Var Models

[legant]

I have a file fr1.txt. A sample of data:

date, price1,price2,price3
2009-11-09 11:05:21pm,3.2,31.4,22.1
2009-11-09 11:07:23pm,40.5,35.8,38.2
2009-11-09 11:43:10pm,11.9,32.1,58.8
2009-11-10 12:22:07am,22.7,65.9,31.8
2009-11-10 1:43:11am,98.32,54.3,21.5
2009-11-10 2:28:59am,95.2,33.4,21.9
2009-11-11 12:33:39am,51.7,32.1,16.9
2009-11-11 12:34:26am,21.5,15.8,10.2
2009-11-11 12:40:16am,31.4,2.4,21.6
2009-11-11 12:45:22am,5.6,2.2,41.9

The code (using the example of here http://statsmodels.sourceforge.net/devel/vector_ar.html#var)

import numpy as np

import pandas as pd

import datetime as datetime

from statsmodels.tsa.api import VAR, SVAR

import matplotlib.pyplot as plt

import statsmodels.api as sm

parse = lambda x: datetime.datetime.strptime(x, '%Y-%m-%d')

dframe = pd.read_table("fr1.txt", delimiter=",", index_col=0, parse_dates=True, date_parser=parse)

ts = dframe.resample("D")

mdata = ts[['price1','price2','price3']]

names = mdata.dtype.names

data = mdata.view((float,3))

data = np.diff(np.log(data), axis=0)

model = sm.tsa.VAR(data, names=names)

results = model.fit(2)

results.summary()

model.plot()

plt.show()

The error:
AttributeError: 'DataFrame' object has no attribute 'dframe'

[josef-pkt]

you are mixing working with pandas and working with recarrays. The doc example uses recarrays throughout.

I cannot replicate your example because of an exception in date parsing. Also, it is better if you include the full traceback, since I don't know where your exception occurs.
I get an exception with names = mdata.dtype.names

If I remove the date parsing, then it works if I convert pandas to recarray

mdata = ts.to_records()[['price1','price2','price3']]

But I think it would be better to work (or make it to work) with pandas all the way.

Aside: the VAR documentation hasn't been updated in a while, and doesn't take advantage of the pandas integration, or provide examples for it, I guess.

[legant]

Thank you. Traceback:

names = mdata.dtype.names
File "C:\Python27\lib\site-packages\pandas\core\frame.py", line 1771, in getattr
(type(self).name, name))
AttributeError: 'DataFrame' object has no attribute 'dtype'

Try this for data set:
date,p1,p2,p3
2009-12-12 11:15:39pm,0.375,0.375,0.5
2009-12-12 11:22:10pm,0.2575,0.375,0.0
2009-12-12 11:43:10pm,0.2,0.1,0.3
2009-12-12 11:44:07pm,0,0,0
2009-12-13 12:23:11am,0,1,0
2009-12-13 12:28:59am,0.0,0.0,0.0
2009-12-13 12:33:39am,1,0,1

Can I use pandas for var model?

[josef-pkt]

I think so, but it needs a datetime index. I get an exception with an integer index. (I don't have a timeseries for pandas handy right now.)

otherwise you can always use

>>> model = sm.tsa.VAR(np.asarray(data), names=data.columns)
E:\Josef\testing\tox\py27b\lib\site-packages\statsmodels-0.5.0-py2.7-win32.egg\statsmodels\tsa\vector_ar\var_model.py:340: FutureWarning: The names argument is deprecated and will be removed in the next release.
  "removed in the next release.", FutureWarning)

I don't know what's the substitute for names argument.

[josef-pkt]

this needs checking for 0.5

What's the status? Do we have pandas support in VAR?

[jseabold]

I don't see why not. If you give a pandas DataFrame to the timeseries models then they expect a DatetimeIndex

https://github.com/statsmodels/statsmodels/blob/master/statsmodels/tsa/base/tsa_model.py#L25

The names are done automatically now like the other models. It uses the fields from a recarray or the column names from a DataFrame.

[josef-pkt]

If it works and we have at least a smoke test, then we can close this issue.

(I'm just trying to figure out what the status with some of the issues is.)

[jseabold]

I think this should be covered. Looks like this was a problem with trying to use a DataFrame as a structured array by following the docs not something internal. Maybe leave it open as a reminder to update the docs?

Thanks for the maintenance work. Am hoping to be able to pitch in in March after I meet a few deadlines and get a release out the door. We were bad about bug-fixes / features in this release cycle (again) and have some stuff that needs to be released (and, of course, some stuff that's not and might not be ready).