Measuring Risk Aversion
本书指出，风险厌恶估计结果不一致的一个被忽视的原因是结果变量（如财富、收入、消费）不同；若能建立这些变量间的函数关系，则可将不同研究的估计值转换到共同基础上，从而消除大部分分歧。
Measuring Risk Aversion, by Donald J. Meyer and Jack Meyer, 2006, NOW Publishers, Inc., 101 pages The empirical literature on risk aversion contains a vast array of estimates undertaken in various contexts, with different types of data, different populations, different models and assumptions regarding utility, and different estimation techniques. The contexts for studying risk have included agricultural production, insurance purchases, financial investments, gambling, and intertemporal consumption, among others; the decision makers in these studies have included college students, working adults, retirees, citizens of less developed countries, and those in developed economies; and cross-sections, time series, laboratory experiments, and survey responses have all been used as data. As a consequence, the results have been quite diverse. Some studies have estimated relative risk aversion to be less than 1 (Hansen and Singleton, 1982), others obtain values that are generally between 1 and 10 (Szpiro, 1986a,1986b; Eisenhauer and Ventura, 2003), still others conclude that relative risk aversion must be well in excess of 10 in order to explain observed behavior (Mehra and Prescott, 1985), and outliers have been reported in the hundreds (Halek and Eisenhauer, 2001). Conflicting results can also be found concerning the slope of the risk aversion function-that is, whether risk aversion increases, remains constant, or decreases as the argument of the utility function rises. Blake (1996) and Levy (1994), for example, found evidence of decreasing relative risk aversion (DRRA), while Eisenhauer and Ventura (2003) reported increasing relative risk aversion (IRRA). A number of observers have been disconcerted by lack of consistency across studies. Mehra and Prescott (1985) and Kocherlakota (1996), finding that stock market investments among households have historically been far too limited to be consistent with widely reported relative risk aversion values below 10, refer to the discrepancy as the equity premium puzzle. Similarly, Gollier (2001, pp. 424-425) has remarked, It is quite surprising and disappointing for me that almost 40 years after the establishment of the concept of risk aversion by Pratt and Arrow, our profession has not been able to attain a consensus about the measurement of risk aversion. Indeed, this confusion is viewed by some as so severe that the existing body of evidence is dismissed in favor of introspection (Meyer and Meyer, 2005, p. 244). In Measuring Risk Aversion, Donald J. Meyer and Jack Meyer point out a previously unrecognized reason for some of the discrepancy: differences in outcome variables, or the arguments of the utility function. Economic models have variously defined utility over wealth, income, consumption, rates of return, profit, payoffs from gambling, and other variables. Moreover, even when two studies appear to apply the same outcome variable, they often define it in different ways-wealth, for example, may be measured more or less broadly by the inclusion or exclusion of components such as human capital. With that as a background, the book's central theme is twofold: neither the magnitude of the risk aversion coefficient nor the slope of the risk aversion function can be compared directly across studies when outcome variables differ, but if there exists a known functional relationship between the outcome variables, then the corresponding risk aversion functions and coefficients can be made comparable. The authors conclude that when existing estimates of risk aversion are converted to a common basis, much of the reported discrepancy disappears. The theorem provides an intriguing and useful insight for cases in which outcome variables can be linked in manner. Of course, not all outcome variables that might be of interest necessarily have either linear or positive relationships, especially when the relationships are behavioral. Consumption may not be linear in income, for example, and simple labor supply models suggest that an increase in exogenous wealth tends to reduce an individual's optimal hours of work, reducing earned income for any given hourly wage. …