Abstract
For addressing the Allis-type anomalies, a fractional degree reference dependent stochastic dominance rule is developed which is a generalization of the integer degree reference dependent stochastic dominance rules. This new rule can effectively explain why the risk comparison does not satisfy translational invariance and scaling invariance in some cases. The rule also has a good property that it is compatible with the endowment effect of risk. This rule can help risk-averse but not absolute risk-averse decision makers to compare risks relative to reference points. We present some tractable equivalent integral conditions for the fractional degree reference dependent stochastic dominance rule, as well as some practical applications for the rule in economics and finance.
Similar content being viewed by others
References
Bawa S (1982) Research bibliography stochastic dominance: A research bibliography. Manage Sci 28:698–712
Bell DE (1985) Disappointment in Decision Making under Uncertainty. Oper Res 33(1):1–27
Bell DE (1988) The value of pre-decision side bets for utility maximizers. Manage Sci 34:797–800
Bi HW, Zhu W (2019) The non-integer higher-order Stochastic dominance. Oper Res Lett 47(2):77–82
Chateauneuf A, Cohen M, Meilijson I (2005) More pessimism than greediness: a characterization of monotone risk aversion in the rank-dependent expected utility model. Econ Theor 25:649–667
Denuit M, Müller A (2002) Smooth generators of integral stochastic orders. Ann Appl Probab 12:1174–1184
Delquié P, Cillo A (2006) Disappointment without prior expectation: a unifying perspective on decision under risk. J Risk Uncertainty 33:197–215
Dentcheva D, Ruszczyński A (2003) Optimization with stochastic dominance constraints. SIAM J Optim 14:548–566
Dentcheva D, Ruszczyński A (2006) Portfolio optimization with stochastic dominance constraints. J Bank Finance 30:433–451
Dittmann I, Maug E, Spalt O (2010) Sticks or carrots? Optimal CEO compensation when managers are loss averse. J Finance 65(6):2015–2050
Fishburn PC (1964) Decision and Value Theory. John Wiley & Sons, New York
Friedman M, Savage LJ (1948) The utility analysis of choices involving risk. J Polit Econ 56:279–304
Guo D, Hu Y, Wang S, Zhao L (2016) Comparing risks with reference points: A stochastic dominance approach. Insurance Math Econom 70:105–116
Gollier C (2016) Stochastic volatility implies fourth-degree risk dominance: Applications to asset pricing. TSE Working Papers
Hanoch G, Levy H (1969) The efficiency analysis of choices involving risk. Rev Econ Stud 36:335–346
Huang R, Tzeng L, Zhao L (2019) Fractional Degree Stochastic Dominance. Management Science forthcoming
Kahnemann D, Tversky A (1979) Prospect Theory: An Analysis of Decision under Risk. Econometrica 47(2):263–291
Köszegi B, Rabin M (2006) A Model of Reference-Dependent Preferences. Quart J Econ 121(4):1133–65
Köszegi B, Rabin M (2007) Reference-dependent risk attitudes. Amer Econ Rev 97:1047–1073
Levy H (1992) Stochastic dominance and expected utility: Survey and analysis. Manage Sci 38:555–593
Machina M (1982) Expected utility analysis without the independence axiom. Econometrica 50:277–323
Mao T, Hu T (2018) Fractional stochastic dominance. Technical Report. University of Science and Technology of China, Hefei
Markowitz H (1952) The utility of wealth. J Polit Econ 60:151–158
Müller A, Stoyan D (2002) Comparison Methods for Stochastic Models and Risks. John Wiley & Sons, Chichester, UK
Müller A, Scarsini M, Tsetlin I, Winkler R (2017) Between first-and second-Order stochatic dominance. Manage Sci 63:2933–2974
Post T, Kopa M (2016) Portfolio choice based on third-degree stochastic dominance. Manage Sci 63(10):3381–3392
Quiggin J (1993) Generalized Expected Utility Theory: the Rank-Dependent Model. Kluwer Academic, Boston
Rabin M (2000) Risk Aversion and Expected-Utility Theory: A Calibration Theorem. Econometrica 68(5):1281–92
Richard TH, Johnson EJ (1990) Gambling with the House Money and Trying to Break Even: The Effects of Prior Outcomes on Risky Choice. Manage Sci 36(6):643–60
Rothschild M, Stiglitz JE (1970) Increasing risk. I. A definition. J Econ Theo 2:225–243
Segal U (1987) Some remarks on Quiggin’s anticipated utility. J Econ Behav Org 8:145–154
Sprenger C (2015) An endowment effect for risk: Experimental tests of stochastic reference points. J Polit Econ 123:1456–1499
Sugden R (2003) Reference-Dependent Subjective Expected Utility. J Econ Theo 11(2):172–191
Tversky A, Kahneman D (1991) Loss aversion in risk-less choice: A reference dependent model. Quart J Econ 106:1039–1061
von Neumann J, Morgenstern O (1947) Theory of games and economic behavior, 2nd edn. Princeton University Press, Princeton, NJ
Yang J, Han S, Zhang X, Zhou D (2019) Optimization with fractional stochastic dominance constraints. Technical Report. Zhejiang Sci-Tech University, Hangzhou
Acknowledgements
J. Yang was supported by the NNSF of China (No. 11701518). S. Han was supported by the NNSF of China (No. 12071436).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendices
1.1 Proof of Theorem 1
Proof
\({\mathrm{[1]} \Rightarrow \mathrm{[2]}}\). Since \(U_{\gamma }^{*}\) is invariant under translations, any \(u\in U_{\gamma }^{*}\) can be approximated by a sequence of functions \(\{u_{n}\in U_{\gamma }, n=1,2,\cdots \}\) as in the proof of Theorem 2.1 in Denuit and Müller (2002). From this result it follows that \({\mathrm{[1]}}\) implies \({\mathrm{[2]}}\).
