Any good book on Statistical Arbitrage?

Quote from ezbentley:

Besides MATLAB, is there any other software that can analyze cointegration with reasonable programming effort? Cointegration calculation seems quite involved and MATLAB is not that accessible to average retails traders.

Also, is there any recommended time frame to re-evaluate the cointegration among traded products? Supposedly, cointegration, which exists for whatever reason, may change or get stronger or weaker over time.
More...

Mathernatica seems to have it also.

Mathematica home edition now costs only $295.
 
Hi Matt,

I am wondering if you can share some insight regarding the effectiveness of cointegration vs Spearman Rank Correlation in pairs trading. Spearman Rank Correlation is A LOT easier computationally. I briefly scanned through some cointegration papers and they all seem to require a certain understanding of time series analysis and modeling, which will take me a while to digest since I have no quant finance background.

In any case, am I correct in stating that both cointegration and Spearman are just "tests" used to determine if pairs tend to mean-revert? In other words, you don't calculate the cointegration "on the fly" during market hours to give you signals. Rather, you use either(or both) method to look for potential pairs, add them to your watch list, and calculate the proper hedge ratio. And then the actual entry and exit signals are still based on the price difference of pairs exceeding a certain amount of standard devation(or other measure of divergence). Please correct me if my general understanding is wrong.

Thanks for generously sharing your stat arb experience.
 
Hi

Stat arb is based on cointegration, not correlation. Chan describes it nicely in the article:
Cointegration is not the same as correlation
http://epchan.blogspot.com/2006/11/cointegration-is-not-same-as.html

Cointegration tests mean reversion, correlation doesn't.
I test for cointegration using the last 252 data points whether that be, monthly close, daily close, 10 minute bars etc.

Then from those cointegration tests I calculate for the pairs that have a dickey fuller (its a cointegration test) of better than 90% the hedge ratio and the standard deviation that the pairs are currently apart. If more than 2 std then I trade.
 
Hi,

I understand that "linear correlation coefficient" is not a good measure of mean reversion. That's why I was asking about "Spearman Rank Correlation," since it was mentioned in your very first reply. Gummy also mentioned "Spearman Rank Correlation" in his pair trading sections. So it seems like Spearman Rank Correlation can also be used as a measure of mean reversion, but maybe not as effective as cointegration.

Thanks,
 
Sorry ezbentley, I misunderstood.

Spearman etc is an OK method but in my opinion not as robust as cointegration. The correlations may break down and not recover - yes that could also happen with cointegration, but the distribution of the cointegration is tested in the dickey fuller so you have an "idea" of the probability of the cointegration remaining going into the future. That’s why I only go in at 90% or better probabilty.

There is another post on elite where a guy swears by his correlation strategy. In fact there are a lot of strategies in Google based on correlation. So I referred to Gummy and Spearman because it is probably a good entry point for start arb beginners. You can modify Gummy's Excel code to either take in a live data feed (10/15/30 min bars) or add your own daily close symbols. You could plot std deviation bands on it. Using correlation is certainly a good start. Also the way gummy has plotted the graph is helpful.
 
Martinghoul - I haven't used any other cointegration tests. I originally studied cointegration in college and then later came across DF in a C++ COM and later in a Matlab toolbox so I have stuck to what I know and can use in my coding.

I have read some work comparing the different cointegration tests and there seems to be subtle differences in lag differencing and the distributions. There is a study in Google on some economics data and they applied three cointegration tests - the results were close enough for me to be happy that DF is Ok.

I'm keen to hear your thoughts on the other tests if you think they may be more robust or accurate.
 
For the selection of stocks to trade against the etf, would it be useful to diagonalize the covariance matrix, to get most important eigenvectors?

I think a similar approach is used in index arbitrage.

I think you should probably wait covariance using the weights of the stocks in the etf.

Your prescription to take the most correlated neglects the importance of weights I think.

The link you give uses the top 10 stocksby weight.Do you do the same?

What are the intraday gains and frequency ( I mean how often) this etf arbitrage can happen?

Let me also ask you why do you publicize it... in this way you will loose your profits...

Did you try to do a forward backtest of this strategy, i.e. to checkif the cointegration pesists?

Do you use half time to get out of bad trades?

Thanks again
Ly
 
You must log in or register to reply here.
Back
Top