Out of sample testing of patterns

Quote from intradaybill:

Thanks, I understand but it is equivalent to a sense, trades from a large number of systems = large number of trades.
More...

It is not equivalent AT ALL.

...what he claims is that a large number of trades will converge to the average expectancy
More...

That is not AT ALL what he is claiming.

More importantly, it says nothing as far as actually trading. It is a theoretical concept with no value to trading.
More...

Interesting perspective.

There is clearly no worthwhile conversation to be had here, so I'm out. Best of luck to you!
 
Quote from Random.Capital:

That is not AT ALL what he is claiming.
More...

No you fool, that is exactly what he is claiming. Read before you spew crap

Quote from Code7:

All 1562 possible trades have the average expectancy of 60.5 % success rate with both stop and target 7 % away from entry. That's a profit factor of about 1.53, assuming avg_win/avg_loss ratio of 1. That's the average expectancy for every single possible trade you goddam idiot.

....The overall expectancy of (many) random trade selections will always average to profit factor 1.53. That's a mathematical fact and doesn't have to be shown empirically.

I'm done with you fools now.
More...

He also makes the wrong assumption of avg_win/avg_loss ratio of 1. This means this guy has done little testing in his life, very little. Ratios of equal percent stops do not always equate to avg_win/avg_loss ratios of 1.

I am basically talking to idiots who have never traded and who misapply statistics.

Good bye. I am done with this forum idiots.
 
Quote from Mike805:

What strikes me is how defensive bill is over this issue...
More...

What strikes me is the obvious now fact that there is a team here with the same agenda. You fool, read the posts to see how insulting this guy was. That was what got me upset. I give you just two examples


Quote from Code7:

That's the average expectancy for every single possible trade you goddam idiot.

...

I'm done with you fools now.
More...

Why are you defending a volgar and insulting personality. What is your vested interest in such behaviour sir? Is it analogous to your vested interest in TS and your attack on MC?

You chose the wrong person to push your agenda.
 
Quote from Code7:

...you are either too stupid to realize it, or you are deliberately deceiving
More...
My vote's on the former. IBilly is not a rocket scientist, your explanations are simply over his head.
 
Quote from Emilio_Lizardo:

My vote's on the former. IBilly is not a rocket scientist, your explanations are simply over his head.
More...

You and code7 are imbeciles. For one, the code he used to test this is wrong. But again, what an idiot like you would now about that silly thing.

Code: 
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;

This code does not take into account exits on the entry bar. Only an idiot would use such code.

Then, you and your buddy code7 know nothing about probability and statistics.

Lastly, you know nothing about trading systems.

In short, you are idiots. I am not going to educate you because there is no hope for you with such attitude you have. You are low lives.
 
You left out important information to deliberately deceive again you lying POS.

Quote from Code7:

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.
More...
Only very few days had >= 7% range from open to high/low and if they did, chances are the result in terms of stopped-out or target-hit was the same for exit on the next day. So the success rate only differed by half a percentage point. And I did point out there was almost no difference. I used both methods of calculation to confirm each other.

Here is proof that you in fact lied.

You claimed I compared 30 trades in 6 years to 2000 trades.
http://www.elitetrader.com/vb/showthread.php?s=&postid=2818730#post2818730

I replied, no I didn't, Harris' patterns had 128 trades.
http://www.elitetrader.com/vb/showthread.php?s=&postid=2819246#post2819246

Then you replied, when you said 30 trades, you were referring to the 27 trades of the 2nd out-of-sample test.
http://www.elitetrader.com/vb/showthread.php?s=&postid=2819276#post2819276

But as it turns out, all 27 trades of your own 2nd out-of-sample test took place in the single year 2009.
http://www.elitetrader.com/vb/showthread.php?s=&postid=2816649#post2816649

I think the reason why you have to lie and lie again and even threaten with legal action is because you are intellectually inferior and can't keep up otherwise. Quite sad.
 
intradaybill cannot backtest correctly and code7 is a stupid geek. This APS thing was so intriguing to me that I spent the money to get a license and I do not regret it. First thing I did was to try to replicated intradaybill's APS results for QQQQ and then try to replicate code7's claims. My results are shown below for QQQQ data from 05/07/2002 to 12/31/2008. I think since APS QQQQ system for these settings is a long-only system, including data for the end of 2008 should do more harm to its performance than good. I had APS generate Amibroker AFL code for each of the patterns and I also enabled multiple backtest mode to get all signals generated by the 8 different patterns, which I combined using the OR Boolean function into a single BUY signal.

