Sizing Ratio Pairs Trades to a Z-Score Stop-Loss Budget
Summary
The document asks how to size two equity legs when a mean-reversion strategy trades their price ratio. Rolling estimates of the ratio’s mean and standard deviation produce a z-score, with entry beyond an outer threshold, exit near the mean, and a more distant z-score intended as a disaster stop. The goal is to choose leg sizes so that reaching the stop corresponds to a specified account loss, assuming the estimated mean and standard deviation remain constant.
The questioner suggests holding one leg fixed and solving backward for the other, but no derivation or final sizing method is supplied. The response instead recommends estimating trade outcomes and probabilities through backtesting and mentions Kelly-based investment sizing. It also offers a trend-following interpretation of Bollinger-style signals and suggests possible loss controls and options hedging. These comments do not calculate the requested loss at the stop, and they mix distinct trading approaches; their claims about indicator behavior are not supported with evidence here.
Key ideas
- The proposed pairs signal is based on z-scores of a rolling price ratio.
- The sizing objective is to align loss at a defined z-score stop with a chosen account risk budget.
- Holding one leg constant and solving for the other is raised as a possible approach but not worked through.
- The response suggests using backtested outcomes and probabilities to inform Kelly-style sizing.
- The answer does not derive sizing that guarantees the specified loss at the stop.
Tags
Full text
# Position Sizing For Ratio Pairs Trade # Position Sizing For Ratio Pairs Trade Ok, let's say I'm trading a spread of two stocks, X & Y, The spread is calculated as a ratio (Spread = X / Y). I use rolling stats to calculate the mean, standard deviation and hence the z-score of the spread, as such, I take the usual entry/exit signals of entering trades when the z-score exceeds 2 the then closing them on return to zero. There are, of course, the usual ways of sizing the legs. (dollar neutral for example). What I want to do is size the legs according to a z-score stop loss. Let's say that I'm going to define a z-score that exceeds 4 as my disaster stop. Given the amount of my account (in dollars) I want to lose if the stop is hit, how do I size the legs on entry of the trade? (Let's assume the mean and standard deviation remain constant). Update: In the meantime I've been pondering this, this only thing I can think of is to hold one leg constant and work backward. This can be done for each leg in turn. But I'd still be interested in other ideas. ## Answer by user12348 (score 1) https://quant.stackexchange.com/a/11113 Interesting idea. I would like to have the reference you are using for this scheme? You are treating X/Y as a stock and then using Bollinger Band kind of signalling mechanism. I would think you would define the entry/exits more strictly to make it profitable. If you are making a lot of trades using this system, then from backtesting, you can figure out your winning and losing trades and probabilities and use Money Management using Kelly Fraction for sizing trade investment. In Bollinger band(+/- 2 sigma from mean), generally 14 or 21 day rolling window, when stock is above mean (z-score 0), it tends to stay up. It is a trend following indicator. With z-score>0, you can stay with positive trend. Of course, you need to build in some loss rule like 5-8% loss from mean to avoid whipsaws. A trend up is higher high and higher lows, trend down the opposite. You initiate a trade only if the trend is established. If the stock moves downward then play that once the trend develops. If the sigma becomes small then it tends to explode either to the up or down side. You could hedge this by buying a strangle one sigma away cheap and then once you make money ride with this system. I have outlaid this as if you are using S=Y/X as a stock. In pair trading, if you buy a spread and it widens it is the same thing.
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.