Pair Trade Strategy In Liquid ETFs (Pairs Trading with ETFs) (Formula, Backtest and Example)
Many traders are pair trading ETFs. Pairs trading is one method of making your strategy less exposed to market fluctuations. Most traders trade stocks, but you can, of course, also trade ETFs. If you are trading ETFs you indirectly have a wide diversification, albeit to different segments of the stock market.
In this article, we look at an ETF pairs trading strategy. We go long the weakest ETF at the close and exit at the close the next day. It looks promising, but probably not a tradeable strategy.Â
An ETF pairs trading strategy in SPY, EEM, EWG, EWU, and EWJ
Today we test ETF pairs trading.
This Sunday morning I was just testing some ideas on these different ETFs: SPY (S&P 500), EEM (emerging markets), EWG (Germany), EWU (UK), and EWJ (Japan). They seek to copy the performance of the most important stock exchanges in the world.
The rules of the pair trading strategy are like these:
- Every day rank each ETF based on the IBS formula: (c-l)/(h-l).
- Buy on the close the one with the lowest value, short the one with the highest value.
- Exit on the close next day.
In other words, this is a 100% “market-neutral” strategy – a daily pair trade in ETFs.
I write “market-neutral” because no strategy is ever completely neutral unless you’re doing arbitrage. Even though you have the same amount invested long and short, or at least adjusted to the expected volatility, you are still exposed to adverse movement in either position. You are liable to black swans.
The ETF pairs strategy returns this equity curve in %:
The test period is from 1. January 2005 until the present. No commission and no slippage.
Does anyone trade something similar to this? Obviously, this one is hard to implement because the MOC needs to be sent 15 mins before close. However, for example in Amibroker you can easily program your strategies to send orders just seconds before the close:
The ETF pair trading strategy presented in this article is very short-term. This means it involves both commissions and slippage.
Â
Hi Oddmund,
according my tests, the strategy is flat recently, in your graph since point 1681 on x-axis. (if you exclude EEM). Until then it worked really well.
So, seems like you had the long short idea too late? Maybe it can be improved upon.
Hi, I’m gonna trade this live just to test. That’s the only way to find out if the test has any value.
What were the results of your live test?
Nowwhere as good as theory! Probably not worth trading.
also, what is very strange, is that strategy works on e.g. EWG but on on DAX. I will have to investigate deeper, but seems to work only on US assets. And since EWG trades only US time, I guess it will be strongly influenced by the US session more than DAX session, on intra session basis.
Jozef
Yes, but as you said, the sessions are completely different so can’t compare them.
Oddmund,
How did you go about testing for cointegration on these pairs? Also, what do you use to backtest?
Cointegration? I’m not into that. I simply picked these ETFs because they are the most important stock markets. Not sure about the quotes the test is build on, been trading it live for one week with 1.2% in gain in total so far.
I’m using Excel. Very simple.
Hi,
I have in my replication for 2012 and 2011 only 0.05% and 0.01% in average return per trade. With costs and slippage this will not be profitable for me. Any thoughts?
Hi, It’s falling equity curve, so I’m not surprised. I tested this on IQFeed data. All theoretical numbers are not to be trusted no matter what (in my opinion). I’m trading this one live now, so we’ll see if it’s any money to be made.
This I agree on, but if the theoretical numbers look bad, I would not even bother testing it live. Its (almost) always worse than it looks. he-he 🙂
Yes, but I got slightly better results than you. Been trading it for 4 weeks now with surprisingly good results. Besides, when trading live, I tend to discover other things. After three weeks of trading it I put on a little twist which I believe has been bossting the results. Time will show.
I have been testing something very similar, buying the bottom x% and shorting the top x%. I find it very interesting that it does not work what so ever if you rank the ETFs by intraday % return, but does work when ranking by IBS. I have a different, larger basket of ETFs but I have found exiting next open superior to exiting next close. I will be probably be going live with it tomorrow. Have you found that the ranking is still correct at the close when submitting orders 15 minutes prior ?
The ranking is often wrong because you have to send the orders before the close. But in te long run I don’t think it will matter. Yes, better to close at open. That was the “twist” I did 🙂
But the best stocks from close to open are those which have regular trading hours when US markets are closed, ie Asia and Europe.
Hi,
how has your return on this been the last month? Im curious because I want to compare with my own “paper trading”.
Since I started in late December I have 6.04% in real trading. But I expect it to decreae.
Hi, could you please tell me your profit on this in march (only if you’ve traded this every day). Im trying to calculate slippage, as this is a quite marginal strategy, at least on the paper..
Hi have changed the ETFs I trade quite a bit. Still, March was 0.45% in plus.
hello~ One simple and important question~how much leverage do u use for the pair trading ?
EOD I trade close to 100% for pairs, and additional positions are extra/bonus. But i trade more than this strategy in this article.
Profit is profit. 🙂
hello Oddmund:
One question is why to use MOC orders ? why not send market orders at 3:59:30 at New York time ?
I can send moc and go away (to close positions). To enter positions it’s more similar to backtesting. Besiden, more volume on the close for stocks.
I would be interested if you could please show a back test of results using this simple formula on weekly rather than daily basis. A rolling five day week so orders can be entered any day. the longer time to set up for a trade may make for a less hectic trade too. thanks in advance for your comments.
Hi, It does not work very well on weekly prices, as far as i remember.
Using ETFs with IBS(Internal bar strength) is heavily covered in QUSMAs blog, he does a great of explaining the concept and testing on various baskets of ETFs.
The IBS effect really is greatest in the US market, its real value is probably as a filter only.
Hello,
Nice discovery!
Except how do you handle the look-ahead bias as you have peeked into the future by submitting a MOC order 15min prior to market close ‘before’ knowing the ranking of each ETF ?
Actual performance would require you to ‘guess’ the final ranking of each ETF 20mins before market close and submit a MOC.
I wonder how is the real-time performance of this strategy? Is it still good?
Hi, not so good as before. I used yo put in orders 15 mins before the close based on ranking at that time. I’m not trading this anymore, not good enough.
I have tried using intraday data to further test the lookahead bias.
Either 1) buy very close to 4:00PM, or 2) enter MOC by 3:45PM by trying to predict the top/bottom ranking…
Choice 1 result in slippage which is will erode the profit
Choice 2 is even harder, within 15 mins the ranking could change quite abit.
They could work prior to 2010 but definitely not going to work from now on.
In short, fail
This has a VERY high degree of lookahead bias IMO. You don’t know the day’s low until trading ends. If you enter before trading ends (obs close, buy the close), you have a degree of error measuring the day’s low/high for the ibs calculation.
If you use next day open to open instead of obs close buy the close, the performance disappears immediately.
Yes, no doubt about that.
However, I trade the “close” a few seconds before the official price and slippage is practically zero in the liquid products.
https://www.quantifiedstrategies.com/what-is-slippage-in-live-trading/