Dynamic Asset Allocation Under Inflation

Dynamic Asset Allocation under Inflation
Author(s): Michael J. Brennan and Yihong Xia

Source: The Journal of Finance , Jun., 2002, Vol. 57, No. 3 (Jun., 2002), pp. 1201-1238
Published by: Wiley for the American Finance Association
Stable URL: https://www.jstor.org/stable/2697777
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
and Wiley are collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Finance
This content downloaded from

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
All use subject to https://about.jstor.org/terms
THE JOURNAL OF FINANCE * VOL. LVII, NO. 3 * JUNE 2002
Dynamic Asset Allocation under Inflation
MICHAEL J. BRENNAN and YIHONG XIA*
ABSTRACT
We develop a simple framework for analyzing a finite-horizon investor's asset al-

location problem under inflation when only nominal assets are available. The in-
vestor's optimal investment strategy and indirect utility are given in simple closed
form. Hedge demands depend on the investor's horizon and risk aversion and on
the maturities of the bonds included in the portfolio. When short positions are
precluded, the optimal strategy consists of investments in cash, equity, and a sin-
gle nominal bond with optimally chosen maturity. Both the optimal stock-bond
mix and the optimal bond maturity depend on the investor's horizon and risk
aversion.
AN INVESTOR WHO HAS A LONG-TERM but finite horizon and can invest only in
nominal bonds or stocks faces a basic problem. Is it better to purchase a zero
coupon bond corresponding to the horizon and bear the inflation risk, to
follow a policy of rolling over short-term bonds, or to adopt some quite dif-
ferent strategy? Despite the simplicity of this issue, there is still no well-
accepted framework for analyzing it because nominal long-term bonds have
two important characteristics that cannot be represented adequately within
the classical static, single-period, framework introduced by Markowitz (1959).
First, the prices of bonds decline as interest rates rise so that, as Merton
(1973) originally pointed out, long-term bonds can provide a hedge against
adverse shifts in the investor's future investment opportunity set. Second,
and somewhat weakening the hedging role of long-term bonds, is the sensi-
tivity of their prices to changes in expectations about future inflation. There-
fore, a satisfactory counterpart to classical static portfolio theory that would
enable us to address the problem faced by the hypothetical long-term inves-
tor must satisfy two criteria. It must yield simple closed-form expressions
for optimal portfolios for investors with different horizons and attitudes to-
wards risk, and it must deal realistically with both the price and return
characteristics of long-term bonds, as well as with inflation. In this paper,
we develop a simple model that satisfies these criteria. The investor's opti-
* Brennan is the Goldyne and Irwin Hearsh Professor of Banking and Finance at th
versity of California, Los Angeles and Xia is an assistant professor of finance at the University
of Pennsylvania. The authors thank John Campbell, John Cochrane, Domenico Cuoco, Richard
Green (the editor), Jun Liu, Pedro Santa-Clara, Jessica Wachter, two anonymous referees, and
participants at the Brown Bag Micro Finance Lunch Seminar at the Wharton School for com-
ments and suggestions. Xia acknowledges financial support from the Rodney L. White Center
for Financial Research. All remaining errors are ours.
1201

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
1202 The Journal of Finance
mal portfolio is shown to be a sum of two components: first, the mean-

variance tangency portfolio and, second, a portfolio that mimics as closely as
possible a hypothetical indexed bond with maturity equal to the investment
horizon.
The analysis of optimal portfolio strategies for long-lived investors starts
with Latane and Tuttle (1967), Mossin (1968), Hakansson (1970), Merton
(1969), and Samuelson (1969). These authors were primarily concerned with
analyzing the investor's optimal allocation between stock and cash when
returns are i.i.d. Merton (1971) was the first to consider the effect of a sto-
chastic investment opportunity set, as well as to demonstrate that this cre-
ates a set of "hedge" demands in addition to the standard myopic demand.
However, further work on optimal strategies under stochastic investment
opportunities languished until revived by empirical work demonstrating ap-
parent asset return predictability in the 1990's. Brennan, Schwartz, and
Lagnado (1997) analyzed numerically the portfolio problem of a long-lived
investor who can invest in bonds, stock, or cash, when there is stochastic
variation in the interest rate, and the equity premium is predictable by the
interest rate and the dividend yield. Kim and Omberg (1996) and Wachter
(1999) have analyzed the optimal strategy of an investor when the interest
rate is constant but the equity premium follows an Ornstein-Uhlenbeck pro-
cess.1 Sorensen (1999), Brennan and Xia (2000), and Omberg (1999) also
compute optimal dynamic strategies when the interest rate follows a Vasicek
(1977) process2 and risk premia are constant. Liu (1999), who provides a
general treatment of models in which the investment opportunity set has an
affine characterization, studies the optimal cash-bond allocation in a model
with an affine term structure. Barberis (2000), Kandel and Stambaugh (1996),
and Xia (2000) have considered the implications of uncertain predictability
of asset returns.
None of the above papers allows for stochastic inflation, takes account of
borrowing and short sales constraints, or provides a rationale for the bond
maturity choice. In this paper, we analyze the portfolio problem of a finite-
lived investor3 who can invest in stock or nominal bonds, when the interest
rate and the expected rate of inflation follow correlated Ornstein-Uhlenbeck
processes and the risk premia are constant.4
The composition of the optimal bond portfolio is not determinate within
the model when there are no constraints; only the optimal loadings on in-
novations in the estimated real interest rate, r, and expected rate of infla-
tion, 1-, can be determined. Calibration of the model to data on U.S. interest
1 Wachter (1999) assumes that the innovation in the equity premium is perfectly correla
with the stock return.
2 Brennan and Xia (1999) allow for a generalized Vasicek process.
3 For completeness, we analyze both a utility of lifetime consumption model and a utility of
final wealth model. These are essentially equivalent.
' While the assumption of constant risk premia may seem restrictive, Bossaerts and Hillion
(1999) find no evidence of out-of-sample excess return predictability in 14 countries using as
potential predictors lagged excess returns, January dummies, bond and bill yields, dividend
yields, and so forth.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Dynamic Asset Allocation under Inflation 1203
rates, stock returns, and inflation yields mixed results. When the calibra-
tion is made to monthly data on bond yields and inflation, strong mean
reversion is found for the shadow real interest rate. This makes the optimal
portfolio holdings relatively insensitive to the investment horizon beyond
five years, and the gains from following a fully dynamic strategy relatively
small, except for high levels of risk aversion. When the calibration is made
to annual interest rates and inflation over a long period, much less mean
reversion is found. With the lower mean reversion parameter, substantial
horizon effects appear in the optimal portfolio strategies, and the gains from
following the optimal dynamic strategy become large. In both cases, as risk
aversion increases, the investor holds less stock and the return on his bond
portfolio tends to become less sensitive to innovations in the expected rate of
inflation. In the limit, as risk aversion becomes infinite, the stock allocation
goes to zero and the investor's dynamic strategy in bonds mimics as closely
as possible the returns on an inflation indexed bond with maturity equal to
the remaining investment horizon.
The foregoing results rely on the assumption that the investor is able to
take unlimited short positions. When the investor is constrained to take
only long positions, the optimal portfolio can be achieved with positions in
only a single bond of the optimally chosen maturity, cash, and stock. The
optimal bond maturity depends on both the investor's horizon and risk aver-
sion. As in the unconstrained case, horizon effects are pronounced only for
the annual data calibration. For both sets of calibrated parameters, the ratio
of bonds to stock increases as the horizon increases, which is at odds with
the popular view that long-horizon investors should hold more stock. The
bond-stock ratio also increases as risk aversion increases, which is consis-
tent with the portfolio recommendations of popular financial advisors that
have puzzled Canner, Mankiw, and Weil (1997). The maturity of the optimal
bond decreases as risk aversion increases.
There are at least two possible explanations for the differences between
the monthly and the annual calibrations. The first is that the expected rate
of inflation series estimated using monthly data on bond yields and price
index changes is too smooth because non-price signals about the expected
rate of inflation are ignored; this would cause the high frequency changes in
the true market assessment of the expected rate of inflation, which are re-
flected in bond yields, to be impounded in our estimates of the real interest
rate. The second possibility is that, whereas we use a one-factor model to
describe the dynamics of the real interest rate, these may be better repre-
sented by a two-factor model in which one factor has high frequency5 and
the other low frequency; in this case, the monthly calibration may be picking
up the high frequency factor while the annual calibration may place more
emphasis on the low frequency component.
In the paper that is closest to this one, Campbell and Viceira (CV; 1999)
develop an approximately optimal portfolio strategy for an infinitely lived
investor with recursive utility, in a discrete time setting in which the real
5 Perhaps because monetary policy has a short run impact on the real interest rate.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
interest rate and the expected rate of inflation follow similar stochastic pro-
cesses. The major differences between their paper and this one are that:
first, unlike CV, we are interested in the problem faced by a finite-horizon
investor and are therefore able to give explicit consideration to the effect of
the horizon on both the optimal bond-stock mix and the maturity of the
optimal bond portfolio. Second, we develop closed form expressions for the
investor's optimal policy and indirect utility function, whereas the CV for-
mulation relies on linear approximation and numerical analysis.6 Our closed-
form solutions enable us to analyze the welfare loss of the myopic policy and
to characterize the dependence of the portfolio hedge demand on the char-
acteristics of the investment opportunity set, such as the horizon and inten-
sity of mean reversion of the real interest rate. Finally, in considering the
problem in which the investor is subject to short sales constraints, CV take
the maturity of the investable bond as given, whereas we allow for the op-
timal choice of bond maturity and are therefore able to consider the effect of
risk aversion and horizon on the optimal maturity. In general, the introduc-
tion of constraints on the size of positions makes the choice of the maturity
of the bonds to be included in the portfolio a critical element of the portfolio
decision.
We present the basic model of stochastic real interest rates, inflation, and
expected stock returns in Section I. The optimal portfolio problem is derived
in Section II. We calibrate the model to the U.S. postwar nominal interest
rate, inflation, and stock return data in Section III. Some representative
calculations and discussions are offered in Section IV. We summarize the
results and conclude the paper in Section V.
I. Investment Opportunities
We assume that the investor can invest in a nominal instantaneous risk

free asset, a stock, and in nominal bonds with different maturities. The real
returns on the nominal bonds are risky, because both the rate of inflation
and the (shadow) real interest rate are stochastic.7
The (commodity) price level, 1I, follows a diffusion process:
dH
g=1dt + or-dz r,, (1)
6 The linear approximation of CV does not bec

except for the special case of unit intertemporal substitution, which corresponds to log utility of
consumption, and, in this case, there is no closed-form solution for the indirect utility function
(Giovannini and Weil (1989)).
7 By the shadow real interest rate, we mean the instantaneous real return that would prevail
for an asset whose instantaneous real return was nonstochastic given the pricing kernel, if
such an asset were to exist. In our model, there is no instantaneous real riskless asset. How-
ever, as we shall show below, when the rate of inflation is spanned by the nominal returns on
assets, as will be the case if the expected rate of inflation is unobservable and must be inferred
from the price level realization, the model allows for the construction of a portfolio of nominal
assets whose instantaneous real return is riskless.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
so that the (realized) rate of inflation is locally stochastic. The instanta-

neous expected rate of inflation, or proportional drift of the price level, r,
follows an Ornstein-Uhlenbeck process:
d7r = a(-fr - r)dt + cr-dz,. (2)
The investment opportunities depend on the (real) pricing kernel of the

economy, Mt, which determines the expected returns on all securities:
dM
-rdt + obsdzs + b,.dz,. + 0,bdz,, + Obudz,
M (3)
= -rdt + b'dz + oudz
where Sb = [OS,Or, ']' and dz = [dz5,dZr,dz']. The expression, /i

