�ݺ�ߣ

International Journal of Business Marketing and Management (IJBMM)
Volume 8 Issue 3 May-June 2023, P.P. 19-30
ISSN: 2456-4559
www.ijbmm.com
International Journal of Business Marketing and Management (IJBMM) Page 19
Mean-Distortion Risk Measure Portfolio Optimization
under the Distribution Uncertainty
Yang Mu1
, PeibiaoZhao1
1
Schoolof Mathematics&Statistics,Nanjing Universityof Scienceand Technology,China Email:
648297379@; pbzhao@njust.edu.cn
Abstract: In this paper, we introduce a universal framework for mean-distortion robust risk measurement and
portfolio optimization. We take accounts for the uncertainty based on Gelbrich distance and another uncertainty
set proposed by Delage & Ye. We also establish the model under the constraints of probabilistic safety
criteria and compare the different frontiers and the investment ratio to each asset. The empirical analysis in the
final part explores the impact of different parameters on the model results.
Keywords: Robust optimization, Distortion risk measure,Probabilistic safety criteria.
I. Introduction
In today's world, the financial system is the foundation for the sustained and stable development of the market
economy. Financial risks, as an inevitable product of rapid economic development, have become the normal
accident. In order to prevent systemic financial risks and maintain the stability of the world financial system,
relevant financial institutions must strengthen their awareness of effective risk identification and improve their
ability to manage risks. From a financial perspective, the risk of financial assets measures the return or potential
losses of investing in different assets, as well as the volatility of assets, which is the degree of deviation from the
mean.The "big bang" of financial risk management is the birth of modern portfolio theory, which is a milestone
in the history of financial development. Its founder, Markowitz [1], published "Portfolio Selection" in 1952 and
won the Nobel Prize in Economics 40 years later. The mean variance portfolio model is a mathematical
technique used to develop the optimal combination of portfolio assets under specific risk measures. Its core is a
statistical analysis, and its optimal selection is determined by the historical returns of asset classes and the
correlation between these returns and other asset returns. Once the mean variance criterion was proposed, it was
widely used by institutional investors such as mutual funds.Based on the shortcomings of risk sensitivity
measurement indicators in traditional risk measurement, J.P. Morgan [2] proposed a new risk measurement VaR
(Value-at-Risk) method in 1994 to meet the needs of his banking business, which was quickly promoted as an
industry standard. However, Var does not satisfy subadditivity, which can not explain the nature of risk
reduction in diversified investments.
Then Artzner et al. [3] (1999) proposed the concept of Coherent risk measure. They believed that a
well-defined risk measure should meet the four axioms of monotonicity, homogeneity, translation invariance and
subadditivity. Then Delbaen and Hochachule [4] extended the finite probability space required by the consistent
risk measure theory to any probability space, and linked the consistent risk measure with game theory and

