Chapter 1 lays out the two pillars to his argument: a theory of investor sentiment, and limits to arbitrage. Both are needed, as without systematic deviations from rationality, irrational decisions cancel each other out, making them somewhat uninteresting (excepting increasing volume); without limits to arbitrage such irrational deviations from `true value' are instantly poached by savvy rational investors.
The first model shows that if arbitrage is limited and noise traders have systematic biases, prices can deviate from fundamental value (DeLong et al, 1990a). That is, Shell can deviate from its fundamental value because no one has enough money and time to put the price into equilibrium, and investor sentiment varies between markets and over time. This model also has the interesting implication that less informed investors might earn a higher rate of return on their total portfolios because they irrationally believe they have a more favorable risk-return opportunity and hence invest in securities with a higher return. In effect, their stupidity effectively diminishes their risk aversion, and in the long run allows a lucky few of them to reap the financial rewards that would accrue to the less risk averse (one could call it the `Forrest Gump' effect). As opposed to speculation weeding out the irrational traders and making only the best opinions matter, the irrational can dominate.
The closed fund puzzle is presented in chapter three and highlights some of the problems of this approach (Lee, Thaler and Shleifer, 1991). The issue to be explained is why 1) funds are issued at premiums to net asset value (overpriced) and 2) funds eventually trade at discount to net asset value (underpriced). While the underpricing is addressed through the mechanism outlined in the first model (limited arbitrage and noise traders), the overpricing effect is addressed by assuming that `noise traders' buy up the initial issuance-not very subtle. Clearly if noise traders can be foisted into overpaying or underpaying within models by assumption, it is hard to avoid the inference that this approach can explain everything and thus nothing.
Of course on occasion EMH proponents also try to explain seemingly everything, where exceptions are assumed order-statistics until they are granted statistical significance, at which time they are instantly seen as an efficient proxy of unmeasured risk (e.g., the value and size effect). Yet explanatory greediness is clearly more of a problem to the inefficient markets camp. For example, in addition to the above example, investor sentiment is used to explain both the equity-premium puzzle (i.e., why stock prices are too low on average) and why recent p/e multiples are too high (page 180): if the equity-premium is a `puzzle' it is difficult to also say that our currently historically high p/e multiple is irrational, but if the p/e ratio `should' be lower (around 15) then the equity premium is not a puzzle.
In chapter 4 a nice approach is taken towards professional arbitrage (Shleifer and Vishny, 1997). By modeling it as a principal agent problem, this model captures some relevant issues usually addressed by the Industrial Organization literature. Clearly there is relevance to modeling the situation where hedge fund managers have uncertain skill and investors have to evaluate them. The failure of famed hedge fund LTCM in 1998 was defended, like almost all bankruptcies, as a failure of investor's patience--outsiders are always much quicker at pulling the plug than insiders would prefer. Modeling, in this case, a liquidity constraint, is a highly relevant issue that seems well suited for asymmetric information and principal-agent modeling.
Chapter 5 introduces the model of investor sentiment, that is, why we should expect noise traders to vary systematically in their buy or sell orders (Barberis, Shleifer, and Vishny, 1998). It derives a straightforward and testable hypothesis based on Bayesian updating of a regime-switching model. For earnings or other surprises that continue a trend, overreaction is predicted, for surprises that counter a trend, underreaction is predicted.
In chapter 6, we see the DeLong, Shleifer, Summers, and Waldeman series on noise traders and positive feedback loops (DSSW, 1990b). This sort of model bothers me because it is a bit disingenuous. It puts superficial rigor onto to the simple idea that "given constraints on arbitrage, irrational trend-following investors can make it rational to follow trends, and thus rational traders can be destabilizing." It is not a compelling model because the results are not derived inevitably and subtly from general assumptions and a friction, but instead from assumptions which guarantee the result (e.g., trend-following noise traders and limited arbitrage). Is it at all helpful to take a straightforward idea that can be clearly expressed in a sentence and model this with contrived algebra? Personally I do not think so, though the realist in me understands that without a model to point to, the idea would not be taken as seriously as it has.
It is stressed throughout the book that risky arbitrage makes taking advantage of pervasive irrationality difficult. Yet if irrationality is a systemic and pervasive phenomenon, then there exist hundreds of scenarios like mispriced Shell, overvalued Amazon, undepriced closed funds, overvalued currencies, overvalued IPOs, etc. Surely over several years these positions, somewhat independent, should make significant abnormal risk-adjusted returns. As is more probable, there are not hundreds of such situations, but perhaps a handful, and with the rational markets assumption ignored hundreds more opportunities appear to exist but actually do not (e.g., the small firm effect). Be sure to monitor Thaler's funds (UBRLX and UBVLX) and Sheifer's fund (LSVEX).