Protecting your strategies

[logic_man]

Quote from trend2009:

I do not agree with this. for example, if I have a 1000 trades out of a mechanical system, at least I can see the common scenario when the 1000 trades take place.

The potential number of variables that go into identifying the common scenario makes it practically impossible.

Imagine trying to guess 1000 encrypted passwords and each password could contain any letter, number, hieroglyphic, ideogram, wingding, astrological symbol, etc. Even a supercomputer would take an eternity to complete the task.

 

[gmst]

Quote from gpiu:

You misunderstand my point, you always have the common scenario, the raw data itself...

I'll give you an example, I challenge you to decipher one of my algos. The attached file is an OOS study of ES trading in 2001 with market order entry/exit and $25 total slippage and $4 commissions. Total ~6500 trades for the year.

Many a times written communication be not clear. Sorry if I misunderstood you.

Please send over the file and ES 1 minute data for 2001. Sorry I don't have that data. I will like to take a stab to see if I can find out anything.

Btw, it might be pretty hard to reverse-engineer your system here, my point was say if have only 50 trades in a year, then it means you are looking for a very specific pattern. It will be relatively easier to decipher the logic.

 

[gpiu]

You misunderstand my point, you always have the common scenario, the raw data itself...

I'll give you an example, I challenge you to decipher one of my algos. The attached file is an OOS study of ES trading in 2001 with market order entry/exit and $25 total slippage and $4 commissions. Total ~6500 trades for the year.

 

[logic_man]

Quote from gmst:

sorry but it is clear what you write makes sense theoretically, but you don't have any real world strategy development experience. Real world strategies that work are typically simple in nature - with most of them containing < 10 parameters. Its not that hard to figure out the basic premise behind the trading strategy given sufficient number of entries/exits.

Btw, thanks epic for that post. Useful!

It doesn't matter how many parameters the strategy has. What matters is how many parameters there are to choose the 10 from. If I can choose 10 parameters out of a set of 12 parameters, I may be able to get all 10 right. If I have to choose the right 10 out of 12,000 parameters or 120,000 or 1.2 million, my chance of getting the right 10 is basically nil.

I'm not sure why this is so difficult to grasp.

I have 4 main parameters in my model. I've been reading ET for nearly 2 years and I've never even seen a single one of those parameters discussed in any depth by any poster. 3 of them have never come up at all. So, how is someone going to guess what my 4 parameters are when they are not even on anyone's radar screen? The answer is that they are not going to. Ever. And if you are just using the same parameters that everyone does discuss, you probably don't have anything anyone who could steal it would want to steal.

What you have to realize is that what is and is not a "parameter" in a trading model is essentially infinite. I mean, you've got guys basing their trading models on the position of astrological phenomena or fucking sunspots. Maybe someone's trading copper based on the meeting schedule of the Chilean copper miner's union or trading wheat on the production schedule at a John Deere factory in Iowa. You can't know.

THAT'S the "real world".

 

[logic_man]

Quote from Swan Noir:

Keep in mind that most rapes go unreported.

In the 70's Italy had stringent (on paper at least) exchange controls. They were designed to prevent money from leaving the country. My partner and I had an office in Lugano, Switzerland less than 50 miles from the Italian border. Lugano's primary industry was handling flight capital from Italy -- the Lira went down a tiny bit nearly every week.

A client of ours had devised a method to help Italians get around the controls. In a fairly short period of time we figured out the basics of his methodology and in less than a month figured out a way to do it in a way that scaled. We approached him with our calculations and we ended up partnering with him in that business. BTW ... we offered him our calculations without charge and warranted not to compete with him. That went a long way toward cementing a relationship of some years ... in fact till his death.

If you own something figure out how to protect it or, better yet, figure out how to scale it and finance the larger size. If you can make enough in a short period losing your edge, while still unwelcome, is less painful.

I don't think that rape is comparable because the victim may choose not to report for personal reasons and there is often gray areas involved regarding consent. If someone straight out steals your strategy (and, implicitly, your money), none of that really plays a role.

Besides, I asked for one high-profile case, not every single instance of these alleged thefts. At some point, someone would have tried this on the wrong person, gotten caught and sued. To suggest that this is an occurrence which goes on continually and never reported defies everything we know about how humans react when they are harmed financially.

One case.

 

[gmst]

Quote from logic_man:

It doesn't matter how many parameters the strategy has. What matters is how many parameters there are to choose the 10 from. If I can choose 10 parameters out of a set of 12 parameters, I may be able to get all 10 right. If I have to choose the right 10 out of 12,000 parameters or 120,000 or 1.2 million, my chance of getting the right 10 is basically nil.

I'm not sure why this is so difficult to grasp.

