Scribd
Upload
753 views1 page

Pair Trading

This document summarizes a statistical arbitrage strategy that trades pairs of stocks. It identifies candidate stock pairs based on fundamental preferences and principal component analysis of the stock covariance matrix. Specifically, it selects the stock pair of Home Depot and Lowe's based on their weight in the 6th eigenvector. It generates trading signals to buy or sell the pair based on their spread moving above or below 1 standard deviation from the mean. Backtesting shows the pair trading strategy achieved higher returns than a simple buy and hold strategy. It also uses Monte Carlo simulations to project expected returns of the strategy across different time horizons and finds a 1% chance of monthly losses exceeding 25 cents on a $1 initial portfolio.

Uploaded by

octaveyahoo
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
753 views1 page

Pair Trading

This document summarizes a statistical arbitrage strategy that trades pairs of stocks. It identifies candidate stock pairs based on fundamental preferences and principal component analysis of the stock covariance matrix. Specifically, it selects the stock pair of Home Depot and Lowe's based on their weight in the 6th eigenvector. It generates trading signals to buy or sell the pair based on their spread moving above or below 1 standard deviation from the mean. Backtesting shows the pair trading strategy achieved higher returns than a simple buy and hold strategy. It also uses Monte Carlo simulations to project expected returns of the strategy across different time horizons and finds a 1% chance of monthly losses exceeding 25 cents on a $1 initial portfolio.

Uploaded by

octaveyahoo
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Statistical Arbitrage: Trading Pairs of stocks – Nilesh N.

Shah 9/28/10

Setup: Candidate Stocks selected: GLD, GDX, Intel,


AMD, Home Depot, Lowes based on fundamental
preference. Data: Daily stock data from '08-'10. The
invariant for Stock Pair Trading Strategy was
identified: Increment in pair log spread values.

The multivariate Ornstein-Uhlenbeck process was


used to identify a Stock Pair to set up for a trade
based on the components of the Eigenvectors from
th
PCA analysis on the stock covariance matrix. The 6 Eigenvector
(which emphasized the pair Home Depot and Lowe’s) was selected.
In the interest of Diversification Management, stocks were
selected in proportion to Eigenvector weights rather than buying all
stock pairs in the same proportion. As verification, cointegration
test was run after calculating
historical Beta measurement
between the 2 stocks using OLS. Beta ratio HD: Lowes = 1: 1.18
(Eigenvector analysis ratio 1: 1.178). Fitting the historical data to the OU
process showed a Half Life of 10
days, seemed like a reasonable
timeframe to unwind the trade for a
small portfolio.

Trading signals (Pair Buy “+”/ Sell


“o”) were generated based on
simple indicators +/- 1 standard
deviation from the spread mean.
Compared to a naïve Buy and
Hold strategy return ($2) over the same period, the Pair trading
strategy returned $13. To measure the expected returns of the
portfolio (assuming all pairs were traded instead of just 1 as
selected above) at various horizon times (day, week and month),
the Gaussian Copula was utilized to perform 10k Monte Carlo
simulations. Procedure: take the covariance matrix of the historical log Pair Spreads (invariants) of all
stocks, find Inverse CDFs by sorting the data, generate multivariate normal random numbers using
Sample mean/Covariance, adjusted for time to horizon, measure the Grade, and compute the Inverse
CDF at each of the points of the grade. The invariant statistics were projected to the horizon (day,
week, and month). Total Portfolio returns were analyzed assuming equal weighted portfolio (even though
Minimum Variance Portfolio heavily weighted by 97 percent the 3rd Pair) out to the horizon. VaR: there is
1% chance of loss greater than 25c on a $1 initial portfolio on a monthly horizon.

Conclusion: Spread returns were identified as invariants while trading Stock Pairs. The trading pairs
selected based on PCA analysis and the Half Life indicators were reasonably accurate indicators while
unwinding trades. The strategy performed better than Naïve Buy/Hold, but could be better optimized for
VaR management by incorporating a Utility function. A Factors-on-demand approach could be applied to
make the strategy more dynamic and adaptive to changes with time and easily broaden stock universe.

You might also like