by Howard Simons, NQLX's Special Academic Advisor.
Covariance and the related measure of correlation are nothing more than
the degree to which two variables, such as two stock prices, move together.
Any stock has a correlation of 1.00 with itself, a correlation of -1.00
with its short position, and a correlation of 0.00 against something totally
random.
The concepts are absolutely critical to both modern portfolio theory
and to the emerging world of single stock futures (SSFs). For portfolio
theory, the covariance is important because of its effect on total risk.
Let's say you own both UAL and AMR. These two airlines are affected by
the same macroeconomic and industry-specific factors, and we should expect
them to have a high covariance. The formula for their combined variance
would be:
Variance (UAL+AMR) = variance (UAL) + variance (AMR) +
2*covariance (UAL+AMR)
Over the past four years, the daily variance of returns for UAL and AMR
has been .114% and .111%, respectively, and their covariance of returns
has been .086% per day. Add these two together in their initial position
sizes, and the daily portfolio variance jumps to a whopping .398%. If
any of this sounds familiar to those of you who diversified by owning
every tech stock available back in 1999, it should.
However, if we go long one and short the other, we now subtract the covariance
term:
Variance (UAL-AMR) = variance (UAL) + variance (AMR) -
2*covariance (UAL+AMR)
The variance of this matched pair trade falls to .052% per day, less
than half of each individual stock's variance. True, you won't make anywhere
near as much as you would if you were right on both, but you won't lose
as much if you're wrong, either.
Let's take the following group of nine pairs of stocks:
- General Motors / Ford
- Qualcomm / Nokia
- Tyco / General Electric
- Pfizer / Merck
- Citigroup / JP Morgan Chase
- Home Depot / Wal-Mart
- Coca-Cola / Pepsico
- Procter & Gamble / Johnson & Johnson
- ChevronTexaco / ExxonMobil
Now let's construct a correlation matrix of each stock's returns, not
only against those of their designated partner, but against the remaining
sixteen stocks as well. The table below can be read like a mileage guide
in a roadmap. Each stock has a 1.00 correlation with itself. The maximum
correlation in each column is highlighted; in all cases, the maximum correlation
is, unsurprisingly, with the designated partner.
Now let's take the pair with the highest correlation, Citigroup / JP
Morgan Chase. We could trade this spread by buying the SSF of one and
selling the SSF of the other. However, a simple one-to-one spread ignores
the different statistical characteristics of these two stocks. A regression
of Citigroup as a function of JP Morgan Chase over the past four years
yields the following equation:
Citigroup = 25.57 + .30 * JP Morgan Chase
The highlighted coefficient, the .30 number, might be interpreted as
saying Citigroup is roughly one-third as volatile in price as JP Morgan
Chase, and therefore we may decide to trade three Citigroup futures for
each JP Morgan Chase future. The two spread trades, one adjusted for this
hedge ratio and the other left as one-for-one, are depicted below.
Which Should You Use?
SSFs can be used both for speculation and for hedging. If you think Citigroup
is going to outperform JP Morgan Chase, the trade that reflects your judgment
is to buy Citigroup futures and sell JP Morgan Chase futures. If you already
own JP Morgan Chase and want to hedge your holdings, you might want to
calculate a hedge ratio for the number of Citigroup futures to sell.
Either way, you'll be able to take advantage of SSFs and their ability
to facilitate the short side of any stock transaction.
|