(i = S, r, 7T, u), representing the constant loadings on the stochastic innova-
tions in the economy, determines the associated prices of risk, As,
and AU, which are constant because the Obs are constant. The vari
represents the increment to the Brownian motion that drives the stock re-
turn in equation (5) below, and dzu, which is defined in equation (7
is proportional to the component of the inflation rate, dEH/H, that is orthog-
onal to dz and therefore to the nominal returns on all assets: It drives the
component of inflation that is not spanned by the asset returns and there-
fore cannot be hedged. We denote the drift of the pricing kernel by -r be-
cause it is well known that, in an economy with a fully indexed riskless
asset, the instantaneous (real) riskless interest rate is equal to (the negative
of) the drift of the pricing kernel.
Following Vasicek (1977), we assume that r follows the Ornstein-Uhlenbeck
process:
dr= K(r - r) dt + o dZr (4)
If an instantaneously riskless real asset existed, then its instantaneous real

rate of return would be r. However, we assume that, in this economy, no
instantaneously riskless real asset exists8 and that the investor is able to
invest only in a single stock index and in nominal bonds with different ma-
turities. The nominal stock price is assumed to follow a Geometric Brownian
motion,
dS
S =(Rf + osAs)dt + sdzs, (5)
where As is the constant unit risk premium associated with the innovation,
dzs, and Rf is the nominal interest rate.
8 Even countries such as Canada, the United States, and the United Kingdom, which
inflation indexed bonds, have them for only a few (long) maturities.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Since the Brownian increments in equat

nal, the pricing kernel relative can be rewritten as the product of two inde-
pendent stochastic integrals:
Ms/Mt exp f(-r(u) - -' Plb) du ?f>,dz
x exp f (-2 2) du +? oudzuj (6)
-1(t,S);2(t,s),
where p is the correlation matrix of dzs, dZr, and dz, with ro

[PSr,I,Pri7], and [pS,,prT,1], and V1(t,s) and ;2(t,s) are orthogonal.
In general, the realized rate of inflation given by equation (1) will not be
perfectly correlated with the change in the expected rate of inflation given
by equation (2). However, it can be shown that if the expected rate of infla-
tion is not observable but must be inferred from observation of the price
level itself (and not from any other information), then the change in (the
investor's assessment of) the expected rate of inflation will be perfectly cor-
related with the realized rate of inflation: For reasons that will be apparent
below, we call this the "complete markets case." In the general case, the
innovation in the rate of inflation can be written as a linear function of the
innovations dzs, dZr, and dz, and the projection residual, (udzu:
dFI
- = Tdt + odzr = 'Tdt + (sdzs + (rdZr ? r ,dzr + (udzu
II (7)
-'mdt + ('dz + (udzu
If S = 0r = ,u =0? drl/rl is perfectly correlated with d7Tfw and the market

is complete.
Equation (7) implies that the price level relative, Hslt (s > t), can also be
written as the product of two independent stochastic integrals:
s /lt =exp {4T(((u) - - P)dp d ?f + 'dz1
x exp jf(-2 2) du +? udzu (8)
-q1(t,S)mq2(t,S),
where q1l(t,s) and i12(t,s) are orthogonal so that o-g = ('pf + (2. q1l(t,s)
component of the price level change that can be hedged by investing in the
available securities, while i%2(t,s) is the unhedgeable component.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
The definition of the pricing kernel implies that P(t, T), the nominal price
at time t of a bond which matures at time T with a nominal payoff of $1, is
given by
M[T/t 1 _ 1(t, T) 1 2(t, T)1

P(t,T) = Et [ 1/11ft Et I (t, T) Et L
It is shown in Appendix A that P(t, T) is an exponential affine function of rt,

the (shadow) real interest rate, and 1Tt, the investor's assessment of the
current expected rate of inflation:
P(t,T) = exp{A(t,T) - B(t,T)rt - C(t,T) 7}T (10)
where A (t, T), B (t, T), and C (t, T) are time-dependent constants, expressions
for which are given in Appendix A.
Using Ito's Lemma and the expressions for A (t, T), B (t, T), and C (t, T), the
stochastic process for the bond price can be written as
dP
= [r + 7T- Bor Ar - Co, A, - s As- (rAr- A,, - (uAj]dt
(11)
- BordZr -Co, dz ,
where the As are market prices of risk associated with stock return, the
innovations of real interest rate, and the innovations of expected and un-
expected inflation. The expressions for As are given in equations (A13)-(A16).
There is a simple linear relation between the market price of risk vector,
A, and the factor loadings of the real pricing kernel in equation (3). Let A be
the vector of nominal risk premiums for the stock and two nominal bonds
with maturities T1 and T2; then A [o-sAs, -B(t,Ti)Ar - C (t, T1) Azr
-B(t,T2)Ar - C(t,T2)AJ]'. Define the factor loadings matrix of the three
securities as o- where the first row is (oSs,0,0), and the second and third rows
are (0, -B (t, Tj) o-r, - C (t, Tj) o-,) (j = 1, 2). The real risk premium on the thr
securities is equal to the covariance between the real return of the (nominal)
security and the real pricing kernel, so the unit market price of the risk
vector is related to the risk premium vector by A = o-A. Equivalently, the
factor loadings of the real pricing kernel are related to the risk premium
vector by = p - 1A.
The instantaneous nominal risk-free interest rate, Rf, is obtained by tak-
ing the limit of the return on the nominal bond in equation (11) by letting
(t - T) -X 0:
Rf =r + 7 - (sAs-_ rAr- rA,, - uAu, (12)

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
and we call -esAs - erAr - A - Au the risk premium for the nominal
instantaneous risk-free asset. The Fisher equation does not hold in this econ-
omy unless all the market prices of risk, As, Ar, A, and Au, are zer
that the nominal risk-free asset has a zero risk premium. Using the defini-
tion of the nominal risk-free interest rate in equation (12), the nominal re-
turn on the nominal bond in equation (11) can also be simplified as
dP
- = [Rf - BorAr -Co,r A, ] dt - Bor dz,. -Co-, dz1, (13)
which shows that the nominal risk premium on a bond depends only on its
exposure to innovations in the real interest rate and expected rate of inflation.
II. Optimal Portfolio Choice
Turning now to the issue of optimal portfolio strategies for long-lived in-
vestors, we shall consider two classical cases. In the first, the investor is
assumed to be concerned with maximizing the expected utility of wealth on
some fixed horizon date, T. This problem has the merits both of simplicity,
for it admits a closed-form solution, and of clarifying the role of the horizon,
for with this simple objective, there is no ambiguity about the duration of
the consumption stream that is being financed. The problem can be thought
of as corresponding to that faced by an individual who has set aside pre-
determined savings for retirement and wishes to maximize the expected util-
ity of wealth on his retirement date;9 we are simplifying the full problem by
ignoring the optimal investment and consumption plan during retirement.
The second case that we consider is that of an investor who is concerned
with maximizing the expected value of a time-additive utility function de-
fined over lifetime consumption. This problem, which is only slightly more
complicated, corresponds to the consumption-portfolio choice problem of an
individual who is retired and faces a known date of death with no bequest
motives.
There are four potential sources of uncertainty in the model economy that
we have described: real interest rate risk represented by the innovations in
r as shown in equation (4); inflation risk due to unanticipated changes in
the price level as shown in equation (7); unanticipated changes in the ex-
pected rate of inflation as shown in equation (2); and finally, the unantici-
pated stock return shown in equation (5).
If (= 0, the change in the price level is an exact linear function of dzs,
dzr, and dzT, and one dimension of risk faced by the investor is eliminat-
ed.10 Then the market is complete if there are at least four securities whose
9 We are simplifying by ignoring labor income and the consumption-investment decision.

10 Formally, the martingale multiplicity of the investor's information structure is
two (Duffie and Huang (1985)). This condition will be satisfied if, for example, the only infor-
mation that is available about expected inflation is derived from the historical inflation series.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
instantaneous variance-covariance matrix has rank three.1" Since the

variance-covariance matrix of real returns on cash, stock, and any two fi-
nite maturity bonds with different maturities has rank three, the market is
complete whenever (u = 0. In the complete market setting, the investo
optimal portfolio problem may be solved using the martingale pricing ap-
proach of Cox and Huang (1989).
If the market is incomplete, (fu 0 0), u cannot be derived from
servable security prices, but any specific value of u determines a
pricing kernel. If a wealth (consumption) plan is feasible-that is, can be
financed by a trading strategy in the available (nominal) securities-then
the static budget constraint can be represented by the expectation of the
product of payoffs and the pricing kernel defined by any given value
Therefore, the original dynamic portfolio choice problem can be mapped into
the static variational problem of Cox and Huang (1989) using an arbitrary
value of U.2 We first solve the optimal terminal wealth and consump
allocation under the static budget constraint for a given u, and then v
that the optimal terminal wealth and consumption are financed by feasible
trading strategies.
Although it is no longer possible to construct a portfolio whose return
is riskless in real terms when the market is incomplete, the unhedge-
able component of inflation is orthogonal to all available asset returns.
Moreover, the homogeneity of both the utility function and the inflation
process implies that the optimal proportional allocation is independent of
real wealth and the price level. Therefore, the market incompleteness caused
by nonhedgeable inflation has no effect on the optimal proportional port-
folio allocation.
LEMMA 1: For an investor with isoelastic utility, the optimal proportional

portfolio allocation is independent of real wealth, wt Wt/rlt, and the price
level, rlt I
The Lemma allows us to characterize the investor's problem in the in-

complete market setting via the extended Cox-Huang method developed by
Pages (1987), He and Pearson (1991), and Karatzas et al. (1991). In order
to represent the market incompleteness caused by nonhedgeable inflation,
fix a probability space (0,17]), where 0 = 01 X 02. The vector of stan-
dard Brownian motions dz [dzs,dZr,dzj]' is defined on 01, while (udzu,
the component of realized inflation that is orthogonal to all security re-
turns and therefore nonhedgeable, is defined on 02. The standard filtra-
tion F1 is generated by {dzs: 0 < s ? t} and F2 by dzu, and F F1 x F2.
It follows from the Lemma that an optimal trading strategy is a process x,
which specifies the proportion of wealth invested in each security and is
adapted to F1.
See Cox and Huang (1989) for a formal analysis.

