State-Dependent Preferences and Insurance Demand∗ Robert Kremslehner† Alexander Muermann‡ July 2014 We compare the demand for insurance under state-dependent and state-independent preferences when allowing for positive premium loading and continuous state space. We show that monotonicity of marginal utility across states is not sufficient to yield unambiguous comparative results–as it is under a fair premium. Nevertheless, unambiguous results can be derived for a specified set of state-independent preferences if both marginal utility and absolute degree of risk aversion are monotonic across states. ∗ We thank Georges Dionne, Meglena Jeleva, Christophe Courbage, and seminar participants at the EGRIE meeting for valuable comments and discussion. † Department of Finance, Accounting and Statistics, Vienna University of Economics and Business, Welthandelsplatz 1, Building D4, A - 1020 Vienna, AUSTRIA, E-mail: [email protected] ‡ Department of Finance, Accounting and Statistics, Vienna University of Economics and Business, and VGSF (Vienna Graduate School of Finance), Welthandelsplatz 1, Building D4, A - 1020 Vienna, AUSTRIA, E-mail: [email protected] 1 1 Introduction In von Neumann and Morgenstern (1944) expected utility theory, preferences of individuals are characterized through a state-independent utility function, i.e. the utility in a state depends only on the consumption in that state but not directly on the state itself. In contrast, Eisner and Strotz (1961) investigate the demand for flight insurance and argue that the utility function of an individual is likely to be different when he is alive compared to when he is dead. Similarly, Arrow (1973) emphasizes that individuals who fell ill may extract less utility from spending money on vacation. Individuals’ preferences therefore directly depend on the state of health. Cook and Graham (1977) apply state-dependent preferences to examine the optimal insurance demand for irreplaceable commodities, such as family heirlooms or a photo album. They show that less (more) than full coverage of the financial loss is optimal if the irreplaceable commodity is normal (inferior). Dionne (1982) and Schlesinger (1984) show that the optimal amount of insurance under state-dependent preferences can be inferred from changes of marginal utility of consumption across states. More specifically, if insurance is actuarially fair then an individual with state-dependent preferences optimally purchases less (more) than full insurance coverage if his marginal utility of consumption decreases (increases) as the irreplaceable commodity is lost.1,2 Furthermore, Karni (1985) and Nordquist (1985) show that the optimal insurance demand under state-dependent preferences is increasing in the risk premium. In the context of nonpecuniary background risk, e.g. health, Rey (2003) determines conditions on the sign of the correlation between insurable losses and background risk such that the results of Dionne (1982) and Schlesinger (1984) hold. For example, if marginal utility of consumption decreases as health deteriorates and health is positively correlated with wealth, then less than full insurance is optimal. There exist few papers that empirically investigate how marginal utility of consumption changes across states of health. Viscusi and Evans (1990) estimate changes in marginal utility of consumption between good and bad health using a survey on pay raise that workers demand for taking on additional risk at work. They conclude that marginal utility is statistically significantly lower if workers are severely injured.3 However, for minor health losses Evans and Viscusi (1991) do not find significant evidence that marginal utility is affected by health status. In a recent 1 Mossin (1968) shows that risk-averse individuals with state-independent preferences optimally purchase full coverage at an actuarially fair premium rate. 2 Eeckhoudt et al. (2007) call individuals whose marginal utility of consumption increases as health deteriorates correlation averse. 