Cracks in the edifice

By 1980 the building was finished. Markowitz had laid the foundation, Sharpe had drawn the plans, Fama had written the rules of the house, and Grossman and Stiglitz had explained the plumbing. Then, in the space of eight years, five papers found things the theory could not explain. The theory survived them, but it was never quite the same again, and the questions they raised are still open. Every serious investor has to decide what to do about each one. This is the chapter where the theory stops being clean.

Shiller, 1981

Robert Shiller asked a question nobody had thought to ask in that form. If a stock price is the market’s best guess of the dividends the company will pay in the future, then the price should not jump around more than those dividends actually do. Dividends are slow and smooth; they change a little each year. So prices should be fairly smooth too.

He took a century of American data, 1871 to 1979, and measured. Prices moved five to thirteen times more than the dividends they were supposed to be forecasting. Not a little more. An order of magnitude more. The market was not calmly predicting the future; it was swinging between moods.

That result split the profession in two, and the split has never healed. One camp said prices are irrational, and built behavioral finance. The other said the swings come from investors rationally demanding more or less return at different times, and built a theory of time-varying discount rates. They fight over the mechanism and agree on the fact, which is what matters to you: most of the ups and downs in a long-term investor’s results come from changes in the market’s mood, not from changes in what companies actually earn. And that has one consequence I have built my whole approach around. What you pay at the start decides most of what you get at the end. I never set a long-horizon expectation from historical averages; I set it from the valuation in front of me today.

Banz, 1981

The same year a doctoral student in Chicago, working under the people who had built the efficient markets theory, found something their model could not explain. Rolf Banz sorted stocks by size and found that the smallest companies had earned about half a percent a month more than their risk, as measured by beta, said they should, over forty years. The effect was concentrated in the very smallest firms, and Banz was careful to say he did not know why.

What Banz did was less important than how he did it. He sorted stocks on a characteristic, size, and looked for a return that beta could not account for. That recipe became the way empirical finance is produced to this day, and it has two faces. It is how markets confess the risks that beta hides. It is also how researchers with a computer and forty years of data turn noise into a “premium.” Small size itself has not held up well as a standalone premium; it survives mainly as a universe, the corner of the public markets where information is still expensive and where other things, value, momentum, quality, work better. When a bank shows you a new premium, remember Banz: ask whether it has held up out of sample, after the paper was published, after costs. Most do not.

De Bondt and Thaler, 1985

This was the first paper to take a psychological bias, make a prediction about the market from it, and test it. The bias is one Kahneman and Tversky had already documented: people extrapolate too far. If investors overreact, then the stocks that have done worst over the past three to five years should, on average, do better than the stocks that have done best. Werner De Bondt and Richard Thaler checked, and they did, by about twenty-five percent over the following three years. The paper was presented into open hostility. It became the empirical flagship of behavioral finance.

The practical version is uncomfortable. Three years of bad performance is, on average, not only a sign that something is wrong. It is also a sign that expected returns have gone up. Every family I know feels the urge to sell what has humiliated it for three years, and this paper is the standing rebuttal, worth rereading before any decision to give up on something. The way to use it is not to trust your nerve. It is to write the contrarian behavior into the rules in advance: rebalancing bands, scheduled additions to whatever has lagged for years, so that the committee’s instinct to run cannot veto the policy in real time. One thing not to confuse: this is about years, not months. Over three to five years, the losers tend to come back. Over the last few months, the opposite is true: what has just gone up tends to keep going up for a while. Both patterns are real, and they live at different time scales. Apply the multi-year rule to a stock that fell last quarter, or the short-term rule to one that has been falling for years, and you will get the worst of both.

Mehra and Prescott, 1985

Rajnish Mehra and Edward Prescott took the most important number in investing, the roughly six percent a year that American stocks had earned above safe bonds for a century, and asked whether the standard theory could explain it. It could not. Not by a little. For a rational investor to demand that much extra return for the risk of stocks, given how smooth consumption actually is, she would have to be thirty or forty times more afraid of risk than any reasonable estimate. The premium was a puzzle.

The puzzle was never fully solved, and the honest position is that several things contribute: disasters that did not happen in the American sample, habits, survivorship. The United States did not lose a world war on its own soil; its stock market history is partly the history of a winner. That is why the lesson of this paper is about planning. If part of the historical premium was luck, then a plan built on the historical premium is built on luck too. I set expected returns two or three points below history, run every long-term calculation at that number, and treat anything above it as a gift. And there is a second, quieter lesson: if the premium really does overpay for the risk, then the investor who can actually bear the risk, who has a long horizon and no one to answer to, is structurally advantaged. That investor can be a family. It is rarely an institution.

Campbell and Shiller, 1988

Shiller came back seven years later with John Campbell and turned the problem into a tool. Start with a simple question: when stocks are cheap, meaning they pay a high dividend for every dollar of price, why are they cheap? There are only two possible reasons. Either the market expects the companies to pay less in the future, so it is right to pay less for them now. Or the market is simply demanding a higher return for holding them, and the price will rise as that return is delivered. Cheap has to mean one or the other. There is no third possibility.

So they went to a century of data to see which one it was. The answer was clear. When stocks were cheap, the companies’ dividends did not grow any slower than usual. What happened instead was that the following ten to fifteen years of returns were higher than usual. When stocks were expensive, the opposite. In other words, the price you pay today does not tell you much about how the businesses will do; it tells you, roughly but reliably, how much you will earn from owning them over the next decade or more.

This is the paper behind every valuation-based expected return, including the ratio that now carries Shiller’s name. Its practical content is one sentence, and it is the sentence I use when a family asks the only question families actually ask: what will this portfolio earn over our planning horizon? Starting yield is destiny at long horizons. It tells you almost nothing about next year, which is why it is not a timing tool, and why low expected returns can persist for years before they matter. But over fifteen years, what you paid is most of what you will get. The exercise from my own reading: build a one-page sheet for each part of your portfolio with the current yield, a trend growth rate, and an allowance for valuation returning toward normal over fifteen years. Compare it with what your bank assumes. Every difference of more than a point deserves a written explanation.

What the cracks teach

Valuation at entry dominates long-run results. Set forward returns conservatively, below history, and treat the rest as luck. Write the rule for buying what has fallen into your plan in advance, because when the moment comes, you will not feel like doing it.

Gala: before you buy anything for the long run, check its price against its own history. Expensive, buy less and expect less. Cheap and not broken, buy more than feels comfortable.