PROFESSOR: Having discussed the properties of individual asset

returns, we will now talk about the joint distribution

of returns.
How can we describe interdependence
of different financial assets?
The core concept here is correlation-- correlation
of returns.
It is a measure of dependence.
It's a linear measure.
It's not the only possible way of describing how returns
are related to each other.
But it's a leading statistic that's very commonly used.
We start with the covariance of returns.
A covariance is defined mathematically
as the expected value of the product of the differences
between returns on asset i and its mean and asset
j and its mean.
The correlation is just the covariance
of returns, normalized by the product of standard deviations.
What this statistic captures is to what degree two assets
co-move with each other.
When correlation is high, close to 1,
both assets tend to go up and down in unison.
When correlation is close to minus 1,
they tend to move in the opposite direction
from each other.
When correlation is zero, there is no linear relation
between the two assets at all.
A closely related concept is the beta of returns.
The beta is the ratio of the covariance between returns
on asset i and asset j and the variance of returns on asset j.
The concept of beta is very important.
We are going to come back to it.
But intuitively, what the beta represents
is the degree to which returns on asset j
affect statistically returns on asset i.
When beta is equal to 1, for each 1% movement in asset j,
we would see on average 1% movement in asset i.
When we plot returns on one asset
versus the other as a scatter plot,
beta is given by the slope of the regression line.
The following graph provides a visual illustration
of how joint distributions may look with positive, negative,
or zero correlation.
What you see in the top left corner
is a scatter plot of the joint distribution
of returns on two assets with zero correlation.
It looks like a ball of points.
There is no obvious linear relation between the two.
On the right, we see two examples
of imperfect correlation, a positive and negative.
At the top, correlation is positive.
It's equal to 0.5.
We see, visually, a clear positive relation between two
assets, which is not perfect.
There is a lot of deviation from the 45 degree line.
But nonetheless, it's clear the two assets, on average,
move together in a positive direction.
Below that is a graph corresponding
to the correlation of minus 0.5.
Now, we see that the typical relation between the two assets
is negative.
When one of them goes up, the other one tends to decline.
At the bottom on the left, we see an example
with a high degree of positive correlation.
Now it is 0.8.
We see that the points now fall very close to the 45 degree
line, indicating that most of the movement
is now highly related between the two assets.
There is very little individual variation.
Next, let's look at a couple of empirical examples
showing us the degree of correlation
between different kinds of financial assets.
On the left, we have a scatter plot
illustrating the joint distribution
of returns on two stocks.
One is Intel.
The other one is IBM.
Both of these are large technology stocks.
What we see from this graph is a clear sense
of positive relation between the two.
These two returns are positively correlated.
This is not surprising and, in fact, fairly typical.
Stocks that are economically related to each other
will tend to have positive correlation.
On the right, we see a scatter plot
of two exchange traded funds.
We encountered them earlier--
the bond fund, AGG, and the stock fund, SPY.
It's a different picture.
We now see that the distribution of points on the scatter plot
does not show an obvious linear relation.
The scatter plot looks like a random cloud of points.
And the regression line is relatively flat.
In this case, correlation between the two assets
is very close to zero.
Stocks and bonds have a low degree of correlation.
This is an important empirical fact
and forms the basis of long-term investment strategies
diversifying risk between stocks and bonds.
We will come back to the concept of diversification
later in this class.
Let's take a look at a broader pattern
of correlations among different types of financial assets.
In the table, we'll look at pair-wise correlations
between stocks, bonds, treasury bills, and inflation.
We see that closely related assets
tend to have high levels of correlation.
For instance, when we look at large cap stocks and small cap
stocks, we observe correlations around 80%.
Different types of equities tend to co-move with each other.
Corporate bonds and Treasury bonds
also show a high degree of co-movement, over 80%.
Stocks versus long-term Treasuries,
as we have seen before, don't have a lot of co-movement.
And their correlation with each other is close to zero.
Treasury bills tend to have positive correlation
with inflation.
It's not terribly high, but it's in the neighborhood of 40%.
The reason for that is that the nominal return
on Treasury bills has an inflation component.
And that induces a positive correlation between the two.
So far, we looked at the pair-wise correlations
between financial assets.
We think of these as cross-sectional correlations,
how returns are related to each other at a point in time.
Another important dimension is time-series correlation
or order correlation.
This concept applies to a single asset over time.
We are now looking at how returns
during different periods are related to each other.
The empirical fact in financial markets
is that there is not a lot of serial correlation in returns.
Autocorrelation, or serial correlation,
of returns on most assets is very close to zero.
This is the case for stocks and long-term bonds.
The reason for that has to do with the fact
that high degree of serial correlation
would imply that it is relatively easy to forecast
future returns by observing past returns;
the two are closely related.
Being able to predict returns in risky assets
would then enable one to design highly profitable trading
strategies.
These are difficult to come by.
As a result, the implication is that one should not
observe high levels of autocorrelation
in returns, particularly at high frequency-- daily,
weekly, or monthly.
The following figure illustrates the point.
We are now looking at the monthly returns on a stock.
It's a stock of IBM.
And the scatter plot shows returns
in any given month against returns in the previous month.
We see that there is no clear linear pattern on this graph.
And in fact, the empirical estimate of the autocorrelation
over this period, which starts at the beginning of '08, is
close to minus 6%.
There is not a lot of autocorrelation
in returns on this stock or most other risky investments.