Mean-Distortion Risk Measure Portfolio Optimization Under The Distribution Uncertainty
distortion probability measure. These results are illustrated with corresponding examples. Then Acerbi et al in
2002 [5] put forward the theory of spectral risk measures, which requires that risk measures not only be
consistent risk measures, but also have a good risk spectral density to characterize the risk aversion of investors.
In this paper, we mainly focus in the distortion risk measure. This measure was firstly introduced by Wang in
1996 [6] when he calculated the insurance premiums by converting cumulative distribution functions. Also,
Wang in 2000 [7] proposed a class of distortion function operators based on normal cumulative distribution,
which took into account both assets and liabilities in the pricing formula, linked the CAPM model, and restored
the Black Scholes option pricing formula, but this pricing was mainly based on insurance risk. In 2012, Li Jun
[8] studied the properties of several specific distortion risk measures within the framework of uncertain
distribution, and based on this, used hybrid intelligent algorithms to calculate the mean risk model. In the same
year, Feng and Tan [9] proposed the Consistent Distortion Risk Measure (CDRM) based on CVaR's portfolio
model and applied it to optimization, comparing the optimal investment portfolios of different transformation
measures. Cai and Wang [10] studied the tail subadditivity of distortion risk measures in 2017. They proposed
multiple risk measures (MTD) and their properties. Finally, these risk measures were applied to the capital
allocation model of venture capital portfolios.
At the same time, many scholars study the robust portfolio optimization based on different kinds of risk
measure. In this paper, we follow the concept of Gelbrich distance which was introduced by Gelbrich in 1990
[11] and define the Gelbrich ambiguity set as the family of all asset return distributions with a given structure
whose mean-covariance paris reside in a Gelbrich ball around an empirical mean-covariance pair estimated from
sample data. Also, we introduced another ambiguity set which was forward by Delage and Ye in 2010 [12].
They use mean and covariance to construct a cone-constrained uncertainty set whose covariance less than
specific times sample data’s covariance.
In general, the constraint of portfolio investment models is the returnconstraint, but there are also other
constraints, such as the safety criterion first proposed by Roy [13] in 1952.Subsequently, Pyle and Turnovsky[14]
compared the results of the Markowitz mean variance model based on safety criteria with the traditional
expected utility maximization model. On this basis, kataoka and Telser[15] proposed different security criteria.
Although the loss objective function is consistent, the optimized objects are different.
II. Problem Statement
In the classic portfolio investment models, it is usually assumed that the market is a complete market. But in
real world, we can only observe incomplete information in the market. The decisions made by the investors
often rely on the unknown distributions and parameters in the model that also called uncertainty.For example,
the distribution of the return shows a fat tail at the tip, so it does not obey the normal distribution. The
uncertainty of this distribution is also called ambiguity. Therefore, many scholars have studied portfolio models
under ambiguity sets, resulting in a series of distribution uncertainty robust models.Delage and Ye [12] studied
the different kinds of uncertainty sets. Chaoui et al. [16] in 2003 discussed the worst-case value-at-risk and
robust portfolio optimization. Zhuet al. [17] in 2009 studied worst-case conditional value-at-risk with
application to robust portfolio management. Then Kanget al. [18] established data-driven robust mean-CVaR
portfolio selection model under distribution ambiguity.
Our research firstly establishes the robust mean-distortion risk measure portfolio selection model and
considers the frontier and best investment ratios of each asset under the condition that the moments of return
satisfy the constraint of Gelbrich distance. Then inspired by the Kang et al. [18], in which they established the
model based on the parameter 𝜃 that is the index of ambiguity attitude. The new detaile d𝜃 − robustmodel is