12 See Pages (1987) and Karatzas et al. (1991) for a rigorous proof. In particular
(1987) Proposition 2.15.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
A. State-Contingent Wealth and Consumption
Consider first the problem of an investor with an isoelastic utility function

who is concerned with maximizing the expected utility of real wealth at time
T. The investor's optimal portfolio choice problem can be mapped into the
following static variational problem:
max ( IWT/Fl) ) (14)

W(T):t?T?T 1 - J
MT
s.t. (1) Et M W/T tItw,(15)
(2) WT is financed by a feasible trading

strategy with initial investment W0,
where equation (15) is the static budget constraint. iThe following theorem
presents the solution of this static variational problem.
THEOREM 1 (OPTIMAL TERMINAL WEALTH ALLOCATION): For an investor con-

cerned with maximizing -the expected value of an isoelastic utiClity function
defined over wealth at time T in the incomplete markets setting in which
eu 0 0:
(i) The optimal terminal real wealth allocation, WT*- W(MT, HT)/1T, is
WT= wt l-'(t, T) (I; (t, T)) -<1/yF1(t, T<) -lF2(t,T) 1, (16)
where
F1(t, T) = Et [1(t, T)11-(ly)] = exp[(1-y)/Y1y[B(t,T)rt?ao(t,T)j (17)
F2 (t, T ) = Et 1;2 (t, T )/92 (t, T )] =exp(>- u )Tt (18)
a(t,T) - (T-t) [(T - t) - B (t, T)]

(19)
+ 4-yK3 r [2K(T - t) - 3 + 4eK(t-T) - e2K(t-T)].
(ii) The indirect utility function, J(Wt, rt, flt, t), is separable in rea
wt, and can be written as

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
J(W,r,7,t) E4 1 " = ((1t)),(r,t,T), (20)

where w* is the real wealth at period T under the optimal policy.
(iii) ql(r,t,T) represents the contribution to the investor's expected utilit
of the remaining investment opportunities up to the horizon:
qf1(r,t,T) = exp (-y)[B(t, T)rt+aL(t,T)] (21)
where B (t, T) K-1(1 - e`K(Tt)) and
al(t,T) = ao(t,T) + [pu - e2 (T- t). (22)
The complete markets problem (eu 0) is a special case of Theorem 1 in

which i12(t,T) = F2(t,T) = 1, so that the optimal real wealth allocation and
indirect utility simplify to
W*(MT, lMT) = wt (;1(t,T)) -11YF(t, T) -1, (23)
J (r, W, XL, t) = (i) exp(l y)[B (t,T) rt +ao(t T)] (24)
The problem for the interim consumption case can be written as
(T (C(S)/ts) 7 j(25
max E ds, (25)
C(s):t?s?T Jt l j
s.t. (1) Et Mt F I ds ]Wt/rt wt, (26)
(2) C is financed by a feasible trading

strategy with initial investment WO,
where C(T) is the investor's (nominal) consumption flow at time T: ET [t,T].

The optimal consumption plan and indirect utility function are character-
ized in the following theorem.
THEOREM 2 (OPTIMAL ALLOCATION OF LIFETIME CONSUMPTION): For an investor

cerned with maximizing the expected value of an isoelastic utility function
defined over lifetime consumption in the incomplete markets setting in which
(u 0 0:

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
(i) The optimal real consumption program, c*(s) CC*(s)/I18, is
c*8(s) - q- (t, s) (;l_(t, s)) - l/Ql 1(t, T )wt, (27

where wt is real wealth at time t, and Q1(t,T) is a constant chosen to
satisfy the budget constraint
rT
Q1(t,T) = F, (t,s)F2(t,s) ds
T ~~~~~~~~~~~~~(28)
fTq(t,s)expJ [B(t,s)rt + aI(t,s)]}ds,
and
q(t, s) expt [ 2 - eu (s - t)j
(ii) The indirect utility function J(Wt, rt, I-It, t) is separable in wt and can
be written as
J(r, W, fl, t) -Etr 1 ds}

1-Y ~~~~~~(29)
= (~~jY>J2 (r, t,T),(9
where we have the following.

(iii) f2(r, t,T) represents the contribution to expected utility of the remain-
ing investment opportunities up to the horizon:
02=[ q (t, s) exp [(I-y)/yr[B (t,s)rt +a(t,s ) ds]y
(30)
X x q '- r(t, s) exp [(I -y)/y I [B (t, s) r + a 1(t,s)] ds]
where B (t, s) = K' -(1 - e K('-t)) and a (t, s) is given in (22) by replac-
ing T with s.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
The complete markets problem (fu = 0) is again a special case of Theo-

rem 2 in which r2(t,s) = F2(t,s) = 1, so that the optimal consumption and
indirect utility simplify to
c*(s) = Q(t,T)-1(;j(t,s))-Y11wt (31)
J (r, W, fl, t) (1t)-Q (t,T)'y (32)
Q(t,T)= exp [(l-y)/y ][B (t,s)rt?+a o(t,s)] ds , (33)
where Q(t,T) is derived from Q1(t,T) by setting 0 0 so that q(t,s) = 1 and

al(t,T) = ao(t,T).
In the complete market setting, the optimal consumption allocation c*(s)
given by equation (31) is of similar form to the optimal terminal wealth
allocation w*(T) in equation (23), and the problem with interim consump-
tion can be interpreted as a summation of terminal wealth problems with
the horizons s varying from t to T and initial wealth allocated to horizon s
equal to [F1(t,s)/f F1 (t,s)ds]wt 13
The interim consumption problem in the incomplete market setting
can also be interpreted as a two-stage problem in which the investor first
allocates his initial wealth across horizons s (s E [t,T]), [F1(t,s)F2(t,s)/
f7'F1(t, s) F2(t,s)ds]wt, and then carries out the terminal-wealth optimiza-
tion problem for each s. The expected utility for the interim consumption
problem is still the sum of the expected utilities of the individual terminal-
wealth problems, although the result no longer simplifies as in the com-
plete market case because of q2.
B. Unconstrained Optimal Portfolio Strategies
The optimal portfolio strategy replicates the optimal terminal wealth and
consumption allocations by dynamically trading the available nominal secu-
rities. Although market incompleteness affects investor utility, the optimal
strategies are of the same form in the complete and incomplete markets
settings, as seen in the following theorems.
THEOREM 3 (OPTIMAL PORTFOLIO STRATEGY FOR TERMINAL WEALTH PROBLEM): The

vector of optimal proportional wealth allocation to the stock and two nominal
bonds for problem (14)-(15), x* (x ,x1 ,x2)', is given by
13 Wachter (1999) provides a similar interpretation in a complete market setting where there
is constant interest rate and stochastic equity risk premium and where the investor only in-
vests in stock and cash.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
1 (1 - ( y)B(t,T) 7 ( ()
= -n A (fY1lape2or) - (f1-UP6 34
= 'f 1A+ (1-- l>' P( s7r-B(t,T)u7.,0)', (35)
where e2 = [0,1,0]', A is the (3 x 1) vector of risk premia of the stock and two
nominal bonds, and Q- upu' is the (3 x 3) variance-covariance matrix of
the nominal security returns. The variable crpe2or> denotes the vector of co-
variances between the security returns and the real interest rate, and the
variable opf represents the vector of covariances between the security returns
and realized inflation.
The balance of the portfolio, 1 - x'i, is invested in the nominal riskless
asset at the rate Rf.
Theorem 3 establishes that there is a feasible trading strategy that fi-

nances the optimal wealth and consumption allocations derived in Theorems
1 and 2. Therefore, the optimal strategies given in Theorem 3 are indeed the
solution to the original portfolio choice problem.
Equation (34) expresses the optimal portfolio as the sum of three port-
folios, in a form that is familiar from Merton (1973). The first portfolio is
proportional to the nominal mean-variance tangency portfolio,14 and the
amount invested in it is inversely related to the investor's relative risk aver-
sion. The second portfolio, fY-lope2o(r, is the one with the largest correlation
with the state variable r, and the third portfolio, fY-lapg, has the highest
correlation with the inflation realization.
Equation (35) expresses the optimal portfolio as the weighted sum of the
nominal mean-variance tangency portfolio and the "minimum risk" port-
folio. This is the portfolio that would be held by an infinitely risk-averse
investor (y -X oo) with horizon T, and it has the highest correlation with the
return on an indexed bond, with a maturity equal to the remaining horizon:
The return on this indexed bond has a real interest rate sensitivity of B (t, T)
and no sensitivity to expected or unexpected inflation.
While the optimal portfolio allocation depends on the maturities of the
two bonds that are chosen for the portfolio, the return characteristics of the
portfolio are completely described by its loadings on the stock return and the
innovations in r and -, and these are determined completely by the inves-
tor's horizon and risk aversion. They are characterized in the following
proposition.
PROPOSITION 1 (OPTIMAL FACTOR LOADINGS FOR NOMINAL PORTFOLIO RETURNS): For

an investor with risk aversion y and horizon T, the optimal stock allocation
and loadings on the innovations in the real interest rate and expected infla-
tion are
14 Where the tangency is from the nominal riskless rate Rf to the nominal risky efficien