\({\mathrm{[2]} \Rightarrow \mathrm{[3]}}\). For a fixed \(t\in \mathfrak {R}\), we define the consumption utility function u(x; t) with the following right derivative:
Obviously, \(u(x;t)\in U_{\gamma }^{*}\). Due to the integration by part, it holds that
Hence, for all \(t\in \mathfrak {R}\), \(\eta ^{*}>0\) and \(\lambda ^{*}> 1\), \(E\left[ v(Y,r,u(x;t))\right] \ge E\left[ v(X,r,u(x;t))\right]\) implies that
\({\mathrm{[3]} \Rightarrow \mathrm{[1]}}\). Let \(u \in U_{\gamma }\). Without loss of generality we can assume
For any fixed \(n\ge 2\), define \(\varepsilon _n=2^{-n}\) and K as the largest integer k for which
and define a partition of a real line into intervals \([x_{k}, x_{k+1}]\) as follows: let \(x_0=-\infty\), \(x_{K+1}=\infty\) and
Then we define
It follows that
This implies that for all \(0\le k\le K\),
and
Let
and
Thus,
Set
Note that for all \(k=0, \ldots , K+1\),
and
and A(x, r) is positive and noningcreasing, which implies that \(\sum _{i=0}^{k}T_i \ge 0\). Furthermore, since \(m_k\) is a decreasing non-negative sequences, \(\sum _{i=0}^{k}m_iT_i\ge 0\). Therefore,
Letting \(n\rightarrow \infty\) yields part [1] holds. This completes the proof of the theorem.
1.2 Proof of Proposition 1
Proof
Let \(F_1(x)\) and \(F_2(x)\) be two cdfs of Z and W, respectively. The proof is main to construct cdfs \(F_1(x)\) and \(F_2(x)\). When \(x_1\le r\le x_2\), define \(F_1(x)\) and \(F_2(x)\) as
and
where \(\zeta _1^{*}\) satisfies that
Obviously, \(0\le \zeta _1^{*}\le \zeta _1\). Hence, \(F_1(x)\le F(x)\) and \(F_2(x)\ge G(x)\) for all \(x\in \mathfrak {R}\), that is, \(X\le _{1-SD}Z\) and \(W\le _{1-SD} Y\). From (13), (19), we have that Z is simple spread of Z. Combing with (21), it holds that \(Z \le _{(1+\gamma )-SD}^{r, \eta ^{*},\lambda ^{*}}Y\) but not \(Z\le _{(1+\gamma )-SD}Y\). And since X is also a simple spread of W based on (13), (20), \(X\le _{(1+\gamma )-SD}^{r, \eta ^{*},\lambda ^{*}}W\) but not \(X\le _{(1+\gamma )-SD}W\) by (21).
When \(x_2<r\le x_3\), we also define cdfs \(F_1(x)\) as (19) and \(F_2(x)\) as (20). But \(\zeta _1^{*}\) satisfies that
Similarly, we can obtain \(X\le _{1-SD}Z\), \(W\le _{1-SD}Y\), \(Z\le _{(1+\gamma )-SD}^{r, \eta ^{*},\lambda ^{*}}Y\) but not \(Z\le _{(1+\gamma )-SD}Y\) and \(X\le _{(1+\gamma )-SD}^{r, \eta ^{*},\lambda ^{*}}W\) but not \(X\le _{(1+\gamma )-SD}W\).