3 Frech III (1994) argues that the tort system and strict liability might cause inefficiencies if marginal utility of consumers is lower in the accident state. 2 paper, Finkelstein et al. (2013) investigate the impact of chronical diseases on the marginal utility of consumption. They find that marginal utility declines by about 10 - 25 percent as health deteriorates, as measured by a one standard deviation increase in the number of chronic diseases. The authors further argue that the estimated 10 - 25 percent decline in marginal utility lowers the optimal amount of medical insurance by about 20 to 45 percentage points compared to the optimal amount of medical insurance under state-independent preferences.4 In this paper, we compare optimal insurance demand between state-independent and statedependent preferences. All literature mentioned above and related to this objective is based on the assumption that insurance is priced at an actuarially fair rate and that individuals face a binary distribution, e.g. staying healthy or falling sick with corresponding medical cost. Under these two assumptions, the sign of the change in marginal utility of consumption as health deteriorates is the only empirically important factor that impacts the demand for insurance. We contribute to the literature by analyzing the demand for insurance under state-dependent preferences for positive loading factors and continuously distributed loss distributions. We find that whether marginal utility of consumption decreases (increases) as health deteriorates is not sufficient to guarantee that individuals with state-dependent preferences purchase less (more) insurance than individuals with state-independent preferences–as it does under a fair premium and binary distribution. An additional property of preferences that is crucial for such unambiguous statements is how the absolute degree of risk aversion of consumption changes as health deteriorates. More specifically, for a fixed loading factor we specify state-independent preferences such that an individual with such preferences purchases more (less) insurance at this loading factor than an individual with state-dependent preferences if marginal utility of consumption decreases (increases) as health deteriorates. If the degree of risk aversion is monotonic in health, then we derive a bound for the degree of risk aversion such that individuals with state-independent preferences and degrees of risk aversion that are larger (smaller) than this bound purchase more (less) insurance at all loading factors than an individual with state-dependent preferences. This result shows that it is relevant not only how marginal utility but also how the degree of risk aversion changes across states of health. The article is structured as follows. In Section 2, we extend the binary insurance model to positive loading factors. In Section 3, we further extend the model to continuous distributions. 4 Finkelstein et al. (2013) derive the optimal amount of insurance by calibrating their two-period model with Epstein-Zin preferences (Epstein and Zin (1989)), a binary distribution of health, and actuarially fair premium. 3 We conclude in Section 4. All proofs are in the Appendix. 2 State-Dependent Preferences: Two-State Space Consider an individual whose preferences depend on wealth ω and health H and are described by a bi-variate utility function U (ω, H).5 We assume that the individual’s utility function is increasing in both wealth and health and concave in wealth, i.e. U1 > 0, U2 > 0, and U11 < 0.6 Let ω0 and H0 = 1 denote the initial levels of wealth and health. The individual faces the risk of falling ill with probability 0 < p < 1. In case of sickness he incurs medical treatment costs L ≥ 0 and his health state is reduced to H = 0. An insurance company offers medical insurance coverage I at a premium P (I) = (1 + λ)pI where λ denotes the proportional loading factor. ∗ The optimal level of insurance coverage ISDEU (λ) that maximizes his