Key components of statistical arbitrage:

• Mean reversion model. It includes spread calculation formulas, statistical techniques, and assessments of price correlations between assets over long time intervals, and more.
• Cointegration. Correlation alone is not enough. Statistical arbitrage requires verifying assets and assessing cointegration to ensure that the spread is stationary and tends to revert to its mean.
• A pair of highly correlated assets. Instruments from the same sector are most often chosen, although other combinations are possible.
• Risk management. Setups depend on market volatility and specific assets. The higher the volatility, the greater the temporary spread divergence may be.

To reduce risk on individual trades and diversify your strategy, you can open dozens of positions on different asset pairs simultaneously. Although trades may be held for several days, statistical arbitrage is often classified as high-frequency trading due to the high speed at which price discrepancies are identified and acted upon.

How Statistical Arbitrage Trading Works

Statistical arbitrage is a form of algorithmic trading that typically involves several key steps.

1. Finding Pairs or Groups of Financial Assets

The algorithm analyzes historical data to identify assets with strong correlations, particularly cointegrated ones.

Examples:

2. Spread Calculation and Cointegration Verification

The spread is not simply the straight-up difference between the prices of two assets. In practice, it is calculated using a mathematical model that factors in the standard deviation, price logarithms, linear regression, beta coefficient, hedging coefficient, and other parameters.

This stage is considered difficult, which explains why manual trading is rarely used in statistical arbitrage. There are many examples of specialized software, analytical platforms, and code (e.g., Python) for working with various asset classes available online. However, the question of their actual efficiency and practical applicability remains open.

Here is an example of Python code from the MQL5 website, which is suggested for use in identifying cointegrated and correlated pairs:

The output is a sorted table with the most suitable pairs.

Based on the data, you can create a trading advisor that spots deviations in real time and automatically opens trades based on predefined conditions.

The risk management model is further optimized using neural networks. It is usually developed in Python or other programming languages.

3. Trading Signal

When the spread deviates from its historical average by a certain amount (for example, by 2 standard deviations), positions are opened. A trader simultaneously buys a cheaper asset and sells a more expensive one. The expectation is that either the cheap asset’s price will converge to the expensive asset’s, or the expensive asset will return to the price of the cheap one.

When prices converge, positions are closed with a profit. If the price of the cheap asset remains unchanged (grows more slowly or, even worse, declines), the asset that is becoming more expensive continues to rise in price, and the spread widens further. In this case, this results in a loss for open positions. Therefore, it is crucial to determine when to open trades and when spread expansion will stop. Such calculations are also based on historical statistics.

Types of Statistical Arbitrage Strategies

The core idea behind all types of statistical arbitrage is exploiting mean reversion, profiting from the tendency of asset prices or their spreads to return to their historical average. The differences lie in the risk management models used, the methods for selecting asset pairs or groups, and the approaches to optimizing the strategy.

Potential returns vary depending on the level of risk taken. For example, market-neutral arbitrage offers minimal returns with relatively low risk.

Pairs Trading Strategy

Pairs trading is one of the most foundational and straightforward types of statistical arbitrage. It involves identifying two assets with a high correlation, typically from the same market segment or class of financial instruments.

A short position is opened on the asset that is growing faster, and a long position is opened on the asset with slower growth. The calculation is based on the subsequent narrowing of the spread between their prices.

Market Neutral Arbitrage

The strategy aims to generate profits regardless of whether the market is rising or falling. The portfolio is structured to have an aggregate beta (sensitivity to market movements) close to zero. In this case, returns are generated solely by relative price changes within the portfolio, rather than by overall market direction.

Example:

• The portfolio is structured so that losses on some positions are offset by profits on others. The allocation of assets and the ratio of short and long positions are selected mathematically to ensure that the overall financial result of the portfolio remains neutral regardless of market conditions.

• Pairs for statistical arbitrage in trading are found among the instruments included in the portfolio.

Investment and hedge funds often use this type of arbitrage. It typically yields a moderate 10–15% annual return, with minimal risk because the strategy is largely unaffected by market trends.

