The risk aversion measure without the independence axiom

The risk premium (conveniently normalized) is defined as the measure of risk aversion. This measure does not require any relevant assumption in the theory of choice under uncertainty except the existence of a certainty equivalent. In particular, the independence axiom is not required. The measure of risk aversion of an action is provided not only for the case with one commodity and two consequences but also for the case with many commodities and consequences. The measure of mean risk aversion of all actions with given consequences is introduced and the local measure of risk aversion is obtained by making all these consequences approach the consequence under consideration. This measure is demonstrated to be zero when the von Neumann-Morgenstern utility function exists. In this case a measure of risk aversion of the second order is introduced, which turns out to be equal to the Arrow-Pratt absolute index when there is only one commodity and similar to the generalized measures proposed by several authors when there are many commodities and two consequences.Helpful comments by I. Gilboa and suggestions by the referee are gratefully acknowledged.     

On the definition of risk aversion

Two definitions of risk aversion have recently been proposed for non-expected utility theories of choice under uncertainty: the former refers the measure of risk aversion (Montesano 1985, 1986 and 1988) directly to the risk premium (i.e. to the difference between the expected value of the action under consideration and its certainty equivalent); the latter defines risk aversion as a decreasing preference for an increasing risk (introduced as mean preserving spreads) (Chew, Karni and Safra 1987, Machina 1987, Röell 1987, Yaari 1987).When the von Neumann-Morgenstern utility function exists both these definitions indicate an agent as a risk averter if his or her utility function is concave. Consequently, the two definitions are equivalent. However, they are no longer equivalent when the von Neumann-Morgenstern utility function does not exist and a non-expected utility theory is assumed. Examples can be given which show how the risk aversion of the one definition can coexist with the risk attraction of the other. Indeed the two definitions consider two different questions: the risk premium definition specifically concerns risk aversion, while the mean preserving spreads definition concerns the increasing (with risk) risk aversion.The mean preserving spreads definition of risk aversion, i.e. the increasing (with risk) risk aversion, requires a special kind of concavity for the preference function (that the derivatives with respect to probabilities are concave in the respective consequences). The risk premium definition of local risk aversion requires that the probability distribution dominates on the average the distribution of the derivatives of the preference function with respect to consequences. Besides, when the local measure of the first order is zero, there is risk aversion according to the measure of the second order if the preference function is concave with respect to consequences.Yaari's (1969) measure of risk aversion is closely linked to the r.p. measure of the second order. Its sign does not indicate risk aversion (if positive) or attraction (if negative) when the measure of the first order is not zero (i.e., in Yaari's language, when subjective odds differ from the market odds).     

Measures of risk aversion with expected and nonexpected utility

In the expected utility case, the risk-aversion measure is given by the Arrow-Pratt index. Three proposals of a risk-aversion measure for the nonexpected utility case are examined. The first one sets “the second derivative of the acceptance frontier as a measure of local risk aversion.” The second one takes into account the concavity in the consequences of the partial derivatives of the preference function with respect to probabilities. The third one measures risk aversion through the ratio between the risk premium and the standard deviation of the lottery. The third proposal catches the main feature of risk aversion, while the other two proposals are not always in accordance with the same crude definition of risk aversion, by which there is risk aversion when an agent prefers to get the expected value of a lottery rather than to participate in it.     

A strong (Ross) characterization of multivariate risk aversion

Using the addition of uncorrelated noise as a natural definition of increasing risk for multivariate lotteries, I interpret risk aversion as the willingness to pay a (possibly random) vector premium in exchange for a reduction in multivariate risk. If no restriction is placed on the sign of any co-ordinate of the vector premium then (as was the case in Kihlstrom and Mirman's (1974) analysis) only pairs of expected utility maximizers with thesame ordinal preferences for outcomes can be ranked in terms of their aversion to increasing risk. However, if we restrict the premium to be a non-negative random variable then comparisons of aversion to increasing risk may be possible between expected utility maximizers withdistinct ordinal preferences for outcomes. The relationship between their utility functions is precisely the multi-dimensional analog of Ross's (1981)global condition forstrongly more risk averse.     

On Bivariate Risk Premia

This note examines the conditions under which the bivariate risk premium for one risk may be negative even if both risks are positively correlated, using a mean variance setting. The link between the bivariate risk premium and the partial bivariate risk premia is also investigated.     

