MACD Value Discrepancy

This topic has 15 replies, 3 voices, and was last updated 3 months ago by druby.

Viewing 15 posts - 1 through 15 (of 16 total)

1 2 →

07/11/2024 at 10:30 AM #235060

VMarkov

Hi all,

I am trying to backtest a strategy that relies on the precise values of the MACD line and signal. However, the ProBacktest values always seem to be slightly different to the indicator (sometimes the difference is quite severe and the actual histogram is wrong).

I’ve attached a screenshot of what this looks like. For this example, I’ve simply graphed the MACD line and signal values at each current bar to see what exactly is going on.

line1 = MACDLine[12, 26, 9](Close)
line2 = MACDLine[12, 26, 9](Close)[1]

signal1 = MACDSignal[12, 26, 9](Close)
signal2 = MACDSignal[12, 26, 9](Close)[1]

graph line1 as "line1" 
graph signal1 as "signal1"

//<Order Logic Below>

line1 = MACDLine[12, 26, 9](Close)

line2 = MACDLine[12, 26, 9](Close)[1]

signal1 = MACDSignal[12, 26, 9](Close)

signal2 = MACDSignal[12, 26, 9](Close)[1]

graph line1 as "line1"

graph signal1 as "signal1"

//<Order Logic Below>

I’ve tried playing around with Preloadbars and the number of units to no avail. This is also run on a single TF.

If anyone can shed some light on this, that would be very much appreciated.

Thank you.

VMarkov

Participant

Topics: 1

Replies: 5

Been thanked: 0 times

07/11/2024 at 12:50 PM #235077

So, there is a (big) difference between your system and the indicator, I see that you use Forex as data, is this difference also there when you use an index, for example…?

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/11/2024 at 1:56 PM #235086

VMarkov

Yes, only dealing with Forex at the moment (I should have mentioned that in the original post).

You are correct, even when previous values of MACD are referenced ([1], [2], etc) the difference is present. It simply spits out the value that it calculated previously, still with a big difference.

VMarkov

Participant

Topics: 1

Replies: 5

Been thanked: 0 times

07/11/2024 at 2:06 PM #235087

Can you show (part of) your strategy?

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/11/2024 at 3:03 PM #235095

VMarkov

It doesn’t quite matter what the strategy is as you will never get the correct MACD values anyway. Take this very basic example that tries to capture the movement of a water line cross.

DEFPARAM CumulateOrders = False
DEFPARAM Preloadbars = 10000

line0 = MACDLine[12, 26, 9](Close)
line1 = MACDLine[12, 26, 9](Close)[1]
line2 = MACDLine[12, 26, 9](Close)[2]

signal0 = MACDSignal[12, 26, 9](Close)
signal1 = MACDSignal[12, 26, 9](Close)[1]
signal2 = MACDSignal[12, 26, 9](Close)[2]

hist0 = MACD[12, 26, 9](Close)
hist1 = MACD[12, 26, 9](Close)[1]
hist2 = MACD[12, 26, 9](Close)[2]

if (hist0 < 0 AND hist1 < 0 AND hist2 < 0 AND line0 < 0 AND signal0 < 0 AND line1 < 0 AND signal1 > 0AND line2 > 0 AND signal2 > 0) then

    entry = Close[0]

    stopLevel = High[0]
    stopUnits = stopLevel - entry

    risk = 500.0 / ((stopUnits) * 10000.0)

    SELLSHORT risk SHARES AT MARKET
    SET STOP LOSS stopUnits
endif

if(OnMarket AND hist0 > 0) then
    BUY AT MARKET
endif

graph line0 as "line0"
graph signal0 as "signal0"
graph hist0 as "hist0"

DEFPARAM CumulateOrders = False

DEFPARAM Preloadbars = 10000

line0 = MACDLine[12, 26, 9](Close)

line1 = MACDLine[12, 26, 9](Close)[1]

line2 = MACDLine[12, 26, 9](Close)[2]

signal0 = MACDSignal[12, 26, 9](Close)

signal1 = MACDSignal[12, 26, 9](Close)[1]

signal2 = MACDSignal[12, 26, 9](Close)[2]

hist0 = MACD[12, 26, 9](Close)

hist1 = MACD[12, 26, 9](Close)[1]

hist2 = MACD[12, 26, 9](Close)[2]

if (hist0 < 0 AND hist1 < 0 AND hist2 < 0 AND line0 < 0 AND signal0 < 0 AND line1 < 0 AND signal1 > 0AND line2 > 0 AND signal2 > 0) then

entry = Close[0]

stopLevel = High[0]

stopUnits = stopLevel - entry

risk = 500.0 / ((stopUnits) * 10000.0)

SELLSHORT risk SHARES AT MARKET

SET STOP LOSS stopUnits

endif

if(OnMarket AND hist0 > 0) then

BUY AT MARKET

endif

graph line0 as "line0"

graph signal0 as "signal0"

graph hist0 as "hist0"

The discrepancies are enough to provide false entries. Compare this to an identical indicator in ProBuilder without the order placement.

