Abstract
This paper introduces a new model of decision making under uncertainty. Aiming to provide a more realistic depiction of decision making, it generalizes the von Neumann–Morgenstern theory by including additional tiers of uncertainty. In this model, beliefs about the probabilities of events are ambiguous and their consequential utilities are vague; both are naturally formulated in the phantom space using phantom numbers. The degree of uncertainty, determined by the decision maker’s beliefs, is distinguished from the attitude toward uncertainty, which is drawn from her preferences. Decision making under ambiguity is a particular case of our model in which probabilities are ambiguous, but resulting utilities of events are knowable.
Similar content being viewed by others
Notes
A detailed discussion of our model with respect to the related literature is presented in Sect. 9.
Formally, a ring is a set of elements equipped with two binary operations, addition and multiplication, such that it is a commutative group with respect to addition and a commutative monoid with respect to multiplication.
The integer numbers, for example, equipped with the standard addition and multiplication form a commutative ring, which is not a field.
Often, we use \({\mathrm{ph }}\left( z \right) \) to denote the phantom term of \(z\) and \({\mathrm{re }}\left( z \right) \) to denote its real term.
In the view of \(\mathbb {PH}\) as \({{\mathbb {R}}}\times {{\mathbb {R}}}\), one can see that the complement of a set of zero divisors is dense in \(\mathbb {PH}.\) This implies that zero divisors do not change much, yet they do need to be dealt with formally.
The innovation of this probability measure is the ability to assign an event with a varying probability while preserving many familiar attributes of classical probability theory, as it is a Bayesian model.
It is important to note that the phantom representation of this problem is unique. The conjugates, \(\mathcal {P}\left( R \right) = 0.6 - \wp \,0.2\) and \(\mathcal {P}\left( R \right) = 0.5 + \wp \,0.3\), provide the phantom representation when the probability distortions are reversed.
Recall that ambiguity in our model is objective, since beliefs are exogenous.
We often abuse the notation and write \(\omega \in \varSigma \), i.e., events are replaced by states.
The condition \(\sum _{x\in \mathcal {X}} \mathcal {P}\left( x \right) =1\) is well defined because lotteries are restricted to have a finite support and the phantom probability measure is additive.
\(\varLambda ^\times \) replaces the open real interval \((0,1)\) in vNM.
In this context, \(x_1\) and \(x_2\) can be viewed as constant lotteries in \(\mathcal {L}\).
In Siniscalchi (2009), for example, the evaluated expected utility is adjusted by aggregating a vector of the covariances of utility and factors representing different sources of ambiguity.
The proof of this theorem follows the steps used by Mas-Colell et al. (1995).
The literature usually refers to this preference as a preference for uncertainty by ignoring the separation of preference for risk from preference for ambiguity.
Note that \({\mathrm{E }}_{\mathcal P} \left( x \right) \in \mathbb {PH}\) can be identified by a constant lottery and thus is comparable with any phantom lottery \(\mathcal Q\in \mathcal {L}\).
The next section investigates the aggregation of real and phantom utilities by the phantom-valued function.
The coefficient \(\gamma \) of risk aversion in the classical CRRA utility function can be any real number, while in the phantom case, it must be a rational number.
Recall that vague utility can be a result of a vague outcome (a phantom consequence) or simply because ex-ante the DM is unsure about the utility of a given consequence.
Note that the projection \(\check{x} = \frac{x + \overline{x}}{2}\in {\mathbb {R}}\) always satisfies \(\check{x}=\check{\overline{x}}\).
A similar notion of comparative preferences is widely used in the literature with respect to risk (see, for example, Pratt 1964). The use of this notion with respect to ambiguity was employed by Ghirardato and Marinacci (2002), who show that higher aversion to ambiguity implies lower capacities, and by Cohen and Meilijson (2013), who extend this notion to Choquet expected utility. See also Cerreia-Vioglio et al. (2011), who study rational preferences for ambiguity, and Castro and Chateauneuf (2011), who study ambiguity aversion in the context of trading.
That is, \({\mathrm{\Gamma }}_A = g \circ {\mathrm{\Gamma }}_B\), where \(g: {\mathbb {R}}\rightarrow {\mathbb {R}}\) is an increasing concave function.
In phantom probability, as in classical probability theory, if two random variables are independent, they are also uncorrelated and mean independent.
See Izhakian an Izhakian (2009, section 3.1).
This class of distributions includes the elliptically distributions whose probability characteristic function is of the form \(e^{it\nu }\Psi \left( \frac{1}{2} t^2 \sigma ^2 \right) \), where \(\Psi \) is a characteristic generator and \(i=\sqrt{-1}\). Normal distribution, student-t distribution, logistic distribution, exponential power distribution, and laplace distribution are all examples of elliptical distributions that can be generalized to phantoms.
The norm \(\left| a + \wp \,b \right| \) is defined by \(\left| a + \wp \,b \right| = \sqrt{\left( a+\frac{b}{2}\right) ^2 + \left( \frac{b}{2} \right) ^2}\), see Izhakian and Izhakian (2009, Equation (1.16)).
In the special case of unambiguous consequences, the negative impact phantoms have on risk implies a negative relation between risk and ambiguity—an insight that coincides with Izhakian (2012c).
In expected utility theory, the DM’s assessments of the likelihoods of \(R\), \(B\), and \(Y\) can be described by some probability measure \(P\). The DM is assumed to prefer a greater chance of winning $9 to a smaller chance of winning $9, such that the choices above imply that \(P\left( R \right) > P\left( B \right) \) and \(P\left( B \cup Y \right) > P\left( R \cup Y \right) \). However, since \(R\), \(B\), and \(Y\) are mutually exclusive events, no such conventional probability measure exists; hence, it is considered a paradox.
Ellsberg (1961) demonstrates that the vNM independent axiom and the Savage P2 axiom are violated.
In the view of phantom probabilities as directed probability intervals, they can be constructed in a canonical way. First, define one possible (real) probability distribution and then build the probability intervals around it by considering all other possible distributions. In some sense, it resembles construction of a set of priors à la Gilboa and Schmeidler (1989), but here the direction of probability distortions plays a major role.
The expected payoff computed with respect to conjugate probabilities is \( {\mathrm{E }}\left( \overline{\mathcal P}_{RY} \right) = 9 \left( \overline{1 -\wp \,\frac{2}{3}} \right) + 0 \left( \overline{\wp \,\frac{2}{3}} \right) = 9\left( \frac{1}{3} + \wp \,\frac{2}{3} \right) +0\left( \frac{2}{3} -\wp \,\frac{2}{3} \right) = 3 + \wp \,6\).
A similar claim can be made for an optimistic DM, whose preferences are \(\mathcal P_{R} \precsim \mathcal P_{B}\) and \(\mathcal P_{BY} \precsim \mathcal P_{RY}\).
In this example, the same qualitative results are obtained for risk neutrality and risk loving.
Kopylov (2006) \(\epsilon \)-contamination suggests the addition of an element of confidence to the generated set of priors.
For a detailed discussion about imprecise risk, see Giraud and Tallon (2011).
Klibanoff et al. (2011), for example, suggest a preference-based definition of ambiguous events. Amarante (2005) studies the measurability of such events, and Machina (2011) studies the separability of the Independence Axiom (under objective uncertainty) and the Sure-Thing Principle (under subjective uncertainty) in the context of ambiguous events.
Continuity and differentiability of phantom functions are discussed in Izhakian and Izhakian (2009, section 1.7).
\(x_1,x_2 \in \mathbb {PH}\) can be considered as vague utilities. Since utility functions are monotonic increasing, we abuse the notation and refer to \(x\in \mathcal {X}\) as its utility \({\mathrm{U }}\left( x \right) \).
Debreu’s theorem: If \(\succsim \) is a complete order over a connected, separable topological space \(X\) such that for some \(z\), the sets \(\left\{ x \in X \ \big | \ x \succsim z \right\} \) and \(\left\{ y \in X \ \big | \ z \succsim y \right\} \) are both closed for all \(x\in X\), then there exists a continuous real-valued function \( {\mathrm{\Gamma }}\) such that \(x \succsim y \ \Leftrightarrow \ {\mathrm{\Gamma }}\left( x \right) \ge {\mathrm{\Gamma }}\left( y \right) \).
