Corporate Finance - Statistics Review

Random Variable Our starting point is the notion of a random variable. As an example, we can think of the return on a stock. If there is uncertainty, for example, because of various possible levels of GNP, which determine whether the firm's products are profitable or not, then the firm may have a number of different returns on its stock depending on what actual incomes and so on are.

The possible outcomes are described by a random variable, which we can call X. For the moment we shall look at discrete random variables for the purposes of illustration. Discrete random variables are those in which the number of possibilities is finite rather than infinite. We denote a specific realization of the random variable X by Xi. State State 1 State 2 Realization X1 - High return 3% X2 - Low return 1%

Probability There are a number of notions as to what probability represents. Perhaps the easiest one to understand is that of relative frequency. In the long run, it measures the proportion of times we get high returns, and the proportion we get low returns.

State State 1 State 2

pi 0.50 0.50

Note probabilities have the following properties in general pi 0 for all pi

pi = 1
i =1

where there are n possible events.

Expectation In order to be able to consider the notion of means, variances and so on it is necessary to define the expectation of a random variable.
n

EX

= pixi
i =1 n

EX
I In general, we have

= pixi2
i =1

E[f(X)]
Two useful formulas are the following: If a is a constant
n

= pif(xi)
i =1

E(aX)

= apixi = a pixi = aEX

E(X + Y)
n

= (pixi (
i =1 n

piyi)

= pixi + piyi = EX + EY
i =1 i =1

The Arithmetic Mean or Average

The first notion we are interested in is the arithmetic mean or average. This is given by __

Mean = EX
e.g.,

= pixi =

EX = 0.5(3) + 0.5(1) = 2%
Variance We are not only interested in the average value, we are also interested in how much returns vary. For example, if you invested the money for your tuition this semester last year you presumably would have liked to know how variable the returns would likely have been. If you had invested in short term government instruments you could have been fairly sure of meeting your tuition payment. If you had put it in stocks you might have made a lot of money or you could have lost a lot of money. To measure variability we use variance
___ n ___

Variance =
e.g.,

X=

E(X X) =
i =1

pi (xi X) 2

Var X = 0.5(3 - 2) 2 + 0.5(1 - 2) 2 = 1

x1* = 4; x2* = 0
so that

Var X* = 0.5(4 - 2) 2 + 0.5(0 - 2) 2 = 4

The diagram below illustrates this.

Standard deviation The problem with variance is that it isn't in the same units as the mean--it's in %2 . In the above example with outcomes of 4 and 0 this means that although the average absolute deviation from the mean is 2 the variance is 4. It is often useful to work with standard deviation that is just the square root of variance. This means that its units are the same as those of the mean. Standard Deviation = Variance = SD = X In our example with X1 * = 4 and X2 * = 0 SD X* = 4 = 2 Since standard deviation has the advantage that the units are the same as the units of mean, we will mostly be using standard deviation as a measure of risk.

Covariance When we construct portfolios of stocks one of the important things of interest is how the stocks move together. For example, suppose two stocks move in opposite directions so that when one has high returns the other has low returns and vice-versa. Then a portfolio of the two stocks would tend to have a low variance since the movements in returns would be offsetting. If on the other hand the two stocks' returns move in the same direction the variance of the portfolio will be high since there will be no offsetting movement.

A measure of how random variables move together is covariance. If we have two random variables, X and Y say, their covariance is defined as follows.

__ __ Cov (X,Y) = XY = E(X X) (Y Y)

=
i =1

pi (xi X) (yi Y)

State State 1 State 2

xi 3 1 EX = 2

yi 6 2 EY = 4

pi 0.5 0.5

Cov (X,Y) = XY = 0.5(32)(64) + 0.5(12)(24) = 2 Note that in this case the variables move together, so that X is above its mean when Y is above its mean. This gives a positive term in the calculation of covariance. When X is below its mean Y is also below its mean so we have a negative times a negative which gives a positive. Hence covariance is positive. This happens whenever two variables move together. What happens if the two variables move in opposite directions?