as below
min (1 − 𝜃) inf
𝒬∈𝒯
𝜌𝒬(−𝑥𝑇
𝜉) + 𝜃 sup
𝒬∈𝒯
𝜉)
𝑠. 𝑡. (1 − 𝜃) sup
𝒬∈𝒯
𝐸𝒬(𝑅(𝑥, 𝜉)) + 𝜃 inf
𝒬∈𝒯
𝐸𝒬(𝑅(𝑥, 𝜉)) ≥ 𝜙
III. Distortion Risk Measure
Wang proposed a pricing method using proportional risk transformation when researching premium pricing in
1996, which involved the theory of distortion risk measurement. The distortion risk measure utilizes the
distortion function proposed by Yaari in 1987. Although the concept of distortion risk measurement originated
from insurance, due to the relationship between insurance and investment risk, distortion risk measurement has
also begun to be used for investment environment and portfolio selection issues (such as Van der Hoek and
Sherris (2001)).
1.1 Definition of the Distortion risk measure
Definition 1 If 𝑔: (0,1) → (0,1)is anon-decreasing function and satisfies𝑔(0) = 0,𝑔(1) = 1(𝑔 is also called
distortion function), the distribution function of𝑋 ∈ 𝒳is 𝐹𝑋(𝑥). Then the distortion risk measure is given by
𝜌𝑔(𝑋) = − ∫ [1 − 𝑔(1 − 𝐹𝑋(𝑥))]𝑑𝑥
0
−∞
+ ∫ 𝑔(1 − 𝐹𝑋(𝑥))𝑑𝑥
∞
0
#(1)
and at least one integral is finite.
When we define the cumulative distribution function (also known as the survival function)𝑆𝑋(𝑥) = 1 −
𝐹𝑋(𝑥) = 𝑃(𝑋 > 𝑥),𝜌𝑔(𝑋)can also be represented as
𝜌𝑔(𝑋) = − ∫ [1 − 𝑔(𝑆𝑋(𝑥))]𝑑𝑥
0
−∞
+ ∫ 𝑔(𝑆𝑋(𝑥))𝑑𝑥
∞
0
#(2)
In some cases, such as issues related to insurance or capital requirements, it can be assumed that the random
variable 𝑋 ∈ 𝒳 is non negative, then
𝜌𝑔(𝑋) = ∫ 𝑔(𝑆𝑋(𝑥))𝑑𝑥
∞
0
#(3)
Usually, assuming that the distribution function of the distortion risk 𝑔 and variable 𝑋 are independent of
each other, the distortion risk measure represents the expectation of a new random variable with a reweighted
probability. By changing the probability of risk tolerance and assigning higher probability weights to high-risk
events while keeping the loss distribution function unchanged, investors can adjust their subjective probability
of tail risk through the distortion function, indicating their aversion to risks at different positions.
The distortion risk measure satisfies the following properties:
Proposition 1 Monotonicity. Suppose 𝑋, 𝑌 ∈ 𝒳，if 𝑋 ≤ 𝑌, then𝜌𝑔(𝑋) ≤ 𝜌𝑔(𝑌).
Proposition 2 Positive homogeneity. For distortion risk measure 𝜌𝑔, Let 𝑋 ∈ 𝒳,𝜆 ≥ 0, then we have𝜌𝑔(𝜆𝑋) =
𝜆𝜌𝑔(𝑋)
Proposition 3 Translation invariance.For distortion risk measure 𝜌𝑔, Let 𝑋 ∈ 𝒳, then we have∀𝑐 ∈ 𝑅: 𝜌𝑔(𝑋 +
𝑐) = 𝜌𝑔(𝑋) + 𝑐
Proposition 4 Distortion risk measures are sub-additive if and only if the distortion function𝑔is concave. That is
𝜌𝑔(𝑋 + 𝑌) ≤ 𝜌𝑔(𝑋) + 𝜌𝑔(𝑌)
Proposition 5 For a given distortion function 𝑔,the distortion risk measure𝜌𝑔(𝑋)is coherent risk measure if and
only if that 𝜌𝑔(𝑋)can be represented as