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
1 ~S -s Is 1
Xs (y) + 1 - - )Xs (GO) (36)
7 +Is 7,
Bp(y,T) 1 - r (1 p i(o)T), (37)
1 ~
C (y, T) - + 1- - CP (o T), (38)
where
Xs(oo) = s/OS Bp(oo,T) (fr -B(t,T) ur)/(or, Cp(oo,T) lo-T,,
(39)
are the stock allocation and loadings for an investor with very large risk
aversion (y -X oo) and horizon T.
To understand the proposition, consider the price process of a hypothetical

indexed bond with maturity equal to the investment horizon T:
dp*
[r [r-B(t,T)orrAr]dt-B(t,T)cordz7, (40)
where Ar CovQ(dM/M),dZr) -4'pe2 is the real risk premium per unit

risk of dzr. Its nominal return is calculated by applying Ito's Lemma to its
nominal value, P8 flp*
dP* ~
d = [r + 7r-B (t,T)orAr]dt + ( S dzs
+ ( (t, T) )or dZr + (?) o,,dzl,, + ?Udzu. (41)
The loadings in equation (41) are the same as those of the portfolio chosen
by the highly risk-averse investor in equations (39), so that a highly risk-
averse investor chooses loadings on the innovations that match those of the
hypothetical indexed bond with maturity T, leaving himself exposed only to
the unhedgeable component of inflation ( dz,. We call the portfolio that
replicates an index bond up to the unhedgeable inflation risk a pseudo index
bond. Proposition 1 shows that the optimal factor loadings for an investor
with finite risk aversion can be written as a weighted average of the load-
ings of the nominal return of the mean variance tangency portfolio and the
loadings of the nominal return of the pseudo index bond of maturity T.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
The component of the loading on r in equation (37) that represents the

hedge against changes in the real opportunity set represented by r,15 (the
"hedge loading on r"), is ((l/'y) - 1)B(t,T), which depends on three param-
eters: risk aversion y, horizon T - t, and the intensity of mean reversion of
the real interest rate, K. This hedge loading is monotonically decreasing in y
and zero for log utility (y = 1). Its elasticity with respect to the horizon is
equal to the elasticity of B(t,T) with respect to T, which is positive, so that
its absolute value is increasing in the horizon. Finally, its elasticity with
respect to the mean reversion intensity, K, iS
e-K(T-t)[K(T - t) + 1 - eK(T-t)]/(K2B(t,T)) < 0.
Thus, the absolute value of the hedge loading increases with the horizon and
decreases with the speed of mean reversion16 in r. This is reasonable, since
the longer the horizon and the slower the mean reversion, the bigger is the
effect of a given innovation in r on future investment opportunities.
The optimal portfolio strategy for the interim consumption problem (25)-
(26) is the one that yields the optimal consumption program c*(s), s E [t,T],
given by equation (27). Since the interim consumption problem can be in-
terpreted as a two-stage terminal wealth optimization problem, the optimal
portfolio strategy has a similar interpretation: It is a weighted average
of the optimal strategies for terminal wealth problems with horizons at s,
s E [t,T].
THEOREM 4 (OPTIMAL PORTFOLIO STRATEGY WITH INTERIM CONSUMPTION): The vector

of optimal portfolio allocations for problem (25,26), x (x , x ", x *)' is given
by expression (34), with B-B (s, T) replaced by B-B (s, T), where
q(s, u)exp [B(s, u)rs + a1(s, u)]

rT y
B(s,T) B(s,u)du
f q(s, P)exp {; [B(s,P)rs+a1(s,pv)j1dv
(42)
is a weighted average of B(s,u) (u E [s,T]), and q(s,u) and a1(s,u) are de-
fined in Theorem 2.
15 The other component of the loading, ,r/Io-,, is part of the portfolio hedge against
in the price level, 11.
16 Note that eK(T-t) can be written as E' =(K(T - t))n/n!, which is 1 + K(T - t) +
[(K(T - t))2/2] + Since K(T - t) is positive, the omitted terms in the series expansion
of eK(T-t) are all positive. Thus, [K(T - t) + 1 - eK(T-t)] < 0.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
C. Welfare Costs of Unhedgeable Inflation and Myopic Investment Strategy
The effect of market incompleteness on the real terminal wealth alloca-

tion, w , in equation (16) is reflected in the terms r 1(t,T) and F2(t,T): The
former represents the realization of the unhedgeable component of inflation
which directly reduces real wealth, while the latter corresponds to the risk
premium for unhedgeable inflation risk. The effect of unhedgeable inflation
risk on investor welfare may be assessed by comparing the investor's cer-
tainty equivalent wealth17 in the incomplete market to that in the complete
market, using the expressions for expected utility given by (24) and (20).
The ratio of the certainty equivalents in the complete and incomplete mar-
kets is equal to exp[(Y/2) u (Tt) Thus, the investor will be made better
off by the unhedgeable inflation component dzu if y < 20u/f
not too risk averse given the risk premium and risk associated with the
unhedgeable inflation.
If the market is complete, indexed bonds are redundant assets and their
introduction does not affect investor welfare. However, if the market is in-
complete because inflation cannot be hedged with nominal assets (fu 0 0),
the introduction of index bonds allows investors to vary their exposure to the
previously unhedgeable component of inflation which, given the pricing ker-
nel, will tend to improve welfare. Calculation of optimal state-contingent
wealth and consumption then follows the proofs of Theorems 1 and 2 by
allowing the investor to vary his exposure to the previously unhedgeable
price level innovation (udzu. Then the expected utility of the terminal
investor when there are indexed bonds and (u 0 0, JIB(W,r, LI, t),
JIB(W r lI t) =Wt t qf1(r, t, T)exp[(lY)/2Y]( LJYf)2(T

1-y
Comparing expressions (20) and (43), we see that, for a given pricing
kernel,18 the introduction of index bonds increases the investor's certainty
equivalent wealth by the factor e (/2y)(?p-_y6)2(T-t) ? 1. Therefore, exc
investors for whom y = 4u/l, the introduction of index bonds in
welfare by permitting trade in the previously unhedgeable inflation com-
ponent (udzU
The investor's expected utility depends, not only on the available invest-
ment instruments, but also on the investment policy that is followed. For
example, the efficiency gain from employing the optimal dynamic strategy
rather than a myopic strategy can be measured by the ratio of the certainty
17 The certainty equivalent wealth is that amount of wealth at the horizon that would leave
the investor indifferent between receiving it for sure and having $1 today to invest in the stock
and bonds up to the horizon.
18 It is important to remember that the introduction of a new security may change the prices
of existing securities.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
equivalent wealth under the optimal dynamic strategy to that under a my-
opic strategy.19 This efficiency gain ratio, EGR, is
EGR = expt ; 2 var(f, r(s)ds) j 1
= ; exp y1 [2 3 (2K(T- t) - 3 e2K(Tt) + 4eK(Tt))]j
The efficiency gain depends only on the risk aversion parameter and vari-
ance of the cumulative real interest rate; for values of y close to unity the
gain is small; the gain is also small if either o-r is small, or the mean rever-
sion parameter K iS large.
D. Constrained Optimal Portfolio Strategies
Since many investors are constrained from taking short positions, it is

important to also consider constrained strategies. Any portfolio strategy can
be characterized in terms of the loadings of the (nominal) portfolio return on
the innovations in the stock return, the real interest rate, and inflation rate,
XS, Bp-(x1B(t,Tl)+ X2B(t,T2)), and Cp-(x1C(t,Tl) + x2C (t,T2)), where
xi is the proportion of wealth invested in bond i, i = 1,2. While Theorem 3
showed that the unconstrained optimal holdings generically involve two bonds,
the following proposition establishes that the investor's constrained optimal
allocation can be achieved by holding only a single bond in combination with
cash and stock.
PROPOSITION 2: An investor who is constrained from borrowing or taking

short positions in bonds, but who has available a continuum of possible
bond maturities up to a maximum, Tmax, can achieve his constrained opti-
mal portfolio allocation by investing in a single bond of an optimally cho-
sen maturity, -* ' -m,x, cash, and stock.
It is easy to see the intuition behind the above proposition if we recall that
the investor is only interested in the optimal loadings, x , Bp, and Cp, on th
stock return, real interest rate, and expected inflation innovations, respec-
tively. To achieve the optimal loadings Bp and C>, the investor can invest in
several bonds subject to the no short sales constraints. Alternatively, the
investor can choose a single bond with the optimal maturity T * such that the
loadings of this bond adjusted for the cash position are optimal. Figure 1
shows the convex feasible region of constrained portfolio loadings and illus-
trates that all feasible combinations of loadings can be achieved by a com-
bination of cash and a single bond of the appropriately chosen maturity. It is
clear from the figure that it will in general be suboptimal to prespecify the
maturity or factor loadings of the bond.
19 The expected utility under the myopic strategy can be simply calculated by inserting
myopic portfolio allocation in the process for real wealth and calculating E [w 1-/y].

