Predictable Forward Performance Processes:

The Binomial Case ∗

Bahman Angoshtari† Thaleia Zariphopoulou‡ Xun Yu Zhou§

March 23, 2017

Abstract

We introduce a new class of forward performance processes that are endogenous and
predictable with regards to an underlying market information set, and are updated at
discrete times. Such performance criteria accommodate short-term predictability of
asset returns, sequential learning and other dynamically unfolding factors affecting
optimal portfolio choice. We analyze in detail a binomial model whose parameters are
random and updated dynamically as the market evolves. We show that the key step in
the construction of the associated predictable forward performance process is to solve a
single-period inverse investment problem, namely, to determine, period-by-period and
conditionally on the current market information, the end-time utility function from a
given initial-time value function. We reduce this inverse problem to solving a single
variable functional equation, and establish conditions for the existence and uniqueness
of its solutions in the class of inverse marginal functions.
Keywords: Portfolio selection, forward performance processes, binomial model, inverse
investment problem, functional equation, predictability.

1 Introduction
The classical portfolio selection paradigm is based on three fundamental ingredients: a given
investment horizon, [0, T ], a performance function (such as a utility or a risk-return trade-
off), UT (·), applied at the end of the horizon, and a market model which yields the random
investment opportunities available over [0, T ) . This triplet is exogenously and entirely spec-
ified at initial time, t = 0.
∗
This work was presented at the SIAM conferences in Financial Mathematics in Chicago and Austin, and
seminars at Oxford University and University of Michigan, Ann Arbor. The authors would like to thank
the participants for fruitful comments and suggestions, as well as the Oxford-Man Institute of Quantitative
Finance for its support and hospitality. Zhou gratefully acknowledges financial support through a start-up
grant at Columbia University and through Oxford–Nie Financial Big Data Lab.
†
Department of Mathematics, University of Michigan, Ann Arbor, MI 48109, USA. Email:
bango@umich.edu.
‡
Department of Mathematics and IROM, The University of Texas at Austin, Austin and the Oxford-Man
Institute, University of Oxford, Oxford, UK. Email: zariphop@math.utexas.edu.
§
Department of IEOR, Columbia University, New York, NY 10027. Email: xz2574@columbia.edu.

1
 Once these ingredients are chosen, one then solves for the optimal strategy π ∗ (·), and
derives the value function U0 (·) at t = 0 as the expectation of the terminal utility of optimal
wealth. The value function thus stipulates the best possible performance value achievable
from each and every amount of initial wealth, and hence it can be in turn considered as a
performance criterion at t = 0. Here, the terminal performance function UT (·) is exogenous,
and the optimal strategy π ∗ (·) and the initial performance function U0 (·) are endogenous.
The model therefore entails a backward approach in time, from UT (·) to U0 (·). This is also in
accordance with the celebrated Dynamic Programming Principle (DPP), or otherwise known
as Bellman’s principle of optimality.1
Despite its classical mathematical foundations and theoretical appeal, this approach
nonetheless has several shortcomings, and hardly reflects what happens in investment prac-
tice.
Firstly, it relies heavily on the model selection for the entire investment horizon. As a
result, once a model is chosen for [0, T ] , no revisions are allowed, for it would lead to time-
inconsistent decisions. Therefore, any additional information coming from realized returns,
and other sources of learning, as the market evolves, cannot be incorporated in the investment
decisions.
This issue has been partially addressed by allowing for optimization over a family of
models, with a wealth of results on the so-called model robustness/ambiguity problem. But
even with this extended modeling approach, one also has to pre-commit to a family of possible
models, which itself may change as time evolves.2 Furthermore, uncertainty might be, more
generally, generated by both changes in asset prices and exogenous factors, and it is quite
difficult to model them accurately, especially for a long time ahead.3
The second difficulty is the pre-commitment at the initial time to a terminal utility.
Indeed, it might be difficult to assess and specify the performance function, especially if the
investment horizon is sufficiently long. It is more plausible to know the utility or the resulting
preferred allocations for now or the immediate future, and then to preserve it under certain
optimality criteria. This was firstly pointed out by Fisher Black in 1968 (see Black (1988))
where he argued that, practically, investors choose their initial/current desired allocation and
proposed a way to update this allocation in the future under time consistent expected utility
criteria. He did so through an equation that the allocation must satisfy through time. This
equation was much later extensively analyzed by Musiela and Zariphopoulou (2010b) and
more recently by Geng and Zariphopoulou (2017) where its long-term (turnpike) behavior is
studied.
Thirdly, it is very seldom the case that an optimal investment problem “terminates” at
a single horizon T. In practice, investment decisions are made for a series of relatively small
time horizons, say [0, T1 ] , [T1 , T2 ] , . . . , [Tn, Tn+1 ] , . . . , forward in time. Thus, a framework of
“pasting” these individual investment problems in a time-consistent manner as the market
evolves is needed. Establishing such a framework is by no means trivial as we will explain
in the sequel, but it captures reality, for it offers substantial flexibility in terms of learning
and model revision, dynamic risk preferences criteria and rolling horizons.
The above considerations have led to the development of the so-called forward perfor-
1
See Dreyfus (2002) for a historical account on dynamic programming.
2
See Kallblad et al. (2017) for a critique on model ambiguity.
3
Nadtochiy and Tehranchi (2015) argue that stochastic factors must be incorporated “forward in time”
and propose a finite-dimensional model for it.

2
mance measurement, initially proposed by Musiela and Zariphopoulou (2006) and later ex-
tended by the same authors in a series of papers (see Musiela and Zariphopoulou (2009,
2011)) and by other authors (see El Karoui and Mrad (2013), and Nadtochiy and Tehranchi
(2015)) in continuous-time market settings. The main idea of the forward approach is that
instead of fixing, as in the classical setting, an investment horizon, a market model and a
terminal utility, one starts with an initial performance measurement and updates it forward
in time as the market and other underlying stochastic factors evolve. The evolution of the
forward process is dictated by a forward-in-time version of the DPP and, thus, it ensures
time-consistency across different times.
With the exception of a special case studied in Musiela and Zariphopoulou (2003) and
Musiela et al. (2016), in the context of indifference prices under exponential forward cri-
teria, the existing results have so far focused exclusively on continuous-time, Itô-diffusion
settings, in which both trading and performance valuation are carried out continuously in
time. It was shown in Musiela and Zariphopoulou (2010a) that the forward performance pro-
cess is associated with an ill-posed infinite-dimensional stochastic partial differential equa-
tion (SPDE), the same way that the classical value function satisfies the finite-dimensional
Hamilton-Jacobi-Bellman equation (HJB). This performance SPDE has been subsequently
studied in El Karoui and Mrad (2013), Nadtochiy and Zariphopoulou (2014), Nadtochiy
and Tehranchi (2015) and, more recently, in Shkolnikov et al. (2016) for asset price factors
evolving at different time scales. Despite the technical challenges that this forward SPDE
presents (ill-posedness, high or infinite dimensionality, degeneracies, and volatility specifica-
tion), the continuous-time cases are tractable because stochastic calculus can be employed
and infinitesimal arguments can be, in turn, developed.
However, the continuous-time setting has a major drawback in that it is hard to see how
exactly the performance criterion evolves from one instant to another. This evolution is lost
at the infinitesimal level and hidden behind the (generally intractable) stochastic PDE.
The aim of this paper is to initiate the development and a systematic study of forward
investment performance processes that are discrete in time, while trading can be either
discrete or continuous in time. We will introduce an iterative mechanism in which an investor
updates/predicts her performance criterion at the next investment period, based on both her
current performance and her assessment of the upcoming market dynamics in the next period.
This predictability will be present in an explicit and transparent manner.
In addition to the conceptual motivation described above, there are also practical con-
siderations in studying the discrete-time predictable forward performance. Indeed, in in-
vestment practice, trading occurs at discrete times and not continuously. More importantly,
performance criteria are directly or indirectly determined by individuals, such as higher-level
managers or clients, and not by the portfolio manager. These “performance evaluators” use
information sets that are different, both in terms of contents and updating frequency, from
the ones used by the portfolio manager. For example, a portfolio manager may have access
to various data sets, proprietary forecasting models, and sophisticated trading strategies
which are out of reach of (or simply deemed as “too detailed” to be considered by) the per-
formance evaluator. Similarly, even if trading can happen at extremely higher frequencies
(hence almost close to continuous trading), performance assessment/update takes place at a
much slower pace, e.g., a senior manager will not keep track of the performance of a portfolio
or update the performance criterion as frequently as the subordinate portfolio manager in
charge of that portfolio.
In this paper, we will consider an (indefinite) series of time points, 0 = t0 , t1 , . . . , tn ,

