Modern portfolio theory is the philosophical opposite of traditional stock picking.

It is the
creation of economists, who try to understand the market as a whole, rather than business
analysts, who look for what makes each investment opportunity unique. Investments are
described statistically, in terms of their expected long-term return rate and their expected short-
term volatility. The volatility is equated with "risk", measuring how much worse than average an
investment's bad years are likely to be. The goal is to identify your acceptable level of risk
tolerance, and then to find a portfolio with the maximum expected return for that level of risk.

This article covers the highlights of modern portfolio theory, describing how risk and its effects
are measured, and how planning and asset allocation can help you do something about it.

Volatility and your Time Horizon

The volatility of an investment is measured by the standard deviation of its rate of return.

(If your statistics is a little rusty, you can think of the standard deviation as measuring how far
away from the average the return rate for any one year is likely to be. The greater the standard
deviation, the more variable the rate of return.)

This interactive graph shows how expected returns and expected fluctuations affect the likely
outcomes of your investment. The graph uses a random process, so there is uncertainty built into
the result - just like life! You choose an investment with a specified return and volatility, and the
graph will produce a bell curve of possible outcomes.

The area of the "negative return zone", shown in red, is proportional to the probability that you
will lose money on your investment.

The Efficient Frontier and Portfolio Diversification

The graph on the previous page shows how volatility increases your risk of loss of principal, and
how this risk worsens as your time horizon shrinks. So all other things being equal, you would
like to minimize volatility in your portfolio.

Of course the problem is that there is another effect that works in the opposite direction: if you
limit yourself to low-risk securities, you'll be limiting yourself to investments that tend to have
low rates of return. So what you really want to do is include some higher growth, higher risk
securities in your portfolio, but combine them in a smart way, so that some of their fluctuations
cancel each other out. (In statistical terms, you're looking for a combined standard deviation
that's low, relative to the standard deviations of the individual securities.) The result should give
you a high average rate of return, with less of the harmful fluctuations.

The science of risk-efficient portfolios is associated with a couple of guys (a couple of Nobel
laureates, actually) named Harry Markowitz and Bill Sharpe.
Suppose you have data for a collection of securities (like the S & P 500 stocks, for example), and
you graph the return rates and standard deviations for these securities, and for all portfolios you
can get by allocating among them. Markowitz showed that you get a region bounded by an
upward-sloping curve, which he called the efficient frontier.

It's clear that for any given value of standard deviation, you would like to choose a portfolio that
gives you the greatest possible rate of return; so you always want a portfolio that lies up along
the efficient frontier, rather than lower down, in the interior of the region. This is the first
important property of the efficient frontier: it's where the best portfolios are.

The second important property of the efficient frontier is that it's curved, not straight. This is
actually significant -- in fact, it's the key
to how diversification lets you improve
your reward-to-risk ratio. To see why,
imagine a 50/50 allocation between just
two securities. Assuming that the year-
to-year performance of these two
securities is not perfectly in sync -- that
is, assuming that the great years and the
lousy years for Security 1 don't
correspond perfectly to the great years
and lousy years for Security 2, but that
their cycles are at least a little off -- then
the standard deviation of the 50/50
allocation will be less than the average
of the standard deviations of the two securities separately. Graphically, this stretches the possible
allocations to the left of the straight line joining the two securities.

In statistical terms, this effect is due to lack of covariance. The smaller the covariance between
the two securities -- the more out of sync they are -- the smaller the standard deviation of a
portfolio that combines them. The ultimate would be to find two securities with negative
covariance (very out of sync: the best years of one happen during the worst years of the other,
and vice versa).

Next: finding the best portfolio in the efficient frontier

The Sharpe Ratio

The previous page showed that the efficient frontier is where the most risk-efficient portfolios
are, for a given collection of securities. The Sharpe Ratio goes further: it actually helps you find
the best possible proportion of these securities to use, in a portfolio that can also contain cash.

The definition of the Sharpe Ratio is:

S(x) = ( rx - Rf ) / StdDev(x)
where
x is some investment
rx is the average annual rate of return of x
Rf is the best available rate of return of a "risk-free" security (i.e. cash)
StdDev(x) is the standard deviation of rx

The Sharpe Ratio is a direct measure of reward-to-risk. To see how it helps you in creating a
portfolio, consider the diagram of the efficient frontier again, this time with cash drawn in.