Fishburn (1970, Theorem 3.1) suggests an alternative proof.
To save on notations, we refer to the phantom random variable \(X\) also as its realization and write \({\mathrm{U }}\left( X \right) \) for \({\mathrm{U }}\left( x \right) \).
References
Ahn, D.S.: Ambiguity without a state space. Rev. Econ. Stud. 75, 3–28 (2008)
Amarante, M.: Ambiguity, measurability and multiple priors. Econ. Theory 26, 995–1006 (2005)
Arrow, K.J.: Aspects of the Theory of Risk Bearing. Yrjo Jahnssonin Saatio, Helsinki (1965)
Bewley, T.F.: Knightian decision theory and econometric inferences. J. Econ. Theory 146, 1134–1147 (2011)
Boyle, P.P., Garlappi, L., Uppal, R., Wang, T.: Keynes meets markowitz: the tradeoff between familiarity and diversification. Manag. Sci. 5, 1–20 (2011)
Brenner, M., Izhakian, Y.: Asset Prices and Ambiguity: Empirical Evidence. Stern School of Business, Finance Working Paper Series, FIN-11-010 (2011)
Brenner, M., Izhakian, Y.: Pricing Systematic Ambiguity in Capital Markets. Stern School of Business, Finance Working Paper Series, FIN-12-008 (2012)
Casadesus-Masanell, R., Klibanoff, P., Ozdenoren, E.: Maxmin expected utility over Savage acts with a set of priors. J. Econ. Theory 92, 35–65 (2000)
Castro, L., Chateauneuf, A.: Ambiguity aversion and trade. Econ. Theory 48, 243–273 (2011)
Cerreia-Vioglio, S., Ghirardato, P., Maccheroni, F., Marinacci, M., Siniscalchi, M.: Rational preferences under ambiguity. Econ. Theory 48, 341–375 (2011)
Chateauneuf, A., Eichberger, J., Grant, S.: Choice under uncertainty with the best and worst in mind: neo-additive capacities. J. Econ. Theory 137(1), 538–567 (2007)
Chen, Z., Epstein, L.: Ambiguity, risk, and asset returns in continuous time. Econometrica 70, 1403–1443 (2002)
Chew, S.H., Sagi, J.S.: Small worlds: modeling attitudes toward sources of uncertainty. J. Econ. Theory 139, 1–24 (2008)
Cohen, M., Meilijson, I.: Preference for safety under the choquet model: in search of a characterization. Econ. Theory 1–24 (2013). doi:10.1007/s00199-013-0762-2
Debreu, G.: Representation of a preference ordering by a numerical function. In: Thrall, R., Coombs, C., Davies, R. (eds.) Decision Processes, pp. 159–175. Wiley, New York (1954)
Dow, J., Werlang, SRdC: Uncertainty aversion, risk aversion, and the optimal choice of portfolio. Econometrica 60(1), 197–204 (1992)
Einhorn, H.J., Hogarth, R.M.: Decision making under ambiguity. J. Bus. 59, 225–250 (1986)
Ellsberg, D.: Risk, ambiguity, and the savage axioms. Q. J. Econ. 75(4), 643–669 (1961)
Epstein, L.G.: A definition of uncertainty aversion. Rev. Econ. Stud. 66(3), 579–608 (1999)
Epstein, L.G., Schneider, M.: Recursive multiple-priors. J. Econ. Theory 113, 1–31 (2003)
Epstein, L.G., Zhang, J.: Subjective probabilities on subjectively unambiguous events. Econometrica 69, 265–306 (2001)
Ergin, H., Gul, F.: A theory of subjective compound lotteries. J. Econ. Theory 144, 899–929 (2009)
Fishburn, P.: Utility Theory for Decision Making. Wiley, New York (1970)
Gajdos, T., Hayashi, T., Tallon, J.M., Vergnaud, J.C.: Attitude toward imprecise information. J. Econ. Theory 140(1), 27–65 (2008)
Gajdos, T., Tallon, J.M., Vergnaud, J.C.: Decision making with imprecise probabilistic information. J. Math. Econ. 40, 647–681 (2004)
Ghirardato, P., Maccheroni, F., Marinacci, M.: Differentiating ambiguity and ambiguity attitude. J. Econ. Theory 118, 133–173 (2004)
Ghirardato, P., Marinacci, M.: Ambiguity made precise: a comparative foundation. J. Econ. Theory 102(2), 251–289 (2002)
Gilboa, I.: Expected utility with purely subjective non-additive probabilities. J. Math. Econ. 16, 65–88 (1987)
Gilboa, I., Schmeidler, D.: Maxmin expected utility with non-unique prior. J. Math. Econ. 18, 141–153 (1989)
Giraud, R.: Objective Imprecise Probabilistic Information, Second Order Beliefs and Ambiguity Aversion: An Axiomatization. Working Paper Series, University of Franche-Comté (2006)
Giraud, R.: Second order beliefs models of choice under imprecise risk: non-additive second order beliefs vs. nonlinear second order utility. Theor. Econ. (2013, forthcoming)
Giraud, R., Tallon, J.M.: Are beliefs a matter of taste? A case for objective imprecise information. Theory Decis. 71, 23–31 (2011)
Hansen, L.P., Sargent, T.J.: Robust control and model uncertainty. Am. Econ. Rev. 91, 60–66 (2001)
Hansen, L.P., Sargent, T.J., Tallarini, T.D.: Robust permanent income and pricing. Rev. Econ. Stud. 66, 873–907 (1999)
Hayashi, T., Miao, J.: Intertemporal substitution and recursive smooth ambiguity preferences. Theor. Econ. 6, 423–472 (2011)
Iwaki, H., Osaki, Y.: The dual theory of the smooth ambiguity model. Econ. Theory. 1–15 (2013). doi: 10.1007/s00199-013-0779-6
Izhakian, Y.: A theoretical foundation of ambiguity measurement. Stern School of Business, Economics Working Paper Series, ECN-12-01 (2012a)
Izhakian, Y.: Capital asset pricing under ambiguity. Stern School of Business, Economics Working Paper Series, ECN-12-02 (2012b)
Izhakian, Y.: Does ambiguity diversification pay? Stern School of Business, Economics Working Paper Series, ECN-12-11 (2012c)
Izhakian, Y., Izhakian, Z.: Phantom probability. Preprint at arXiv:0901.0902 (2009)
Jaffray, J.Y.: Linear utility theory for belief functions. Oper. Res. Lett. 8, 107–112 (1989)
Ju, N., Miao, J.: Ambiguity, learning, and asset returns. Econometrica 80, 559–591 (2012)
Klibanoff, P., Marinacci, M., Mukerji, S.: A smooth model of decision making under ambiguity. Econometrica 73, 1849–1892 (2005)
Klibanoff, P., Marinacci, M., Mukerji, S.: Definitions of ambiguous events and the smooth ambiguity model. Econ. Theory 48(2–3), 399–424 (2011)
Kopylov, I.: A Parametric Model of Hedging Under Ambiguity. Mimeo, UC Irvine. [829] (2006)
Maccheroni, F., Marinacci, M., Ruffino, D.: Alpha as ambiguity: Robust mean-variance portfolio analysis. Econometrica 81, 10751113 (2013)
Maccheroni, F., Marinacci, M., Rustichini, A.: Ambiguity aversion, robustness, and the variational representation of preferences. Econometrica 74, 1447–1498 (2006)
Machina, M.J.: Event-separability in the ellsberg urn. Econ. Theory 48, 425–436 (2011)
Mas-Colell, A., Whinston, M.D., Green, J.R.: Microeconomic Theory. Oxford University Press, Oxford (1995)
Meyer, P.A.: Quantum Probability for Probabilists, 2nd edn. Springer, Berlin (1995)
Mukerji, S.: Understanding the nonadditive probability decision model. Econ. Theory 9, 23–46 (1997)
Nascimento, L., Riella, G.: Second-order ambiguous beliefs. Econ. Theory 52, 1005–1037 (2013)
Nau, R.F.: Uncertainty aversion with second-order utilities and probabilities. Manag. Sci. 52, 136–145 (2006)
