The Nobel Prize winning creator of the Capital Asset Pricing Model takes another look at his work.

William Sharpe won the Nobel Memorial Prize in Economics in 1990 for his contribution to Modern Portfolio Theory. Sharpe shared the award with Harry Markowitz. Markowitz won the prize because of his ideas about diversification. Sharpe won it because he devised the Capital Asset Pricing Model (CAPM), a formula for determining the theoretically required rate of return-and, therefore, the price-of an asset when added to a well-diversified portfolio. These ideas form the basis for Modern Portfolio Theory, and that is the theoretical underpinning for many of the world's investors, including many of the smartest independent financial advisors.

Gluck: So, what got into you? What made you write the book?
Sharpe: I was asked some years ago to give a series of lectures at Princeton University. Once a year they'd have somebody come in and give three lectures on a subject of their choosing, and then it's turned into a book. It seemed to be a good opportunity to put together what my thoughts were about the area I've been working in heavily for many, many, years-asset prices, risk and return, equilibrium-and to try to get my arms around both what I'd done and what other people have done, and maybe have some new thoughts. So I gave the lectures in 2003 and then worked on turning it into a book. For a number of years, I've tried to think about finance in terms of this State-Preference approach, to think of the future as one in which the world can be in a number of discrete states, and each one has a probability...

Gluck: Sorry to interrupt, but what do you mean by discrete states?
Sharpe: As opposed to continuous. I'm sorry, that's mathematician jargon. It's either A or B or C or D. The market is either up 10% or 11% or 12%. It rains or it shines. It's cold or it's warm, or it's really hot. An easy way to think about it is a spreadsheet. You've got a column in the spreadsheet, and each row is a different thing that could happen. Only one of them will happen, but you don't know which one. The best you can do is say, "There's a 10% chance it'll be this row and an 8% chance it'll be that row."
That's opposed to the kind of Mean Variance notion, the traditional paradigm that we've used a lot in finance for a long time, the Markowitz portfolio theory/Capital Asset Pricing Model (CAPM), in which you have a continuous probability distribution. The market could be up 10.12346% or 10.12347%, etc. Which row will it be? And then you can think in terms of each column. If you end up in row 12, then stocks will do this, bonds will do that, hedge funds will do this, General Motors will do that, General Electric will do that. You think about it in that manner, and that has a couple of advantages. First of all you can be much more general in certain respects than you can with traditional theories, and second, the math's a lot easier. You don't have to have very well tuned mathematical and statistical skills to even approach the subject. So I've tended to think of things in that way for a number of years, and I've taught some of my courses in that way. So I thought, let's take that approach and let's think about broader and more realistic worlds than we usually think about.

Gluck: How is this more realistic than the traditional way of thinking about investing, the way that you helped invent, in which you rely on Mean Variance Optimization and the Capital Asset Pricing Model? How did you go from here to there?
Sharpe: Theory is always an abstraction and a simplification of reality and, at best, an approximation. You can't get your arms around enough of reality to do a totally realistic theory. So with theory, you always abstract and approximate. With the Markowitz portfolio analysis and CAPM we made a lot of simplifications and approximations, hoping that we're looking at the essence of the problem and not missing something really important. Then, over the course of the last few decades, people have been saying, "What if you think about the impact of taxes, both on how you choose a portfolio and how the market will price assets?" There's been this progressive attempt to bring in more elements of reality and trying to focus on the most important ones because you're never going to get them all in. I view these ideas as part of that progression even while it looks like a paradigm shift...

Gluck: It does look like you've made a major shift in how you look at investing.
Sharpe: Because it's a different apparatus. But it really isn't a huge shift in thinking. For example, if you take the simulator, which I'll talk about in a second, and you just tell the simulator that people have certain preferences, voilà, you get all the results of Markowitz and CAPM. So it's a more powerful mechanism, which as a special case gives you all the key results of the traditional theories. But whereas the traditional theories have been limited because you can't deal with a lot of complexity, you can deal with more complexity with this approach. As I was preparing the lectures on this, I thought I'd really like to have some examples of what happens when people disagree about the future, because we know people do disagree about the probabilities of different things. What happens when people have preferences that are not only different from what Markowitz and the CAPM assumes, but more like the kinds of things that the behaviorists are now telling us characterize human beings.