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Aiea of feasible BC,xs combinations

for bond with maturity T
D~=(0,0,1)
Cp (T)
CaSh Feasible region of(B,C)

0=(0,0,0)
Figure 1. Illustration of the feasible region for a portfolio strategy with short sales
1g-
constraint. The curve OEM -
is the locus of (B (t),..g,
C (t)) combinations as t is varied from zero
to Tmax. The point M corresponds to (B (Tmax), C(Tmax)). The area OEM is the feasible region
of (B, C) combinations that are attainable with cash and bonds. Given a selected bond maturity
T, the area DOE is the feasible region of (B, C,x s) combinations.
For a CRRA utility function, the indirect utility J (r, W, H, t) is homo-

geneous in real wealth and can be written as [(W/17I)1-y/(l - y)]qf/(r, t)
Bellman equation for the constrained portfolio problem is then
max - jrO-r + 1/Irl -)(X.
+ qf _/3 (1 Y)2(Xu....S.. - X b B(t,T )r x b C(t,r)O )
+ (-y) (r ? x5u5-,,A, - X b B(t,r)Jor Ar7 - X bC (t,7) o-,A~
es- - erAr -g -
1 1 ~~~~~~~~~~~~~~~(45)
-- 2
(1 - y)(y
2
- 2)o -r2 - y(l - y)
X [X42o - 2XsXBTrs + C (t,r) sn)
+ 2(B(t, ) 2o E + 2B (t, )C (t,T)Jr' ? C (t',) 2 )1
+ 1/f

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
where x. and Xb are the proportion of

respectively, and T = T - t is the maturity of the bond chosen by the investor.
Equation (45) is solved numerically to yield the constrained portfolio strat-
egies that are reported in Section IV below.
The introduction of indexed bonds will generally increase investor welfare
when the investor is subject to short-selling or borrowing constraints. With
index bonds, the investor's portfolio choice problem involves the nominal
bond maturity, the indexed bond maturity, and the proportions of wealth
invested in stocks and nominal and indexed bonds.20
III. Model Calibration
To provide illustrative calculations of horizon effects on optimal portfolio

choice, the model parameters were estimated using a Kalman filter in which
the unobserved state is described by r and -T. There are n observation equa-
tions provided by the relation between the yields on n bonds of different
maturities and the state variables that follow from the bond pricing equa-
tion (10); the final observation equation follows from the discretized relation
between the realized inflation rate and the expected inflation rate.21
The system was estimated using monthly data on 11 constant maturity
U.S. treasury discount bond yields with maturities of 1, 3, 6, and 9 months,
and 1, 2, 3, 4, 5, 7, and 10 years, and CPI inflation for the period January
1970 to December 1995.22 Table I reports the parameter estimates, along
with their standard errors. The mean reversion coefficients imply half-lives
for innovations in the real interest rate and expected inflation of 1.1 and
25.7 years, respectively.23 The market prices of both interest rate risk and
inflation risk are negative and significant.24 Since the loadings of bond re-
turns on innovations in these variables are negative and grow with matu-
rity, estimated bond risk premia are positive and increasing with maturity.
The standard deviation of unexpected inflation, oyn, is about 133 ba
per year, which compares with 411 basis points for the unconditional stan-
dard deviation of inflation and with 136 basis points for the standard devi-
ation of innovations in expected inflation, o-,. The standard devi
20 The constrained optimization problem can only be solved numerically and the
of indexed bonds significantly increases the dimension of the problem. Therefore, we do not
consider indexed bonds in our calculations in Section IV. Campbell and Viceira (1999) offer a
welfare analysis of indexed bonds for constrained investors, but do not allow for maturity choice.
21 See Harvey (1989) for the discussion of the Kalman filter and its estimation. See, for
example, de Jong (1998) for a detailed discussion of estimating term structure parameters
using the Kalman filter technique.
22 The data on yields were kindly provided by David Backus. The CPI data are from the
Bureau of Labor Statistics.
23 These estimates are comparable to those of Campbell and Viceira (1999) for the period of
1952 to 1996. The slow mean reversion in the estimated expected rate of inflation is consistent
with Fama and Gibbons (1982, p. 305) who report that "the short term expected inflation rate
is close to a random walk."
24 Note that by ignoring the standard errors of 1- and ir, the standard errors of all the pa-
rameters are understated.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Table I
Estimates of Model Parameters

Maximum likelihood parameter estimates for the joint process of real interest rate, expected
rate of inflation, and stock returns estimated by implementing Kalman filter using monthly
yields of 11 U.S. constant maturity Treasury bonds, CPI data, and CRSP value-weighted stock
returns for the period from January 1970 to December 1995. Parameter estimates using annual
nominal interest rate and CPI data from 1890 to 1985 are reported in parentheses.
Parameter Estimate Standard Error
Stock return process: dS/S (Rf + AS (S) dt + o-s dzs

-Ss 0.158
As 0.343 0.057
Real interest rate: dr =(r - r)dt + o-rdz.
r 0.012 (0.017) 0.002
K 0.631 (0.105) 0.003
O-r 0.026 (0.013) 0.004
Ar -0.209 0.077
Expected inflation: d17 = a (ir - Ir)dt + o-T,dz,
vT 0.054 0.023
a 0.027 0.009
-1T 0.014 0.005
A1T -0.105 0.005
Realized inflation: dr/rl = Tdt + o-ndzn
0rI 0.013 0.005
Parameters of the pricing kernel process: 1
'Os -0.333
0.170
0.120
PSr -0.129
PS,T-0.024
Pr,7r -0.061 0.002
Standard Errors of Bond Yield Errors by Bond Maturities

1 month 0.0099
3 months 0.0066
6 months 0.0043
9 months 0.0030
1 year 0.0020
2 years 0.0004
3 years 0.0003
4 years 0.0002
5 years <0.0001
7 years 0.0005
10 years 0.0012
innovations in the real interest rate, or, of 260 basis points seems high, but
should be considered in conjunction with the strong mean reversion, which
implies that only about 53 percent of any innovation remains after one year.
The correlation between innovations in the real interest rate and in infla-
tion (expected and realized) is -0.06, which is consistent with the Mundell-

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
0.12 Fi -r
0. I - I I IIld 1 --
A')
0.08 -
0.06 - I
0.04 - I '. `
0.02
0V
-0.02-
1970 1975 1980 1985 1990 1995
Figure 2. Estimated state variables, r and ir. This figure shows the estimated state vari-
ables, r, the real interest rate, and ii-, the expected rate of inflation, derived from the Kalman
filter. The sample period is January 1970 to December 1995. The series is estimated using
monthly observations on 11 U.S. Treasury constant maturity bond yields and U.S. CPI data.
Tobin model and with the empirical findings of Fama and Gibbons (1982) for
the period 1953 to 1977.
The standard errors for the bond yield estimates include model error as
well as sampling error. The standard errors of these "estimation errors" are
quite low, except for maturities up to one year, which vary from 20 basis
points for one year to 99 basis points for one month; for longer maturities,
the standard errors are in the range of 0 to 12 basis points.
Figure 2 plots the time series of the estimated real interest rate, r, and the
expected rate of inflation, r-. The estimated real interest rate varies between
-4 and 7 percent. The series exhibits high short-run variability and strong
mean reversion. In contrast, the expected rate of inflation series exhibits
much less mean reversion. It is possible that the high frequency variability
in r is due to the model's attempts to fit variation in the yields of medium to
long-term bonds-this would account also for the relatively poor fit of the
model at the short end of the term structure.
The estimate of the real interest rate mnean reversion parameter, K, iS
likely to be too high in light of the high frequency oscillations in the esti-
mated real interest rate series that it implies, as seen in Figure 2. Figure 3
plots the annual "realized" real interest rates for the period 1890 to 1985.25
It is clear that there is much less mean reversion even in this noisy series
than that in the estimated series in Figure 2. Therefore, we reestimated the
25 The "realized" real interest rate for a year is the difference between the nominal i
rate and the realized rate of inflation. All data are from Shiller (1989).

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
0.25 I X - -
- estimated r
0.2 ^ |-- realizedr
0.15 '
-O. 1I_ II
-0.15 -0. I~~~~~~~~~~~~~~~~~~~~I

-,
-0.2 -
1890 1900 1910 1920 1930 1940 1950 1960 1970 1980
Figure 3. Estimated state variables, r, and its realized counterpart. This figure shows
the estimated state variables, r, the real interest rate derived from the Kalman filter, and the
realized real interest rate derived by subtracting actual inflation rate from the nominal inter-
est rate. The sample period is 1890 to 1985. The series is estimated using observations on
annual U.S. nominal interest rate and U.S. CPI data.
parameters of the stochastic processes for r and ir with a Kalman filter

using only annual data on the nominal interest rate and inflation for the
period 1890 to 1985. The results are striking-the new value of K iS only
0.105 in contrast to 0.631, while the estimate of or is only half of its previous
estimate.
IV. Unconstrained and Constrained Dynamic Strategies
Optimal investment strategies were determined for the parameter values

calibrated from the monthly data set when the whole of the inflation inno-
vation is unhedgeable ((s = (r = 5, = 0),26Table II summarizes the optim
strategies and the certainty equivalent wealth for different horizons and
risk aversion parameters, y.27 Since a f K, any two bonds with different
maturities are sufficient to span the (nominal or real) returns on all possible
bond portfolios, which are characterized by their loadings on dr and d7r; the
26 No significant relation was found in the data between unexpected inflation and innova-
tions in the state variables or stock return.
27 We analyze the optimal strategies only for the terminal wealth problem and n
lifetime consumption problem since, as shown in Theorem 4, the optimal stock holding is the
same for the two problems and the bond allocation for the second problem with horizon T is a
weighted average of the allocations for the terminal wealth problem for horizons from 0 to T.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Table II
An Unconstrained Optimal Portfolio Strategy for Investors

with the Terminal Wealth Objective (K = 0.631)
This table reports the unconstrained optimal strategy for an investor with different values of
the risk aversion parameter, y, and the investment horizon, T. Bp (Cp) is the sensitivity of
optimal portfolio to innovations in r (ir); xs is the proportional portfolio allocation to the stock
CE is the certainty equivalent wealth at the horizon. The calculation is based on the parameter
estimates in Table I for a current interest rate of three percent. The variable EGR is the ratio
of the CE under the optimal strategy to the CE under a myopic strategy: It is calculated from
equation (44). The variable K iS the mean reversion coefficient for the shadow real interest rate
process.
Risk Aversion Parameter, y
Horizon 0.8 1.5 3.0 5.0 7.0 10.0 15.0
1 month
CE 1.01 1.01 1.00 1.00 1.00 1.00 1.00
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 2.51 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.37 -4.50 -2.29 -1.41 -1.03 -0.74 -0.52
CP -9.64 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
1 year
CE 1.13 1.08 1.05 1.04 1.04 1.03 1.03
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 2.51 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.21 -4.72 -2.73 -1.94 -1.59 -1.34 -1.14
CP -9.64 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
5 years
CE 1.77 1.43 1.27 1.21 1.18 1.16 1.14
EGR 1.00 1.00 1.00 1.01 1.01 1.02 1.03
XS 2.51 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.01 -4.98 -3.25 -2.56 -2.26 -2.04 -1.86
CP -9.64 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
10 years
CE 3.05 2.00 1.58 1.43 1.37 1.32 1.29
EGR 1.00 1.00 1.01 1.02 1.03 1.05 1.09
XS 2.51 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.00 -5.00 -3.29 -2.61 -2.32 -2.10 -1.92
CP --9.64 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
20 years
CE 9.02 3.92 2.43 2.01 1.84 1.72 1.63
EGR 1.00 1.00 1.02 1.05 1.08 1.13 1.21
XS 2.51 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.00 -5.00 -3.29 -2.61 -2.32 -2.10 -1.93
CP -9.64 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
amounts invested in the bonds depend on which bonds are used to achieve
the portfolio loadings. Therefore, in the table, we report the loadings of the
nominal portfolio returns, Bp and Cp; expressions for these loadings fol
immediately from equation (B8). Both the optimal stock allocation and in-
flation loadings are independent of the horizon. While the absolute magni-
tude of the interest rate loading is increasing in the horizon for -y > 1, is

