In this chapter we return to the quantification of financial risk in terms of monetary risk measures. As we saw in Chapter 4, a monetary risk measure specifies the capital which is needed in order to make a given financial position acceptable. In our present dynamic setting, the capital requirement at a given time will depend on the available information and will thus become a random variable. In Section 11.1 we introduce the notion of a conditional monetary risk measure. In the convex case we prove a conditional version of the robust representation theorem which involves the conditional expected losses under various models Q and a conditional penalization of these models.
As time varies, both the capital requirement and the penalization of a given probabilistic model become stochastic processes. The structure of these processes depends on how the risk assessments at different times are connected with each other. In Section 11.2 we focus on a strong notion of time consistency which amounts to a recursive property of the successive capital requirements. For a typical model Q, time consistency implies that the corresponding penalization process is a super-martingale under Q, and that its Doob decomposition takes a special form. Time consistency is also characterized by a combined supermartingale property of the capital requirements and the penalization process.
As in the preceding chapters we fix a filtration (Ft)t=0,...,T on our probability space (Ξ©,F, P) such that F0 = {β , Ξ©} and FT = F. As the set X of all financial positions we take the space Lβ := Lβ(Ξ©,F, P). The subspace βconsists of those positions whose outcome only depends on the history up to time t. All inequalities and equalities applied to random variables are meant to hold P-a.s. if not stated otherwise.
Definition 11.1. A map will be called a monetary conditional risk measure if it satisfies the following properties for all X, Y β Lβ:
βConditional cash invarianceβ: Οt(X + Xt) = Οt(X) β Xt for any
βMonotonicity: X β€ Y β Οt(X) β₯ Οt(Y).
βNormalization: Οt(0) = 0.
A monetary conditional risk measure will be called convex if it satisfies
βConditional βconvexity: with 0 β€ Ξ» β€ 1.
A convex conditional risk measure will be called coherent if it satisfies in addition
ββConditional positive homogeneity: Οt(Ξ»X) = Ξ»Οt(X) for such that Ξ» β₯ 0.
β
For t = 0 we recover our previous definition of (unconditional and normalized) monetary, convex, and coherent risk measures given in Section 4.1, since the space reduces to the real line.
To any conditional monetary risk measure Οt we associate its acceptance set
One easily checks that At has the following properties:
βX β At , Y β₯ X β Y β At ,
βess infand 0 β At.
Note that Οt is uniquely determined by its acceptance set since
A conditional monetary risk measure can thus be viewed as the conditional capital requirement needed at time t to make a financial position X acceptable at that time. Moreover, Οt is a convex conditional risk measure if and only if At satisfies the additional property of
βConditional convexity: X, Y β At and with 0 β€ Ξ» β€ 1 implies Ξ»X + (1 β Ξ») βY β At.
Conversely, one can use acceptance sets to define conditional monetary risk measures: If a given acceptance set At β Lβ satisfies the above conditions then the functional defined via (11.2) is a conditional monetary risk βmeasure.
Exercise 11.1.1. Let be a conditional monetary risk βmeasure. For X β Lβ we define
Show that for X, Y β Lβ
β
Exercise 11.1.2. Show that every conditional βmonetary risk measure has the following property:
Then show that (11.3) is equivalent to the the following local property: if X, Y β Lβ and A β Ft, then
Hint: It can be helpful to use Exercise 11.1.1.
β
By M1(P) we denote the set of all probability measures on (Ξ©,F) that are absolutely continuous with respect to P. As we have shown in Theorem 4.33, an unconditional convex risk measure that has the Fatou property admits a robust representation of the form
with some penalty function Ξ± : M1(P) β β βͺ {+β}. In fact we can take the minimal penalty function
In this section we are going to prove a robust representation for conditional convex risk measures which is analogous to (11.4). Here the penalty function will depend on the history up to time t, and so it will be given by a map Ξ±t from (a subset of) M1(P) to the set of Ft-measurable random variables with values in β βͺ{+β}. In analogy to (11.5), we are going to take the minimal penalty function, defined as the worst conditional loss over all positions which are acceptable at time t:
The conditional expectations on the right are a priori defined only Q-a.s., but they also need to be well-defined under the reference measure P. This is the case when Q belongs to the set
Theorem 11.2. For a convex conditional risk measure Οt the following properties are equivalent:
(a) Οt has the representation
where for Q β Qt the penalty function is given by (11.6). In fact the representation (11.7) also holds if we replace Qt in (11.7) by the smaller set
(b) Οt has the following Fatou property:
for any bounded sequence (Xn) β Lβ which converges P-a.s. to X β Lβ.
