Error: ValueError: operands could not be broadcast together with shapes (12,) (11,) about quantstats HOT 1 CLOSED

I've run into this sort of error too and it has also been sporadic. After some digging, it seems that if the first year has only one data point then the code in quanstats._plotting.wrappers.yearly_returns:

returns = _utils._prepare_returns(returns).resample('A')
if compounded:
    returns = returns.apply(_stats.compsum)  <-----
else:
    returns = returns.apply(_df.cumsum)  <-----
returns = returns.resample('A').last()

returns a Series with values that range from scalars to arrays that don't have the same shape (leap year, trading day, etc. differences) instead of just scalars. The compsum or cumsum seems to group by day. Perhaps the sampler samples along the wrong axis? I wanted to suggest a fix to @ranaroussi but I couldn't figure it out.

This code reproduces the problem:

import pandas as pd
import pandas_datareader.data as pdr
import quantstats as qs

SP500 = pdr.get_data_yahoo("^GSPC", pd.to_datetime('2009-12-31'), pd.to_datetime('2020-01-02'), interval='d')
SP500["mon_ret"] =SP500["Adj Close"].pct_change()

qs.reports.html(SP500["mon_ret"], output='report2.html', title = "Monthly portfolio rebalancing")

A frustrating one because it can crop up anytime when a user-specified date range is just right (one day/month/...) in a year.

from quantstats.