This action might not be possible to undo. Are you sure you want to continue?

**Real Interest Rates:
**

Theory and Evidence from

UK Index-Linked Bonds

Juha Sepp¨ al¨ a

∗

Department of Economics, University of Illinois,

1206 South Sixth Street, Champaign, IL 61820, USA

tel: (217) 244-9506 fax: (217) 244-6678

email: seppala@uiuc.edu

October 15, 2003

∗

I am grateful to Fernando Alvarez, Marco Bassetto, Marco Cagetti, John Cochrane, Bernard Dumas, Sylvester

Eijﬃnger, Alex Monge, Casey Mulligan, Marcelo Navarro, Antti Ripatti, Thomas J. Sargent, Pietro Veronesi, Anne

Villamil, Jouko Vilmunen, and especially Lars Peter Hansen, Urban Jermann, and an anonymous referee for comments

and discussions. In addition, Lili Xie provided excellent research assistance for the latest revision of the paper. I

also received useful comments from seminar participants at the Hansen–Sargent Macro lunch group, J.P. Morgan,

Bank of Finland, University of Chicago Financial Engineering workshop, Lehman Brothers, Stockholm School of

Economics, University of Michigan, McGill University, CEPR/LIFE/Wharton Conference on International Finance

and Economic Activity (Vouliagmeni, 2000), Society for Economic Dynamics Meetings (Costa Rica, 2000), the World

Congress of the Econometric Society (Seattle, 2000), and the University of Illinois. Juha Niemel¨a and Ulla Rantalahti

helped me obtain the macro data. Martin Evans kindly shared his data on nominal and real yields. The British

Household Panel Survey data were made available through The Data Archieve. The data were originally collected

by the ESRC Research Center on Micro-Social Change at the University of Essex. I would like to thank the Yrj¨o

Jahnsson Foundation for ﬁnancial support. Parts of this paper were written while I was visiting the Bank of Finland

Research Department. I would like to thank the organization for its support. All errors are mine.

Abstract

This paper studies the behavior of the default-risk-free real term structure and term premia

in two general equilibrium endowment economies with complete markets but without money.

In the ﬁrst economy there are no frictions as in Lucas (1978) and in the second risk-sharing is

limited by the risk of default as in Alvarez and Jermann (2000, 2001). Both models are solved

numerically, calibrated to UK aggregate and household data, and the predictions are compared

to data on real interest rates constructed from the UK index-linked data. While both models

produce time-varying risk or term premia, only the model with limited risk-sharing can generate

enough variation in the term premia to account for the rejections of expectations hypothesis.

JEL classiﬁcation: E43, E44, G12.

Keywords: Term structure of interest rates, General equilibrium, Default risk, Term pre-

mia, Index-linked bonds.

1 Introduction

One of the oldest problems in economic theory is the interpretation of the term structure of interest

rates. It has been long recognized that the term structure of interest rates conveys information

about economic agents’ expectations about future interest rates, inﬂation rates, and exchange rates.

In fact, it is widely agreed that the term structure is the best source of information about economic

agents’ inﬂation expectations for one to four years ahead.

1

Since it is generally recognized that

monetary policy can only have eﬀect with “long and variable lags” as Friedman (1968) put it,

the term structure is an invaluable source of information for monetary authorities.

2

Moreover,

empirical studies indicate that the slope of the term structure predicts consumption growth better

than vector autoregressions or leading commercial econometric models.

3

Empirical research on the term structure of interest rates has concentrated on the (pure) expec-

tations hypothesis. That is, the question has been whether forward rates are unbiased predictors of

future spot rates. The most common way to test the hypothesis has been to run a linear regression

(error term omitted):

r

t+1

−r

t

= a +b(f

t

−r

t

),

where r

t

is the one-period spot rate at time t and f

t

is the one-period-ahead forward rate at time t.

The pure expectations hypothesis implies that a = 0 and b = 1. Rejection of the ﬁrst restriction

a = 0 is consistent with the expectations hypothesis with a term premium that is nonzero but

constant.

By and large the literature rejects both restrictions.

4

Rejection of the second restriction, b =

1, requires a risk or term premium that varies through time and is correlated with the forward

premium, f

t

− r

t

. Many studies—e.g., Fama and Bliss (1987) and Fama and French (1989)—take

this to indicate the existence of time-varying risk or term premium.

5

In order for policy makers

to extract information about market expectations from the term structure, they need to have a

general idea about the sign and magnitude of the term premium. Obviously, as emphasized by

Cochrane (1999), the sign and the magnitude of the term premium are also interesting for market

professionals. Therefore, it is interesting to ask if there are models that are capable of generating

term premia that are similar to the ones observed in actual time series. Unfortunately, as S¨ oderlind

and Svensson (1997) note in their review,

“We have no direct measurement of this (potentially) time-varying covariance [term

1

See, e.g., Fama (1975, 1990) and Mishkin (1981, 1990a, 1992) for studies on inﬂation expectations and the term

structure of interest rates using U.S. data. Mishkin (1991) and Jorion and Mishkin (1991) use international data.

Abken (1993) and Blough (1994) provide surveys of the literature.

2

Svensson (1994ab) and S¨ oderlind and Svensson (1997) discuss monetary policy and the role of the term structure

of interest rates as a source of information.

3

See, e.g., Harvey (1988), Chen (1991), and Estrella and Hardouvelis (1991).

4

The literature is huge. Useful surveys are provided by Melino (1988), Shiller (1990), Mishkin (1990b), and

Campbell, Lo, and MacKinlay (1997). Shiller (1979), Shiller, Campbell, and Schoenholtz (1983), Fama (1984, 1990),

Fama and Bliss (1987), Froot (1989), Campbell and Shiller (1991), and Campbell (1995) are important contributions

to this literature.

5

The literature is somewhat inconsistent in the deﬁnitions of risk and term premium. In the discussion below,

both terms are used interchangeably. In Section 3.3, I will deﬁne the risk premium as the term that accounts for

the rejections of expectations hypothesis in prices and term premium as the term that accounts for the rejections of

expectations hypothesis in rates.

3

premium], and even ex post data is of limited use since the stochastic discount factor is

not observable. It has unfortunately proved to be very hard to explain (U.S. ex post)

term premia by either utility based asset pricing models or various proxies for risk.”

The question whether utility-based asset pricing models are capable of generating risk premia

similar to the ones observed in actual time series was originally posed in Backus, Gregory, and

Zin (1989). They use a complete markets model ﬁrst presented by Lucas (1978), and their answer

is that the model can account for neither sign nor magnitude of average risk premia in forward

prices and holding-period returns.

6

In addition, they show that one cannot reject the expectations

hypothesis with data generated via the Lucas model. Donaldson, Johnsen, and Mehra (1990) obtain

the same result for a general equilibrium production economy, and den Haan (1995) investigates

the issue further.

The problem is that given the variability in U.S. aggregate consumption series, the stochastic

discount factor derived from frictionless utility-based asset pricing models is not suﬃciently volatile.

This is closely related to the equity premium puzzle ﬁrst posed by Mehra and Prescott (1985) and

the risk-free rate puzzle posed by Weil (1989). Mehra and Prescott (1985) conjecture that the most

promising way to resolve the equity premium puzzle is to introduce features that make certain

types of intertemporal trades among agents infeasible. Usually this has meant that markets are

exogenously incomplete. Economic agents are allowed to trade only in certain types of assets, and

the set of available assets is exogenously predetermined.

Heaton and Lucas (1992) use a three-period incomplete markets model to address the term

premium puzzle. Their answer is that “uninsurable income shocks may help explain one of the

more persistent term structure puzzles” but “the question remains whether the prediction of a

relatively large forward premium will obtain in a long horizon model.”

Alvarez and Jermann (2000, 2001) study the asset pricing implications of an endowment econ-

omy when agents can default on contracts. They show how endogenously determined solvency

constraints that prevent the agent from defaulting on his own contracts help explain the equity

premium and the risk-free rate puzzles. For the purpose of dynamic asset pricing, this framework

has three advantages over the standard incomplete markets approach described above.

7

First, allocations do not depend on a particular arbitrary set of available securities. Second,

the markets are complete and hence any security can be priced. This is particularly important in

addressing questions related to the term structure of interest rates. Finally, ﬁnding the solution

to an incomplete markets problem involves solving a very diﬃcult ﬁxed point problem, whereas

solving the model in Alvarez and Jermann can be very fast and easy to implement.

However, it should be noted that, while empirical research has concentrated on the nominal

term structure, both Lucas (1978) and Alvarez and Jermann (2000, 2001) are models of endowment

economies without money. This paper studies the implications of the Lucas and Alvarez-Jermann

models on the behavior of the default-risk-free ex-ante real term structure and term premia. Both

models are solved numerically and calibrated to UK aggregate and household data, and the predic-

tions are compared to data on ex-ante real interest rates constructed from UK index-linked data.

6

LeRoy (1973), Rubinstein (1976), and Breeden (1979) were important early contributions to the literature on

dynamic general equilibrium asset pricing models.

7

The work by Alvarez and Jermann builds on contributions by Kehoe and Levine (1993), Kocherlakota (1996),

and Luttmer (1996). See also Aiyagari (1994) and Zhang (1997) for studies on endogenous solvency constraints.

4

While both models produce time-varying risk premia, only the model with limited risk-sharing

can generate enough variation in the term premia to account for the rejections of the expectations

hypothesis.

8

It should also be emphasized that in both the Lucas and Alvarez-Jermann economies the object

of the study is the term structure of default-risk-free real interest rates. Even though in the Alvarez-

Jermann economy risk-sharing is limited by the risk of default, the instruments that are traded

in state-contingent markets are default-risk-free.

9

Therefore, it is natural to compare the Lucas

and Alvarez-Jermann term structures with the term structure of real bonds issued by the UK

government. At least in principle, these bonds are default-risk free. Hence, the thesis of this paper

is not that default risk directly explains the term premium in the real term structure, but rather

that default risk limits risk-sharing in such a way as to produce a “realistic” term premium.

The rest of the paper is organized as follows. Section 2 presents the data on nominal and

index-linked bonds. Section 3 presents the basic features of the Lucas and Alvarez-Jermann models.

Section 4 calibrates the models given the data on the risk-free rate, aggregate consumption, business

cycles, and invidual incomes in the UK Section 5 presents the numerical results for both models

and compares the models’ behavior to the behavior of UK nominal and real term structures.

Section 6 summarizes the sensitivity analysis of the Alvarez-Jermann model. Section 7 concludes.

Appendix A explains how the models are numerically solved.

2 UK Index-Linked Bonds

The main complication in analyzing ex-ante real interest rates is that in most economies they

simply are unobservable. The most important exception is the UK market for index-linked debt.

It constitutes a signiﬁcant proportion of marketable government debt, and its daily turnover is by

far the highest in the world. The UK market for index-linked debt was started in 1981, and by

March 1994 it accounted approximately 15% of outstanding issues by market value.

10

Unfortunately, UK index-linked bonds do not provide a correct measure of ex-ante real interest

rates. The reason is that the nominal amounts paid by index-linked bonds do not fully compensate

the holders for inﬂation; the indexation operates with a lag. Both coupon and principal payments

are linked to the level of the RPI (Retail Price Index) published in the month seven months prior

to the payment date. In addition, the RPI number relates to a speciﬁc day in the previous month,

so that the eﬀective lag is approximately eight months. The motivation behind this procedure is

that it always allows the nominal value of the next coupon payment to be known, and nominal

8

There are several other interesting models that oﬀer at least partial resolution of the equity premium and/or

risk-free rate puzzles. A few examples are Constantinides (1990), Constantinides and Duﬃe (1996), Bansal and

Coleman (1996), Campbell and Cochrane (1999), and Abel (1999). In principle, any of these models could be used

to study the Backus-Gregory-Zin “term premium puzzle” in UK data. Since the previous version of this paper was

completed, Wachter (2002) and Buraschi and Jiltsov (2003) have successfully applied habit formation to the “term

premium puzzle.” Duﬀee (2002) and Dai and Singleton (2002) study the issue in the context of aﬃne term structure

models.

9

In the Alvarez-Jermann economy, the agents have an incentive to default on their contracts if honoring their

contracts would leave them worse oﬀ than they would be in autarky. Since everyone knows this and there is no

private information, nobody is willing to lend more than the debtor is willing to pay back. Therefore, in equilibrium

nobody defaults but risk-sharing is limited by the risk of default.

10

See Brown and Schaefer (1996) for more details.

5

accrued interest can always be calculated.

