Avoiding Curve fitting

[Code7]

Quote from bashatrader:

I did my own analysis and I calculated a success rate slightly better than 50% in the case of long-only trades. This makes sense. It appears that you selected a sample biased enough to match the performance of the Harris system so you can declare it random.

No, it doesn't make sense. QQQQ went up from 29.27 to 47.49 in the testing period. Please provide full disclosure of your analysis and please be very specific. Put up or shut up.

I'll go first. I used daily QQQQ data from TradeStation and this code for my test.

Inputs: Num(1);
Vars: EP(0);

if CurrentBar = Num then
	Buy next bar on Open;

if CurrentBar = Num + 1 then
	EP = Open;

if CurrentBar > Num then
begin
	Sell next bar at EP * 1.07 Limit;
	Sell next bar at EP * 0.93 Stop;
end;

I raised Num from 1 to 2000 by 1 to buy on all days and then looked up the results by sorting the optimization report. So in essence, I just weighted all days equally without any bias whatsoever. Random entries will average to the same winning %.

I verified the results of the above code by comparing against the built-in SetProfitTarget and SetStopLoss. Almost identical outcome, winning % very close to 60. I also accounted for the MaxBarsBack value and made sure my first trade was on 05/07/2002.

Unlike the built-in TS stop and target, the above code doesn't exit on the entry day. However, the difference in results was minimal due to a total sample size of > 1500. We are talking about different winning % of +- 0.5 points here.

I'm very interested to learn how you got slightly better than 50 and not 60.
Beyond that, your twisted drivel and bizarre claims are not worth discussing.

 

[Jerry030]

Quote from Muskoka Joe:

Quote from Jerry030:


Components 14 day RSI:
1 - 14 : daily gains each day
15 to 28 ; daily losses each day
29; Average gain
30 Average loss

total components = 30

==================================

I'll just disagree with this, as it seems totally off to me, but maybe I just can't figure it out. Sometimes I wish I was smarter, but I wonder if it would help in an arena that is really not a very intellectual pursuit.

It helps if you map out all the detail calculations for any TI in Excel so that they are all visible.

For work with non-linear models more information is much better then the dumbed down final TIs. So while the RSI appears to be a single number, a length parameter and a rule for interpretation, what is actually required for calculation is some 30 separate components or numbers: gains daily for each day in the time window, loss values and average values.

If you lay out the calculating the average with D1 + D2 + D3 .... D14 then divide by 14 as opposed to using the Excel average function you will need 30 active cells for calculation.

These should be input to nonlinear analysis in addition to the final RSI value. If you use software that can rate the predictive contribution of each element you will find that for most active markets the RSI calculation sub components have more predictive capacity than the RSI itself.

 

[Stoxtrader]

If you use software that can rate the predictive contribution of each element you will find that for most active markets the RSI calculation sub components have more predictive capacity than the RSI itself.

What software have you used to calculate predictive coefficients? Is there a name for this calculation or is it an umbrella term for regression, forecasting, classification, etc, combined with cross-validation?

Data mining software poll:
http://www.kdnuggets.com/polls/2009/data-mining-tools-used.htm

 

[Jerry030]

Quote from Stoxtrader:

What software have you used to calculate predictive coefficients? Is there a name for this calculation or is it an umbrella term for regression, forecasting, classification, etc, combined with cross-validation?

Data mining software poll:
http://www.kdnuggets.com/polls/2009/data-mining-tools-used.htm

Like a good carpenter I use a fair number of software tools. Of those listed I occasionally use Rapid Miner and S-Plus. I also use a number of others. What you are looking for in a NN application is the ability to rank the final weights given to each input variable. In general terms this can be done by systematically varying the values of the independent variables and measuring their contribution to the output signal

 

[bashatrader]

Quote from Code7:

I'll go first. I used daily QQQQ data from TradeStation and this code for my test.

Inputs: Num(1);
Vars: EP(0);

if CurrentBar = Num then
	Buy next bar on Open;

if CurrentBar = Num + 1 then
	EP = Open;

if CurrentBar > Num then
begin
	Sell next bar at EP * 1.07 Limit;
	Sell next bar at EP * 0.93 Stop;
end;

ROFLMAO!

I just thought you may be just a kid playing around with a computer and a language you do not understand how it works. You also appear vicious and insulting when confronted and you seem to totally miss the essence of the points made. I suggest you first learn EL before exposing yourself. It is clear to me now you are not a trader. So long.

 

[Code7]

Looks like you decided to not put up and fully disclose how you

Quote from bashatrader:

calculated a success rate slightly better than 50% in the case of long-only trades.

Probably because you are a liar.

You also gave not a single valid reason why my winning percentage of 60.5 +- 0.5 points was allegedly not accurate. Just some more useless drivel without any substance or content.

 

[Rocko Bonaparte]

Quote from Jerry030:

I've found that in some cases a NN will learn more from the mathematical components used to calculate the indicator along with the indicator itself. Keep in mind that indicators were invented to dumb down complex market dynamics so it can be shown on a 2 dimensional chart. NN are inherently N dimensional and don't have this limitation. For example the components in the RSI model much better than the RSI itself.

Yes this is what happened when I got the fools gold. I was normalizing the inputs to a span of zero to one to make it more suitable for the NN models I was using. I was doing that by creating a normal distribution of my test window, which involved getting a mean and standard deviation.

The two flaws:
1. I screwed up the call to calculate the standard deviation; the way it worked was counter-intuitive to how I thought I was supposed to give it stuff.
2. I calculated the mean of the entire backtest window up front, when it should have been a rolling window. I failed because I was incorporating future information into the mean.

With those two errors, the neural network was able to take the crap indicators and make something out of them. Something that wouldn't forward test, but something that wasn't 100% perfect yet portrayed ridiculous returns.

 

[bashatrader]

Quote from Code7:

Looks like you decided to not put up and fully disclose how you

Probably because you are a liar.

You also gave not a single valid reason why my winning percentage of 60.5 +- 0.5 points was allegedly not accurate. Just some more useless drivel without any substance or content.

Liar? Another insult? Does your mom know you are insulting people using her computer while she is working trying to feed you?

Listen to me. You do not understand how EL works. I have no time posting results to educate you. What you have done is program a trend following system. It is reasonable that you will get a winning bias since there was a trend, but only in hindsight. The system you are attemtping to compare it to was a position/swing trading one. You are comparing apples and oranges. I have come accross many like you before. Your "know it all" attitude and autistic behavior is probably due to a well-known syndrome:

http://en.wikipedia.org/wiki/Asperger_syndrome

ROFLMAO!

P.S. Thinking about it now, what you have done is to prove that the APS system did as well as a trend following system. Actually, you proved that the pattern system was as good as a system developed in hindsight that followed the trend but did so unconsciously due an erroneous application of EL.

 

[Kevin Schmit]

Quote from Code7:

Looks like you decided to not put up and fully disclose how you
Probably because you are a liar.

You are wasting your time responding to bashatrader, an exceptionally moronic poster even by ET standards.

Put him on ignore.

 

[bashatrader]

Quote from Kevin Schmit:

You are wasting your time responding to bashatrader, an exceptionally moronic poster even by ET standards.

Put him on ignore.

The crossdresser is moaning at himself