When \(x_3<r\le x_4\), we define
and
where \(\zeta _2^{*}\) satisfies that
Clearly, \(\zeta _2^{*}\ge \zeta _2\). Hence, \(F_1(x)\le F(x)\) and \(F_2(x)\ge G(x)\) for all \(x\in \mathfrak {R}\). That is, \(X\le _{1-SD}Z\) and \(W\le _{1-SD} Y\). Based on (13), (23), (25), we have \(Z \le _{(1+\gamma )-SD}^{r, \eta ^{*},\lambda ^{*}}Y\) but not \(Z\le _{(1+\gamma )-SD}Y\). And from (13), (24), (25), it holds that \(X\le _{(1+\gamma )-SD}^{r, \eta ^{*},\lambda ^{*}}W\) but not \(X\le _{(1+\gamma )-SD}W\). Combing these three cases, we complete the proof.
1.3 Proof of Proposition 2
Proof
Without loss of generality, we assume \(r_1< r_2\). Let X be a random variable with probability mass function \(p_1(x)\). We take \(x_1< x_2 \le x_3 <x_4\) with \(x_2=r_1\) and \(x_4=r_2\). Continue to take \(\eta _1>0\), \(\eta _2>0\), \(0<\gamma \le 1\) such that \(\gamma \eta _1\left( x_2-x_1\right) \frac{1+\eta ^{*}\lambda ^{*}}{1+\eta ^{*}}=\eta _2(x_4-x_3)\). Define random variable Y with probability mass function \(p_2(x)\) as
Let F(x) and G(x) be the cdfs of X and Y, respectively. Obviously, we have
Therefore, X is a simple spread of Y with a single crossing point \(x_2\). And since \(\gamma \eta _1\left( x_2-x_1\right) \frac{1+\eta ^{*}\lambda ^{*}}{1+\eta ^{*}}=\eta _2(x_4-x_3)\), we have
Thus, \(X\le _{(1+\gamma )-SD}^{r_1, \eta ^{*},\lambda ^{*}} Y\). And since
and
we have
Thus, \(X \nleqslant _{(1+\gamma )-SD}^{r_2,\eta ^{*},\lambda ^{*}} Y\). This completes the proof of proposition 2.
1.4 Proof of Theorem 2
To prove Theorem 2, we need the next lemma.
Lemma 1
Given cdfs F and G, let \(F^{-1}(s)=\inf \{x:F(x)\ge s\}\) and \(G^{-1}(s)=\inf \{x:G(x)\ge s\}\). \(F\le _{(1+\gamma )-SD}^{r, \lambda ^{*},\eta ^{*}} G\) if and only if, for all \(p\in [0,1]\),
Proof
We can use the similar proof of Müller et al. (2017). Define the function
Then, \(F\le _{(1+\gamma )-SD}^{r, \lambda ^{*},\eta ^{*}} G\) if and only if for all x, \(h(x)\ge 0\). Since \(H_{\gamma ,r}^{'}(t)\ge 0\), the function h assumes local minima in the points \(b_k\) where the distribution functions F and G cross, going from \(F\le G\) left of \(b_k\) to \(F>G\) right of \(b_k\). If we define \(p_k:=G(b_k)\) and
then we have \(h(b_k)=\tilde{h}(p_k)\) and the function \(\tilde{h}\) assumes its local minima in the points \(p_k\). Therefore, \(h\ge 0\) if and only if \(\tilde{h}\ge 0\).
Proof
Based on the idea of Müller et al. (2017), define
and
Without the loss of generality, we assume that \(A_{2}(1)>0\). Let \(\alpha (a)\) and \(\beta (a)\) be the smallest probabilities that solve
It follows from Lemma 1 that \(A_{1}(p)\ge A_{2}(p)\). Hence, \(\alpha (a)\le \beta (a)\) for all \(0<a<A_{2}(1)\). We set
and
Since X and Y assume only finitely many values, there is a sequence \(0=a_1<\cdots <a_k\le A_{2}(1)\) such that \(a\mapsto x_1(a), \cdots , x_4(a)\) are constant on \((a_{i-1}, a_i)\). Denote the corresponding values of these functions as
Moreover, for \(i=1,\ldots ,k\), at the points \(x_{1,i}\) and \(x_{4,i}\) the function F has jumps of sizes at least \(\zeta _{1i}\) and \(\zeta _{2i}\), and at the corresponding points \(x_{2,i}\) and \(x_{4,i}\) the function G has jumps of sizes at least \(\zeta _{1i}\) and \(\zeta _{2i}\), where \(\zeta _{1i}\) and \(\zeta _{2i}\) are given by the equation
For \(x>x_{4,k}\), we have \(F(x)>G(x)\). Thus, G is obtained from F by a sequence of k \((\gamma ,r)\)-transfers described by the corresponding x’s and \(\zeta\)’s above, plus a finite number of increasing transfers moving the mass from F to G right of \(x_{4,k}\). This completes the proof.