Normalized measures of concavity and Ross’s strongly more risk averse order

The Arrow-Pratt (A-P) definitions of absolute and relative risk aversion dominate the discussion of risk aversion and defining “more risk averse”. Ross (Econometrica 49:621–663, 1981) notes, however, that being A-P more risk averse is not sufficient for addressing many important comparative static questions. Consequently he introduces “a new and stronger measure for comparing two agents’ attitudes towards risk…”. Ross does not provide a corresponding measure of risk aversion. This paper uses a normalized measure of concavity to characterize the Ross definition of strongly more risk averse on bounded intervals. Other properties and uses of these normalized measures of concavity are also presented.     

Horizon Length and Portfolio Risk

In this paper, we compare the attitude towards current risk of two expected-utility-maximizing investors who are identical except that the first investor will live longer than the second one. It is often suggested that the young investor should take more risks than the old investor. We consider as a benchmark the case of complete markets with a zero risk-free rate. We show that a necessary and sufficient condition to assure that younger is riskier is that the Arrow-Pratt index of absolute tolerance (T) be convex. If we allow for a positive risk-free rate, the necessary and sufficient condition is T convex, plus T(0) = 0. It extends the well-known result that rational investors can behave myopically if and only if the utility function exhibits constant relative risk aversion.     

The Schwarzian derivative as a ranking of downside risk aversion

It is observed that the measure S _u  = u′′′/u′ − (3/2)(u′′/u′)², previously shown to be a relevant measure of the degree of downside risk aversion, is known in the mathematics literature as the Schwarzian derivative. The Schwarzian derivative has invariance properties under composition of functions that make it particularly well-behaved as a ranking of downside risk aversion. Indeed, it has the same invariance properties as the measure R _u  = −u′′/u′, familiar to economists as a ranking of utility functions by degree of Arrow-Pratt risk aversion.     

Total and partial Bivariate Risk Premia: An Extension

This paper investigates the link between the total bivariate risk premium and the sum of partial bivariate risk premia. Whereas in the case of small risks, the non interaction between risks is a sufficient condition to obtain the equality between the total risk premium and the sum of partial risk premia, the paper shows that this condition is not sufficient for large risks. The non interaction between risks occurs in two cases: if risks are independent or if individual's marginal utility of one good is independent of the endowment in the other. Without restriction on the utility function, none of these two conditions is sufficient for large risks. If attention is restricted to preferences that exhibit constant absolute risk aversion, the non variability of the marginal utility of good one with respect to variations in endowment in the other remains a sufficient condition, while the independence between risks does not.     

Ross' measure of risk aversion and portfolio selection

This article shows that if Ross' definition of riskier is replaced by a more traditional definition, such as a mean-preserving spread or second-degree stochastic dominance, then the application of Ross's stronger measure of risk aversion to the portfolio problem may no longer produce the desired result. It is also shown that the stronger measure may not perform satisfactorily when applied to exponential utility functions.The authors are grateful to John Pratt for his helpful comments.     

Proper and Standard Risk Aversion in Two-Moment Decision Models

For linear distribution classes, mean-variance and expected utility specifications have been shown in the literature to be fully compatible when studying the concepts of risk aversion, prudence, risk vulnerability and temperance. This paper shows that such compatibility does hold for the concept of standard risk aversion but not for the concepts of proper risk aversion and proper prudence.Jel Classification: D81     

Endogenous risk and protection premiums

Endogenous risk implies an individual perceives he can influence the likelihood that a state of nature will occur. To add structure to endogenous risk models, I define a protection premium for reduced uncertainty about protection efficiency when a stochastic variable enters the probability functionp(x) rather than the utility function. For a binary lottery, a measure of aversion of uncertain protection efficiency(x) =-p(x)/p(x) is defined to unambiguously determine the effects of increased risk on an individual's voluntary contribution to public good supply earmarked to reduce the probability of an undesirable state. Finally, I examine the protection premium in ann-state discrete lottery and when uncertainty exists in both the probability and utility function.     

Greater Downside Risk Aversion

Although investors are concerned foremost with mean and variance, they are also sensitive to downside risk. In this paper, we introduce an index of downside risk aversion to distinguish risk aversion from higher-order aspects of risk preference, including prudence. We show that the index of downside risk aversion S increases with monotonic downside risk averse transformations of utility, thereby directly linking S to the definition of downside risk aversion introduced by Menezes et al. (American Economic Review, 70, 921–932, 1980). Although the index S applies equally to risk averse and risk loving decision makers, for a given positive degree of risk aversion, S is greater when the index of prudence is greater and vice versa.     