Nehring, K.D.: Bernoulli Without Bayes: A Theory of Utility-Sophisticated Preferences Under Ambiguity. Mimeo. University of California, Davis (2007)
Nehring, K.D.: Imprecise probabilistic beliefs as a context for decision-making under ambiguity. J. Econ. Theory 144, 1054–1091 (2009)
Neilson, W.S.: Ambiguity Aversion: An Axiomatic Approach Using Second Order Probabilities. Mimeo. A&M University, Texas (1993)
Olszewski, W.: Preferences over sets of lotteries. Rev. Econ. Stud. 74, 567–595 (2007)
Pratt, J.W.: Risk aversion in the small and in the large. Econometrica 32, 122–136 (1964)
Quiggin, J., Chambers, R.G.: Invariant risk attitudes. J. Econ. Theory 117, 96–118 (2004)
Rébillé, Y.: Decision making over necessity measures through the choquet integral criterion. Fuzzy Sets Syst. 157, 3025–3039 (2006)
Roberts, K.W.S.: Interpersonal comparability and social choice theory. Rev. Econ. Stud. 47, 421–439 (1980)
Rothschild, M., Stiglitz, J.E.: Increasing risk: I. A definition. J. Econ. Theory 2, 225–243 (1970)
Schmeidler, D.: Subjective probability and expected utility without additivity. Econometrica 57, 571–587 (1989)
Seo, K.: Ambiguity and second-order belief. Econometrica 77, 1575–1605 (2009)
Siniscalchi, M.: A behavioral characterization of plausible priors. J. Econ. Theory 128, 91–135 (2006)
Siniscalchi, M.: Vector expected utility and attitudes toward variation. Econometrica 77, 801–855 (2009)
Stinchcombe, M.: Choice and Games with Ambiguity as Sets of Probabilities. Mimeo, University of Texas, Austin (2003)
Takesaki, M.: Theory of Operator Algebras I, Volume Operator Algebras and Non-commutative Geometry of Encyclopaedia of Mathematical Sciences, vol. 124. Springer, Berlin (2000)
Tversky, A., Kahneman, D.: Advances in prospect theory: cumulative representation of uncertainty. J. Risk Uncertain 5, 297–323 (1992)
Ui, T.: The ambiguity premium vs. the risk premium under limited market participation. Rev. Finance 15, 245–275 (2011)
Von-Neumann, J., Morgenstern, O.: Theory of Games and Economic Behavior. Princeton University Press, Princeton (1944)
Walley, P.: Statistical Reasoning with Imprecise Probabilities. Chapman and Hall, London (1991)
Wang, T.: A Class of Multi-priors Preferences. University of British Columbia, British Columbia (2003)
Wang, Z., Klir, G.J.: Fuzzy Measure Theory. Plenum Press, New York (1991)
Yaari, M.E.: The dual theory of choice under risk. Econometrica 55, 95–115 (1987)
Youssef, S.: Quantum mechanics as bayesian complex probability theory. Mod. Phys. Lett. 9(28), 2571–2586 (1994)
Zhang, J.: Subjective Ambiguity, Expected Utility and Choquet Expected Utility. Carleton Economic Papers 99–19, Carleton University, Department of Economics (1999)
Author information
Authors and Affiliations
Corresponding author
Additional information
The authors thank two anonymous referees, Adam Brandenburger, Xavier Gabaix, Sergiu Hart, Ruth Kaufman, Efe Ok, Benjamin Polak, Jacob Sagi, Larry Samuelson, and especially Itzhak Gilboa and Thomas Sargent for valuable discussions and suggestions. We would also like to thank the seminar participants at Lund University, New York University, Technion, Tel Aviv University, The Hebrew University of Jerusalem, and Yale University. The first author acknowledges the support of the AXA research fund. The second author acknowledges the support of the Chateaubriand scientific fellowship, Ministry of Science France.
Appendix
Appendix
1.1 Phantom spaces: special attributes
The special properties of the phantom framework are explored in this section. For this purpose, we use the following definition. A function \(f\) taking phantom arguments is said to have a realization form if it can be written as
where \(f_{ {\mathrm{re }}}\) and \(\widehat{f}\) are real functions, called the real component and the reduction of \(f\), respectively.
Recall that the division of a nonnegative real number by a greater or equal positive number lies in the interval \(\left[ 0,1 \right] \). The phantom analog of this definition, with respect to \(\bar{\varLambda }\), is as follows.
Claim
Suppose \(z_2 \ \underline{\gg } \ z_1\) are, respectively, pseudo-positive and pseudo-nonnegative, then \(\frac{z_1}{z_2} \in \bar{\varLambda }\).
Proof
Write the fraction \( \frac{z_1}{z_2} = \frac{a_1}{a_2} + \wp \,\left( \frac{ \widehat{z_1}}{\widehat{z_2}} - \frac{a_1}{a_2} \right) \), then \(0 \le \frac{a_1}{a_2} \le 1\) and \(0 \le \frac{\widehat{z_1}}{\widehat{z_2}} \le 1\). Since \(\frac{\widehat{z_1}}{\widehat{z_2}} = \widehat{(\frac{{z_1}}{{z_2}} )}\), then \(- \frac{a_1}{a_2} \le \widehat{(\frac{{z_1}}{{z_2}} )}- \frac{a_1}{a_2} \le 1- \frac{a_1}{a_2} \), implying that \(\frac{z_1}{z_2} \in \bar{\varLambda }\). \(\square \)
Later, when using phantom affinity and convexity, we also need the following.
Claim
Let \(z_1,z_2,z_3 \in \mathbb {PH}\), then there exists \(\lambda \in \mathbb {PH}\) such that \(z_2 = \lambda z_1 + \left( 1-\lambda \right) z_3\). If \(z_1 \ne z_3\) and at least one of them is not a zero divisor, then \(\lambda \) is unique.
Proof
Let \(z_i = a_i + \wp \,b_i\), \(i = 1,2,3\), and let \(\lambda = \alpha + \wp \,\beta \). Expanding the product \( \lambda z_1 + \left( 1-\lambda \right) z_3\), we have
This shows that each term is linear in \(\alpha \) and in \(\beta \), and that there exists \(\lambda \) that satisfies the requirement. By the same token, when \(z_1 \ne z_3\), i.e., \(a_1 \ne a_3\), then \(\alpha \) is unique, and if \(z_1\) or \(z_2\) are not zero divisors, then \(\beta \) is also unique. \(\square \)
The following definition is analogous to the known definition of affinity over the real numbers.
Definition 10
A phantom-valued function \(f: \mathbb {PH}\rightarrow \mathbb {PH}\) is affine if it is of the form
for some \(\alpha _1, \ldots , \alpha _n,\beta \in \mathbb {PH}.\)
The class of affine functions has subclasses of functions \(f: \mathbb {PH}\rightarrow {\mathbb {R}}\) and \(f: {\mathbb {R}}\rightarrow {\mathbb {R}}\), which satisfy the following lemma.
Lemma 1
(Phantom affinity) If the function \(f: \mathbb {PH}\rightarrow \mathbb {PH}\) is affine, then for any \(\lambda \in \varLambda \)
Proof
By affinity, \(f\left( \lambda z_1 + (1-\lambda )z_2 \right) = \lambda \left( \alpha z_1 + \beta \right) + (1-\lambda )\left( \alpha z_2+ \beta \right) = \lambda f(z_1) + (1-\lambda ) f(z_2).\) \(\square \)
Lemma 2
Let \(f: \mathbb {PH}\rightarrow {{\mathbb {R}}}\) be a continuous function.Footnote 46 For any real \(\alpha \in \left[ 0,1 \right] \), there exists \(\lambda \in \bar{\varLambda }\) such that
Conversely, if \(f\) is monotonic, for any \(\lambda \in \bar{\varLambda }\), there exists a real \(\alpha \in [0,1]\) for which Eq. (2) holds.