3
. . . , at which the performance measurement is evaluated and updated. The (short) period
between any given two neighboring points is called an evaluation period. We then introduce
forward performance processes that are predictable with regards to the information at the
most recent evaluation time.
We are motivated to introduce this class of criteria for two reasons. Firstly, it is natural
to infer at the beginning of the evaluation period the criterion we use for the end of it.
Conceptually different from the expected utility framework, here the criterion to be used at
the end of the period will be endogenous, not exogenous.
Secondly, it is more feasible to estimate the market parameters for just one evaluation
period ahead, than for longer periods. For example, the volatility can be reliably forecasted
for a short time ahead using the so-called realized volatility introduced by the seminal work
of Barndorff-Nielsen and Shephard (2002). On the other hand, short-term predictability of
equity risk premium has been subject to extended studies in the last three decades (see,
among others, Fama and French (1989)).
We stress that by looking at short-term predictability, we do not discard any long-term
forecasting modeling input. On the contrary, the approach we propose accommodates both
short- and long-term predictability modeling, by tracking the market and incorporating the
unfolding information as time moves along. This cannot be done in the classical setting in
which the market model, which may contain both short- and long-term factors, is pre-chosen
at the initial time.
To highlight the key ideas of predictable forward performance processes, we start our
analysis with a simple yet still rich enough setting. The market consists of two securities,
a riskless asset and a stock whose price evolves according to a binomial model at times
0 = t0 , t1 , ., tn , . . . , at which the forward performance evaluation also occurs. The market
model is more general than the standard binomial tree, in that the asset returns and their
probabilities can be estimated/determined only one period ahead. Such a setting allows for
genuine dynamic updating of the underlying parameters, as the market evolves from one
period to the next.
In generating a predictable forward performance process, the investor starts at t = 0
and chooses (i.e. estimates) the market parameters for the upcoming trading/evaluation
period [0, t1 ) . She also chooses her initial performance function U0 (·) , and derives a utility
(performance) function U1 (·) that is deterministic, such that the pair (U0 , U1 ) is consistent
with the investment problem in [0, t1 ] , with U0 , U1 being respectively the value function and
terminal utility function. Then, at t1 , the agent repeats exactly the same procedure for
the next period. Proceeding iteratively forward in time, and conditionally on the current
information, a predictable performance process is constructed together with the optimal
allocations and their wealth processes.4
Therefore, technically, we are left to solve a single-period investment problem where the
value function is given and the terminal utility function is to be found. We term this as a
single-period inverse investment problem, which needs to be solved sequentially “period-by-
period,” conditionally on the information at the beginning of this trading period. It turns
out that the key to solving this problem is a linear functional equation, which relates the
4
Here we assume that the updating and trading take place at the same time. As discussed above, this
does not have to be the case. However, we choose to study this parsimonious model in order to highlight
the significance of updating the performance measurement in discrete times without getting into much
technicality.

4
inverse marginal processes, at the beginning and the end of each evaluation period, with
coefficients depending on the random market inputs. We analyze this equation in detail,
and establish conditions for existence and uniqueness of the solutions in the class of inverse
marginal functions.
The paper is structured as follows. In Section 2, we introduce the notion of predictable
forward performance processes in a general market setting. We then formulate a binomial
model with random, dynamically updated parameters in Section 3. In Section 4, we apply
the definition of predictable forward performance processes to the binomial model, and show
that their construction reduces to solving an inverse investment problem. In Section 5,
this inverse problem is shown to be equivalent to a functional equation. We derive sufficient
existence and uniqueness conditions as well as the explicit solution to the functional equation
in Section 6. Finally, we present the general construction algorithm in Section 7, and conclude
in Section 8. Proofs of the main results are relegated to an Appendix.

2 Predictable forward performance processes: A gen-

eral definition
In this section, we define discrete-time predictable forward performance processes in a general
market model. Starting from the next section, we will restrict the market setting to a
binomial model with random parameters, and provide a detailed discussion on the existence
and construction of such performance processes.
The investment paradigm is cast in a probability space (Ω, F, P) augmented with a fil-
tration (Ft ), t ≥ 0. We denote by X (t, x) the set of all the admissible wealth processes Xs ,
s ≥ t, starting with Xt = x and such that Xs is Fs -measurable. The term “admissible” is
for now generic and will be specified once a specific market model is introduced in the sequel.
We call a function U : R+ → R+ a utility (or performance) function if U ∈ C 2 (R+ ),
U 0 > 0, U 00 < 0, and satisfies the Inada conditions: limx→0+ U 0 (x) = ∞ and limx→∞ U 0 (x) =
0.
For any σ-algebra G ⊆ F, the set of G-measurable utility (or performance) functions is
defined as

U(G) = U : R+ × Ω → R U (x, ·) is G -measurable for each x ∈ R+ ,


and U (·, ω) is a utility function a.s.} .

In other words, the elements of U (G) are entirely known (predicted) based on G, as they are
predictable by the information contained in G. Alternatively, we may think of U ∈ U (G) as
a deterministic utility function, given the information in G.
Next, we introduce the predictable forward performance processes. To ease the notation,
we skip the ω-argument throughout.

Definition 1. Let discrete time points 0 = t0 < t1 < · · · < tn < · · · be given. A family of
random functions {U0 , U1 , U2 , · · · } is a predictable forward performance process with respect
to (Ft ) if, for Xn = Xtn and Fn = Ftn , n = 1, 2, . . . , the following conditions hold:

(i) U0 is a deterministic utility function and Un ∈ U(Fn−1 ).

5
 (ii) For any initial wealth x > 0 and any admissible wealth process X ∈ X (0, x),
Un−1 (Xn−1 ) ≥ EP [Un (Xn )| Fn−1 ] .

(iii) For any initial wealth x > 0, there exists an admissible wealth process X ∗ ∈ X (0, x)
such that
∗
= EP [Un (Xn∗ )| Fn−1 ] .