Gluck: What were the preferences assumed, by the way, by Markowitz and CAPM? And how does the new State-Preference approach you are now working on differ from the Markowitz/CAPM approach?

Gluck: And that program is on your Web site?
Sharpe: Yes, and it's available for anybody to use at www.wsharpe.com. In order to make it free, I had to write it in C and C++ so you could compile it for free. Anybody who wants to could look at the source code and documentation and enhance it or modify it and distribute it freely.

Gluck: Had you ever coded some other things like this before?

Sharpe: Oh, I write programs. I've been writing programs probably every day of my life. And I did a little C programming way back when. It's a difficult language for research, but it produces very fast code because you're getting close to the machine level. So it's great if a program will be widely distributed. I use Matlab for most of my work, which is much more research-oriented.

Gluck: Getting back to the theory, you're saying Mean Variance Optimization and CAPM are all still valid, and your recent work simply builds on that.

Gluck: So then, what you're saying is we knew the CAPM system was imperfect. And now this is an effort to make it more real-world?
Sharpe: Yeah, that's fair. No theory will be fully real-world. There's no way you can bring all of the elements of the real world into a set of equations or a computer program and get it all in there. So the question is, at what point do you have enough of the real world in your theory or your computer program so that you're comfortable going ahead with the implications as a matter of how you run portfolios?

Gluck: How has your real-world experience at Financial Engines influenced your ideas? [Financial Engines, founded by Sharpe, provides expert financial advice for investors, particularly those planning for retirement.

Sharpe: Over the course of the years, we've worked with a lot of people to see how they choose among different investment approaches and help them narrow their choices of investment approaches to alternatives that seem, based on theory and empirical work, to be more desirable. Now I don't do all the day-to-day empirical and analytic work, but seeing what happens when you deal with a real world of tens of thousands of investment alternatives and all the gory details and trying to predict into the future, and then show people those predictions and see how they react to them and what choices they make in terms of their savings, in terms of how they choose among riskier or less risky alternatives. We do that, obviously. We do it online with thousands of people. But we've also, over the course of years, done focus groups and test cases and all the rest. So I've learned a lot. There's no question about that.

Gluck: Could you use your simulation software to make real-world simulations?

Sharpe: No. It's to test ideas. To actually advise somebody as to how to invest a 401(k) plan, you need the fruits of ten years of many, many people's labors and a lot of money.

Gluck: So what are the practical implications that you hope are going to be realized from looking at markets the way you are now and using the State-Preference theory to make investment decisions?

Sharpe: Well, I'm just beginning to explore that. I have a paper on my Web site (www.wsharpe.com) under Working Papers called, Expected Utility Asset Allocation, and I'm working with a fund, a large endowment and university retirement plan fund. We're doing some experiments to try to bring some of these ideas into the asset allocation decision for a large institutional fund. It's in its early days, so I wouldn't predict how practical that's going to work out. We're getting some ideas about how we could actually use this kind of approach. But it's in its early days. As I tried to figure out for whom I was going to write this book, I ended up deciding I was going to write it for academics to try to convince them to change the way they taught finance. In using this approach, you can look at a lot more issues in what are actually simpler ways. I'm trying to influence people who will be teaching MBAs and undergraduates, trying to bring ideas that are taught now at the Ph.D. level down to a broader student body. If that's successful, then hopefully in time the industry will start shifting and implementing ideas that use this approach as well. But that's in the future. A lot of the work in financial engineering already uses the State-Preference approach.