I have 4 main parameters in my model. I've been reading ET for nearly 2 years and I've never even seen a single one of those parameters discussed in any depth by any poster. 3 of them have never come up at all. So, how is someone going to guess what my 4 parameters are when they are not even on anyone's radar screen? The answer is that they are not going to. Ever. And if you are just using the same parameters that everyone does discuss, you probably don't have anything anyone who could steal it would want to steal.

What you have to realize is that what is and is not a "parameter" in a trading model is essentially infinite. I mean, you've got guys basing their trading models on the position of astrological phenomena or fucking sunspots. Maybe someone's trading copper based on the meeting schedule of the Chilean copper miner's union or trading wheat on the production schedule at a John Deere factory in Iowa. You can't know.

THAT'S the "real world".

Your strategy is truly unique. You shouldn't be concerned about reverse engineering, rather by theft. For many others the strategy is not based on unique parameters. Also we are not talking about datamining. When you see 60 trades and look at their graphs side by side (even without doing any number crunching), you start getting an idea about what the stuff is.

In any case, discussion in this thread has not been focussed only on reverse engineering but on a bunch of topics.

I am waiting for response of MC about them uploading the strategies during the crash process.

 

[gpiu]

Quote from gmst:

Btw, it might be pretty hard to reverse-engineer your system here, my point was say if have only 50 trades in a year, then it means you are looking for a very specific pattern. It will be relatively easier to decipher the logic.

I think it is more difficult than you realize. You are correct in that the number of parameters tends to be low for a system to be robust. However, even if you happen to know the exact number of parameters, the problem is akin to unlocking a password in which each character has an infinite number of possibilities, just like logic_man pointed out.

I think your comments reveal your strategy development approach. I am guessing you tend to focus on some pattern and ride the wave until it stops working. In my experience, a robust successful strategy is inherently non-stationary in nature in order to adapt to a changing market.

For example, the trades I depicted in 2001 while in the thousands, only give you an after the fact view of the algo during that time period. The volatility clustering in ES for 2005 for example was very different. My algo adapts to it...You would not have the information to infer this. If I show you my 2005 results, you would think it is a different strategy entirely, when in reality they are subsets of the same more general feedback system.

 

[gmst]

Quote from gpiu:


For example, the trades I depicted in 2001 while in the thousands, only give you an after the fact view of the algo during that time period. The volatility clustering in ES for 2005 for example was very different. My algo adapts to it...You would not have the information to infer this. If I show you my 2005 results, you would think it is a different strategy entirely, when in reality they are subsets of the same more general feedback system.

Probably, you are correct here to some extent at least. It will be hard to figure out the various filters that are being used. For example, exact same conditions might happen, but trade not taken because of say vol filter. hmmm

Thanks.

 

[Epic]

Quote from gmst:

sorry but it is clear what you write makes sense theoretically, but you don't have any real world strategy development experience. Real world strategies that work are typically simple in nature - with most of them containing < 10 parameters. Its not that hard to figure out the basic premise behind the trading strategy given sufficient number of entries/exits.

Btw, thanks epic for that post. Useful!

With my strat, the complexity doesn't come from the final handful of parameters, it comes from what it takes to identify those parameters. Anyway, let's say that the trigger is based on even just 3 simultaneous events. You then say that, with the raw data, numerical analysis would easily identify these three events. That is only true if you know which data, beyond the transaction data, to analyze.

Let's assume that a system with only 3 parameters trades ES long the open and always offset at the close as the first parameter.

The next parameter, is to only trade the open if there has been a gap larger than 1/2%.

So now any decent trader will say that these first two parameters would be easy to figure out and that only leaves one more parameter, which should be very easy.

The last parameter is to trade a 10-lot long when both of the following two conditions are met;
-USD/EUR is up more than 0.2% from previous close.
-ZN has moved less than 0.02% from previous close.

You have absolutely no idea that you should even be looking at USD/EUR or ZN, so you are automatically accepting the fact that you must scan every possible related security, looking for a condition that you don't know how to define. And in the case of ZN, you are actually looking for a "non-move" which is infinitely harder than looking for some quantifiable event in the price action.

I think the point is obvious, and that was a system with only 3 very basic parameters. Mine has more than a dozen parameters, and runs both numerical and statistical analysis in real-time to locate and execute trades. Someone would literally have to spend years trying to figure it out, which is how long it took me to figure it out in the first place. The only benefit that they get is in knowing which underlying to start analyzing.

It really is just as easy to develop your own system.

 

[tenthousandmen]

Quote from themickey:

I totally disagree with this theory.
It would be very easy to reverse engineer, I have done it. However not illegally. It can be done without too many smarts.
The solution as one poster mentioned is dropping in some illogical discretionary trades to throw off the scent, but I'd assume this will cost a little in wasted trades, probably small change though.

No it's not, and no you couldn't. If it were, then you could generate your own buy-sell signals then reverse engineer them.