Cross Asset Arbitrage

The strategy involves identifying and exploiting correlations across asset classes. For example, correlations between the shares of gold mining companies and gold futures, the currencies of exporting countries and commodity futures, and between stock index futures and the shares included in the index basket.

Therefore, if the price of gold has risen by 5% while the quotes of gold mining companies have remained unchanged, this discrepancy can serve as a trading signal.

ETF Arbitrage

ETF arbitrage is a strategy that leverages the spread between an exchange-traded fund (ETF) market price and its Net Asset Value (NAV).

NAV is the value of all assets in the fund’s portfolio, excluding total liabilities, divided by the number of shares outstanding. In theory, this indicator should correspond to the fair price of one share (unit) of the fund.

An arbitrage opportunity arises when investors start buying ETF shares en masse. In this case, the market price may temporarily exceed the value of the fund’s underlying assets, creating a premium.

Conversely, during sell-offs, an ETF’s price may fall below the value of its assets, resulting in a discount.

Essentially, statistical arbitrage in trading boils down to calculating that the market price of ETF shares will eventually return to NAV.

Advantages and Disadvantages of Statistical Arbitrage

Statistical arbitrage is one of the most popular strategies in algorithmic trading. However, like any strategy, it has its strengths and weaknesses. Below is a table showing the main advantages and disadvantages of statistical arbitrage.

How to Get Started Trading with Statistical Arbitrage

At first glance, statistical arbitrage seems like a fairly simple strategy: you need to find assets with high correlation and wait for the moment when their price spread begins to widen. However, the main difficulty lies in putting this approach into practice. For novice traders, the following basic step-by-step plan can be outlined.

1. Learning the Basics

The first step is to learn the basics of econometrics, statistics, and programming languages such as Python or C++. Although there are ready-made algorithmic solutions, using them without understanding how they work can lead to errors and financial losses.

The goal of this stage is to learn how to determine whether a pair of assets is statistically stable and whether it makes sense to trade its spread.

2. Choosing an Analytical Platform and Tools

Specialized platforms or libraries are used to analyze cointegration and find suitable pairs. The goal is to select assets that meet the conditions for statistical arbitrage: high correlation and cointegration.

This requires access to historical data on a large number of instruments. This data can be obtained through analytical services or free APIs.

3. Developing a Trading Algorithm

Based on the selected platform and historical data, a trading robot can automatically search for correlated assets and open trades when their spreads widen.

Thus, the first three points form the preparatory stage, which should lead to an algorithm capable of applying the statistical arbitrage strategy in practice.

4. Trading Automation

The next step is to connect the trading robot to the API of an exchange, broker, or cryptocurrency platform to execute trades in real time.

5. Strategy Testing

Before using your robot in live trading, it should be tested on historical data and a demo account. It is important to take into account commissions, slippage, and market conditions.

The goal is to evaluate the effectiveness of a trading strategy in both past and current market environments.

6. Designing a Risk Management System

The final stage involves building a risk management system: setting limits on maximum drawdown, factoring in the probability of correlation breaks, and accounting for the impact of unexpected events and black swans.

Some sources also offer examples of basic trading system parameters depending on the type of strategy used:

High-Frequency Trading

Long-Term Arbitrage

Conclusion

Statistical arbitrage is based on the idea that price differences (spreads) between highly correlated assets are temporary. In the long term, these instruments tend to revert to their historical averages, creating trading opportunities.

This strategy is geared toward professional market participants because it requires knowledge of mathematical statistics, econometrics, and basic programming. In addition, it is necessary to integrate trading algorithms with exchange platforms via APIs and account for the specifics of trade execution.

The average return on this strategy is about 15% per year. Despite its lower performance compared to more aggressive approaches, it has a significant advantage: most operations are performed by a trading robot.

Statistical arbitrage is a complex yet promising algorithmic trading strategy that leverages market inefficiencies. Mastery of it contributes to broadening one’s horizons, developing analytical thinking, and forming a professional approach to investing.