|
1
|
|
|
""" |
|
2
|
|
|
Reference implementation for EarningsCalendar loaders. |
|
3
|
|
|
""" |
|
4
|
|
|
from itertools import repeat |
|
5
|
|
|
|
|
6
|
|
|
from numpy import full_like, full |
|
7
|
|
|
import pandas as pd |
|
8
|
|
|
from six import iteritems |
|
9
|
|
|
from six.moves import zip |
|
10
|
|
|
from toolz import merge |
|
11
|
|
|
|
|
12
|
|
|
from .base import PipelineLoader |
|
13
|
|
|
from .frame import DataFrameLoader |
|
14
|
|
|
from ..data.earnings import EarningsCalendar |
|
15
|
|
|
from zipline.utils.numpy_utils import np_NaT |
|
16
|
|
|
from zipline.utils.memoize import lazyval |
|
17
|
|
|
|
|
18
|
|
|
|
|
19
|
|
|
class EarningsCalendarLoader(PipelineLoader): |
|
20
|
|
|
""" |
|
21
|
|
|
Reference loader for |
|
22
|
|
|
:class:`zipline.pipeline.data.earnings.EarningsCalendar`. |
|
23
|
|
|
|
|
24
|
|
|
Does not currently support adjustments to the dates of known earnings. |
|
25
|
|
|
|
|
26
|
|
|
Parameters |
|
27
|
|
|
---------- |
|
28
|
|
|
all_dates : pd.DatetimeIndex |
|
29
|
|
|
Index of dates for which we can serve queries. |
|
30
|
|
|
announcement_dates : dict[int -> pd.Series or pd.DatetimeIndex] |
|
31
|
|
|
Dict mapping sids to objects representing dates on which earnings |
|
32
|
|
|
occurred. |
|
33
|
|
|
|
|
34
|
|
|
If a dict value is a Series, it's interpreted as a mapping from the |
|
35
|
|
|
date on which we learned an announcement was coming to the date on |
|
36
|
|
|
which the announcement was made. |
|
37
|
|
|
|
|
38
|
|
|
If a dict value is a DatetimeIndex, it's interpreted as just containing |
|
39
|
|
|
the dates that announcements were made, and we assume we knew about the |
|
40
|
|
|
announcement on all prior dates. This mode is only supported if |
|
41
|
|
|
``infer_timestamp`` is explicitly passed as a truthy value. |
|
42
|
|
|
|
|
43
|
|
|
infer_timestamps : bool, optional |
|
44
|
|
|
Whether to allow passing ``DatetimeIndex`` values in |
|
45
|
|
|
``announcement_dates``. |
|
46
|
|
|
""" |
|
47
|
|
|
def __init__(self, |
|
48
|
|
|
all_dates, |
|
49
|
|
|
announcement_dates, |
|
50
|
|
|
infer_timestamps=False, |
|
51
|
|
|
dataset=EarningsCalendar): |
|
52
|
|
|
self.all_dates = all_dates |
|
53
|
|
|
self.announcement_dates = announcement_dates = ( |
|
54
|
|
|
announcement_dates.copy() |
|
55
|
|
|
) |
|
56
|
|
|
dates = self.all_dates.values |
|
57
|
|
|
for k, v in iteritems(announcement_dates): |
|
58
|
|
|
if isinstance(v, pd.DatetimeIndex): |
|
59
|
|
|
if not infer_timestamps: |
|
60
|
|
|
raise ValueError( |
|
61
|
|
|
"Got DatetimeIndex of announcement dates for sid %d.\n" |
|
62
|
|
|
"Pass `infer_timestamps=True` to use the first date in" |
|
63
|
|
|
" `all_dates` as implicit timestamp." |
|
64
|
|
|
) |
|
65
|
|
|
# If we are passed a DatetimeIndex, we always have |
|
66
|
|
|
# knowledge of the announcements. |
|
67
|
|
|
announcement_dates[k] = pd.Series( |
|
68
|
|
|
v, index=repeat(dates[0], len(v)), |
|
69
|
|
|
) |
|
70
|
|
|
self.dataset = dataset |
|
71
|
|
|
|
|
72
|
|
|
def get_loader(self, column): |
|
73
|
|
|
"""Dispatch to the loader for ``column``. |
|
74
|
|
|
""" |
|
75
|
|
|
if column is self.dataset.next_announcement: |
|
76
|
|
|
return self.next_announcement_loader |
|
77
|
|
|
elif column is self.dataset.previous_announcement: |
|
78
|
|
|
return self.previous_announcement_loader |
|
79
|
|
|
else: |
|
80
|
|
|
raise ValueError("Don't know how to load column '%s'." % column) |
|
81
|
|
|
|
|
82
|
|
|
@lazyval |
|
83
|
|
|
def next_announcement_loader(self): |
|
84
|
|
|
return DataFrameLoader( |
|
85
|
|
|
self.dataset.next_announcement, |
|
86
|
|
|
next_earnings_date_frame( |
|
87
|
|
|
self.all_dates, |
|
88
|
|
|
self.announcement_dates, |
|
89
|
|
|
), |
|
90
|
|
|
adjustments=None, |
|
91
|
|
|
) |
|
92
|
|
|
|
|
93
|
|
|
@lazyval |
|
94
|
|
|
def previous_announcement_loader(self): |
|
95
|
|
|
return DataFrameLoader( |
|
96
|
|
|
self.dataset.previous_announcement, |
|
97
|
|
|
previous_earnings_date_frame( |
|
98
|
|
|
self.all_dates, |
|
99
|
|
|
self.announcement_dates, |