(c) Οt is continuous from above, i.e.,
for any sequence (Xn) β Lβ and X β Lβ.
Proof. (a)β(b): Lebesgueβs dominated convergence theorem for conditional expectations yields
and hence P-a.s., for each Q β Qt. Thus, P-a.s.,
and so the representation (11.7) implies Οt(X) β€ lim infn Οt(Xn).
(b)β(c): The Fatou property yields lim infn Οt(Xn)β₯ Οt(X). On the other hand we have lim supn Οt(Xn)β€ Οt(X) by monotonicity, and this implies Οt(Xn) β Οt(X).
(c)β(a): Since X + Οt(X) β At for any X β Lβ by conditional cash invariance, the definition (11.6) of implies
hence
for any Q β Qt. This yields the inequality
In order to prove the equality in (11.7), both for Qt and for the smaller set Pt, it is therefore enough to show that
To this end, note first that
Next, we consider the map Ο : Lβ β β defined by
It is easy to check that Ο is a convex risk measure. Property (c) implies that Ο is continuous from above; equivalently, it has the Fatou property. Thus, Theorem 4.33 states that Ο has the robust representation
where the penalty function Ξ±(Q) is given by
Next we prove that Q = P on Ft , and hence Q β Pt, if Ξ±(Q) < β. Indeed, take A β Ft and Ξ» > 0. Then
hence
for any Ξ» > 0. Thus Ξ±(Q) < β implies P[ A ]β€ Q[ A ] for any A β Ft, and hence P = Q on Ft.
Furthermore,
holds for every Q β Pt. Indeed,
holds by equation (11.17) in Lemma 11.3 below, and so (11.12) and (11.11) imply (11.13) since Ο(X) β€ 0 for all X β At.
We can now conclude the proof of inequality (11.9) as follows, using first (11.14) and then (11.13):
The following lemma was used in the preceding proof.
Lemma 11.3. For Q β Qt and 0 β€ s β€ t,
and in particular
Proof. First we show that the family
is directed upward for any Q β (see Appendix A.5). Indeed, for X, Y β we can Qt At take where A := {EQ[ βX |Ft ]β₯ EQ[ βY |Ft ]} β Ft. Conditional convexity of Οt implies Z β At, and clearly we have
Thus the family is directed upward. Hence, Theorem A.37 implies that its essential supremum can be computed as the supremum of an increasing sequence within this family, i.e., there exists a sequence in At such that
By monotone convergence we get
The converse inequality follows directly from the definition of This shows (11.16), and for s = 0 we get (11.17).
Remark 11.4. The inequalities (11.8) and (11.9) in the proof of the implication β(c)β(a)β in Theorem 11.2 show that the representation (11.7) also holds if we replace Qt in (11.7) not only by Pt but by the even smaller set
In view of (11.8) and (11.9), we can conclude that the representation (11.7) holds for any set Q such that
β
Example 11.5. Suppose that preferences are characterized by an exponential utility function u(x) = 1 β exp(βΞ²x) with Ξ² > 0. At time t the conditional expected utility of a financial position X β Lβ is then given by the Ft-measurable random variable
The set
satisfies the conditions required from an acceptance set. In analogy to Example 4.34, the induced convex conditional risk measure Οt is given by
i.e.,
Clearly, Οt is a monetary conditional risk measure. It will be shown in Exercise 11.1.3 that it is also convex and admits a robust representation with minimal penalty function
where
denotes the conditional relative entropy of Q β Pt with respect to P, given Ft. For this reason, Οt is called the conditional entropic risk measure.
β
Exercise 11.1.3. Argue as in Lemma 3.29 to show that for any Q β Pt the conditional relative entropy (11.20) satisfies the following variational identity:
where the second supremum is attained by Then proceed as in Example 4.34 to conclude that the conditional entropic risk measure is convex and that (11.19) describes indeed its minimal penalty function for Q β Pt.