Un−1 Xn−1

We stress that there are no specific assumptions on the market model and how often
trading occurs. The asset price processes can be discrete or continuous in time and, for
the latter case, trading can be discrete or continuous. Furthermore, if trading takes place
at discrete times, the rebalancing periods do not need to be aligned with the performance
assessment times.
In practice, as mentioned in Introduction, it is typically the case that trading occurs more
frequently than the performance evaluation, but the above definition accommodates cases
when there is perfect alignment - as in the binomial model we will study herein - and the
less realistic case when trading occurs less frequently than the performance measurement.
Compared with the continuous-time counterpart initially proposed by Musiela and Za-
riphopoulou (2009), the fundamentally distinctive element of Definition 1 is condition (i),
which explicitly requires that the performance function at the next upcoming assessment
time be entirely determined from the information up to the present time.
On the other hand, as in the continuous-time case, properties (ii)-(iii) draw from Bell-
man’s principle of optimality, which stipulates that the processes Un (Xn ) and Un (Xn∗ ),
n = 0, 1, . . . , are, respectively, a supermartingale and a martingale with respect to the
filtration (Fn ) . Since the Bellman principle underlines time-consistency, properties (ii)-(iii)
directly ensure that the investment problem is time-consistent under the predictable forward
performance criterion.
Hence, the above performance measurement is essentially endogenized by the time-
consistency requirements (ii)-(iii).
We also note that the predictability of risk preferences is implicitly present in the classical
expected utility in finite horizon settings, say [0, T ] , in which a deterministic utility for T is
pre-chosen at initial time t0 = 0, and it is thus F0 -measurable. The fundamental difference,
however, is that the terminal utility function in the classical theory is exogenous, instead of
endogenous.
Definition 1 suggests a general scheme for constructing predictable forward performance
functions in discrete times. Indeed, starting from an initial datum U0 , given at time t0 = 0,
the entire family U1 , .., Un , .., can be obtained by determining Un from Un−1 iteratively,
n = 1, 2, . . . , in the way described below.
Properties (ii)–(iii) dictate that, for each trading period [tn−1 , tn ), we have
∗


Un−1 Xn−1 = sup EP [Un (Xn )| Fn−1 ] . (1)
∗
Xn ∈X (tn−1 ,Xn−1 )

At instant tn−1 , since Fn−1 is realized, the random functions Un−1 and Un are both
∗
deterministic and so is Xn−1 . This, in turn, suggests that we should consider the following
“single-period ” investment problem
Un−1 (x) = sup EP [Un (Xn )| Fn−1 ] , (2)
Xn ∈Xn−1,n (x)

6
for x > 0, where, with a slight abuse of notation, we use Xn−1,n (x) to denote the set of
admissible wealths at tn starting at tn−1 with wealth x.
Therefore, if we are able to determine, for each n = 1, 2, . . . , a performance function Un
∈ U (Fn−1 ) , such that the pair (Un , Un+1 ) satisfies (2), then we will have an iterative scheme
to construct the entire predictable forward performance process, starting from U0 .
One readily recognizes that (2) would be the classical expected utility problem if the ob-
jective were to derive Un−1 from Un , with Un being a deterministic utility function. Therefore,
what we consider now is an inverse investment problem in that we are given its initial value
function and we seek a terminal utility that is consistent with it, with both these functions
being deterministic (conditionally on Fn−1 ).
To our best knowledge, such inverse discrete-time problems have not been considered in
the literature. The aim herein is to initiate a concise study of such performance criteria for
general market settings. We start with the binomial case in which, however, the parameters -
including the transition probabilities and price levels - are not known a priori but are updated
as the market moves. Recall that, while this is very much in accordance with real investment
practice, such a model is not implementable in the classical expected utility settings because
model commitment occurs once, at the initial time. As we will see, while the binomial case
is one of the simplest discrete-time market models, its analysis is sufficiently rich and its
results reveal the key economic insights regarding the predictable performance criteria.

3 A binomial market model with random, dynamically

updated parameters
We consider a market with two traded assets, a riskless bond and a stock. The bond is taken
to be the numeraire and assumed to offer, without loss of generality, zero interest rate. The
stock price at times t0 , t1 , . . . , evolves according to a binomial model that we now specify.
Let Rn be the total return of the stock over period [tn−1 , tn ). Here, Rn is a random variable
with two values Rnu > Rnd . We assume that Rn , Rnu , and Rnd , n = 1, 2, . . . , are all random
variables in a measurable space (Ω, F) augmented with a filtration (Fn ) , n = 1, 2, . . . ,
with Fn representing the information available at tn . Moreover, we assume that Rn is Fn -
measurable, and that its values, Rnu and Rnd , are taken to be Fn−1 -measurable. In other
words, the high and low return levels for each investment period are known at the beginning
of this period, while the realized return is known at its end.
Finally, the historical measure P is a probability measure on (Ω, F) and the following
standard no-arbitrage assumption is satisfied.

Assumption 2. For all n = 1, 2, . . . :

(i) 0 < Rnd < 1 < Rnu , P -almost surely; and,



(ii) 0 < EP 1{Rn =Rnu } |Fn−1 = 1 − EP 1{Rn =Rnd } |Fn−1 < 1, P -almost surely.

For n = 1, 2, . . . , the Fn−1 -measurable random variable

   
pn := EP 1{Rn =Rnu } |Fn−1 = 1 − EP 1{Rn =Rnd } |Fn−1 ,

7
represents the best estimate of the probability of an upward jump over [tn−1 , tn ], given the
information available at tn−1 . In practice, pn corresponds to the outcome of a sequential
learning procedure that is conducted at tn−1 .
We assume no further information about the physical measure P . In particular, we do
not assume that P is known, other than it satisfies Assumption 2.
The investor trades between the stock and the bond using self-financing strategies. She
starts at t0 = 0 with total wealth x > 0 and rebalances her portfolio at times tn , n = 1, 2, . . . .
At the beginning of each period, say [tn , tn+1 ), she chooses the amount πn+1 to be invested
in the stock (and the rest in the bond) for this period. In turn, her wealth process, denoted
by Xnπ , n = 1, 2, .., evolves according to the wealth equation
π
Xn+1 = Xnπ + πn+1 (Rn+1 − 1),

with X0 = x.
The investor is allowed to short the stock but her wealth can never become negative; thus
πn+1 must satisfy
Xnπ Xnπ
− u ≤ πn+1 ≤ d
; n = 1, 2, . . . (3)
Rn+1 − 1 1 − Rn+1

We call an investment strategy π = {πn }∞ n=1 admissible if it is self-financing, πn is Fn−1 -

mble, and (3) is satisfied P -almost surely. A wealth process X = {Xnπ }∞ n=0 is then admissible
if the strategy π that generates it is admissible.
We recall that X (n, x) is the set of admissible wealth processes {Xm }∞
m=n , starting with
Xn = x.
We also introduce the auxiliary ”single-step” set of admissible portfolios πn+1 , chosen at
tn for the trading period [tn , tn+1 ) assuming wealth x at tn , by
 
x x
An,n+1 (x) = πn+1 : πn+1 is Fn -measurable, − u ≤ πn+1 ≤ d
, x>0 ,
Rn+1 − 1 1 − Rn+1
as well as the corresponding set of admissible wealth processes

Xn,n+1 (x) = {x + πn+1 Rn+1 : πn+1 ∈ An,n+1 (x), x > 0} .

4 Problem statement and reduction to the single-period

inverse investment problem
In this section, we consider predictable forward performance processes in the binomial model,
and show that their construction reduces to solving a series of single-period inverse invest-
ment problems.
The investor starts with an initial utility U0 and updates her performance criteria at
times t1 , t2 , . . . , with the associated performance functions U1 , U2 , . . . satisfying Definition
1.
We now present the procedure that yields the construction of a predictable forward
performance process starting from U0 , and determining Un from Un−1 , iteratively for n =
1, 2, . . . .

