ENH: Let hold_back and last_obs work with dates

Provide functionality so that these can accpet wither int or dates
Add testing for first_obs and last_obs
Improve appearance of plot

arch/base.py

@@ -931,6 +931,7 @@ def plot(self, annualize=None, scale=None):
         ax = fig.add_subplot(2, 1, 1)
         ax.plot(self._index, self.resid / self.conditional_volatility)
         ax.set_title('Standardized Residuals')
+        ax.axes.xaxis.set_ticklabels([])
 
         ax = fig.add_subplot(2, 1, 2)
         vol = self.conditional_volatility

arch/mean.py

@@ -520,7 +520,9 @@ def _fit_no_arch_normal_errors(self, cov_type='robust'):
         resids = np.empty_like(self._y, dtype=np.float64)
         resids.fill(np.nan)
         resids[first_obs:last_obs] = e
-        vol = np.zeros_like(resids) * np.sqrt(sigma2)
+        vol = np.zeros_like(resids)
+        vol.fill(np.nan)
+        vol[first_obs:last_obs] = np.sqrt(sigma2)
         names = self._all_parameter_names()
         loglikelihood = self._static_gaussian_loglikelihood(e)
 

arch/tests/test_mean.py

@@ -255,10 +255,20 @@ def test_ar(self):
         ar.__repr__()
         ar = ARX(self.y_series, lags=5)
         ar.__repr__()
+        res = ar.fit()
+        assert_true(isinstance(res.resid, pd.Series))
+        assert_true(isinstance(res.conditional_volatility, pd.Series))
+        # Smoke tests
         summ = ar.fit().summary()
         ar = ARX(self.y, lags=1, volatility=GARCH(), distribution=StudentsT())
-        res = ar.fit(iter=5)
+        res = ar.fit(iter=5, cov_type='mle')
+        res.param_cov
         res.plot()
+        res.plot(annualize='D')
+        res.plot(annualize='W')
+        res.plot(annualize='M')
+        res.plot(scale=360)
+
 
 
     def test_arch_model(self):
@@ -307,6 +317,10 @@ def test_arch_model(self):
         am = arch_model(self.y, vol='arch')
         assert_true(isinstance(am.volatility, ARCH))
 
+        am = arch_model(self.y, vol='egarch')
+        assert_true(isinstance(am.volatility, EGARCH))
+
+
         assert_raises(ValueError, arch_model, self.y, mean='unknown')
         assert_raises(ValueError, arch_model, self.y, vol='unknown')
         assert_raises(ValueError, arch_model, self.y, dist='unknown')
@@ -445,3 +459,52 @@ def test_multiple_lags(self):
                             cm.volatility = process(p=p, o=o, q=q)
                             cm.fit(iter=0, disp='off')
 
+    def test_first_last_obs(self):
+        ar = ARX(self.y,lags=5, hold_back=100)
+        res = ar.fit(iter=0)
+        resids = res.resid
+        resid_copy = resids.copy()
+        resid_copy[:100] = np.nan
+        assert_equal(resids, resid_copy)
+
+        ar.volatility = GARCH()
+        res = ar.fit(iter=0)
+        resids = res.resid
+        resid_copy = resids.copy()
+        resid_copy[:100] = np.nan
+        assert_equal(resids, resid_copy)
+
+        ar = ARX(self.y,lags=5, last_obs=500)
+        ar.volatility = GARCH()
+        res = ar.fit(iter=0)
+        resids = res.resid
+        resid_copy = resids.copy()
+        resid_copy[500:] = np.nan
+        assert_equal(resids, resid_copy)
+
+        ar = ARX(self.y,lags=5, hold_back=100, last_obs=500)
+        ar.volatility = GARCH()
+        res = ar.fit(iter=0)
+        resids = res.resid
+        resid_copy = resids.copy()
+        resid_copy[:100] = np.nan
+        resid_copy[500:] = np.nan
+        assert_equal(resids, resid_copy)
+
+        vol = res.conditional_volatility
+        vol_copy = vol.copy()
+        vol_copy[:100] = np.nan
+        vol_copy[500:] = np.nan
+        assert_equal(vol, vol_copy)
+        assert_equal(self.y.shape[0], vol.shape[0])
+
+
+        ar = ARX(self.y,lags=5, last_obs=500)
+        ar.volatility = GARCH()
+        res = ar.fit(iter=0)
+        resids = res.resid
+        resid_copy = resids.copy()
+        resid_copy[:5] = np.nan
+        resid_copy[500:] = np.nan
+        assert_equal(resids, resid_copy)
+

arch/tests/test_utils.py

@@ -3,9 +3,10 @@
 
 from numpy.testing import assert_equal, assert_raises
 import numpy as np
-from pandas import Series, DataFrame
+from pandas import Series, DataFrame, date_range
 