Example 2 State State 1 State 2 xi 3 1 EX = 2 yi 2 6 EY = 4 pi 0.5 0.5

Cov (X,Y) = XY = 0.5(32)(24) + 0.5(12)(64) = 2 In this case we have a negative covariance. This is because when X is above its mean Y is below its mean and vice-versa so in each term we have a positive times a negative

which gives a negative. Thus when variables move in opposite directions they have a negative covariance. If they are independent and don't tend to move around together or in opposite directions, then we get some positive terms and some negative terms. When there is no systematic relationship at all these positive and negative terms tend to cancel out and we get a zero covariance. When two random variables are independent they have zero covariance.

Correlation One of the problems with covariance is that its magnitude depends on the units of measurement. It is not possible to simply deduce from the size of the number how much they move together. A concept which allows us to do this is the correlation coefficient. Correlation Coefficient = Corr(X,Y) = XY =

Cov (X,Y) SD X SD Y

Now it can be shown -1 XY +1 If XY = +1, they are perfectly correlated and move in the same direction in proportion. If XY = -1, they are perfectly negatively correlated, they move in opposite directions and in proportion. If XY = 0, then on average they don't move together or in opposite directions.

Example 1 Var Y = 0.5(2)2 + 0.5(-2) 2 = 4; SD Y = 2 XY =

2 1x2

= +1

In this case, they're perfectly correlated: they move in the same direction and in proportion. When X is 1 above its mean, Y is 2 above its mean; when X is 1 below its mean, Y is 2 below its mean.

Example 2 XY =

-2 1x2

= -1

In this example they're perfectly negatively correlated. When X is 1 above its mean, Y is 2 below its mean; when X is 1 below its mean, Y is 2 above its mean.

Example 3 State State 1 State 2 State 3 xi 1 2 3 EX = 1.7 SD X = 0.781 yi 9 11 25 EY = 12.8 SD Y = 6.161 pi 0.5 0.3 0.2

Cov(X,Y) = 0.5(1 - 1.7)(9 - 12.8) + 0.3(2 - 1.7)(11 - 12.8) + 0.2(3 - 1.7)(25 - 12.8) = 4.34 XY =

4.34

= 0.902

0.781 x 6.161

In this case, when X is 0.7 below average, Y is 3.8 below; when X is 0.3 above average, Y is 1.8 below average; when X is 1.3 above average, Y is 12.2 above average. Thus when one goes up, the other tends to go up, but doesn't always do so. Hence XY is positive but less than one.

Evidence on Covariance and Correlation As far as the empirical evidence is concerned there are two important conclusions that can be drawn from the data: (i) Stock returns are serially uncorrelated or in other words if stock returns are high one year then you can't use this information to predict whether returns in the subsequent year will be high or low. This evidence will be important when we talk about market

efficiency later in the course. However, for the moment the one conclusion that is relevant as far as models of asset pricing is concerned is: (ii) Within a period most stocks are positively correlated to the market portfolio (e.g. a value-weighted portfolio of all the stocks on the NYSE).

Regression Suppose we have two random variables and we plot them on a diagram. Regression theory allows us to fit a line y = a + bx representing how the two variables are related. y

x The estimate of the slope of the line, b* , is given by b* = Cov (y,x) Var x What does this remind you of? If you remember initially when we talked about , we said it was Cov(Stock, Market) /Variance(Market). As we will be seeing one interpretation of is that it is the slope of the regression when you plot the returns on a stock against returns on the market.

Results Needed for Theories of Asset Pricing Diversification Suppose you have a stock X1 with payoff distribution: Payoff 0 2 so Prob. 0.5 0.5

Mean = 1; Variance = 1; S.D. = 1

The stock can be thought of as like a coin toss. With tails (T) the payoff is 0 and with Heads (H) the payoff is 2.