𝜌𝑔(𝑋) = ∫ 𝜔(𝛼)𝜙𝛼𝑑𝛼
1
𝛼=0
#(4)
Where 𝜔 ∶ [0,1] ↦ [0,1],𝜔(𝛼) = 𝑔′(𝑥)|𝑥=1−𝛼and 𝜙𝛼(𝑋)is 𝐶𝑉𝑎𝑅𝛼(𝑥, 𝑃).
This proposition indicates that coherent risk measure can be expressed as the convex combination of 𝐶𝑉𝑎𝑅𝛼.
The proof of this see propositioncan be found in Feng and Tan's article.
1.2 Examples of the Distortion risk measure
Example 1 Suppose 𝑋 ∈ 𝒳，𝛼 ∈ (0,1)，and
𝑔(𝑥) = {
0 𝑖𝑓 0 ≤ 𝑥 ≤ 1 − 𝛼
1 𝑖𝑓 1 − 𝛼 ≤ 𝑥 ≤ 1
Then𝑉𝑎𝑅𝛼can be expressed as a distortion risk measure, that is
𝑉𝑎𝑅𝛼(𝑋) = 𝜌𝑔(𝑋)
Example 2 Suppose 𝑋 ∈ 𝒳，𝛼 ∈ (0,1)
𝑔(𝑥) = 𝑚𝑖𝑛 (
𝑥
1 − 𝛼
, 1)
If 𝑥 ∈ [0,1],then 𝐶𝑉𝑎𝑅𝛼(𝑥, 𝑃)can be expressed as a distortion risk measure, that is
𝐶𝑉𝑎𝑅𝛼(𝑥, 𝑃) = 𝜌𝑔(𝑋)
IV. Robust model under Gelbrich uncertainty set
This section considers the Gelbrichambiguity set defined on the Gelbrich distance, where the distribution of
returns is uncertain, but its first-order moment mean and second-order moment variance have certain limitations.
In this paper, we suppose that there are 𝑛 risky assets in the market. The return on the portfolio is𝑅(𝑥, 𝜉) =
𝑥𝑇
𝜉.The ratio of investment in the variousrisk asset can be represented as𝑥 = (𝑥1, 𝑥2, … , 𝑥𝑛)𝑇
, and the return
is𝜉 = (𝜉1, 𝜉2, … , 𝜉𝑛).
1.3 Gelbrich Uncertainty Set
Definition 2[19]Gelbrich distance: The Gelbrich distance between two mean-variance paris(𝜇1, Σ1),(𝜇2, Σ2) ∈
𝑅𝑛
× 𝕊+
𝑛
can be given by
𝐺((𝜇1, Σ1), (𝜇2, Σ2)) = √‖𝜇1 − 𝜇2‖2 + 𝑇𝑟 [Σ1 + Σ2 − 2 (Σ2
1
2Σ1Σ2
1
2)
1
2
] #(5)
We can proof that Gelbrich distance non-negative, symmetric and sub-additive. Only when(𝜇1, Σ1) = (𝜇2, Σ2),
this distance equals to 0.
Definition 3[19]Gelbrich ambiguity set: suppose
𝒰𝜋(𝜇̂, Σ
̂) = ,(𝜇, Σ) ∈ 𝑅𝑛
× 𝕊+
𝑛
: 𝐺 ((𝜇, Σ), (𝜇̂, Σ
̂)) ≤ 𝜋 - #(6)
Then Gelbrichambiguity set is given by
𝒢𝜋(𝜇̂, Σ
̂) = ,𝒬 ∈ 𝒮: (𝐸𝒬[𝜉], 𝐸𝒬 *(𝜉 − 𝐸𝒬[𝜉])(𝜉 − 𝐸𝒬[𝜉])
𝑇
+) ∈ 𝒰𝜋(𝜇̂, Σ
̂)- #(7)
Proposition 6 For the set 𝒰𝜋(𝜇̂, Σ
̂) = ,(𝜇, Σ) ∈ 𝑅𝑛
× 𝕊+
𝑛
: 𝐺 ((𝜇, Σ), (𝜇̂, Σ
̂)) ≤ 𝜋 -and𝒱𝜋(𝜇̂, Σ
̂) = {(𝜇, 𝑀) ∈
𝑅𝑛
× 𝕊+
𝑛
: (𝜇, M − 𝜇𝜇𝑇
) ∈ 𝒰𝜋(𝜇̂, Σ
̂)}, they are compact and convex.
1.4 Robust distortion risk measure model with Gelbrich Uncertainty Set
In this section, we establish the robust mean-distortion risk measure model with Gelbrich uncertainty set and
introduce the simplified expression of the objective function.

