how to protect trading strategies for a colocated server

[Corey]

Quote from Jerry030:


Who really cares about the details of the process if you can clone it in terms of results?

Quite simply because any curve can be fit perfectly with enough degrees of freedom -- but it doesn't mean that it can predict the next point with any measure of accuracy (in fact, quite usually, it will do worse than a more robust fit with less degrees of freedom).

The primary issue is that even though the input and output are easily measured, it is impossible to define the set of methods I am using to work with that data. Am I tossing out data older than 5 minutes? Am I fitting gamma-distributions from samples and trading spreads? Am I building interpolating spines with variable time windows to reduce noise? Am I removing the time-frame and translating into constant tick or volume measures?

The point being, with all these possibilities, unless you know something about my strategy (and that is a big IF), you most likely won't be able to reverse engineer it with such simple input / output. In fact, the more complex the input / output, the EASIER the job would be to a certain degree (e.g. translating Japanese codes in WWII). But with only price / volume and BUY / SELL, it becomes near impossible to define a domain of methods to replicate with...


Unless ... and this just struck me -- could it be possible to define a set of 'eigen indicators' -- a set of processes that almost any strategy can be broken down into a (non)linear combination of? Wouldn't that be strange... If so, well, maybe it would be possible to reverse engineer rather easily...

 

[Jerry030]

Quote from Corey:

Quite simply because any curve can be fit perfectly with enough degrees of freedom -- but it doesn't mean that it can predict the next point with any measure of accuracy (in fact, quite usually, it will do worse than a more robust fit with less degrees of freedom).

The primary issue is that even though the input and output are easily measured, it is impossible to define the set of methods I am using to work with that data. Am I tossing out data older than 5 minutes? Am I fitting gamma-distributions from samples and trading spreads? Am I building interpolating spines with variable time windows to reduce noise? Am I removing the time-frame and translating into constant tick or volume measures?

The point being, with all these possibilities, unless you know something about my strategy (and that is a big IF), you most likely won't be able to reverse engineer it with such simple input / output. In fact, the more complex the input / output, the EASIER the job would be to a certain degree (e.g. translating Japanese codes in WWII). But with only price / volume and BUY / SELL, it becomes near impossible to define a domain of methods to replicate with...


Unless ... and this just struck me -- could it be possible to define a set of 'eigen indicators' -- a set of processes that almost any strategy can be broken down into a (non)linear combination of? Wouldn't that be strange... If so, well, maybe it would be possible to reverse engineer rather easily...

Bingo...we’re getting it. As I said we aren't looking for the specifics of the machine in the black box...only the ability to create the same results. Part of the problem on this thread is the gulf between linear and non-linear approaches. They are almost different worlds.

In terms of curve fitting, would you agree that if the cloned machine consistently over many trades produces the same results as the black box machine on data unknown to either system then it’s not curve fitting but a data mining discover of equivalent functionality?

A simple analogy: there are multiple ways to cut a piece of paper into the same shape: with a knife, with a laser beam, with a punch press template, etc. As long as they all create the same shape they are equivalent in functional output.

 

[Algo_Design_Kid]

IMO there are far too many possibilities to be able to predict the code unless you are sitting at the server looking at the formulas. Think about it..

How many different technical indicators are there, and how many ways can I use them? Almost infinitely I am guessing.

Another thing would be capital startup. How would you propose if this was traded over 50 different securities to gain the capital to only be risking their exact risk tolerance of somewhere between 2-5%?

Lets just say you see all the signals.. unless the signals are just sent at a particular pattern in TIME SEQUENCE, it's not possible IMO.

If the OP was strictly about someone "hacking" into your server and stealing your information, well I am sure there are programs that could at least paralyze the attack to the point where it would be impossible, esp. if you are running a true "HF Strat".

Again, IMO a real HF Strat is more of a liquidity provider than anything, but all my programs do not involve this, so I can't really speak too much about it.

GL

 

[Jerry030]

Quote from Algo_Design_Kid:

IMO there are far too many possibilities to be able to predict the code unless you are sitting at the server looking at the formulas. Think about it..

How many different technical indicators are there, and how many ways can I use them? Almost infinitely I am guessing.

Another thing would be capital startup. How would you propose if this was traded over 50 different securities to gain the capital to only be risking their exact risk tolerance of somewhere between 2-5%?

Lets just say you see all the signals.. unless the signals are just sent at a particular pattern in TIME SEQUENCE, it's not possible IMO.

If the OP was strictly about someone "hacking" into your server and stealing your information, well I am sure there are programs that could at least paralyze the attack to the point where it would be impossible, esp. if you are running a true "HF Strat".

Again, IMO a real HF Strat is more of a liquidity provider than anything, but all my programs do not involve this, so I can't really speak too much about it.

GL

Nice opinion......what is you backgound in Analytics?...just so we know the skills behind the opinion.

If you read the thread carefully, you'll note that we aren't talking about recre4ating the system in plain text, only dulpicating it's outup singals. For this knowing the TA isn't necessary as they are being replaced by other equations which produce the output.
That is the whole point of using GA/EA in the first place.

Why hack into the server when you can use the known input and know output to recreate the system without the danger or effort of breaking and entering?

 

[Algo_Design_Kid]

Quote from Jerry030:

Nice opinion......what is you backgound in Analytics?...just so we know the skills behind the opinion.