state-dependent expected utility (SDEU) is determined by the following optimization problem max SDEU (I) = max E[U (ω(I, H̃), H̃)], I I (1) where ω (I, 1) = ω0 − P and ω (I, 0) = ω0 − P − L + I. For an actuarially fair premium (λ = 0), Dionne (1982) and Schlesinger (1984) show that the optimal amount of insurance coverage for individuals with state-dependent preferences relative to one implied by state-independent preferences depends on whether marginal utility of wealth increases or decreases in health, i.e. on the sign of U12 . If marginal utility of wealth decreases (increases) as health deteriorates U12 > (<) 0 less (more) than full coverage is optimal. An individual with state-dependent preferences therefore purchases less (more) insurance than any risk-averse individual with state-independent preferences (see Mossin (1968)). In this section, we extend this comparative result for positive loading factors. For λ = 0, full insurance is optimal for any risk-averse individual with state-independent preferences. It is thus possible to compare the optimal insurance demand between state-dependent preferences and all state-independent preferences that exhibit risk aversion. In contrast, for λ > 0, the optimal ∗ (λ), depends on the degree of amount of insurance under state-independent preferences, IEU risk aversion, i.e. on the specific utility function U . When comparing the optimal amount of 5 6 The interpretation of H as the state of health is arbitrary. Other interpretations are possible. We use the common notation Ui for the partial derivative of U with respect to the i’s argument, e.g. U1 = ∂U and U2 = ∂H . 4 ∂U ∂ω ∗ insurance under state-dependent preferences, ISDEU (λ), with the optimal insurance under state∗ independent preferences, it seems natural in a first step to compare ISDEU (λ) against the optimal ∗ ∗ amounts of insurance, IEU(0) (λ) and IEU(1) (λ), implied by the two state-independent utility functions U (·, 0) and U (·, 1) that form the basis for the state-dependent preferences. In the following proposition, we show that the result of Dionne (1982) and Schlesinger (1984) can be extended to positive loading factors when comparing against these two specific state-independent preferences. From hereon, we focus on the case in which marginal utility of wealth is increasing in health, i.e. U12 > 0.7 Proposition 1. Suppose U12 > 0. An individual with state-dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than an individual with stateindependent preferences characterized by either utility function U ∈ {U (·, 0), U (·, 1)}. That is n o ∗ ∗ ∗ (λ) < min IEU(0) (λ) , IEU(1) (λ) ISDEU for all λ ≥ 0. Figure 1 illustrates the result of Proposition 1. The plot shows the optimal insurance coverage I ∗ for state-dependent and state-independent preferences as function of the loading factor λ. I∗ ∗ ISDEU (λ) 1 ∗ (λ) IEU(0) ∗ IEU(1) (λ) λ 0 Figure 1: Comparison of optimal insurance coverage of individuals with state-dependent and state-independent preferences at different loading factors λ. Since the optimal amount of insurance of EU-maximizers8 is increasing in the degree of risk 7 U12 > 0 is consistent with the empirical evidence of Viscusi and Evans (1990) and Finkelstein et al. (2013) and imply correlation loving preferences between wealth and health (see Eeckhoudt et al. (2007)). We obtain analogous results for U12 < 0 which is covered in the appendix. 8 We use the term “SDEU-maximizers” for individuals with state-dependent preference and “EU-maximizers” for individuals with state-independent preferences. 