|
1
|
|
|
# |
|
2
|
|
|
# Copyright 2015 Quantopian, Inc. |
|
3
|
|
|
# |
|
4
|
|
|
# Licensed under the Apache License, Version 2.0 (the "License"); |
|
5
|
|
|
# you may not use this file except in compliance with the License. |
|
6
|
|
|
# You may obtain a copy of the License at |
|
7
|
|
|
# |
|
8
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0 |
|
9
|
|
|
# |
|
10
|
|
|
# Unless required by applicable law or agreed to in writing, software |
|
11
|
|
|
# distributed under the License is distributed on an "AS IS" BASIS, |
|
12
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
|
13
|
|
|
# See the License for the specific language governing permissions and |
|
14
|
|
|
# limitations under the License. |
|
15
|
|
|
|
|
16
|
|
|
from zipline.errors import ( |
|
17
|
|
|
InvalidBenchmarkAsset, |
|
18
|
|
|
BenchmarkAssetNotAvailableTooEarly, |
|
19
|
|
|
BenchmarkAssetNotAvailableTooLate |
|
20
|
|
|
) |
|
21
|
|
|
|
|
22
|
|
|
|
|
23
|
|
|
class BenchmarkSource(object): |
|
24
|
|
|
def __init__(self, benchmark_sid, env, trading_days, data_portal, |
|
25
|
|
|
emission_rate="daily"): |
|
26
|
|
|
self.benchmark_sid = benchmark_sid |
|
27
|
|
|
self.env = env |
|
28
|
|
|
self.trading_days = trading_days |
|
29
|
|
|
self.emission_rate = emission_rate |
|
30
|
|
|
self.data_portal = data_portal |
|
31
|
|
|
|
|
32
|
|
|
if self.benchmark_sid: |
|
33
|
|
|
self.benchmark_asset = self.env.asset_finder.retrieve_asset( |
|
34
|
|
|
self.benchmark_sid) |
|
35
|
|
|
|
|
36
|
|
|
self._validate_benchmark() |
|
37
|
|
|
|
|
38
|
|
|
self.precalculated_series = \ |
|
39
|
|
|
self._initialize_precalculated_series( |
|
40
|
|
|
self.benchmark_sid, |
|
41
|
|
|
self.env, |
|
42
|
|
|
self.trading_days, |
|
43
|
|
|
self.data_portal |
|
44
|
|
|
) |
|
45
|
|
|
|
|
46
|
|
|
def get_value(self, dt): |
|
47
|
|
|
return self.precalculated_series.loc[dt] |
|
48
|
|
|
|
|
49
|
|
|
def _validate_benchmark(self): |
|
50
|
|
|
# check if this security has a stock dividend. if so, raise an |
|
51
|
|
|
# error suggesting that the user pick a different asset to use |
|
52
|
|
|
# as benchmark. |
|
53
|
|
|
stock_dividends = \ |
|
54
|
|
|
self.data_portal.get_stock_dividends(self.benchmark_sid, |
|
55
|
|
|
self.trading_days) |
|
56
|
|
|
|
|
57
|
|
|
if len(stock_dividends) > 0: |
|
58
|
|
|
raise InvalidBenchmarkAsset( |
|
59
|
|
|
sid=str(self.benchmark_sid), |
|
60
|
|
|
dt=stock_dividends[0]["ex_date"] |
|
61
|
|
|
) |
|
62
|
|
|
|
|
63
|
|
|
if self.benchmark_asset.start_date > self.trading_days[0]: |
|
64
|
|
|
# the asset started trading after the first simulation day |
|
65
|
|
|
raise BenchmarkAssetNotAvailableTooEarly( |
|
66
|
|
|
sid=str(self.benchmark_sid), |
|
67
|
|
|
dt=self.trading_days[0], |
|
68
|
|
|
start_dt=self.benchmark_asset.start_date |
|
69
|
|
|
) |
|
70
|
|
|
|
|
71
|
|
|
if self.benchmark_asset.end_date < self.trading_days[-1]: |
|
72
|
|
|
# the asset stopped trading before the last simulation day |
|
73
|
|
|
raise BenchmarkAssetNotAvailableTooLate( |
|
74
|
|
|
sid=str(self.benchmark_sid), |
|
75
|
|
|
dt=self.trading_days[0], |
|
76
|
|
|
end_dt=self.benchmark_asset.end_date |
|
77
|
|
|
) |
|
78
|
|
|
|
|
79
|
|
|
def _initialize_precalculated_series(self, sid, env, trading_days, |
|
80
|
|
|
data_portal): |
|
81
|
|
|
""" |
|
82
|
|
|
Internal method that precalculates the benchmark return series for |
|
83
|
|
|
use in the simulation. |
|
84
|
|
|
|
|
85
|
|
|
Parameters |
|
86
|
|
|
---------- |
|
87
|
|
|
sid: (int) Asset to use |
|
88
|
|
|
|
|
89
|
|
|
env: TradingEnvironment |
|
90
|
|
|
|
|
91
|
|
|
trading_days: pd.DateTimeIndex |
|
92
|
|
|
|
|
93
|
|
|
data_portal: DataPortal |
|
94
|
|
|
|
|
95
|
|
|
Notes |
|
96
|
|
|
----- |
|
97
|
|
|
If the benchmark asset started trading after the simulation start, |