Diﬀerent authors have made diﬀerent assumptions in order to overcome the “indexation lag

problem.” Woodard (1990), Deacon and Derry (1994) and Brown and Schaefer (1994) impose the

Fisher hypothesis: nominal yields move one for one with changes in inﬂation, which means that

there is no “inﬂation risk premia.” On the other hand, Kandel, Ofer, and Sarig (1996) and Barr and

Campbell (1997) assume that diﬀerent versions of the expectations hypothesis hold. Finally, using

the properties of stochastic discount factors, Evans (1998) isolates the “indexation lag problem”

to a conditional covariance term between the future (maturity less the inﬂation lag) inﬂation and

nominal bond prices. He then estimates this term using a VAR model and derives the real interest

rates.

Evans also tests for the versions of expectations hypothesis used by Kandel, Ofer, and Sarig (1996)

and Barr and Campbell (1997) and rejects both versions at the 1 percent signiﬁcance level.

11

Since

this paper is mainly concerned on the expectations hypothesis, the methodology of Evans seemed

best suited for my purposes. The data discussed below was provided by him.

Evans’ (1998) estimates of the term structure of real interest rates in the UK from January 1984

until August 1995 reveal the following stylized facts:

1. While the average term structure is upward sloping, the average real term structure is down-

ward sloping.

2. While the long-term of nominal term structure is quite volatile, the long-term of the real term

structure appears to be highly stable.

12

However, the reverse is true for the short-term: the

short-term of the real term structure is highly volatile compared with that of the nominal

term structure.

3. Both the short-term and and long-term of the real term structure seem to be less autocorre-

lated than the corresponding maturities in the real term structure.

4. Shapes of the real term structure are relatively simple compared with the nominal term

structure.

Similar observations have been obtained by Woodward (1990) and Brown and Schaefer (1994,

1996) who made diﬀerent assumptions to overcome the “indexation lag problem” and estimated

the yield curve using diﬀerent methods than Evans used. With the exception of the average shape

of real term structure, the results in Brown and Schaefer are consistent with results presented

here. Brown and Schaefer (1994, 1996) estimated the term structure of real interest rates to be

upward-sloping on average. Following the suggestion by an anonymous referee, I investigated this

issue using real rates estimated from the nominal data as in den Haan (1995) and Chapman (1997).

These data also produce on average downward-sloping term structures. More details are available

upon request.

13

11

He also rejects the Fisher hypothesis. He ﬁnds that expected inﬂation is negatively correlated with real yields.

12

This observation is very interesting since Dybvig, Roll, and Ross (1996) showed that under very general conditions

the limiting forward rate, if it exists, can never fall. In aﬃne-yield models, such as Cox, Ingersoll, and Ross (1985),

this means that the long-term of the term structure should converge, and these models have been criticized on the

grounds that the long-term of the nominal term structure does not appear to be stable.

13

One interesting observation should be mentioned. If one calculates the average term structure using only the

6

3 The Lucas and Alvarez-Jermann Models

3.1 The Environment

Alvarez and Jermann (1996, 2000, 2001) consider a pure exchange economy with two agents.

14

Agents have identical preferences represented by time-separable expected discounted utility. Their

endowments follow a ﬁnite-state ﬁrst-order Markov process. The diﬀerence between the Alvarez-

Jermann economy and the Lucas economy is that in the former agents cannot commit to their

contracts. The agents have an incentive to default on their contracts if honoring their contracts

would leave them worse oﬀ than they would be in autarky. Since everyone knows this and there

is no private information, nobody is willing to lend more than the debtor is willing to pay back.

Therefore, in equilibrium nobody defaults but risk-sharing is limited by the risk of default.

In the planning problem, the possibility of default is prevented by participation constraints. The

economy can be decentralized through complete asset markets where the positions that the agents

can take are endogenously restricted by person, state, and time-dependent solvency constraints.

From now on, the discussion concentrates on the Alvarez-Jermann economy, the Lucas economy

being a special case of the Alvarez-Jermann economy where participation or solvency constraints

do not ever bind.

Let i = 1, 2 denote each agent and {z

t

} denote a ﬁnite-state Markov process z ∈ Z =

{z

1

, . . . , z

N

} with transition matrix Π. z

t

determines the aggregate endowment e

t

, and the in-

dividual endowments, e

i

t

, i = 1, 2 as follows:

e

t+1

= λ(z

t+1

)e

t

and e

i

t

= α

i

(z

t

)e

t

for i = 1, 2,

where λ(z

t+1

) is the growth rate of aggregate endowment between t and t + 1 when the state in

period t + 1 is z

t+1

and α

i

(z

t

) determines agent i’s share of the aggregate endowment when the

state in period t is z

t

.

Let z

t

= (z

1

, . . . , z

t

) denote the history of z up to time t. The matrix Π determines the

conditional probabilities for all histories π(z

t

|z

0

). Households care only about their consumption

streams, {c} = {c

t

(z

t

) : ∀t ≥ 0, z

t

∈ Z

t

}, and rank them by discounted expected utility,

U(c)(z

t

) =

∞

¸

j=0

¸

z

t+j

∈Z

t+j

β

j

u(c

t+j

(z

t+j

))π(z

t+j

|z

t

),

where β ∈ (0, 1) is a constant discount factor and the one-period utility function is of the constant

relative risk aversion type

u(c) =

c

1−γ

1 −γ

, 1 < γ < ∞,

where γ is the agents’ constant coeﬃcient of relative risk-aversion.

In the planning problem, the participation constraints force the allocations to be such that

under no history will the expected utility be lower than that in autarky

U(c

i

)(z

t

) ≥ U(e

i

)(z

t

) ∀t ≥ 0, z

t

∈ Z

t

, i = 1, 2. (1)

data after the UK left the European Exchange Rate Mechanism in October 1992, the term structure has been

upward-sloping on average regardless whether one uses Evans data or real rates estimated from nominal data.

14

Alvarez and Jermann (2000) consider the more general case with I ≥ 2 agents.

7

In other words, in no time period and in no state of the world can either agent’s expected discounted

utility be less than what the agent would obtain in autarky. If this were the case, the agent would

not commit to his contract and would choose to default. The punishment for default would be

permanent exclusion from the risk-sharing arrangement. The motivation for autarky constraints

of the form (1) is quite natural. In the real world, it is sometimes diﬃcult to make a debtor pay.

People do not always keep their promises and debt collection can be costly and possibly useless.

One can argue that the model is unrealistic in two respects. On the one hand, the punishment is

too severe; typically an agent who declares bankruptcy can return to the risk-sharing arrangement

after a ﬁnite number of periods. On the other hand, the agent who reverts to autarky can keep

his endowment stream; there is no conﬁscation of assets as punishment for default. Trying to

relax both of these unrealistic features of the model probably would not change the asset pricing

implications too much. Relaxing the ﬁrst assumption would reduce risk-sharing, since punishment

for default would not be so serious. Relaxing the second assumption would increase risk-sharing,

since punishment for default would be more serious.

Note that the model abstracts completely from game-theoretic issues related to bargaining and

renegotiation. Punishing one agent by forever excluding him from the risk-sharing arrangement

makes the punishing agent much worse oﬀ than with ﬁnite punishment. The justiﬁcation is that as

in Alvarez and Jermann (2000), there are actually many more agents than just one and each agent

has a very small weight in the collective bargaining problem. The model is easiest to solve when

there are only two agents. Section 3.3 shows that adding more agents need not change the asset

pricing implications of the model.

3.2 Constrained Optimal Allocations

The constrained optimal allocations are deﬁned as processes, {c

i

}, i = 1, 2, that maximize agent 1’s

expected utility subject to feasibility and participation constraints at every date and every history,

given agent 2’s initial promised expected utility. The recursive formulation of the constrained

optimal allocations is given by the functional equation

TV (w, z, e) = max

c

1

,c

2

,{w

z

}

u(c

1

) +β

¸

z

∈Z

V (w

z

, z

, e

)π(z

|z)

subject to

c

1

+c

2

≤ e

u(c

2

) +β

¸

z

∈Z

w

z

π(z

|z) ≥ w

V (w

z

, z

, e

) ≥ U

1

(z

, e

) ∀z

∈ Z (2)

w

z

≥ U

2

(z

, e

) ∀z

∈ Z, (3)

where primes denote next-period values, V (w, z, e) is agent 1’s value function, w is agent 2’s

promised utility this period, w

z

is agent 2’s promised utility when the next-period state of the

world is z

**, and (2) and (3) are the participation constraints for agent 1 and agent 2, respectively.
**

The second welfare theorem holds for the economy. Hence, one can solve for the allocations by

solving the planning problem and read the prices oﬀ the ﬁrst-order conditions of the competitive

equilibrium.

8

3.3 Asset Pricing

To analyze asset prices, the economy must be decentralized. The markets are complete, so that

in every state z there are one-period Arrow securities for each next period state of nature, z

.

However, the solvency constraints prevent agents from holding so much debt in any state that they

would like to default on their debt contracts. The solvency constraints aﬀect each state diﬀerently,

since the relative value of autarky compared to honoring the contract is diﬀerent in every state.

Let q(z

**, z) be the price of an Arrow security that pays one unit of consumption good at the
**

beginning of the next period when the next-period state is z

**and the current-period state is z.
**

Agent i’s holdings of this asset are denoted by a

i

z

. Finally, let B

i

(z

**, z) be the minimum position
**

agent i can take in the asset that pays oﬀ when the next-period state is z

**and the current-period
**

state is z. Hence,

Deﬁnition 1. The household’s problem given the current state (a, z) is to maximize the expected

utility

H

i

(a, z) = max

c,{a

z

}

z

∈Z

u(c) +β

¸

z

∈Z

H

i

(a

z

, z

)π(z

|z)

subject to solvency and budget constraints

a

z

≥ B

i

(z

, z) ∀z

∈ Z

¸

z

∈Z

q(z

, z)a

z

+c ≤ a +e

i

(z).

The equilibrium can now be deﬁned.

Deﬁnition 2. The equilibrium is a set of solvency constraints {B

i

t

}, prices {q

t

}, and alloca-

tions {c

i

t

, a

i

t+1

} such that

1. Taking the constraints and prices as given, the allocations solve both households’ optimization

problems.

2. Markets clear:

a

1

(z

t+1

) +a

2

(z

t+1

) = 0 ∀t ≥ 0, ∀z

t+1

∈ Z

t+1

.

3. When the solvency constraints are binding, the continuation utility equals autarchy utility:

H

i

(B

i

t+1

(z

t+1

), z

t+1

) = U

i

(e

i

)(z

t+1

) ∀t ≥ 0, ∀z

t+1

∈ Z

t+1

. (4)

Condition (4) means that solvency constraints are endogenously generated. This ensures that

the solvency constraints prevent default and it allows as much insurance as possible. A binding

solvency constraint means that the agent is indiﬀerent between defaulting and staying in the risk-

sharing regime. A slack solvency constraint means that the agent’s expected discounted utility

is strictly higher than what he would obtain in autarky. Hence, the model provides endogenous

justiﬁcation for debt, solvency, short-selling, and other exogenous constraints that are commonly

used in the literature on incomplete markets.

9

From the perspective of asset pricing, the crucial point of the model is that the prices of Arrow

securities are given by the maximum of the marginal rates of substitution for agents 1 and 2.

15

That is, if the current state is z

t

, the price of an Arrow security that pays one unit of consumption

good at the beginning of the next period if the next-period state is z

t+1

is given by

q(z

t+1

, z

t

) = max

i=1,2

β

u

(c

i

t+1

)

u

(c

i

t

)

π(z

t+1

|z

t

). (5)

The economic intuition is that the unconstrained agent in the economy does the pricing. Since

B(z

t+1

) gives the minimum amount of an asset one can buy, the constrained agent would like to sell

that asset and hence his marginal valuation of the asset is lower. In other words, the constrained

agent has an internal interest rate that is higher than the market rate. Therefore, he would like

to borrow more than is feasible, to keep the autarky constraints satisﬁed. In the full risk-sharing

regime (Lucas economy), the two marginal rates of substitution are equalized.

Notice that one can introduce as many new agents into the economy as desired without changing

the asset pricing implications, provided that the new agents’ marginal valuations are always less

than or equal to market valuations. A corollary of this is that each new agent whose income process

is perfectly correlated with aggregate income has no eﬀect on asset prices.

In addition, let q(z

t+j

, z

t

) be the price of an Arrow security from state z

t

to state z

t+j

, which

is given by

q(z

t+j

, z

t

) =

t+j−1

¸

k=t

q(z

k+1

, z

k

), (6)

and let q(z

k+1

, z

k

) be given by (5).

Let m

t+1

denote the real stochastic discount factor

m

t+1

≡ max

i=1,2

β

u

(c

i,t+1

)

u

(c

i,t

)

.

