def __init__(self, data, lags, intercept=True):
self._data = DataFrame(_combine_rhs(data))
self._p = lags
self._columns = self._data.columns
self._index = self._data.index
self._intercept = intercept
def __init__(self, data, p=1, intercept=True):
import scikits.statsmodels.tsa.var as sm_var
self._data = DataFrame(_combine_rhs(data))
self._p = p
self._columns = self._data.columns
self._index = self._data.index
self._intercept = intercept
def beta(self):
"""
Returns a DataFrame, where each column x1 contains the betas
calculated by regressing the x1 column of the VAR input with
the lagged input.
Returns
-------
DataFrame
"""
d = dict([(key, value.beta)
for (key, value) in self.ols_results.iteritems()])
return DataFrame(d)
def resid(self):
"""
Returns the DataMatrix containing the residuals of the VAR regressions.
Each column x1 contains the residuals generated by regressing the x1
column of the input against the lagged input.
Returns
-------
DataMatrix
"""
d = dict([(col, series.resid)
for (col, series) in self.ols_results.iteritems()])
return DataFrame(d, index=self._index)
def granger_causality(self):
"""Returns the f-stats and p-values from the Granger Causality Test.
If the data consists of columns x1, x2, x3, then we perform the
following regressions:
x1 ~ L(x2, x3)
x1 ~ L(x1, x3)
x1 ~ L(x1, x2)
The f-stats of these results are placed in the 'x1' column of the
returned DataMatrix. We then repeat for x2, x3.
Returns
-------
Dict, where 'f-stat' returns the DataMatrix containing the f-stats,
and 'p-value' returns the DataMatrix containing the corresponding
p-values of the f-stats.
"""
from pandas.stats.api import ols
from scipy.stats import f
d = {}
for col in self._columns:
d[col] = {}
for i in xrange(1, 1 + self._p):
lagged_data = self._lagged_data[i].filter(self._columns -
[col])
for key, value in lagged_data.iteritems():
d[col][_make_param_name(i, key)] = value
f_stat_dict = {}
p_value_dict = {}
for col, y in self._data.iteritems():
ssr_full = (self.resid[col]**2).sum()
f_stats = []
p_values = []
for col2 in self._columns:
result = ols(y=y, x=d[col2])
resid = result.resid
ssr_reduced = (resid**2).sum()
M = self._p
N = self._nobs
K = self._k * self._p + 1
f_stat = ((ssr_reduced - ssr_full) / M) / (ssr_full / (N - K))
f_stats.append(f_stat)
p_value = 1 - f.cdf(f_stat, M, N - K)
p_values.append(p_value)
f_stat_dict[col] = Series(f_stats, self._columns)
p_value_dict[col] = Series(p_values, self._columns)
f_stat_mat = DataFrame(f_stat_dict)
p_value_mat = DataFrame(p_value_dict)
return {
'f-stat': f_stat_mat,
'p-value': p_value_mat,
}