Proof
In the view of \(f\) as a function \(f: {\mathbb {R}}^2 \rightarrow {\mathbb {R}}\), the proof is obtained by the continuity of \(f\) over \({\mathbb {R}}^2\). \(\square \)
In order to compare phantom numbers, one needs to equip \(\mathbb {PH}\) with a weak order \(\succsim \), which is assumed to be given with the phantom structure. In our case, this order corresponds to the DM’s preferences and, by Theorem 1, has an order-preserving function \({\mathrm{\Gamma }}\).
Assumption 1
The order \(\succsim \) is taken to be compatible with the familiar order of \({\mathbb {R}}\). That is, if \(z_1 \succsim z_2\) for \(z_1,z_2 \in {\mathbb {R}}\), then \(z_1 \ge z_2\).
Definition 11
A function \(f: \mathbb {PH}\rightarrow \mathbb {PH}\) is (i) Monotonically increasing if \(f(z_1) \succsim f(z_2)\) for all \(z_1 \succsim z_2\), and monotonically decreasing if \(f(z_1) \precsim f(z_2)\); (ii) Strictly increasing if \(f(z_1) \succ f(z_2)\) for all \(z_1 \succ z_2\), and strictly decreasing if \(f(z_1) \prec f(z_2)\).
In the special case when \(f:\mathbb {PH}\rightarrow {\mathbb {R}}\), the inequality between the values of functions is given by the standard order of \({\mathbb {R}}\). This leads us to define the convexity of the phantom-valued function.
Definition 12
A continuous function \(f: \mathbb {PH}\rightarrow \mathbb {PH}\) is convex on a set \(X \subseteq \mathbb {PH}\) if for any nonzero divisor \(z_1,z_2 \in X\) and any \(\lambda \in \varLambda \): \( f \left( \lambda z_1 + \left( 1-\lambda \right) z_2 \right) \precsim \lambda f \left( z_1 \right) + \left( 1-\lambda \right) f\left( z_2 \right) . \) \(f\) is said to be concave on \(X\) if \( f \left( \lambda z_1 + \left( 1-\lambda \right) z_2 \right) \succsim \lambda f \left( z_1 \right) + \left( 1-\lambda \right) f\left( z_2 \right) .\)
When the target of \(f\) is restricted to \({{\mathbb {R}}}\), the weight \(\lambda \) belongs to the real interval \([0,1]\). Definition 12 leads to a generalization of the Jensen inequality of functions having a finite support.
Lemma 3
Let \(f: \mathbb {PH}\rightarrow \mathbb {PH}\) be a continuous convex function on \(X \subseteq \mathbb {PH}\), and let \(\lambda _i \in \varLambda \), \(i = 1,2,\ldots ,n\), be phantom weights such that \(\sum \lambda _i =1\). Then, \( f \left( {\sum \lambda _i z_i} \right) \precsim {\sum \lambda _i f(z_i)}, \) assuming \(z_i\) is a nonzero divisor for every \(i\). In the case where \(f: \mathbb {PH}\rightarrow {\mathbb {R}}\) and \(\lambda _i \in [0,1]\) are real weights, \( f \left( {\sum \lambda _i z_i)} \right) \le {\sum \lambda _i f(z_i)}\), assuming each \(z_i\) is a nonzero divisor for every \(i\).
Proof
We prove the lemma by induction on \(n\). Assume that \(f\) has a finite support, when \(n=2\), the convexity of \(f\) implies \( f \left( \lambda _1 z_1 + \lambda _2 z_2 \right) \precsim \lambda _1 f \left( z_1 \right) + \lambda _2 f \left( z_2 \right) .\) For \(n+1\), write \(\sum _{i=1}^{n+1} \lambda _i = \mu + \lambda _{n+1}\). Then, \(\sum _{i=1}^{n+1} \lambda _i z_i = \mu (\sum _{i=1}^n \frac{\lambda _i}{\mu } z_i) + \lambda _{n+1}z_{n+1}\), and by convexity
The induction step shows that the right-hand side is
\(\square \)
1.2 Properties of preferences
This section studies additional properties of the preference relation \(\succsim \) over phantom lotteries in \(\mathcal {L}\), defined in Axioms 1–6.
Lemma 4
(Intermediacy) If \(\mathcal P\succ \mathcal Q\) and \(\alpha \in \varLambda \), then \(\mathcal P\succsim \alpha \mathcal P+ (1 - \alpha ) \mathcal Q\ \succsim \mathcal Q\). When \(\alpha \ne 0\) and \(\alpha \ne 1\), the relations are strict.
Proof
Since \( \mathcal P\succ \mathcal Q\), Independence (Axiom 4) implies that \(\mathcal P= \alpha \mathcal P+ (1 - \alpha ) \mathcal P\succsim \alpha \mathcal P+ (1 - \alpha ) \mathcal Q\succsim \alpha \mathcal Q+(1 - \alpha ) \mathcal Q= \mathcal Q\). \(\square \)
Lemma 5
(Mixture monotonicity) Suppose that \(\mathcal P\succ \mathcal Q\) and \(\alpha , \beta \in [0,1]\), then
Proof
\((\Leftarrow )\): Assume \(\beta > \alpha \). Define \(\gamma \in [0,1] \) to be \( \gamma = \frac{ \beta - \alpha }{1 - \alpha } \) and write \(\beta \mathcal P+ (1 - \beta ) \mathcal Q= \gamma \mathcal P+ (1 - \gamma ) \left( \alpha \mathcal P+ (1 - \alpha ) \mathcal Q \right) .\) Since \(\mathcal P\succ \mathcal Q\) and \(\alpha \in [0,1] \subseteq \varLambda \), Lemma 4 implies that \(\gamma \mathcal P+ (1 - \gamma ) \left( \alpha \mathcal P+ (1 - \alpha ) \mathcal Q \right) \succ \alpha \mathcal P+ (1 - \alpha ) \mathcal Q\). We conclude that \( \beta \mathcal P+ (1 - \beta ) \mathcal Q\succ \alpha \mathcal P+ (1 - \alpha ) \mathcal Q\).
\((\Rightarrow ):\) Assume \(\beta \le \alpha \) and \(\beta \mathcal P+ (1 - \beta ) \mathcal Q\succ \alpha \mathcal P+ (1 - \alpha ) \mathcal Q\). Interchanging \(\alpha \) and \(\beta \) in the previous argument, one gets \(\alpha \mathcal P+ (1 - \alpha ) \mathcal Q\succsim \beta \mathcal P+ (1 - \beta ) \mathcal Q\)—a contradiction. \(\square \)
Lemma 6
(Unique solvability) For \(\mathcal P\succsim \mathcal Q\succsim \mathcal R\), there exists a unique \(\alpha ^*\in [0,1]\) such that \(\mathcal Q\backsim \alpha ^*\mathcal P+ (1 - \alpha ^*) \mathcal R\).
Proof
When \(\mathcal P\backsim \mathcal Q\) (\(\mathcal Q\backsim \mathcal R\)), take \(\alpha ^*= 1\) (\(\alpha ^*= 0).\) Otherwise, consider the sets \({\mathcal {H}}_{A} = \left\{ \alpha \in [0,1] \ \big | \ \mathcal Q\succ \alpha \mathcal P+ (1 - \alpha ) \mathcal R \right\} \) and \({\mathcal {H}}_{B} = \left\{ \alpha \in [0,1] \ \big | \ \alpha \mathcal P+ (1 - \alpha ) \mathcal R\succ \mathcal Q \right\} .\) By Archimedean (Axiom 3), these are disjoint nonempty open sets, which do not cover \([0,1]\). Thus, there exists \(\alpha ^*\in [0,1]\) for which \(\mathcal Q\backsim \alpha ^*\mathcal P+ (1 - \alpha ^*) \mathcal R\), which is unique by mixture monotonicity (Lemma 5). \(\square \)
Theorem 5
Let \(x_1,x_2 \in \mathbb {PH}\) be consequences in \(\mathcal {X}\).Footnote 47 Then, there exists a continuous real order-preserving function (phantom-valued representation) \({\mathrm{\Gamma }}: \mathbb {PH}\rightarrow {\mathbb {R}}\), such that
The function \({\mathrm{\Gamma }}\) is unique up to a strictly increasing transformation.