The price of an n-period zero-coupon bond is given by

p

b

n,t

=

¸

z

t+n

∈Z

t+n

q(z

t+n

, z

t

) = E

t

¸

n

¸

j=1

m

t+j

¸

. (7)

Using (5), (6), and the equation above, note that

p

b

n,t

=

¸

z

t+1

∈Z

t+1

max

i=1,2

β

u

(c

i,t+1

)

u

(c

i,t

)

p

b

n−1,t+1

π(z

t+1

|z

t

) = E

t

[m

t+1

p

b

n−1,t+1

]. (8)

The bond prices are invariant with respect to time, and hence equation (8) gives a recursive formula

for pricing zero-coupon bonds of any maturity.

Forward prices are deﬁned by

p

f

n,t

=

p

b

n+1,t

p

b

n,t

,

15

This result was also derived by Cochrane and Hansen (1992) and Luttmer (1996).

10

and the above prices are related to interest rates (or yields) by

f

n,t

= −log(p

f

n,t

) and r

n,t

= −(1/n) log(p

b

n,t

). (9)

To deﬁne the risk premium as in Sargent (1987), write (8) for a two-period bond using the

conditional expectation operator and its properties:

p

b

2,t

= E

t

[m

t+1

p

b

1,t+1

]

= E

t

[m

t+1

]E

t

[p

b

1,t+1

] + cov

t

[m

t+1

, p

b

1,t+1

]

= p

b

1,t

E

t

[p

b

1,t+1

] + cov

t

[m

t+1

, p

b

1,t+1

],

which implies that

p

f

1,t

=

p

b

2,t

p

b

1,t

= E

t

[p

b

1,t+1

] + cov

t

¸

m

t+1

,

p

b

1,t+1

p

b

1,t

. (10)

Since the conditional covariance term is zero for risk-neutral investors, I will call it the risk premium

for the one-period forward contract, rp

1,t

, given by

rp

1,t

≡ cov

t

¸

m

t+1

,

p

b

1,t+1

p

b

1,t

= p

f

1,t

−E

t

[p

b

1,t+1

],

and similarly rp

n,t

is the risk premium for the n-period forward contract:

rp

n,t

≡ cov

t

¸

n

¸

j=1

m

t+j

,

p

b

1,t+n

p

b

1,t

¸

= p

f

n,t

−E

t

[p

b

1,t+n

].

In addition, I will call a diﬀerence between the one-period forward rate and the expected value of

one-period interest rate next period the term premium for the one-period forward contract, tp

1,t

:

tp

1,t

≡ f

1,t

−E

t

[r

1,t+1

]

and similarly tp

n,t

is the term premium for the n-period forward contract:

tp

n,t

≡ f

n,t

−E

t

[r

1,t+n

].

4 Calibration

4.1 Free Parameters

In the spirit of Backus, Gregory, and Zin (1989), I will solve the model for the simplest possible

case that produces nonconstant interest rates and has income heterogeneity.

16

This is obtained by

introducing uncertainty in the growth rate of aggregate endowment while treating the agents in a

symmetric fashion. Similar techniques can be applied for more complicated cases.

16

Without income heterogeneity, the Alvarez-Jermann and Lucas models would be identical.

11

In particular, there will be three exogenous states: z

hl

, z

e

, and z

lh

. The states z

hl

and z

lh

are

associated with a recession, and following Mankiw (1986) and Constantinides and Duﬃe (1996),

the recessions are associated with a widening of inequality in earnings.

17

During the expansion

both agents have the same endowment z

e

. The states are ordered so that

α

1

(z

hl

) = α

2

(z

lh

) ≥ α

1

(z

e

) = α

2

(z

e

) ≥ α

1

(z

lh

) = α

2

(z

hl

)

and the transition matrix Π preserves symmetry between the agents.

Thus, there are ﬁve free parameters associated with aggregate and individual endowment. These

are α

1

hl

for individual incomes,

α

1

=

α

1

hl

0.5

1 −α

1

hl

¸

¸

and α

2

=

1 −α

1

hl

0.5

α

1

hl

¸

¸

,

λ

e

and λ

r

for the growth rates of aggregate endowment,

λ =

λ

r

λ

e

λ

r

¸

¸

,

and π

r

and π

e

for the transition matrix

Π =

π

r

1 −π

r

0

(1 −π

e

)/2 π

e

(1 −π

e

)/2

0 1 −π

r

π

r

¸

¸

.

The order of calibration is as follows. First, the transition matrix for the aggregate states can

be expressed as a function of the fraction of time spent in the expansion state, π, and the ﬁrst-order

autocorrelation of aggregate consumption, θ:

18

Π =

¸

(1 −θ)π +θ (1 −θ)(1 −π)

(1 −θ)π (1 −θ)(1 −π) +θ

.

Clearly,

π =

1 −π

r

2 −(π

e

+π

r

)

θ = π

e

+π

r

−1.

Next, given the transition matrix for the aggregate states, λ

e

and λ

r

determine the average growth

rate of aggregate consumption and its standard deviation. Finally, α

1

hl

determines the standard

deviation of individual income. Note that with this parameterization one cannot pin down the

persistence of individual income. An alternative would be to allow individual incomes to take

diﬀerent values during expansions. I discuss in Section 4.2.2 why I chose not to do so.

17

Storesletten, Telmer, and Yaron (1999) document this for U.S. data.

18

See Barton, David, and Fix (1962).

12

In addition, there are two free parameters, β and γ, associated with preferences. Section 4.2

explains the ﬁve equations that determine endowment parameters, and Section 4.3 shows how the

preference parameters are pinned down to match the ﬁrst and second moments of the risk-free rate

in the UK Appendix A gives details on how the model is solved numerically, and Appendix B of

Sepp¨ al¨ a (2000) shows that the main results are highly robust with respect to measurement error

in moment conditions.

4.2 Aggregate and Individual Endowment

4.2.1 Aggregate Consumption and Business Cycles

The ﬁrst step is to calibrate the law of motion for the aggregate endowment so that it matches

a few facts about aggregate consumption (in the model, aggregate endowment equals aggregate

consumption) and business cycles. Campbell (1998) reports that in the annual UK data for 1891–

1995, the average growth rate of aggregate consumption is 1.443%, the standard deviation of the

growth rate of aggregate consumption is 2.898%, and the ﬁrst-order autocorrelation is 0.281.

Since the UK does not publish oﬃcial deﬁnitions of expansions and recessions, I followed Chap-

man (1997), who used the following deﬁnitions in his study of the cyclical properties of U.S. real

term structure. Business cycle expansions are deﬁned as at least two consecutive quarters of posi-

tive growth

19

in a three quarter equally-weighted, centered moving average of the real GDP/capita

(“output”). Cycle contractions are at least two consecutive quarters of negative growth in the

moving average of output. A peak is the last quarter prior to the beginning of a contraction, and a

trough is the last quarter prior to the beginning of a expansion. Business cycles thusly deﬁned are

presented in Figure 1 for the UK from the ﬁrst quarter of 1957 until the last quarter of 1997. The

quarterly observations on the GDP at 1990 prices and annual observations on the population were

obtained from the CD-ROM July 1999 version of the International Monetary Fund’s International

Financial Statistics. The quarterly population series were constructed by assuming that population

grows at constant rate within the year and that the original annual data are as at December 31

of each year. According to the deﬁnitions, expansions are 3.8824 times more likely to occur than

recessions in the UK from the ﬁrst quarter of 1957 until the last quarter of 1997.

4.2.2 Individual Endowment

The next step is to calibrate the process for the individual endowments. As mentioned above, the

current speciﬁcation only allows one to match the standard deviation of individual income. An

alternative would be to allow individual incomes to take diﬀerent values also during expansions.

Unfortunately, in the UK there are only two data sets that provide information on household income.

The ﬁrst is the Family Expenditure Survey (FES), which covers the years 1968–1992 and the second

is the British Household Panel Survey (BHPS), which covers the period September 1990 to January

1998. Since I am interested in household income heterogeneity for particular households and the

FES is a survey, I chose to use the BHPS. This approach is similar to that taken in previous studies,

such as Heaton and Lucas (1996) and Storesletten, Telmer, and Yaron (1999), which calibrate asset

19

The growth (rate) is the diﬀerence in the logarithm of the series.

13

Growth Rate of Real GDP/Capita

1957 1962 1967 1972 1977 1982 1987 1992 1997

-3

-2

-1

0

1

2

3

4

5

Figure 1: The growth rate of real GDP/capita in the UK 1957:1–1997:4 and its moving average

(quarterly observations).

pricing models to the U.S. data on households.

20

The BHPS provides a panel of monthly observations of individual and household income and

other variables from September 1990 until January 1998 for more than 5,000 households, giving a

total of approximately 10,000 individuals.

21

I use the subset of the panel that has reported positive

income in every year since 1991. This gives a total of 2,391 individuals in the panel. For households

with more than one member with reported income, I constructed the individual income as the total

household income divided by the number of the members of the household with reported income.

Following Heaton and Lucas (1996), the individual annual income dynamics are assumed to follow

an AR(1)-process:

log(η

i

t

) = ¯ η

i

+ρ

i

log(η

i

t−1

) +

i

t

, (11)

where η

i

t

= e

i

t

/

¸

n

i=1

e

i

t

. This provides a close connection for the standard deviation of

i

t

in the

data with the standard deviation of α’s in the model. Note that it also imposes a cointegration

relationship between aggregate and individual income. Unfortunately, given the length of data it

seems unlikely that one could test whether this relationship is reasonable.

20

Heaton and Lucas (1996) and Storesletten, Telmer, and Yaron (1999) also provide examples on how to estimate

the standard deviation of individual income. To implement the Storesletten, Telmer, and Yaron method, one needs

to know how the cross-sectional variance in household income varies over business cycles. The BHPS was started

only in September 1990 when the UK was in depression, and the depression ended in the ﬁrst quarter of 1992. Thus,

the data does not cover a full cycle and it has only one data point for the recession. Therefore, I concentrate on the

simplest possible speciﬁcation (the least free parameters) that would provide diﬀerent results from the Lucas model.

21

For more details on the BHPS, see Taylor (1998).

14

Table 1: Cross-sectional means and standard deviations of the coeﬃcient estimates in the regres-

sion (11). Data: British Household Panel Survey.

Coeﬃcient Cross-Sectional Mean Cross-Sectional Standard Deviation

¯ η

i

−6.4331 5.2376

ρ

i

0.2035 0.6315

σ

i

0.2830 0.2853

Table 2: Cross-sectional means and standard deviations of the coeﬃcient estimates in the regres-

sion (11) obtained by Heaton and Lucas (1996). Data: Panel Study of Income Dynamics

Coeﬃcient Cross-Sectional Mean Cross-Sectional Standard Deviation

¯ η

i

−3.354 2.413

ρ

i

0.529 0.332

σ

i

0.251 0.131

Table 1 reports cross-sectional means and standard deviations of the coeﬃcient estimates in the

regression (11). The relevant numbers are the ﬁrst-order autocorrelation coeﬃcient, ρ

i

, and the

standard deviation of the error term, σ

i

=

E[(

i

t

)

2

]. Heaton and Lucas (1996) estimate the same

coeﬃcients using a sample of 860 U.S. households in the Panel Study of Income Dynamics (PSID)

that have annual incomes for 1969 to 1984. Table 2 reports cross-sectional means and standard

deviations of the coeﬃcient estimates obtained by Heaton and Lucas. The estimated autocorrelation

coeﬃcient in the BHPS, 0.2035, is signiﬁcantly smaller than that in the PSID, 0.529. On the other

hand, the standard deviation of the error term in the BHPS, 0.2830, is roughly the same as that in

the PSID, 0.251. Storesletten, Telmer, and Yaron (1999) obtain a higher number for the standard

deviation of individual income.

22

The question of the persistence of individual income is, unfortunately, an unresolved issue.

Heaton and Lucas (1996) estimate a relatively low number while Storesletten, Telmer, and Yaron

(1999) obtain estimates much closer to one. Finally, Blundell and Preston (1998) impose the pres-

ence of a random walk component. Baker (1997) provides analysis that questions the assumption

of a unit root. In order to reasonably pin down the persistence of individual income one would

need to have very long panels of data, which are currently unavailable.

22

For previous estimates of the standard deviation of individual income using the UK data, see Meghir and White-

house (1996) and Blundell and Preston (1998). Both studies use the FES data. Blundell and Preston (1998), which

is closer to exercise here, decompose the variance in income into permanent and transitory components, and assume

is that the process is composed of a pure transitory component and a random walk. They show strong growth in

transitory inequality toward the end of this period, while young cohorts are shown to face signiﬁcantly higher levels

of permanent inequality. At the beginning of the sample, their estimates are close to mine, but at the end their

estimates are higher, as in Storesletten, Telmer, and Yaron (1999).