If you read the thread carefully, you'll note that we aren't talking about recre4ating the system in plain text, only dulpicating it's outup singals. For this knowing the TA isn't necessary as they are being replaced by other equations which produce the output.
That is the whole point of using GA/EA in the first place.

Why hack into the server when you can use the known input and know output to recreate the system without the danger or effort of breaking and entering?

I have absolutely no background in analytics. I am strictly about pattern recognition, I literally sit and watch replays continuously until I puke for many different securities/derivatives with many different time frames/tick charts/vol charts. Then of course I just up with a formula for what I see. Definitely not genius work by any means although with the proper risk management it's not hard to make double digit ROR's year/year in futures contracts.

I am not familiar with ga/ea. My point is how would you know without seeing the code plus all technical specifics exactly how this would trade in the future? Are you talking about strictly picking up the signal as they output? Then I guess that would be easy enough if somehow you could just front run their trades if you were filtering through their output. Or I should say easier said than done. I of course would have no idea how to even start to derive that though.

Are you speaking about coming up with an algorithm to PREDICT their future trades? IMO that would be impossible again because of the fact of almost endless securities and ways you could possibly trade them.

There are many many many more smarter people than I, so more power to you if you could actually predict with a high certainty their future output signals. I just do not think that is possible without doing what I outlined above. If that is indeed what you are referring to, well then it's a cutthroat world my friend isn't it?

Referring to the hacking part, again I have no idea how that would be accomplished. I am guessing for the best of the best there always could possibly be a way without being noticed. Although I have no idea and that is just based off watching the movie "hackers" haha.

GL

 

[trend2009]

If you colocate the server with your broker, for sure he can have your executable code that generates trading signals. the question is if he can reverse engineer your executable.

Quote from Algo_Design_Kid:

I have absolutely no background in analytics. I am strictly about pattern recognition, I literally sit and watch replays continuously until I puke for many different securities/derivatives with many different time frames/tick charts/vol charts. Then of course I just up with a formula for what I see. Definitely not genius work by any means although with the proper risk management it's not hard to make double digit ROR's year/year in futures contracts.

I am not familiar with ga/ea. My point is how would you know without seeing the code plus all technical specifics exactly how this would trade in the future? Are you talking about strictly picking up the signal as they output? Then I guess that would be easy enough if somehow you could just front run their trades if you were filtering through their output. Or I should say easier said than done. I of course would have no idea how to even start to derive that though.

Are you speaking about coming up with an algorithm to PREDICT their future trades? IMO that would be impossible again because of the fact of almost endless securities and ways you could possibly trade them.

There are many many many more smarter people than I, so more power to you if you could actually predict with a high certainty their future output signals. I just do not think that is possible without doing what I outlined above. If that is indeed what you are referring to, well then it's a cutthroat world my friend isn't it?

Referring to the hacking part, again I have no idea how that would be accomplished. I am guessing for the best of the best there always could possibly be a way without being noticed. Although I have no idea and that is just based off watching the movie "hackers" haha.

GL

 

[Algo_Design_Kid]

If the outcome is just to reverse engineer I guess I would go the way of not being possible.

We still have to define how this trades anyway. IF it is a true HF Strat as I stated above, to me *IMO* that is more of a liquidity provider than anything and you would definately need the funds, etc. to back something like that in the first place.

Plus now you automatically have someone going against you as well fighting for the same ticks w/your (stolen) method, so IMO this would be an avoidable strat. Again, there is no reason for programs that I create to be in this situation in the first place, so 99% of this is all hypothetical from my POV.

If this only trades like my programs (maybe a couple of times a week at most) it would most likely be an easier outcome to replicate if indeed you could figure out a way to reverse engineer that strat. Again, I would not have the server space to be in this position anyway.

 

[Jerry030]

Quote from Algo_Design_Kid:

I have absolutely no background in analytics. ....<snip>
I am not familiar with ga/ea. ....<snip>
There are many many many more smarter people than I, ...<snip>
I have no idea how that would be accomplished. .... <snip>
I am guessing for the ..... <snip>
Although I have no idea .... <snip>
GL

Algo Kid,

Thank you for the response on your expertise on this topic. At least you are honest enough to admit that you don't have any. I have to respect that. As strange as it sounds there are rumors that on occasion one or two individuals have posted on ET material presented as expertise which turned out ot be nonsensical opinion. Astounding, I know!

 

[Jerry030]

Quote from trend2009:

If you colocate the server with your broker, for sure he can have your executable code that generates trading signals. the question is if he can reverse engineer your executable.

That would be even easier then extracting functional equivalence for the system from the input price time series and the known trading signals output, with no access to the code.

What we are talking about here in both cases is vaguely similar to very advanced curve fitting…..the dread of automated system developers everywhere.

 

[Algo_Design_Kid]

Quote from Jerry030:

Algo Kid,

Thank you for the response on your expertise on this topic. At least you are honest enough to admit that you don't have any. I have to respect that. As strange as it sounds there are rumors that on occasion one or two individuals have posted on ET material presented as expertise which turned out ot be nonsensical opinion. Astounding, I know!

Im not going to write any checks I cannot cash, but just state my POV. But until you (not specifically) can define what HF Strat means to you personally, I think it's only 50% of the equation. Because even if they handed you everything you wanted on a silver platter it does not mean you could take advantage of it.