|
98
|
|
|
or finished trading before the simulation end, exceptions are raised. |
|
99
|
|
|
|
|
100
|
|
|
If the benchmark asset started trading the same day as the simulation |
|
101
|
|
|
start, the first available minute price on that day is used instead |
|
102
|
|
|
of the previous close. |
|
103
|
|
|
|
|
104
|
|
|
We use history to get an adjusted price history for each day's close, |
|
105
|
|
|
as of the look-back date (the last day of the simulation). Prices are |
|
106
|
|
|
fully adjusted for dividends, splits, and mergers. |
|
107
|
|
|
|
|
108
|
|
|
Returns |
|
109
|
|
|
------- |
|
110
|
|
|
A pd.Series, indexed by trading day, whose values represent the % |
|
111
|
|
|
change from close to close. |
|
112
|
|
|
""" |
|
113
|
|
|
if sid is None: |
|
114
|
|
|
# get benchmark info from trading environment, which defaults to |
|
115
|
|
|
# downloading data from Yahoo. |
|
116
|
|
|
daily_series = \ |
|
117
|
|
|
env.benchmark_returns[trading_days[0]:trading_days[-1]] |
|
118
|
|
|
|
|
119
|
|
|
if self.emission_rate == "minute": |
|
120
|
|
|
# we need to take the env's benchmark returns, which are daily, |
|
121
|
|
|
# and resample them to minute |
|
122
|
|
|
minutes = env.minutes_for_days_in_range( |
|
123
|
|
|
start=trading_days[0], |
|
124
|
|
|
end=trading_days[-1] |
|
125
|
|
|
) |
|
126
|
|
|
|
|
127
|
|
|
minute_series = daily_series.reindex( |
|
128
|
|
|
index=minutes, |
|
129
|
|
|
method="ffill" |
|
130
|
|
|
) |
|
131
|
|
|
|
|
132
|
|
|
return minute_series |
|
133
|
|
|
else: |
|
134
|
|
|
return daily_series |
|
135
|
|
|
elif self.emission_rate == "minute": |
|
136
|
|
|
minutes = env.minutes_for_days_in_range(self.trading_days[0], |
|
137
|
|
|
self.trading_days[-1]) |
|
138
|
|
|
benchmark_series = data_portal.get_history_window( |
|
139
|
|
|
[sid], |
|
140
|
|
|
minutes[-1], |
|
141
|
|
|
bar_count=len(minutes) + 1, |
|
142
|
|
|
frequency="1m", |
|
143
|
|
|
field="price", |
|
144
|
|
|
ffill=True |
|
145
|
|
|
) |
|
146
|
|
|
|
|
147
|
|
|
return benchmark_series.pct_change()[1:] |
|
148
|
|
|
else: |
|
149
|
|
|
start_date = env.asset_finder.retrieve_asset(sid).start_date |
|
150
|
|
|
if start_date < trading_days[0]: |
|
151
|
|
|
# get the window of close prices for benchmark_sid from the |
|
152
|
|
|
# last trading day of the simulation, going up to one day |
|
153
|
|
|
# before the simulation start day (so that we can get the % |
|
154
|
|
|
# change on day 1) |
|
155
|
|
|
benchmark_series = data_portal.get_history_window( |
|
156
|
|
|
[sid], |
|
157
|
|
|
trading_days[-1], |
|
158
|
|
|
bar_count=len(trading_days) + 1, |
|
159
|
|
|
frequency="1d", |
|
160
|
|
|
field="price", |
|
161
|
|
|
ffill=True |
|
162
|
|
|
)[sid] |
|
163
|
|
|
return benchmark_series.pct_change()[1:] |
|
164
|
|
|
elif start_date == trading_days[0]: |
|
165
|
|
|
# Attempt to handle case where stock data starts on first |
|
166
|
|
|
# day, in this case use the open to close return. |
|
167
|
|
|
benchmark_series = data_portal.get_history_window( |
|
168
|
|
|
[sid], |
|
169
|
|
|
trading_days[-1], |
|
170
|
|
|
bar_count=len(trading_days), |
|
171
|
|
|
frequency="1d", |
|
172
|
|
|
field="price", |
|
173
|
|
|
ffill=True |
|
174
|
|
|
)[sid] |
|
175
|
|
|
|
|
176
|
|
|
# get a minute history window of the first day |
|
177
|
|
|
first_open = data_portal.get_spot_value( |
|
178
|
|
|
sid, 'open', trading_days[0], 'daily') |
|
179
|
|
|
first_close = data_portal.get_spot_value( |
|
180
|
|
|
sid, 'close', trading_days[0], 'daily') |
|
181
|
|
|
|
|
182
|
|
|
first_day_return = (first_close - first_open) / first_open |
|
183
|
|
|
|
|
184
|
|
|
returns = benchmark_series.pct_change()[:] |
|
185
|
|
|
returns[0] = first_day_return |
|
186
|
|
|
return returns |
|
187
|
|
|
|
quantopian /
zipline
