Exponentially concave functions and
multiplicative cyclical monotonicity
W. Schachermayer
University of Vienna Faculty of Mathematics11th December, RICAM, Linz
Based on the paper S. Pal, T.L. Wong:
Exponential concavity
Fix a convex subsetU⊆Rn.TypicallyUwill equal the unit simplex ∆n={(p1, . . . ,pn)∈[0,1]n:
n X i=1
pi= 1} or a convex subset of ∆n,e.g.int(∆n).
Definition:
A functionϕ:U7→Ris calledexponentially concaveif Φ = exp(ϕ) is a concave function.
An exponentially concave function is concave but not vice versa.
For example, considerU= (0,1).An affine functionϕ(x) =ax+bonUis concave but not exponentially concave. On the other hand, the functionϕ(x) = log(x) is the arch-example of an exponentially concave function onU.
Exponential concavity
Fix a convex subsetU⊆Rn.TypicallyUwill equal the unit simplex ∆n={(p1, . . . ,pn)∈[0,1]n:
n X i=1
pi= 1} or a convex subset of ∆n,e.g.int(∆n).
Definition:
A functionϕ:U7→Ris calledexponentially concaveif Φ = exp(ϕ) is a concave function.
An exponentially concave function is concave but not vice versa.
For example, considerU= (0,1).An affine functionϕ(x) =ax+bonUis concave but not exponentially concave. On the other hand, the functionϕ(x) = log(x) is the arch-example of an exponentially concave function onU.
Exponential concavity
Fix a convex subsetU⊆Rn.TypicallyUwill equal the unit simplex ∆n={(p1, . . . ,pn)∈[0,1]n:
n X i=1
pi= 1} or a convex subset of ∆n,e.g.int(∆n).
Definition:
A functionϕ:U7→Ris calledexponentially concaveif Φ = exp(ϕ) is a concave function.
An exponentially concave function is concave but not vice versa.
For example, considerU= (0,1).An affine functionϕ(x) =ax+bonUis concave but not exponentially concave. On the other hand, the functionϕ(x) = log(x) is the arch-example of an exponentially concave function onU.
Lemma:
LetU be a convex subset ofRn. A concave functionϕ:U7→Ris
exponentially concave iff for everyµ0, µ1∈U the function
f(t) =ϕ(tµ0+ (1−t)µ1) satisfies
f00(t)≤ −(f0(t))2,
for almost allt in [0,1].
Lemma:
LetU be a convex subset ofRn. A concave functionϕ:U7→Ris
exponentially concave iff for everyµ0, µ1∈U the function
f(t) =ϕ(tµ0+ (1−t)µ1) satisfies
f00(t)≤ −(f0(t))2,
for almost allt in [0,1].
Definition:
LetU⊆RnandT :U7→Rn a (possibly multi-valued) map (interpreted
as a transport map).
We callT multiplicatively cyclically monotoneif, for all
µ1, . . . , µm, µm+1 =µ1∈U and all valuesT(µ1), . . . ,T(µm) we have
hT(µj), µj+1−µji>−1 and m Y j=1 (1 +hT(µj), µj+1−µji)≥1, (1) or, equivalently n X j=1 log(1 +hT(µj), µj+1−µji)≥0. (2)
Recall thatT iscyclically monotoneif n
X
j=1
Theorem [PW14]:
LetU⊆Rnbe a convex set andϕ:U 7→
Ra concave function. Thenϕ
is exponentially concave iff its super-differentialT :=∂ϕonU is multiplicatively cyclically monotone. In this case,ϕis unique up to an additive constant.
Sketch of proof
Assume thatϕis exponentially concave so that Φ(µ) = exp(ϕ(µ)) is a concave function onU. Denote byS(µ) the super-differential of Φ for which we have Φ(µj+1)≤Φ(µj) +hS(µj), µj+1−µji, or Φ(µj+1) Φ(µj) ≤1 + DS(µj) Φ(µj), µ j+1 −µjE.