15

2

3

4

5

6

7

0.2

0.4

0.6

0.8

1

0

5

10

15

20

25

30

35

Standard Deviation of (Individual Consumption/Aggregate Consumption)

Risk aversion

Discount Factor

%

Figure 2: Standard deviation of (individual consumption/aggregate consumption) as a function of

the discount factor and coeﬃcient of relative risk aversion.

4.2.3 Calibrated Values

Solving the system of ﬁve unknowns in ﬁve equations leads to the endowment vectors and growth

rates

23

α

1

=

0.7706

0.5

0.2294

¸

¸

, α

2

=

0.2294

0.5

0.7706

¸

¸

, and λ =

0.9573

1.0291

0.9573

¸

¸

,

and the transition matrix

Π =

0.4283 0.5717 0

0.0736 0.8527 0.0736

0 0.5717 0.4283

¸

¸

.

4.3 Preferences

The next step is to match agents’ preference parameters, the discount factor β, and the coeﬃcient

of relative risk aversion, γ, to the key statistics of the asset market data. To illustrate the features

of the model, Figure 2 plots the standard deviation of individual consumption, std(log(

c

i

¸

j

c

j

)), as

a function of β and γ.

23

Due to highly nonlinear nature of the model, the results are quite sensitive to the parameter values. Therefore,

I will report all the parameter values with four decimal precision.

16

2

3

4

5

6

7

0.2

0.4

0.6

0.8

1

−50

0

50

100

150

200

250

300

Expected Riskfree Rate

Risk aversion

Discount Factor

%

Figure 3: Average risk-free rate as a func-

tion of the discount rate and risk aversion

in an Alvarez-Jermann economy.

2

3

4

5

6

7

0.2

0.4

0.6

0.8

1

−50

0

50

100

150

200

250

300

Expected Riskfree Rate

Risk aversion

Discount Factor

%

Figure 4: Average risk-free rate as a func-

tion of the discount rate and risk aversion

in a Lucas economy.

The ﬂat segment in the upper-left corner of the Figure corresponds to autarky and the ﬂat

segment in the lower-right corner to perfect risk-sharing. The parameter values which are interesting

for the purpose of asset pricing are those that generate allocations between these two extremes.

To accomplish this, one has to choose either relatively low risk aversion and a relatively high

discount factor or relatively high risk aversion and a relatively low discount factor. In Section 5, I

report numerical results only for one pair of coeﬃcients of risk aversion and discount factor, and

in Appendix B of Sepp¨ al¨ a (2000) I show that these results are very robust to measurement error

in risk-free rates, aggregate consumption, business cycles, and individual income. In addition, I

show how increasing the coeﬃcient of risk-aversion, the discount factor, or the standard deviation

of individual income moves the results in the direction of more risk-sharing (the Lucas model).

Figures 3–6 present the average and standard deviations of the risk-free rate in the Alvarez-

Jermann and Lucas economies. Note that in the Alvarez-Jermann economy in autarky no trade is

allowed because one of the agents would default on his contract, and hence all the statistics have

been set to zero. The shapes in the Lucas economy are relative easy to understand using log-linear

approximations, as was done in Section 2.

Recall that the price of an n-period zero-coupon bond in a Lucas economy is given by

p

b

n,t

= β

n

E

t

¸

e

t+n

e

t

−γ

¸

.

17

2

3

4

5

6

7

0.2

0.4

0.6

0.8

1

0

20

40

60

80

100

Standard Deviation of Riskfree Rate

Risk aversion

Discount Factor

%

Figure 5: Standard deviation of the risk-

free rate as a function of the discount rate

and risk aversion in an Alvarez-Jermann

economy.

2

3

4

5

6

7

0.2

0.4

0.6

0.8

1

0

20

40

60

80

100

Standard Deviation of Riskfree Rate

Risk aversion

Discount Factor

%

Figure 6: Standard deviation of the risk-

free rate as a function of the discount rate

and risk aversion in a Lucas economy.

Taking a log-linear approximation of the n-period interest rate,

24

one obtains

r

n,t

= −log(β) +

γ

n

E

t

¸

log

e

t+n

e

t

−

γ

2

2n

var

t

¸

log

e

t+n

e

t

, (12)

E[r

n,t

] = −log(β) +

γ

n

E

¸

log

e

t+n

e

t

−

γ

2

2n

var

¸

log

e

t+n

e

t

. (13)

Therefore, the interest rate decreases exponentially in the discount factor and increases slowly in

the risk-aversion coeﬃcient.

In the Alvarez-Jermann model, the relationship between parameter values and interest rates is

more complicated because the interest rates depend not only on aggregate endowment but also on

the consumption share of the unconstrained agent:

−exp(r

t

) = p

b

t

= βE

t

¸

e

t+1

e

t

−γ

max

i=1,2

ˆ c

i

t+1

ˆ c

i

t

−γ

¸¸

, (14)

where ˆ c

i

t

is agent i’s consumption share in period t. When the discount factor is lowered, the

incentive to participate in the risk-sharing arrangement is reduced. This leads to an increase in

variability in consumption shares and hence an increase in the “max” operator, thereby increasing

bond prices and reducing one-period interest rates.

Campbell (1998) reports that in annual UK data for 1891–1995, the average real risk-free rate

was 1.198 and its standard deviation was 5.446. In the Alvarez-Jermann model, matching these

values leads to β = 0.3378 and γ = 3.514. β = 0.3378 is considerably less than what either

24

Again, the approximation is exact only if consumption growth has a log-normal distribution. See Campbell (1986).

18

complete markets or incomplete markets literature typically use. The reason is that, in order to be

able to match asset market data, one has to reduce risk-sharing. In the Alvarez-Jermann endowment

economy, an incentive to participate in risk-sharing is very high so that only by lowering the discount

factor can autarky become tempting. In addition, note that the persistence of individual income

is lower than the values estimated by either Heaton and Lucas (1996) or Storesletten, Telmer, and

Yaron (1999). The quantitative results in Alvarez and Jermann (2001) indicate that increasing the

persistence does not change the asset pricing implications, provided that one is allowed to increase

the discount factor. That is, the more persistent the individual income, the more tempting is

default (the agents would like to accumulate state-contingent debt during bad times and default

during good times) and hence the risk-sharing is reduced for more patient agents.

Obviously, in the Lucas economy risk-sharing is never limited, so it is not possible to match both

the average risk-free rate and the standard deviation simultaneously. Since the standard deviation

is more important for explaining the behavior of the term premium, I chose the discount factor and

risk-aversion coeﬃcients in the Lucas economy

(0.99, 6.1149) = arg min

β∈(0,0.99],γ∈(1,100)

{|E[r(β, γ)] −1.198| s.t. std[r(β, γ)] = 5.446} .

Table 3 summarizes the main statistics for the models, given the above discount factor and

risk aversion values, and in the data. Notice that, in the Alvarez-Jermann economy, the standard

deviation of individual consumption is close to the standard deviation of individual income. In

other words, the allocations are close to autarky allocations. In Appendix B of Sepp¨ al¨ a, I show

that this result is also very robust. In order to be able to match the basic asset pricing data, one has

to reduce risk-sharing considerably from the full risk-sharing benchmark. It is diﬃcult to say how

reasonable this result is: In a recent paper, Brav, Constantinides, and Geczy (1999) conclude that

the observation error in the consumption data makes it impossible to test the complete consumption

insurance assumption against the assumption of incomplete consumption insurance.

The mechanism for limited risk-sharing works as follows. Both agents are oﬀ the solvency

constraint when the current state is the same as the previous state. The average duration of

expansion is about six years and the average duration of depression is about two years. Only when

expansion changes to recession or vice versa is somebody constrained. When expansion changes

to recession, the agent who got the lower share during the recession has the higher consumption

growth rate. Hence, he would like to default. When recession changes to expansion, the agent

who got the lower share during the recession has the higher consumption growth and would like

to default.

25

The next section shows how this mechanism translates to the behavior of the term

structure of interest rates.

5 Results

5.1 The Term Structure of Interest Rates

Figures 7–10 present the interest rates for maturities of 1 to 30 years and the forward rates of 1

to 30-year forward contracts during expansions and recessions in the Alvarez-Jermann and Lucas

25

Clearly, with only two agents, one cannot interpret the agents literally, but rather how the fraction of population

is aﬀected by solvency constraints over the business cycle.

19

Table 3: Selected statistics for the Lucas model when β = 0.99 and γ = 6.1149, for the Alvarez-

Jermann model when β = 0.3378 and γ = 3.514, and for the data.

Lucas Model Alvarez-Jermann Model Data

E[r] (%) 7.93 1.198 1.198

std[r] (%) 5.446 5.446 5.446

E(∆c) (%) 1.443 1.443 1.443

std[∆c] (%) 2.898 2.898 2.898

corr[∆c] 0.281 0.281 0.281

Pr(exp.)/ Pr(rec.) 3.882 3.882 3.882

std(log(c

i

/

¸

j

(c

j

))) (%) 0.0 26.71 —

corr(log(c

i

/

¸

j

(c

j

)) 1.0 0.4336 —

std(log(e

i

/

¸

j

(e

j

))) (%) 28.3 28.3 28.3

corr(log(e

i

/

¸

j

(e

j

)) 0.4193 0.4193 0.2035

economies. A few things are worth noting from the Figures. First, both models produce both

upward and downward-sloping term structures. However, the models’ cyclical behavior is exactly

the opposite. In the Alvarez-Jermann model, the term structure of interest rates is downward-

sloping in recessions and upward-sloping during expansions. In the Lucas model, the term structure

of interest rates is upward-sloping in recessions and downward-sloping during expansions. Also, in

both models upward-sloping term structures are always uniformly below downward-sloping term

structures.

26

The cyclical behavior of the term structure is of particular interest since empirical

and theoretical results from previous studies have been contradictory.

Fama (1990) reports that

“A stylized fact about the term structure is that interest rates are pro-cyclical. (. . .)

[I]n every business cycle of the 1952–1988 period the one-year spot rate is lower at the

business trough than at the preceding or following peak. (. . .) Another stylized fact is

that long rates rise less than short rates during business expansions and fall less during

contractions. Thus spreads of long-term over short-term yields are counter-cyclical.

(. . .) [I]n every business cycle of the 1952–1988 period the ﬁve-year yield spread (the

ﬁve-year yield less the one-year spot rate) is higher at the business trough than at the

preceding or following peak.”

Notice that this statement applies to the term structure of nominal interest rates. On the other

hand, Donaldson, Johnsen, and Mehra (1990) report that in a stochastic growth model with full

depreciation the term structure of (ex-ante) real interest rates is rising at the top of the cycle and

falling at the bottom of the cycle. In addition, at the top of the cycle the term structure lies

uniformly below the term structure at the bottom of the cycle.

26

See Proposition 1 below for the explanation.

20

0 5 10 15 20 25 30

−2

0

2

4

6

8

10

12

Years

%

Interest Rates and Forward Rates (state 1)

Figure 7: Interest rates (solid line) and

forward rates in the Alvarez-Jermann

economy during recessions.

0 5 10 15 20 25 30

−3

−2

−1

0

1

2

3

4

5

6

7

Years

%

Interest Rates and Forward Rates (state 1)

Figure 8: Interest rates (solid line) and

forward rates in the Lucas economy during

recessions.

0 5 10 15 20 25 30

−2

0

2

4

6

8

10

Years

%

Interest Rates and Forward Rates (state 3)

Figure 9: Interest rates (solid line) and

forward rates in the Alvarez-Jermann

economy during expansions.

0 5 10 15 20 25 30

6

6.5

7

7.5

8

8.5

9

9.5

10

10.5

11

Years

%

Interest Rates and Forward Rates (state 3)

Figure 10: Interest rates (solid line) and

forward rates in the Lucas economy during

expansions.

21

In the Lucas economy, the cyclical behavior of the term structure will depend on the autocor-

relation of consumption growth. Recall equation (12):

r

n,t

= −log(β) +

γ

n

E

t

¸

log

e

t+n

e

t

−

γ

2

2n

var

t

¸

log

e

t+n

e

t

.

It implies that if consumption growth is positively autocorrelated, then a good shock today will

forecast good shocks in the future and consequently high interest rates for the near future. As the

maturity increases, however, the autocorrelation decreases and the maturity term in the denom-

inator starts to kick in reducing the interest rates. The interest rates move one-for-one with the

business cycle, exactly as Donaldson, Johnsen, and Mehra (1990) report.

27

In the Alvarez-Jermann economy, the prices of multiple-period bonds are determined by

−nexp(r

n,t

) = p

b

n,t

= β

n

E

t

¸

e

t+n

e

t

−γ

×max

i=1,2

ˆ c

i

t+1

ˆ c

i

t

−γ

¸

×· · · ×max

i=1,2

ˆ c

i

t+n

ˆ c

i

t+n−1

−γ

¸¸

.