|
100
|
|
|
), |
|
101
|
|
|
adjustments=None, |
|
102
|
|
|
) |
|
103
|
|
|
|
|
104
|
|
|
def load_adjusted_array(self, columns, dates, assets, mask): |
|
105
|
|
|
return merge( |
|
106
|
|
|
self.get_loader(column).load_adjusted_array( |
|
107
|
|
|
[column], dates, assets, mask |
|
108
|
|
|
) |
|
109
|
|
|
for column in columns |
|
110
|
|
|
) |
|
111
|
|
|
|
|
112
|
|
|
|
|
113
|
|
|
def next_earnings_date_frame(dates, announcement_dates): |
|
114
|
|
|
""" |
|
115
|
|
|
Make a DataFrame representing simulated next earnings dates. |
|
116
|
|
|
|
|
117
|
|
|
Parameters |
|
118
|
|
|
---------- |
|
119
|
|
|
dates : pd.DatetimeIndex. |
|
120
|
|
|
The index of the returned DataFrame. |
|
121
|
|
|
announcement_dates : dict[int -> pd.Series] |
|
122
|
|
|
Dict mapping sids to an index of dates on which earnings were announced |
|
123
|
|
|
for that sid. |
|
124
|
|
|
|
|
125
|
|
|
Returns |
|
126
|
|
|
------- |
|
127
|
|
|
next_earnings: pd.DataFrame |
|
128
|
|
|
A DataFrame representing, for each (label, date) pair, the first entry |
|
129
|
|
|
in `earnings_calendars[label]` on or after `date`. Entries falling |
|
130
|
|
|
after the last date in a calendar will have `np_NaT` as the result in |
|
131
|
|
|
the output. |
|
132
|
|
|
|
|
133
|
|
|
See Also |
|
134
|
|
|
-------- |
|
135
|
|
|
previous_earnings_date_frame |
|
136
|
|
|
""" |
|
137
|
|
|
cols = {equity: full_like(dates, np_NaT) for equity in announcement_dates} |
|
138
|
|
|
raw_dates = dates.values |
|
139
|
|
|
for equity, earnings_dates in iteritems(announcement_dates): |
|
140
|
|
|
data = cols[equity] |
|
141
|
|
|
if not earnings_dates.index.is_monotonic_increasing: |
|
142
|
|
|
earnings_dates = earnings_dates.sort_index() |
|
143
|
|
|
|
|
144
|
|
|
# Iterate over the raw Series values, since we're comparing against |
|
145
|
|
|
# numpy arrays anyway. |
|
146
|
|
|
iterkv = zip(earnings_dates.index.values, earnings_dates.values) |
|
147
|
|
|
for timestamp, announce_date in iterkv: |
|
148
|
|
|
date_mask = (timestamp <= raw_dates) & (raw_dates <= announce_date) |
|
149
|
|
|
value_mask = (announce_date <= data) | (data == np_NaT) |
|
150
|
|
|
data[date_mask & value_mask] = announce_date |
|
151
|
|
|
|
|
152
|
|
|
return pd.DataFrame(index=dates, data=cols) |
|
153
|
|
|
|
|
154
|
|
|
|
|
155
|
|
|
def previous_earnings_date_frame(dates, announcement_dates): |
|
156
|
|
|
""" |
|
157
|
|
|
Make a DataFrame representing simulated next earnings dates. |
|
158
|
|
|
|
|
159
|
|
|
Parameters |
|
160
|
|
|
---------- |
|
161
|
|
|
dates : DatetimeIndex. |
|
162
|
|
|
The index of the returned DataFrame. |
|
163
|
|
|
announcement_dates : dict[int -> DatetimeIndex] |
|
164
|
|
|
Dict mapping sids to an index of dates on which earnings were announced |
|
165
|
|
|
for that sid. |
|
166
|
|
|
|
|
167
|
|
|
Returns |
|
168
|
|
|
------- |
|
169
|
|
|
prev_earnings: pd.DataFrame |
|
170
|
|
|
A DataFrame representing, for (label, date) pair, the first entry in |
|
171
|
|
|
`announcement_dates[label]` strictly before `date`. Entries falling |
|
172
|
|
|
before the first date in a calendar will have `NaT` as the result in |
|
173
|
|
|
the output. |
|
174
|
|
|
|
|
175
|
|
|
See Also |
|
176
|
|
|
-------- |
|
177
|
|
|
next_earnings_date_frame |
|
178
|
|
|
""" |
|
179
|
|
|
sids = list(announcement_dates) |
|
180
|
|
|
out = full((len(dates), len(sids)), np_NaT, dtype='datetime64[ns]') |
|
181
|
|
|
dn = dates[-1].asm8 |
|
182
|
|
|
for col_idx, sid in enumerate(sids): |
|
183
|
|
|
# announcement_dates[sid] is Series mapping knowledge_date to actual |
|
184
|
|
|
# announcement date. We don't care about the knowledge date for |
|
185
|
|
|
# computing previous earnings. |
|
186
|
|
|
values = announcement_dates[sid].values |
|
187
|
|
|
values = values[values <= dn] |
|
188
|
|
|
out[dates.searchsorted(values), col_idx] = values |
|
189
|
|
|
|
|
190
|
|
|
frame = pd.DataFrame(out, index=dates, columns=sids) |
|
191
|
|
|
frame.ffill(inplace=True) |
|
192
|
|
|
return frame |
|
193
|
|
|
|
quantopian /
zipline