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
4 -
stock
- - 1 year bond
- cash
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
O-2 _ __
-2
-4 0\ -
-6 1 , , . . .
0 2 4 6 8 10 12 14 16 18 20
Investment Horizon (years)
Figure 4. Unconstrained optimal asset allocation for different horizons (y = 3, K =

0.631, and ro = 3%). The unconstrained optimal proportion of wealth invested in the stock
index, the 1-year bond, the 10-year bond, and cash. The optimal allocations are derived using
the parameter estimates in Table V for a current interest rate of three percent.
decreasing in the horizon for -y < 1, and is independent of the horizon for
-y 1= (log utility), this loading is relatively insensitive to the horizon for
T > 5. Thus, in contrast to Xia (2000), who finds strong horizon effects even
at long horizons in models with excess return predictability, horizon effects
here are limited to about five years. Nevertheless, the hedge component of
the optimal bond portfolio is significant. For example, when -y = 3, the my-
opic strategy has a loading of -2.29 on dr, while the optimal strategy for a
five-year investment horizon has a loading of -3.25, so the hedge demand as
measured by the loading on dr is about 42 percent of the myopic demand.
When the horizon increases from 5 to 20 years, the absolute value of the
hedge demand only increases by a further 2 percent of the myopic demand.
The importance of the hedge demand increases rapidly as risk aversion in-
creases; for example, when y = 5, the optimal loading on dr for a five-year
horizon is 182 percent of the loading for a myopic investor.
Figure 4 plots the optimal portfolio holdings of cash, stock, and 1-year and
10-year nominal bonds as a function of the investment horizon for an inves-
tor with -y = 3. The figure confirms the discussion in terms of factor load-
ings. The optimal bond allocation changes little beyond year four, although
there is a big difference between the myopic and the optimal allocation for a
long horizon investor: the myopic allocation in the 1-year bond is 3.24, while
the optimal allocation is 4.94 for T = 5, so that the hedge demand is as high
as 52 percent of the myopic allocation.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Although not shown here, the optimal portfolio allocations are also quite
sensitive to the risk aversion parameter: The stock allocation and the load-
ings on both dr and dr decrease with the risk aversion. In the limit, as risk
aversion becomes infinite, the stock allocation goes to zero and the investor's
dynamic strategy in bonds synthesizes the returns on a pseudo inflation
indexed bond with maturity equal to the remaining investment horizon. In-
vestment in a bond with positive maturity increases the investor's exposure
to inflation risk. If , were equal to zero so that there was no rewar
bearing inflation risk, the investor would eliminate all inflation risk by tak-
ing a short position in at least one of the bonds. For the parameter estimates
in Table I, all of the optimal portfolios in Table II involve at least one short
bond or cash position.
Perhaps surprisingly, the costs imposed by the unhedgeability of the in-
flation surprise are quite small. Assuming that the reward for bearing in-
flation risk is zero (0b, = 0), the certainty equivalent cost imposed by
unhedgeable inflation risk is exp( u/2)(Tt); this amounts to only about
1 percent of wealth for a 20-year horizon investor with -y = 5 because (, the
volatility of unhedgeable inflation, is only 1.3 percent per year.
Table III reports the constrained optimal strategies for the same param-
eter values as Table II: The portfolio allocation is now shown as the propor-
tion of wealth allocated to bonds and stock together with the optimal maturity
of the bond. The constraints modestly reduce the investor's certainty equiv-
alent wealth. For example, an investor y = 3 and T = 20 years requires only
about 13 percent additional wealth to compensate for the constraints. The
portfolio allocation is relatively insensitive to the investment horizon: The
stock-bond ratio is virtually independent of the horizon, and the maturity of
the optimal bond shows little variation beyond year five. However, the op-
timal portfolio is quite sensitive to the risk aversion parameter. As risk aver-
sion increases, the ratio of bonds to stock increases at all horizons; moreover,
the maturity of the optimal bond decreases as shown in Figure 5.
The efficiency gain from following a dynamic strategy is calculated from
equation (44). Since the estimate of K from the first data set is very high, the
estimates of efficiency gains reported in Table II are very small except for
long horizons and strong risk aversion-for -y equal to 15, the efficiency gain
over 20 years is around 21 percent. When the investor faces constraints,
EGR is calculated numerically. Estimates of EGR reported in Table III are
all close to one, reflecting the fact that the investor's optimal portfolio strat-
egy is close to the myopic one when there are constraints.
Tables IV and V report the unconstrained and constrained portfolio strat-
egies for the value of K corresponding to the set of annual data, holding the
other parameters unchanged. There are no horizon effects in Cp and
the unconstrained strategies now exhibit strong horizon effects in Bp e
long horizons: When y is less than unity, Bp decreases with the horizon,
while the reverse is true for -y greater than unity. The strong horizon effect
is also evident in Figure 6, which plots the optimal portfolio holdings for an
investor with y = 3 who can invest in cash, stock, and 1-year and 10-year
bonds. Most significantly, the efficiency gain over the myopic strategy is

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Table III
A Constrained Optimal Portfolio Strategy for Investors

This table reports the constrained optimal strategy for an investor with different values of the
risk aversion parameter, y, and the investment horizon, T. The variables xs and XB denote the
proportional allocations to the stock and the optimal bond. The variable r represents the ma-
turity of the optimally chosen bond. The variable CE is the certainty equivalent wealth at the
horizon. The calculation is based on the parameter estimates in Table I for a current interest
rate of three percent. The variable EGR is the ratio of the CE under the optimal strategy to the
CE under a myopic strategy: It is calculated numerically. The variable K iS the mean reversion
coefficient for the shadow real interest rate process.
Horizon 0.8 1.5 3.0 5.0 7.0 10.0 15.0
1 month
CE 1.01 1.01 1.00 1.00 1.00 1.00 1.00
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 1.00 1.00 0.64 0.40 0.29 0.20 0.13
XB 0.00 0.00 0.36 0.60 0.71 0.80 0.87
r n.a. n.a. 7.49 3.08 1.86 1.09 0.62
1 year
CE 1.07 1.06 1.05 1.04 1.04 1.03 1.03
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 1.00 1.00 0.63 0.40 0.29 0.20 0.14
XB 0.00 0.00 0.37 0.60 0.71 0.80 0.86
T n.a. n.a. 7.24 3.34 2.35 1.71 1.27
5 years
CE 1.36 1.31 1.23 1.18 1.16 1.14 1.12
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.01
XS 1.00 1.00 0.62 0.39 0.29 0.20 0.14
XB 0.00 0.00 0.38 0.61 0.71 0.80 0.86
T n.a. n.a. 7.00 3.54 2.71 2.18 1.81
10 years
CE 1.81 1.68 1.47 1.37 1.31 1.27 1.22
EGR 1.00 1.01 1.00 1.00 1.00 1.01 1.03
XS 1.00 1.00 0.62 0.39 0.28 0.20 0.14
XB 0.00 0.00 0.38 0.61 0.72 0.80 0.86
T n.a. n.a. 7.00 3.54 2.73 2.21 1.85
20 years
CE 3.22 2.77 2.16 1.82 1.65 1.49 1.31
EGR 1.00 1.00 1.00 1.00 1.00 1.01 1.02
XS 1.00 1.00 0.63 0.40 0.29 0.21 0.14
XB 0.00 0.00 0.37 0.60 0.71 0.79 0.86
T n.a. n.a. 7.00 3.55 2.74 2.22 1.85
n.a.: not applicable.
now substantial: When the horizon is 20 years, the gain is 119 percent for -y
equal 5 and 252 percent for y = 7 in the unconstrained case, and 54 percent
and 121 percent, respectively, in the constrained case.
Horizon effects are now also evident for the constrained strategies shown
in Table V. First, in contrast to the popular view that long-horizon investors

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
8 l l l
* I
- 1
7 - - 5 year horizon
O- 10year hor
20 year horio
2 -~
2 4 6 8 10 12 14 16
risk aversion, y
Figure 5. Optimal bond maturity and risk aversion for different horizons (r
0.631). This figure plots the optimal bond maturity chosen by an investor who faces
and short sales constraints for different values of the risk aversion parameter, y,
ment horizon.
should hold more stock than short-horizon investors,28 the optimal stock
holding decreases with the horizon; for example, for -y =3, the optimal hold-
ing of stock is 66 percent for a one-month horizon and only 40 percent for a
20-year horizon. Similarly, the stock-bond ratio also decreases with the ho-
rizon beyond year one. These results are obviously sensitive to the assump-
tion of the model that the equity risk premium is constant. In contrast to the
previous example, a myopic investor with a large enough risk aversion pa-
rameter may want to hold cash, and, in such circumstances, the constraints
are not binding. However, when the investment horizon is longer than one
year, a mix of stocks and bonds (no cash) dominates portfolios with cash. The
second horizon effect, which is shown in Figure 7, is that the maturity of the
optimally chosen bond increases with the horizon-the effect is much more
pronounced with the reduced degree of mean reversion in r, which generally
leads the long-term investor to hold a much longer maturity bond: The ma-
turity of the optimally chosen bond for a 20-year investor with -y =3 rises
from 7 years when K iS 0.63 to 13.5 years when K iS 0.11.
28 See, for example, Siegel (1998, p. 283), "Stocks should constitute the overwhe
portion of all long-term financial portfolios."