8
 At t0 = 0, equation (1) becomes
h  i h  i
U0 (x) = sup E P U1 (X1 )F0 = sup E P U1 x + π1 (R1 − 1) ; x > 0. (4)

X1 ∈X (0,x) π1 ∈A0,1 (x)

Since the market parameters (R1u , R1d , p1 ) and the initial datum U0 are known at t0 , finding
a deterministic (F0 -measurable) U1 reduces to the single-period inverse investment problem
discussed in Section 2. Let us for the moment assume that we are able to solve this inverse
problem to obtain U1 .
At t = t1 , the investor observes the realization of the stock return R1 and estimates the
parameters (R2u , R2d , p2 ) for the second trading period [t1 , t2 ). Setting n = 2 in (1) then yields

U1 (X1∗ (x)) = sup EP [U2 (X2 )| F1 ] , (5)

X2 ∈X (1,X1∗ (x))

where X1∗ (x) is the optimal wealth generated at t1 , starting at x at t0 = 0, from the previous
period.
It follows from the classical expected utility theory (see also Theorem 4 below) that
X1∗ (x) = I1 (ρ1 U00 (x)), x > 0, where I1 = (U10 )−1 and ρ1 is the pricing kernel over the period
[0, t1 ), given by

1 − R1d R1u − 1
ρ1 = 1 u
{R1 =R1 } + 1 d .
p1 (R1u − R1d ) (1 − p1 )(R1u − R1d ) {R1 =R1 }

The mapping x → X1∗ (x) is strictly increasing for each x > 0 and of full range, since I1 and
U00 are both strictly decreasing functions, ρ1 > 0, and the Inada conditions yield X1∗ (0) = 0
and X1∗ (∞) = ∞.
Since X1∗ (x) is F1 -measurable and the parameters (R2u , R2d , p2 ) together with U1 are all
known at t = t1 , we deduce that (5) reduces, with a slight abuse of notation, to finding
U2 (·) ∈ U (F1 ) such that

U1 (x) = sup EP [U2 (x + π2 (R2 − 1))| F1 ] ; x > 0,

π2 ∈A1,2 (x)

with U1 given. In other words, one needs to solve yet another inverse investment problem
that is mathematically identical to (4).
At t = tn , in exactly the same manner as above, we have to solve

Un (x) = sup EP [Un+1 (x + πn+1 (Rn+1 − 1))| Fn ] ; x > 0,

πn+1 ∈An,n+1 (x)

thereby deriving Un+1 from Un , with Un+1 ∈ U (Fn+1 ) and with the parameters (Rnu , Rnd , pn )
known.
Thus, all the terms of a predictable forward performance process can be obtained, starting
from any arbitrary initial wealth x > 0 and proceeding iteratively solving a “period-by-
period” inverse optimization problem. Moreover, as we will show in the next section, we will
also derive the optimal portfolio and wealth processes at the same time.
To summarize, the crucial step in the entire predictable forward construction is to solve
this single-period inverse investment problem. We do this in the next section.

9
5 The single-period inverse investment problem
We focus on the analysis of the inverse investment problem (4). To ease the presentation,
we introduce a simplified notation. We set t0 = 0, t1 = 1 and R1 = R taking values u and d,
u > 1 and 0 < d < 1, with probability 0 < p < 1 and 1 − p, respectively. We recall the risk
neutral probabilities
1−d u−1
q= and 1 − q = ,
u−d u−d
and the pricing kernel
q 1−q
ρ1 = ρu 1{R=u} + ρd 1{R=d} := 1{R=u} + 1{R=d} . (6)
p 1−p

The investor starts with wealth X0 = x > 0, and invests the amount π in the stock. Her
wealth at t = 1 is then given by the random variable X = x + π(R − 1). The no-bankruptcy
constraint (3) becomes π(x) ≤ π ≤ π(x), with
x x
π(x) = − < 0 and π(x) = > 0.
u−1 1−d
We denote the set of admissible portfolios as
A(x) = {π ∈ R, and π(x) ≤ π ≤ π(x), x > 0}.
Given an initial utility function U0 , we then seek another performance function U1 , such that
U0 (x) = sup EP [U1 (x + π(R − 1))] ; x > 0. (7)
π∈A(x)

Let U be the set of deterministic utility functions. We introduce the set of inverse
marginal functions I,
 
1 + 0
I := I ∈ C (R ) : I < 0, lim I(y) = 0, lim+ I(y) = ∞ . (8)
y→∞ y→0

Note that if functions U and I satisfy I = (U 0 )−1 , then U is a utility function if and only if
I is an inverse marginal function.
Assuming for now that a utility function U1 satisfying (7) exists, we consider the inverse
marginal functions
I0 = (U00 )−1 and I1 = (U10 )−1 .
Our main goal in this section is to show that the inverse investment problem (7) reduces to
a functional equation in terms of I0 and I1 ; see (9) below.
The following theorem is one of the main results herein, establishing a direct relationship
between the inverse marginals at the beginning and at the end of the trading period [0, 1] ,
when the corresponding utilities are related by (7).
Theorem 3. Let U0 , U1 ∈ U satisfy the optimization problem (7). Then, their inverse
marginals I0 and I1 must satisfy the linear functional equation
I1 (ay) + bI1 (y) = (1 + b) I0 (c y); y > 0, (9)
where
1−p q 1−q 1−p
a= , b= and c= . (10)
p 1−q q 1−q

10
Proof. From standard arguments, for all x > 0, there exists an optimizer π ∗ (x) for (7)
satisfying the first-order condition

p(u − 1)U10 (x + π ∗ (x)(u − 1)) + (1 − p)U10 (x + π ∗ (x)(d − 1)) = 0. (11)

On the other hand, it follows from (7) that

p U1 (x + π ∗ (x)(u − 1)) + (1 − p) U1 (x + π ∗ (x)(d − 1)) = U0 (x).

Differentiating the above equation and using (11) yield

p U10 (x + π ∗ (x)(u − 1)) + (1 − p) U10 (x + π ∗ (x)(d − 1)) = U00 (x). (12)

Solving the linear system (11)-(12) gives

(1 − d) 0
U10 (x + π ∗ (x)(u − 1)) = U (x)
p(u − d) 0

and
(u − 1)
U10 (x + π ∗ (x)(d − 1)) = U00 (x).
(1 − p)(u − d)
Therefore, the optimal allocation function π ∗ (x) satisfies
  
x + π ∗ (x)(u − 1) = I1 1−d U00 (x) ,
 p(u−d)  (13)
∗
x + π (x)(d − 1) = I1 u−1 0
U
(1−p)(u−d) 0
(x) ,

from which we obtain the solution

    
∗ 1 1−d 0 u−1 0
π (x) = I1 U (x) − I1 U (x) ; x > 0.
u−d p(u − d) 0 (1 − p)(u − d) 0

Substituting the above in either of the equations in (13) yields

   
1−d 1−d 0 u−1 u−1 0
I1 U (x) + I1 U (x) = x.
u−d p(u − d) 0 u−d (1 − p)(u − d) 0
 
Changing variables x = I0 (1−p)(u−d)
u−1
y , y > 0, the above becomes
   
(1 − p)(1 − d) u−1 u−d (1 − p)(u − d)
I1 y + I1 (y) = I0 y ; y > 0.
p(u − 1) 1−d 1−d u−1

Noting (10) we conclude.

Next, we show by an explicit construction how to recover U1 from I1 . At the same time
we derive the optimal portfolio π ∗ (x) and its wealth X ∗ (x).

Theorem 4. Let U0 be a utility function and I0 be its inverse marginal, and I1 be an inverse
marginal solving the functional equation (9). Let also ρ1 be the pricing kernel given by (6).
Then, the following statements hold.