However, to study the slope of the term structure it is suﬃcient to know whether interest rates

are procyclical or countercyclical. To see this, note that the following version of Dybvig, Roll, and

Ross (1996) holds in the Alvarez-Jermann economy.

Proposition 1. If the transition matrix for the economy is ergodic, then as n approaches inﬁnity,

the forward price, p

f

n,t

, converges to a constant.

Proof. Let m

ij

denote the pricing kernel between states i and j. In state i, the price of a one-period

bond is

¸

j

π

ij

m

ij

. This can be expressed as

¸

j

b

ij

, where b

ij

deﬁnes a matrix B. Similarly, the

price of a two-period bond is

¸

j

¸

k

π

ij

m

ij

π

jk

m

jk

or

¸

j

b

(2)

ij

, where b

(2)

ij

denotes the (i, j) element of B

2

. In general, the price of an n-period bond is

given by

¸

j

b

(n)

ij

. Since the transition matrix is ergodic, the Perron-Frobenius theorem guarantees

that the dominant eigenvalue of B is positive and that any positive vector operated on by powers

of B will eventually approach the associated eigenvector and grow at the rate of this eigenvalue.

Recall that the n-period forward price is the ratio of the price of an (n + 1)-period bond to the

price of an n-period bond. As n gets large, the ratio converges to the dominant eigenvalue of B

regardless of the current state.

An immediate corollary of Proposition 1 is that the risk premia will converge. Since the tran-

sition matrix is ergodic, the expected spot price, E

t

[p

b

1,t+n

], will converge and the limiting risk

premium is the diﬀerence between the limiting forward price and the limiting expected spot price.

Therefore, it is enough to study one-period bonds that are determined by equation (14)

−exp(r

t

) = p

b

t

= βE

t

¸

e

t+1

e

t

−γ

max

i=1,2

ˆ c

i

t+1

ˆ c

i

t

−γ

¸¸

.

27

However, the question of the cyclical behavior of the term structure is more complicated for production economies.

See den Haan (1995) and Vigneron (1999).

22

0 5 10 15 20 25 30

−5

−4

−3

−2

−1

0

1

2

3

4

5

Years

%

Risk and Term Premium (state 1)

Figure 11: Risk premium (solid line) and

term premium in the Alvarez-Jermann

economy during recessions.

0 5 10 15 20 25 30

−2

−1.5

−1

−0.5

0

0.5

1

1.5

2

Years

%

Risk and Term Premium (state 1)

Figure 12: Risk premium (solid line) and

term premium in the Lucas economy dur-

ing recessions.

Recall that the variability of consumption shares increases in recessions and that aggregate income

growth is positively autocorrelated. Hence, the two terms inside the conditional expectation work

opposite to each other. In the current parameterization, the dominant term is consumption het-

erogeneity. The greatest variability inside the “max” operator occurs when one moves from the

expansion state to the recession state. This means that bond prices increase during expansions or

that interest rates decline. Interest rates are countercyclical even though consumption growth is

positively autocorrelated.

28

Next, Figures 11–14 present the risk and term premia for maturities of 1 to 30 years during

expansions and recessions in the Alvarez-Jermann and Lucas economies. Note how the signs of

risk and term premia are opposite in the Lucas vs. Alvarez-Jermann economies. A positive sign

on the term premium means that forward rates tend to overpredict future interest rates and a

negative sign means that the term premium underpredicts. In addition, in the Lucas model the

term premium is very stable, and in the Alvarez-Jermann model, although the term premium is

also stable over the cycle for long maturities, it varies considerably and negatively with the level of

interest rates for short maturities. This result indicates that the Alvarez-Jermann model may be

useful in accounting for rejections of the expectations hypothesis.

Figures 15–20 present the mean and the standard deviation of the interest rates, forward rates,

28

In the British data presented in Section 2, the correlation between one-year real interest rate and the cyclical

component of real GDP/capita (obtained using a Hodrick-Prescott ﬁlter with a smoothing parameter of 1,600) is

−0.17. The correlation between yield spread (ﬁve-year yield minus one-year yield) and the cyclical component is 0.43.

The correlations between nominal data and the cyclical component are 0.19 and −0.48, respectively. In other words,

the Alvarez-Jermann model seems to be consistent with the real data and the Lucas model with the nominal data.

These estimates, unfortunately, are not very reliable, as the Britain had time to go through only one business cycle in

the sample. It is interesting to note that King and Watson (1996) obtained the same result for the cyclical behavior

of nominal and real interest rates in U.S. data. They obtained real interest rates by estimating expected inﬂation

using VAR.

23

0 5 10 15 20 25 30

−8

−6

−4

−2

0

2

4

6

8

Years

%

Risk and Term Premium (state 3)

Figure 13: Risk premium (solid line) and

term premium in the Alvarez-Jermann

economy during expansions.

0 5 10 15 20 25 30

−2

−1.5

−1

−0.5

0

0.5

1

1.5

2

Years

%

Risk and Term Premium (state 3)

Figure 14: Risk premium (solid line) and

term premium in the Lucas economy dur-

ing expansions.

risk, and term premia. The average term structure is upward-sloping in the Alvarez-Jermann

economy and downward-sloping in the Lucas economy. In both economies, the standard deviations

decrease with the maturity as in the UK nominal and real data. In the Lucas economy, the average

shape of the term structure is easy to explain. Recall equation (13):

E[r

n,t

] = −log(β) +

γ

n

E

¸

log

e

t+n

e

t

−

γ

2

2n

var

¸

log

e

t+n

e

t

.

If the E

¸

log

e

t+n

et

**≈ n¯ e, then the average shape of the term structure is determined by the
**

ratio of the variance term to maturity. If consumption growth is positively autocorrelated, the

variance term grows faster than the maturity since shocks in the growth rate are persistent.

29

In

the Alvarez-Jermann economy, unconditional expectations are more diﬃcult to obtain, but it is

suﬃcient to note that the term structure is upward-sloping during the expansions and the economy

is growing most of the time.

The relationship between the term premium and the shape of the term structure is as follows.

Yields can be expressed as averages of forward rates:

r

n,t

=

1

n

n−1

¸

j=0

f

j,t

.

Therefore, the average term structure can be expressed as

E[r

n,t

] −E[r

1,t

] =

1

n

n−1

¸

j=0

E[f

j,t

−E

t

[r

1,t+j

]] =

1

n

n−1

¸

j=0

tp

j,t

.

29

See den Haan (1995).

24

0 5 10 15 20 25 30

1

2

3

4

5

6

7

Years

%

Average Interest Rates and Forward Rates

Figure 15: Average interest rates (solid

line) and forward rates in the Alvarez-

Jermann economy.

0 5 10 15 20 25 30

6

6.2

6.4

6.6

6.8

7

7.2

7.4

7.6

7.8

8

Years

%

Average Interest Rates and Forward Rates

Figure 16: Average interest rates (solid

line) and forward rates in the Lucas econ-

omy.

0 5 10 15 20 25 30

−6

−4

−2

0

2

4

6

Years

%

Average Risk and Term Premium

Figure 17: Average risk premium (solid

line) and term premium in the Alvarez-

Jermann economy.

0 5 10 15 20 25 30

−2

−1.5

−1

−0.5

0

0.5

1

1.5

2

Years

%

Average Risk and Term Premium

Figure 18: Average risk premium (solid

line) and term premium in the Lucas econ-

omy.

25

0 5 10 15 20 25 30

0

1

2

3

4

5

6

Years

%

Standard Deviation of Interest Rates, Forward Rates, and Premia

Figure 19: Standard deviation of in-

terest rates (solid line), forward rates

(dashed line), risk premium (dash-dot

line), and term premium (dotted line) in

the Alvarez-Jermann economy.

0 5 10 15 20 25 30

0

1

2

3

4

5

6

Years

%

Standard Deviation of Interest Rates, Forward Rates, and Premia

Figure 20: Standard deviation of interest

rates (solid line), forward rates (dashed

line), risk premium (dash-dot line), and

term premium (dotted line) in the Lucas

economy.

Using the log-linear approximation and assuming homoscedastic errors, the term premium can be

expressed as

tp

n,t

= −

1

2

var

t

[log(p

b

1,t+n

)] −cov

t

[log(p

b

1,t+n

), log(m

t+n

)].

In the Lucas economy, this reduces to

tp

n,t

= −

γ

2

2

var

t

¸

E

t+n

log

e

t+n+1

e

t+n

−γ

2

cov

t

¸

E

t+n

log

e

t+n+1

e

t+n

, log(e

t+n

)

.

Suppose that ∆log(e

t

) = ρ∆log(e

t−1

) +

t

, where

t

∼ N(0, σ

2

). Then

tp

1,t

= −

γ

2

2

ρ

2

σ

2

−γ

2

ρσ

2

= −γ

2

σ

2

ρ

2

¸

1

2

+

1

ρ

,

which is less that zero, if ρ > 0. For example, in the British data

tp

1,t

= −(6.1149)

2

(0.02898)

2

(0.281)

2

¸

1

2

+

1

0.281

= −1%.

Alvarez and Jermann (2001) show that in the Alvarez-Jermann economy the sign of the term

premium depends on one-period ahead individual income variance, conditional on current aggregate

state or “heteroscedasticity ex-ante”:

σ

r

σ

e

≡

std[log(α

i

(z

t+1

))|λ

t

= λ

r

]

std[log(α

i

(z

t+1

))|λ

t

= λ

e

]

.

26

0 5 10 15 20 25 30

0.08

0.1

0.12

0.14

0.16

0.18

0.2

0.22

Years

Autocorrelation of Interest Rates

Figure 21: Autocorrelation of interest

rates in the Alvarez-Jermann Economy.

0 5 10 15 20 25 30

0

0.1

0.2

0.3

0.4

0.5

Years

Autocorrelation of Interest Rates

Figure 22: Autocorrelation of interest

rates in the Lucas Economy.

When

σr

σe

> 1, the term premium is negative and the average term structure is downward-sloping;

when

σr

σe

< 1, the term premium is positive and the average term structure is upward-sloping.

σr

σe

< 1 means that in expansions one expects more idiosyncratic risk in the future. From (14)

it follows that the max-term in the stochastic discount factor becomes more volatile and hence

bond prices are higher (interest rates lower) in expansions. Therefore, a positive term premium

is required to compensate bond holders. The next section studies whether this term premium is

volatile enough to account for rejections of the expectations hypothesis.

Finally, Figures 21 and 22 present autocorrelations for interest rates in both economies. In the

Alvarez-Jermann economy, autocorrelations are U-shaped between 0.08 and 0.2, as in the empirical

term structure of real interest rates. However, in the Lucas economy the autocorrelation is constant

and determined by the autocorrelation of consumption growth (0.281).

5.2 The Expectations Hypothesis

There are two main versions of the expectations hypothesis. While most of the empirical litera-

ture has concentrated on the expectations hypothesis in rates, it is pedagogical to start with the

expectations hypothesis in prices. Recall equation (10)

p

f

1,t

= E

t

[p

b

1,t+1

] + cov

t

¸

m

t+1

,

p

b

1,t+1

p

b

1,t

.

Backus, Gregory, and Zin (1989) tested the expectations model in the Lucas economy by starting

with (10), assuming that the risk premium was constant, i.e.,

E

t

[p

b

1,t+1

] −p

f

1,t

= a,

and regressing

p

b

1,t+1

−p

f

1,t

= a +b(p

f

1,t

−p

b

1,t

) (15)

27

Table 4: The number of rejects with regressions in the Lucas economy.

y

t+1

p

b

1,t+1

−p

f

1,t

p

b

1,t+1

−p

f

1,t

−rp

1,t

p

b

1,t+1

−p

f

1,t

p

b

1,t+1

−p

f

1,t

−rp

1,t

x

t

p

f

1,t

−p

b

1,t

p

f

1,t

−p

b

1,t

p

b

1,t

−p

f

1,t

p

b

1,t

−p

f

1,t

Wald(a = b = 0) 648 71 651 68

Wald(b = 0) 109 67 105 62

Wald(b = −1) 1000 1000 1000 1000

Table 5: The number of rejects with regressions in the Alvarez-Jermann economy.

y

t+1

p

b

1,t+1

−p

f

1,t

p

b

1,t+1

−p

f

1,t

−rp

1,t

p

b

1,t+1

−p

f

1,t

p

b

1,t+1

−p

f

1,t

−rp

1,t

x

t

p

f

1,t

−p

b

1,t

p

f

1,t

−p

b

1,t

p

b

1,t

−p

f

1,t

p

b

1,t

−p

f

1,t

Wald(a = b = 0) 1000 54 1000 62

Wald(b = 0) 1000 65 1000 62

Wald(b = −1) 1000 1000 1000 1000

to see if b = 0. They generated 200 observations 1000 times and used the Wald test with

White (1980) standard errors to check if b = 0 at the 5% signiﬁcance level. They could reject

the hypothesis only roughly 50 times out of 1000 regressions, which is what one would expect from

chance alone. On the other hand, for all values of b except −1, the forward premium is still useful

in forecasting changes in spot prices. The hypothesis b = −1 was rejected every time.