Proposition 7 If 𝜌(𝑋)is a law-invariant, positive homogenous risk measure, then the Gelibrich risk if the
portfolio loss function 𝑙(𝜉) = −𝑥𝑇
𝜉is given by
sup
𝒬∈𝒢𝜋(𝜇
̂,Σ
̂)
𝜉) = −𝜇̂𝑇
𝑥 + 𝛼√𝑥𝑇Σ
̂𝑥 + 𝜌√1 + 𝛼2‖𝑥‖#(8)
where
𝛼(𝜇, Σ, 𝑥) = sup𝒬∈(𝒞(𝜇,Σ)) 𝜌𝒬 (−
𝑥𝑇(𝜉−𝜇)
√𝑥𝑇Σ𝑥
) #(9)
is called standard risk coefficient.
Proposition 8[20] If the distortion risk function 𝑔 is a right continuous function, 𝜌𝒬is the distortion risk
measure based on the distortion risk function 𝒬 ∈ ℳ,𝒮 = ℳ2, then the standard risk coefficient for the
distortion risk measure is given by
𝛼 = .∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2
𝑑𝜏 − 1
1
0
/
1
2
#(10)
where 𝑔′
𝑐𝑣𝑥
denotes the derivative of the convex envelope of 𝑔, which exists almost everywhere. The proof of
thisproposition can be found in the article of Li.
Therefore, when the distortion risk measure satisfies the properties of coherent and law-invariant, then
sup𝒬∈𝒢𝜋(𝜇
̂,Σ
̂) 𝜌𝒬(−𝑥𝑇
𝜉) describes the worst-case distortion risk, the robust distribution uncertainty model based
on Gelbrich uncertainty sets can be represented as
𝑚𝑖𝑛
𝑥∈Ω
sup
̂,Σ
̂)
𝜉) = 𝑚𝑖𝑛
𝑥∈Ω
− 𝜇̂𝑇
𝑥 + 𝛼√𝑥𝑇Σ
̂𝑥 + 𝜋√1 + 𝛼2‖𝑥‖
= 𝑚𝑖𝑛
𝑥∈Ω
− 𝜇̂𝑇
𝑥 + .∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2
𝑑𝜏 − 1
1
0
/
1
2
√𝑥𝑇Σ
̂𝑥 + 𝜋√∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2𝑑𝜏
1
0
‖𝑥‖#(11)
If the constraint of returnis added, this model can be written
𝑚𝑖𝑛
𝑥∈Ω
sup
̂,Σ
̂)
𝜉) =
𝑚𝑖𝑛
𝑥∈Ω
− 𝜇̂𝑇
𝑥 + .∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2
𝑑𝜏 − 1
1
0
/
1
2
√𝑥𝑇Σ
̂𝑥 + 𝜋√∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2𝑑𝜏
1
0
‖𝑥‖#(12)
𝑠. 𝑡. 𝐸𝒬[𝜉] ≥ 𝑟0
According to the proposition raised by Viet [21](2019), any external distribution𝒬that attains the Gelbrich
risk of loss function 𝑙(𝜉) = −𝑥𝑇
𝜉has the same mean 𝜇∗
and covariance matrix Σ∗
.
Then the constraint of return is changed to this form
𝐸𝒬[𝜉] = 𝑥𝑇
𝜇̂ −
𝜋𝑥𝑇
𝑥
√1 + 𝛼2‖𝑥‖
= 𝑥𝑇
𝜇̂ −
𝜋𝑥𝑇
𝑥
‖𝑥‖ ∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2𝑑𝜏
1
0
≥ 𝑟0#(13)
V. Robust model with safety criteria
1.5 Robust model with original safety criteria constraint
This section will refer to the safety criteria proposed by Li (2017) as constraints for robust optimization
models. This constraint mainly indicates that investors assign different weights to the best and worst return rates,
and comprehensively calculate their expected return rate under their risk preference, and require that the return
meet certain goals. Similarly, if the objective function is a linear combination of the best- and worst-case
scenarios, the final model will be as follows:

min (1 − 𝜃) inf
𝒬∈𝒯
𝜉) + 𝜃 sup
𝒬∈𝒯
𝜉)
𝑠. 𝑡. (1 − 𝜃) sup
𝒬∈𝒯
𝒬∈𝒯
𝐸𝒬(𝑅(𝑥, 𝜉)) ≥ 𝜙
#(14)
Among them, the return function 𝑅(𝑥, 𝜉) = 𝑥𝑇
𝜉, 𝜙is the target for investors, 𝜃 ∈ [0,1], that characterizes
the risk aversioncoefficient of an investor for uncertainty. When 𝜃 = 0,investors prefer more aggressive
investment strategies, which only consider the maximum return in the best case. Once the maximum return
exceeds the established target, this investment portfolio at least satisfies investors' risk preferences.
When𝜃approaches 1, it indicates that investors are more conservative and pay more attention to the certainty
and stability of returns, considering the achievement of return targets in worst-case scenarios.
Definition 4 Assuming a random vector of returns 𝜉(𝜇, Σ) ∈ 𝑅𝑛
is derived from the following uncertain set
𝒰(𝜇̂, Σ
̂) = {(𝜇̅, Σ
̅) ∈ 𝑅𝑛
× 𝑆𝑛
:
(𝜇̅ − 𝜇̂)𝑇
Σ
̂−1(𝜇̅ − 𝜇̂) ≤ 𝛾1
Σ
̅ ≼ 𝛾2Σ
̂
} #(15)
𝒰(𝜇̂, Σ
̂) includes the information on the mean vector and covariance matrix, 𝛾1,𝛾2represent the degree of the
uncertainty. The higher figures of 𝛾1,𝛾2, the higher level of ambiguity.
Proposition 9 Suppose a random vector of return 𝜉(𝜇, Σ) ∈ 𝑅𝑛
is derived from the uncertain distribution set 𝒯,
and the robust optimization model under original safety criteria constraints can be expressed in the following
form
min
𝑥∈𝑅𝑛
− 𝜇̂𝑇
𝑥 +[(2𝜃 − 1)√𝛾1 + 𝛿𝜃√𝛾2] √𝑥𝑇Σ
̂𝑥
𝑠. 𝑡. 𝜇̂𝑇
𝑥 + (1 − 2𝜃)√𝛾1
√𝑥𝑇Σ
̂𝑥 ≥ 𝜙#(16)
𝛿 = (∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2
𝑑𝜏 − 1
1
0
)
1
2
denotes risk appetites of investors induced by different distortion risk functions is
related to the parameters of the distortion risk function.
1.6 Robust model with probabilistic safety criteria constraint
In real world, investors do not strictly pursue a return level that meets their personal investment goals with a
probability of 100%. Investors often exhibit a certain risk preference and overconfidence. They believe that a
certain investment portfolio can exceed a predetermined return rate with a probability of more than 95%, which
is acceptable. Therefore, this section will refer to Li's (2017) article and change the expected return constraint to
a constraint under probability weighting, which introduces probability and parameters 𝜃. The specific constraint
formula for representing the risk preference of investors' investment strategies is
𝑃 {(1 − 𝜃) sup
𝒬∈𝒯
𝒬∈𝒯
𝐸𝒬(𝑅(𝑥, 𝜉)) ≤ 𝜙} ≤ 𝜀 #(17)
𝜙is the predetermined return target for investors, while the probability constraint 𝜀is generally (0,0.5).
Proposition 10[21]The constraint of probabilistic safety criteria can be given by the following form
𝛾2
𝜀
𝑥𝑇
Σ
̂𝑥 − ∑ ,(𝜇̂𝑘
𝑇
𝑥 − 𝜇̂𝑘−1
𝑇
𝑥) *𝜇̂𝑘
𝑇
𝑥 + 𝜇̂𝑘−1
𝑇
𝑥 + 2(1 − 2𝜃)√𝛾1
√𝑥𝑇Σ
̂𝑥 − 2𝜙+-
𝑛
𝑘=1
≤ 0
The process of the proof uses the Chebyshev’s inequality and Bhat [22]’s result.
Proposition 11 Suppose a random vector of return 𝜉(𝜇, Σ) ∈ 𝑅𝑛
is derived from the uncertain distribution set
𝒯, the distortion risk measure robust optimization model based on theprobabilistic safety criterion can be
transformed into the following form:
min
𝑥∈𝑅𝑛
−𝜇̂𝑇
𝑥 + [(2𝜃 − 1)√𝛾1 + 𝛿𝜃√𝛾2]√𝑥𝑇Σ
̂𝑥

𝑠. 𝑡.
𝛾2
𝜀
𝑥𝑇
Σ
̂𝑥 − ∑ ,(𝜇̂𝑘
𝑇
𝑥 − 𝜇̂𝑘−1
𝑇
𝑥) *𝜇̂𝑘
𝑇
𝑥 + 𝜇̂𝑘−1
𝑇
𝑥 + 2(1 − 2𝜃)√𝛾1
√𝑥𝑇Σ
̂𝑥 − 2𝜙+-
𝑛
𝑘=1
≤ 0
𝛿 = (∫ 𝑔′
𝑐𝑣𝑥
(𝜏)2
𝑑𝜏 − 1
1
0
)
1
2
denotes risk appetites of investors induced by different distortion risk functions is
related to the parameters of the distortion risk function.
VI. Empirical analysis
1.7 Empirical data
The sample data in this empirical analysis is mainly based on the Shanghai and Shenzhen stock markets, and
10 representative stocks from different industries, different listing times, and low yield correlation are selected
from China's A-share market. They are Longji Machinery (002363), Yiling Pharmaceutical (002603), Suzhou
Bank (002966), Compass (300803), Nongfa Seed Industry (600313), China People's Insurance Corporation
(601319), Laiyifen (603777), Sanqi Mutual Entertainment (002555), Qingdao Beer (600600), and Jingsheng
Electromechanical (300316). The sample data has the time range of 1 year, from January 1, 2022 to December
31, 2022, and the corresponding yield is calculated based on the opening and closing prices of each trading day.
The following pictures mainly depict the closing price trends of the above 10 stocks. It can be seen that there are
still differences in the trends of these stocks, which also verifies the low correlation between the selected stocks.
Figure 1 Closing price trend figures of 10 stocks
Figure 2 Daily return rate fluctuation figures of 10 stocks
Table 1 and Figure 3 depict the annual return, annual standard deviation, minimum, maximum, kurtosis,
skewness, and correlation matrices of each stock, respectively.

Table 1 Statistics of 10 stocks
Code N Return SD min max kurtosis skewness
002363 241 35.39% 0.558 -10.07% 10.08% 4.35 0.284
002603 241 65.40% 0.719 -10.00% 10.02% 2.93 0.318
002966 241 29.84% 0.239 -4.79% 5.21% 3.54 0.081
300803 241 30.89% 0.563 -11.03% 20.00% 7.39 1.187
600313 241 78.10% 0.657 -10.01% 10.07% 3.40 0.247
601319 241 16.33% 0.250 -6.34% 6.04% 5.24 0.322
603777 241 41.52% 0.539 -10.00% 10.01% 3.82 0.148
002555 241 -30.87% 0.436 -10.00% 10.03% 5.33 0.169
600600 241 17.23% 0.416 -8.75% 9.69% 4.99 0.608
300316 241 9.20% 0.457 -9.57% 11.96% 4.78 0.398
Figure 3 Correlation matrices of each stock
1.8 Empirical results under Gelbrich Uncertainty set
We take these two functions 𝑔(𝑥) = 𝑚𝑖𝑛 (
𝑥
1−𝛼
, 1)and 𝑔(𝑥) = 1 − (1 − 𝑥)1+𝜆
with 𝛼 = 0.95as example.
The left figure in Figure 4 depicts the changes of effective frontier when π is taken as 0.1, 0.2and 0.3. The figure
on the right shows the changes in the investment ratio of each asset under different Gelbrich distances. It can be
observed that the smaller the value of Gelbrich distance, the more its effective frontier spreads outward. This
property can be
derived from the
Gelbrich distance
formula.

Figure 4 The results of different Gelbrich distances π
The parameter 𝜆 measures the loss probability of investors adjusting different positions under subjective
conditions, indicating their risk preference. Therefore, the empirical results analyzed that under the fixed
Gelbrich distance𝜆. Please refer to Figure 5 for the specific changes in the effective frontier and investment ratio.
Figure 5 The results of different parameter 𝜆
1.9 Empirical results under safety criteria
In this section, we use the bootstrapping method to estimate the parameters 𝛾1and 𝛾2. Also, to well show the
change of the weight of each asset, we refer to the research of Li and consider only 3 stocks here. Through the
bootstrapping procedure, we get 𝛾
̂1 = 0.2140and 𝛾
̂2 = 0.1003.
When𝛼 and 𝜆equal to 0.95 and 5, the results are shown in the following figure. Figure 6 shows the change of
the effective frontier and investment proportion when 𝛼 = 0.95 and the aversion for uncertainty𝜃 with values

of 0.5, 0.8and 1. It can be clearly seen from the figure that when 𝜃gets higher, the higher the expected return
rate, and the smaller the value of the best and worst distortion risk measure. Investors have a stronger attitude
towards avoiding uncertainty.
Figure 6 The results of different parameter𝜃
For the probabilistic safety criteria, the empirical results are shown in the figure 7.It can be observed that
when 𝜃get larger, the more the effective frontier moves to the right, and the proportion of investment in stock
002966 also increases.

Figure 7 The results of different parameter𝜃
VII. Conclusion
The article studies the relevant properties of distortion risk measures based on distortion functions, and
analyzes and solves the optimization model of mean distortion risk measures on this basis. Before model
analysis, uncertain sets with different definitions were introduced to examine the changes in the effective
frontier of the model under different uncertain sets. Further improve the constraint conditions of the model,
introduce safety criteria constraints, and solve the changes in the optimal solution of the model under changes in
relevant parameters. The results indicate that the effective frontier of the probability safety criterion shifts to the
right relative to the original model, meaning that investors bear greater risk at the same rate of return. At the
same time, the smaller the risk aversion coefficient of the uncertainty set, the more frontier moves to the left and
up, which is consistent with the conclusion under general safety criteria.
References
[1] Markowitz H M. Portfolio selection[J]. The Journal of Finance, 1952, 7(1):77.
[2] Duffie D, Pan J. An overview of value at risk[J]. Journal of derivatives, 1997, 4(3): 7-49.
[3] Artzner P, Delbaen F, Eber J M, et al. Coherent Measures of Risk[J]. Mathematical Finance, 1999,
9(3):203-228.
[4] Delbaen,Freddy. Coherent risk measures on general probability spaces[J]. Advances in finance and
stochastics: essays in honour of Dieter Sondermann, (2002): 1-37.
[5] Acerbi C. Spectral measures of risk: A coherent representation of subjective risk aversion[J]. Journal of
Banking & Finance, 2002, 26(7):1505-1518.
[6] Wang S . Premium Calculation by Transforming the Layer Premium Density[J]. ASTIN Bulletin, 1996,
26.
[7] Wang S S . A Universal Framework for Pricing Financial and Insurance Risks[J]. Cambridge

University Press, 2002(02).
[8] A Balbás, Garrido J , Mayoral S . Properties of Distortion Risk Measures[J]. Springer US, 2009(3).
[9] Feng M B, Tan K S. Coherent Distortion Risk Measures in Portfolio Selection[J]. Systems Engineering
Procedia, 2012, 4:25-34.
[10] Cai J , Wang Y , Mao T . Tail subadditivity of distortion risk measures and multivariate tail
distortion risk measures[J]. Insurance Mathematics and Economics, 2017, 75(jul.):105-116.
[11] Gelbrich M. On a formula for the L2 Wasserstein metric between measures on Euclidean and Hilbert
spaces[J]. MathematischeNachrichten, 1990, 147(1): 185-203.
[12] Delage, Erick, and Yinyu Ye. Distributionally Robust Optimization Under Moment Uncertainty with
Application to Data-Driven Problems. Operations Research, 2010, 58(3): 595–612.
[13] Roy, A.D. (1952) Safety First and the Holding of Assets. Econometrica, 20, 431-449.
[14] TurnovskyP . The Dynamics of Government Policy in an Inflationary Economy" An `Intermediate-Run'
Analysis. 1975.
[15] Kataoka, Shinji. A Stochastic Programming Model. Econometrica 31, no. 1/2 (1963): 181–96.
[16] Ghaoui L E, Oks M, Oustry F. Worst-case value-at-risk and robust portfolio optimization: A conic
programming approach[J]. Operations research, 2003, 51(4): 543-556.
[17] Zhu M S. Worst-Case Conditional Value-at-Risk with Application to Robust Portfolio Management[J].
Operations Research, 2009, 57(5):1155-1168.
[18] Kang Z, Li X, Li Z. Mean-CVaR portfolio selection model with ambiguity in distribution and
attitude[J]. Journal of Industrial and Management Optimization, 2020, 16(6): 3065-3081.
[19] Nguyen, Viet Anh, et al. Mean-covariance robust risk measurement. arXiv preprint
arXiv:2112.09959 (2021).
[20] Li J Y M. Closed-form solutions for worst-case law invariant risk measures with application to robust
portfolio optimization[J]. Operations Research, 2018, 66(6): 1533-1541.
[21] Shihan Di, Dong Ma, Peibiao Zhao. α-robust portfolio optimization problem under the distribution
uncertainty. Journal of Industrial and Management Optimization, 2023, 19(4): 2528-2548.
[22] Rujeerapaiboon N, Kuhn D, Wiesemann W. Chebyshev inequalities for products of random variables[J].
Mathematics of Operations Research, 2018, 43(3): 887-918.

�ݺ�ߣ

Mean-Distortion Risk Measure Portfolio Optimization under the Distribution Uncertainty

More Related Content

Mean-Distortion Risk Measure Portfolio Optimization under the Distribution Uncertainty