11
 (i) The function U1 defined by
"Z #
x
U1 (x) := U0 (1) + EP I1−1 (ξ)dξ ; x>0 (14)
I1 (ρ1 U00 (1))

is a well-defined utility function.

(ii) We have
U0 (x) = sup EP [U1 (x + π(R − 1))] ; x > 0.
π∈A(x)

(iii) The optimal wealth X1∗ (x) and the associated optimal investment allocation π ∗ (x) are
given, respectively, by

X1∗ (x) = I1 (ρ1 U00 (x)) = X ∗,u (x)1{R=u} + X ∗,d (x)1{R=d}

and
X ∗,u (x) − X ∗,d (x)
π ∗ (x) = ,
u−d
with    
∗,u q 0 ∗,d 1−q 0
X = I1 U (x) and X = I1 U (x) .
p 0 1−p 0

Proof. See Appendix A.

Remark 5. As shown in the proof of Theorem 4, we can replace (14) with

"Z #
x
U1 (x) := U0 (c) + EP I1−1 (ξ)dξ ; x > 0,
I1 (ρ1 U00 (c))

for any arbitrary constant c > 0. The choice of c does not change the value of U1 (x), neither
the optimal policies.

Theorem 4 reduces the inverse investment problem (7) to the functional equation (9).
We study this functional equation in the next section.

6 A functional equation for inverse marginals

In this section, we analyze the linear functional equation (9), with I0 given and I1 to be
found, for positive constants a, b, c, given by (10). We provide conditions for the existence
and uniqueness of its solutions and, in particular, solutions in the class of inverse marginal
functions.
When a = 1, the unique solution is trivially I1 (y) = I0 (y). This is economically intuitive.
If p = q, then essentially there is no risk premium to exploit. As a result, when r = 0 as
assumed herein, the pricing kernel becomes a constant, ρ = 1, and the optimal wealth reduces
to X ∗ (x) = x. In turn, the value function (at t = 0) coincides with the terminal utility. So
the forward performance remains constant, U0 (x) = U1 (x) , or their inverse marginals I0

12
and I1 coincide.5 Indeed, there is no reason to modify the performance function in a market
with no investment opportunities.

Henceforth we assume that a 6= 1. We start with an example showing that a general

solution of (9) may not be unique, even if we restrict the solutions to inverse marginals.

Example 6. Let loga b < 0 and I0 (y) = y loga b , y > 0. It is easy to check that the function
I1 (y) = δy loga b , y > 0, with δ = 2b (1+b)
c− loga b
> 0, is a solution to (9).
However, this particular solution is not the only solution. Indeed, consider any differ-
entiable anti-periodic function, say Θ(z) = −Θ(z + ln a), for which there exists a constant
M > 0 such that
− loga b (1 + b) loga b
sup (|Θ(z)|, |Θ0 (z)|) < M < δ =− −
.
z∈R 1 − loga b 2b c loga b (1 − loga b)

For instance, Θ(x) = M sin( lnxa π) is such a function. One can then directly check that the
function
I˜1 (y) = y loga b (δ + Θ(ln y)) ; y > 0
is a solution.
As a matter of fact, both solutions I1 and I˜1 are inverse marginals. This is obvious for I1 .
As for I˜1 , we have limy→∞ I˜1 (y) = 0 since loga b < 0. Moreover, it follows from the inequality
I˜1 (y) ≥ y loga b (δ − M ), y > 0, that limy→0+ I˜1 (y) = ∞. Furthermore,
0
 
Θ (ln y)
I˜1 (y) = y a
0 log b−1
loga b δ + Θ(ln y) +
loga b
 
loga b−1 M loga b − M
≤y loga b δ − < 0; y > 0.
loga b

Thus, in general, there is no uniqueness even among inverse marginals.

The above example suggests that we need additional conditions to ensure uniqueness. To
identify these conditions, we first note that (9) is a functional equation of the more general
form
F (f (y)) = g(y)F (y) + h(y), (15)
with f , g, and h given functions, y ∈ Y ⊆ R and F to be found. The equations of this type
have been studied in the literature; see Kuczma et al. (1990) and the references therein for
a general exposition.
In general, such equations have many solutions. A trivial example is F (y + 1) = F (y),
y ∈ R, for which any periodic function with period 1 is a solution. Such non-uniqueness
often renders the underlying equation inapplicable for concrete problems, where a single
well-defined solution is usually needed. For the general equation (15), conditions for the
uniqueness of solutions usually limit the set of solutions by imposing additional assumption
on F (y0 ), where y0 is a fixed point for f : f (y0 ) = y0 . In the example of the equation
5
This is also in accordance with the so-called time-monotone forward processes in the continuous-time
setting. For example, in Musiela and Zariphopoulou (2010b), it is shown that this forward performance
Rt 2
is given by U (x, t) = u x, 0 |λs | ds , with u (x, t) a deterministic function and the process λ being the
market price of risk. If λ ≡ 0, then U (x, t) = u (x, 0) = U (x, 0) , for all t > 0.

13
F (y + 1) = F (y), y ∈ R, if we require a solution to be such that limy→∞ F (y) exists, then
F ≡ 0 becomes the only possible solution. Note here that ∞ is a fixed point of the function
f (y) = y + 1.
For equation (9), f (y) = ay, g (y) = −b and h (y) = (1 + b) G(c y). Therefore, uniqueness
conditions should impose additional assumptions on F at y1 = 0 and y2 = ∞, which are the
fixed points of f (y) = ay.

We start with the following auxiliary result in which we provide general uniqueness
conditions for equation (9). Afterwards, we will strengthen the results for the family of
inverse marginals.
Lemma 7. Let I0 be given. Then there exists at most one solution to (9), say I, sat-
isfying limy→0+ y − loga b I(y) = 0. Similarly, there exists at most one solution satisfying
limy→∞ y − loga b I(y) = 0.

Proof. Let F1 and F2 be two solutions of (9) that both satisfy either conditions given in the
lemma. We show that their difference w := F1 − F2 ≡ 0.
The function w satisfies the homogenous equation w(ay) = −bw(y), y > 0. Therefore,
for k = 1, 2, . . . ,
w(ay) w(a2 y) w(ak y)
w(y) = = = · · · = ,
−b (−b)2 (−b)k
and y  y y
2 k
w(y) = −bw = (−b) w 2 = · · · = (−b) w k .
a a a
It then follows that for k = ±1, ±2, . . . and y > 0,
  y   y − loga b   y 
|w(y)| = bk w k  = y loga b k w k 
   
a a a
 y − loga b   y    y 
≤ y loga b k  1 k   2 k  .
F + F
   
a a a
The right side vanishes as either k → ∞ or k → −∞, and we conclude.

We note that the function I˜1 in Example 6 satisfies neither conditions in Lemma 7, and
thus uniqueness fails.

Next, we state the main result for this section, which provides sufficient conditions for
existence and uniqueness of solutions to (9) that are inverse marginal functions.

Theorem 8. Let I0 in (9) be an inverse marginal utility, i.e. I0 ∈ I with I defined in (8).
Define the functions
Φ0 (y) = I0 (a c y) − bI0 (c y) and Ψ0 (y) = y − loga b I0 (c y); y > 0. (16)
The following assertions hold:

(i) If Φ0 is strictly increasing and, either a > 1 and limy→∞ Ψ0 (y) = 0 or a < 1 and
limy→0+ Ψ0 (y) = 0, then a solution of (9) is given by
∞
1+b X
I1 (y) = (−1)m b−m I0 ( am c y); y > 0. (17)
b m=0

14
 (ii) If Φ0 is strictly decreasing and, either a > 1 and limy→0+ Ψ0 (y) = 0 or a < 1 and
limy→∞ Ψ0 (y) = 0, then a solution of (9) is given by
∞
X
I1 (y) = (1 + b) (−1)m bm I0 (a−(m+1) c y); y > 0. (18)
m=0

(iii) In parts (i) and (ii), the corresponding I1 satisfies the uniqueness condition(s) of
Lemma 7 and, moreover, I1 ∈ I, i.e., I1 preserves the inverse marginal properties.

