A Very Simple System...

keeptradin' · May 1, 2012

Quote from bwolinsky:

Your parameters are too literal to work going forward. Any brief read about what you're doing would immediately be dismissed as curve fitting, and even the 10 and 10 you started with could also be considered curve fitting.

When I come up with the right parameters, I'll report in to you how much better my version of your model is doing than yours.

Thanks Beau, I look forward to hearing from you.

Mike805 · May 1, 2012

Quote from bwolinsky:

I can probably improve your results 3-4 fold by adding trail stops, targets, and stops by genetically optimizing them. Would you like to see?

Did you add if o[1]>c[1]?

Baby steps here. The technological advantage you'd have with just some easylanguage would have no compare to the random guessing you might be doing by hand.

...

Your parameters are too literal to work going forward. Any brief read about what you're doing would immediately be dismissed as curve fitting, and even the 10 and 10 you started with could also be considered curve fitting.

When I come up with the right parameters, I'll report in to you how much better my version of your model is doing than yours.

Okay, for the sake of gullible readers I want to say something here...

So, you're recommending the use of GA? I'm assuming you're talking about Genetic Algo's right? It sounds like you want to add 3 degrees of freedom and then let the optimizer run wild... ok, I think I get the angle here.

Assuming I'm right, the amount of stuff wrong with this process can fill a very large landfill; in fact - I don't even know where to begin...

Let's start with intent, since intent is legally as important as the actual action. By curve fitting the crap out of a simple rule, you intend to compare that result to something you claim to be "random guessing"? From that comparison, you believe you'll attain more information?

Does anyone else see this for what it is? For goodness sake, I'm actually in awe that BoWo just wrote this...

dom993 · May 1, 2012

Quote from bwolinsky:

The technological advantage you'd have with just some easylanguage would have no compare to the random guessing you might be doing by hand.

I have to agree with this - and *only* with this

It is much faster to backtest trading ideas than forward test them. Of course, backtesting results are not always indicative of future performance, but neither are forward testing results, especially when the sample size is small (say less than a few hundred trades).

I know you already stated you are not a programmer, but it would be easy enough for you to learn the basics required to backtest & enhance your VSS. For this I would use a platform supporting EasyLanguage (TradeStation, Multichart) as it is much faster learning curve than C# / java / etc. Another option would be to use Excel for backtesting, if you are more comfortable with it.

I do trade a somewhat similar system. Live P/F around 1.35 for ~115 trades since July 2011. I developed, "optimized" & validated that system in a couple weeks only.

bwolinsky · May 1, 2012

Quote from keeptradin':

Thanks Beau, I look forward to hearing from you.

Incidentally, if you ever want to use Tradestation or Multicharts, I recommend Multicharts.

Here is your code:

inputs:
tgt(7.5),
stp(11);

variables:
bull(1),
shit(1);

if marketposition>0 then begin
bull=entryprice+tgt;
shit=entryprice-stp;
end;

if marketposition=0 and time>0929 and time<0931 and o[1]>c[1] then begin buy next bar at market;end;
if marketposition>0 then begin
sell next bar at bull limit;
sell next bar at shit stop;
end;
if marketposition>0 and time>1599 then begin sell next bar at market;end;

Below is the equity curve with a realistic 2 tick slippage and $3 per side commission.

From -$24,690, to +$2800. I increased your profit margin infinitely with 30 minutes of optimization.

Your most optimal parameters were a 7.25 point target in ES, 11 point stop, and a $375 threshold with 95% give back trailing stop. I guess the trailing stop is nonfunctional and did not improve results. Could be a breakeven trail stop what with the 95% give back.

bwolinsky · May 1, 2012

Quote from bwolinsky:

Incidentally, if you ever want to use Tradestation or Multicharts, I recommend Multicharts.

Here is your code:

inputs:
tgt(7.5),
stp(11);

variables:
bull(1),
shit(1);

if marketposition>0 then begin
bull=entryprice+tgt;
shit=entryprice-stp;
end;

if marketposition=0 and time>0929 and time<0931 and o[1]>c[1] then begin buy next bar at market;end;
if marketposition>0 then begin
sell next bar at bull limit;
sell next bar at shit stop;
end;
if marketposition>0 and time>1599 then begin sell next bar at market;end;

Below is the equity curve with a realistic 2 tick slippage and $3 per side commission.

From -$24,690, to +$2800. I increased your profit margin infinitely with 30 minutes of optimization.