Ambiguity Aversion in the Field of Insurance: Insurers’ Attitude to Imprecise and Conflicting Probability Estimates

This article presents the results of a survey designed to test, with economically sophisticated participants, Ellsberg’s ambiguity aversion hypothesis, and Smithson’s conflict aversion hypothesis. Based on an original sample of 78 professional actuaries (all members of the French Institute of Actuaries), this article provides empirical evidence that ambiguity (i.e. uncertainty about the probability) affect insurers’ decision on pricing insurance. It first reveals that premiums are significantly higher for risks when there is ambiguity regarding the probability of the loss. Second, it shows that insurers are sensitive to sources of ambiguity. The participants indeed, charged a higher premium when ambiguity came from conflict and disagreement regarding the probability of the loss than when ambiguity came from imprecision (imprecise forecast about the probability of the loss). This research thus documents the presence of both ambiguity aversion and conflict aversion in the field of insurance, and discuses economic and psychological rationales for the observed behaviours.     

Comparative Mixed Risk Aversion: Definition and Application to Self-Protection and Willingness to Pay

We analyze the optimal choices of agents with utility functions whose derivatives alternate in sign, an important class that includes most of the functions commonly used in economics and finance (Mixed Risk Aversion, MRA, Caballé and Pomansky, 1996). We propose a comparative mixed risk aversion definition for this class of utility functions, namely, More Risk Averse MRA, and provide a sufficient condition to compare individuals. We apply the model to optimal prevention and willingness to pay. More risk averse MRA agents spend less to reduce accident probabilities that are above 1/2. They spend more only when accident probabilities are below 1/2. Explanations in terms of risk premiums are provided. The results presented also allow for the presence of background risk.     

The valuation of contingent claims markets

This article studies an agent's valuation of the right to trade in a complete contingent claims market. The proposed measure generalizes the Pratt (1964) risk premium, which captures the willingness to pay to replace a given risky wealth prospect with an actuarially equivalent, nonrisky wealth. Specifically, we define ageneralized risk premium to be the willingness to pay to trade at going market prices. If state prices are actuarially fair, the Pratt premium is obtained as a special case. We derive several properties of this generalized premium and note its relationship to the option price of a public project under uncertainty.     

Risk and risk aversion for state-dependent utility

This paper analyses risk and risk aversion in the state-dependent utility model, which is useful for modelling health or life insurance purchase. We use Karni's (1983) definition of risk aversion, and extend the class of risks to which it can be applied.Research supported by the ESRC postdoctoral fellowship scheme. I would like to thank Jerry Nordquist for arousing my interest in this subject. For helpful comments on an earlier draft I am grateful to an anonymous referee and the editor of this journal.     

The preservation of multivariate comparative statics in nonexpected utility theory

This article investigates the preservation of multivariate expected utility comparative statics for “smooth” nonexpected utility representations. Specifically, we answer the following question: if an expected utility comparative statics property depends only on preferences over sure prospects, then when will a nonexpected utility maximizer with identical sure preferences also satisfy that property? We demonstrate that the effects of increased risk aversion are preserved under the “Almost Degenerate Independence” axiom, but that those of distribution changes of exogenous risks are not preserved under stringent assumptions. Hence, nonexpected utility comparative statics may diverge from expected utility, even for “first-order” properties—those whose effect is determinable from restrictions on “local” utility functions.     

Background Risks and the Value of a Statistical Life

We examine the effects of background mortality and financial risks on an individual's willingness to pay to reduce his mortality risk (the value of statistical life or VSL). Under reasonable assumptions about risk aversion and prudence with respect to wealth in the event of survival and with respect to bequests in the event of death, background mortality and financial risks decrease VSL. The effects of large mortality or financial risks on VSL can be substantial, but the effects of small background risks are negligible. These results suggest that the commonplace failure to account for background risk in evaluating VSL is unlikely to produce substantial bias in most applications.     

Uncertainty aversion and aversion to increasing uncertainty

According to the original Ellsberg (1961) examples there is uncertainty version if the decision maker prefers to bet on an urn of known composition rather than on an urn of unknown composition. According to another definition (Schmeidler, 1989), there is uncertainty aversion if any convex combination of two acts is preferred to the least favorable of these acts. We show that these two definitions differ: while the first one truly refers to uncertainty aversion, the second one refers to aversion to increasing uncertainty. Besides, with reference to Choquet Expected Utility theory, uncertainty aversion means that there exists the core of a capacity, while aversion to increasing uncertainty means that the capacity is convex. Consequently, aversion to increasing uncertainty implies uncertainty aversion, but the opposite does not hold. We also show that a completely analogous situation holds for the case of risk and we define a set of risk and uncertainty premiums according to the previous analysis.