(iv) The function I1 in parts (i) and (ii), respectively, is the only positive solution of (9).
It is also the only inverse marginal that solves (9).

Proof. See Appendix B.

Next, we apply the above result to the case when the initial utility is a power function.
The following example provides results complementary to the ones in Example 6 where
uniqueness lacks as the result of not satisfying the conditions of Lemma 7.

−1 1− 1
Corollary 9. Let U0 (x) = 1 − 1θ x θ , x > 0, and assume that 1 6= θ > 0, θ 6= − loga b,
with a, b, c > 0 given by (10). Then, the following assertions hold:

(i) The unique marginal utility function that satisfies the functional equation (9) with the
initial I0 (y) = y −θ is given by

I1 (y) = δy −θ ; y > 0, (19)

1+b
where δ = cθ (a−θ +b)
.

(ii) The unique utility function U1 that satisfies the inverse investment problem (7) is given
by  −1
1 1 1 1
U1 (x) = δ θ 1 − x1− θ = δ θ U0 (x) ; x > 0.
θ

(iii) The corresponding optimal allocation is given by

δ(p/q)θ − 1
π ∗ (x) = x; x > 0.
u−1

So, if we start with an initial power utility U0 , then the forward utility at t = 1 is a
1
multiple of the initial datum, with the constant given by δ θ . Proceeding iteratively, the
utilities for all the future periods remain to be power functions. In other words, in the
binomial setting, the (predictable) power utility preferences are preserved throughout.

15
7 Construction of the predictable forward performance
process
We are now ready to present the general algorithm for the construction of forward perfor-
mance processes as well as the associated optimal investment strategies and their wealth
processes. We stress that one of the main strengths of our approach is that for every

trading
u d
period, say (tn , tn+1 ] , we do not update the model parameters pn+1 , Rn+1 , Rn+1 for this
period until tn . Thus we take full advantage of the incoming information up to time tn .
The algorithm is based on repeatedly applying the following result on the single-period
inverse investment problem (7).

Theorem 10. For the inverse investment problem (7), assume that the initial inverse
marginal I0 = (U00 )−1 satisfies condition (i) (resp. condition (ii)) in Theorem 8, and de-
fine I1 by (17) (resp. (18)). Then, the unique solution to (7) is given by
"Z #
x
U1 (x) = U0 (1) + EP I1−1 (ξ)dξ ; x > 0,
I1 (ρ1 U00 (1))

where ρ1 as in (6). Moreover, the optimal wealth X1∗ (x) and the associated optimal investment
allocation π ∗ (x) are given, respectively, by

X1∗ (x) = I1 (ρ1 U00 (x)) = X1∗,u (x)1{R1 =u} + X1∗,d (x)1{R1 =d} ,

∗ X1∗,u (x) − X1∗,d (x)

π (x) = ,
u−d
where    
q 0 1−q 0
X1∗,u (x) = I1 U (x) and X1∗,d (x) = I1 U (x) .
p 0 1−p 0

Proof. The results follow directly from Theorem 8 and Theorem 4.

Given an initial performance function U0 and initial wealth X0 , the following is the algo-
rithm for constructing the predictable forward performance process {U1 , U2 , · · · } along with
the associated optimal portfolio process {π1∗ , π2∗ , · · · } and the wealth process {X1∗ , X2∗ , · · · }
in the binomial market model.


• At t = 0 : Assess the market parameters R1u , R1d , p1 for the first investment period,
[0, t1 ) . Compute

1 − R1d q1 (1 − p1 ) 1 − q1 1 − p1
q1 = , a1 = , b1 = , and c1 = ,
R1u − R1d p1 (1 − q1 ) q1 1 − q1
and
q1 1 − q1
ρu1 = and ρd1 = .
p1 1 − p1

16
 Using (a1 , b1 , c1 ) check the conditions in part (i) (resp. (ii)) of Theorem 8, and obtain I1
from (17) (resp. (18)). Then, apply Theorem 10 to compute
Z x
U1 (x) = U0 (1) + p1 I1−1 (ξ)dξ
I1 (ρu 0
1 U0 (1))
Z x
+ (1 − p1 ) I1−1 (ξ)dξ; x > 0,
I1 (ρd1 U00 (1))

X1∗,u (X0 ) − X1∗,d (X0 )

π1∗ = ,
u−d
and
X1∗ = X0 + π1∗ (R1 − 1) ,
where
   
q1 0 1 − q1 0
X1∗,u (x) = I1 U (x) and X1∗,d (x) = I1 U (x) ; x > 0.
p1 0 1 − p1 0

• At t = tn (n = 1, 2, · · · ): We have already obtained {U1 , · · · , Un ; I1 , · · · , In }, {π1∗ , · · · , πn∗ }

and {X1∗ , · · · , Xn∗ }. Estimate the market parameters (Rn+1
u d
, Rn+1 , pn+1 ) for the upcom-
ing investment period [tn , tn+1 ). Let

d
1 − Rn+1 qn+1 (1 − pn+1 ) 1 − qn+1 1 − pn+1
qn+1 = u d
, an+1 = , bn+1 = , and cn+1 = ,
Rn+1 − Rn+1 pn+1 (1 − qn+1 ) qn+1 1 − qn+1
and
qn+1 1 − qn+1
ρun+1 = and ρdn+1 = .
pn+1 1 − pn+1
Check the conditions in part (i) (resp. (ii)) in Theorem 8, using (an+1 , bn+1 , cn+1 ) (instead
of (a, b, c)) and In instead of I0 , and obtain In+1 from (17) (resp. (18)).6
Compute
Z x
−1
Un+1 (x) = Un (1) + pn+1 In+1 (ξ)dξ
In+1 (ρu 0
n+1 Un (1))
Z x (20)
−1
+ (1 − pn+1 ) In+1 (ξ)dξ; x > 0,
In+1 (ρdn+1 Un0 (1))

∗,u ∗,d
∗ Xn+1 (Xn∗ ) − Xn+1 (Xn∗ )
πn+1 = u d
,
Rn+1 − Rn+1
and
n+1
X
∗
Xn+1 = Xn∗ + ∗
πn+1 (Rn+1 − 1) = X0 + πi∗ (Ri − 1) ,
i=1

6
If both conditions in part (i) and (ii) do not hold, then the functional equation (9) may not have a
1−1/θ
solution, or the solution may not be unique. For the case of initial power utility U0 (x) = x1−1/θ , θ > 0,
Example 6 and Corollary 9 show that both condition fail at tn if and only if θ = − loga b > 0, in which
case the solution exists but is not unique. This case is pathological, but to solve it remains a technically
interesting question.

17
where,
   
∗,u qn+1 0 ∗,d 1 − qn+1 0
Xn+1 (x) = In+1 U (x) and Xn+1 (x) = In+1 U (x) ; x > 0.
pn+1 n 1 − pn+1 n

In summary, starting with an initial datum U0 , we have constructed for (the end of)
each trading period, say (tn , tn+1 ] , n = 1, 2, ..., a performance criterion Un+1 at tn+1 that
is indeed Fn −measurable. This measurability is inherited by the same measurability of the
inverse marginal In+1 that enters in the lower part of the integration in (20). Moreover, as
∗
expected, the optimal wealth Xn+1 is Fn+1 −measurable, given that the pricing kernel ρn+1
∗
is Fn+1 −measurable. The optimal portfolio πn+1 is Fn −measurable, chosen at the beginning
of the period (tn , tn+1 ] .

