Equity · United States · ETFs · Daily · United States Free

Synthetic Lending Rates Predict Subsequent Market Return

Academic paper

AuthorsMatus Padysak

Institute

SKComenius University Bratislava
?Comenius University - Faculty of Mathematics, Physics and Informatics
?Quantpedia.com

Paper linkhttps://papers.ssrn.com/sol3/papers.cfm?abstract_id=3976307

Strategy in a nutshell

Universe: SPY ETF. Use CBOE borrow intensity data (45-day maturity) to measure synthetic shorting costs. Daily, calculate the change in aggregate borrow intensity:

Positive change → buy SPY at 15:59

Negative change → short SPY at 15:59

Positions are closed the next day at 15:58.

Economic rationale

Changes in borrow intensity reflect shorting demand and market sentiment:

Increase in intensity → lower shorting fees → positive subsequent returns

Decrease in intensity → higher shorting fees → negative subsequent returns

The signal is short-lived, with partial reversal over two days. Correlations with VIX suggest borrow intensity also proxies fear/sentiment, and intraday negative correlation indicates a short-term reversal effect linked to liquidity provision.

Backtest performance

Annualised return15.47%

Volatility19.52%

Beta0.018

Sharpe ratio0.79

Maximum drawdown-28.51%

Win rate50%

Full Python code

1from AlgorithmImports import *
2#endregion
3
4class SyntheticLendingRatesPredictSubsequentMarketReturn(QCAlgorithm):
5
6def Initialize(self):
7self.SetStartDate(2016, 1, 1)
8self.SetCash(100000)
9
10self.spy_symbol:Symbol = self.AddEquity('SPY', Resolution.Minute).Symbol  
11
12self.lending_data_symbol:Symbol = self.AddData(
13    QuantpediaLendingRates,
14    'lending_rate', 
15    Resolution.Minute).Symbol
16
17self.last_lending_mean = None          
18
19def OnData(self, data: Slice):
20curr_time:datetime.datetime = self.Time
21
22# liquidate at 15:58
23if curr_time.hour == 15 and curr_time.minute == 58:
24    self.Liquidate(self.spy_symbol) 
25
26# lending rate data came in
27if self.lending_data_symbol in data and data[self.lending_data_symbol]:
28    curr_lending_mean:float = data[self.lending_data_symbol].Value
29
30    if self.last_lending_mean:
31        # calculate daily change in lending rate
32        diff:float = curr_lending_mean - self.last_lending_mean
33
34        if diff > 0:
35            self.SetHoldings(self.spy_symbol, 1)
36        else:
37            self.SetHoldings(self.spy_symbol, -1)
38
39    self.last_lending_mean = curr_lending_mean    
40
41# Quantpedia data.
42# NOTE: IMPORTANT: Data order must be ascending (datewise)
43class QuantpediaLendingRates(PythonData):
44def GetSource(self, config, date, isLiveMode):
45return SubscriptionDataSource("data.quantpedia.com/backtesting_data/options/lending_rates_day_close_matur_45_days.csv".format(config.Symbol.Value), SubscriptionTransportMedium.RemoteFile, FileFormat.Csv)
46
47def Reader(self, config, line, date, isLiveMode):
48data:QuantpediaLendingRates = QuantpediaLendingRates()
49data.Symbol = config.Symbol
50
51if not line[0].isdigit(): return None
52
53split:list = line.split(';')
54
55datetime_str:str = split[0] + ', 15:59'
56
57data.Time = datetime.strptime(datetime_str, "%Y-%m-%d, %H:%M")
58valid_values:list = list(filter(lambda value: value != '', split[1:]))
59valid_values:list = list(map(lambda str_value: float(str_value), valid_values))
60data.Value = np.mean(valid_values)
61
62return data