Table 4 presents the number of rejections of diﬀerent Wald tests in the regressions

y

t+1

= a +bx

t

in the Lucas economy calibrated to UK data. Table 5 presents the same tests for the Alvarez-

Jermann model. Unlike in the Lucas model, the results with the Alvarez-Jermann model are

consistent with empirical evidence on the expectations hypothesis. The model can generate enough

variation in the risk premia to account for rejections of the expectations hypothesis. When the risk

premium is subtracted from p

b

1,t+1

−p

f

1,t

, b is equal to zero with 5% signiﬁcance level.

In Table 6 the results of regression (15) are presented for one realization of 200 observations

for the Lucas model and for the Alvarez-Jermann model, and for UK real interest rate data. In

Table 6, Wald rows refer to the marginal signiﬁcance level of the corresponding Wald test. It is

worth noting how close the values of the regression coeﬃcients are for the Alvarez-Jermann model

and the real UK data.

Recent empirical literature has concentrated on the Log Pure Expectations Hypothesis. Ac-

cording to the hypothesis, the n-period forward rate should equal the expected one-period interest

rate n periods ahead:

f

n,t

= E

t

[r

1,t+n

].

To test the hypothesis, one can run the regression

(n −1) ∗ (r

n−1,t+1

−r

n,t

) = a +b(r

n,t

−r

1,t

) for n = 2, 3, 4, 5, 6, 11. (16)

28

Table 6: Tests of the expectations hypothesis with a single regression.

Variable/Test Lucas Alvarez-Jermann Data

a −0.0118 0.0279 −0.0047

se(a) 0.0041 0.0032 0.002

b 0.0070 −0.4970 −0.4982

se(b) 0.2711 0.0481 0.1046

R

2

0.0067 0.4036 0.3457

Wald(a = b = 0) 0.0013 2e−104 3e−7

Wald(b = 0) 0.9569 4e−19 2e−6

Wald(b = −1) 5e−15 1e−19 2e−6

According to the Log Pure Expectations Hypothesis, one should ﬁnd that b = 1.

30

Table 7 sum-

marizes the results from this regression for the models and data. The expectations hypothesis is

clearly rejected in all cases except for the Lucas model.

6 Sensitivity Analysis

Sepp¨ al¨ a (2000, Appendix B) provides extensive sensitivity analysis on the Alvarez-Jermann model

with respect to estimated and calibrated parameters of the model. The calibrate parameters are

the discount factor, β, and the coeﬃcient of risk aversion, γ. The results are not surprising. When

the discount factor is increased, the allocations move closer to full sharing, and hence rejecting the

expectations hypothesis becomes more and more diﬃcult. Similarly, the allocations move closer to

full sharing as the coeﬃcient of risk aversion is increased. In addition, the term premium changes

sign and becomes less and less volatile, so that rejecting the expectations hypothesis becomes more

and more diﬃcult.

Sensitivity analysis are provided with respect to estimated parameters can be done in two

diﬀerent ways. Once one parameter has been changed, the calibrated parameters can be either

recalibrated or not. Both results are reported in Sepp¨ al¨ a (2000). Without recalibration, the allo-

cations move closer to full sharing when the standard deviation of individual income is increased.

On the other hand, when the preference parameters are recalibrated, it is possible to explain

the rejections of the expectations hypothesis when the standard deviation of individual income

(std(log(e

i

/

¸

j

(e

j

)))) is as high as 65%, expansions are four times more likely to occur than re-

cessions, expansions and recessions are equally likely to take place, autocorrelation of consumption

varies between 0.45 and −0.15, the standard deviation of consumption growth varies between 1 and

7%, and the average consumption growth varies between 0.5 and 3.5%.

31

30

See, e.g., Campbell, Lo, and McKinley (1997).

31

There is some evidence in the U.S. data that average consumption growth has become less volatile after the

WWII, see Chapman (2001) and Otrok, Ravikumar, and Whiteman (2002). Following a suggestion by an anonymous

referee, I divided the UK data into several subperiods, 1900–1914, 1915–1938, 1939–1945 (WWII), 1946–1969 (post-

WWII), 1970–1979 (the British Disease), 1980–1990, (Thatcherism), and 1991-2001. With the expectation of WWII

period, the unconditional mean and standard deviation of consumption growth during those periods varied between

29

Table 7: Expectations hypothesis regressions in rates.

Regression a se(a) b se(b) R

2

Lucas (n = 2) 1.3841 0.0007 1.1830 0.0003 0.1443

Lucas (n = 3) 1.9277 0.0009 1.1609 0.0002 0.1808

Lucas (n = 4) 2.1219 0.0010 1.1449 0.0002 0.2120

Lucas (n = 5) 2.1896 0.0010 1.1339 0.0001 0.2370

Lucas (n = 6) 2.2131 0.0010 1.1264 0.0001 0.2564

Lucas (n = 11) 2.2265 0.0010 1.1103 0.0001 0.3056

Alvarez-Jermann (n = 2) −2.5150 0.0009 −0.0586 0.0001 0.0024

Alvarez-Jermann (n = 3) −0.8688 0.0004 0.5663 0.0001 0.5629

Alvarez-Jermann (n = 4) −1.8223 0.0007 0.2910 0.0001 0.1152

Alvarez-Jermann (n = 5) −1.3232 0.0005 0.4714 0.0001 0.3786

Alvarez-Jermann (n = 6) −1.6003 0.0006 0.3983 0.0001 0.2522

Alvarez-Jermann (n = 11) −1.5077 0.0006 0.4567 0.0001 0.3517

Data (n = 2) 0.2752 0.2336 0.2931 0.1448 0.0365

Data (n = 3) 0.2316 0.2543 0.4033 0.1113 0.1

Data (n = 4) 0.2055 0.2820 0.4199 0.1029 0.1121

Data (n = 5) 0.2056 0.3135 0.3135 0.4168 0.1060

Data (n = 6) 0.2225 0.3470 0.4059 0.1034 0.0942

Data (n = 11) 0.3891 0.5198 0.3891 0.1336 0.0292

30

7 Conclusions and Further Research

With risk-averse agents, the term structure contains expectations plus term premia. In order for

policy makers to extract information about market expectations from the term structure, they need

to have a general idea about the sign and magnitude of the term premium. But as S¨ oderlind and

Svensson (1997) note in their review

“We have no direct measurement of this (potentially) time-varying covariance [term

premium], and even ex post data is of limited use since the stochastic discount factor is

not observable. It has unfortunately proved to be very hard to explain (U.S. ex post)

term premia by either utility based asset pricing models or various proxies for risk.”

This paper studied the behavior of the default-risk free real term structure and term premia in

two general equilibrium endowment economies with complete markets but without money. In the

ﬁrst economy there were no frictions, as in Lucas (1978) and in the second the risk-sharing was

limited by the risk of default, as in Alvarez and Jermann (2000, 2001). Both models were solved

numerically, calibrated to UK aggregate and household data, and the predictions were compared to

the data on real interest rate constructed from UK index-linked data. While both models produce

time-varying term premia, only the model with limited risk-sharing can generate enough variation

in the term premia to account for the rejections of expectations hypothesis.

I conclude that the Alvarez-Jermann model provides one plausible explanation for the Backus-

Gregory-Zin term premium puzzle in real term structure data. What is needed now is a theory to

explain the behavior of the term structure of nominal interest rates. Since it is usually recognized

that monetary policy can only have eﬀect with “long and variable lags” as Friedman (1968) put

it, it is crucial to understand what are the inﬂation expectations that drive the market. Once we

understand the behavior of both nominal and real interest rates, we can get correct estimates of

these inﬂation expectations. An interesting topic for further research is whether a nominal version

of the Alvarez-Jermann model is consistent with the nominal data.

Another interesting topic would be to analyze the cyclical behavior of nominal and real term

structures, both in data and in theory. King and Watson (1996) provide an example of how to

do this, but they had to use real interest rates that they constructed using a VAR framework.

In my opinion, the British data would provide a better approximation for ex-ante real interest

rates. However, since the British data are still relatively short and neither the Lucas model nor the

Alvarez-Jermann model were built to confront this question, this topic is left for further research.

A Algorithm

This section explains how the Alvarez-Jermann model can be solved numerically. The aggregate

endowment is growing over time, but the CRRA utility function implies that the value function,

V (·), autarky values of utility, U

i

(·), and the policies, {C

1

(·), C

2

(·), W(·)}, satisfy the following

homogeneity property.

0.63 and 3.45 (the mean) and 1.12 and 4.38 (the standard deviation). During the WWII, the average consumption

growth was negative (−0.89). Unfortunately, the model produces strange results with negative growth, but that is

hardly typical for technologically advanced countries during the last 150 years.

31

Proposition 2. For any y > 0 and any (w, z, e),

V (y

1−γ

w, z, ye) = y

1−γ

V (w, z, e)

U

i

(z, ye) = y

1−γ

U

i

(z, e) for i = 1, 2

C

i

(y

1−γ

w, z, ye) = yC

i

(w, z, e) for i = 1, 2

W(y

1−γ

w, z, ye) = y

1−γ

W(w, z, e).

Proof. See the proof of Proposition 3.9 in Alvarez and Jermann (1996).

Deﬁning a new set of “hat” variables as

u(c) = e

1−γ

u(

c

e

) = e

1−γ

u(ˆ c)

U

i

(z

, e

) = (e

)

1−γ

U

i

(z

, 1) =

ˆ

U

i

(z

, 1) for i = 1, 2

w =

e

1−γ

e

1−γ

w = e

1−γ

ˆ w

w

=

(e

)

1−γ

(e

)

1−γ

w

= (e

)

1−γ

ˆ w

,

and using the above proposition in the following way:

V (w, z, e) = V (

e

1−γ

e

1−γ

w, z,

e

e

e) = e

1−γ

V ( ˆ w, z, 1)

V (w

, z

, e

) = (e

)

1−γ

V (

w

(e

)

1−γ

, z

, 1) = (λ(z

)e)

1−γ

V ( ˆ w, z, 1),

the functional equation can be rewritten with stationary variables as

TV ( ˆ w, z, 1) = max

ˆ c

1

,ˆ c

2

,{ ˆ w(z

)}

u(ˆ c

1

) +β

¸

z

∈Z

V ( ˆ w(z

), z

, 1)λ(z

)

1−γ

π(z

|z)

subject to

ˆ c

1

+ ˆ c

2

≤ 1 (17)

u(ˆ c

2

) +β

¸

z

∈Z

ˆ w(z

)λ(z

)

1−γ

π(z

|z) ≥ ˆ w (18)

V ( ˆ w(z

), z

, 1) ≥

ˆ

U

1

(z

, 1) ∀z

∈ Z,

ˆ w(z

) ≥

ˆ

U

2

(z

, 1) ∀z

∈ Z. (19)

In order to guarantee the above maximization problem is well-deﬁned, it is needed to assume that

max

z∈Z

β

¸

z

∈Z

λ(z

)

1−γ

π(z

|z)

¸

< 1.

The results in Alvarez and Jermann (1996) indicate that during recession the allocations do

not depend on past history. However, during the expansion there are two endogenous states: one

32

where z

t

= z

e

and z

t−1

= z

hl

, and another where z

t

= z

e

and z

t−1

= z

lh

. From now on, the states

are ordered as follows:

z

1

z

2

z

3

z

4

z

5

¸

¸

¸

¸

¸

¸

=

(z

t

= z

hl

)

(z

t

= z

e

, z

t−1

= z

hl

)

(z

t

= z

e

, z

t−1

= z

e

)

(z

t

= z

e

, z

t−1

= z

lh

)

(z

t

= z

lh

)

¸

¸

¸

¸

¸

¸

.

Moreover, it is possible to solve for the allocations and prices simply by solving for at most two

systems of nonlinear equations. The ﬁrst system corresponds to the case in which the participation

constraints are not binding, and the second system corresponds to the case in which the agents are

constrained when entering the boom period. In addition to the nonlinear equation, there is a set

of inequalities that determines which case is valid.

From feasibility (17) and non-satiation, it follows that agent 1’s consumption is always the

aggregate endowment less agent 2’s consumption. Hence, both systems have 10 equations in 10 un-

knowns: agent 2’s consumption and continuation utility in each of the ﬁve states. From now on,

c

n

will denote ˆ c

2

(z

n

) and “hats” will be dropped from other variables as well.

