OU Pair Trading Thresholds and Shifts in the Long-Run Mean
Summary
The document asks whether optimal entry and exit thresholds for an Ornstein–Uhlenbeck mean-reversion strategy should move one-for-one when the process mean changes. It frames the question through the thresholds relative to the mean and the symmetry between long-short and short-long pairs, whose OU parameters differ only in the sign of the mean.
The author compares threshold curves from two figures in a book on optimal mean-reversion trading with transaction costs and stop-loss exits. The curves appear to shift differently near a mean of zero, prompting a question about what causes the asymmetry. The document does not provide an explanation or derive the thresholds, so it offers a useful modeling question rather than a resolved method. Its observations are based on plotted results and do not specify the model assumptions or parameter settings needed to assess the cause.
Key ideas
- The author questions whether optimal thresholds relative to the OU mean should remain constant as that mean shifts.
- The question invokes sign symmetry between long-short and short-long pairs.
- Plotted thresholds appear to depart from a simple vertical shift, especially near a mean of zero.
- The document poses the issue but does not establish its cause or provide a derivation.
Tags
Full text
# Why aren't the optimal entry/exit thresholds for OU pairs trading relatively invariant to shifts in the OU mean? # Why aren't the optimal entry/exit thresholds for OU pairs trading relatively invariant to shifts in the OU mean? The optimal entry/exit thresholds for mean reversion trading (assuming an underlying Ornstein-Uhlenbeck (OU) process) is derived in the paper "Optimal Mean Reversion Trading with Transaction Costs and Stop-Loss Exit" by Tim Leung and Xin Li, and these results are also included in the book "Optimal Mean Reversion Trading Mathematical Analysis and Practical Applications" by the same authors. I would have expected that the values of these threshold levels ($b^*$ and $d^*$) relative to the OU mean $\theta$ would not change if you changed $\theta$. That is, for some arbitrary shift in the OU mean, $\triangle \theta$, it would always be true that $$b^*(\theta) - \theta = b^*(\theta + \triangle \theta) - (\theta + \triangle \theta)$$ and $$d^*(\theta) - \theta = d^*(\theta + \triangle \theta) - (\theta + \triangle \theta)$$ This must be true since buying the long-short pair is equal to selling the short-long pair, and the OU process parameters for the short-long pair are identical other than the sign of $\theta$. However, this is clearly not the case when $\theta$ is around zero for some reason, as seen in Figures 2.3 and 2.5 in the book: In all of the charts, there are curves for three different values of $\theta$ in increments of 0.3, and so I would expect that the three lines would be exactly the same, but shifted up by exactly 0.3 each time. In Figure 2.3, this is more or less the case for the lines corresponding to $\theta=0.3$ and $\theta=0$, but the line corresponding to $\theta=-0.3$ seems to be trying to avoid the values around $b^*=0$? Meanwhile in Figure 2.5, the lines look like vertically shifted versions of each other, however the distance between them is not exactly 0.3. The underlying OU process can go negative, as can the price of a long-short pair of assets being modelled, so it can't be anything to do with a positive price constraint. What am I missing here?
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.