There are three important things to notice in this diagram:

1. If you take some investment like "x" and combine it with cash, the resulting portfolio will
lie somewhere along the straight line joining cash with x. (This time it's a straight line,
not a curve; cash is riskless, so there's no "damping out" effect between cash and x.)
2. Since you want the rate of return to be as great as possible, you want to select the x that
gives you the line with the greatest possible slope (like we have done in the diagram).
3. The slope of this line is equal to the Sharpe Ratio of x.
Putting this all together gives you the method for finding the best possible portfolio from this
collection of securities: First, find the investment with the highest possible Sharpe Ratio (this
part requires a computer); Next, take whatever linear combination of this investment and cash
will give you your desired value for standard deviation. The result will be the portfolio with the
greatest possible rate of return.

Next: really use the Sharpe Ratio to build a risk-efficient portfolio.

Index Funds and Optimal Portfolios

The portfolio demo was easy to use because it assumes that the investment universe consists only
of two market securities, plus riskless cash. But of course the real investment universe is a lot
bigger than that, with thousands of choices among U.S. stocks alone. In theory you could find the
optimal point on the efficient frontier generated by this many securities, but doing that wouldn't
be practical. For one thing, you'd have to calculate the covariance between every pair of
securities: thousands of securities means millions of covariance calculations. But even if you
could do all those calculations, you wouldn't really want to. That's because the efficient frontier
is based on an idealized model of the way investments work; and when you apply a huge number
of calculations to a model you tend to amplify the error between the model and reality, leaving
you with more "noise" than anything else.

So as a practical matter, putting portfolio theory to work means reducing the problem to
something about as simple as the portfolio demo, and investing in a small number of index funds
rather than a huge number of individual stocks and bonds.

Index investing is where portfolio theory starts to rely on the efficient market hypothesis. When
you buy an index you're allocating your money the same way the whole market is - which is a
good thing if you believe the market has a plan. This is why portfolio theory really is a branch of
economics rather than finance: instead of studying financial statements you study the aggregate
behavior of investors, some of whom presumably have studied financial statements so that
market valuations will reflect their due diligence.

(This viewpoint also gives rise to some bad blood that's pretty entertaining if you aren't involved.
The economists see business analysts as necessary to market efficiency, but otherwise rather
beneath them as a life form, like the bacteria that make yogurt: they're useful, but they're
basically germs. The "germs" respond that the economists are delusional eggheads whose
theories collapse whenever real money is involved.)

Tobin Separation Theorem

The pioneering result that helped popularize index investing was Tobin's "separation theorem",
which Bill Sharpe summarized this way in an interview:

James Tobin ... in a 1958 paper said if you hold risky securities and are able to
borrow - buying stocks on margin - or lend - buying risk-free assets - and you do so
at the same rate, then the efficient frontier is a single portfolio of risky securities
plus borrowing and lending....

Tobin's Separation Theorem says you can separate the problem into first finding that optimal
combination of risky securities and then deciding whether to lend or borrow, depending on your
attitude toward risk. It then showed that if there's only one portfolio plus borrowing and lending,
it's got to be the market.

The reasoning behind this is easy to understand from the same kind of diagram we have already
been looking at:

As usual you're trying to build an optimal portfolio for your risk tolerance; and as before, it will
lie somewhere on the straight line joining the cash rate Rf to some optimal mix on the efficient
frontier. We're specifically assuming what Sharpe said, that high risk investors can and will buy
on margin, with money borrowed at the low rate Rf. That's why there is just one straight line in
the picture, and one unique optimal mix on the efficient frontier; so the problem of building an
optimal portfolio is "separated" into somehow finding the optimal mix and then combining it
with cash to give you your desired risk tolerance.

Now for the part that's really interesting. Assume that everybody is facing the same efficient
frontier that you are, and that the market is efficient in the specific sense that it behaves in the
aggregate as if everybody is trying to build an efficient portfolio this way. That means it behaves
as if everybody is on your straight line, with the same optimal mix as you. So the mix that the
market is holding - the index - is guaranteed to be your own personal optimal mix.

That's an incredibly elegant result... but it requires you to accept some really strong hypotheses.
(Two quick jabs: real investors can't afford to be so cavalier about the special risks of margin
buying; and different tax brackets mean different people face different efficient frontiers.
Goodbye single straight line; so long universal optimal mix.)

