PandasRollingOLS does not actually work with use_const=False about pyfinance HOT 6 CLOSED

Looks good Brad. Thank you. I really like the rolling functionality - didn't think it was possible to make it this fast!

from pyfinance.

example:

from pyfinance import ols
import pandas as pd
import numpy as np
X = pd.DataFrame(index=list(range(5)), data=list(range(5)), columns=['X'])
Y = pd.DataFrame(index=list(range(5)), data=list(np.arange(0,10,2)+1), columns=['Y'])
window = 2
reg_df = pd.concat([Y, X],axis=1)
rolling_reg = ols.PandasRollingOLS(y=reg_df.iloc[:,0],  # Series
                                   x=reg_df.iloc[:,1:],  # DataFrame
                                   window=window,
                                   has_const=False,
                                   use_const=False,
                                   names=['Const','X'])
print(rolling_reg.beta)

output:

   Const    X
1    1.0  2.0
2    1.0  2.0
3    1.0  2.0
4    1.0  2.0

if you try to run without specifying a name for the constant in "names" parameter:
ValueError: Shape of passed values is (2, 4), indices imply (1, 4)

from pyfinance.

Thank you for catching this @prd1 and I'm sorry it's taken me so long to have a look. Hopefully better late than never. It seems to be a legitimate bug. The signature for ols._clean_xy() is:

def _clean_xy(y, x=None, has_const=False, use_const=True):

However, in ols.RollingOLS.__init__(), it gets called just as:

    def __init__(self, y, x=None, has_const=False, use_const=True):
        self.x, self.y, self.k = _clean_xy(y, x)

Noticeably leaving the other two parameters unpassed and so ignoring them.

I'll make a correction, push the updated changes and post a new package version on PyPI.

from pyfinance.

Closed via 5d07d1e and 4d7d7b5 - please let me know if that doesn't clear everything up @prd1. These changes are reflected on PyPI as well.

from pyfinance.

Hi Brad. Sorry still having a couple issues with constants. Using 1.2.1.

Issue1:

from pyfinance import ols
import pandas as pd
import numpy as np
X = pd.DataFrame(index=list(range(5)), data=list(range(5)), columns=['X'])
X.loc[:,'Const'] = 1
Y = pd.DataFrame(index=list(range(5)), data=list(np.arange(0,10,2)+1), columns=['Y'])
reg_df = pd.concat([Y, X],axis=1)
rolling_reg = ols.OLS(y=reg_df.iloc[:,0],  # Series
                                   x=reg_df.iloc[:,1:],  # DataFrame
                                   has_const=False,
                                   use_const=True,
                     )
print(rolling_reg.beta)

returns: ValueError: data already contains a constant
As it should. But, executing the same code with has_const = True returns:
ValueError: Data does not contain a constant; specify has_const=False

Issue 2: (I think)

from pyfinance import ols
import pandas as pd
import numpy as np
X = pd.DataFrame(index=list(range(5)), data=list(range(5)), columns=['X'])
# X.loc[:,'Const'] = 1
Y = pd.DataFrame(index=list(range(5)), data=list(np.arange(0,10,2)+1), columns=['Y'])
reg_df = pd.concat([Y, X],axis=1)
rolling_reg = ols.OLS(y=reg_df.iloc[:,0],  # Series
                                   x=reg_df.iloc[:,1:],  # DataFrame
                                   has_const=False,
                                   use_const=True,
                     )
print(rolling_reg.beta)

Above does not add a constant. Is it supposed to?

Currently, adding the constant manually and specifying has_const and use_cost both as False will give the desired result with constant.

from pyfinance.

Hi @prd1, thanks for bearing with me here. I've made a few more commits with minor fixes and tests that should correct the issues you pointed out here, too. Current version is 1.2.2 and the commits are 6de1db1 and 5046dd9. Please let me know if this clears everything up, and I'll leave things open til then. Using pip install --upgrade pyfinance should give you the latest version.

from pyfinance.