-from arch.utils import ensure1d, parse_dataframe, DocStringInheritor
+from arch.utils import ensure1d, parse_dataframe, DocStringInheritor, \
+    date_to_index
 from arch.base import implicit_constant
 from arch.compat.python import add_metaclass, range
 
@@ -86,6 +87,16 @@ class B(A):
             """
         assert_equal(B.__doc__,ds)
 
+    def test_date_to_index(self):
+        dr = date_range('20000101',periods=3000, freq='W')
+        y = Series(np.arange(3000.0), index=dr)
+        date_index = y.index
+        import datetime as dt
+        #TODO: Finish Test
+        # date_to_index(dt.datetime(2001, 3, 3), date_index)
+
+
+
 
 
 

arch/utils.py

@@ -77,3 +77,57 @@ def __new__(meta, name, bases, clsdict):
                         attribute.__doc__=doc
                         break
         return type.__new__(meta, name, bases, clsdict)
+
+def date_to_index(date, date_index, method='first'):
+    """
+    Looks up a
+
+    Parameters
+    ----------
+    date : datetime or datetime64
+    date_index : 1-d array of datetime64
+    method : str, optional
+        Method to use.  Choices are 'first', 'last', 'exact'  First will return
+        the index where date_index >= date.  Last will return the index where
+        date_index <= date.  Exact will find the first instance where
+        date_index == date.
+    """
+    import datetime as dt
+    import pandas as pd
+    from pandas.core.common import is_datetime64_dtype
+
+    if not is_datetime64_dtype(date_index):
+        raise ValueError('date_index must be a datetime64 array')
+    if not isinstance(date, (dt.datetime, np.datetime64, str)):
+        raise ValueError("date must be a datetime, datetime64 or string")
+    elif isinstance(date, dt.datetime):
+        date = np.datetime64(date)
+    elif isinstance(date, str):
+        orig_date = date
+        date = pd.to_datetime(date, coerce=True)
+        if date == pd.NaT:
+            raise ValueError('date:' + orig_date +
+                             ' cannot be parsed to a date.')
+
+    date_index = np.asarray(date_index)
+
+    if method == 'first':
+        locs = np.nonzero(date_index >= date)[0]
+        if locs.shape[0] == 0:
+            raise ValueError('All dates in date_index occur before date')
+        else:
+            return locs.min()
+    elif method == 'last':
+        locs = np.nonzero(date_index <= date)[0]
+        if locs.shape[0] == 0:
+            raise ValueError('All dates in date_index occur after date')
+        else:
+            return locs.max()
+    elif method == 'exact':
+        locs = np.nonzero(date_index == date)[0]
+        if locs.shape[0] == 0:
+            raise ValueError('date not in date_index')
+        else:
+            return locs.max()
+    else:
+        raise ValueError('method is unknown')

doc/source/index.rst

@@ -15,13 +15,6 @@ as
 
 An complete ARCH model is divided into three components:
 
-.. toctree::
-    :maxdepth: 1
-
-    Examples <examples>
-    Mean Models <mean>
-    Volatility Processes <volatility>
-    Distributions <distribution>
 
 ..
 .. Theoretical Background <background>
@@ -45,10 +38,10 @@ blocks of an ARCH model
 
     from arch.mean import ConstantMean
     from arch.volatiltiy import GARCH
-    from arch.distribution impot StudentsT
+    from arch.distribution import Normal
     am = ConstantMean(returns)
     am.volatility = GARCH(1,0,1)
-    am.distribution = StudentsT()
+    am.distribution = Normal()
 
 In either case, model parameters are estimated using
 
@@ -57,6 +50,15 @@ In either case, model parameters are estimated using
     res = am.fit()
 
 
+.. toctree::
+    :maxdepth: 1
+
+    Examples <examples>
+    Mean Models <mean>
+    Volatility Processes <volatility>
+    Distributions <distribution>
+
+
 Model Constructor
 -----------------
 While models can be carefully specified using the individual components, most common specifications can be specified