Next suppose you have another stock just like X1 which we will call X2 . It has the same payoff distribution and the same price. It's payoffs are independent of X1 's. Instead of using your money to buy 1 share of X1 , suppose instead you buy 1/2 a share of X1 and 1/2 a share of X2 . In other words, you buy a portfolio of stocks. What is the mean and standard deviation of the portfolio?

State 1 2 3 4

Payoff 2T:0 TH: (1/2) 0+(1/2) 2 = 1 HT: (1/2) 2+(1/2) 0 = 1 2H: (1/2) 2+(1/2) 2 = 2 Mean = 1

Prob. 0.25 0.25 0.25 0.25

Variance = 0.25(0-1)2 + 0.5(1-1) 2 + 0.25(2-1) 2 = 0.5 S.D. = 0.707 The mean stays the same but because you can spread your investment across the two stocks the variance and S.D. go down. This is because the heads and tails cancel out in states 2 and 3. It is this basic idea that underlies most analysis of risk in finance. If you add more stocks and split your investment between them you keep on getting a reduction in variance and standard deviation. Eventually, it follows from what is known as the law

of large numbers that all the risks go as long as the stocks are independent and so have zero covariance and correlation.

Nonzero covariance and correlation Next consider what happens if the two stocks X1 and X2 are perfectly negatively correlated i.e., = -1. In that case what happens to the portfolio mean, variance and standard deviation? State 1 2 3 4 Payoff 2T: 0 TH: (1/2) 0+(1/2) 2 = 1 HT: (1/2) 2+(1/2) 0 = 1 2H: (1/2) 2+(1/2) 2 = 2 Mean = 1 Variance = 0(0-1)2 + 1(1-1) 2 + 0(2-1) 2 = 0 S.D. = 0 Here you can never get two tails or two heads you always get one head and one tail. Risk is completely eliminated. In terms of the probability distribution you are putting all the weight on the mean at 1. What would happen if you had a negative correlation with -1 < < 0? This case lies in between the two cases above. For example, if having obtained 1 T the probability of a second T is O.25 and similarly having obtained 1 H the probability of a second is also 0.25 it can be shown the distribution of payoffs is as below and it can be straightforwardly checked the correlation is -0.5. State 1 2 3 4 Payoff 2T: 0 TH: (1/2) 0+(1/2) 2 = 1 HT: (1/2) 2+(1/2) 0 = 1 2H: (1/2) 2+(1/2) 2 = 2 Mean = 1 Variance = 0.125(0-1) + 0.75(1-1) 2 + 0.125(2-1) 2 = 0.25 S.D. = 0.5
2

Prob. 0 0.5 0.5 0

Prob. 0.125 0.375 0.375 0.125

Here with = -0.5 the weight put on the extremes of 2T and 2H is half way between the cases where = -1 and = 0. As a result the degree of risk in terms of the standard deviation of the portfolio is also between the two cases above. With positive correlation, the analysis is similar. With = +1 all the weight is at the extremes so the probability of 2T and 2H is 0.5 and there is no benefit from diversification. With = +0.5, the probability weight on 2T and 2H is 0.375 and the weight on HT and TH is 0.125. In conclusion, except in the case where = +1 there is always some benefit to spreading your investment among different stocks. As correlation goes down this benefit increases. As we will see this is the basic idea behind diversification. We shall see below we can extend it to a wide range of situations with many stocks and differing probability distributions of returns.

Means and Variances of Portfolios In the case above it was quite a simple task to take the individual stock returns combine them with the (1/2) and (1/2) weights on the two stocks and work out the payoffs of the portfolio in each state. From these portfolio payoffs we calculated the mean, variance and standard deviation of the portfolio. As the number of states becomes larger this procedure of calculating portfolio payoffs state by state and then calculating the statistics for the portfolio payoff distribution becomes more and more cumbersome. An alternative which is considerably easier is just to use the means and standard deviations of the individual stocks and their covariances or correlations and use them to calculate the mean and standard deviation of the portfolios. There are some useful formulas which allow us to do this. We will be using them a lot in the coming weeks. They are derived in the appendix at the end of the section. These formulas look fearsome but there are only two basic things to remember.

