Just dabbling with pairs trading at the moment. The chart below shows the spread of a highly cointegrated pair (p value = 5e-7). Data is based on 1 minute bar close, over a period of 6 months. Regression was done over the whole 6 months for simplicity. I'm surprised to see long periods of 'drift' from the centreline given such a strong relationship between the two. Cointegrated_pair Anything I should be aware of? Any pitfalls and tips for pairs trading? Thx.
Ample unused capital to survive periods where they digress. Read the Blair Hull interview in Market Wizards. Might not be identical to what you're doing, but well worth a read. A lot depends on when you settle the trades. If you are not a dealer and not subject to daily settlement - that's better, but will reflect in your returns. You also want to dig deep into LTCM - not just the gossip and uninformed beliefs. Blair was all US markets and derivatives. LTCM positioned cross-border risk. Much more lucrative until it goes wrong. A broker who fully understands is a must.
Be aware that you can find correlations between two series in even random data. You can easily demonstrate this by writing software to create 10,000 series of random numbers. Find the two that correlate the most. You can't then pair trade the remainder of those two random series. So make sure what you have found is statistically significant and not just the most interesting sample out of a large set of pairs.
Pairs of what? Stocks, index futures, rates, ETFs, EM, commodities... It's surprising you ask for advice and don't even specify any details. Don't even know if it's a time spread or cross-hedged position. It could be any of a million different trades...
There is no real relationship (or pair) just based on statistics. If you only look for cointegration you will find many spurious and non stable pairs. You need to look for really fundamentally related relationships. But then you do not need to measure cointegration because you already know they will mean-revert from some point. So the whole cointegration thing is more or less a kind of joke, because there are no such mathematical relationships in reality when it comes to trading. Only fundamentals do matter. So you also need to understand what you trade and why. And you also need to factor in any relevant news. My two cents.
Of course. But math is a tool for finding these things. That's the point. Yes, there are clear mathematical relationships in financial instruments. Options are the most obvious example. But there are many others.
I have only meant and spoken about pairs based on statistical arbitrage. Of course there are mathematical relationships among financial instruments. But it is a totally different game to profit from mispricings among Put-Call-Parity or Cash-Futures relationships (and much more difficult with more tiny profits too).
Most of the larger option MM's today trade some form of correlation trading. Either to an index - like Blair - or to a(n) ETF or both. Blair mostly traded back in days when the SPX was the index standout and of the volume in Single names was not a fragmented as today. Didn't really work out that well as he was only able to sell his firm to Goldman(SLK) for about half a billion dollars.
I think that is a different game doing this as MM because it is very short term orientated. Doing that kind of scalping is usually for the retailer not profitable because of transaction costs. But this is different when being a MM. Let us not discuss strategies here the average trader cannot implement at all. The original question was about statistical arbitrage for a normal trader (not being at a hedge fund or maket maker position which is a clear exception).