Gluck: This approach is already used at the institutional level by the rocket scientists at the giant firms, right?
Sharpe: That's right, exactly. In some ways, what I'm trying to encourage people to bring [is] the rocket science of the Ph.Ds. in math at the giant firms-and the abstract models used by Ph.D.-level courses in some major universities-to a broader audience. If it works, that'll be done in the academy initially and then it'll spread to your readership.

Gluck: Please explain some of the basics of State-Preference theory. It's been around for some years, right?

Sharpe: It goes back to Gérard Debreu in the '50s, when Harry was writing. Think of that spreadsheet I spoke about. Each row in that spreadsheet is something that could happen in the future. Only one row, however, is actually going to happen, but you don't know which one. What people do a lot in the industry, they get 20 years of history or 40 years or 100 years, and you know each row is a year. Say you've got 50 years of history and each row is a year, and each column is the return on an asset class-bonds, stocks, non-U.S. stocks, real estate, what have you. What people often do is, they take that table and they compute the average return for each of the asset classes and the standard deviation and the correlations, and then maybe they modify them to take into account their current views of the future or something about theory. Then they run it through an optimizer, a là Markowitz. What this approach says is that maybe you want to just leave that table as it is, make your decisions as to how likely you think it is that next year will be like 1939 or instead like 1941, and take into account how you feel about losing 30% of your money and run a different kind of optimizer. That's a different way of using the data that many people already use.

Gluck: And you're using judgment.
Sharpe: And use judgment by saying, in effect, "Well, next year is as likely to be like 1939 as it is to be 2005." But you could say, "I think, given my view of the world, that 1939 returns are more likely than 2005." The great advantage is it allows you more freedom. The disadvantage is it allows you more freedom. But at least you know what you've assumed.
Whereas in a traditional procedure that some people follow, you in effect assume that next year is as likely to be 1939 as it is 2005, and maybe that's not really what you want to assume. Now you can make those assumptions and still use the traditional methods. You also assume that you have a particular view about a really bad outcome vis-à-vis a really good one. Whereas with this broader view, you can say, "No, no, no, for me losing 20% is the end of the world. I'll get fired from the investment committee, my client will go elsewhere and I might get sued. I don't want to take a chance, however small, of that happening." In Traditional Mean Variance approaches you can't really do that.

Gluck: But you break these judgments down into different preferences. Could you explain some of that?

Sharpe: Let's take it now to the financial advisor. If you think about the whole relationship between an investment professional and a client, basically the client has positions, preferences and predictions. I myself have positions-for instance, a house in Carmel and I consult in the investment industry. A lot of my financial situation is different from that of my neighbor. My neighbor was a journalist and he's now retired. Other friends of mine are in different professions and they're different from people who live in France, for example. So we're all different in various important ways.

Gluck: Especially the French.
Sharpe: Especially the French. [Laughing] Viva la difference! So people are different and every good advisor knows that. So the first thing you need to do is to mold an investment portfolio to take into account the differences between your client and sort of the great average client out there in the world at large. People have different preferences. One client says, "Look, this is only a small amount of my wealth, I've got all this other wealth. So let's go for it." Another client says, "I'm willing to take some risk but I can't take any risk of having less than this amount, because if I have less than this I have to go to a lousy nursing home instead of a good one." So I put in preferences.
That's where the client is the expert. The client knows how he or she feels about various kinds of risk. The client knows what his other sources of income and wealth are. The client is the expert. On the other hand, the client typically doesn't know much at all about what's available in the capital markets: What kind of instruments are there, what kind of risks are associated, future risks, with this kind of portfolio or that-what can you reasonably expect to get in the long run from this kind of portfolio. So the advisor is the expert on the capital markets and the opportunities, the trade-offs and the ways of implementing.

