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

[toe]

Testing a system vs a randomised system is much better than simply looking for a positive expectancy, in much the same way as medical researchers measure the results of a double blind study using a new drug vs a placebo rather that just looking to see if patients get better on the new drug (they likely will get better without any drug).

The reason using randomised trades as your benchmark is better is because you can get a random system to make money if the market is good enough. So simply looking for a positive expectancy doesn't proove non random events.

Even forward testing has its drawbacks because market conditions can change between initial run and out of sample runs.

If you want to determine whether a system has an edge over random then 'Acrary' has already done very good work on this and posted a lot on it, why re-invent the wheel.

 

[OddTrader]

Quote from toe:

...

Thanks Toe for your contribution and independent thinking.

 

[rosy]

your edge is the amount over expected value you think you'll make. you can't quantify it. with it, you can use the kelly criterion to optimize you bet size.

 

[OddTrader]

Quote from rosy:

your edge is the amount over expected value you think you'll make.

So you are saying actually there is a measurable and quantitative value for an edge.

Perhaps I wouldn't mind to keep trading a system even if it produces a small negative value when comparing to an expected value which is an extremely huge amount.

 

[rosy]

Quote from OddTrader:

So you are saying actually there is a measurable and quantitative value for an edge.

your_expectation -expected_value=edge

however, your expectation might be way off.

 

[OddTrader]

Quote from rosy:

your_expectation -expected_value=edge

however, your expectation might be way off.

There are two issues:

How to determine YourExpection, quantitatively?

It seems illogical that: With a fixed value of YourExpectation, a (poor performance) system with a smaller (or negative) ExpectedValue would produce a higher/ better EdgeValue.

 

[ElectricSavant]

Yes Equalizer is correct... Acrary has measured it, I can't find it though...do a search and do us a favor...post the link here.

Measuring time, consistency, profitability and risk...is Acrary's playground.

Recently I suspect he has suspended teaching as this business of trading may be a game of theft....But what do I know? I could be completley wrong.

Michael B.

Quote from Equalizer:

Emm..., not quite, you'll find that many of these publicly available "potentially profitable" strategies/methods presented fail miserably under the cold harsh scrutiny of backtesting. If you add a discretionary dimension then you change it completely. So then next question is usually why can't you simply tell a computer what it is that you do? The answer is that it is extremely difficult to do that - for numerous reasons. Then again computers can do things that are incredibly difficult for human trader to do.

Regardless of the approach, what you are trying to determine is whether your profitability is based purely on chance alone - which is what my post was alluding to. For a simplistic example, if you have a collection of 100 trades, and 95 were loses, but the 5 winners were huge, then do you have an edge? Can you trade this way?

Once you have a profitable method that you have verified is statistically significant based on your measurements - then you can begin to discuss various measures as you point out.

What these are would be dependent on your approach in addition to your own personal utility. How much of a drawdown can you tolerate? What level of risk is acceptable to you (and of course how do you define and measure risk)? How many consecutive losses can you handle before you are "psyched-out". Sharpe ratio? Sortino ratio? How do you know if the edge has disappeared, or the string of losses was simply due to pure chance? Annual return? etc... The list goes on.

You might wish to search for some of acrary's posts.

 

[Trader666]

Quote from OddTrader:

Can we measure Edge?

Do you have anything in mind that could be used to measure an Edge with quantitative terms?

Any pointers?

All comments would be appreciated.

Depends on what you're trying to measure the edge in. For example, if you want to measure the edge of a short term entry technique you're trading on a daily basis, use time exits. Get distributions of trades for your entry method with 1,2,3,4,5,... day exits and compare them to the trade distributions for the same duration time exits but with random entries. Do this visually and with (preferably) nonparametric statistical tests, but the t-test does an OK job.

 

[OddTrader]

Quote from Trader666:

Depends on what you're trying to measure the edge in. For example, if you want to measure the edge of a short term entry technique you're trading on a daily basis, use time exits. Get distributions of trades for your entry method with 1,2,3,4,5,... day exits and compare them to the trade distributions for the same duration time exits but with random entries. Do this visually and with (preferably) nonparametric statistical tests, but the t-test does an OK job.

Personally I like this idea very much, since I've been using a very similar concept (but quite different method) in order to measure the edge value of a system for my own evaluation.

I guess the key point would be making a comparison against a benchmark chosen for a pre-defined purpose.

Many thanks!

 

[nononsense]

Quote from OddTrader:

[]
I guess the key point would be making a comparison against a benchmark chosen for a pre-defined purpose.

Many thanks!

Took a bit long before trying to arrive at something sensible in this thread. Simply restate the above without the fuzzy gobbledygook. Read further:

Quote from nononsense:

Only way to measure an edge: your bank account.