Theorem [PW14]:
LetU⊆Rnbe a convex set andϕ:U 7→
Ra concave function. Thenϕ
is exponentially concave iff its super-differentialT :=∂ϕonU is multiplicatively cyclically monotone. In this case,ϕis unique up to an additive constant.
Sketch of proof
Assume thatϕis exponentially concave so that Φ(µ) = exp(ϕ(µ)) is a concave function onU. Denote byS(µ) the super-differential of Φ for which we have Φ(µj+1)≤Φ(µj) +hS(µj), µj+1−µji, or Φ(µj+1) Φ(µj) ≤1 + DS(µj) Φ(µj), µ j+1 −µjE.
Sketch of proof contd.
Assuming that Φ is differentiable, the super-differentialS(µ) equals
∇Φ(µ) so thatT :=∇ϕ=∇(log(Φ) = ∇ΦΦ = S Φ.Therefore Φ(µj+1) Φ(µj) ≤1 +hT(µ j), µj+1−µji. Ifµ1, µ2, . . . , µm, µm+1=µ1is a roundtrip we have 1 = Φ(µ m+1) Φ(µ1) ≤ m Y i=1 (1 +hT(µj), µj+1−µji).
Applications in Finance
Stochastic portfolio Theory
R. Fernholz (1999), Karatzas & Fernholz (2009),. . .
We considernstocks with relative market capitalization at timet
µ(t) = (µ1(t), . . . , µn(t))∈int(∆n).
Aportfolio is a map
π:int(∆n)7→∆n.
Interpretation:
The agent invests her wealthV(t) according toπ(µ(t)) during ]t,t+ 1].
We associate toπtheweights w(µ) = (π1(µ)
µ1
, . . . ,πn(µ) µn
Stochastic portfolio Theory
R. Fernholz (1999), Karatzas & Fernholz (2009),. . .
We considernstocks with relative market capitalization at timet
µ(t) = (µ1(t), . . . , µn(t))∈int(∆n).
Aportfolio is a map
π:int(∆n)7→∆n.
Interpretation:
The agent invests her wealthV(t) according toπ(µ(t)) during ]t,t+ 1].
We associate toπtheweights
w(µ) = (π1(µ)
µ1
, . . . ,πn(µ) µn
Example:
a) themarket portfolio:
π(µ) =µ w(µ) = (1, . . . ,1)
b) Theequal weight portfolio: π(µ) = (1 n, . . . , 1 n) w(µ) = 1 n( 1 µ1 , . . . , 1 µn ) c) Let 0<p<1 and π(µ) = µ p 1 Pn i=1µ p i , . . . , µ p n Pn i=1µ p i ,w(µ) = µ p−1 1 1 p Pn i=1µ p i , . . . , µ p−1 n 1 p Pn i=1µ p i
Example:
a) themarket portfolio:
π(µ) =µ w(µ) = (1, . . . ,1)
b) Theequal weight portfolio:
π(µ) = (1 n, . . . , 1 n) w(µ) = 1 n( 1 µ1 , . . . , 1 µn ) c) Let 0<p<1 and π(µ) = µ p 1 Pn i=1µ p i , . . . , µ p n Pn i=1µ p i ,w(µ) = µ p−1 1 1 p Pn i=1µ p i , . . . , µ p−1 n 1 p Pn i=1µ p i
Example:
a) themarket portfolio:
π(µ) =µ w(µ) = (1, . . . ,1)
b) Theequal weight portfolio:
π(µ) = (1 n, . . . , 1 n) w(µ) = 1 n( 1 µ1 , . . . , 1 µn ) c) Let 0<p<1 and π(µ) = µ p 1 Pn i=1µ p i , . . . , µ p n Pn i=1µ p i ,w(µ) = µ p−1 1 1 p Pn i=1µ p i , . . . , µ p−1 n 1 p Pn i=1µ p i
Question:
Can you beat themarket portfolio?
