Standard MACD vs Talib MACD - Different Values

Rodrigo Brito

Talib MACD returns a different value than std MACD with re-sample data.
A report of @mr-m0nst3r

Using this example: https://gist.github.com/rodrigo-brito/4f779001647ce1ce7235a7d254b7e75b
With this dataset: https://github.com/rodrigo-brito/backtrader-binance-bot/blob/master/dataset/binance_nov_18_mar_19_btc.csv

Without re-sample, it is ok:

EMA12 = 6960.21 EMA26 = 6949.12 DIFF = 11.09
STD MACD = 11.09
TALIB MACD = 11.09
--------------
EMA12 = 6961.31 EMA26 = 6950.47 DIFF = 10.84
STD MACD = 10.84
TALIB MACD = 10.84
--------------
EMA12 = 6961.38 EMA26 = 6951.31 DIFF = 10.08
STD MACD = 10.08
TALIB MACD = 10.08
--------------

But, with a re-sample of 30 minutes, it returns a different value:

cerebro.resampledata(data, timeframe=bt.TimeFrame.Minutes, compression=30)

--------------
EMA12 = 7031.31 EMA26 = 7057.54 DIFF = -26.23
STD MACD = -26.23
TALIB MACD = -83.05
--------------
EMA12 = 7012.22 EMA26 = 7046.41 DIFF = -34.19
STD MACD = -34.19
TALIB MACD = -90.54
--------------
EMA12 = 7004.46 EMA26 = 7040.14 DIFF = -35.68
STD MACD = -35.68
TALIB MACD = -88.64
--------------

The graphic result is also different:

alt text

backtrader

@Rodrigo-Brito said in Standard MACD vs Talib MACD - Different Values:

Talib MACD returns a different value than std MACD with re-sample data.

A good reason not to use ta-lib.

mr-m0nst3r

@backtrader said in Standard MACD vs Talib MACD - Different Values:

on not to use ta-lib.

Hi @backtrader ,

there're differences with standard MACD after resample the data.
Reference: https://github.com/rodrigo-brito/backtrader-binance-bot/issues/3

backtrader

Indeed. You seem to be extremely new to indicators, timeframes, periods and resampling. ta-lib has a problem with the implementation of indicators with an "unstable" period, because of a cumbersome implementation, which is easily solved (industry de-facto standard) by using a seed value.

An "Exponential Moving Average" is an example, because the definition of that average is that of a first order infinite impulse response filter. All values are always taken into account which leads to a recursive definition for updating values in real-time.

Please read: https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average

You can see that the problem is a ta-lib well-known issue: https://github.com/mrjbq7/ta-lib/issues/114

Back to square one: a good reason not to use ta-lib