-
Notifications
You must be signed in to change notification settings - Fork 653
/
synthetic-lending-rates-predict-subsequent-market-return.py
72 lines (54 loc) · 3.22 KB
/
synthetic-lending-rates-predict-subsequent-market-return.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
# https://quantpedia.com/strategies/synthetic-lending-rates-predict-subsequent-market-return/
#
# The investment universe consists of SPY ETF. Synthetic shorting costs data are obtained from Borrow Intensity Indicators by the CBOE (and includes 4877 stocks/ETFs).
# The paper utilizes the constant maturities of 45 days. Intraday SPY data are obtained from FirstRate Data. The aggregate (mean) borrow intensity is calculated as equally weighted borrow intensity of each stock/ETF in the sample at day t.
# The shorting costs data are estimated at a timestamp of 15:57. Calculate the change in the aggregate intensity at day t as the difference of aggregate borrowing intensity at day t and t-1.
# Buy the SPY ETF at 15:59 if the difference is positive and short the SPY if the difference is negative. The positions are held for one day and are closed at 15:58 at next day.
#
# QC Implementation changes:
# - Signal calculation and trade opening is done each day at 15:59.
#region imports
from AlgorithmImports import *
#endregion
class SyntheticLendingRatesPredictSubsequentMarketReturn(QCAlgorithm):
def Initialize(self):
self.SetStartDate(2016, 1, 1)
self.SetCash(100000)
self.spy_symbol:Symbol = self.AddEquity('SPY', Resolution.Minute).Symbol
self.lending_data_symbol:Symbol = self.AddData(
QuantpediaLendingRates,
'lending_rate',
Resolution.Minute).Symbol
self.last_lending_mean = None
def OnData(self, data: Slice):
curr_time:datetime.datetime = self.Time
# liquidate on 15:58
if curr_time.hour == 15 and curr_time.minute == 58:
self.Liquidate(self.spy_symbol)
# lending rate data came in
if self.lending_data_symbol in data and data[self.lending_data_symbol]:
curr_lending_mean:float = data[self.lending_data_symbol].Value
if self.last_lending_mean:
# calculate daily change in lending rate
diff:float = curr_lending_mean - self.last_lending_mean
if diff > 0:
self.SetHoldings(self.spy_symbol, 1)
else:
self.SetHoldings(self.spy_symbol, -1)
self.last_lending_mean = curr_lending_mean
# Quantpedia data.
# NOTE: IMPORTANT: Data order must be ascending (datewise)
class QuantpediaLendingRates(PythonData):
def GetSource(self, config, date, isLiveMode):
return SubscriptionDataSource("data.quantpedia.com/backtesting_data/options/lending_rates_day_close_matur_45_days.csv".format(config.Symbol.Value), SubscriptionTransportMedium.RemoteFile, FileFormat.Csv)
def Reader(self, config, line, date, isLiveMode):
data:QuantpediaLendingRates = QuantpediaLendingRates()
data.Symbol = config.Symbol
if not line[0].isdigit(): return None
split:list = line.split(';')
datetime_str:str = split[0] + ', 15:59'
data.Time = datetime.strptime(datetime_str, "%Y-%m-%d, %H:%M")
valid_values:list = list(filter(lambda value: value != '', split[1:]))
valid_values:list = list(map(lambda str_value: float(str_value), valid_values))
data.Value = np.mean(valid_values)
return data