Given the portfolio mapπ:int(∆n)7→∆n and a sequence (µ(t))m t=1, we
obtain for therelative wealth(in terms of the market portfolio) V(t+ 1) V(t) = n X i=1 πi(µ(t)) µi(t+ 1) µi(t) = n X i=1 wi(µ(t)) µi(t+ 1). =hw(µ(t)), µ(t+ 1)i = 1 +hw(µ(t)), µ(t+ 1)−µ(t)i
Question:
Can you beat themarket portfolio?
Given the portfolio mapπ:int(∆n)7→∆n and a sequence (µ(t))m t=1, we
obtain for therelative wealth(in terms of the market portfolio)
V(t+ 1) V(t) = n X i=1 πi(µ(t)) µi(t+ 1) µi(t) = n X i=1 wi(µ(t)) µi(t+ 1). =hw(µ(t)), µ(t+ 1)i = 1 +hw(µ(t)), µ(t+ 1)−µ(t)i
Question:
Can you beat themarket portfolio?
Given the portfolio mapπ:int(∆n)7→∆n and a sequence (µ(t))m t=1, we
obtain for therelative wealth(in terms of the market portfolio)
V(t+ 1) V(t) = n X i=1 πi(µ(t)) µi(t+ 1) µi(t) = n X i=1 wi(µ(t)) µi(t+ 1). =hw(µ(t)), µ(t+ 1)i = 1 +hw(µ(t)), µ(t+ 1)−µ(t)i
Suppose that the market makes a“round trip” µ1, µ2, . . . , µm, µm+1=µ1. Then V(m+ 1) V(1) = m Y t=1 V(t+ 1) V(t) = m Y t=1 [1 +hw(µ(t)), µ(t+ 1)−µ(t)i] (4)
is precisely the term appearing in the definition ofmultiplicative cyclical monotonocity. Taking logarithms yields
logh m Y t=1 V(t+ 1) V(t) i = m X t=1 logh1 +hw(µ(t)), µ(t+ 1)−µ(t)ii (5)
Note thatw ismultiplicatively cyclically monotoneiff (4) is always≥1 or, equivalently, (5) is always≥0.
Suppose that the market makes a“round trip” µ1, µ2, . . . , µm, µm+1=µ1. Then V(m+ 1) V(1) = m Y t=1 V(t+ 1) V(t) = m Y t=1 [1 +hw(µ(t)), µ(t+ 1)−µ(t)i] (4)
is precisely the term appearing in the definition ofmultiplicative cyclical monotonocity. Taking logarithms yields
logh m Y t=1 V(t+ 1) V(t) i = m X t=1 logh1 +hw(µ(t)), µ(t+ 1)−µ(t)ii (5)
Note thatw ismultiplicatively cyclically monotoneiff (4) is always≥1 or, equivalently, (5) is always≥0.
Suppose that the market makes a“round trip” µ1, µ2, . . . , µm, µm+1=µ1. Then V(m+ 1) V(1) = m Y t=1 V(t+ 1) V(t) = m Y t=1 [1 +hw(µ(t)), µ(t+ 1)−µ(t)i] (4)
is precisely the term appearing in the definition ofmultiplicative cyclical monotonocity. Taking logarithms yields
logh m Y t=1 V(t+ 1) V(t) i = m X t=1 logh1 +hw(µ(t)), µ(t+ 1)−µ(t)ii (5)
Note thatw ismultiplicatively cyclically monotoneiff (4) is always≥1 or, equivalently, (5) is always≥0.
Suppose that the market makes a“round trip” µ1, µ2, . . . , µm, µm+1=µ1. Then V(m+ 1) V(1) = m Y t=1 V(t+ 1) V(t) = m Y t=1 [1 +hw(µ(t)), µ(t+ 1)−µ(t)i] (4)
is precisely the term appearing in the definition ofmultiplicative cyclical monotonocity. Taking logarithms yields
logh m Y t=1 V(t+ 1) V(t) i = m X t=1 logh1 +hw(µ(t)), µ(t+ 1)−µ(t)ii (5)
Note thatw ismultiplicatively cyclically monotoneiff (4) is always≥1 or, equivalently, (5) is always≥0.