Proof
The preference \(\succsim \) on \(\mathcal {X}\) is induced by the order on \(\mathcal {L}\). To see this, one can construct a constant lottery for each element in \(\mathcal {X}\). Archimedean (Axiom 3) provides the convexity of \(\mathcal {X}\), implying that \(\mathcal {X}\) is connected. Let \(\mathcal {Y}\) be the set of the phantom consequence in \(\mathcal {X}\). Since \(\mathcal {Y}\) is a connected subset of the topological space \(\mathbb {PH}\), realized as the Cartesian product \({{\mathbb {R}}}\times {{\mathbb {R}}}\), and since the rational numbers are dense in \({{\mathbb {R}}}\), then \(\mathcal {Y}\) contains a countable dense subset. That is, \(\mathcal {Y}\) is a separable. Thus, by Debreu (1954) Theorem I, there is a function \({\mathrm{\Gamma }}: \mathcal {Y}\rightarrow {\mathbb {R}}\) that preserves \(\succsim \).Footnote 48 Since \(\succsim \) is assumed to coincide with the order of \({\mathbb {R}}\), any strictly increasing transformation of the value function \({\mathrm{\Gamma }}\) preserves \(\succsim \).Footnote 49 \(\square \)
Proposition 10
Suppose \(\mathcal P\) and \(\mathcal R\) are phantom lotteries whose outcomes \(X\) and \(Z\) are independent and \({\mathrm{E }}_{\mathcal R} \left( Z \right) = 0\). Let \(\mathcal Q_i\), \(i = 1,2\), be the phantom lotteries having the outcomes \(X + \lambda _i Z\), where \(\lambda _1, \lambda _2 \in \mathbb {PH}\). Then, \(\mathcal Q_1\) is strictly more uncertain than \(\mathcal Q_2\) for every pseudo-positive \(\lambda _1 \gg \lambda _2\).
Proof
Let \(\mu =\frac{\lambda _2}{\lambda _1}\) and write \(X + \lambda _2 Z = \mu (X + \lambda _1 Z) + (1-\mu ) X\), which by Claim 11.1 is in \(\varLambda \). Then,Footnote 50
since \({\mathrm{U }}: \mathcal {X}\rightarrow \mathbb {PH}\) is strictly concave. Taking expectations implies \( {\mathrm{E }}_{\mathcal Q_2} \left( {\mathrm{U }}\left( X + \lambda _2 Z \right) \right) >_{\mathrm{\Gamma }}\mu {\mathrm{E }}_{\mathcal Q_1} \left( {\mathrm{U }}\left( X + \lambda _1 Z \right) \right) + (1 -\mu ){\mathrm{E }}_{\mathcal P} \left( {\mathrm{U }}\left( X \right) \right) \). Since \(\mathcal Q_1\), whose outcome is \(X + \lambda _1 Z\), is strictly more uncertain than \(\mathcal P\), we have \({\mathrm{E }}_{\mathcal P} \left( {\mathrm{U }}\left( X \right) \right) >_{\mathrm{\Gamma }}{\mathrm{E }}_{\mathcal Q_1} \left( {\mathrm{U }}\left( X + \lambda _1 Z \right) \right) \). Together with Eq. (3), this yields, \( {\mathrm{E }}_{\mathcal Q_2} \left( {\mathrm{U }}\left( X + \lambda _2 Z \right) \right) >_{\mathrm{\Gamma }}{\mathrm{E }}_{\mathcal Q_1} \left( {\mathrm{U }}\left( X + \lambda _1 Z \right) \right) \). Then, Theorem 3 asserts that \(\mathcal Q_1\) is strictly more uncertain than \(\mathcal Q_2\). \(\square \)
Corollary 2
The phantom lottery \(\mathcal Q'\) with outcomes \(\lambda Y\) is strictly more uncertain than lottery \(\mathcal Q\) with outcomes \(Y\) for any \(\lambda \gg 1\).
Proof
The proof is obtained immediately by applying Proposition 10 to a constant phantom lottery \(\mathcal P\) with outcome 0 and to phantom lottery \(\mathcal R\) whose outcome is given by \(Y - {\mathrm{E }}_{\mathcal Q} \left( Y \right) \). \(\square \)
1.3 Proofs
Proof of Theorem 1
Recall that \({\mathrm{C }}: \mathcal {L}\rightarrow {\mathbb {R}}\) is given by \({\mathrm{C }}: \mathcal P\mapsto {\mathrm{\Gamma }}\left( \sum {\mathrm{U }}\left( x \right) \mathcal {P}\left( x \right) \right) \), i.e., it is a composition of the utility function \({\mathrm{U }}: \mathcal {X}\rightarrow \mathbb {PH}\) and the value function \({\mathrm{\Gamma }}: \mathbb {PH}\rightarrow {\mathbb {R}}\).
\((\Leftarrow ):\) The following two-step proof first proves the assertion for a closed and bounded (by a worst and a best consequence) subset of lotteries \(\mathcal {L}\) and then extends to the entire \(\mathcal {L}\).Footnote 51
I) Assume \(x^* \) and \(x_* \), where \(x^* \succsim x_*\) are a best and a worst consequences, respectively. Note that if \(x^* \thicksim x_*\), then \(\mathcal P\backsim \mathcal Q\) for any \(\mathcal P, \mathcal Q\in \mathcal {L}\), and the proof is accomplished by taking a constant value function.
We first show, inductively, that each (not necessarily constant) phantom lottery \(\mathcal P\in \mathcal {L}\) is equivalent to a compound lottery constructed of \(x^* \) and \(x_*\). Suppose \(\mathcal P: x_i \mapsto \mathfrak {p}_i\), \(i =1,\ldots ,n\), then for the sequence of constant lotteries \(\mathcal Q_i^{\mathrm{C }}: x_i \mapsto 1\), we have
By Archimedean (Axiom 3), \(\mathcal Q^{\mathrm{C }}_i \backsim \alpha _i x^* + (1-\alpha _i) x_*\) for some \(\alpha _i \in \varLambda \). Therefore,
Recall that \(\mathfrak {p}_i= \mathcal P(x_i)\) and set each phantom value \(\alpha _i ={\mathrm{U }}\left( x_i \right) \), thus
i.e., \({\mathrm{U }}\left( x \right) \mathcal {P}\left( x \right) \in \varLambda \) for any \(x \in \mathcal {X}\).
On the other hand, by Unique Solvability (Lemma 6), there is a unique (real) \(\alpha ^*\in [0,1]\) such that
By Theorem 5, we have \({\mathrm{\Gamma }}\left( \alpha ^* \right) = \alpha ^*= {\mathrm{\Gamma }}\left( \sum {\mathrm{U }}\left( x \right) \mathcal {P}\left( x \right) \right) \in [0,1]\). Hence,
To complete this part of the proof, suppose \(\mathcal P\succsim \mathcal Q\) are two lotteries in \( \mathcal {L}\). By the above considerations (see Eq. (4)), for \(\mathcal Q\), we have
Then, Lemma 5 (Mixture monotonicity) completes the proof of this part:
II) Next, we extend the result of part I to the case where there is no best or worst lottery. If there is no best or worst lottery, we take an arbitrary pair of constant phantom lotteries \(\mathcal P_0 \succ \mathcal Q_0\) in \(\mathcal {L}\) and define the set of lotteries \(\mathcal {T}_0 = \left\{ {\mathcal R} \in \mathcal {L}\big | \mathcal P_0 \succsim \mathcal R\succsim \mathcal Q_0 \right\} \). If \(\mathcal P_0 \backsim \mathcal Q_0\), for all \(\mathcal P_0,\mathcal Q_0 \in \mathcal {L}\), then the conclusion holds trivially. Otherwise, following the reasoning above, we can obtain a functional representation \({\mathrm{C }}_0\) for all the lotteries in \(\mathcal {T}_0\).