8 Conclusion
We have introduced a new approach to optimal portfolio management that allows for dy-
namic model specification and adaptation, flexible investment horizons, and stochastic risk
preferences. These risk preferences are modeled as a discrete-time predictable process, which
is a rather natural and intuitive property of performance measurement criteria in practical
applications. The frequency of performance evaluation is allowed to be different from or the
same as the one at which the portfolio is rebalanced.
Specifically, at the beginning of each evaluation period, the investor assesses the market
parameters only for this period (during which trading may take place many times, in both
discrete or continuous fashion). Then, she solves an inverse single-period investment model
which yields the utility at the end of the period, given the one at the beginning. The
martingality and supermartingality requirements of the forward performance process ensure
that this construction, ”period-by-period forward in time” and adapted to the new market
information, yields time-consistent policies.
We have implemented this new approach in a binomial model with random parameters,
including both the probabilities and the levels of the stock returns. Such a setting is consid-
erably flexible, as it accommodates short-term predictability of the asset returns, sequential
learning and other dynamically evolving factors affecting optimal investments. We have
then discussed in detail how the construction of predictable forward performance processes
essentially reduces to a single-period inverse investment problem. We have, in turn, shown
that the latter is equivalent to solving a functional equation involving the inverse marginal
functions at the beginning and the end of trading period, and have established conditions
for the existence and uniqueness of solutions in the class of inverse marginal functions.
We have finally provided an explicit algorithm that yields the forward performance pro-
cess, the optimal portfolio and the associated optimal wealth processes.
There are a number of possible future research directions. Firstly, one may depart from
the binomial model to study general discrete-time models allowing for trading to be discrete
or continuous. Such models are inherently incomplete and additional difficulties are expected
to arise with regards to the derivation of the functional equation for the inverse marginals
as well as the existence and uniqueness of its solutions among suitable classes of functions.

18
 A second direction is to enrich the predictable framework by incorporating model ambi-
guity. This will allow for the specification of all possible market models only one evaluation
period ahead, thus offering flexibility to narrow down the most realistic models period-by-
period as the market evolves.
From the theoretical point of view, an interesting question is to investigate whether
predictable forward performance processes converge to their continuous-time counterparts.
While this is naturally and intuitively expected, conditions on the appropriate convergence
scaling need to be imposed, which might be quite challenging due to the ill-posedness of
the problem. Such results may also shed light to deeper questions on the construction
of continuous-time forward performance criteria related to the appropriate choice of their
volatility, finite-dimensional approximations, Markovian or path-dependent cases, among
others.

A Proof of Theorem 4
We start with the following auxiliary result, showing that (7) is equivalent to

U0 (I0 (y)) = EP (U1 (I1 (ρ1 y))) ; y > 0. (21)

Lemma 11. Suppose that U0 , U1 ∈ U and let I0 and I1 be respectively their inverse marginals.
Then, (7) holds if and only if (21) holds.

Proof. We first show that (7) implies (21). Indeed, standard results yield that (7) implies
h 
i
U0 (x) = EP U1 I1 ρ1 U00 (x) ; x > 0,

and (21) is obtained by the change of variable y = U00 (x).

Next, we show that (21) yields (7). Define the value function Ũ by

Ũ (x) = sup EP [U1 (X)] ; x > 0.

A(x)

We claim that Ũ ≡ U0 . Let I˜ be the inverse marginal of Ũ . By (i), one must then have
h
i
˜


Ũ I(y) = EP U1 I1 (ρ1 y) ; y > 0,

˜



and it follows that Ũ I(y) = U0 I0 (y) , for y > 0.
Differentiating with respect to y yields I˜0 ≡ I00 . Therefore I(y)
˜ = I0 (y) + C, y > 0,
for some constant C. Taking the limit as y → ∞ and using the Inada condition I(∞) ˜ =
I0 (∞) = 0 yields C = 0. Therefore, we obtain I˜ ≡ I0 , which implies Ũ (x) = U0 (x), for all
0 0

x > 0. Finally, we obtain

h
i h
i
Ũ (x) = EP U1 I1 (ρŨ 0 (x)) = EP U1 I1 (ρU00 (x)) = U0 (x); x > 0.

Proof of Theorem 4. (i): From (14) it follows that

Z x Z x
−1
U1 (x) := U0 (1) + p I1 (ξ)dξ + (1 − p) I1−1 (ξ)dξ; x > 0,
xu (1) xd (1)

19
where xu (·) and xd (·) are given by
xi (c) = I1 ρi U00 (c) ;


c > 0, i = u, d. (22)
Thus,
U10 (x) = p I1−1 (x) + (1 − p) I1−1 (x) = I1−1 (x); x > 0.
It then follows that I1 is the inverse marginal of U1 and that U1 is a utility function.

(ii): Define the function F by

Z x Z x
−1
F (x, c) := U0 (c) + p I1 (ξ)dξ + (1 − p) I1−1 (ξ)dξ; (x, c) ∈ R+ × R+ , (23)
xu (c) xd (c)

with xu (c) and xd (c) as in (22). We claim that

∂F
(x, c) = 0; x, c > 0.
∂c
Differentiating (23) with respect to c and then using that I1−1 xi (c) = ρi U00 (c), for c > 0,

we have
∂F
(x, c) = U00 (c) − px0u (c)G xu (c) − (1 − p)x0d (c)G xd (c)



∂c
= U00 (c) − px0u (c)ρu U00 (c) − (1 − p)x0d (c)ρd U00 (c)
 
= U00 (c) 1 − pρu x0u (c) − (1 − p)ρd x0d (c) = 0.
To obtain the last equation, note that (9) is equivalent to
I0 (y) = pρu I1 (y ρu ) + (1 − p)ρd I1 (y ρd ); y > 0.
Therefore, substituting y = U0 (c) and differentiating with respect to c yield
 
d  0  0  d u

0

d

0
1= I U (c) = pρ I1 ρu U0 (c) + (1 − p)ρ I1 ρd U0 (c)
dc 0 0 dc
   
= p(ρu )2 I10 ρu U00 (c) U000 (c) + p(ρd )2 I10 ρd U00 (c) U000 (c)
= pρu x0u (c) + (1 − p)ρd x0d (c).

Note that, by definition, U1 (x) = F (x, 1). Since we have showed that ∂F ∂c
≡ 0, we must
+
have U1 (x) = F (x, c), for all x > 0 and c > 0. In other words, for all x, c ∈ R , U1 satisfies
Z x Z x
−1
U1 (x) = U0 (c) + p I1 (ξ)dξ + (1 − p) I1−1 (ξ)dξ.
xu (c) xd (c)

On the other hand, as it was shown in (i), U10 ≡ I1−1 . Therefore, for all x > 0 and c > 0,

 

U1 (x) = U0 (c) + p U1 (x) − U1 xu (c) + (1 − p) U1 (x) − U1 xd (c) ,
which, in turn, yields that
h
i
U0 (c) = pU1 xu (c) + (1 − p)U1 xd (c) = EP U1 I1 (ρ1 U00 (c)) ;



c > 0.
This is equivalent to (21). Hence, (ii) follows from Lemma 11.
(iii): This part follows easily from existing results in the classical expected utility problems,
if we view (7) as a terminal expected utility problem with U1 now given and U0 being its
value function.

