Mean-Reversion Trading in SOFR Swap Curve Butterflies
Summary
The note outlines a proposed mean-reversion trade in a five-, ten-, and thirty-year SOFR swap butterfly. The trader regresses the ten-year swap rate on the shorter and longer rates without an intercept, then uses the residual as a signal. The suggested position scales the swap legs using regression coefficients and DV01 ratios to seek a duration-hedged exposure. The question raises whether this construction is preferable to approaches such as principal component analysis and how to handle a possible breakdown in historical relationships.
The response considers the residual strategy reasonable, particularly in rangebound markets where it may reflect short-term supply and demand. It cautions that broad moves driven by fundamental changes can make the residual less reliable as a mean-reversion signal. It also suggests holding periods of days or weeks rather than months and questions the need for a factor of two in the proposed position. This is a practitioner view, not a tested strategy: no empirical results, entry rules, stop criteria, or detailed hedge validation are provided.
Key ideas
- A regression residual can serve as a signal for a SOFR swap curve butterfly trade.
- DV01 ratios can scale the swap legs to target a hedged position.
- Rangebound markets may be more favorable when residuals reflect temporary supply and demand.
- Fundamental market moves can disrupt the relationship and undermine mean reversion.
- The response suggests shorter holding periods and questions one scaling factor in the proposed formula.
Tags
Full text
# Swaps hedged curves and butterflies strategies
# Swaps hedged curves and butterflies strategies
I want to try to trade Butterflies/Curves strategies on sofr swaps. These strategies will be mean reverting. My idea is as follow for the 5-10-30 butterfly:
- I run a regression between the swap rates $C_i$ without an intercept: $C_{10} = \beta_1 \cdot C_5 + \beta_2 \cdot C_{30} +\epsilon$.
- To be fully hedged, I construct the following position: $N \cdot \left( 2\times\text{10y SWAP} - \beta_1 \cdot \frac{DV01_{10}}{DV01_{5}} \cdot \text{30y SWAP} - \beta_2 \cdot \frac{DV01_{10}}{DV01_{30}} \cdot \text{5y SWAP} \right) $
where $N$ is a dollar notional and where $DV01_i$ is the DV01 of the swap $i$.
The idea is that if the above position is positive, it suggests mean reversion, so I would short the position; otherwise, I would go long. Essentially, I'm trading the residual 𝜖 ϵ from the regression.
I'm wondering:
- Is this a good way to hedge the position? (or PCA shows better mean reverting patterns?)
- Are there better ways to approach this? Additionally, does anyone have experience running similar mean-reverting strategies? Specifically, does the residual $\epsilon$ ever get significantly large, indicating that the correlation has "broken"? I would love to hear about potential pitfalls or lessons from those who have traded these kinds of strategies.
## Answer by dm63 (score 1)
https://quant.stackexchange.com/a/80988
The strategy you mention is reasonable. (I don’t think you need the 2 in the formula). In my experience this works well when the market is relatively rangebound , so that the regression residual represents short term supply and demand. It works less well when there is a larger move caused by a fundamental change in the market. The holding period for these trades is usually days/weeks rather than months, I would say.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.