In both the unconstrained case and the constrained case, ﬁve equations are given by (18);

during the expansion neither agent has a reason to trade: c

3

= 0.5; (17) and symmetry imply that

c

1

+ c

5

= 1; and the participation constraint (19) holds with equality when agent 2 receives the

most favorable shock: w(z

5

) = U

2

(z

5

). The two missing equations depend on whether the agents

are constrained when entering the expansion state. In the unconstrained case, c

2

= c

1

and c

4

= c

5

,

and, in the constrained case, w(z

2

) = U

2

(z

2

) and c

2

+c

4

= 1.

To solve for the allocations, one needs only to solve for the unconstrained system and check

whether w(z

2

) ≥ U

2

(z

2

). If this is not the case, the solution is given by the constrained case.

References

Abel, A.B. (1999), “Risk Premia and Term Premia in General Equilibrium”, Journal of Monetary

Economics, 43, 3–33.

Abken, P.A. (1993), “Inﬂation and the Yield Curve”, Federal Reserve Bank of Atlanta Economic

Review, May/June, 13–31.

Aiyagari, S.R. (1994), “Uninsured Idiosyncratic Risk and Aggregate Saving”, Quarterly Journal of

Economics, 109, 659–684.

Alvarez, F. and U.J. Jermann (1996), “Asset Pricing when Risk Sharing is Limited by Default”,

Mimeo, The Wharton School.

Alvarez, F. and U.J. Jermann (2000), “Eﬃciency, Equilibrium, and Asset Pricing with Risk of

Default”, Econometrica, 68, 775–797.

Alvarez, F. and U.J. Jermann (2001), “Quantitative Asset Pricing Implications of Endogenous

Solvency Constraints”, Review of Financial Studies, 14, 1117–1151.

Backus, D.K., A.W. Gregory, and S.E. Zin (1989), “Risk Premiums in the Term Structure: Evidence

from Artiﬁcial Economies”, Journal of Monetary Economics, 24, 371–399.

33

Baker, M. (1997), “Growth-Rate Heterogeneity and the Covariance Structure of Life-Cycle Earn-

ings”, Journal of Labor Economics, 15, 338–375.

Bansal, R. and W.J. Coleman II (1996), “A Monetary Explanation of the Equity Premium, Term

Premium, and Risk-Free Rate Puzzles”, Journal of Political Economy, 104, 1135–1171.

Barr, D.G. and J.Y. Campbell (1997), “Inﬂation, Real Rates, and the Bond Market: A Study of

UK Nominal and Index-Linked Government Bond Prices”, Journal of Monetary Economics, 39,

361–383.

Barton, E., F. David, and E. Fix. (1962), “Persistence in a Chain of Multiple Events when There

Is Simple Persistence”, Biometrika, 49, 351–357.

Brav, A., G.M. Constantinides, and C.G. Geczy (1999), “Asset Pricing with Heterogeneous Con-

sumers and Limited Participation: Empirical Evidence”, Mimeo, University of Chicago.

Blough, S.R. (1994), “Yield Curve Forecasts of Inﬂation: A Cautionary Tale”, New England Eco-

nomic Review, May/June, 3–15.

Blundell R. and I. Preston (1998), “Consumption Inequality and Income Uncertainty” Quarterly

Journal of Economics, 113, 603–640.

Breeden, D.T. (1979), “An Intertemporal Asset Pricing Model with Stochastic Consumption and

Investment Opportunities”, Journal of Financial Economics, 7, 265–296.

Brown, R.H. and S.M. Schaefer (1994), “The Term Structure of Real Interest Rates and the Cox,

Ingersoll, and Ross Model”, Journal of Financial Economics, 35, 3–42.

Brown, R.H. and S.M. Schaefer (1996), “Ten Years of the Real Term Structure: 1984–1994”, Journal

of Fixed Income Research, March, 6–22.

Buraschi, A. and A. Jiltsov (2003), “Habit Persistence and the Nominal Term Structure of Interest

Rates ”, Mimeo, London Business School.

Campbell, J. Y. (1986), “Bond and Stock Returns in a Simple Exchange Model”, Quarterly Journal

of Economics, 101, 785–803.

Campbell, J. Y. (1995), “Some Lessons from the Yield Curve”, Working Paper No. 5031, National

Bureau of Economic Research.

Campbell, J.Y. (1998), “Asset Prices, Consumption, and the Business Cycle”, Mimeo, Harvard

University.

Campbell, J.Y. and J.H. Cochrane (1999), “By Force of Habit: A Consumption-Based Explanation

of Aggregate Stock Market Behavior”, Journal of Political Economy, 107, 205–251.

Campbell, J.Y. and R.J. Shiller (1991), “Yield Spreads and Interest Rate Movements: A Bird’s

Eye View”, Review of Economic Studies, 58, 495–514.

Campbell, J.Y., A.W. Lo, and A.C. MacKinlay (1997), The Econometrics of Financial Markets,

Princeton University Press, Princeton, New Jersey.

Chapman, D.A. (1997), “The Cyclical Properties of Consumption Growth and the Real Term

Structure”, Journal of Monetary Economics, 39, 145–172.

Chapman, D.A. (2001), “Does Intrinsic Habit Formation Actually Resolve the Equity Premium

Puzzle?”, Mimeo, University of Texas.

Chen, N.-F. (1991), “Financial Investment Opportunities and the Macroeconomy”, Journal of

Finance, 46, 529–554.

Cochrane, J.H. (1999), “New Facts in Finance”, Mimeo, University of Chicago.

Cochrane, J.H. and L.P. Hansen (1992), “Asset Pricing Explorations for Macroeconomics”, Working

Paper No. 4088, National Bureau of Economic Research.

34

Constantinides, G. (1990), “Habit Formation: A Resolution of the Equity Premium Puzzle”, Jour-

nal of Political Economy, 98, 519–543.

Constantinides, G. and D. Duﬃe (1996), “Asset Pricing with Heterogeneous Consumers”, Journal

of Political Economy, 104, 219–240.

Cox, J., J. Ingersoll, and S. Ross (1985), “A Theory of the Term Structure of Interest Rates”,

Econometrica, 53, 385–408.

Dai, Q. (2003), “Term Structure Dynamics in a Model with Stochastic Internal Habit”, Mimeo,

New York University.

Dai, Q. and K.J. Singleton (2002), “Expectations Puzzle, Time-varying Risk Premia, and Aﬃne

Models of the Term Structure”, Journal of Financial Economics, 63, 415–441.

Deacon, M.P. and A.J. Derry (1994), “Deriving Estimates of Inﬂation Expectations from the Prices

of UK Government Bonds”, Working Paper No. 23, Bank of England.

Donaldson, J.B., T. Johnsen, and R. Mehra (1990), “On the Term Structure of Interest Rates”,

Journal of Economic Dynamics and Control, 14, 571–596.

Duﬀee, G.R. (2002), “Term premia and interest rate forecasts in aﬃne models”, Journal of Finance,

57, 405–443.

Dybvig, P., J. Ingersoll, Jr., and S. Ross (1996), “Long Forward and Zero-Coupon Rates Can Never

Fall”, Journal of Business, 69, 1–25.

Estrella A. and G.A. Hardouvelis (1991), “The Term Structure as a Predictor of Real Economic

Activity”, Journal of Finance, 46, 555–580.

Evans, M.D.D. (1998), “Real Rates, Expected Inﬂation, and Inﬂation Risk Premia”, Journal of

Finance, 53, 187–218.

Fama, E.F. (1975), “Short Term Interest Rates as Predictors of Inﬂation”, American Economic

Review, 65, 269–282.

Fama, E.F. (1984), “The Information in the Term Structure”, Journal of Financial Economics, 13,

509–528.

Fama, E.F. (1990), “Term-Structure Forecasts of Interest Rates, Inﬂation, and Real Returns”,

Journal of Monetary Economics, 25, 59–76.

Fama, E.F. and R.R. Bliss (1987), “The Information in Long-Maturity Forward Rates”, American

Economic Review, 77, 246–273.

Fama, E.F. and K.R. French (1989), “Business Conditions and Expected Returns on Stocks and

Bonds”, Journal of Financial Economics, 25, 23–50.

Friedman, M. (1968), “The Role of Monetary Policy”, American Economic Review, 63, 1–17.

Froot, K.A. (1989), “New Hope for the Expectations Hypothesis of the Term Structure of Interest

Rates”, Journal of Finance, 44, 283–304.

den Haan, W.J. (1995), “The Term Structure of Interest Rates in Real and Monetary Economies”,

Journal of Economic Dynamics and Control, 19, 909–940.

Harvey, C.R. (1988), “The Real Term Structure and Consumption Growth”, Journal of Financial

Economics, 22, 305–333.

Heaton, J. and D. Lucas (1992), “The Eﬀects of Incomplete Insurance Markets and Trading Costs

in a Consumption-Based Asset Pricing Model”, Journal of Economic Dynamics and Control, 16,

601–620.

Heaton, J. and D. Lucas (1996), “Evaluating the Eﬀects of Incomplete Markets on Risk Sharing

and Asset Pricing”, Journal of Political Economy, 104, 443–487.

35

Jorion, P. and F. Mishkin (1991), “A Multicountry Comparison of Term-Structure Forecasts at

Long Horizons”, Journal of Financial Economics, 29, 59–80.

Kandel, S., A.R. Ofer, and O. Sarig (1996), “Real Interest Rates and Inﬂation: An Ex-Ante

Empirical Analysis”, Journal of Finance, 51, 205–225.

Kehoe, T.J. and D.K. Levine (1993), “Debt-Constrained Asset Markets”, Review of Economic

Studies, 60, 865–888.

King, R.G. and M.W. Watson (1996), “Money, Prices, and the Business Cycle”, Review of Eco-

nomics and Statistics, 78, 35–53.

Kocherlakota, N.R. (1996), “Implications of Eﬃcient Risk Sharing without Commitment”, Review

of Economic Studies, 63, 595–609.

LeRoy, S.F. (1973), “Risk Aversion and the Martingale Property of Stock Prices”, International

Economic Review, 14, 436–446.

Lucas, R.E., Jr. (1978), “Asset Prices in an Exchange Economy”, Econometrica, 46, 1429–1446.

Luttmer, E. (1996), “Asset Pricing in Economies with Frictions”, Econometrica, 64, 1439–1467.

Mankiw, G. “The Equity Premium and the Concentration of Aggregate Shocks”, Journal of Fi-

nancial Economics, 17, 211–219.

Meghir, C. and E. Whitehouse (1996), “The Evolution of Wages in the United Kingdom: Evidence

from Micro Data”, Journal of Labor Economics, 14, 1–25.

Mehra, R. and E. C. Prescott (1985) “The Equity Premium: A Puzzle”, Journal of Monetary

Economics, 15, 145–161.

Melino, A. (1988) “The Term Structure of Interest Rates: Evidence and Theory”, Journal of

Economic Surveys, 2, 335–366.

Mishkin, F.S. (1981), “The Real Interest Rate: An Empirical Investigation”, Carnagie-Rochester

Conference Series on Public Policy, 15, 151–200.

Mishkin, F.S. (1990a), “The Information in the Longer Maturity Term Structure about Future

Inﬂation”, Quarterly Journal of Economics, 105, 815–828.

Mishkin, F.S. (1990b), “Yield Curve”, Working Paper No. 3550, National Bureau of Economic

Research.

Mishkin, F.S. (1991), “A Multi-Country Study of the Information in the Shorter Maturity Term

Structure about Future Inﬂation”, Journal of International Money and Finance, 10, 2–22.

Mishkin, F.S. (1992), “Is the Fisher Eﬀect for Real?”, Journal of Monetary Economics, 30, 195–215.

Otrok, C., B. Ravikumar, and C.H. Whiteman (2002), “Habit Formation: A Resolution of The

Equity Premium Puzzle?”, Journal of Monetary Economics, 49, 1261–1288.

Rubinstein, M. (1976), “The Valuation of Uncertain Income Streams and the Pricing of Options”,

Bell Journal of Economics and Management Science, 7, 407–425.

Sargent, T.J. (1987), Dynamic Macroeconomic Theory, Harvard University Press, United States of

America.

Sepp¨ al¨ a, J. (2000), “The Term Structure of Real Interest Rates: Theory and Evidence from UK

Index-Linked Bonds”, Oﬃce of Research Working Paper 00–0120, University of Illinois.

Shiller, R.I. (1979), “The Volatility of Long-Term Interest Rates and Expectations Models of the

Term Structure”, Journal of Political Economy, 87, 1190–1219.

Shiller, R.I. (1990), “The Term Structure of Interest Rates” in B. M. Friedman and F. H. Hahn

(eds.), Handbook of Monetary Economics, 1, North-Holland, Netherlands.