β
In the coherent case we obtain the following representation result:
Corollary 11.6. Let Οt be a coherent conditional risk measure that has the Fatou property. Then Οt has the representation
where the setis given by
and coincides with the set in (11.18).
Proof. In the coherent case the penalty function can only take the values 0 or β for Q β Qt. Indeed, for and any Ξ» > 0 we have Ξ»A X β At due to conditional positive homogeneity of Οt, hence
and the lower bound converges to infinity on A as Ξ» β β. Thus on A.Next, letbe as in (11.18). Then and this implies P[ A ] = 0. Thus coincides with the set of all Q β Qt for which P-a.s., and so the representation (11.7), where we are free to take instead of Qt by Remark 11.4, reduces to (11.21).
Example 11.7. For any Ξ» β (0, 1), the acceptance set
defines a monetary conditional risk measure, which we call conditional Value at Risk at level Ξ»:
The conditional risk measure V@RΞ»( Β· |Ft) is conditionally positively homogenous, but it is not conditionally convex. While the terminology βconditional Value at Riskβ seems quite natural, one should be aware of a possible confusion because the unconditional risk measure AV@R, which was introduced in Definition 4.48, is sometimes also called Conditional Value at Risk and denoted by CV@R.
β
Definition 11.8. For Ξ» β (0, 1] let denote the set of all measures Q β Pt whose density dQ/dP is P-a.s. bounded by 1/Ξ». The resulting coherent conditional risk measure
is called conditional Average Value at Risk at level Ξ».
β
We have the following representation result for the coherent conditional risk measure AV@RΞ»( Β· |Ft).
Proposition 11.9. For X β Lβ, the essential supremum in (11.23) is attained for any whose density satisfies
and the set of such maximizers is nonempty. In particular,
Proof. Fix X β Lβ and note that
where we have used (11.22) in the last step. On the other hand, (11.22) also implies that
for each n, and hence that
The two inequalities (11.26) and (11.27) imply that the set of measures satisfying (11.24) is nonempty. Now let ZX = dQX /dP be the density of such a measure, and let Z be the density of any other We furthermore define := X +V@RΞ»(X|Ft). Then V@RΞ»(|Ft) = 0 and, arguing as in the proof of the general Neyman-Pearson lemma in the form of Theorem A.35,
with P-a.s. equality if and only if Z is also of the form (11.24). The conditional cash invariance of AV@RΞ»(Β·|Ft) thus yields the first part of the assertion. The identity (11.25) now follows from the following chain of identities:
In this section we consider a sequence of convex conditional risk measures
In such a dynamic setting the key question is how the risk assessments of a financial position at different times are connected to each other.
Definition 11.10. A sequence of conditional risk measures ( Οt)t=0,1,...,T is called (strongly) time-consistent if for any X, Y β Lβ and for all t β₯ 0 the following condition holds:
β
Lemma 11.11. Time consistency is equivalent to each of the following two properties:
(a) Οt+1(X) = Οt+1(Y) Οt(X) = Οt(Y) for t = 0,1, . . . T β 1.
(b) Recursiveness: Οt = Οt(βΟt+1) for t = 0,1, . . . , T β 1.
Proof. Time consistency clearly implies (a).
(a) β (b): Note that Οt+1(βΟt+1(X)) = Οt+1(X), due to conditional cash-invariance and normalization. Applying (a) with Y := βΟt+1(X) we thus obtain Οt(X) = Οt(βΟt+1(X)).
(b) implies time consistency: If Οt+1(X) β€ Οt+1(Y) then Οt(βΟt+1(X)) β€ Οt(βΟt+1(Y)) by monotonicity, and so Οt(X) β€ Οt(Y) follows from (b).
Example 11.12. For fixed Ξ² > 0, the sequence of conditional entropic risk measures
is time-consistent. Let us check recursiveness:
Note, however, that time consistency will be lost if the constant risk aversion parameter Ξ² is replaced by an adapted process (Ξ²t). Moreover, under a suitable condition of dynamic law-invariance, the conditional entropic risk measure is in fact the only time-consistent dynamic convex risk measure; see [196], where this result is reduced to an application of Proposition 2.46. Here we include the ordinary conditional expectation with respect to P as the limiting case Ξ² = 0.
