Value Works Because People Extrapolate: Contrarian Investment, Extrapolation, and Risk

By 1994 nobody seriously disputed the fact: cheap stocks beat expensive ones. Buy companies trading at low prices relative to their book value, earnings, or cash flow, and historically you did better than buying the glamorous, high-priced ones. Fama and French had just built a whole factor model around it.

The fight was about why. And the two answers had wildly different implications for whether you should actually do it.

• The risk story (Fama and French): value stocks earn more because they are genuinely more dangerous. They are distressed, fragile, and likely to blow up in a recession. The extra return is your fair wage for eating that danger. It is not a free lunch, it is hazard pay.
• The mistake story: value stocks earn more because investors systematically misjudge them. The premium is somebody else's error, and you get to collect it.

Lakonishok, Shleifer and Vishny came down hard on the second, and they did it by attacking the first on its own terms.

The problem: is value a risk premium or a bad habit?

This matters enormously in practice. If value is a risk premium, it should keep paying, but it will hurt you exactly when you can least afford it, in the bad states of the world. If value is an error, it should pay you without being especially dangerous, and it should slowly disappear as people wise up.

The debate was stuck because both theories predict the same headline fact (value wins). To break the tie you need a test where the two theories predict different things.

The key idea via analogy: the hot restaurant

Imagine a restaurant has three spectacular years. Lines around the block, glowing reviews. A developer looks at this and thinks: this place will be packed forever, and pays an enormous multiple for it.

But restaurants are competitive. Rivals copy the menu, the chef gets poached, the neighbourhood shifts. Almost every restaurant's growth reverts toward the average after a hot streak. The buyer paid for a straight line and got a curve.

Meanwhile down the street a tired old diner has had three terrible years. Everyone assumes it is dying and it sells for scrap value. But it is not dying, it is just having a rough patch, and rough patches also revert toward the average. The buyer paid for a funeral and got a modest recovery, which is a great deal.

The mistake in both cases is the same: extrapolation. People take the recent trend and project it forward far too confidently, forgetting that business performance mean-reverts. This is representativeness (from Tversky and Kahneman) applied to corporate earnings.

Lakonishok, Shleifer and Vishny turned this story into a series of tests, and this is where the paper gets clever. The heart of it: they classified stocks not just by how cheap they were, but by cheapness combined with recent growth. A stock that is cheap and has had poor recent growth is exactly the one that the extrapolation story says is most likely to be underpriced, because investors have written off a company that will in fact mean-revert. And that is exactly where the biggest returns showed up.

They then went hunting for the fingerprints of the mistake:

• Investors' expectations were demonstrably too extreme. The market was pricing glamour stocks as though their spectacular growth would continue, and value stocks as though their poor growth would continue. In reality, the growth rates of both groups converged sharply afterward. The forecast embedded in prices was simply wrong, in the direction extrapolation predicts.
• Value stocks did not behave like riskier assets. This is the direct assault on Fama and French. If value stocks are compensated for risk, they should do especially badly in bad times, when investors are hurting most. Lakonishok, Shleifer and Vishny looked at down markets, recessions, and the worst months for the market, and found that value stocks did not underperform in those states. If anything they held up fine. A "risk premium" that doesn't hurt you in bad times is not a risk premium. It is a gift.
• The strategy worked without extreme volatility. They found value's advantage was not accompanied by the kind of higher volatility or downside sensitivity that the risk story requires.

Why it mattered

• It gave value investing a psychological engine. Graham and Dodd said buy cheap. Fama and French said cheap is risky. Lakonishok, Shleifer and Vishny said cheap is mispriced, because human beings cannot stop drawing straight lines through the recent past. That is a mechanism, and mechanisms are what let you predict where else the effect should show up.
• It reframed value as a bet against extrapolation. This is the most useful practical takeaway. When you buy a value stock you are not buying "cheapness" in the abstract, you are betting that the market's forecast of this company's future growth is too pessimistic, and that mean reversion is stronger than the crowd believes. Frame it that way and you immediately know when value should fail: when the extrapolation turns out to be right, and the loser really does keep losing.
• It opened the second front of the great debate. This paper and Fama-French are the two poles of one of the most productive arguments in finance. Fama and French replied. LSV replied to the reply. The question is still not settled, and both sides have made everyone smarter.
• They put their money where their paper was. Lakonishok, Shleifer and Vishny founded LSV Asset Management, running quantitative value strategies on essentially this thesis. That is either admirable conviction or an enormous conflict of interest, and honestly it is both.

The honest limitations

• Proving the absence of risk is much harder than proving its presence. LSV showed value stocks don't underperform in the recessions and down markets in their sample. Fama and French's reply is essentially: you are measuring the wrong risk. Maybe value stocks are exposed to a distress factor, or a rare-disaster risk, that simply did not fire in the sample period. This is a frustrating, unfalsifiable-feeling defence, but it is not stupid: a risk that has not yet materialized still looks like free money right up until it does.
• Extrapolation is inferred, not observed. The paper infers that investors extrapolate from the fact that prices behave as if they do. It does not have a survey of investors saying "yes, I projected the last five years forward". Later work using analyst forecasts has supported the story, but the 1994 paper's evidence is circumstantial.
• The 2010s were brutal. Value endured a decade-plus of savage underperformance after this paper became famous. If value is a mistake, why didn't it correct quickly? If it's a risk premium, that decade is exactly the risk showing up. Awkwardly, the long drawdown is more comfortable for Fama and French's story than for LSV's.
• Crowding. LSV's own explanation contains the seed of its destruction. If value works because of an error, and you publish the error, the error should shrink. Value premia have been thinner since publication, which is precisely what their own theory predicts.

The one-line takeaway

Lakonishok, Shleifer and Vishny argued that value beats glamour because investors extrapolate recent growth far too confidently in both directions, and they showed that value stocks did not actually behave like riskier assets, which reframes value investing as a systematic bet against the human inability to expect mean reversion.