Thread: Nonparametric Association Measures in Systematic Trading

Dear Fellow Traders,

I am collecting links and other resources which include examples of using nonparametric association steps in egy development and risk management. Two chief examples of nonparametric association steps are http://www.stanfordphd.com/KendallsTau.html and http://www.stanfordphd.com/SpearmansRho.html. The two steps are based on ranks. As such, they're naturally suitable for trading, where people care more about the advantage going up/down and less about the true size of the motion.

The steps are strong in the sense that they are not sensitive to big infrequent spikes in the data. For this reason they're more secure compared to correlation#8230; The robustness may be a blessing in some scenarios (in which #8220;smoothing#8221; is necessary) and it may be a curse in others. I use nonparametric dependence measures in a couple of places in my egies but their effectiveness is surely much wider. I would like to boost my scope and, perhaps, write a wonderful systematic web-resource one day... Any information would be appreciated.

Thank you.

Originally Posted by ;

quote His maths are wrong. He begins from a incorrect equation he then takes (correct) deductions starting from his error... Let's begin with the suitable equation y=#946;x#963;²#949; lt;- he overlooked the doubt (aka sound ) #963;²#949;!! The assumptions are that - The expectancy of the noise E[#963;²#949;]=0 since it's the model assumption of a linear regression. - The variance of the noise V[#963;²#949;]=#963;² because a linear regression implicitely assumes white noise. - It's supposed to remain like that. It's not actually the...

interesting journey onto the stats arb in Currency Market. I've read some posts on your pairs trading threads; also it rings bells all the way. Been there; completed that; and that I know your fruions. After a while I've concluded that simplicity wins.

Say for example that you have discovered 2 cointegrated tools; the white noise is gaussian; and also the relationship was steady historically. There is not any guarantee that it will remain so. How do you trade it? Presumebly you will specify some deviation threshold and then fade the moves.

After a while; this cointegration relationship will break forever or temporarily. How do you discover that so that you may avoid the losses?

The response to that is going to be to use a standard deviations (aka bollinger bands). That won't work either; unless you use a shing standard deviation; and et voila you end up using a moving average.

Because a moving average is used by many to solve the expression drift issue of the spread; why not use it for non-cointegrated pairs then. Basically throw away those cointegrated tests; and stick to the remedy that solved the expression drift issue.

The moving average is indeed a powerful tool which solves a very tough problem in statistics. You will find quite a variation of moving averages; such as Kalman filter; to which they attempt to fix the same problem.

My favourate read on this topic; is that chapter two by James Altrucher. Though he isn't a hedgy; but he did hit the nail in the head when he described the unilateral pairs trading egy within his publiion.

Originally Posted by ;

quote Hello Stass, at the above mentioned equity curve you uploaded; just how many trades each year?

Is Dependent on the specifiion. In the chart shown above, the number of trades each year is approximately 244. The majority of the time is spent in the idle state, as every trade takes only several minutes.... Not saying that this is the best way to exchange AUD/CAD. Just illuing the possibility in machine learning.

How do you use these stuff to Forex?

Originally Posted by ;

we care more about the asset heading up/down and less about the true magnitude of the movement.

I recall a post by (I can't loe it) where he showed that an up fashion containing more down pubs compared to up bars. The tendency was created by a minority of big up moves.

Originally Posted by ;

quote I recall a place by (I can't find it) where he showed an upward trend comprising more down bars than bars up. The tendency was created by a minority of large up moves.

Thanks for this example, PipMeUp. Yes, such situations happen all of the time (when one enormous player must work a large notional, for example). But, seeing down more bars than bars up in your case is just a matter of placing the bar frequency. If you had chosen a time step large enough you would have seen just up one bar, do you not? ... At the end of the day/week/month, first of all, we care if we're any good as prop traders. We care if our signals are powerful, whether our call was correct and we've made money on the transaction... And then we care how much.

Sorry if I wasn't too clear in my initial article. I was just trying to express a notion that, overall, the leadership is more significant. However there are scenarios on all amounts where magnitudes are crucial as well... And non-parametric smoothing is just one possible appliion (suitable only in certain corners of prop research). As I also mentioned, oftentimes ignoring magnitudes is a curse.

Originally Posted by ;

how can you use these stuff to Forex?

Software of nonparametric dependence measures in pair selection was only the initial block of links that I submitted. This really isn't the only area with potential (rather than the most promising one). here they are connections with the macro world. In particular, we all are interested in finding pairs in which the k-th lag of indior (cross) A leads indior (cross) B. Some of the aforementioned statistical methods may serve as a building block in a more complex analytical frame.

Originally Posted by ;

In case you'd selected a time step large enough you would have seen only up one bar, would you not?

Obviously. So what?

Originally Posted by ;

first and foremost, we all care whether we are any good as prop traders

Why should I do that?

Originally Posted by ;

We care whether our signals are powerful

there's not any such thing as a A installment. If such a thing existed you can filter from the non-A . This post-filtering of your egy would only generate a new egy emitting normal signals.

Originally Posted by ;

whether our call was correct and we've made money on the trade...

I disagree. That really is a whim of prepared to be correct. The illusion of control. The outcome of the next trade is irrelevant. The most important thing is your expectation (will the egy make money in the long term ) and the variance (to gauge the risk of ruin) of the subsequent 1000 trades.

Originally Posted by ;

we're interested in finding pairs at which the k-th lag of indior (cross) A leads indior (cross) B.

I don't believe I do that but I don't really understand what you mean.



I will try to use Spearman's correlation to see how it compares with Pearson's for the risk alloion problem I discribed here https://www.nigeriaforextrading.com/...g-journal.html

Originally Posted by ;

quote I remember a post by (I can not find it) where he revealed an upward trend containing more down bars than bars up. The trend was made by a minority of big up moves.

There's also data out there that maybe 10 percent of market days make up 90 percent of market movement.

PipMeUp, you're missing the point that this thread isn't about well-documented asymmetry of yields (interplay between magnitude and direction) but about interesting prop ideas based on Spearman's rho, Kendall's tau and these.

Originally Posted by ;

There isn't any such thing as a A setup.

Sure, but there is a difference between being appropriate 50.5percent of their time (precarious) and 75 percent of their time (better). Additionally, unfortunately I cannot concur with this

Originally Posted by ;

The results of the following trade is immaterial. What is important is that the expectation

Expectation is just another performance metric of the next random trade. And how do you gauge the expectation anyhow, in this constantly changing environment? ... Any trade is applicable, especially an ineffective one, because it may indie structural changes in the market. Now, having that said, structural changes are somewhat tricky to find with nonparametric association steps. Better tools exist. So let us return this discussion into the following area: where are nonparametric institution steps useful and where do they have suboptimal possessions? ... Any examples of egies or risk management tools would be appreciated. Thanks.