β
Example 11.13. The sequence of coherent conditional risk measures
given by conditional Average Value at Risk at some level Ξ» β (0, 1) is not time-consistent, and the same is true for conditional Value at Risk and for conditional mean-standard deviation risk measures defined in terms of Sharpe ratios. To see this, note first that all these risk measures are well defined on L2(P), and that they are of the form
if X has a conditional Gaussian distribution with respect to Ft (compare Examples 4.11, 4.12 and Exercise 4.4.3 (b)). Now consider a position of the form X = X1+ X2 for two independent Gaussian random variables X i with distribution N (0, Ο) and i2assume that F1 is the Ο-field generated by X1. Then
hence
On the other hand,
and so we have Ο0(βΟ1(X)) > Ο0(X) unless X1 or X2 are constant and therefore Ο1 = 0 or Ο2 = 0. Note that the preceding argument works also if it is possible to represent Οt in the form
for some set M of Borel probability measures on (0, 1]. Such a representation would extend the representation of law-invariant coherent risk measures given in Corollary 4.63 to the conditional setting.
β
Exercise 11.2.1. Show that recursiveness as defined in part (b) of Lemma 11.11 is equivalent to the following condition: for 0 β€ s < t β€ T and X β Lβ,
β
The next exercise relies on and generalizes the preceding one.
Exercise 11.2.2. For a stopping time Ο : Ξ© β {0, . . . , T} we define
Show that recursiveness is equivalent to the following property: if Ο is another stopping time such that Ο β€ Ο, then
Hint: Use Exercise 11.2.1 and (11.3).
β
Exercise 11.2.3. Let Οt, t = 0, . . . , T, be any sequence of convex conditional risk measures. Show that the recursive definition
yields a time-consistent sequence of convex conditional risk measures.
β
We are now going to characterize time consistency of the sequence ( Οt)t=0,1,...,T both in terms of the corresponding acceptance sets (At)and in terms of the penalty t=0,1,...,T processes Our main goal is to prove a supermartingale criterion which characterizes time consistency in terms of the joint behavior of the stochastic processes ( Οt(X))t=0,1,...,T and (Ξ±t(Q))t=0,1,...,T.
To this end we introduce some notation. Suppose that we look just one step ahead and assess the risk only for those positions whose outcome will be known by the end of the next period. This means that we restrict the conditional convex risk measure Οt to the space The βcorresponding one-step acceptance set is given by
and
is the resulting one-step penalty function.
The following lemma holds for any sequence of monetary conditional risk measures. The equivalences between set inclusions for the acceptance sets and inequalities for the risk measures can be used as starting points for various departures from the strong notion of time consistency which we are considering here; see, e.g., [234], [271], and [274]. For two sets A , B β Lβ we will use the following notation:
Lemma 11.14. Let ( Οt)t=0,1,...,T be a sequence of monetary conditional risk measures. Then the following equivalences hold for all t = 0, . . . , T β 1 and all X β Lβ:
(a) X β At,t+1 + At+1 ββ βΟt+1(X) β At,t+1.
(b) At β At,t+1 + At+1 ββ Οt(βΟt+1) β€ Οt.
(c) At β At,t+1 + At+1 ββ Οt(βΟt+1) β₯ Οt.
Proof. (a) To prove βββ take X = Xt,t+1 + Xt+1 with Xt,t+1 β At,t+1 and Xt+1 β At+1. Then
by cash invariance, and monotonicity implies
hence βΟt+1(X) β At+1. The converse implication follows immediately from the decomposition X = X + Οt+1(X) β Οt+1(X), since X + Οt+1(X) β At+1 for all X β Lβ and βΟt+1(X) β At,t+1 by assumption.
(b) In order to show βββ, take X β Lβ. Since X + Οt(X) β At β At,t+1 + At+1, we obtain
by (a) and by cash invariance. This implies
To prove βββ take X β At. Then βΟt+1(X) β At,t+1 by the right-hand side of (b), and hence X β At,t+1 + At+1 by (a).
(c) Take X β Lβ and assume At β At,t+1 + At+1. Then
by cash invariance, and so we have shown βββ. For the converse implication take X β At,t+1 + At+1. Since βΟt+1(X) β At,t+1 by (a), we obtain
hence X β At.