Probably due to problems like those, results about index investing have trended away from
proofs that index funds are optimal toward statistical models confirming that index funds are
hard to beat. That's a trend we'll be following on the next three pages, with CAPM (a theoretical
model that looks like a statistical model) and the three factor model (a pure statistical model with
a little theory suggested as an afterthought).

Next: how CAPM relates individual securities to the index.

Capital Asset Pricing Model

Bill Sharpe made his first big breakthrough by taking the picture on the previous page and
showing how the market must price individual securities in relation to their asset class (a.k.a. the
index, or the "optimal mix" in the picture). The derivation isn't exactly a walk in the park
(yikes!), but the result is a simple linear relationship known as the Capital Asset Pricing Model:

r = Rf + beta x ( Km - Rf )
where
r is the expected return rate on a security;
Rf is the rate of a "risk-free" investment, i.e. cash;
Km is the return rate of the appropriate asset class.

Beta measures the volatility of the security, relative to the asset class. The equation is saying that
investors require higher levels of expected returns to compensate them for higher expected risk.
You can think of the formula as predicting a security's behavior as a function of beta: CAPM
says that if you know a security's beta then you know the value of r that investors expect it to
have.

Naturally, somebody has to verify that this simple relationship actually holds true in the market.
Part of the question is how few classes you can get away with: whether you can use a very coarse
division into just "stocks" and "bonds", or whether you need to divide much further (into
"domestic mid-cap value stocks", and so on). There are also ongoing attempts at "building better
betas" that incorporate company debt and other traditional valuation measures, instead of relying
solely on past volatility, to measure risk. All of this is a full-time job for academic modern
portfolio theorists (and deriding the whole effort is a popular hobby for some traditional stock
analysts: how could a magnificent company equal a mediocre one times beta? To them, CAPM
seems like a very blunt instrument.)
CAPM has a lot of important consequences. For one thing it turns finding the efficient frontier
into a doable task, because you only have to calculate the covariances of every pair of classes,
instead of every pair of everything.

Another consequence is that CAPM implies that investing in individual stocks is pointless,
because you can duplicate the reward and risk characteristics of any security just by using the
right mix of cash with the appropriate asset class. This is why followers of MPT avoid stocks,
and instead build portfolios out of low cost index funds.

(One point about that last paragraph. If you are trying to duplicate an expected return that's
greater than that of the asset class, you have to hold "negative" cash, meaning you have to buy
the index on margin. This is consistent with the big message of MPT - that trying to beat the
index is inherently risky).

Next: finding CAPM from linear regression.

Regression, Alpha, R-Squared

One use of CAPM is to analyze the performance of mutual funds and other portfolios - in
particular, to make active fund managers look bad. The technique is to compare the historical
risk-adjusted returns (that's the return minus the return of risk-free cash) of the fund against those
of an appropriate index, and then use least-squares regression to fit a straight line through the
data points:

Each data point in this graph shows the risk-adjusted return of the portfolio and that of the index
over one time period in the past. (For example, you might make a graph like this with twenty
data points, showing the annual returns for each of the past twenty years.)

The general equation of this type of line is

r - Rf = beta x ( Km - Rf ) + alpha
where r is the fund's return rate, Rf is the risk-free return rate, and Km is the return of the index.

Note that, except for alpha, this is the equation for CAPM - that is, the beta you get from
Sharpe's derivation of equilibrium prices is essentially the same beta you get from doing a least-
squares regression against the data. (Also note that alpha and beta are standard symbols that
statisticians use all the time for this type of regression; Sharpe and his followers weren't trying to
be obscure, as some people like to believe.)

Beta is the slope of this line. Alpha, the vertical intercept, tells you how much better the fund did
than CAPM predicted (or maybe more typically, a negative alpha tells you how much worse it
did, probably due to high management fees).

The quality of the fit is given by the statistical number r-squared. An r-squared of 1.0 would
mean that the model fit the data perfectly, with the line going right through every data point.
More realistically, with real data you'd get an r-squared of around .85. From that you would
conclude that 85% of the fund's performance is explained by its risk exposure, as measured by
beta. (Then you'd punch your fist in the air and say "And the other 15% is due to pure luck!"
MPT never believes in investor skill: an investment's behavior equals that of its asset class,
minus management fees, plus-or-minus unpredictable luck.)

Next: three factor regression.