5 aversion, we can extend the result obtained above from the comparison set of the two stateindependent utility functions U (·, 0) and U (·, 1) to the set of all state-independent utility functions which exhibit degrees of risk aversion that are larger than the one implied by either U (·, 0) or U (·, 1). Corollary 2. Suppose U12 > 0. An individual with state-dependent preferences optimally chooses less insurance coverage for all loading factors λ than any individual with state-independent preferences whose degree of risk aversion is higher at all levels of wealth ω than the degree of risk aversion implied by either of the two utility functions U ∈ {U (·, 0), U (·, 1)}. That is ∗ ∗ ISDEU (λ) < IEU (λ) U (0) for all λ ≥ 0 and U with RaU (ω) ≥ Ra U (1) (ω) for all ω or RaU (ω) ≥ Ra (ω) for all ω where RaU (ω) denotes the coefficient of absolute risk aversion in wealth under the state-independent utility function U and U (0) and U (1) denote the utility functions U (·, 0) and U (·, 1), respectively.9 In this section, we have shown a SDEU-maximizer whose marginal utility of wealth decreases as health deteriorates, i.e. U12 > 0, optimally purchases less insurance coverage at all loading factors λ than all EU-maximizers whose degree of risk aversion is globally greater than the one implied by either U (·, 0) or U (·, 1).10 3 State-Dependent Preferences: Continuous State Space In this section, we examine the case in which both health status H̃ and medical cost L̃ are continuously distributed random variables. As in the previous section, we associate a deterioration in health with an increase in medical costs. More specifically, we define the function h← : H → L that assigns to each health status H ∈ H a loss of size L ∈ L and assume that h← is strictly decreasing and left continuous.11 Furthermore, let h← be the generalized inverse of function h which is continuous and decreasing due to the properties of h← . We additionally assume that h is non-constant a.s. and define the individuals utility function as U (ω, H = h(L)) where the medical 9 Risk aversion in wealth is meant in the sense of Pratt (1964) and Arrow (1971). Equivalently, correlation averse SDEU maximizers, i.e. with U12 < 0, optimally purchase more insurance coverage at all loading factors λ than all EU-maximizers whose degree of risk aversion is smaller than the one implied by either U (·, 0) or U (·, 1). 11 H and L are thus the sets of all realizations of health status H and medical cost L, respectively. 10 6 treatment cost L̃ is distributed according to a cumulative distribution function F : [0, L̄] → R+ with F (0) = 0 and F (L̄) = 1, where L̄ denotes the maximum loss.12 The individual is offered a coinsurance contract, i.e. the indemnity schedule I : [0, L̄] → R+ is given by I(L) ≡ αL where L is a realization of L̃ and α is the coinsurance rate. The premium is determined by Pα = α P (λ) with a loading factor λ ≥ 0 where P (λ) = (1 + λ)E[L̃]. The ∗ individual chooses the optimal level of insurance coverage αSDEU (λ) by maximizing expected state-dependent utility, i.e. max SDEU (α) = max E[U (ω(α, L̃), h(L̃))], α α (2) where ω(α, L) = ω0 − (1 + λ)αE[L̃] − (1 − α)L. In the following proposition we show that the result of Dionne (1982) and Schlesinger (1984) under their assumption that the premium is actuarially fair, i.e. λ = 0, generalizes to the continuous state space. Proposition 3. Suppose U12 > 0 and λ = 0. Then an individual with state-dependent preferences optimally chooses less than full insurance coverage, i.e. ∗ αSDEU (λ = 0) < 1. Next, for positive loading factors λ ≥ 0, we compare the insurance purchase behavior of SDEUmaximizers with the one of EU-maximizers whose preferences are characterized by the stateindependent utility functions of the set U ≡ {U (·, H) : H ∈ H} that constitutes the statedependent preferences. For the two state case, we have shown in Proposition 1 that an individual with state-dependent preferences optimally purchases less insurance coverage for all loading factors λ than an individual with state-independent utility function U for all U ∈ U. This result does not generalize to a continuous state space.13 Nevertheless, we show in the following proposition that for each loading factor λ there exists a state-independent utility function U (λ) ∈ U such that a SDEU-maximizer purchases less insurance at this loading factor λ than an EUmaximizer with utility function U (λ). 