The preceding lemma implies immediately the following result.
Proposition 11.15. Let ( Οt)t=0,1,...,T be a sequence of convex conditional risk measures such that each Οt has the Fatou property. Then the following conditions are equivalent:
(a) ( Οt)t=0,1,...,T is time-consistent.
(b) At = At,t+1 + At+1 for t = 0, . . . , T β 1.
In the sequel, we will investigate the time consistency of a sequence ( Οt)t=0,1,...,T of convex conditional risk measures in terms of the dynamics of their penalty functions. In doing this, we need to assume that every element of the sequence can be represented in terms of the same set Q of probability measures:
These representations are only well-defined if Q β Qt for all t. Since T < β, every Q β Q must be equivalent to P on FT = F, and so we may as well assume
Definition 11.16. A sequence ( Οt)t=0,1,...,T of convex conditional risk measures is called sensitive if the representation (11.28) holds in terms of the set Q in (11.29).
β
Sensitivity is also called relevance. It generalizes the concept of sensitivity introduced in Section 4.3 to the case of conditional risk measures, and it can be characterized along the lines of Theorem 4.43. In fact, it can be shown that a time-consistent sequence ( Οt)t=0,1,...,T of conditional convex risk measures is sensitive as soon as Ο0 is sensitive in the sense of Definition 4.42; see [128].
The following theorem, and in particular the equivalence of its conditions (a) and (c), is the main result of this section.
Theorem 11.17. Let ( Οt)t=0,1,...,T be a sensitive sequence of convex conditional risk measures. Then the following conditions are equivalent:
(a) ( Οt)t=0,1,...,T is time-consistent.
(b) For any Q β Q,
(c) For any Q β Q and any X β Lβ, the process
satisfies
and is a Q-supermartingale for any Q with
We prepare the proof with the following two lemmas. The first lemma deals with the pasting of two probability measures Q1 and Q2 in the deterministic time t. It is a special case of the results in Section 6.4, but we state and prove it here to keep the discussion self-contained.
Lemma 11.18. For Q1, Q2 β Q and t β {0, . . . , T} there exists Q β Q such that Q = Q1 on Ft and
In particular we have
Proof. When setting
then EQ1 [ Z | Ft ] = 1, and so dQ = Z dQ1 defines a measure Q β Q that coincides with Q1 on Ft. Moreover, Proposition A.16 yields (11.30). The identity (11.31) now follows from the definition of
is directed upward.
Proof. By applying Lemma 11.18 with Q1 := P we see that it is enough to show that the family
is directed upward. To this end, fix Q1, Q2 β Q β© Pt and let Yi := EQi[ βX | Ft ] β and A := {Y2 β₯ Y1}. We then define
This random variable is P-a.s. strictly positive and satisfies
It follows that Z is the density of a probability measure Q β Q β© Pt. Under Q, we have
Similarly,
Therefore,
which concludes the proof.
Proof of Theorem 11.17: (a) β (b): By Proposition 11.15, any X β At can be written as the sum of some Xt,t+1 β At,t+1 and some Xt+1 β At+1. Hence, we obtain
for any Q β Q, using Lemma 11.3 in the last step.
(b) β (c): Fix X β Lβ. By Lemma 11.19 and Theorem A.37 we can find a sequence Qn β Q such that Οt+1(X) can be identified as the limit of an increasing sequence:
Now take Q β Q and write We have to show that Ut β₯ EQ[ Ut+1 | Ft ]. Using property (b) we see that
Now we use the sequence of measures Qn appearing in (11.32). By Lemma 11.18, we are free to assume that Qn = Q on Ft+1, and this implies
Using the approximation (11.32) and monotone convergence for conditional expectations, and applying property (b) together with equation (11.34) to each Qn, we obtain
Together with (11.33) this yields
If in addition Ξ±0(Q) < β then (b) implies Moreover, Οt(X) is bounded by Xβ, and so Ut is integrable and hence a supermartingale.