1.5 Proof of Theorem 3
Proof
Note that for random variables \(X_n\), X with distribution functions \(F_n\), F the convergence \(X_n\Rightarrow X\) mentioned in the theorem holds if and only if
and since \(0<H_{\gamma ,r}^{'}\le \frac{1+\eta ^{*}\lambda ^{*}}{1+\eta ^{*}}\), it holds that
This implies that, for any \(t\in \mathfrak {R}\),
The if-part thus follows from (14).
For the only-if-part, if X, Y are are bounded, then the proof is similar to Müller et al. (2017). We can define for any \(n\in \mathbb {N}\)
and
Then \(X_n\) and \(Y_n\) have finite support with \(X_n\le _{1-SD} X\) and \(Y\le _{1-SD} Y_n\). Therefore,
and \(X_n\Rightarrow X\) and \(Y_n\Rightarrow Y\).
If X and Y are unbounded, then we define
and
where \(x_n^{*}\) satisfies that
An easy calculation for the corresponding distribution functions \(F_n\), \(G_n\) shows that
Thus \(X\le _{(1+\gamma )-SD}^{r, \lambda ^{*},\eta ^{*}}Y\), that is,
implies that
Then, it follows that \(X\le _{(1+\gamma )-SD}^{r, \lambda ^{*},\eta ^{*}}Y\) implies that \(X_n\le _{(1+\gamma )-SD}^{r, \lambda ^{*},\eta ^{*}}Y_n\), and obviously \(X_n\), \(Y_n\) are bounded and \(X_n\Rightarrow X\), and \(Y_n\Rightarrow Y\).
For each fixed n we can approximate \(X_n\) and \(Y_n\) by sequences \(\{X_{nn}\}\) and \(\{Y_{nn}\}\) as in (27) and (28). Then, the sequences \(\{X_{nn}\}\) and \(\{Y_{nn}\}\) fulfill the conditions of the theorem. This completes the proof.
1.6 Proof of Theorem 4
To prove Theorem 4, we need Lemma 2.
Lemma 2
Define
Then, A(x; H) is nonegative and nonincreasing.
Proof
Obviously, A(x; H) is greater than zero. Since
and cdf H(x) is nondecreasing in x, to prove A(x; H) is nonincreasing is just to prove
is less than zero.
From \(\eta \ge \eta ^{*}\ge 0\), \(\lambda \ge \lambda ^{*}\ge 1\), we have
This completes the proof of Lemma 2.
Proof
Let \(X\vee R = \max \left\{ X, R\right\}\) and \(X\wedge R = \min \left\{ X,R\right\}\) with cdfs \(F_{X\vee R}(x)\) and \(F_{X\wedge R}(x)\), respectively. Then,
Thus,
Combing with Lemma 2, the equivalent integral condition of
for all \(u\in U_{\gamma }\), and \(\eta \ge \eta ^{*}, \lambda \ge \lambda ^{*}\) follows in a similar manner of the proof of Theorem 1.
1.7 Proof of Proposition 3
Proof
Since \(X_1\) is a simple spread of \(Y_1\) with crossing point \(x_0\) and
it holds that \(X_1\le _{(1+\gamma _0)-SD}^{Y_1,\eta ^{*},\lambda ^{*}}Y_1\). But we have
That is, \(X_2\nleq _{(1+\gamma _0)-SD}^{Y_2,\eta ^{*},\lambda ^{*}}Y_2\). This completes the proof.
1.8 Proof of Proposition 4
Proof
Since X is a simple spread of Y with crossing point \(x_0\) and \(\gamma _1\le 1\), we have \(X\le _{(1+\gamma _1)-SD}^{R_1\eta ^{*},\lambda ^{*}} Y\). For case [1], it holds that \(H_2(x)\ge H_1(x)\) on \([-\infty ,x_0)\) and \(H_2(x)\le H_1(x)\) on \([x_0,\infty )\). Thus, we have
That is, \(X\nleq _{(1+\gamma _1)-SD}^{R_2, \eta ^{*},\lambda ^{*}} Y\). For case [2], it holds that \(H_2(x)\le H_1(x)\) on \([-\infty ,x_0)\) and \(H_2(x)\ge H_1(x)\) on \([x_0,\infty )\). Thus, in contrast to (30), we have
That is, \(X\le _{(1+\gamma _1)-SD}^{R_2, \eta ^{*},\lambda ^{*}} Y\). This completes the proof.
Rights and permissions
About this article
Cite this article
Yang, J., Zhao, C., Chen, W. et al. Fraction-Degree Reference Dependent Stochastic Dominance. Methodol Comput Appl Probab 24, 1193–1219 (2022). https://doi.org/10.1007/s11009-022-09939-0
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11009-022-09939-0