VectorBT vs TA-Lib | RSI discrepancies

[danigarod]

Hi there.
I am trying to understand how VectorBT calculates the RSI indicator. I have the following code to compare VectorBT RSI values againts talib:

import vectorbt as vbt
import talib as ta
import pandas as pd
import yfinance as yf

symbol ='SPY'
data = yf.download(symbol)
data = data.dropna()

##################################

RSI Check

##################################
rsi_1 = ta.RSI(data.Close, 3)
rsi_1 = pd.Series(rsi_1, name="rsi_1")

rsi_2= vbt.RSI.run(data.Close, 3, ewm=True).rsi
rsi_2 = pd.Series(rsi_2, name="rsi_2")

merged = pd.merge(rsi_1, rsi_2, on='Date', how='outer')

merged.head(50)
And these are the results:

rsi_1 rsi_2

Date

1993-01-29 NaN NaN

1993-02-01 NaN NaN

1993-02-02 NaN NaN

1993-02-03 100.000000 100.000000

1993-02-04 100.000000 100.000000

1993-02-05 94.267516 89.333333

1993-02-08 94.267516 89.333333

1993-02-09 41.168289 16.962025

1993-02-10 49.672814 37.284895

1993-02-11 71.387503 76.885130

1993-02-12 35.392258 25.761511

1993-02-16 10.184007 4.813165

I am trying to dig into VectorBT code but it is hard to follow. Anyone has already looked into this before?

Thanks a lot!

[polakowo]

Duplicate of #471