If there is a quantitative measure for Edge, what is it?

[DrChaos]

I like Acrary's methods.

In modern statistics, Monte Carlo randomization types of tests are quite important and powerful. In my opinion, these types of things which can get at the real problem at hand, as opposed to a perfect analytic solution for a problem which isn't really the right one.

You can, at a simple level, attempt bootstraps of your actual returns (bootstraps are a randomized sequence taken from the original time series, and you should probably use a method which includes whatever sequential correlation appears to exist) and then compare to bootstraps of returns from a "null" (zero expectation) series.

You can get the second from randomized trading, or even just subtracting the sample mean of returns from yours (though that is 'less independent' in some ways).

Then a figure of merit could be the fraction of times, and degree that bootstrapped returns (or cumulative sum thereof, or some measure including drawdowns) exceed the simulated returns from a no-skill setup.

 

[inCom]

Quote from OddTrader:

Sometimes even worse, a truly good and reasonably profitable system is neglected.
That evaluation can be resulted by probably due to lack of independent thinking and analysis, I would guess.

Maybe. However if I used little or no backtest at all, trading with my own money I'd risk to run out of it or spend too much time before I find profitable strategies. I try to regard backtest just as a preliminary phase in system development. Personally I used to believe much more in it than I do now.

PS: I googled for "can be resulted" and "by probably due to" and found very few hits. I thought that 1) your written english is worster than mine or 2) you are kindly pushing my leg or 3) never mind, I'm wrong.

GS

 

[OddTrader]

Quote from DrChaos:

I like Acrary's methods.

In modern statistics, Monte Carlo randomization types of tests are quite important and powerful. In my opinion, these types of things which can get at the real problem at hand, as opposed to a perfect analytic solution for a problem which isn't really the right one.

You can, at a simple level, attempt bootstraps of your actual returns (bootstraps are a randomized sequence taken from the original time series, and you should probably use a method which includes whatever sequential correlation appears to exist) and then compare to bootstraps of returns from a "null" (zero expectation) series.

You can get the second from randomized trading, or even just subtracting the sample mean of returns from yours (though that is 'less independent' in some ways).

Then a figure of merit could be the fraction of times, and degree that bootstrapped returns (or cumulative sum thereof, or some measure including drawdowns) exceed the simulated returns from a no-skill setup.

Just 2 cents:

I believe we can find and see a large number of papers covering this and related aspects for trading, although not knowing how many of the writers are profitable traders, whether before or after writing up their papers.

How/ why are we going to decide using the 70% for long trades or 60% for short trades as mentioned by ACrary?

How do we know these % figures are objective ones? Are they statistically significant, or arbitrary?

 

[OddTrader]

Quote from inCom:

Maybe. However if I used little or no backtest at all, trading with my own money I'd risk to run out of it or spend too much time before I find profitable strategies. I try to regard backtest just as a preliminary phase in system development. Personally I used to believe much more in it than I do now.

GS

Personally, I strongly recommend to do good backtesting - some Proper ones.

 

[ecritt]

Quote from DrChaos:

I like Acrary's methods.

In modern statistics, Monte Carlo randomization types of tests are quite important and powerful. In my opinion, these types of things which can get at the real problem at hand, as opposed to a perfect analytic solution for a problem which isn't really the right one.

You can, at a simple level, attempt bootstraps of your actual returns (bootstraps are a randomized sequence taken from the original time series, and you should probably use a method which includes whatever sequential correlation appears to exist) and then compare to bootstraps of returns from a "null" (zero expectation) series.

You can get the second from randomized trading, or even just subtracting the sample mean of returns from yours (though that is 'less independent' in some ways).

Then a figure of merit could be the fraction of times, and degree that bootstrapped returns (or cumulative sum thereof, or some measure including drawdowns) exceed the simulated returns from a no-skill setup.

How can you select at random (sample & replace) and maintain sequential correlation at the same time? I have always been under the impression that the two are mutually exclusive.

 

[nononsense]

Quote from ecritt:

How can you select at random (sample & replace) and maintain sequential correlation at the same time? I have always been under the impression that the two are mutually exclusive.

cause it sounds great to a lot of losers.

 

[nononsense]

Quote from OddTrader:

Personally, I strongly recommend to do good backtesting - some Proper ones.

OddTrader,
I like that. Of course, in my viewpoint, backtesting goes TOGETHER with system development. Point already often taken up at ET's: No competent profitable trader will ever reveal anything about either.

 

[Opra]

Quote from OddTrader:

My guess is if a trader's method/ system does not have any (trading) edge (including market directions, MM skills, etc.) against all other players, probably some of the above infrastructure edges would not help much.

Yes you are right. I shuold have included method/system in the list of "edges".

All I was trying to say is, edge is not the steady upward-sloping equity curve, which is the result of a trader exploiting his "edges" here and there effectively.

In other words, having "edges" does not garrantee trading success and lacking some "edges" does not spelll trading doom either. It all comes down to recognize one's strengths and weakness and trade accordingly.

 

[RunTrade]

Quote from OddTrader:

Personally I like this approach and I think that could be practical, useful and helpful to find and (I think, indirectly) define an edge which has been discussed on ET so many times so far.

Adding significantly more capital on top of an initial one would be relatively safe and feasible by this approach, I would guess. Many thanks.

Everyone seems to be agreeing with ecritt's idea of quantifying edge. I havent gotten a chance to get the whole way through this thread, so i apologize if this has be mentioned already.

While ecritt's idea is logical and probably mathematically correct, I believe it should not be pursued. In fact, quantifying an edge should not be attempted either. (Which is why it has little research on it)....

Many traders seem to forget that the market is constantly changing, evolving, yada yada...this is why we all know that systems will eventually fail without modifications. The same applies to an edge. Its strength/weakness will constantly flucuate along with the market (unless you have the perfect system that adjusts to the market 100%). So quantifying your edge could only lead to a false sense of security making one believe his edge is stronger and more secure simply b/c it is defined. The quality of an edge will change as the markets change, therefore the only signifigant point is that you keep an edge at all times.

~RT

 

[OddTrader]

Quote from nononsense:

Point already often taken up at ET's: No competent profitable trader will ever reveal anything about either.

Thanks for your revelation above.