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Table IV
An Unconstrained Optimal Portfolio Strategy for Investors

This table reports the unconstrained optimal strategy for an investor with different values of
the risk aversion parameter, y, and the investment horizon, T. The variable Bp (Cp) is the
sensitivity of the optimal portfolio to innovations in r (v-); xS is the proportional portfolio al-
location to the stock. The variable CE is the certainty equivalent wealth at the horizon. The
calculation is based on the parameter estimates in Table I for a current interest rate of three
percent. The variable EGR is the ratio of the CE under the optimal strategy to the CE under a
myopic strategy: It is calculated from equation (44). The variable K is the mean reversion
coefficient for the shadow real interest rate process.
Horizon 0.8 1.5 3.0 5.0 7.0 10.0 15.0
1 month
CE 1.01 1.01 1.00 1.00 1.00 1.00 1.00
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 2.52 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.37 -4.50 -2.29 -1.41 -1.03 -0.75 -0.52
CP -9.63 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
1 year
CE 1.14 1.09 1.06 1.05 1.04 1.04 1.04
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 2.52 1.34 0.67 0.40 0.29 0.20 0.13
Bp -8.15 -4.79 -2.87 -2.10 -1.77 -1.53 -1.33
CP -9.63 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
5 years
CE 1.85 1.51 1.35 1.28 1.26 1.24 1.22
EGR 1.00 1.00 1.01 1.03 1.05 1.08 1.13
XS 2.52 1.34 0.67 0.40 0.29 0.20 0.13
Bp -7.42 -5.77 -4.83 -4.45 -4.29 -4.17 -4.08
CP -9.63 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
10 years
CE 3.31 2.27 1.83 1.68 1.61 1.57 1.53
EGR 1.00 1.01 1.08 1.19 1.33 1.56 2.05
XS 2.52 1.34 0.67 0.40 0.29 0.20 0.13
Bp --6.84 -6.54 -6.36 -6.29 -6.27 -6.24 -6.23
CP -9.63 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
20 years
CE 10.19 5.02 3.35 2.84 2.64 2.49 2.37
EGR 1.01 1.04 1.39 2.19 3.52 7.24 24.41
XS 2.52 1.34 0.67 0.40 0.29 0.20 0.13
Bp --6.30 -7.26 -7.81 -8.03 -8.12 -8.19 -8.25
CP -9.63 -5.14 -2.57 -1.54 -1.10 -0.77 -0.51
V. Conclusion
In this paper, we have derived the optimal dynamic strategies for an in-
vestor with power utility in an economy with stochastic inflation and real
interest rates, as well as a constant equity premium, when there exists no
riskless security. Closed-form expressions were obtained for the optimal port-

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Table V
A Constrained Optimal Portfolio Strategy for Investors with the

Terminal Wealth Objective (K = 0.105)
This table reports the constrained optimal strategy for an investor with different values of the
risk aversion parameter, y, and investment horizon, T. xs and XB are the proportional alloca-
tions to the stock and the optimal bond. r is the maturity of the optimally chosen bond. CE is
the certainty equivalent wealth at the horizon. The calculation is based on the parameter es-
timates in Table I for rD = 3 percent. EGR is the ratio of the CE under the optim
the CE under a myopic strategy: It is calculated numerically. K is the mean reversion coefficient
for the shadow real interest rate process.
Horizon 0.8 1.5 3.0 5.0 7.0 10.0 15.0
1 month
CE 1.01 1.01 1.01 1.00 1.00 1.00 1.00
EGR 1.00 1.00 1.00 1.00 1.00 1.00 1.00
XS 1.00 1.00 0.66 0.40 0.29 0.20 0.13
XB 0.00 0.00 0.34 0.44 0.31 0.22 0.15
r n.a. n.a. 9.71 3.72 3.72 3.72 3.72
1 year
CE 1.08 1.07 1.06 1.05 1.04 1.04 1.04
EGR 1.00 1.00 1.01 1.00 1.00 1.00 1.00
XS 1.00 1.00 0.64 0.40 0.29 0.20 0.14
XB 0.00 0.00 0.36 0.60 0.71 0.80 0.86
r n.a. n.a. 10.81 3.69 2.40 1.72 1.30
5 years
CE 1.48 1.42 1.33 1.27 1.24 1.21 1.18
EGR 1.04 1.04 1.01 1.01 1.04 1.06 1.10
XS 1.00 1.00 0.55 0.38 0.28 0.20 0.13
XB 0.00 0.00 0.45 0.62 0.72 0.80 0.87
r n.a. n.a. 12.60 7.15 5.61 4.72 4.21
10 years
CE 2.14 1.97 1.73 1.56 1.46 1.27 1.28
EGR 1.06 1.10 1.03 1.11 1.20 1.36 1.72
XS 1.00 0.66 0.47 0.32 0.23 0.16 0.09
XB 0.00 0.34 0.53 0.68 0.77 0.84 0.91
r n.a. 27.77 13.20 9.07 7.72 6.94 6.58
20 years
CE 4.39 3.61 2.63 2.00 1.66 1.37 n.a.
EGR 1.11 1.24 1.12 1.54 2.21 3.97 n.a.
xS 1.00 0.63 0.40 0.24 0.16 0.09 n.a.
XB 0.00 0.37 0.60 0.76 0.84 0.91 n.a.
r n.a. 25.98 13.53 10.26 9.15 8.50 n.a.
n.a.: not available.
folio when the investor is free to take unconstrained portfolio positions, and
it was shown that the optimal portfolio position can be achieved by invest-
ments in stocks, cash, and two nominal bonds. In this setting, the optimal
allocation to stocks and the optimal portfolio loading on the innovation in
inflation are independent of the horizon, while the optimal loading on the

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
30
20 -
stock
- - 1 year bond
10 - 10 | _IOyear bond
o . . . .. . .... ...... .... ... .......... ................... .. .... ...... .......... ......... ..... ...... ..... ............. .....
i: ~ ~ - -- - - - - - -- - - -- - - - - - - - - - - -
-10 -
-20 -
-301
0 2 4 6 8 10 12 14 16 18 20
Investment Horizon (years)
Figure 6. Unconstrained optimal asset allocation for different horizons (y = 10, K =

0.105, and ro = 3%). This figure plots the unconstrained optimal proportion of wealth
in the stock, the 1-year bond, the 10-year bond, and cash. The optimal allocations are derived
using the parameter estimates in Table V.
innovation in the shadow real interest rate is increasing in the horizon for
investors more risk averse than the log. The efficiency gain of the optimal
dynamic strategy over the myopic strategy was shown to be a function of
both the investor's risk aversion and the variance of the cumulative real
interest rate over the investor's horizon: The gain is small for risk aversion
close to the log, as well as when either the variance of innovations in the
real interest rate is small or the mean reversion in the real interest rate is
large. When the investor is constrained from holding short positions or bor-
rowing, the optimal portfolio was shown to be achievable with an investment
in stocks, cash, and a single bond with an optimally chosen maturity.
The model was calibrated to monthly data on U.S. Treasury bond yields
and inflation for the period 1970 to 1995. The resulting parameter estimates
implied an unreasonably high degree of mean reversion in the real interest
rate and yielded very small estimates of the efficiency gain of the dynamic
strategy and only limited horizon effects in optimal portfolios. A striking
characteristic of the optimal constrained portfolios is that, not only does the
allocation to bonds increase with risk aversion, but the maturity of the op-
timal bond decreases as risk aversion increases.
When the real interest rate mean reversion parameter is calibrated to a
long history of annual data, it falls from 0.63 to 0.11. With this parameter-
ization, both horizon effects and the efficiency gains of the optimal dynamic
strategy become large. Thus, the importance of dynamic considerations in

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
30
I* 1 month horizon
I- year horizon
- - 5 year horizon
25 -- 10 year horizon
20 year horizo
'15 .2O
>1
0 20
15
0
2 4 6 8 10 12 14 16
risk aversion, y
Figure 7. Optimal bond maturity and risk aversion for different horizons (ro = 3%, K =
0.105). This figure plots the optimal bond maturity chosen by an investor who faces borrow-
ing and short sales constraints for different values of the risk aversion parameter, y, and in-
vestment horizon.
optimal asset allocation depends critically on the stochastic characteristics

of the investment opportunity set. Further work is required to assess more
precisely the dynamics of the real interest rate in the United States.
Appendix A: Nominal Bond Price
The nominal bond price, P(t,T), is given by
P (t, T) exp Et[In(M )- In( )]+ -Vart[In(M )- In(~)}

([(Mt ) (tI) 2 [(Mt ) (tI)
(Al)
Letting b'b,
Lettig 02 ,P0, V + =02
2 _'P,VM V_1
+ 02, = ~ +
VEI=( U ~2
ln(MT/Mt) = fT(-r(s) - 2 VM)
ln(tI7T/-It) J;T (s) - 2 i) ds +fJ fdz + (u dz (A3)

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Then define
B(t,T) = K ( - eK(t-T)) (A4)
C(t,T) = a-1(1 - e a(t-T)) (A5)
Using equation (A2) and (A3),
Et[In(MT/Mt)1 = -r(T- t) + (r - rt)B(t,T) - - VM(T - t) (A6)

2
Vart [In(MT/M)1 = -2 [2K(B (t,T) - (T - t)) + K2B2(t,T)] + VM(T - t)

2K 3
- 2 r (OlS Psr + Ot'r + 'r pir)[T - t - B (t, T)]; (A7)

K
and
Et[In(UIT/UIt)1 = T(T- t) - (T- -Tt)C(t,T)2 - - ( u + 5)(T - t) (A8)
Vart[ln(IET/)1 = -2 [2a(C(t,T) - (T- t)) + a2C2(t,T)]
+ ((2 + >2) (T - t) (A9)
+2o-
a
((SPs1r + (r Prir + $7r
In addition, we have
CV- cov [ln I n (T)
= - [r L(T-t) - B(t,T),-C(t,T) + +
aK a + K
(A10)
- -(SSPSr + (r + r prr)[(T - t) - B(t,T)] + Ou u(T-t)

K
+ -(OSPss7r
a
+ OrPrir + 4v) [(T - t) - C(t,T)] + O'p (T - t).