Your most optimal parameters were a 7.25 point target in ES, 11 point stop, and a $375 threshold with 95% give back trailing stop. I guess the trailing stop is nonfunctional and did not improve results. Could be a breakeven trail stop what with the 95% give back.

keeptradin' · May 1, 2012

Quote from Mike805:

Okay, for the sake of gullible readers I want to say something here...

So, you're recommending the use of GA? I'm assuming you're talking about Genetic Algo's right? It sounds like you want to add 3 degrees of freedom and then let the optimizer run wild... ok, I think I get the angle here.

Assuming I'm right, the amount of stuff wrong with this process can fill a very large landfill; in fact - I don't even know where to begin...

Let's start with intent, since intent is legally as important as the actual action. By curve fitting the crap out of a simple rule, you intend to compare that result to something you claim to be "random guessing"? From that comparison, you believe you'll attain more information?

Does anyone else see this for what it is? For goodness sake, I'm actually in awe that BoWo just wrote this...

Thanks, Mike.

keeptradin' · May 1, 2012

Quote from bwolinsky:

Incidentally, if you ever want to use Tradestation or Multicharts, I recommend Multicharts.

Here is your code:

inputs:
tgt(7.5),
stp(11);

variables:
bull(1),
shit(1);

if marketposition>0 then begin
bull=entryprice+tgt;
shit=entryprice-stp;
end;

if marketposition=0 and time>0929 and time<0931 and o[1]>c[1] then begin buy next bar at market;end;
if marketposition>0 then begin
sell next bar at bull limit;
sell next bar at shit stop;
end;
if marketposition>0 and time>1599 then begin sell next bar at market;end;

Below is the equity curve with a realistic 2 tick slippage and $3 per side commission.

From -$24,690, to +$2800. I increased your profit margin infinitely with 30 minutes of optimization.

Your most optimal parameters were a 7.25 point target in ES, 11 point stop, and a $375 threshold with 95% give back trailing stop. I guess the trailing stop is nonfunctional and did not improve results. Could be a breakeven trail stop what with the 95% give back.

Thanks, Beau. Much appreciated.

I especially like how you included the word "bullshit" in your code. Very clever, in a passive/agressive sort of way.

Anyway, thanks for the time and effort, looking forward to learning the "language" so as to be able to do these types of exercises myself.

keeptradin' · May 1, 2012

Quote from dom993:

I have to agree with this - and *only* with this

It is much faster to backtest trading ideas than forward test them. Of course, backtesting results are not always indicative of future performance, but neither are forward testing results, especially when the sample size is small (say less than a few hundred trades).

I know you already stated you are not a programmer, but it would be easy enough for you to learn the basics required to backtest & enhance your VSS. For this I would use a platform supporting EasyLanguage (TradeStation, Multichart) as it is much faster learning curve than C# / java / etc. Another option would be to use Excel for backtesting, if you are more comfortable with it.

I do trade a somewhat similar system. Live P/F around 1.35 for ~115 trades since July 2011. I developed, "optimized" & validated that system in a couple weeks only.

Thanks, Dom 993. Looks like some lessons in basic programming are in my future.

Thanks to all who continue to provide their input and suggestions as to how to improve this "Very Simple System."

bwolinsky · May 1, 2012

Quote from keeptradin':

Thanks, Beau. Much appreciated.

I especially like how you included the word "bullshit" in your code. Very clever, in a passive/agressive sort of way.

Anyway, thanks for the time and effort, looking forward to learning the "language" so as to be able to do these types of exercises myself.

You're welcome. Take a look at Multicharts sometime. Easylanguage really is easy if you have basic high school programming abilities, and most people do. Finding functions isn't difficult either. The worst problem in MC is data. Without data, you'll be left with too limited of a dataset to do a minimum 2 year genetic optimization or exhaustive once you narrow down your parameters.

And, no, not even with the most optimal parameters would you ever want to trade this.

keeptradin' · May 1, 2012

Quote from bwolinsky:

You're welcome. Take a look at Multicharts sometime. Easylanguage really is easy if you have basic high school programming abilities, and most people do. Finding functions isn't difficult either. The worst problem in MC is data. Without data, you'll be left with too limited of a dataset to do a minimum 2 year genetic optimization or exhaustive once you narrow down your parameters.

And, no, not even with the most optimal parameters would you ever want to trade this.

Kewl, thanks.