Original paper
Abstract
We propose an investment-momentum strategy of buying past winners with low investment and selling past losers with high investment, which simultaneously exploits two dimensions of market inefficiencies. The new strategy generates twice the monthly returns earned by either the price momentum or investment strategy (1.44% vs 0.75% or 0.61%). Despite the diminishing anomalies in recent decades, the investment-momentum stays persistent. The mispricing-based strategy performs better in periods of high investor sentiment or for stocks with high limits-to-arbitrage, which is consistent with our expectations. Overall, we show that one can simultaneously use multiple dimensions of market inefficiency to attain superior performance.
Keywords:Â Investment momentum, price momentum, corporate investment, mispricing, behavioural finance
Trading rules
- Investment universe: Focus on common stocks from NYSE and AMEX with share codes 10 and 11.
- Filter out stocks that have a starting price below $5 during the initial holding duration.
- Categorize stocks into three segments based on bid-ask spreads; prioritize the segment with the smallest spread for better liquidity.
- Combine past returns momentum and I/A ratios to construct the InvMom strategy
- Momentum: based on past six-month returns (from month -5 to month 0)
- I/A ratio: annual change in gross property, plant, and equipment + annual change in inventories / lagged book value of assets
- Organize stocks into five tiers based on both momentum and I/A ratios.
- Through double-sorting, create 5x5 portfolio mixes based on momentum and investment levels.
- Take long position in winner portfolio with low investment; short position in loser portfolio with high investment.
- Hold these positions for the next six months (stretching from month 2 to month 7), skipping the most recent month.
- Adopt an equal-weight approach, adjusting every half-year.
Python code
Backtrader
import backtrader as bt
import pandas as pd
import numpy as np
class InvMom(bt.Strategy):
params = (
('lookback', 5),
('hold_period', 6),
)
def __init__(self):
self.inv_mom_data = dict()
def next(self):
if len(self.datas) < self.p.lookback + self.p.hold_period + 1:
return
if self.datetime.date(0).month != self.datetime.date(-1).month:
self.calculate_inv_mom()
for data, (position, weight) in self.inv_mom_data.items():
if position == 'long':
self.order_target_percent(data, target=weight)
elif position == 'short':
self.order_target_percent(data, target=-weight)
def calculate_inv_mom(self):
stocks_data = []
for data in self.datas:
if data.close[0] > 5 and data.share_code in (10, 11):
bid_ask_spread = data.bid[0] - data.ask[0]
momentum = data.close[0] / data.close[-self.p.lookback] - 1
ia_ratio = (data.gross_property_change[0] + data.inventory_change[0]) / data.book_value_assets[0]
stocks_data.append({
'data': data,
'bid_ask_spread': bid_ask_spread,
'momentum': momentum,
'ia_ratio': ia_ratio,
})
sorted_stocks = sorted(stocks_data, key=lambda x: x['bid_ask_spread'])
lowest_spread_group = sorted_stocks[:len(sorted_stocks) // 3]
momentum_quintiles = pd.qcut([stock['momentum'] for stock in lowest_spread_group], 5, labels=False)
ia_quintiles = pd.qcut([stock['ia_ratio'] for stock in lowest_spread_group], 5, labels=False)
long_portfolio = []
short_portfolio = []
for i, stock in enumerate(lowest_spread_group):
if momentum_quintiles[i] == 4 and ia_quintiles[i] == 0:
long_portfolio.append(stock['data'])
elif momentum_quintiles[i] == 0 and ia_quintiles[i] == 4:
short_portfolio.append(stock['data'])
long_weight = 1 / len(long_portfolio)
short_weight = 1 / len(short_portfolio)
for stock_data in lowest_spread_group:
data = stock_data['data']
if data in long_portfolio:
self.inv_mom_data[data] = ('long', long_weight)
elif data in short_portfolio:
self.inv_mom_data[data] = ('short', short_weight)
else:
self.inv_mom_data[data] = (None, 0)
if __name__ == '__main__':
cerebro = bt.Cerebro()
# Add stocks datafeeds to cerebro here
cerebro.addstrategy(InvMom)
cerebro.run()