[BUG?] Default MACD, Tulip MACD and Talib MACD all different results
#1
Exclamation 
Hey guys,

I posted in discord but realised that it might get lost of hijacked easily so I created a thread here.

I just cloned stable branch, installed production dependencies and npm installed talib and tulind.

After that, I ran the default MACD, Tulip-MACD and Talib-MACD. All of them with default settings over the exact same dataset. 
To my surprise, all 3 of them returned different results. And they arent even close to each other.

Im wondering if anyone faces the same issue (it would help if you could do a quick test. After all, they are all build in strategies with default params).
And if there is a solution to this. Because its going to be difficult to build an accurate strategy if 3 supposedly identical indicators provides different results.

Attached are the results from each:
MACD

.png   macd.png (Size: 57.43 KB / Downloads: 14)

TULIP MACD

.png   tulip-macd.png (Size: 57.05 KB / Downloads: 9)

TALIB MACD

.png   talib-macd.png (Size: 57.14 KB / Downloads: 11)
  Reply
#2
Hi
Yes these backtest Results..!
Did you tryed different Timespans on the same Strategie.
  Reply
#3
(01-31-2018, 04:29 PM)susitronix Wrote: Hi
Yes these backtest Results..!
Did you tryed different Timespans on the same Strategie.

Yes I did. 

I tried 2 tests: 
- 1 hour with 1 minute candles
- 1 day with 1 hour candles

Did u encounter the same issue as well?
  Reply
#4
Aha just recognized with the StochiRSI Strat that the Results can differ somehow even if its a POSITIVE-intrest-Strat-setting!
i loaded 3month of data and then backtested with 10-12 different spans, just to see what happen.

Try
1,2,3,5,7,10 -Days
2,3,4,6,9,13 -Weeks
2,3 -Month

i startet the third Stratrunner (3days)
Its hard to wait 2-3 Day for to see if it works. (even in paperTrader-mode)
A Backtest takes 1-2 Minutes!!!

Only downside: i guess Gekko only uses the CandleData for Backtest
therefor if the Strat uses a other indicator, the Backtest could be incomplete maybe. not shure...
  Reply


Forum Jump:


Users browsing this thread: