SEARCH

SEARCH BY CITATION

Keywords:

  • Distortion;
  • risk appetite;
  • risk measures;
  • subadditivity;
  • tails

Abstract

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

We propose a new family of risk measures, called GlueVaR, within the class of distortion risk measures. Analytical closed-form expressions are shown for the most frequently used distribution functions in financial and insurance applications. The relationship between GlueVaR, value-at-risk, and tail value-at-risk is explained. Tail subadditivity is investigated and it is shown that some GlueVaR risk measures satisfy this property. An interpretation in terms of risk attitudes is provided and a discussion is given on the applicability in nonfinancial problems such as health, safety, environmental, or catastrophic risk management.

1. INTRODUCTION

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

Financial and insurance risk management practitioners typically have to deal with two opposing demands: on the one hand, they want business units to achieve or outperform the objectives fixed by the firm's executive committee, yet, on the other, they are responsible for controlling their economic risks. Finding a tradeoff between these two demands is the challenging task that risk managers face on a daily basis. At the same time, they need to decide how risk should be quantified.

Financial and insurance firms are subject to the capital requirements established by regulators' guidelines and directives. These requirements are typically equal to, or proportional to, a risk measure value that determines a minimum cushion of economic liquidity. The selection of such risk measures and tolerance levels is crucial, therefore from the regulators' point of view.

Our aim is to propose a new family of risk measures, which we name GlueVar, and that have analytical closed-form expressions for many statistical distributions that are frequently used in financial and insurance applications. This new family combines the most popular risk measures and considers more than just one parameter to capture managerial and regulatory attitudes toward risk.

Financial institutions and insurance companies prefer to minimize the level of capital reserves required by solvency regulations because they must contend with many restrictions on how this capital can be invested and, as such, the return on their capital reserves is usually lower than that provided by other opportunities. For this reason, companies typically favor regulations that impose risk measures and tolerance levels that are not overly conservative. Managers also prefer simple, straightforward risk measures rather than more complicated alternatives, since they claim that the former are more easily communicated.

From the regulators' perspective, controlling the risk of financial institutions and insurance companies is fundamental in order to protect consumers and investors, which may have conflicting objectives. Strict solvency capital requirements may limit the capacity of firms, but they also reassure consumers and guarantee the position of the financial industry in the economy. Thus, the debate as to what constitutes a suitable risk measure and what represents a suitable tolerance level is interminable, without their apparently having been much investigation as to what might represent an appropriate compromise.

We contend that the GlueVaR family could be useful in helping regulators and practitioners reach a consensus. As we discuss below, the GlueVaR family should enhance the way in which regulatory capital requirements are calculated, as GlueVaR can incorporate more information about agents' attitudes to risk. It is our belief that the incorporation of qualitative information in decision-making tools is essential for risk managers and, as such, the GlueVaR risk measures can play a key role in achieving this goal.

2. BACKGROUND AND MOTIVATION

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

Value-at-risk (VaR) has been adopted as a standard tool to assess the risk and to calculate capital requirements in the financial industry. Value-at-risk at level α is the α-quantile of a random variable X (which we often call loss), that is, VaRinline image, where inline image is the cumulative distribution function (cdf) of X and α is the confidence or the tolerance level inline image. However, VaR is known to present a number of pitfalls when applied in practice. A disadvantage when using VaR in the financial context is that the capital requirements for catastrophic losses based on the measure can be underestimated, that is, the necessary reserves in adverse scenarios may well be less than they should be. The underestimation of capital requirements may be aggravated when fat-tailed losses are incorrectly modeled by mild-tailed distributions such as the normal distribution. There are attempts to overcome this kind of model risk when using VaR or, at least, to quantify the risk related to the modeling. [1] A second drawback is that the VaR may fail the subadditivity property. A risk measure is subadditive when the aggregated risk is less than or equal to the sum of individual risks. Subadditvity is an appealing property when aggregating risks in order to preserve the benefits of diversification. VaR is subadditive for elliptically distributed losses.[2] However, the subadditivity of VaR is not granted.[3, 4]

Tail value-at-risk (TVaR) may be interpreted as the mathematical expectation beyond VaR, and is defined as TVaRinline image. The TVaR risk measure does not suffer the two drawbacks discussed above for VaR and, as such, would appear to be a more powerful measure for assessing the actual risks faced by companies and financial institutions. However, TVaR has not been widely accepted by practitioners in the financial and insurance industry. VaR is currently the risk measure contemplated in the European solvency regulation for the insurance sector (Solvency II), and this is also the case of solvency regulation for the banking sector (Basel accords). The TVaR measures average losses in the most adverse cases rather than just the minimum loss, as the VaR does. Therefore, capital reserves based on the TVaR have to be considerably higher than those based on VaR and significant differences in the size of capital reserves can be obtained depending on which risk measure is adopted.

This article is motivated, therefore, by an attempt to respond to the following question. Can a risk measure be devised that would provide a risk assessment that lies somewhere between that offered by the VaR and the TVaR? To this end, we propose a new family of risk measures (GlueVaR), which forms part of a wider class referred to as distortion risk measures. We analyze the subadditivity properties of these GlueVaR risk measures and show that a subfamily of GlueVaR risk measures satisfies tail-subadditivity.

GlueVaR risk measures are defined by means of a four-parameter function. By calibrating the parameters, GlueVaR risk measures can be matched to a wide variety of contexts. Specifically, once a confidence level has been fixed, the new family contains risk measures that lie between those of VaR and TVaR and that may adequately reflect the risk of mild-tailed distributed losses without having to resort to VaR. In certain situations, however, more conservative risk measures even than TVaR may be preferred. We show that these highly conservative risk measures can also be defined by means of the GlueVaR family. We derive analytical closed-form expressions of GlueVaR for commonly used statistical distributions in the financial context. These closed-form expressions should enable practitioners to undertake an effortless transition from the use of VaR and TVaR to GlueVaR.

3. DISTORTION RISK MEASURES

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

Consider a probability space and the set of all random variables defined on this space. Any risk measure[5] ρ is a mapping from the set of random variables to the real line inline image, inline image. Distortion risk measures were introduced by Wang [6, 7] and are closely related to the distortion expectation theory. A review on how risk measures can be interpreted from several perspectives is provided in Ref. [8], and a clarifying explanation of the relationship between distortion risk measures and distortion expectation theory is included. A detailed literature review of distortion risk measures is available in Refs. [9] and [10]. There are two key elements to define a distortion risk measure: first, the associated distortion function; and, second, the concept of the Choquet[11] Integral. The distortion function, Choquet Integral, and the distortion risk measure can be defined as follows:

  • Distortion function. Let inline image be a function such that inline image, inline image, and g is nondecreasing. Then g is called a distortion function.
  • Choquet Integral. The (asymmetric) Choquet Integral with respect to a set function μ of a μ-measurable function inline image is denoted as inline image and is equal to inline image, if inline image, where inline image denotes the survival function of X with respect to μ. Note that Ω denotes a set, which in financial and insurance applications is the sample space of a probability space. A set function μ in this context is a function defined from 2Ω (the set of all subsets of Ω) to inline image. A μ-measurable function X is, widely speaking, a function defined on Ω so that expressions like inline image or inline image make sense. See Ref. [12] for more details.
  • Distortion risk measure. Let g be a distortion function. Consider a random variable X and its survival function inline image. Function inline image defined by inline image is called a distortion risk measure.

From the previous definitions, it is straightforward to see that for any random variable X, inline image is the Choquet Integral of X with respect to the set function inline image, where P is the probability function associated with the probability space in which X is defined.

The mathematical expectation is a distortion risk measure whose distortion function is the identity function, [9] inline image. Therefore, a straightforward way to interpret a distortion risk measure is as follows: first, the survival function of the random variable is distorted inline image; second, the mathematical expectation of the distorted random variable is computed. From a theoretical point of view, note that this interpretation fits the discussion that risk may be defined as an expected value in many situations. [13]

VaR and TVaR measures are in fact distortion risk measures. The associated distortion functions of these risk measures are shown in Table 1.

Table 1. VaR and TVaR Distortion Functions
Risk MeasureDistortion Function
VaRinline image
TVaRinline image
For a confidence level inline image.

Based on the distortion functions shown in Table 1, once α is fixed it can be proved that VaRinline image TVaRinline image for any random variable X.

Remark 3.1. Let g and inline image be two distortion functions and let inline image and inline image be their respective distortion risk measures. Suppose that inline image for all inline image. Then inline image for any random variable X.

This result follows immediately from the definition of distortion risk measures because:

  • display math

4. A NEW FAMILY OF RISK MEASURES: GlueVaR

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

We define a new family of risk measures, named GlueVaR. Any GlueVaR risk measure can be described by means of its distortion function. Given a confidence level α, the distortion function for GlueVar is:

  • math image(1)

where inline image so that inline image, inline image, and inline image. Parameter β is the additional confidence level besides α. The shape of the GlueVaR distortion function is determined by the distorted survival probabilities h1 and h2 at levels inline image and inline image, respectively. We call parameters h1 and h2 the heights of the distortion function.

A wide range of risk measures may be defined under this framework. Note that VaRα and TVaRα are particular cases of this new family of risk measures. Namely, for a random variable X, VaRinline image and TVaRinline image correspond to distortion functions inline image and inline image, respectively. By establishing suitable conditions on the heights h1 and h2, the GlueVaR family is very flexible. For example, risk managers might like to select α, β, h1, and h2 so that VaRinline image GlueVaRinline image TVaRinline image: this can be achieved by selecting a set of parameters for their associated distortion functions to ensure that inline image for any inline image following Remark 3.1, that is, by forcing condition inline image. An example of such a case is shown in Fig. 1 (left-hand side).

image

Figure 1. Examples of GlueVaR distortion functions. Left. Distortion function is concave in inline image and VaRinline image GlueVaRinline image TVaRinline image for a random variable X; Right. Distortion function is concave in the whole range [0, 1] and TVaRinline image GlueVaRinline image TVaRinline image for a random variable X.

Download figure to PowerPoint

The GlueVaR family also allows us to define a highly conservative risk measure GlueVaRinline image, so that TVaRinline image TVaRinline image for any X and that the associated distortion function inline image is concave in [0, 1]. In this case, inline image and inline image must be fulfilled, as occurs in the example shown in Fig. 1 (right-hand side).

5. LINEAR COMBINATION OF RISK MEASURES

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

Given a random variable X and for fixed tolerance levels α and β so that inline image, GlueVaRinline image can be expressed as a linear combination1 of TVaRinline image, TVaRinline image, and VaRinline image. This result allows us to translate the initial graphical-based construction of GlueVaR risk measures into an algebraic construction based on standard risk measures.

If the following notation is used,

  • math image(2)

then the distortion function inline image in Equation (1) may be rewritten as (details can be found in Appendix A1):

  • display math(3)

where inline image, inline image, inline image are the distortion functions of TVaR at confidence levels β and α and of VaR at confidence level α, respectively (see Table 1). Therefore, GlueVaR is a risk measure that can be expressed as a linear combination of three risk measures: TVaR at confidence levels β and α and VaR at confidence level α,

  • display math(4)

Given this relationship, some abuse of notation may be employed for GlueVaRinline image and its related distortion function. The notation GlueVaRinline image or inline image may, on occasions, be preferred to that based on heights h1 and h2. The bijective relationship between pairs inline image and inline image is shown in Appendix B.

5.1. Analytical Closed-Form Expressions of GlueVaR

A useful consequence of Equation (4) is that when analytical closed-form expressions of VaRinline image and TVaRinline image are known for a random variable X, we can automatically derive the closed-form expression of GlueVaRinline image without further complications. Otherwise, using the definition of GlueVaR as a distortion risk measure, the Choquet Integral of X with respect to the set function inline image should be calculated.

5.1.1. Illustration: GlueVaR Expression for Student t Distribution

If X is a random variable such that inline image is distributed as a Student t random variable with ν degrees of freedom (df)2, then:

  • display math
  • display math

where inline image is the α-quantile of a Student t distribution with ν df and τ is its density function.

Using Equation (4) the GlueVaR of X random variable is:

  • math image
5.1.2. Analytical Expressions for Other Frequently Used Distributions

Normal (inline image), log-normal (inline image), and generalized Pareto (inline image) distributions3 have simple closed-form expressions of GlueVaR. Notation conventions are used. Namely, ϕ and Φ stand for the standard normal pdf and cdf, respectively. The standard normal distribution α and β quantiles are denoted as inline image and inline image. For the inline image distribution, the definition provided in Ref. [18] is considered, where the scale parameter is denoted by σ and k is the shape parameter. The inline image distribution contains the uniform (inline image), the exponential (inline image), the Pareto (inline image), and the type II Pareto (inline image) distributions as special cases. Closed-form expressions of GlueVaR for several distributions are presented in Table 2.

Table 2. Closed-Form Expressions of GlueVaR for Some Selected Distributions
DistributionGlueVaRinline image Expression
Normal: inline imageinline image
Log-normal: inline imageinline image
Exponential: inline image, with inline imageinline image
Pareto: inline image, with inline imageinline image
Type II Pareto: inline image, with inline imageinline image
 inline image

6. SUBADDITIVITY IN THE TAIL

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

This section is devoted to an analysis of the properties of the GlueVaR family of risk measures, with special attention to subadditiviy. Our reason for defining these GlueVaR risk measures is a response to the concerns expressed by risk managers regarding the choice of risk measures in the case of regulatory capital requirements. However, an axiomatic approach to define or represent risk measures is more frequent in the literature.[3, 19-26]

In a seminal article,[3] the following set of axioms that a risk measure should satisfy was established: positive homogeneity, translation invariance, monotonicity, and subadditivity. Authors referred to such risk measures as “coherent risk measures.” Distortion risk measures always satisfy the first three properties, but subadditivity is only guaranteed when the distortion function is concave.[12, 27, 28] Therefore, VaR, unlike TVaR, is not coherent. In some situations, coherence of risk measures is a requirement [29] but, nonetheless, some criticisms can be found. [30] Additional properties for distortion risk measures are provided in Refs. [31] and [11].

As shown in the previous section, GlueVaR risk measures may be interpreted as a linear combination of VaR and TVaR risk measures. Therefore, a GlueVaR risk measure is coherent when the weight assigned to VaR is zero and the weights of TVaR are nonnegative. In terms of the parameters of the distortion function, GlueVaR is subadditive (and thus coherent) if inline image and inline image. More generally, any property satisfied by TVaR but not by VaR will be inherited by GlueVaR if inline image and inline image in expression (2).

Subaddtitivity in the whole domain is a strong condition. When dealing with fat tail losses (i.e., low-frequency and large-loss events), risk managers are especially interested in the tail region. Fat right tails have been extensively studied in insurance and finance. [32-37] To the best of our knowledge, however, previous studies of the subadditivity of risk measures in the tail region are scarce.[38, 39] The milder condition of subadditivity in the tail region is investigated here.

We introduce the concept of subadditivity in the right tail for a pair of risks. Note that if interested in the left—as opposed to the right tail, a simple change of sign in the random variable suffices. Subadditivity in the right tail is defined in this discussion for distortion risk measures. Consider a probability space with sample space Ω. Let inline image the α-quantile of random variable Z, inline image. Let inline image be defined by inline image, so inline image means here the tail region of random variable Z given a confidence level α. Let inline image be two risks defined on the same probability space. When aggregating two risks, the common tail for both risks must be taken into account. This common tail region is defined here as follows: inline image.

Definition 6.1. Given a confidence level inline image, a distortion risk measure inline image is subadditive in the tail for the pair inline image if inline image and

  • display math

where the integral symbol stands for Choquet Integrals with respect to the set function inline image.

When there is no ambiguity as to which confidence level α and random variables X, Y are taken into account, tail-subadditivity is used to refer to this property. If notation inline imageinline image is introduced, the integral condition used in the definition can be rewritten, in terms of survival functions, as:

  • math image

Theorem 6.1. Given a confidence level α and a pair of risks X and Y so that inline image, a GlueVaR risk measure is tail-subadditive if its associated distortion function inline image is concave in inline image.

The proof is contained in Appendix C.

Tail-subadditivity is a desirable property because it implies that the benefits of diversification may not be valid in every situation but, at least, they hold in extreme cases.

Note that, in terms of parameters h1 and h2, a GlueVaR risk measure may be tail-subadditive if, and only if, inline image, as a corollary of Theorem 6.1.

7. RISK ATTITUDES IN GlueVaR

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

An interesting interpretation in the context of decision making and risk management is that GlueVaR risk measures arise as a linear combination of three possible scenarios. So, two levels of severity can be fixed, namely, α and β, with inline image. Then, the risk can be measured in the highly conservative scenario with TVaR at level β; in the conservative scenario with TVaR at level α; and in the less conservative scenario with VaR at level α.

Each combination of these risk scenarios reflects a specific risk attitude. Therefore, we can say that the combination of these risk scenarios in this context is something that is directly identified by an explicit GlueVaR risk measure. To some extent, these risk attitudes could be related to risk appetite. [40]

From the practitioner's point of view, four parameters must be fixed to define the GlueVar risk measure. The α and β values correspond to the confidence levels used for bad and very bad scenarios, respectively. We could select, for instance, inline image and inline image, which is equivalent to one bad event every 20 years or one bad event every 200 years, respectively. The other two parameters are directly related to the weights given to these scenarios. For instance, we could say that the three components of GlueVaR in Equation (4) are equally important. This would imply inline image, so we could find the corresponding h1 and h2 parameters. When inline image and inline image, inline image, these parameters are inline image and inline image.

7.1. Geometrical Discussion on Risk Attitudes

Given α and β, the shaded areas in Fig. 2 delimit feasible weights inline image for GlueVaRinline image. The point inline image corresponds to a balanced risk attitude on the part of risk managers when faced by the three components shown in Equation (4). The corresponding distortion function inline image is concave on inline image in the lightly shaded area and, thus, the associated GlueVaR risk measure can be tail-subadditive. Yet, the distortion function is not concave on inline image in the darkly shaded area and, thus, the associated GlueVaR risk measure cannot be tail-subadditive. The distortion function is concave in [0, 1] in the boldest continuous segment and, thus, the associated GlueVaR risk measure is subadditive.

image

Figure 2. Given α and β, the shaded areas delimit feasible weights inline image for GlueVaRinline image.

Download figure to PowerPoint

If inline image, risk managers are optimistic regarding the impossibility of the occurrence of the worst-case scenario, and so attach a negative weight to it.

Note that any pair of weights inline image on the boldest line in Fig. 2 leads to inline image. This means that a zero weight is allocated to the less conservative scenario, that is, the one associated with the VaRinline image. This is indicative of the decisionmaker's conservative approach. Nonetheless, differences in just how restrictive this conservative attitude is can be found among the weights lying on this line: the nearer to inline image, the less restrictive it is, while the nearer to inline image, the more conservative it is.

7.2. Illustration

Data for the cost of claims for property damage from a major Spanish motor insurer are used to illustrate the applicability of these results. These data contain inline image observations of the cost of individual claims in thousands of euros, and were analyzed in Refs. [41] and [42]. The risk measures for these data are displayed in Table 3. In the first row, our results obtained using the empirical distribution are presented. In subsequent rows normal, log-normal, Student t with 4 df and generalized Pareto distributions are fitted and their respective risk values are shown.

Table 3. Risk Measures for the Data on Claim Costs
 VaRinline imageTVaRinline imageTVaRinline imageGlueVaRinline image
Empirical38.8112.5440.0197.1
Normal78.996.1130.4101.8
Log-normal42.5106.3364.0170.9
Student t (4 df)99.0143.2272.1171.4
Pareto38.382.4264.5128.4
X stands for “cost of individual claims in thousands of euros.”
Notation GlueVaRinline image is used in this table.
For inline image, inline image, inline image, inline image, and inline image

The values in Table 3 indicate that the cost of individual claims is fat right-tailed: differences between TVaRinline image and VaRinline image, and also between TVaRinline image and TVaRinline image are huge for the empirical distribution (and also for the rest of the selected distributions). In this case, it seems clear that GlueVaR is more conservative than TVaR at the 95% level but less so than TVaR at the 99.5% level, independently of the selected distribution. On the other hand, GlueVaR is not, unlike TVaR, subadditive, but it is a candidate to be tail-subbadditive in many situations because its associated distortion function is concave in [0, 0.05).

Calculations have been made in R. R programs are available from the authors.

8. OTHER NONFINANCIAL APPLICATIONS

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

New risk measures based on distortion functions can be valuable outside the scope of finance and insurance. There is a natural bridge from financial applications to any discipline where the choice of a risk measure plays a role for decision making. GlueVaR risk measures can be applied to health, safety, environmental, adversarial risks, or catastrophic risks including terrorism. Health or safety regulations report quantile risks and could be enhanced with GlueVaR, as the latter allows to combine risk measures and risk levels.

A crucial feature is that in GlueVaR risk measures, the decisionmaker sets up two tolerance levels: one for the “bad cases” and another one for the “very bad cases.” Then, weights are set up according to a decisionmaker's or a regulator's risk attitude. If a decisionmaker is very conservative, then he should give all weight to the “very bad case” outcome, whereas a less cautious decisionmaker would assign all weight to a lesser “very bad case” outcome.4 An intermediate decisionmaker could set up a reasonable position, where he could balance a position between VaR versus TVaR, and a tradeoff between a lower tolerance level versus a higher tolerance level.

One good example of difficulties in agreeing upon a suitable risk measure is found in a recent article [43] on the optimal security investments and extreme risks with an application to terrorism risks. These authors focus on both amount and timing of security investment, but they fundamentally model risk based on the principle of expected net benefit of investment in security. An analogous principle is also used [44] to investigate security in information systems and was earlier proposed [45] for the same purpose. The expected net benefit of investment in security is a simple analytical model that maximizes the gain when investing in security procedures. Optimization is straightforward once a level of loss severity and a probability of occurrence are assumed (see section 'BACKGROUND AND MOTIVATION' in Ref. [43]). When addressing terrorism risk in form of intentional attacks on the food sector using chemical, biological, and radionuclear tools, several scenarios for casualties resulting from a terrorist attack are established in Ref. [43], which have a very low probability. The choice of those scenarios is done subjectively, even if there is a technical part that relies on extreme value theory to approximate the tail probabilities. At least two scenarios are necessary to carry out the optimization procedure and their corresponding anticumulative distributions. A GlueVaR risk measure could be used directly in the optimization procedure. Moreover, GlueVaRs could help to compare different contexts, such as terrorist attacks on the food sector versus other sectors, where the number of casualties could be much lower. GlueVaRs could provide a single value for every phenomenon and there would be no need to define interval scenarios, whose choice could potentially be controversial.

Extensions can be found in many other applications. Let us imagine an employer who has to manage worker compensation reserves. In order to assess the risk of being short in reserves in a one-year horizon, the classical procedure is to estimate VaR, which would equal the minimum reserve amount that would be likely to cover the compensations to be paid. If the manager has a tolerance of one in 100 years, that would correspond to a confidence level equal to 99% and VaR at that level should provide the estimate of sufficient reserve. In the GlueVaR framework, the manager could be slightly more cautious. He would recalculate the risk with a higher confidence and raise reserves accordingly, in case he might face a bad scenario. He could fix a confidence level of 99.5% or tolerate shortfall of reserves once in 200 years. The fund manager could select transitional position and assign weights so that a GlueVaR reflects his risk attitude. He could give equal weight to the two scenarios and be equally positioned with regard to the classical risk measure. Not only risk managers, but regulators could leave freedom to fund managers to choose their preferred GlueVaR risk measure to set up their reserves and thus, regulators could inform about the risk measure that is being used by the worker compensation fund managers.

Generalizations and extensions to disaster management could be implemented. A public safety agency could use GlueVaR risk measures to help better plan how many resources to set aside to meet the needs of next year's hurricane season. Let us assume that resources are established proportional to a VaR at certain level α. Similarly, as in the example concerning worker compensation, agencies could allow for a risk attitude, so that there would be two tolerance levels, α and β, and then weights would determine whether the agency would rather have a position in between those two levels, as it is natural to do with GlueVaRs.

Managers of scarce resources with uncertain supply and demand (e.g., strategic petroleum reserve, antibiotic stockpiles, blood bank, etc.) could use the GlueVaR measures instead of classical measures to improve decisions. Quantile-based risk measures are applied to solve a variety of optimization issues in presence of uncertainty in supply and/or demand, such as strategic planning for hospital care services, [46] operational planning of chemical and petrochemical plants,[47, 48] level of capacity in auto industry facilities, [49] water resources management[50, 51] or hydrocarbon supply-chain designs.[52, 53] Many decisions are taken based on an optimization procedure where an expected loss is minimized subject to the risk being lower than an upper bound. Risk could be defined by a GlueVaR rather than by a traditional risk measure. In that respect, GlueVaR would allow for defining the bad and very bad scenarios and would weight them according to risk aversion.

A related area where the application of GlueVaR measures might be useful is the reservoir management decision making. A set of percentiles are commonly computed to subdivide hydrocarbon reserve estimates into categories that describe the probability of extracting a certain volume, such as proven (P1), probable (P2), and possible (P3) reserves. [54-56] Management decisions are taken under these alternative scenarios. For instance, they might represent the bad, average, and good cases to evaluate the impact of additional drillings. A unified evaluation may be performed by means of the GlueVaR measures. These three scenarios may be jointly considered by the selection of tolerance levels and weights for the VaR and TVaR measures in accordance with the management decisionmaker's risk profile.

The choice of a risk measure for regulatory purposes in a matter of strong debate in the financial and the insurance sector, as solvency requirements limit the potential benefits of a firm; however, the discussion is not unique to that sector.

9. CONCLUSIONS

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

The VaR measure is broadly used to calculate financial requirements in finance and insurance. However, the VaR measure shows drawbacks when the risk is assessed in extremely adverse scenarios and whether risks are aggregated. The TVaR measure overcomes previous limitations but capital requirements based on the TVaR measure are often observed as excessively conservative by practitioners. A new family of flexible risk measures, called GlueVar, is proposed in this article as a bridge between both approaches.

We have shown that GlueVaR measures can be expressed as linear combinations of standard risk measures and that, similarly, they can be defined based on a straightforward distortion function. Attractive properties of a GlueVaR risk measure are, therefore, readily derived from the definition of its associated distortion function. This is the case of the tail-subadditivity property defined in this article. Basically, concavity of the distortion function on the subrange inline image assures tail-subadditivity. This milder condition in the distortion function than concavity over the whole range might be a sufficient requisite for risk measures when fat right-tail risks are assessed: the benefits of diversification are attained in adverse scenarios but capital requirements are not excessively high. An advantage of this new approach compared to previous types of VaR measures including different extensions available in the literature is flexibility and simplicity, as it is based on standard measures. We note, however, that we have limited the construction of the GlueVaR measures to the linear combination of two standard measures and two tolerance levels, but this could be extended to accommodate a larger number of elements.

The results provided in this article are directly applicable in the financial industry. Closed-form expressions of GlueVaR risk measures are shown for commonly used distributions in finance and insurance. We encourage regulators and financial and insurance risk managers to seek an equilibrium between their different demands. The two levels of qualitative information that GlueVaR risk measures incorporate (one related to the confidence levels of bad- and worst-case scenarios; the other related to the plausibility of those scenarios) can help achieve this goal. We believe that GlueVaR risk measures can play a leading role in helping to reach a satisfactory consensus.

There is potential for extending the application of GlueVaR measures to nonfinancial disciplines where the choice of a risk measure matters.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

The authors gratefully acknowledge the comments and suggestions from the editor and anonymous referees. This research is sponsored by the Spanish Ministry of Science ECO2010-21787-C03-01 and ECO2012-35584. Montserrat Guillén thanks ICREA Academia.

Appendix A

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

EQUIVALENT EXPRESSION FOR THE GlueVar DISTORTION FUNCTION

We provide details as to how to define the GlueVaR distortion function inline image as a linear combination of the distortion functions of TVaR at confidence levels β and α, and VaR at confidence level α, that is, details as to how to obtain expression (3). Expression (1) of the distortion function inline image can be rewritten as,

  • display math(A1)

where inline image is an indicator function so that it takes a value of 1 if inline image and 0 otherwise.

Note that:

  • display math(A2)
  • display math(A3)
  • math image(A4)

Taking into account expressions (A2), (A3), and (A4), expression (A1) may be rewritten as,

  • math image(A5)

Given that inline image, inline image, and inline image, expression (3) follows directly from Equation (A5).

Appendix B

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

BIJECTIVE RELATIONSHIP BETWEEN HEIGHTS AND WEIGHTS

Pairs of GlueVaR heights inline image and weights inline image are linearly related to each other. The parameter relationships are inline image and, inversely, inline image, where H and inline image matrices are inline image and inline image, respectively.

Appendix C

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References

TAIL-SUBADDITIVITY FOR GlueVar RISK MEASURES

This appendix is devoted to the proof of Theorem 6.1. Given a confidence level α and a pair of random variables X and Y so that inline image, a GlueVaR risk measure is tail-subadditive if its associated distortion function inline image is concave in inline image.

The subadditivity theorem and the integration on subsets of Ω are defined as: [12]

  • Subadditivity theorem. Let inline image be a monotone, submodular set function. Then for functions inline image being μ-essentially inline image
    • display math
    If μ is continuous from below the assumption on inline image being μ-essentially inline image can be dropped.
  • Integration on subsets. Let μ be a monotone set function on a set system inline image with inline image and closed under intersection. For inline image define inline image, inline image. Then inline image is a monotone set function on inline image and we define inline image.

A set system is, generally speaking, a collection of sets. Definitions of monotone, modular, or submodular set functions, as well as the definition of continuity from below, are given in next paragraphs.5

According to Definition 6.1, given a confidence level α and taking into account that inline image for the fixed pair of random variables, that is, inline image, the tail-subadditivity property is satisfied by a distortion risk measure inline image if the subadditivity theorem can be applied to the set function inline image, that is, the set function so that for any inline image, inline image.

Therefore, subadditivity in the tail for a pair of risks is proven if inline image is submodular and continuous from below.

If inline image is a distortion risk measure so that its associated distortion function g is concave in inline image, then it is shown that inline image is submodular.6 Consider the set function ν defined by inline image, for any inline image. Note that inline image because inline image and P is a monotone set function. The set function ν is modular because P is modular, that is, inline image for any inline image. Given inline image suppose, without loss of generality, that inline image. Let us rename inline image, inline image, inline image and inline image. Because ν is monotone then it holds that inline image due to inline image. The modularity of ν implies that inline image, i.e., inline image and inline image have common centers, inline image. Then, because g is concave in inline image we can conclude that inline image or, equivalently, that inline image is submodular.

The property of continuity from below of inline image must also be satisfied to use the subadditivity theorem. An arbitrary set function μ is continuous from below if for any increasing collection of subsets in the set system (inline image, inline image for inline image) so that inline image then equality inline image holds. So inline image is continuous from below because inline image and g is continuous on inline image.

Given that inline image is submodular and continuous from below, applying the subadditivity theorem and using integration on subsets, it is true that, given X and Y:

  • math image

which prove that the associated risk measure inline image is tail-subadditive.

Consider a GlueVaR risk measure so that weights inline image belong to the lightly shaded area in Fig. 2. This is a sufficient condition to guarantee concavity of the distortion function on inline image. Therefore, these GlueVaR risk measures are candidates to satisfy the tail-subadditivity property.

References

  1. Top of page
  2. Abstract
  3. 1. INTRODUCTION
  4. 2. BACKGROUND AND MOTIVATION
  5. 3. DISTORTION RISK MEASURES
  6. 4. A NEW FAMILY OF RISK MEASURES: GlueVaR
  7. 5. LINEAR COMBINATION OF RISK MEASURES
  8. 6. SUBADDITIVITY IN THE TAIL
  9. 7. RISK ATTITUDES IN GlueVaR
  10. 8. OTHER NONFINANCIAL APPLICATIONS
  11. 9. CONCLUSIONS
  12. Acknowledgments
  13. Appendix A
  14. Appendix B
  15. Appendix C
  16. References
  • 1
    Alexander C, Sarabia JM. Quantile uncertainty and value-at-risk model risk. Risk Analysis, 2012; 32(8):12931308.
  • 2
    McNeil A, Frey R, Embrechts P. Quantitative Risk Management. New York: Princeton Series in Finance, Princeton University Press, 2005.
  • 3
    Artzner P, Delbaen F, Eber J-M, Heath D. Coherent measures of risk. Mathematical Finance, 1999; 9(3):203228.
  • 4
    Acerbi C, Tasche D. On the coherence of expected shortfall. Journal of Banking and Finance, 2002; 26(7):14871503.
  • 5
    Szëgo G. Measures of risk. Journal of Banking and Finance, 2002; 26(7):12531272.
  • 6
    Wang S. Insurance pricing and increased limits ratemaking by proportional hazard transforms. Insurance: Mathematics and Economics, 1995; 17(1):4354.
  • 7
    Wang S. Premium calculation by transforming the layer premium density. ASTIN Bulletin, 1996; 26(1):7192.
  • 8
    Tsanakas A, Desli E. Measurement and pricing of risk in insurance markets. Risk Analysis, 2005; 25(6):16531668.
  • 9
    Denuit M, Dhaene J, Goovaerts M, Kaas R. Actuarial Theory for Dependent Risks. Measures, Orders and Models. Chichester: John Wiley & Sons Ltd, 2005.
  • 10
    Balbás A, Garrido J, Mayoral S. Properties of distortion risk measures. Methodology and Computing in Applied Probability, 2009; 11(3):385399.
  • 11
    Choquet G. Theory of capacities. Annales de l'Institute Fourier, 1954; 5:131295.
  • 12
    Denneberg D. Non-Additive Measure and Integral. Dordrecht: Kluwer Academic Publishers, 1994.
  • 13
    Aven T. Foundational issues in risk assessment and risk management. Risk Analysis, 2012; 32(10):16471656.
  • 14
    Grabisch M, Marichal J-L, Mesiar R, Pap E. Aggregation functions: Construction methods, conjunctive, disjunctive and mixed classes. Information Sciences, 2011; 181(1):2343.
  • 15
    Grabisch M, Marichal J-L, Mesiar R, Pap E. Aggregation functions: Means. Information Sciences, 2011; 181(1):122.
  • 16
    Belles-Sampera J, Merigó J-M, Guillén M, Santolino M. The connection between distortion risk measures and ordered weighted averaging operators. Insurance: Mathematics and Economics, 2013; 52(2):411420.
  • 17
    Dhaene J, Kukush A, Linders D, Tang Q. Remarks on quantiles and distortion risk measures. European Actuarial Journal, 2012; 2(2):319328.
  • 18
    Hosking J, Wallis J. Parameter and quantile estimation for the generalized Pareto distribution. Technometrics, 1987; 29(3):339349.
  • 19
    llmer H, Schied A. Convex measures of risk and trading constraints. Finance and Stochastics, 2002; 6(4):429447.
  • 20
    Frittelli M, Gianin E. Putting order in risk measures. Journal of Banking and Finance, 2002; 26(7):14731486.
  • 21
    Denuit M, Dhaene J, Goovaerts M, Kaas R, Laeven R. Risk measurement with equivalent utility principles. Statistics & Decisions, 2006; 24(1):125.
  • 22
    Song Y, Yan J-A. Risk measures with comonotonic subadditivity or convexity and respecting stochastic orders. Insurance: Mathematics and Economics, 2009; 45(3):459465.
  • 23
    Cerreia-Vioglio S, Maccheroni F, Marinacci M, Montrucchio L. Risk measures: Rationality and diversification. Mathematical Finance, 2011; 21(4):743774.
  • 24
    Ekeland I, Galichon A, Henry M. Comonotonic measures of multivariate risks. Mathematical Finance, 2012; 22(1):109132.
  • 25
    Goovaerts M, Linders D, Van Weert K, Tank F. On the interplay between distortion, mean value and Haezendonck-Goovaerts risk measures. Insurance: Mathematics and Economics, 2012; 51(1):1018.
  • 26
    Grechuk B, Molyboha A, Zabarankin M. Mean-deviation analysis in the theory of choice. Risk Analysis, 2012; 32(8):12771292.
  • 27
    Wang S, Dhaene J. Comonotonicity, correlation order and premium principles. Insurance: Mathematics and Economics, 1998; 22(3):235242.
  • 28
    Wirch J, Hardy M. Distortion Risk Measures: Coherence and Stochastic Dominance. Lisbon: IME Conference, 2002.
  • 29
    Cox T. Confronting deep uncertainties in risk analysis. Risk Analysis, 2012; 32(10):16071629.
  • 30
    Dhaene J, Laeven R, Vanduffel S, Darkiewicz G, Goovaerts M. Can a coherent risk measure be too subadditive? Journal of Risk and Insurance, 2008; 75(2):365386.
  • 31
    Jiang L. Convexity, translation invariance and subadditivity for g-expectations and related risk measures. Annals of Applied Probability, 2008; 18(1):245258.
  • 32
    Wang S. An actuarial index of the right-tail risk. North American Actuarial Journal, 1998; 2(2):88101.
  • 33
    Embrechts P, Lambrigger D, Wuethrich M. Multivariate extremes and the aggregation of dependent risks: Examples and counter-examples. Extremes, 2009; 12(2):107127.
  • 34
    Embrechts P, Neslehova J, Wuethrich M. Additivity properties for value-at-risk under Archimedean dependence and heavy-tailedness. Insurance: Mathematics and Economics, 2009; 44(2):164169.
  • 35
    Degen M, Lambrigger D, Segers J. Risk concentration and diversification: Second-order properties. Insurance: Mathematics and Economics, 2010; 46(3):541546.
  • 36
    Nam H, Tang Q, Yang F. Characterization of upper comonotonicity via tail convex order. Insurance: Mathematics and Economics, 2011; 48(3):368373.
  • 37
    Chen D, Mao T, Pan X, Hu T. Extreme value behavior of aggregate dependent risks. Insurance: Mathematics and Economics, 2012; 50(1):99108.
  • 38
    Danielsson J, Jorgensen B, Sarma M, de Vries C. Sub-additivity Re-Examined: The Case for Value-at-Risk. CiteSeerx: Technical report 2005.
  • 39
    Hua L, Joe H. Tail comonotonicity: Properties, constructions, and asymptotic additivity of risk measures. Insurance: Mathematics and Economics, 2012; 51(2):492503.
  • 40
    Aven T. On the meaning and use of the risk appetite concept. Risk Analysis, 2013; 33(3):462468.
  • 41
    Bolancé C, Guillén M, Pelican E, Vernic R. Skewed bivariate models and nonparametric estimation for the CTE risk measure. Insurance: Mathematics and Economics, 2008; 43(3):386393.
  • 42
    Guillén M, Prieto F, Sarabia J-M. Modelling losses and locating the tail with the Pareto positive stable distribution. Insurance: Mathematics and Economics, 2011; 49(3):454461.
  • 43
    Mohtadi H, Agiwal S. Optimal security investments and extreme risk. Risk Analysis, 2012; 32(8):13091325.
  • 44
    Hausken K. Returns to information security investment: The effect of alternative information security breach functions on optimal investment and sensitivity to vulnerability. Information Systems Frontiers, 2006; 8(5):338349.
  • 45
    Gordon L, Loeb M. The economics of information security investment. ACM Transactions on Information System Security, 2002; 5(4):438457.
  • 46
    Dehlendorff C, Kulachi M, Mersen S, Andersen K. Conditional value at risk as a measure for waiting time in simulations of hospital units. Quality Technology & Quantitative Management, 2010; 7(3):321336.
  • 47
    Verderame P, Floudas C. Integration of operational planning and medium-term scheduling for large-scale industrial batch plants under demand and processing time uncertainty. Industrial & Engineering Chemistry Research, 2010; 49(10):49484965.
  • 48
    Pongsakdi A, Rangsunvigit P, Siemanond K, Bagajewicz M. Financial risk management in the planning of refinery operations. International Journal of Production Economics, 2006; 103(1):6486.
  • 49
    Eppen G, Martin R, Schrage L. OR practice—A scenario approach to capacity planning. Operations Research, 1989; 37(4):517527.
  • 50
    García-González J, Parrilla E, Mateo A. Risk-averse profit-based optimal scheduling of a hydro-chain in the day-ahead electricity market. European Journal of Operational Research, 2007; 181(3):13541369.
  • 51
    Webby R, Adamson P, Boland J, Howlett P, Metcalfe A, Piantadosi J. The Mekong-applications of value at risk (VaR) and conditional value at risk (CVaR) simulation to the benefits, costs and consequences of water resources development in a large river basin. Ecological Modelling, 2007; 201(1):8996.
  • 52
    Gebreslassie B, Yao Y, You F. Design under uncertainty of hydrocarbon biorefinery supply chains: Multiobjective stochastic programming models, decomposition algorithm, and a comparison between CVaR and downside risk. AIChE Journal, 2012; 58(7):21552179.
  • 53
    Carneiro M, Ribas G, Hamacher S. Risk management in the oil supply chain: A CVaR approach. Industrial & Engineering Chemistry Research, 2010; 49(7):32863294.
  • 54
    Bret-Rouzart N, Favennec J-P. Oil and Gas Exploration and Production: Reserves, Costs, Contracts. 3rd ed. Paris: Editions Technip, 2011.
  • 55
    Owen N, Inderwildi O, King D. The status of conventional world oil reserves: Hype or cause for concern? Energy Policy, 2010; 38(8):47434749.
  • 56
    Campbell C, Laherrère J. The end of cheap oil. Scientific American Magazine, 1998; 278:7883.
  1. 1

    An interpretation of GlueVaR risk measures as aggregation operators can be undertaken. An aggregation operator is a function that combines inputs into a single value, where inputs may be degrees of preference, membership or likelihood, or support of a hypothesis. Therefore, a linear combination of risk measures may be understood as an aggregation operator. A complete state of the art on aggregation operators can be found in Refs. [14] and [15]. Additionally, VaR and TVaR may be understood as aggregation operators for discrete distributed random variables. [16] A discussion of the relationships between quantiles and distortion risk measures is found in Ref. [17].

  2. 2

    In a case such as this, X has μ mean and a SD equal to inline image.

  3. 3

    There are some exceptions to this rule. When X follows a Pareto distribution with inline image and for any confidence level α, TVaRinline image. But when inline image GlueVaRinline image is finite. There is a compensation effect between TVaRinline image and TVaRinline image. This is taken into account in Table 2.

  4. 4

    For instance, a conservative decisionmaker would set inline image (i.e., inline image), while a much less cautious decisionmaker would set inline image (i.e., inline image).

  5. 5

    A proper definition of a function inline image-essentially inline image is not needed in the proof and, thus, not provided. Interested readers can find this definition in Ref. [12].

  6. 6

    A set function μ is modular if inline image, and it is submodular if inline image. A set function μ is monotone if inline image for any inline image in 2Ω.