line0 = MACDLine[12, 26, 9](Close)
line1 = MACDLine[12, 26, 9](Close)[1]
line2 = MACDLine[12, 26, 9](Close)[2]

signal0 = MACDSignal[12, 26, 9](Close)
signal1 = MACDSignal[12, 26, 9](Close)[1]
signal2 = MACDSignal[12, 26, 9](Close)[2]

hist0 = MACD[12, 26, 9](Close)
hist1 = MACD[12, 26, 9](Close)[1]
hist2 = MACD[12, 26, 9](Close)[2]

enter = 0

if (hist0 < 0 AND hist1 < 0 AND hist2 < 0 AND line0 < 0 AND signal0 < 0 AND line1 < 0 AND signal1 > 0AND line2 > 0 AND signal2 > 0) then
    enter = 1
endif

return enter

line0 = MACDLine[12, 26, 9](Close)

line1 = MACDLine[12, 26, 9](Close)[1]

line2 = MACDLine[12, 26, 9](Close)[2]

signal0 = MACDSignal[12, 26, 9](Close)

signal1 = MACDSignal[12, 26, 9](Close)[1]

signal2 = MACDSignal[12, 26, 9](Close)[2]

hist0 = MACD[12, 26, 9](Close)

hist1 = MACD[12, 26, 9](Close)[1]

hist2 = MACD[12, 26, 9](Close)[2]

enter = 0

if (hist0 < 0 AND hist1 < 0 AND hist2 < 0 AND line0 < 0 AND signal0 < 0 AND line1 < 0 AND signal1 > 0AND line2 > 0 AND signal2 > 0) then

enter = 1

endif

return enter

I hope someone can clarify this strange behaviour.

VMarkov

Participant

Topics: 1

Replies: 5

Been thanked: 0 times

07/11/2024 at 3:36 PM #235100

When I use your last example/information, I get (almost) the same values for the indicator as for the system for the MACD…

1 user thanked author for this post.

VMarkov

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/11/2024 at 3:58 PM #235103

VMarkov

I get areas that are relatively close and some that are not. Are yours the same?

In theory, shouldn’t they be exactly the same? I still don’t want to accept best case scenario as they are “close enough” as this is quite annoying for what I am trying to do.

VMarkov

Participant

Topics: 1

Replies: 5

Been thanked: 0 times

07/11/2024 at 4:13 PM #235104

The difference you see are “rounding errors” and only noticeable after the fourth decimal place… (six decimal places doesn’t make sense anyway)

I’ve checked the other areas, and the values are the same…

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/11/2024 at 4:29 PM #235105

You can use this mathematical function in your strategy, to adjust the decimal places (e.g. 2 decimal places) …

https://www.prorealcode.com/documentation/round/

2 users thanked author for this post.

Iván,

VMarkov

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/11/2024 at 4:43 PM #235113

druby

The stock, custom and custom indicator called from another file all align up.

However, results from the backtest versions do slightly deviate.

Can’t shed any further light on that.

On a different observation, you can write the code this way as well. Because most variables store its value for each bar, you can use the [‘n’] to get previous values. This way you only have to repeatedly call external code for one since there all got the same settings.

iMacd   = MACDLine[12, 26, 9](Close)
iSignal = MACDSignal[12, 26, 9](Close)
iHist   = MACD[12, 26, 9](Close)

enter = 0
if (iHist < 0 AND iHist[1] < 0 AND iHist[2] < 0 AND iMacd < 0 AND iSignal < 0 AND iMacd[1] < 0 AND iSignal[1] > 0 AND iMacd[2] > 0 AND iSignal[2] > 0) then
enter = 100
endif

return enter

iMacd = MACDLine[12, 26, 9](Close)

iSignal = MACDSignal[12, 26, 9](Close)

iHist = MACD[12, 26, 9](Close)

enter = 0

if (iHist < 0 AND iHist[1] < 0 AND iHist[2] < 0 AND iMacd < 0 AND iSignal < 0 AND iMacd[1] < 0 AND iSignal[1] > 0 AND iMacd[2] > 0 AND iSignal[2] > 0) then

enter = 100

endif

return enter

1 user thanked author for this post.

VMarkov

druby

Participant

Topics: 1

Replies: 348

Been thanked: 261 times

07/11/2024 at 5:10 PM #235115

VMarkov

This is quite a shame as I do need exact values since some instruments have MACD values only beginning at the 4th decimal (in some cases 5th). So it’s not an option for me to round.

Is it possible to get an official confirmation as to why this is happening, or a workaround?

Thanks for your help so far guys.

VMarkov

Participant

Topics: 1

Replies: 5

Been thanked: 0 times

07/11/2024 at 8:31 PM #235118

It seems that the BackTest module calculates with a maximum of 5 decimal places, while an Indicator calculates with 6 decimal places…

5 decimal places should be enough for Forex, because here, with most pairs, 5 decimal places are also used (tick size = 0.00001) …

3 users thanked author for this post.

Iván,

VMarkov,

druby

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/11/2024 at 9:33 PM #235123

druby

That solves one of life’s great mystery’s…

iMacd = round(MACDline[12,26,9](close)),5) in an indicator matches

iMacd = MACDline[12,26,9](close) in backtest.

3 users thanked author for this post.

Iván,

VMarkov,

druby

Participant

Topics: 1

Replies: 348

Been thanked: 261 times

07/11/2024 at 9:43 PM #235125

Plenty of other “Great Mystery’s” in PRT to solve… 🙂

Participant

Topics: 26

Replies: 1695

Been thanked: 766 times

07/12/2024 at 3:46 AM #235128

VMarkov

Glad we were able to get to the bottom of this. This discrepancy in decimal precision surely affects other indicators as well, and not just MACD.

I’ve checked both the IG and PRT Trading platforms and this looks like a fundamental issue that impacts the accuracy of all calculations.

Accurate indicators are the backbone of any reliable strategy, so it is legitimately beyond me why this is something we have to discuss and “deal with” in the first place.

VMarkov

Participant

Topics: 1

Replies: 5

Been thanked: 0 times

Author

Posts

Viewing 15 posts - 1 through 15 (of 16 total)

1 2 →

Find exclusive trading pro-tools on