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Therefore,
P(t,T) = exp{A(t,T) - B(t,T)rt - C(t,T)wt}, (All)
where
A(t,T) = [B(t,T) - (T - t)]i* + [C(t,T) - (T- t)]-T*
- 43r [2K(B(t, T) - (T - t)) + K2B2(t, T)]
- 4 3 [2a(C(t,T) - (T - t)) + a2C2(t,T)] (A12)
+ La [(T- t)-C(t,T) - B(t,T) + a+ K 1

+ ^ a
+ ((sA
In the above equation, r* = - Ar(-r/K), and T = - A,(o-(/a), where
As-( S + (rPSr + (wrPsr T)- (S + OrrPSr + 0Pisi), (A13)
Ar- (SPSr + (r + (rPrw) -(OSPSr + Oir + (/7TPrFr) (A14)
AvT((s Ps1 + (r Prir + (3)O-(ls Psr + Or Prir + ()r), (A15)
Au- - 4u (A16)
The variable r* can be interpreted as the long-run mean of the real interest
rate under the risk-neutral measure and I* as the long-run mean of the
expected inflation rate.
Appendix B: Proof of Theorems
Proof of Theorem 1: Since -q,, and ;, are orthogonal to 71

IT =q ft771(t,T?2(t,T), we can write the Lagrangian for feasible wealth
processes as
C (WT/77l(t,T))17 lE; L )1
L (Bi)t(tT)Ifl-
- iEtL1(tl-) Et L2(t,T) m(tT - (Bl
- b4E 01(t T) t T 7 Et0; (t, T ) 772(t, T ) H 0t ntt

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
because the investor can vary nominal wealth across states in 01, but not
across states in 02. The first order conditions are
(r,(t,i)T ) I Et L X2(t, T) y = 61(t,T)Et k2(t,T) (t,
Et[j (t, T) WT t ]Et ;2(t, T) r(tT]=wt (1B3)
Equation (B2) can be rewritten as
WT = (6)-/(1(,T)-/71(,T t.(1B4)
Substitute equation (B4) into (B3) and solve for 5, then eliminate 8 from (B4)
by substituting the value of 5 to obtain
WT- -WT-T = %1 (t,T) wtVj(t,T) -11-FI(t,T) -1F2(t, T) -1, (B5
where F1(t, T) =Et [;(t, T)1-(l)] and F2(t, T) = Et 1;2(t, T)l-q2(t, T)]
ex(62-6"0,)(T-t)
Substituting wT given in equation (B5) into the investor's J function (20)
yields
J (wt),y eF (t, T) -/F2(t,T)7-1-F3(tT), (1B6)

-y
where F3(t,T) = Et[1'2(t,T)Y-'] = exp[(-Y)(2u)/2lf(Tt). The expression for

q,l(t,T) is derived by substituting for Fl(t,T), F2(t, T), and F3(t, T). Q. E.D.
The proof of Theorem 2, which is similar to that of Theorem 1, is omitted

and is available from the authors on request.
Proof of Theorem 3: Define G, G[MS,r,s] EJ[(WT/1T)(MT/M,)]

(real) value at time s of the optimally chosen terminal payoff WT. Then,
using the definition of WT, equation (16),
G-Wt F, (s, T) F2(s, T) Sj (t, s) - 1/yB7

s rl.t F, (t, T) F2(t, T) 712(t,5S)
where F1(t,T) and F2(t,T) are given in equations (17)-(18). Note that F2(t,T)
and F2(s,T) are functions of horizon only, while F1(t,T) is a function of the
horizon and the realization of state variable at time t, rt. In contrast, Fl(s
is a function of the current state variable r5. More specifically, lnF1(s,T) is
a summation of a function of T - s and -(1 - (1/y))B(t,T)r5. Therefore, the

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
stochastic terms of Gs come from l1(t,s)'1/), %2(t,s), and rS, while the other
terms are deterministic functions of the horizon T - s. Using Ito's Lemma,
we know that the stochastic terms of d In G. will be from those of
-(1/y)d ln 1, and -(1 - (1/y)) B(t,T)dr. Thus, the (log) instantaneous real
return on optimally invested wealth is
Kskr !1(,)jlZ
d ln G. = g1(r T-s)dt- -dz, - + (
Y LY y(B8)
- + dz -$ dzu,
Y
where the drift term g1(r, s) is a function of current real interest rate r. and
the remaining investment horizon T - s.
Now consider the log real return on portfolio x*, which is a vector of op-
timal proportion of wealth invested in stock, a bond with maturity T1, and a
bond with maturity T2. The remaining wealth, 1 - i'x*, is invested in cash
(a nominal instantaneous risk-free asset). The nominal wealth process is
given by
dW
W= (Rf + x'A)dt + x'odz. (B9)
The log real wealth process, ln w = ln W - ln U, is derived by using the

inflation process (7) and Ito's Lemma:
dlnw= g2(r,T-s)dt + (xl1u- rS)dz -[(x2B + X3B2) err + Ar]dZr
- [(X2C1 + X3C2)ofs + 6J]dz - (udzu. (B10)
Since strategy x* yields the terminal payoff w+, it follows that the coeffi-
cients in equations (B8) and (B10) are identical. Equating coefficients yields
(34). Q.E.D.
The proof of Theorem 4, which is similar to that of Theorem 3, is omitted

and is available from the authors on request.
Proof of Proposition 2: Suppose without loss of generality that K > a. Then,

allowing bond maturity, T, to vary, a bond's return loading on the real inter-
est rate, B, is an increasing convex function of its loading on inflation, C.
Then the set of achievable factor loading combinations is defined by
S = z{ (B, C)BTmaB-C, 1- (1 - KB) aKj

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Any point in this set, (Bp,Cp), can be achieved by a co

cash (0,0) and the loadings of a single bond with maturity r* ' Tmax, such
that C-/BB* = Cp/Bp, where the weight on the bond is Bp/B,*. Q.E.D.
REFERENCES
Barberis, Nicholas, 2000, Investing for the long run when returns are predictable, Journal of
Finance 55, 225-264.
Bossaerts, Peter, and Pierre Hillion, 1999, Implementing statistical criteria to select return
forecasting models, The Review of Financial Studies 12, 405-428.
Brennan, Michael J., Eduardo S. Schwartz, and Ronald Lagnado, 1997, Strategic asset alloca-
tion, Journal of Economic Dynamics and Control 21, 1377-1403.
Brennan, Michael J., and Yihong Xia, 2000, Stochastic interest rates and the bond-stock mix,
European Finance Review 4, 197-210.
Campbell, John Y., and Luis M. Viceira, 1999, Who should buy long term bonds? Working paper,
Harvard University.
Canner, Niko, N. Gregory Mankiw, and David N. Weil, 1997, An asset allocation puzzle, Amer-
ican Economic Review 87, 181-191.
Cox, John C., and Chi-fu Huang, 1989, Optimum consumption and portfolio policies when asset
prices follow a diffusion process, Journal of Economic Theory 49, 33-83.
de Jong, Frank, 1998, Time-series and cross-section information in affine term structure mod-
els, Working paper, University of Amsterdam.
Duffie, Darrell, and Chi-fu Huang, 1985, Implementing Arrow-Debreu equilibria by continuous
trading in a few long-lived securities, Econometrica 53, 1337-1356.
Fama, Eugene F., and Michael R. Gibbons, 1982, Inflation, real returns and capital investment,
Journal of Monetary Economics 9, 297-323.
Giovannini, Alberto, and Philippe Weil, 1989, Risk aversion and intertemporal substitution in
the capital asset pricing model, NBER Working Paper No. 2824, Cambridge, Massachusetts.
Hakansson, Nils H., 1970, Optimal investment and consumption strategies under risk for a
class of utility functions, Econometrica 38, 587-607.
Harvey, Andrew, 1989, Forecasting, Structural Time Series Models and the Kalman Filter (Cam-
bridge University Press, Cambridge, Great Britain).
He, Hua, and Neil D. Pearson, 1991, Consumption and portfolio policies with incomplete mar-
kets and short-sale constraints: The infinite dimensional case, Journal of Economic Theory
54, 259-304.
Kandel, Shumael, and Robert F. Stambaugh, 1996, On the predictability of stock returns: An
asset-allocation perspective, Journal of Finance 51, 385-424.
Karatzas, loannis, John P. Lehoczky, Steven E. Shreve, and Ganlin Xu, 1991, Martingale and
duality methods for utility maximization in an incomplete market, SIAM Journal of Con-
trol and Optimization 29, 702-730.
Kim, Tong, and Edward Omberg, 1996, Dynamic non-myopic portfolio behavior, Review of Fi-
nancial Studies 9, 141-161.
Latane, Henry A., and Donald L. Tuttle, 1967, Criteria for portfolio building, Journal of Fi-
nance 22, 359-373.
Liu, Jun, 1999, Portfolio selection in stochastic environments, Working paper, Stanford University.
Markowitz, Harry, 1959, Portfolio Selection: Efficient Diversification of Investment (Wiley, New
York).
Merton, Robert C., 1969, Lifetime portfolio selection under uncertainty: The continuous-time
case, Review of Economics and Statistics 51, 247-257.
Merton, Robert C., 1971, Optimum consumption and portfolio rules in a continuous-time model,
Journal of Economic Theory 41, 867-887.
Merton, Robert C., 1973, An intertemporal capital asset pricing model, Econometrica 41, 867-887.
Mossin, Jan, 1968, Optimal multi-period portfolio policies, Journal of Business 41, 215-229.
Mundell, Robert, 1963, Inflation and real interest, Journal of Political Economy 71, 280-283.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00
Omberg, Edward, 1999, Non-myopic asset-allocation with stochastic interest rates, Working
paper, San Diego State University.
Pages, Henri, 1987, Optimal consumption and portfolio rules in a continuous time model, Work-
ing paper, MIT.
Samuelson, Paul A., 1969, Lifetime portfolio selection by dynamic stochastic programming,
Review of Economics and Statistics 51, 239-246.
Shiller, Robert, 1989, Market Volatility (M.I.T. Press, Cambridge, MA).
Siegel, Jeremy, 1998, Stocks for the Long Run (McGraw-Hill, New York).
Sorensen, Carsten, 1999, Dynamic asset allocation and fixed income management, Journal of
Financial and Quantitative Analysis 34, 513-532.
Vasicek, Oldrich, 1977, An equilibrium characterization of the term structure, Journal of Fi-
nancial Economics 5, 177-188.
Wachter, Jessica, 1999, Optimal allocation for an investor with utility over consumption: A
complete markets approach, Working paper, Harvard University.
Xia, Yihong, 2000, Learning about predictability: The effect of parameter uncertainty on dy-
namic asset allocation, The Journal of Finance 56, 205-246.

130.225.27.190 on Mon, 12 Feb 2024 16:14:49 +00:00

Dynamic Asset Allocation Under Inflation

Uploaded by

Copyright:

Available Formats

Dynamic Asset Allocation Under Inflation

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dynamic Asset Allocation Under Inflation

Uploaded by

Copyright:

Available Formats

Dynamic Asset Allocation under Inflation

Author(s): Michael J. Brennan and Yihong Xia

Stable URL: https://www.jstor.org/stable/2697777

This content downloaded from