(c) β (a): Take X, Y β Lβ such that Οt+1(X) β€ Οt+1(Y); we have to show Οt(X) β€ Οt(Y). For each Q β Q,
The first inequality follows from the supermartingale property in (c), and in the third step we have used the inequality
which is valid for any Q β Q; see (11.8). Thus
for all Q β Q, and this implies the desired inequality Οt(Y) β₯ Οt(X) due to our assumption that Οt admits the representation (11.28).
Remark 11.20. It follows from property (b) of Theorem 11.17 that
This in turn implies for all Thus, the process is a Q-supermartingale for any such Q. This can be seen as a built-in learning effect: If the dynamics are in fact driven by the probability measure Q then the penalization of these measures decreases βon averageβ in the sense that it is a supermartingale under Q. Note that property (b) provides more information than just the supermartingale property: It actually yields an explicit description of the predictable increasing process in the Doob decomposition of the supermartingale in terms of the βone-stepβ penalty functions More precisely,
where M Q is a martingale under Q.
β
Exercise 11.2.4. Show that the cost of superhedging defines a time-consistent sequence of conditional convex risk measures. More precisely, in the situation and with the notation of Chapter 9, let
for Y Sβ Lβ. Show that is a time-consistent sequence of conditional convex risk measures if the conditions of Corollary 9.32 are satisfied.
β
Remark 11.21. The weaker version (11.35) of property (b) is in fact equivalent to a weaker notion of time consistency, called weak time consistency, that is defined by the implication
and which is equivalent to the following weaker version
of property (b) in Proposition 11.15; see Exercise 11.2.5
β
Exercise 11.2.5. Let ( Οt)t=0,1,...,T be a sensitive sequence of convex conditional risk measures. Show that the following conditions are equivalent.
(a) ( Οt)t=0,1,...,T is weakly time consistent in the sense that Οt+1(X) β€ 0 implies Οt(X) β€ 0.
(b) The acceptance sets satisfy the inclusion At+1 β At for t = 0, . . . , T β 1.
(c) The penalty functions satisfy
β
We will now consider the coherent case in Theorem 11.17. An initial discussion of this case was already given at the end of Section 6.5 in connection with our analysis of stability under pasting. Recall from Definition 6.39 that the pasting of two equivalent probability measures Q1 and Q2 in a stopping time Ο β€ T is the probability measure
According to Definition 6.42, a set QΟ of equivalent probability measures is called stable if, for any Q1, Q2 β Q and Ο β T , also their pasting in Ο is belongs to Q. We can now state the following characterization of time consistency for coherent dynamic risk measures. It extends, and provides a converse of, Theorem 6.52.
Theorem 11.22. Let ( Οt)t=0,1,...,T be a sensitive sequence of convex conditional risk measures, and assume that Ο0 is coherent. Then the following conditions are equivalent.
(a) ( Οt)t=0,1,...,T is time-consistent.
(b) There exists a stable set QΟ β Q such that
In particular, each Οt is coherent if these equivalent conditions hold.
Proof. It was shown in Theorem 6.52 that (b) implies recursiveness and in turn (a).
To show the implication β(a)β(b)β, let
Then is a nonnegative Q-supermartingale for β QΟ any Q by Theorem 11.17. Since Ο0 is coherent, we have and hence Q-a.s. (and hence P-a.s.) for each t, due to the supermartingale property (compare Exercise 6.1.2). Therefore the representation 11.36 follows with Remark 11.4. In particular, each Οt is coherent.
To show the stability of QΟ, we recall from Proposition 6.44 that the stability of QΟ is equivalent to the following fact: For Q1, Q2 β QΟ, t β {0, . . . , T}, and B β Ft, also the probability measure
belongs to QΟ. Clearly, this measure Q is equivalent to P and hence belongs to Q. To prove that and in turn Q β QΟ, we show first that Indeed, for X β At,
because for i = 1,2
This shows our claim Next, part (b) of Theorem 11.17 yields that
where we have used the fact that Q = Q1 on Ft in the fourth identity. An iteration of this argument shows
Remark 11.23. As in the other chapters of Part II, we have limited the discussion to a finite horizon T < β. If we pass to an infinite horizon, the characterization of time consistency in terms of supermartingale properties in Theorem 11.17 yields convergence results for risk measures which can be seen as nonlinear extensions of martingale convergence; cf. [128].
β
18.218.238.134