Next, consider lotteries \(\mathcal P_1 \succ \mathcal P_0\) and \(\mathcal Q_0 \succ \mathcal Q_1\), i.e., they are outside \(\mathcal {T}_0\) and, as before, define the set of lotteries \(\mathcal {T}_1 = \left\{ \mathcal R\in \mathcal {L}\big | \mathcal P_1 \succsim \mathcal R\succsim \mathcal Q_1 \right\} \). Clearly, \(\mathcal {T}_0 \subset \mathcal {T}_1\), and therefore, the choice function \({\mathrm{C }}_0\) on \(\mathcal {T}_0\) agrees with the choice function \({\mathrm{C }}_1\), representing the preference on \(\mathcal {T}_1\). That is, \({\mathrm{C }}_0(\mathcal R) = {\mathrm{C }}_1(\mathcal R)\) for every \(\mathcal R\in \mathcal {T}_1 \cap \mathcal {T}_0\). Proceeding inductively, we obtain a sequence \(\mathcal {T}_0 \subset \mathcal {T}_1 \subset \mathcal {T}_2 \subset \cdots \subset \mathcal {L}\) of sets of lotteries, each having a preference representative \({\mathrm{C }}_i\) with \( {\mathrm{C }}_i(\mathcal R) = {\mathrm{C }}_{i+1}(\mathcal R) = {\mathrm{C }}_{i+2}(\mathcal R)= \cdots \) for all \(\mathcal R\in \mathcal {T}_{i}\), and thereby construct a choice function \({\mathrm{C }}\) whose domain is \(\mathcal {L}\).
\((\Rightarrow )\): Suppose that \({\mathrm{C }}: \mathcal {L}\rightarrow {\mathbb {R}}\) is a choice function representing \(\succsim \). We need to prove that Axioms 1–6 are satisfied. By the continuity of \({\mathrm{C }}\) (since \({\mathrm{U }}\) and \({\mathrm{\Gamma }}\) are continuous), for any phantom lottery \(\mathcal P\in \mathcal {L}\), one can find a constant lottery \(\mathcal P^{\mathrm{C }}\) such that \({\mathrm{C }}\left( \mathcal P^{{\mathrm{C }}} \right) = {\mathrm{C }}\left( \mathcal P \right) \). Therefore, one may consider only constant lotteries, which are identified with consequences \(x\in \mathcal {X}\), and \({\mathrm{C }}\) can be viewed as a function sending \(\mathcal {X}\) to \({\mathbb {R}}\).
Completeness (Axiom 1) and Transitivity (Axiom 2) are immediately obtained by the fact that any function \({\mathrm{\Gamma }}: \mathbb {PH}\rightarrow {\mathbb {R}}\) induces a weak order on \(\mathbb {PH}\) and that \(\mathcal {X}\) is contained in the domain of \({\mathrm{C }}\). Archimedean (Axiom 3): The representation of \(\mathcal P\succsim \mathcal Q\succsim \mathcal R\) implies \({\mathrm{C }}\left( \mathcal P \right) \ge {\mathrm{C }}\left( \mathcal Q \right) \ge {\mathrm{C }}\left( \mathcal R \right) \), which are reals. Thus, there is \(\lambda \in (0, 1)\) for which \(\lambda {\mathrm{C }}\left( \mathcal P \right) + (1-\lambda ){\mathrm{C }}\left( \mathcal R \right) \ge {\mathrm{C }}\left( \mathcal Q \right) \). By Lemma 2, there is \(\alpha \in \varLambda \) such that \(\lambda {\mathrm{C }}\left( \mathcal P \right) + (1-\lambda ){\mathrm{C }}\left( \mathcal R \right) = {\mathrm{C }}\left( \alpha \mathcal P+ (1-\alpha )\mathcal R \right) \), hence \({\mathrm{C }}\left( \alpha \mathcal P+ (1-\alpha )\mathcal R \right) \ge {\mathrm{C }}\left( \mathcal Q \right) \), which by representation (converse direction) yields \(\alpha \mathcal P+ (1-\alpha )\mathcal R\succsim \mathcal Q\). The same argument shows that \(\mathcal Q\succsim \beta \mathcal P+ (1-\beta )\mathcal R\) for some \(\beta \in \varLambda \). Independence (Axiom 4) is then obtained by the Archimedean (Axiom 3), the continuity of \({\mathrm{C }}\), and its monotonicity. Consequence monotonicity (Axiom 5) is obtained since \({\mathrm{\Gamma }}\) is monotonic function, i.e., monotonic in phantom numbers. Uniformity (Axiom 6) is obtained immediately, since one can find a constant phantom lottery \(\mathcal P^{{\mathrm{C }}}\) such that \({\mathrm{C }}\left( \mathcal P^{{\mathrm{C }}} \right) = {\mathrm{\Gamma }}\left( {\mathrm{U }}\left( \mathcal P^{{\mathrm{C }}} \right) \right) = {\mathrm{\Gamma }}\left( \sum {\mathrm{U }}\left( x \right) \mathcal P\left( x \right) \right) = {\mathrm{C }}\left( \mathcal P \right) \). \(\square \)
Proof of Proposition 1
Since, by Theorem 5, \({\mathrm{\Gamma }}\) is invariant to strictly increasing transformations, then \({\mathrm{C }}= {\mathrm{\Gamma }}\circ {\mathrm{U }}\) is also invariant to strictly increasing transformations. \(\square \)
Proof of Proposition 2
The proof is straightforward: \({\mathrm{C }}\) is a composition of two order-preserving functions \({\mathrm{\Gamma }}\) and \({\mathrm{U }}\). \(\square \)
Proof of Proposition 3
The proof is derived directly from the concavity of the representation functions (Definition 12) and the convexity of preferences, together with the representation theorems (Theorems 5 and 1). \(\square \)
Proof of Proposition 4
We prove for risk aversion. The proof for risk loving and risk neutrality is obtained by similar considerations. Write \((i)\) \({\mathrm{E }}\left( \mathcal P \right) \succsim \mathcal P\) \(\ \Leftrightarrow \ \) \((ii)\) \({\mathrm{U }}\) is concave \(\ \Leftrightarrow \ \) \((iii)\) \({\mathrm{U }}\left( {\mathrm{E }}_{\mathcal {P}} \left( x \right) \right) \ge _{\mathrm{\Gamma }}{\mathrm{E }}_{\mathcal {P}} \left( {\mathrm{U }}\left( x \right) \right) \). \((i) \Rightarrow (ii)\): Assume that \(\succsim \) exhibits risk aversion, represented by \({\mathrm{U }}\). That is, \({\mathrm{E }}_{\mathcal {P}} \left( x \right) = \sum x\mathcal {P}\left( x \right) \succsim \mathcal P\). The latter can be written as a compound phantom lottery \(\bigoplus _i \alpha _i \mathcal P_i^{\mathrm{C }}\) of constant lotteries (cf. Remark 1) \(\mathcal P_1^{\mathrm{C }}, \ldots , \mathcal P_n^{\mathrm{C }}\), each identified with a consequence \(x_i \in \mathcal {X}\). By Theorem 1, \({\mathrm{U }}\left( {\mathrm{E }}_{\mathcal {P}} \left( x \right) \right) \ge _{\mathrm{\Gamma }}\sum _{i=1}^n \alpha _i {\mathrm{U }}\left( x_i \right) \), where \(\alpha _i = \mathcal {P}\left( x_i \right) \), and thus, by Lemma 3, \({\mathrm{U }}\) is concave. \((ii) \Rightarrow (iii)\): Directly by Lemma 3. \((iii) \Rightarrow (i)\): If \({\mathrm{U }}\) is concave, then \({\mathrm{U }}\left( {\mathrm{E }}_{\mathcal {P}} \left( x \right) \right) \ge _{\mathrm{\Gamma }}{\mathrm{E }}_{\mathcal {P}} \left( {\mathrm{U }}\left( x \right) \right) \) by Lemma 3; thus, \({\mathrm{E }}_{\mathcal P} \left( x \right) \succsim \mathcal P\). \(\square \)
