Constructing Macro-Fundamental Factors for Cross-Asset Momentum Trading

In the world of quantitative finance, the concept of stationarity is of paramount importance. A time series is said to be stationary if its statistical properties, such as its mean and variance, are constant over time. Many statistical models and trading strategies, particularly those based on mean-reversion, rely on the assumption of stationarity. However, most financial time series, such as stock prices, are non-stationary. The most common method for achieving stationarity is to take the first difference of the series. But what if this is too blunt an instrument? This article explores the technique of fractional differentiation, a more nuanced approach to achieving stationarity that can be particularly useful for developing mean-reversion strategies.

The Problem with Integer Differencing

When we take the first difference of a price series, we are essentially transforming it into a series of returns. While this often results in a stationary series, it comes at a cost: we lose the memory of the original price series. The returns series only tells us the change in price from one period to the next, but it doesn't tell us anything about the long-term trend or the 'memory' of the series. For mean-reversion strategies, this memory can be important. A stock that has a tendency to revert to its long-term mean will exhibit this behavior in its price series, but not necessarily in its returns series.

This is where fractional differentiation comes in. Instead of taking an integer difference (e.g., d=1), we can take a fractional difference (e.g., d=0.5). This allows us to find the minimum amount of differencing required to make the series stationary, while preserving as much of the original series' memory as possible.

The Mathematics of Fractional Differentiation

The concept of fractional differentiation is an extension of the familiar integer differentiation. The fractional difference operator is defined as:

$(1-L)^d = \sum_{k=0}^{\infty} \binom{d}{k} (-L)^k$

where L is the lag operator and d is the order of differentiation. When d is an integer, this formula reduces to the standard integer difference. But when d is a fraction, we get a weighted sum of all past values of the series, with the weights decaying over time.

The key to applying fractional differentiation is to find the optimal value of d. This is typically done by searching for the smallest d that results in a stationary series. A common approach is to use the Augmented Dickey-Fuller (ADF) test. We can apply the fractional difference operator for a range of d values (e.g., from 0 to 1 in increments of 0.01) and choose the smallest d for which the ADF test fails to reject the null hypothesis of a unit root.

Fractional Differentiation in a Mean-Reversion Strategy

Once we have a fractionally differentiated series that is stationary, we can use it to develop a mean-reversion trading strategy. The basic idea is to identify periods when the fractionally differentiated series deviates significantly from its mean and then take a position in the opposite direction. For example, if the series is significantly below its mean, we would buy the asset, expecting it to revert back to its mean. Conversely, if the series is significantly above its mean, we would sell the asset.

The trading signals can be generated using a variety of methods. A simple approach is to use a z-score. We can calculate the z-score of the fractionally differentiated series and then buy when the z-score is below a certain threshold (e.g., -2) and sell when it is above a certain threshold (e.g., 2). The positions would be closed when the z-score reverts back to zero.

The Hurst Exponent

Another useful tool for identifying mean-reverting series is the Hurst exponent. The Hurst exponent is a measure of the long-term memory of a time series. It can be used to classify a series as either mean-reverting (H < 0.5), random (H = 0.5), or trending (H > 0.5). By calculating the Hurst exponent of a fractionally differentiated series, we can get a better sense of its mean-reverting properties. A series with a low Hurst exponent is a good candidate for a mean-reversion strategy.

Conclusion

Fractional differentiation is a effective technique that allows us to achieve stationarity in a time series while preserving its memory. This can be particularly useful for developing mean-reversion trading strategies. By finding the optimal amount of differencing, we can create a stationary series that is still able to capture the long-term mean-reverting behavior of the original price series. When combined with tools like the ADF test and the Hurst exponent, fractional differentiation can be a valuable addition to any quantitative trader's toolkit.