12 Health status H̃ and medical cost L̃ are thus assumed to be countermonotonic. We note that this assumption does not necessarily imply that H̃ and L̃ are perfectly negatively correlated as measured by linear correlation. 13 We provide an example below to illustrate that an EU-maximizer might purchase less insurance even if marginal utility is decreasing as health deteriorates. 7 Proposition 4. Suppose U12 > 0. For each loading factor λ ≥ 0, we define the state-independent utility function U (λ) ≡ U (·, H̄(λ)) ∈ U where H̄(λ) ≡ h((1 + λ)E[L̃]). Then for each loading factor λ ≥ 0, the SDEU-maximizer optimally purchases less insurance than an EU-maximizer with utility function U (λ), i.e. ∗ ∗ αSDEU (λ) < αEU( H̄(λ)) (λ) . Proposition 4 implies that the higher the loading factor λ the worse the necessary fixed health status H̄(λ) to ensure that an EU-maximizer with utility function U (·, H̄(λ)) purchases more insurance than the SDEU-maximizer. As in the case with two states, the degree of risk aversion can be used to make further statements. First, suppose that the degree of risk aversion of EU-maximizers with utility function U (·, H) ∈ U is independent of the fixed health status H, i.e. ∂ a ∂H (RU (·,H) (ω)) = 0 for all ω.14 The optimal ∗ amount of insurance αEU(H) is therefore also independent of the fixed health status H. In this case, the result of Proposition 1 generalizes to the continuous state space. Corollary 5. Suppose U12 > 0 and ∂ a ∂H (RU (H) (ω)) = 0 for all ω. Then an individual with state- dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than an individual with state-independent preferences characterized by any utility function U (·, H) ∈ U, i.e. ∗ ∗ (λ) < αEU(H) (λ) αSDEU for all λ ≥ 0 and U (·, H) ∈ U. Furthermore, an individual with state-dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than any EU-maximizing individual who exhibits a degree of risk aversion which, at all levels of wealth, is larger than a RU (H) , i.e. ∗ ∗ αSDEU (λ) < αEU (λ) a (ω) ≥ Ra for all λ ≥ 0 and U with RU U (H) (ω) for all ω. This result is illustrated in Figure 2 which depicts the optimal coinsurance rates α∗ as function of the loading factor λ. For the two loading factors λ1 < λ2 , the two utility functions U (·, H̄(λ1 )) and U (·, H̄(λ2 )) exhibit identical degrees of risk aversion. The optimal amount of insurance of 14 The utility function used in Finkelstein et al. (2013) satisfies this property. 8 an EU-maximizer with utility function U (·, H̄(λ1 )) thus coincides with the optimal amount of insurance of an EU-maximizer with utility function U (·, H̄(λ2 )) for all λ ≥ 0. Since this holds for any health status H̄(λ), Proposition 4 implies Corollary 5. Note that the result that statedependence implies lower insurance coverage as derived in ? only extends to positive loading factors under the condition that the degree of risk aversion is independent of the health status. α∗ ∗ αSDEU (λ) 1 ∗ αEU λ, H̄ (λ) 0 ∗ αEU λ, H̄ (λ1 ) ∗ αEU λ, H̄ (λ2 ) λ1 λ2 λ Figure 2: Comparison of optimal insurance coverage of individuals with state-dependent and state-independent preferences at different loading factors λ if risk aversion is unaffected by health state. Now suppose that the degree of risk aversion of EU-maximizers with utility functions U (·, H) ∈ U increases as health deteriorates, i.e. ∂ a ∂H (RU (H) (ω)) < 0 for all ω. Then the optimal amount of ∗ insurance αEU( is increasing in the loading factor λ. Proposition 4 then implies the following H̄(λ)) corollary. Corollary 6. Suppose U12 > 0 and ∂ a ∂H (RU (H) (ω)) < 0 for all ω. Then an individual with state-dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than an individual with state-independent preferences characterized by the utility function U (·, H̄(λmax )) ∈ U, i.e. ∗ ∗ αSDEU (λ) < αEU( H̄(λmax )) (λ) ∗ for all λ ≥ 0 where λmax is defined through αEU( (λmax ) = 0. Furthermore, an individual H̄(λmax )) with state-dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than any EU-maximizing individual who exhibits a degree of risk aversion which, at all a