Proof of Proposition 5
The second equality is obtained by applying proposition 1.18 of Izhakian and Izhakian (2009) to \({\mathrm{U }}\left( x \right) = -\frac{e^{-\gamma x}}{\gamma }\). The Arrow-Pratt measure \(A\left( x \right) =-\frac{{\mathrm{U }}''\left( x \right) }{{\mathrm{U }}'\left( x \right) }\), computed using phantom derivatives, implies that \(A\left( x \right) =\gamma \), cf. Izhakian and Izhakian (2009, section 1.7). \(\square \)
Proof of Proposition 6
Suppose \(\gamma = \frac{m}{n}\). Then, \(\frac{x^{1-\gamma }}{1-\gamma }\) can be written \({\mathrm{U }}\left( x \right) = \frac{n}{n-m}\root n \of {x^{m-n}}\). The second equality is then obtained by the realization property for exponents and roots of phantoms, cf. Izhakian and Izhakian (2009, section 1.4). The phantom derivatives imply \(A\left( x \right) =-\frac{{\mathrm{U }}''\left( x \right) }{{\mathrm{U }}'\left( x \right) }x=\gamma \), cf. Izhakian and Izhakian (2009, section 1.7). \(\square \)
Proof of Proposition 7
We prove the case (i); the other two cases are proved by the same considerations. Suppose \({\mathrm{\Gamma }}\) is concave, then by Lemma 3, \({\mathrm{\Gamma }}\left( \check{x} \right) \ge \frac{{\mathrm{\Gamma }}\left( x \right) +{\mathrm{\Gamma }}\left( \overline{x} \right) }{2}\), which by Theorem 1 implies \(\check{x} \succsim \frac{1}{2} x \oplus \frac{1}{2} \overline{x}\). Conversely, suppose that \(\succsim \) exhibits phantom aversion, then \(\check{x} \succsim \frac{1}{2} x \oplus \frac{1}{2} \overline{x}\), which by Theorem 1 implies \({\mathrm{\Gamma }}\left( \check{x} \right) \ge \frac{{\mathrm{\Gamma }}\left( x \right) +{\mathrm{\Gamma }}\left( \overline{x} \right) }{2}\), i.e., \({\mathrm{\Gamma }}\) is concave. \(\square \)
Proof of Theorem 2
Recall that since DMs are assumed to have identical risk attitudes, phantom lotteries can be represented by their expected utilities \(x \in \mathbb {PH}\). In particular, \(\check{x}\) can be referred to as a constant real lottery in \(\mathcal {L}\) and \(\frac{1}{2}x \oplus \frac{1}{2}\overline{x}\) as a phantom lottery in \(\mathcal {L}\). Recall also that \({\mathrm{\Gamma }}: \mathbb {PH}\rightarrow {\mathbb {R}}\) is assumed to be monotonically increasing with respect to an order that coincides with the order on \({{\mathbb {R}}}\). Therefore, \({\mathrm{\Gamma }}^{-1}\), is well defined on the restriction \({\mathrm{\Gamma }}|_{{\mathbb {R}}}\) of \({\mathrm{\Gamma }}\) to \({{\mathbb {R}}}\). (\(\Rightarrow \)): Suppose that DM \(A\) is at least as phantom averse as DM \(B\) and assume \(x,y,z \in \mathbb {PH}\) such that \({\mathrm{\Gamma }}_B\left( x \right) < {\mathrm{\Gamma }}_B\left( y \right) < {\mathrm{\Gamma }}_B\left( z \right) \). Let \(\lambda \in (0,1)\) satisfy \({\mathrm{\Gamma }}_B\left( y \right) = \lambda {\mathrm{\Gamma }}_B\left( x \right) + (1-\lambda ){\mathrm{\Gamma }}_B\left( z \right) \). One have to show that \({\mathrm{\Gamma }}_A\left( y \right) \ge \lambda {\mathrm{\Gamma }}_A\left( x \right) + (1-\lambda ){\mathrm{\Gamma }}_A\left( z \right) \). Suppose by way of negation that \({\mathrm{\Gamma }}_A\left( y \right) < \lambda {\mathrm{\Gamma }}_A\left( x \right) + (1-\lambda ){\mathrm{\Gamma }}_A\left( z \right) \). Then, by monotonicity and continuity, there is \(y' \succ y\) close enough to \(y\) for which \(y' \prec _A \lambda x \oplus (1-\lambda )z\) and \(y' \succ _B \lambda x \oplus (1-\lambda )z\)—a contradiction. Therefore, \(y' \succsim _A \lambda x \oplus (1-\lambda )z\), and by Theorem 1 \({\mathrm{\Gamma }}_A\left( y \right) \ge \lambda {\mathrm{\Gamma }}_A\left( x \right) + (1-\lambda ){\mathrm{\Gamma }}_A\left( z \right) \). Then, for \(g = {\mathrm{\Gamma }}_A|_{{\mathbb {R}}}\circ {\mathrm{\Gamma }}_B^{-1}|_{{\mathbb {R}}}\), it follows that \(g\left( {\mathrm{\Gamma }}_B\left( y \right) \right) \ge \lambda g\left( {\mathrm{\Gamma }}_B\left( x \right) \right) + (1-\lambda )g\left( {\mathrm{\Gamma }}_B\left( z \right) \right) \). That is, \(g\) is concave. (\(\Leftarrow \)): Assume that \({\mathrm{\Gamma }}_A\) is a concave transformation of \({\mathrm{\Gamma }}_B\). Then, by definition, \(\frac{{\mathrm{\Gamma }}_B\left( x \right) +{\mathrm{\Gamma }}_B\left( \overline{x} \right) }{2}={\mathrm{\Gamma }}_B\left( {\mathrm{RCE }}_B(\mathcal P) \right) ,\) i.e., \( {\mathrm{RCE }}_B(\mathcal P) = {\mathrm{\Gamma }}_{B}^{-1} \left( \frac{{\mathrm{\Gamma }}_B\left( x \right) +{\mathrm{\Gamma }}_B\left( \overline{x} \right) }{2} \right) \). Let \(g: {{\mathbb {R}}}\rightarrow {{\mathbb {R}}}\) be defined by \(g = {\mathrm{\Gamma }}_A|_{{\mathbb {R}}}\circ {\mathrm{\Gamma }}_B^{-1}|_{{\mathbb {R}}}\). Since \(g\) is concave, by the Jensen inequality,
The monotonicity of \({\mathrm{\Gamma }}_A\) implies that \( {\mathrm{RCE }}_B \left( \mathcal P \right) \ge {\mathrm{RCE }}_A \left( \mathcal P \right) \). Now, suppose a constant lottery \(\mathcal Q^{\mathrm{C }}\) satisfying \(\mathcal P\succsim _A \mathcal Q^{\mathrm{C }}\), then \( {\mathrm{RCE }}_A\left( \mathcal P \right) \succsim _A {\mathrm{RCE }}_A\left( \mathcal Q^{\mathrm{C }} \right) \). This implies \( {\mathrm{RCE }}_A\left( \mathcal P \right) \ge {\mathrm{RCE }}_A\left( \mathcal Q^{\mathrm{C }} \right) \), since \(\succsim _A\) is compatible with the order of \({{\mathbb {R}}}\). But \( {\mathrm{RCE }}_B\left( \mathcal P \right) \ge {\mathrm{RCE }}_A\left( \mathcal P \right) \) and \( {\mathrm{RCE }}_A\left( \mathcal Q^{\mathrm{C }} \right) = {\mathrm{RCE }}_B\left( \mathcal Q^{\mathrm{C }} \right) \) which, by Theorem 5, implies that \( {\mathrm{RCE }}_B\left( \mathcal P \right) \succsim _B {\mathrm{RCE }}_B\left( \mathcal Q^{\mathrm{C }} \right) \) and by transitivity \(\mathcal P\succsim _B \mathcal Q^{\mathrm{C }}\). \(\square \)
Proof of Corollary 1