Gluck: And those factors were never part of investment decisions in your earlier approach?
Sharpe: Well, they were, but we summarized the client. We said Client A has a risk aversion of this and Client B has a risk aversion of that, and that's about all we did. But if you look at Mean Variance theory, if you look at the CAPM, everybody is just a little risk aversion model: I'm conservative, you're aggressive, someone else is moderate. That's about it. Now, I'm in many ways trying to bring in to formal procedures and computational procedures things that every good financial advisor has known and practiced for years. Is this news to a good financial advisor? Of course it's not. Yes, we've known this all along. This is trying to be a little more rigorous about how we do this process. It's thinking more about how capital markets function and what the impact might be on the trade-offs available, in terms of risks and returns.

Gluck: Are there any software products on the market now that are moving in this direction?
Sharpe: Not that I'm aware of, certainly nothing that I'm aware of that a financial advisor could pick up and use off the shelf.

Gluck: Is there another way to get at it, to maybe approximate it and do it yourself? Maybe do it with the kind of spreadsheet, the kind of table you were talking about before?
Sharpe: That's what I'm working on in this Expected Utility Asset Allocation paper. It is basically a set of recipes. How that will work in practice, I have no notion yet. But I'm working on it. As a matter of fact, I've been working on some examples for a talk I'm giving next month.

Gluck: So it'll give advisors a basic framework?
Sharpe: Yes, but it is a very basic framework. Again, I'm just beginning to experiment with real data to see what happens when you actually try to apply this approach, and it's too early for me to predict it's going to be the be-all and end-all approach. Just as it took us decades to figure out how to use Markowitz and the CAPM in ways that would actually give plausible results, it may take a while for this as well.

Gluck: By the way, State-Preference theory, you said it was around in the early '50s?
Sharpe: It's evolved. In the '50s Harry said, "Here's a way to advise a client as to what portfolio to hold. Here's a way to choose a portfolio when you have predictions of risks and correlations and expected returns." Harry's focus was, "You get your predictions from wherever you get them and here's how you use them to come up with a good portfolio." It's what I call normative economics, meaning that it tells you, "Here's what you should do." It's advice-oriented. Ken Arrow and Gerard Debreu were asking a different question. In an efficient market, what determines the prices of financial assets and therefore the risks and returns? It was what I call positive economics.
Here is how the world would work in this abstract kind of world. Those of us that worked in the CAPM were in the Arrow-Debreu camp, in the sense that we were asking how the world worked, positive economics. Of course, the trick is to give your advice with a gestalt as to how the world works. In other words, Harry said, "Wherever you get your predictions, here's how to use them to give advice." CAPM, and before that Arrow-Debreu, said, "Well, here's what the predictions ought to look like because they come from a capital market." Then, the trick is to say, let's get predictions that make sense and use them using something like Harry's approach, or this broader expected utility approach. So there are both sides of the coin. Ken and Gerard were very much in an equilibrium economics context, and not even in a heavily financial sense. The financial aspect, the uncertainty aspect, was kind of a page or two at the end of Ken's article and four pages at the end of Gerard's book. But over the years economists and financial economists sort of started developing the Arrow-Debreu approach and bringing it more and more to the point where you have what you have now.

Meanwhile, of course, the mainline in the investments industry was working off Harry's Mean Variance approach, the CAPM, which built on that, and all the developments that came thereafter. So you had these parallel tracks. Meanwhile the financial engineers are coming up with things that are basically State-Preference but coming at it in a different way, and in many cases not even knowing that philosophically they're in the same camp. One of the things I was trying to do in the book is to show that all the Mean Variance can be thought of as a special case of this broader State-Preference approach and try to get an integration of some sort.

Sharpe: What happened was State-Preference was growing mainly in the economic departments and the finance departments and the business schools were coming predominantly from the Mean Variance tradition-until in the last decade or two, at which point the advanced research tended to move over to the State-Preference side.