levels of wealth, is larger than RU , i.e. (H̄(λmax )) ∗ ∗ αSDEU (λ) < αEU (λ) 9 a (ω) ≥ Ra for all λ ≥ 0 and U with RU (ω) for all ω. U (H̄(λmax )) This result is illustrated in Figure 3. α∗ ∗ αSDEU (λ) 1 ∗ αEU λ, H̄ (λ) ∗ αEU λ, H̄ (λmax ) ∗ αEU λ, H̄ (λ1 ) ∗ αEU λ, H̄ (0) 0 λmax λ1 λ Figure 3: Comparison of optimal insurance coverage of individuals with state-dependent and state-independent preferences at different loading factors λ if risk aversion increases as health deteriorates. ∗ (λ) of the SDEU-maximizer as a The thick solid line plots the optimal coinsurance rate αSDEU ∗ (λ, H̄(λ)) of an EU-maximizer as a function function of λ. The optimal coinsurance rate αEU of λ is depicted by the dashed line for all 0 ≤ λ ≤ λmax . The three thin solid lines plot ∗ (λ, H̄) as a function of λ for the three fixed levels of health the optimal coinsurance rates αEU status H̄ = H̄(0), H̄(λ1 ), and H̄(λmax ) for all 0 ≤ λ ≤ λmax . Note that for each fixed loading ∗ (·, H̄(·)) intersects with the thin solid line α∗ (·, H̄(λ)). Since the factor λ, the dashed line αEU EU ∗ (λ, H̄(0)) ≤ α∗ (λ, H̄(λ )) ≤ degree of risk aversion increases as health deteriorates, we have αEU 1 EU ∗ (λ, H̄(λmax )) and α∗ ∗ max . αEU SDEU (λ) < αEU(H̄(λmax )) (λ) for all 0 ≤ λ ≤ λ We obtain the analogous result if the degree of risk aversion of EU-maximizers with utility functions U (·, H) ∈ U decreases as health deteriorates, i.e. Corollary 7. Suppose U12 > 0 and ∂ a ∂H (RU (H) (ω)) ∂ a ∂H (RU (H) (ω)) > 0 for all ω. > 0 for all ω. Then an individual with state- dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than an individual with state-independent preferences characterized by the utility function U (·, H̄(0)) ∈ U, i.e. ∗ ∗ αSDEU (λ) < αEU( H̄(0)) (λ) for all λ ≥ 0. Furthermore, an individual with state-dependent preferences optimally chooses less insurance coverage for all loading factors λ ≥ 0 than any EU-maximizing individual who exhibits 10 a a degree of risk aversion which, at all levels of wealth, is larger than RU , i.e. (H̄(0)) ∗ ∗ αSDEU (λ) < αEU (λ) a (ω) ≥ Ra for all λ ≥ 0 and U with RU (ω) for all ω. U (H̄(0)) This result is illustrated in Figure 4. α∗ ∗ αSDEU (λ) 1 ∗ αEU λ, H̄ (λ) ∗ αEU λ, H̄ (0) ∗ αEU λ, H̄ (λ1 ) ∗ αEU λ, H̄ (λmax ) 0 λmax λ1 λ Figure 4: Comparison of optimal insurance coverage of individuals with state-dependent and state-independent preferences at different loading factors λ if risk aversion decreases as health deteriorates. Example. In this example, we illustrate that an EU-maximizer might purchase less insurance even if marginal utility is decreasing as health deteriorates. Suppose that the set of utility 1 functions is defined as U (ω, H) = −e− 9 H ω . Note that this utility function satisfies the conditions of Corollary 7, i.e. marginal utility is increasing in health for all H and ω such that H ω < 9 and the degree of risk aversion is increasing in health. Let ω0 = 3 and suppose that the density distribution function of the loss is given by f (L) = 10 1−e−10 e−10 L , with L ∈ [0, 1]. Further, suppose that the mapping from losses to health levels is given by h(L) = 2.9 − L. Figure 5 plots the optimal amount for insurance for the SDEU-maximizer (thick line) and for EU maximizers with fixed health status H = 2.9, H = H̄(0) ≈ 2.8, and H = 1.9 (thin lines). We observe that EU-maximizers with fixed health status H = 2.9 and H = H̄(0) purchase more insurance at all loading factors than the SDEU-maximizer. In contrast, the EU-maximizers with fixed health status H = 1.9 purchase less insurance at high loading factors than the SDEUmaximizer. For λ = 0.02, the optimal amount of insurance for the SDEU-maximizer is given by ∗ αSDEU (λ = 0.02) = 0.1464. For EU-maximizers with fixed health status H = 2.9, H = H̄(0) ≈ ∗ 2.8, and H = 1.9, the optimal amounts of insurance are given by αEU(H=2.9) (λ = 0.02) = 0.3888, 11 ∗ ∗ αEU(H= (λ = 0.02) = 0.3670, and αEU(H=1.9) (λ = 0.02) = 0.0671, respectively. H̄(0)) α∗ ∗ αSDEU (λ) 1 ∗ αEU (λ, H = 2.9) 0.75 ∗ αEU λ, H = H̄(0) ∗ αEU (λ, H = 1.9) 0.5 0.25 0 0 0.01 0.02 0.03 λ Figure 5: Illustration of the example: an individual with preferences characterized by a utility function U (·, H = 1.9) optimally chooses less insurance coverage for high loading factors than an individual with state-dependent preferences. 