20
B Proof of Theorem 8
We only show part (i) and the corresponding statements in parts (iii) and (iv), since (ii)
follows from similar arguments.
(i) Direct substitution shows that if the infinite series in (17) converges, then I1 satisfies
(9). Thus, to show (i), it only remains to show that the series converges. Note that (17) can
be written, for y > 0, as
∞
b loga b X
I1 (y) = y (−1)m Ψ0 (am y), (24)
1+b m=0

which, by the Leibniz test for alternating series, converges if limm→∞ Ψ0 (am y) = 0 mono-
tonically. The fact that limm→∞ Ψ0 (am y) = 0 follows directly from either of the conditions
in (i) on a and Ψ0 . To show that the convergence is monotonic, note that by (16)

Ψ0 (am+1 y) − Ψ0 (am y) = b−m−1 y − loga b Φ0 (am y); y > 0, m = 0, 1, . . . . (25)

On the other hand, since Φ0 is increasing and limy→∞ Φ0 (y) = limy→∞ I0 (a c y)−b I0 (c y) = 0
by the Inada condition, we must have Φ0 (y) < 0, for y > 0. Thus, by (25), we have that
Ψ0 (am y) > Ψ0 (am+1 y) and limm→∞ Ψ0 (am y) = 0 monotonically.
(iii) First, we prove that I1 is strictly decreasing. Indeed, (24) and (25) yield
∞ ∞
b loga b X   1 X −2m
I1 (y) = y Ψ0 (a2m y) − Ψ0 (a2m+1 y) = − b Φ0 (a2m y).
1+b m=0
1 + b m=0

It then follows that, for y < y 0 ,

∞
1 X −2m  
I1 (y 0 ) − I1 (y) = b Φ0 (a2m y) − Φ0 (a2m y 0 ) < 0,
1 + b m=0

where the inequality holds because Φ0 is strictly increasing.

Using equation (9), that a, b, c > 0 and limy→∞ I0 (y) = 0, and the monotonicity of I1 ,
we deduce that limy→∞ I1 (y) = 0, and, hence, that I1 (y) > 0, y > 0. Similarly, the fact that
limy→0+ I0 (y) = ∞ yields limy→0+ I1 (y) = ∞. Thus, we have shown that I1 ∈ I.
Finally, conditions in Lemma 7 follow from Ψ0 (y) → 0, as either y → 0+ or y → ∞, and
from
I1 (y) b+1
0 < y loga b I1 (y) = Ψ0 (y) < Ψ0 (y); y > 0,
I0 (c y) b
where we used (9) and that I1 (y) > 0 to obtain

I1 (y) (1 + b)I1 (y) 1+b

= < .
I0 (c y) I1 (a y) + b I1 (y) b

(iv) Repeating the last part of the argument in part (iii) for any solution I˜ > 0 yields
that I˜ satisfies the same uniqueness condition for (9) as I1 . The result then follows directly
from Lemma 7.

21
C Proof of Corollary 9
Assertion (ii) follows from (i) and Theorem 4. Also, one can easily check that I1 given by
(19) is an inverse marginal satisfying (9).
It only remains to show the uniqueness of solutions that are inverse marginals. To this
end, it suffices to check that the condition of Theorem 8 holds for all possible values of the
parameters. Setting G(y) = y −θ , y > 0, in (16) yields

Φ0 (y) = (a−θ − b)c−θ y −θ and Ψ0 (y) = y −(θ+loga b) .

Since θ 6= − loga b and a 6= 1, we have the following dichotomy:

a) Either θ < − loga b and a < 1 or θ > − loga b and a > 1. Then, one can show that
conditions (i) of Theorem 8 hold.

b) Either θ < − loga b and a > 1 or θ > − loga b and a < 1. Then, one can show that
conditions (ii) of Theorem 8 hold.

References
O.E. Barndorff-Nielsen and N. Shephard. Econometric analysis of realized volatility and its
use in estimating stochastic volatility models. Journal of the Royal Statistical Society:
Series B (Statistical Methodology), 64(2):253–280, 2002.

F. Black. Individual investment and consumption under uncertainty. In D. L. Luskin, editor,

Portfolio Insurance: A Guide to Dynamic Hedging, pages 207–225. John Wiley and Sons,
New York, 1988. First version: November 1, 1968, Financial Note No. 6B, Investment and
consumption through time.

S. Dreyfus. Richard Bellman on the birth of dynamic programming. Operations Research,

50(1):48–51, 2002.

N. El Karoui and M. Mrad. An exact connection between two solvable SDEs and a nonlinear
utility stochastic PDE. SIAM Journal on Financial Mathematics, 4(1):697–736, 2013.

E. F. Fama and K. R. French. Business conditions and expected returns on stocks and bonds.
Journal of financial economics, 25(1):23–49, 1989.

T. Geng and T. Zariphopoulou. Temporal and spatial turnpike-type results under forward
time-monotone performance criteria. submitted for publication, 2017.

S. Kallblad, J. Oblój, and T. Zariphopoulou. Time–consistent investment under model

uncertainty: the robust forward criteria. to appear in Finance and Stochastics, 2017.

M. Kuczma, B. Choczewski, and R. Ger. Iterative Functional Equations. Encyclopedia of

Mathematics and its Applications. Cambridge University Press, 1990.

M. Musiela and T. Zariphopoulou. Backward and forward utilities and the associated pricing
systems: The case study of the binomial model. In R. Carmona, editor, Indifference
Pricing. Princeton University Press, 2003.

22
M. Musiela and T. Zariphopoulou. Investments and forward utilities. Technical report, 2006.
URL http://www.ma.utexas.edu/users/zariphop/pdfs/TZ-TechnicalReport-4.pdf.

M. Musiela and T. Zariphopoulou. Portfolio choice under dynamic investment performance

criteria. Quantitative Finance, 9(2):161–170, 2009.

M. Musiela and T. Zariphopoulou. Stochastic partial differential equations in portfolio choice.

In C. Chiarella and A. Novikov, editors, Contemporary Quantitative Finance, pages 195–
216. Springer-Verlag Berlin Heidelberg, 2010a.

M. Musiela and T. Zariphopoulou. Portfolio choice under space-time monotone performance

criteria. SIAM Journal on Financial Mathematics, 1(1):326–365, 2010b.

M. Musiela and T. Zariphopoulou. Initial investment choice and optimal future allocations
under time-monotone performance criteria. International Journal of Theoretical and Ap-
plied Finance, 14(01):61–81, 2011.

M. Musiela, E. Sokolova, and T. Zariphopoulou. Forward exponential indifference valuation

in an incomplete binomial model. In J. Kallsen and A. Papapantoleon, editors, Advanced
Modelling in Mathematical Finance, In Honour of Ernst Eberlein, pages 277–302. Springer,
2016.

S. Nadtochiy and M. Tehranchi. Optimal investment for all time horizons and Martin
boundary of space-time diffusions. Mathematical Finance, 2015.

S. Nadtochiy and T. Zariphopoulou. A class of homothetic forward investment performance

processes with non-zero volatility. In Y. Kabanov, M. Rutkowski, and T. Zariphopoulou,
editors, Inspired by Finance: The Musiela Festschrift, pages 475–504. Springer, 2014.

M. Shkolnikov, R. Sircar, and T. Zariphopoulou. Asymptotic analysis of forward performance

processes in incomplete markets and their ill-posed HJB equations. SIAM Journal on
Financial Mathematics, 7(1):588–618, 2016.

23