Gluck: Will advisors be able to utilize this book?
Sharpe: Some of the reviews of the book have said absolutely correctly, "Well you know this is too hard for practitioners and it's too verbose for academics. Academics would have much preferred the formula, the math, instead of all this verbiage. They're absolutely right. I was trying to say, "Look, Ph.Ds. and finance professors teaching Ph.D courses, this is a way of doing something differently when you go to the MBA classroom and the undergraduate classroom. You can bring these ideas to the MBAs and the undergraduates and here's the way to do it." And so to some extent, I'm trying to illustrate how easy it is to make these ideas understandable. So the book has this strange, ambivalent focus.

Gluck: One of the points you make late in the book is that advisors are good.
Sharpe: You better believe it. I like to differentiate it and trivialize investing versus betting. Investing is saying, "You are my client. You are a unique individual with a unique situation, a unique family and so on. If you're coming to me because I can find you a hot stock or a way to beat the market, forget it. That's not what I do. But I know how the markets work, and I can find the way to invest your money so that it fits your situation given what you can get in the capital markets." I call that investing. It's really important for everybody concerned to know where the investing is and where the betting is. I think most personal financial advisors would say their main function is the investing function, helping clients understand how to use the capital markets to fit their personal situation. That's a remarkably valuable function.

Gluck: Getting people to use Financial Engines has been a challenge, from what I can surmise, and you changed your model. Why?
Sharpe: We changed to a management model. With the original model, every employee with an account online could get his or her predictions. They could experiment with different portfolios, and we gave them our advice. We still do all that. For a certain part of the work force, this is great. They love it, they use it. But you sort of peak at some point. With some companies it's with 70% of the employees, and with other companies it's 20% of employees. We wanted to reach more than even 70%. We wanted to provide something that would work for everybody.
So our typical engagement now is, everybody gets on paper the personalized forecasts, recommendations, etc. at least once a year. Everybody has access to an online account of the traditional sort. And those who choose can just check a box and we will manage their account for them. It turns out the managed accounts are a major part of our business model. So we can reach everybody, one way or the other. We're much happier with that, and employers are much happier with that.

Gluck: In your book, you raise doubts going forward about value and small-cap stocks, and comment about Eugene Fama and Ken French's work. I thought it was interesting that you were taking that on and being out front with that.
Sharpe: Are there times when value really does better than growth? Yup. Are there times when growth does really better than value? Yup. So, are value and growth factors in terms of risks and correlations in the markets? You better believe it, and I did a paper on this myself many years ago. We've known this for many, many years. Barr Rosenberg, I think, probably was the first to really identify that. So are value-growth risk factors, is that a risk factor, is small-large a risk factor? Yup. Absolutely. So that's one element, and there's nothing inconsistent with the standard theories or new theories of that. The controversial part is somehow or other can you expect to do better with a value tilt or a small tilt than you "should."

Gluck: Permanently.
Sharpe: Yeah, is there some sort of a premium here, and in the future? Will there be, in the future, a premium for tilting towards value and/or tilting towards small? If the answer is yes, why would that be? There are two approaches. One is "Well, because you get the higher expected return from value, because there's something about value that a lot of people don't like." Like maybe if there's a real depression all those companies, or most of those companies, are going to disappear from sight, in which case it's a rational reward for a kind of risk that doesn't manifest itself very often and maybe hasn't been seen since the early '30s. That's one possibility. If so, fine.

The other alternative is, well, you know, people are dumb. The markets are screwed up. Value stocks are stocks, many of which have very low prices and they're too low, and ultimately the market will wake up and say, "No, no, we overdid it on that one," and the price will rise. So the question is whether this is a market inefficiency or just a manifestation of a risk, which you know is remote but there. And again, different people have different approaches. I'm very cautious about saying there is a free lunch out here. Here's a reward for which there's no commensurate risk. What I tried to point out in the book is if you look at the evidence from the Fama and French empirical studies, it's really evidence that comes from a very small part of the market. The way they did their studies, basically, most of that result comes from 10% or 5% of the value of securities out there. The idea that value investing is a free lunch, I would be reluctant to advise my clients to operate on that premise.