1. MEANS ADD. In other words if you invest half your money in stock X1 and the other half in stock X2 , the mean return of the portfolio is simply half the mean of stock X1 and half the mean of stock X2 .

2. VARIANCES DON'T ADD - COVARIANCES MATTER. In other words if we take the same example of half our money in stock X1 and half in stock X2 , the variance of the portfolio is not just half the variance of stock X1 and half the variance of stock X2 . The reason is that correlations play an important role too as the simple examples above demonstrated. This is why there is a correlation or covariance term when calculating the variance of a portfolio. For the moment the thing to remember is means add but variances and also standard deviations don't. We will come back to this later on and consider examples to illustrate it.

Statistical Review: Conclusion This is all the statistics you'll basically need to know. The main thing is to have an intuitive understanding of what they mean. Arithmetic mean is a measure of how much you can expect to receive if you hold a stock for a year. The variance and standard deviation are measures of how variable the returns are likely to be. The higher the variance or standard deviation the greater the variation. Covariance and correlation are measures of whether two variables move together or in opposite directions. If they move together covariance and correlation are positive; if they move in opposite directions covariance and correlation are negative and if they are independent covariance and correlation are zero. The lower the correlation the greater the reduction in risk from putting stocks together in a portfolio.

Appendix

Results Needed for Theories of Asset Pricing Note: To derive the formulas we need the following. If is a constant

E(X)
Also

= pixi = pixi = EX
i =1 i =1 n

E(X + Y)
n

= (pixi (
i =1 n

piyi)

= pixi + piyi = EX + EY
i =1 i =1

Mean and Variance of a Portfolio of 2 assets Suppose that you put 1 in stock X1 and 2 in stock X2.

Return on the portfolio = X = 1 X1 + 2 X2

This implies

Mean of portfolio = EX = 1 EX1 + 2 EX2 i.e. MEANS ADD

Using the definition of variance E(X - EX)2 gives Variance of Portfolio = E[ 1 X1 + 2 X2 - ( 1 EX1 + 2 EX2 )] 2 = E[ 1 (X1 - EX1 ) + 2 (X2 - EX2 )] 2 = E[ 1 2 (X1 - EX1 ) 2 + 2 2 (X2 - EX2 )2 + 2 1 2 (X1 - EX1 )(X2 - EX2 )]

This simplifies to

Variance of Portfolio = 1 2 E(X1 -EX1 )2 + 2 2 E (X2 -EX2 )2 +2 1 2 E(X1 -EX1 )(X2 - EX2 ) = 1 2 Var X1 + 2 2 Var X2 + 2 1 2 Cov (X1 ,X2 )

= i2 Var Xi + i j Cov (Xi ,Xj)

or writing it in full

i2 Var X1
+ 1 2 Cov (X1,X2 )

+ 1 2 Cov (X1,X2) + 22 Var X2

Notice that in this format which will be useful later on the diagonal terms are variances and the off-diagonal terms are covariances. Note also VARIANCES DON'T ADD

Mean and Variance of a portfolio of N assets Suppose that you put i in stock Xi for i = 1,...,N. The formulas for mean and variance are just extensions of the case with 2 assets.

Mean of portfolio = 1 EX1 + 2 EX2 + . . . + NEXN

MEANS ADD
N N N

Variance of Portfolio =

i
i =1

Var Xi +

i j Cov (Xi ,Xj)

i2 Var X1

+ 1 2 Cov (X1,X2) + 22 Var X2 + N 2 Cov (X2,XN)

+ 1 N Cov (X1,XN) + 2 N Cov (X2,XN) 2 + 2 Var XN

+ 2 1 Cov (X1,X2) + N 1 Cov (X1,XN)

Again the diagonal terms are variances and the off-diagonal terms are covariances.

VARIANCES DON'T ADD