I found your posts about this on the other forum. It's interesting how using a low pass filter instead of an MA worked better. A moving average is really just a low pass filter anyway...
An EMA is a 1-pole lowpass filter; the LowPass function is a 2-pole filter. A 2-pole filter has constant amplification - and thus zero lag - below the filter frequency. That's why it works better. EMA lags at any frequency.
Been following your thread on babypips and learning your lessons on programming c for backtesting. Thanks for the great work your doing I appreciate it.
I have one question in regards to your code you posted above:
var *Price = series(price()); var *LP5 = series(LowPass(Price,5)); var *LP10 = series(LowPass(Price,optimize(10,6,20))); var *RSI10 = series(RSI(Price,10)); Stop = optimize(5,1,10)*ATR(30);
The 'Delay' variable is not defined and wont work in Zorro, I understand that it needs to be identified just having a little trouble working out what needs to be done here to rectify your code to acheive that optimised result you posted.
Regards TankWolf
Edit: Looking over this again and reading what the crossOver and crossUnder functions do is it correct that delay shouldnt actually be there and it should be 1 or -1 because the next part of the code looks to go long if crossed is > 0 since the crossOver function looks if data1 crosses over data2 so then if it does it should assign crossed = 1?
The variable "crossed" is set to 3 or -3, then counted down or up until it's 0. This has the purpose that the signal stays valid until 3 bars after the crossover.
The "static" keyword is important here, as it keeps the content of the "crossed" variable when the function is left. Otherwise, crossed could not be used as a counter because it would be set to 0 every time the run() function is called.
This is a programming trick that was not mentioned in the course. It's however described in the "Variables" chapter of the manual.
I understand what you are saying about making the int static so it keeps its variable when the function is left, what I dont understand is how the "crossed" variable is counted down or up to 0 from -3,3.
Quote:
static int crossed = 0; int Delay = 3; if(crossOver(LP5,LP10)) crossed = Delay; else if(crossUnder(LP5,LP10)) crossed = -Delay;
The way I read that code is that crossed would become either 3 or -3 after completing the first if/else statement then the next part of the code looks to see if crossed is greater than 0 and the RSI has crossOver 50 from below. If crossed is greater ie (3) the algo goes long. Where does the count down actually take place? Shouldnt there be a loop statement somewhere to count the crossed variable down or up to 0?
If a trade takes place, crosses is immediately set to 0 for not triggering a trade again. Otherwise, its sign is subtracted, which means that its absolute value is reduced by 1. The sign is -1 or 1 dependent on whether the number is negative or positive.
I have used all of the rules except for the profit & stop rules posted. I tried to incorporate them but when I did it made the results worse. This could of been however because of my poor coding. Anyway here is the code & results if anyone wants to play with it further without the profit & stop rules.
var *Price = series(priceClose()); var *EMA10 = series(EMA(Price,optimize(10,5,20))); var *EMA20 = series(EMA(Price,optimize(20,10,40))); var Stoch50 = Stoch(optimize(14,10,20),optimize(3,1,5),MAType_SMA,optimize(3,1,5),MAType_SMA);
static int crossed = 0; int Delay = optimize(3,1,5); if(crossOver(EMA10,EMA20)) crossed = Delay; else if(crossUnder(EMA10,EMA20)) crossed = -Delay;
Sorry for waking up an old thread, but I wanted to see if I could reproduce these results. I'm running with Zorro 1.58, I've copy pasted the code above, and I'm not seeing anywhere near the same equity curve.
Hi Pewi - since the English book version comes out soon, I've tested 2 weeks ago all the scripts with the latest Zorro version 1.58. You should get the same results as in the book, with some differences due to different trading costs.
Make sure you have Zorro version 1.58 and also the latest book scripts from the financial hacker website - a few things have been changed in the scripts.
@Dalla, I did not mean you, but it's the same in your case: Differences are usuaally caused by either different test periods, or by different trading costs. You can see that the curves are somewhat similar, only the profit is lower. I suppose that the result back in 2012 was based on no commission. Also the test periods look different and the default test/training split was probably different in that early version.
thanks for your quick reply. I will try with Zorro 1.58.
I am a little bit surprised that a relative short script like your optimized version of Huck's trend catcher depends so much of the Zorro version?
Also thanks for the tip with the updated book coming soon. Is there a german version planned also? I am not so familiar with english nor with such statistics nor with the special trading stuff that I am happy if at least the language barrier misses ...
BTW: Are there enough members for a german speaking sub forum?
The german book has been around for some time I think http://www.amazon.de/Das-B%C3%B6rsenhackerbuch-Finanziell-algorithmische-Handelssysteme/dp/1530310784
also with 1.58 I get for Huck's trend catcher and your optimized script only a result of 1.6k and not 16k as you (FXCM demo account). How can such a big difference happen?
I don't know, but aside from the other mentioned reasons, a 10 times smaller balance usually results from a 10 times smaller investment. I guess FXCM introduced micro accounts somewhat after 2012.
