Backtesting, lets talk about it!

[TipsyWisdom]

So, I do this often, and am wondering why and how it happens...I understand things that may cause it, but I just dont understand the fundamental reasoning or difference I reckon.

I do not calculate anything in pips for my projects. I do calculations based on candles OHLC, I draw lines based on OHLC of candles and I enter trades based on those lines.

I take profit in various means, RR, % of balance, and opposite conditions.

I terminate EAs based on consecutive losses and 10% DD.

With that, I am able to achieve with 1m OHLC data a graph that looks like this;

however, when I switch to real ticks. I get a EA that terminates.

and every tick

Now, I was not expecting the every tick EA to even get as far as it did....anyways, what I am left wondering is; how did every tick and real ticks produce different results if my EA is entirely controlled via a bar filter before it can do anything?

[TipsyWisdom]

...and how can I produce better 1m ohlc results. I dont care to operate on ticks, I have a brain for bars, but it must be an error in my logic that I am overlooking

[l'andorrà]

What happens while a candle is in formation is fundamental. Personally I would only focus on working with real tick only. You'll save a lot of time and pain, trust me.

[TipsyWisdom]

but I dont care for ticks...I dont care how many orders flow into the market is the premise for why ticks are irrelevant.

I only care for candle position, and when OHLC is done, then I am reacting...again, not reacting to every single trade flowing in.

I am under the impression that backtesting forces the ticks to my EA when in reality they would not be utilized. Is that a correct assumption to make?

[TipsyWisdom]

@l-andorrà said in Backtesting, lets talk about it!:

What happens while a candle is in formation is fundamental. Personally I would only focus on working with real tick only. You'll save a lot of time and pain, trust me.

im rereading what you said, and you are saying 2 different things.

candle formation is fundamental. but ticks are not any part of candle formation. this is where I fail to understand why I must have them, I am trying to avoid using them but it seems to give different results no matter what method I use...leading me to believe an all ticks or real ticks test of my candle reliable EA, is not a good way to test because its forcing false data

[TipsyWisdom]

bringing in a Youtube expert since I know ive been over this before...
https://www.youtube.com/watch?v=IYEJOTzQw5U

part 2
https://www.youtube.com/watch?v=qqVoHMg-Q_Y

[TipsyWisdom]

[seb 0]

@TipsyWisdom
Do you use stop levels ? They induce the holy grail tester inaccuracy when the price is not in ticks modelling.
Every ticks modeling seems to model the price, every tick size change, with a variation equal to the tick size. While real tick modeling, models the price every tick that actually happened, with the change that actually happened(greater or equal to tick size). That explains why the results are different.

[TipsyWisdom]

my stops are based on drawn lines of OHLC, so, yes.

[TipsyWisdom]

funny that you say that though, because i will notice that the one version of this likes to trade a lot....it will have 2 losses of what visually appears to be maybe 20 points each, will still make me lose more than a single winning trade that appears to be 3 or 4 times bigger. Ive never stopped and measure the size of them in backtest....interesting.

[TipsyWisdom]

@TipsyWisdom said in Backtesting, lets talk about it!:

@l-andorrà said in Backtesting, lets talk about it!:

What happens while a candle is in formation is fundamental. Personally I would only focus on working with real tick only. You'll save a lot of time and pain, trust me.

im rereading what you said, and you are saying 2 different things.

candle formation is fundamental. but ticks are not any part of candle formation. this is where I fail to understand why I must have them, I am trying to avoid using them but it seems to give different results no matter what method I use...leading me to believe an all ticks or real ticks test of my candle reliable EA, is not a good way to test because its forcing false data

what I said is fundamentally wrong...interesting.

I think I have some more thought to do with candle control...because I rely on Candle 0 high to be utilized, and that will significantly affect using 1m VS real ticks because it will be out of order.

Thanks for class @l-andorrà haha

[l'andorrà]

@TipsyWisdom You're welcome, man!

[TipsyWisdom]

@seb-0 said in Backtesting, lets talk about it!:

@TipsyWisdom
Do you use stop levels ? They induce the holy grail tester inaccuracy when the price is not in ticks modelling.
Every ticks modeling seems to model the price, every tick size change, with a variation equal to the tick size. While real tick modeling, models the price every tick that actually happened, with the change that actually happened(greater or equal to tick size). That explains why the results are different.

wait a minute....are you trolling me or are you being serious? lol...I was pondering this last night and was wondering if you were being serious or not. I mean, its truly terrible advice to say, trade without a stoploss. But you only mean for in terms of calculations correct? Are you saying instead, dont use a predetermined stoploss level and instead use a stoploss line, or other closing conditions...anything but SL works better from a calculation standpoint and modelling as you prescribed?

[seb 0]

@TipsyWisdom
Yes I was serious, predetermined stop levels(SL-TP) are problematic when not on ticks modelling. Or it is more problematic because there will always be inaccuracies in the tester compared to live anyway, even on ticks.
But on 1m OHLC, the code is executed every OHLC points, that are pretty far each one from each other.
So, except if by chance the next execution models the price at your exact stop levels, it will likely be exceeded by much. And I don't know exactly how but the tester in these cases "correct" this gap rather in your direction.
Such a way that if you traded randomly with stop levels on OHLC modelling, you would be winning.
To overcome this I advise, instead of using predetermined stop levels, using conditions :
Not -> buy(TP=1.2)
But -> buy(no TP) , and if price is above 1.2 next OHLC execution close it.

[roar]

@seb-0 I agree, I've thought about this effect many times myself... In essence, if your TP (or trailing SL) triggers at such a price point where the tester doesnt have data, you get extra profit all the way to the nearest available price point.

The effect is negligible if you trade big timeframes, or check your triggers on candle close only (in this case "open prices only" testing mode is perfectly fine to use), but it will cause serious problems when you make your trading actions on every tick and small scale, especially with trailing stop.

Naturally, 99% of those glorious mql5.com Market EAs abuse the heck out of this.

[seb 0]

@roar
Do you mean that stop levels are less problematic on "open prices" modelling than "1m OHLC" or "on ticks" ?
I understand that the results are more impacted when decreasing the amplitude of the trades, because of the ratio inaccuracy/amplitude. But I thought the more tight and frequent are the modelling points, the better.
I guess they can trick you this way by showing a pic of their tester grail. That's why I prefer relying only on the live results of the signals.

[roar]

@seb-0 no, I meant fixed stops are probably just as problematic on "open prices" modelling, but if you dont use fixed stops and instead do every entry and exit action once per bar, then you can use the rough modelling without noticeable difference.

I find it more my style to quickly test a lot of EA iterations on rough data, rather than focusing on 100% tick data at the cost of much slower development. Can't do any HFT strategies using this approach, naturally.

[seb 0]

@roar
I do the same, I feel it's pretty common. Stuffing on open prices to find promising and then optimizing strategies, then launch the strategy on ticks to see of it's ready for live.

[TipsyWisdom]

@roar said in Backtesting, lets talk about it!:

@seb-0 no, I meant fixed stops are probably just as problematic on "open prices" modelling, but if you dont use fixed stops and instead do every entry and exit action once per bar, then you can use the rough modelling without noticeable difference.

I find it more my style to quickly test a lot of EA iterations on rough data, rather than focusing on 100% tick data at the cost of much slower development. Can't do any HFT strategies using this approach, naturally.

ah ok, accounting for what has been said so far, I could also put all closing conditions, under a once per minute and it would also solve all this?

[TipsyWisdom]

this is good stuff gents, probably the fineer details of ea verification that really matter and no effort or thought goes into it