Let \(g: {\mathbb {R}}\rightarrow {\mathbb {R}}\) be a twice-differentiable function such that \({\mathrm{\Gamma }}_A = g \circ {\mathrm{\Gamma }}_B\). Thus, \(\partial _{\wp \,}{{\mathrm{\Gamma }}_A} = g'({\mathrm{\Gamma }}_B) \partial _{\wp \,}{{\mathrm{\Gamma }}_B}\) and \(\partial _{\wp \,\wp \,}{{\mathrm{\Gamma }}_A} = g''({\mathrm{\Gamma }}_B) (\partial _{\wp \,}{{\mathrm{\Gamma }}_B})^2 + g'({\mathrm{\Gamma }}_B) \partial _{\wp \,\wp \,}{{\mathrm{\Gamma }}_B}\). Since \({\mathrm{\Gamma }}_A\) and \({\mathrm{\Gamma }}_B\) are strictly increasing, \(g'({\mathrm{\Gamma }}_B)=\frac{\partial _{\wp \,}{{\mathrm{\Gamma }}_A}}{\partial _{\wp \,}{{\mathrm{\Gamma }}_B}}>0\). Putting all together:
Recall that, by consequence monotonicity (Axiom 5), \(\partial _{\wp \,}{{\mathrm{\Gamma }}_B} \ge 0\). Therefore, \(\vartheta _A\left( x \right) \ge \vartheta _B\left( x \right) \) iff \(g''({\mathrm{\Gamma }}_B)/g'({\mathrm{\Gamma }}_B) < 0\), i.e., \(g\) is concave. The proof is then completed by Theorem 2. \(\square \)
Proof of Proposition 8
We prove the case of pessimism; the cases of optimism and apathy are proved similarly. Take \(x = a + b \wp \,\in \mathbb {PH}\) with \(\check{x} = a + \frac{1}{2}b \in {\mathbb {R}}\). Then, by Theorem 1, \(\check{x} \succ x\) iff \({\mathrm{\Gamma }}\left( a + \frac{1}{2}b \right) > {\mathrm{\Gamma }}\left( a + b \wp \, \right) \). Since \({\mathrm{\Gamma }}\) is assumed to be differentiable,
which implies that if \(b > 0\), then \(\theta = \frac{\partial _{r \,}{{\mathrm{\Gamma }}}}{\partial _{\wp \,}{{\mathrm{\Gamma }}}} >2\), and if \(b < 0\), then \(\theta = \frac{ \partial _{r \,}{{\mathrm{\Gamma }}}}{\partial _{\wp \,}{{\mathrm{\Gamma }}}}<2\). \(\square \)
Proof of Proposition 9
Proposition 8 is a special case of Proposition 7, and the proof is derived from the latter. \(\square \)
Proof of Theorem 3
\((\Rightarrow )\): Assume \(\mathcal P\) is more uncertain than \(\mathcal Q\). Let \(x=X-{\mathrm{E }}_{\mathcal P} \left( X \right) \) and \(y=Y-{\mathrm{E }}_{\mathcal Q} \left( Y \right) \) be random variables. By Definition 6, \(x =_dy+z\) for some phantom lottery \(\mathcal R\in \mathcal {L}\) with \({\mathrm{E }}_{\mathcal R} \left( z \right) =0\). Considering a DM’s preference \(\succsim \), characterized by \({\mathrm{U }}:\mathcal {X}\rightarrow \mathbb {PH}\) and \({\mathrm{\Gamma }}:\mathbb {PH}\rightarrow {\mathbb {R}}\), we have
Risk aversion implies a concavity of \({\mathrm{U }}\) and thus, by Lemma 3,
Taking expectations implies \({\mathrm{E }}_{\mathcal P} \left( {\mathrm{U }}\left( x \right) \right) \le _{\mathrm{\Gamma }}{\mathrm{E }}_{\mathcal Q} \left( {\mathrm{U }}\left( y \right) \right) \). Applying the same arguments with respect to \({\mathrm{\Gamma }}\) provides \({\mathrm{\Gamma }}\left( {\mathrm{E }}_{\mathcal P} \left( {\mathrm{U }}\left( x \right) \right) \right) \le {\mathrm{\Gamma }}\left( {\mathrm{E }}_{\mathcal Q} \left( {\mathrm{U }}\left( y \right) \right) \right) \). Hence, \(\mathcal P\precsim \mathcal Q\).
\((\Leftarrow )\): On the contrary, assume that \(\mathcal P\succsim \mathcal Q\), but \(\mathcal P\) is more uncertain than \(\mathcal Q\), i.e., \(x =_dy+z\). However, by the arguments of the first part of the proof, if the DM is uncertainty averse, then \({\mathrm{E }}_{\mathcal P} \left( {\mathrm{U }}\left( x \right) \right) \le _{\mathrm{\Gamma }}{\mathrm{E }}_{\mathcal Q} \left( {\mathrm{U }}\left( y \right) \right) \), and therefore \(\mathcal P\precsim \mathcal Q\), which is a contradiction. \(\square \)
Proof of Theorem 4
Recall that a phantom variance is pseudo-positive (Izhakian and Izhakian 2009, Lemma 1.10). (\(\Rightarrow \)): Suppose \(\mathcal P\) is strictly more uncertain than \(\mathcal Q\). Then, \(X -{\mathrm{E }}_{\mathcal P} \left( X \right) =_dY - {\mathrm{E }}_{\mathcal Q} \left( Y \right) + Z\), for some phantom lottery \(\mathcal R\) with \({\mathrm{E }}_{\mathcal R} \left( Z \right) = 0\). Since \({\mathrm{E }}_{\mathcal P} \left( X \right) = {\mathrm{E }}_{\mathcal Q} \left( Y \right) \), then \(X =_dY + Z\). Because \(Y\) and \(Z\) are independent, \(\sigma _{\mathcal P}^2 = \sigma _{\mathcal Q}^2 + \sigma _{\mathcal R}^2\), yielding \(\sigma _{\mathcal P}^2 \gg \sigma _{\mathcal Q}^2\), since \(\sigma _{\mathcal R}^2\) is pseudo-positive.
(\(\Leftarrow \)): The phantom lottery \(\mathcal Q'\) having the outcomes \(Y' = \lambda \left( Y-{\mathrm{E }}_{\mathcal Q} \left( Y \right) \right) \) is symmetric with mean 0 and variance \(\lambda ^2 \sigma _{\mathcal Q}^2\). Let \(\lambda =\frac{\sigma _{\mathcal P}^2}{\sigma _{\mathcal Q}^2}\), which is pseudo-positive, and \(\lambda \gg 1\) by hypothesis. Clearly, \(\sigma _{\mathcal P}^2 = \lambda \sigma _{\mathcal Q}^2\), and therefore, the distribution of \(\lambda \left( Y -{\mathrm{E }}_{\mathcal Q} \left( Y \right) \right) \) is identical to the distribution of \(X-{\mathrm{E }}_{\mathcal P} \left( X \right) \). By Corollary 2, \(\lambda \left( Y-{\mathrm{E }}_{\mathcal Q} \left( Y \right) \right) \) is strictly more uncertain than \(Y-{\mathrm{E }}_{\mathcal Q} \left( Y \right) \), and hence \(X-{\mathrm{E }}_{\mathcal P} \left( X \right) \) is strictly more uncertain than \(Y-{\mathrm{E }}_{\mathcal Q} \left( Y \right) \). Since expectations play no role here, \(\mathcal P\) is strictly more uncertain than \(\mathcal Q\). \(\square \)
Proof of Claim 1
Pessimism implies
, but for the LHS, we have
, cf. Table (1). Thus, \({\mathrm{E }}_{\mathcal P_{R}} \left( X \right) \succsim {\mathrm{E }}_{\mathcal P_{B}} \left( X \right) \), which by Axiom 6 implies \({\mathcal P_{R}} \succsim {\mathcal P_{B} }\). By the same consideration, starting with
, one can see that \(\mathcal P_{BY} \succsim \mathcal P_{RY}\). Therefore, Independence (Axiom 4) is satisfied. \(\square \)
Rights and permissions
About this article
Cite this article
Izhakian, Y., Izhakian, Z. Decision making in phantom spaces. Econ Theory 58, 59–98 (2015). https://doi.org/10.1007/s00199-013-0798-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00199-013-0798-3
Keywords
- Phantom probability
- Decision making under uncertainty
- Expected utility
- Imprecise risk
- Ambiguity
- Uncertainty
- Ellsberg paradox