Do Trading Models Count?

[qazwsxedc]

Quote from stephencrowley:

Well, imho, a models profitability doesn't really matter.. because theoretically if you develop something that has a consistent negative expectancy then you can just do the opposite of what it says or find out why it is negative and reverse it in the model and have consistent positive expectancy.

Not true.

Test a series of random entries, 2-bar trailing stop loss, target 3 x initial risk on 10-minute bars in a reasonably liquid market. You'll see that it loses money hand over fist.

Now check the same series of entries, but trade in the opposite direction each time, using the same timeframe, stop loss criterion and profit target. Still loses money hand over fist.

For bonus points, figure out why that is so.

 

[stephencrowley]

Bars have no meaning in my world.. can you reframe a 2-bar trailing stop loss in terms of trailing stop percentage of volatility? Also, I don't have profit targets, they make no sense.

Long return's limit is [-100%,inf]
Short return's limit is [-inf,100%]

Long from 30->70 gives you ~133% return but shorting from 70->30 only gives you ~57% obviously.

The point is, random entry will *always* be the same as buy(short)/hold over the long run minus transaction costs, whatever stop-loss/bar/whatever policy you set.

If your system has any expectation besides 0 and it's absolute value exceeds transaction costs you can turn it into profit.

Quote from qazwsxedc:

Not true.

Test a series of random entries, 2-bar trailing stop loss, target 3 x initial risk on 10-minute bars in a reasonably liquid market. You'll see that it loses money hand over fist.

Now check the same series of entries, but trade in the opposite direction each time, using the same timeframe, stop loss criterion and profit target. Still loses money hand over fist.

For bonus points, figure out why that is so.

 

[nitro]

Quote from nononsense:

stephen,

You really found the answer:
(1) Scratch together any kind of a gismo you call "MODEL";
(2) Test it for its "expectancy";
(3) If it is positive, you're done;
(4) If it is negative, replace SELL by BUY and BUY by SELL and you're also done.
Thanks a lot pal,
nononsense

Yeah,

Ok, Good night and GOOOOOD LUCK

nitro

 

[stephencrowley]

Haha, dude, you kill me.

Quote from nitro:

Yeah,

Ok, Good night and GOOOOOD LUCK

nitro

 

[qazwsxedc]

Stephen,

You make four statements. I don't want to just flatly contradict you four times because that's likely to get your back up rather than make you look in the right places. So I won't.

Instead, can I suggest you take some sample data of your choice, a copy of Excel, then during a weekend with an unlimited supply of coffee you test your assertions? You'll be surprised.

Oh, and a 2-bar trailing stop loss is just based on a snapshot of local volatility. For testing purposes just pick what seems sensible by eyeballing.

Quote from stephencrowley:

Bars have no meaning in my world.. can you reframe a 2-bar trailing stop loss in terms of trailing stop percentage of volatility? Also, I don't have profit targets, they make no sense.

Long return's limit is [-100%,inf]
Short return's limit is [-inf,100%]

Long from 30->70 gives you ~133% return but shorting from 70->30 only gives you ~57% obviously.

The point is, random entry will *always* be the same as buy(short)/hold over the long run minus transaction costs, whatever stop-loss/bar/whatever policy you set.

If your system has any expectation besides 0 and it's absolute value exceeds transaction costs you can turn it into profit.

 

[stephencrowley]

You won't get my back up.. so don't be shy.

You are your basing your assumptions on random entry points.

Can you tell me, theoretically, why any model(not counting for transaction costs) that generates any expectancy(deviation from entire market movement) other than 0 should not be able to be reversed?

Obviously, if a model is generating buy/sell signals in the wrong direction then there is either 1) something fundamentally wrong in the model or 2) something near the 'output' of the model causing a simple sign reversal or somesuch.

Anything that can consistently extract signal from noise is money imho.

Quote from qazwsxedc:

Stephen,

You make four statements. I don't want to just flatly contradict you four times because that's likely to get your back up rather than make you look in the right places. So I won't.

Instead, can I suggest you take some sample data of your choice, a copy of Excel, then during a weekend with an unlimited supply of coffee you test your assertions? You'll be surprised.

Oh, and a 2-bar trailing stop loss is just based on a snapshot of local volatility. For testing purposes just pick what seems sensible by eyeballing.