4 Conclusion We analyze the impact of state-dependent preferences on insurance demand when allowing for positive loading factors and continuously distributed states. We find that the results do not only depend on how marginal utility of consumption changes in health–as pointed out by Dionne (1982) and Schlesinger (1984) under fair premium rates and binary distribution–but also on other characteristics of preferences such as how absolute degree of risk aversion changes in health. This has important implications for the conclusions that can be drawn from empirical research on state-dependent preferences. If insurance includes a loading, then the decline of marginal utility does not unambiguously imply a decline in the optimal amount of insurance. For such conclusions it is necessary to jointly estimate how the degree of risk aversion changes as health deteriorates. 12 Appendix - Proofs Proof of Proposition 1. We face the maximization problem stated in equation 1 max SDEU (I) with I h i SDEU (I) = E U (ω(I, H̃), H̃) = pU (ω (I, 0) , 0) + (1 − p) U (ω (I, 1) , 1) , where H = 0 denotes bad health and H = 1 denotes good health. Final wealth levels in bad and good health are given by ω (I, 0) = ω0 − P − L + I and ω (I, 1) = ω0 − P , respectively, where P = (1 + λ) pI is the premium at loading factor λ. The FOC is given by SDEU0 (I) = p (1 − (1 + λ) p) U1 (ω (I, 0) , 0) − (1 − p) (1 + λ) pU1 (ω (I, 1) , 1) = 0. The SOC SDEU00 (I) = p (1 − (1 + λ) p)2 U11 (ω (I, 0) , 0) + (1 − p) (1 + λ)2 p2 U11 (ω (I, 1) , 1) < 0 ∗ ∗ is satisfied. Thus, the unique global maximum ISDEU (λ) for all λ is determined by SDEU0 (ISDEU (λ)) = 0. Now we compare insurance purchase behavior of state-dependent preferences with state-independent preferences. The FOC for individuals with preferences characterized by U (ω, H̄) with fixed H̄ is given by EU0U (·,H̄) (I) = p (1 − (1 + λ) p) U1 (ω (I, 0) , H̄) − (1 − p) (1 + λ) pU1 (ω (I, 1) , H̄) = 0. ∗ The SOC is satisfied. The unique global maximum IEU( (λ) is therefore determined by H̄) ∗ EU0U (·,H̄) (IEU( H̄) (λ)) = 0. ∗ ∗ Evaluation of the FOC for ISDEU (λ) at I = IEU( (λ) yields H̄) ∗ ∗ SDEU0 (IEU( H̄) (λ)) = p (1 − (1 + λ) p) U1 (ω(IEU(H̄) (λ) , 0), 0) ∗ − (1 − p) (1 + λ) pU1 (ω(IEU( H̄) (λ) , 1), 1). ∗ IEU( (λ) satisfies H̄) ∗ ∗ EU0U (·,H̄) (IEU( H̄) (λ)) = p (1 − (1 + λ) p) U1 (ω(IEU(H̄) (λ) , 0), H̄) ∗ − (1 − p) (1 + λ) pU1 (ω(IEU( H̄) (λ) , 1), H̄) = 0. Substitution yields ∗ ∗ SDEU0 (IEU(1) (λ)) = p (1 − (1 + λ) p) U1 (ω(IEU(1) (λ) , 0), 0) ∗ − U1 (ω(IEU(1) (λ) , 0), 1) and ∗ ∗ SDEU0 (IEU(0) (λ)) = (1 − p) (1 + λ) p U1 (ω(IEU(0) , 1), 0) ∗ − U1 (ω(IEU(0) , 1), 1) . 13 ∗ ∗ We conclude that if U12 S 0 then ISDEU (λ) T IEU( (λ) for all λ ≥ 0. H̄) Proof of Proposition 3. We face the maximization problem stated in equation 2 max SDEU (α) with α h i SDEU (α) = E U (ω(α, L̃), h(L̃)) , where ω(α, L̃) = ω0 − (1 + λ)αE[L̃] − (1 − α)L. The FOC is given by h i SDEU0 (α (λ)) = E (L̃ − P )U1 (ω(α (λ) , L̃), h(L̃)) = 0. The SOC h i SDEU00 (α (λ)) = E (L̃ − P )2 U11 (ω(α (λ) , L̃), h(L̃)) < 0 ∗ ∗ is satisfied. The unique global maximum αSDEU (λ) is therefore determined by SDEU0 (αSDEU (λ)) = 0. For λ = 0, evaluating the FOC at α = 1 yields h i SDEU0 (1) = E (L̃ − E[L̃])U1 (ω0 − E[L̃], h(L̃)) = h i = Cov L̃, U1 (ω0 − E[L̃], h(L̃)) ∗ More than full (full, partial) insurance αSDEU (0) T 1 is therefore optimal if h i Cov L̃, U1 (ω0 − E[L̃], h(L̃)) T 0 which is equivalent to U12 S 0. Proof of Proposition 4. We face the maximization problem stated in equation 2. For λ > 0, we compare insurance purchase behavior of SDEU preferences with EU preferences. The FOC for individual with preferences characterized by U (·, H̄) with fixed H̄(λ) ≡ h((1 + λ)E[L̃]) is given by h i EU0U (·,H̄) (α (λ)) = E (L̃ − P )U1 (ω(α (λ) , L̃), H̄ (λ)) . ∗ The SOC is satisfied. The unique global maximum αEU( (λ) is therefore determined by H̄) 0 ∗ EUU (·,H̄) (αEU(H̄) (λ)) = 0. ∗ ∗ Evaluating the FOC for αSDEU (λ) at αEU( (λ) yields H̄) h i ∗ ∗ SDEU0 (αEU( (λ)) = E ( L̃ − P )U (ω(α (λ) , L̃), h( L̃)) 1 H̄) EU(H̄) ˆP ∗ (L − P ) U1 (ω(αEU( H̄) (λ) , L), h(L))dF (L) = 0 ˆL̄ ∗ (L − P ) U1 (ω(αEU( H̄) (λ) , L), h(L))dF (L). + P 14 ∗ The FOC EU0U (·,H̄) (α) at αEU( (λ) satisfies H̄) h i ∗ ∗ EU0U (·,H̄) (αEU( (λ)) = E ( L̃ − P )U (ω(α (λ) , L̃), H̄) 1 H̄) EU(H̄) ˆP ∗ (L − P ) U1 (ω(αEU( H̄) (λ) , L), H̄)dF (L) = 0 ˆL̄ ∗ (L − P ) U1 (ω(αEU( H̄) (λ) , L), H̄)dF (L) = 0. + P Suppose U12 > 0. Then, for L < (1 + λ)E[L̃] we have ∗ ∗ U1 (ω(αEU( H̄) (λ) , L), h (L)) ≥ U1 (ω(αEU(H̄) (λ) , L), h((1 + λ) E[L̃])) ∗ ∗ (L − P ) U1 (ω(αEU( H̄) (λ) , L), h (L)) ≤ (L − P ) U1 (ω(αEU(H̄) (λ) , L), h((1 + λ) E[L̃])) and for L > (1 + λ)E[L̃] we have ∗ ∗ U1 (ω(αEU( H̄) (λ) , L), h (L)) ≤ U1 (ω(αEU(H̄) (λ) , L), h((1 + λ) E[L̃])) ∗ ∗ (L − P ) U1 (ω(αEU( H̄) (λ) , L), h (L)) ≤ (L − P ) U1 (ω(αEU(H̄) (λ) , L), h((1 + λ) E[L̃])) ∗ ∗ with strict inequality for some L. Thus SDEU0 (αEU( (λ)) < EU0U (·,H̄) (αEU( (λ)) = 0. SimH̄) H̄) ilar reasoning applies to U12 < 0 and U12 = 0. Hence, we conclude that if U12 S 0 then ∗ ∗ ∗ ∗ (λ) T αEU( (λ) for all SDEU0 (αEU( (λ)) T EU0U (·,H̄) (αEU( (λ)) = 0 and therefore αSDEU H̄) H̄) H̄) λ ≥ 0 with H̄ ≡ h((1 + λ) E[L̃]). 15 References Arrow, K. J. (1971). Essays in the Theory of Risk-Bearing. Markham Publishing Company, Chicago. Arrow, K. J. (1973). Optimal insurance and generalized deductibles. Technical report, Office of Economic Opportunity. Cook, P. J. and Graham, D. A. (1977). The demand for insurance and protection: The case of irreplaceable commodities. The Quarterly Journal of Economics, 91(1):143–156. Dionne, G. (1982). Moral hazard and state-dependent utility function. The Journal of Risk and Insurance, 49(3):405–422. Eeckhoudt, L., Rey, B., and Schlesinger, H. (2007). A good sign for multivariate risk taking. Management Science, 53(1):117–124. Eisner, R. and Strotz, R. H. (1961). Flight insurance and the theory of choice. The Journal of Political Economy, 69(4):355–368. Epstein, L. G. and Zin, S. E. (1989). Substitution, risk aversion, and the temporal behavior of consumption and asset returns: A theoretical framework. Econometrica, 57(4):937–969. Evans, W. N. and Viscusi, W. K. (1991). Estimation of state-dependent utility functions using survey data. The Review of Economics and Statistics, 73(1):94–104. Finkelstein, A., Luttmer, E. F. P., and Notowidigo, M. (2013). What good is wealth without health? the effect of health on the marginal utility of consumption. Journal of European Economic Association, 11(S1):221–258. Frech III, H. E. (1994). State-dependent utility and the tort system as insurance: Strict liability versus negligence. International Review of Law and Economics, 14:261–271. Karni, E. (1985). Decision Making under Uncertainty: The Case of State-Dependent Preferences. Harvard University Press, Cambridge, Massachusetts and London, England. Mossin, J. (1968). Aspects of rational insurance purchasing. The Journal of Political Economy, 76(4-1):553–568. Nordquist, G. L. (1985). On the risk-aversion comparability of state-dependent utility functions. Theory and Decision, 18:287–300. Pratt, J. W. (1964). Risk aversion in the small and in the large. Econometrica, 32(1/2):122–136. Rey, B. (2003). A note on optimal insurance in the presence of a nonpecuniary background risk. Theory and Decision, 54:73–83. Schlesinger, H. (1984). Optimal insurance for irreplaceable commodities. The Journal of Risk and Insurance, 51(1):131–137. Viscusi, W. K. and Evans, W. N. (1990). Utility functions that depend on health status: Estimates and economic implications. The American Economic Review, 80(3):353–374. von Neumann, J. and Morgenstern, O. (1944). Theory of Games and Economic Behavior. Princeton University Press, Princeton. 16
© Copyright 2026 Paperzz