APS results: profit factor 1.65, win rate 61.73% trades 277

http://www.4shared.com/photo/MxwLXkDO/AB_APS_2002_2008.html

Then I tested a system that buys all the open bars in the same period with same R:R

http://www.4shared.com/photo/1XthiQfE/AB_ALLBARS_2002_2008.html

profit factor 1.38, win rate 58.11% trades 1676

So, after about 1 hour of hard work:) I have come to the conclusion that intradaybill - although a backtesting idiot - was nevertheless correct and in that time period APS patterns performed very well.

I have also tested several other R:R ratios and APS produced much better results and also generated several short patterns.

By the way, Amibroker is a terrific tool. Just because you intrabill you used MS and you code7 you used TS I declare you both idiots.
 
Thanks goodgoing, it's quite refreshing to see some actual content here.

I'd like to add two things.

First, please note that my results are always for the very same period of 05/07/2002 to 08/22/2008 that Harris also used for his test in his article.

Second, Harris himself got 128 trades up to 08/22/2008.
http://www.tradingpatterns.com/Literature/article3/article3.html#update

So it's a bit strange that you got more than double that amount, 277 trades when you only tested a few months longer up to 12/31/2008. I'm not sure if you really backtested "the same thing" I compared my results to.

Edit: I just saw you used some kind of position sizing. In this case, your profit factors are screwed. APS backtests one share only. The results in the article were done with one share only.
 
Code7,

You actually made 3 comments.

- About the number of trades. I allowed multiple entries for each pattern. This is why the number if greater. Harris allows multple pattern entries but not multiple entries for each patterns. This IMO is based on a certain philosophy about treating signals. I think Harris' method works aginst him in this particular example but for other examples it may work for him.

- About testing period and position sizing. I used the same position sizing for both test (Harris patterns and all bars) but I repeated the results for constant shares =1, per entry. I also repeated the tests for the period in the Harris article (up to 8/22/2008). I did that for multiple positions and for no multiple. I do not have time to do it so that you get multiple from different patterns but not from same, as Harris does (not that it would matter that much).

- Trading all bars equally gives 60.72% win rate and pf = 1.41, 1586 trades

http://www.4shared.com/photo/Ouc3wAHH/AB_ALLBARS_20080822.html

- But Harris and APS outperform that with win rate = 64.59% and pf = 1.71, 257 trades (multiple entries).

http://www.4shared.com/photo/whvnb_wS/AB_APS_20080822_1share_multipl.html

Now, if no multiple entries are allowed at all, the win rate for Harris and APS drops to 56.25% and pf = 1.28, trades = 48.

http://www.4shared.com/photo/vwxqMVN1/AB_APS_20080822_1share_no_mult.html

In this last case, although the profit factor is less than "all bars", that is done at a fraction of the number of trades and required capital, naturally allowing equity to be invested in other systems.

Thus, surprisingly enough, I conclude that even in the case of constant share trading during the test period of the article, the APS pattern performance could NOT be obtained by chance and there must be some timing involved to result in a 21% increase in pf over trading that equally weighs all bars.
 
OK, thanks. Your way of trading the patterns is very interesting and I'm surprised there isn't an option for this already built into APS. Still, I think the results are no more out-of-sample if you change the interpretation method afterwards to what would have been best.

I always referred to the results Harris published on his website. By just taking the numbers from his table and calculating the trade-weighted PF, I get 1.50 for his 128 trades. This is done under the assumption that the amounts of avg_win and avg_loss for each pattern are comparable, the actual PF for all 8 patterns combined might be different.

If 1.50 is correct, then yes, that's slightly better than what would have been expected due to chance alone, your number is 1.41 here. It could still be argued how convincing that difference of 6.4% is. Anyway, none of that tells us if Harris' results were in fact random or not because that's just correlation, not causation. Even a truly random system can give you great results for one sample of 128 trades over several years but the chances of that happening are very slim.
 
You must log in or register to reply here.
Back
Top