I have a df called "log_returns_5d". Here's a small sample:
Gasoil Gasoline Oil Gas
Date
2011-01-07 0.009059 -0.007125 -0.037390 0.003793
2011-01-10 -0.013459 0.011151 -0.025067 -0.054915
2011-01-11 0.032774 0.027211 0.019039 -0.040233
For all assets (column headers) I want to calculate pairwise ewm correlations. I tried:
st_correlations = log_returns_5d.ewm(alpha=0.01, adjust=False).corr(method='pearson', min_periods=2000)
but I got a 3-dimensional df, with pairwise correlations being calculated for every single date:
Gasoil Gasoline \
Date
2010-12-31 Gasoil NaN NaN
Gasoline NaN NaN
Oil NaN NaN
Gas NaN NaN
Canola NaN NaN
... ... ...
2021-07-28 CNHJPY 0.295644 0.189685
SEKJPY 0.452436 0.418866
NOKJPY 0.588999 0.591803
NOKSEK 0.497179 0.542514
MXNJPY 0.322246 0.349632
In order to get only the most updated correlations, I wrote the following:
st_correlations = st_correlations.reset_index()
max_date = st_correlations.Date. max()
st_correlations = st_correlations.loc[(st_correlations['Date'] == max_date)]
which works. The whole code is quite slow to run though. Is there a way to get what I need from the begininning?
I think my problem is similar to the one described here, but I can't get to apply the proposed solution.