Theorem:
((i)⇒(ii): Fernholz (1999), (ii)⇒(i): Pal, Wong (2014))
Fix a portfolio mapπ:int(∆n)7→∆nand letw(µ) = (π1(µ)µ1 , . . . ,πn(µ)
µn ) be the weight function. TFAE
(i) There is an exponentially concave function ϕ:int(∆n)7→R such thatw(µ) is in the super-gradient ofϕ.
Theorem:
((i)⇒(ii): Fernholz (1999), (ii)⇒(i): Pal, Wong (2014))
Fix a portfolio mapπ:int(∆n)7→∆nand letw(µ) = (π1(µ)µ1 , . . . ,πn(µ)
µn ) be the weight function. TFAE
(i) There is an exponentially concave function
ϕ:int(∆n)7→R
such thatw(µ) is in the super-gradient ofϕ.
Theorem:
((i)⇒(ii): Fernholz (1999), (ii)⇒(i): Pal, Wong (2014))
Fix a portfolio mapπ:int(∆n)7→∆nand letw(µ) = (π1(µ)µ1 , . . . ,πn(µ)
µn ) be the weight function. TFAE
(i) There is an exponentially concave function
ϕ:int(∆n)7→R
such thatw(µ) is in the super-gradient ofϕ.
Mass Transport
LetPbe a Probability measure onint(∆n) (e.g. normalized Lebesgue)
and letQ be a probability measure onRn+.We interpretQ as a
probability measure on theweights w ∈Rn+which arenot yet normalized.
Note that forπ:int(∆n)7→∆nand
w(µ) =π1(µ) µ1 , . . . ,πn(µ) µn we have hµ,w(µ)i= 1 (6)
Hence: If we associate (via the desired mass transport)µ∈(∆n, P) with w(µ)∈Rn
Mass Transport
LetPbe a Probability measure onint(∆n) (e.g. normalized Lebesgue)
and letQ be a probability measure onRn+.We interpretQ as a
probability measure on theweights w ∈Rn+which arenot yet normalized.
Note that forπ:int(∆n)7→∆nand
w(µ) =π1(µ) µ1 , . . . ,πn(µ) µn we have hµ,w(µ)i= 1 (6)
Hence: If we associate (via the desired mass transport)µ∈(∆n,
P) with w(µ)∈Rn
Consider the cost function
c(µ,w) = log(hµ,wi) µ∈int(∆n),w ∈Rn
+.
For the probabilitiesPonint(∆n) andQ onRn+we consider the optimal
transport problem
E[c(µ,w(µ))]7→min!
Where we optimize over all (non-normalized) functions
Normalization Lemma:
Leth:Rn+7→R+ be a function and
Sh:Rn+7→Rn+ Sh(y) =h(y)y
Givenw :int(∆n)7→Rn+ we definew
h=Sh◦w.
Thenw is an optimal transport for the pair (P,Q) iffwh is an optimal transport for the pair (P,S#h(Q)).
Proof.
EP[log(hµ,w
h(µ)i)] =
EP[log(h(w(µ))hµ,w(µ)i]
Normalization Lemma:
Leth:Rn+7→R+ be a function and
Sh:Rn+7→Rn+ Sh(y) =h(y)y
Givenw :int(∆n)7→Rn+ we definew
h=Sh◦w.
Thenw is an optimal transport for the pair (P,Q) iffwh is an optimal transport for the pair (P,S#h(Q)).
Proof.
EP[log(hµ,w
h(µ)i)] =
EP[log(h(w(µ))hµ,w(µ)i]
Theorem [PW14]:
LetP(a.c. with respect to Lebesgue) onint(∆n) and Q on
Rn+be as
above such that
EP[log(hµ,w(µ)i)]7→min!
has a finite value. Then there is an optimal transportwb :int(∆n)7→
Rn+.
Assuming (w.l.g.) thathµ,wb(µ)i= 1,for allµ, the functionw is in the supergradient of anexponentiallyconcave functionϕand thereforew is