Shiller, R.I., J.Y. Campbell, and K. Schoenholtz (1983), “Forward Rates and Future Policy: In-

36

terpreting the Term Structure of Interest Rates”, Brookings Papers on Economic Activity, 2,

173–225.

Storesletten, K., C.I. Telmer, and A. Yaron (1999), “Asset Pricing with Idiosyncratic Risk and

Overlapping Generations”, Mimeo, Carnegie-Mellon University.

Svensson, L.E.O. (1994a), “Estimating and Interpreting Forward Interest Rates: Sweden 1992–

1994”, Working Paper No. 4871, National Bureau of Economic Research.

Svensson, L.E.O. (1994b), “Monetary Policy with Flexible Exchange Rates and Forward Interest

Rates as Indicators”, Discussion Paper No. 941, Centre for Economic Policy Research.

S¨ oderlind, P. and L.E.O. Svensson (1997), “New Techniques to Extract Market Expectations from

Financial Instruments”, Journal Monetary Economics, 40, 383–429.

Taylor, M.F. (ed.) with J. Brice, N. Buck, and E. Prentice-Lane (1998), British Household Panel

Survey User Manual Volume A: Introduction, Technical Report and Appendices, University of

Essex, Colchester.

Vigneron, O. (1999), “The Real Term Structure of Interest Rates and Aggregate Economic Fluc-

tuations”, Mimeo, University of Chicago.

Wachter, J.A. (2002), “Habit Formation and Returns on Bonds and Stocks”, Mimeo, New York

University.

Weil, P. (1989), “The Equity Premium Puzzle and the Risk-Free Rate Puzzle”, Journal of Monetary

Economics, 24, 401–421.

White, H. (1980), “A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test

for Heteroskedasticity”, Econometrica, 48, 817–838.

Woodard, T.G. (1990), “The Real Thing: A Dynamic Proﬁle of the Term Structure of Real Interest

Rates and Inﬂation Expectations in the United Kingdom 1982–1989, Journal of Business, 63,

373–398.

Zhang, H.H. (1997), “Endogenous Borrowing Constraints with Incomplete Markets”, Journal of

Finance, 52, 2187–2209.

37

Abstract This paper studies the behavior of the default-risk-free real term structure and term premia in two general equilibrium endowment economies with complete markets but without money. In the ﬁrst economy there are no frictions as in Lucas (1978) and in the second risk-sharing is limited by the risk of default as in Alvarez and Jermann (2000, 2001). Both models are solved numerically, calibrated to UK aggregate and household data, and the predictions are compared to data on real interest rates constructed from the UK index-linked data. While both models produce time-varying risk or term premia, only the model with limited risk-sharing can generate enough variation in the term premia to account for the rejections of expectations hypothesis. JEL classiﬁcation: E43, E44, G12. Keywords: Term structure of interest rates, General equilibrium, Default risk, Term premia, Index-linked bonds.

1

Introduction

One of the oldest problems in economic theory is the interpretation of the term structure of interest rates. It has been long recognized that the term structure of interest rates conveys information about economic agents’ expectations about future interest rates, inﬂation rates, and exchange rates. In fact, it is widely agreed that the term structure is the best source of information about economic agents’ inﬂation expectations for one to four years ahead.1 Since it is generally recognized that monetary policy can only have eﬀect with “long and variable lags” as Friedman (1968) put it, the term structure is an invaluable source of information for monetary authorities.2 Moreover, empirical studies indicate that the slope of the term structure predicts consumption growth better than vector autoregressions or leading commercial econometric models.3 Empirical research on the term structure of interest rates has concentrated on the (pure) expectations hypothesis. That is, the question has been whether forward rates are unbiased predictors of future spot rates. The most common way to test the hypothesis has been to run a linear regression (error term omitted): rt+1 − rt = a + b(ft − rt ), where rt is the one-period spot rate at time t and ft is the one-period-ahead forward rate at time t. The pure expectations hypothesis implies that a = 0 and b = 1. Rejection of the ﬁrst restriction a = 0 is consistent with the expectations hypothesis with a term premium that is nonzero but constant. By and large the literature rejects both restrictions.4 Rejection of the second restriction, b = 1, requires a risk or term premium that varies through time and is correlated with the forward premium, ft − rt . Many studies—e.g., Fama and Bliss (1987) and Fama and French (1989)—take this to indicate the existence of time-varying risk or term premium.5 In order for policy makers to extract information about market expectations from the term structure, they need to have a general idea about the sign and magnitude of the term premium. Obviously, as emphasized by Cochrane (1999), the sign and the magnitude of the term premium are also interesting for market professionals. Therefore, it is interesting to ask if there are models that are capable of generating term premia that are similar to the ones observed in actual time series. Unfortunately, as S¨derlind o and Svensson (1997) note in their review, “We have no direct measurement of this (potentially) time-varying covariance [term

See, e.g., Fama (1975, 1990) and Mishkin (1981, 1990a, 1992) for studies on inﬂation expectations and the term structure of interest rates using U.S. data. Mishkin (1991) and Jorion and Mishkin (1991) use international data. Abken (1993) and Blough (1994) provide surveys of the literature. 2 Svensson (1994ab) and S¨derlind and Svensson (1997) discuss monetary policy and the role of the term structure o of interest rates as a source of information. 3 See, e.g., Harvey (1988), Chen (1991), and Estrella and Hardouvelis (1991). 4 The literature is huge. Useful surveys are provided by Melino (1988), Shiller (1990), Mishkin (1990b), and Campbell, Lo, and MacKinlay (1997). Shiller (1979), Shiller, Campbell, and Schoenholtz (1983), Fama (1984, 1990), Fama and Bliss (1987), Froot (1989), Campbell and Shiller (1991), and Campbell (1995) are important contributions to this literature. 5 The literature is somewhat inconsistent in the deﬁnitions of risk and term premium. In the discussion below, both terms are used interchangeably. In Section 3.3, I will deﬁne the risk premium as the term that accounts for the rejections of expectations hypothesis in prices and term premium as the term that accounts for the rejections of expectations hypothesis in rates.

1

3

this framework has three advantages over the standard incomplete markets approach described above. and den Haan (1995) investigates the issue further. This is closely related to the equity premium puzzle ﬁrst posed by Mehra and Prescott (1985) and the risk-free rate puzzle posed by Weil (1989). Rubinstein (1976). 2001) are models of endowment economies without money. and their answer is that the model can account for neither sign nor magnitude of average risk premia in forward prices and holding-period returns. Mehra and Prescott (1985) conjecture that the most promising way to resolve the equity premium puzzle is to introduce features that make certain types of intertemporal trades among agents infeasible. and Zin (1989). Finally. They use a complete markets model ﬁrst presented by Lucas (1978). Johnsen. and Mehra (1990) obtain the same result for a general equilibrium production economy. They show how endogenously determined solvency constraints that prevent the agent from defaulting on his own contracts help explain the equity premium and the risk-free rate puzzles. Both models are solved numerically and calibrated to UK aggregate and household data. Their answer is that “uninsurable income shocks may help explain one of the more persistent term structure puzzles” but “the question remains whether the prediction of a relatively large forward premium will obtain in a long horizon model. and even ex post data is of limited use since the stochastic discount factor is not observable. Usually this has meant that markets are exogenously incomplete.premium].” The question whether utility-based asset pricing models are capable of generating risk premia similar to the ones observed in actual time series was originally posed in Backus. the markets are complete and hence any security can be priced. ﬁnding the solution to an incomplete markets problem involves solving a very diﬃcult ﬁxed point problem. This is particularly important in addressing questions related to the term structure of interest rates. Economic agents are allowed to trade only in certain types of assets. The problem is that given the variability in U. they show that one cannot reject the expectations hypothesis with data generated via the Lucas model. 6 4 . 7 The work by Alvarez and Jermann builds on contributions by Kehoe and Levine (1993).7 First.S. the stochastic discount factor derived from frictionless utility-based asset pricing models is not suﬃciently volatile. and the predictions are compared to data on ex-ante real interest rates constructed from UK index-linked data. whereas solving the model in Alvarez and Jermann can be very fast and easy to implement. Second. Kocherlakota (1996).S. ex post) term premia by either utility based asset pricing models or various proxies for risk. Donaldson.” Alvarez and Jermann (2000. 2001) study the asset pricing implications of an endowment economy when agents can default on contracts. LeRoy (1973).6 In addition. For the purpose of dynamic asset pricing. aggregate consumption series. Heaton and Lucas (1992) use a three-period incomplete markets model to address the term premium puzzle. and Breeden (1979) were important early contributions to the literature on dynamic general equilibrium asset pricing models. both Lucas (1978) and Alvarez and Jermann (2000. This paper studies the implications of the Lucas and Alvarez-Jermann models on the behavior of the default-risk-free ex-ante real term structure and term premia. See also Aiyagari (1994) and Zhang (1997) for studies on endogenous solvency constraints. it should be noted that. However. while empirical research has concentrated on the nominal term structure. It has unfortunately proved to be very hard to explain (U. allocations do not depend on a particular arbitrary set of available securities. Gregory. and the set of available assets is exogenously predetermined. and Luttmer (1996).

UK index-linked bonds do not provide a correct measure of ex-ante real interest rates. At least in principle. so that the eﬀective lag is approximately eight months. The motivation behind this procedure is that it always allows the nominal value of the next coupon payment to be known. nobody is willing to lend more than the debtor is willing to pay back. Even though in the AlvarezJermann economy risk-sharing is limited by the risk of default. the RPI number relates to a speciﬁc day in the previous month. only the model with limited risk-sharing can generate enough variation in the term premia to account for the rejections of the expectations hypothesis. Campbell and Cochrane (1999). The most important exception is the UK market for index-linked debt.9 Therefore. Section 7 concludes. Section 2 presents the data on nominal and index-linked bonds. In addition. Hence. business cycles. but rather that default risk limits risk-sharing in such a way as to produce a “realistic” term premium. the thesis of this paper is not that default risk directly explains the term premium in the real term structure.10 Unfortunately. Wachter (2002) and Buraschi and Jiltsov (2003) have successfully applied habit formation to the “term premium puzzle. 8 5 . the instruments that are traded in state-contingent markets are default-risk-free. The rest of the paper is organized as follows. the indexation operates with a lag. The reason is that the nominal amounts paid by index-linked bonds do not fully compensate the holders for inﬂation. and its daily turnover is by far the highest in the world. The UK market for index-linked debt was started in 1981. Both coupon and principal payments are linked to the level of the RPI (Retail Price Index) published in the month seven months prior to the payment date. it is natural to compare the Lucas and Alvarez-Jermann term structures with the term structure of real bonds issued by the UK government. Section 3 presents the basic features of the Lucas and Alvarez-Jermann models. Bansal and Coleman (1996). 9 In the Alvarez-Jermann economy.8 It should also be emphasized that in both the Lucas and Alvarez-Jermann economies the object of the study is the term structure of default-risk-free real interest rates. Appendix A explains how the models are numerically solved. Since the previous version of this paper was completed. and invidual incomes in the UK Section 5 presents the numerical results for both models and compares the models’ behavior to the behavior of UK nominal and real term structures. aggregate consumption. Section 4 calibrates the models given the data on the risk-free rate. Therefore. the agents have an incentive to default on their contracts if honoring their contracts would leave them worse oﬀ than they would be in autarky. It constitutes a signiﬁcant proportion of marketable government debt. in equilibrium nobody defaults but risk-sharing is limited by the risk of default. and nominal There are several other interesting models that oﬀer at least partial resolution of the equity premium and/or risk-free rate puzzles. these bonds are default-risk free. any of these models could be used to study the Backus-Gregory-Zin “term premium puzzle” in UK data. A few examples are Constantinides (1990).” Duﬀee (2002) and Dai and Singleton (2002) study the issue in the context of aﬃne term structure models. Since everyone knows this and there is no private information. and by March 1994 it accounted approximately 15% of outstanding issues by market value. In principle.While both models produce time-varying risk premia. 2 UK Index-Linked Bonds The main complication in analyzing ex-ante real interest rates is that in most economies they simply are unobservable. and Abel (1999). Section 6 summarizes the sensitivity analysis of the Alvarez-Jermann model. 10 See Brown and Schaefer (1996) for more details. Constantinides and Duﬃe (1996).

Similar observations have been obtained by Woodward (1990) and Brown and Schaefer (1994.11 Since this paper is mainly concerned on the expectations hypothesis. I investigated this issue using real rates estimated from the nominal data as in den Haan (1995) and Chapman (1997). He ﬁnds that expected inﬂation is negatively correlated with real yields. Evans (1998) isolates the “indexation lag problem” to a conditional covariance term between the future (maturity less the inﬂation lag) inﬂation and nominal bond prices. Brown and Schaefer (1994. Following the suggestion by an anonymous referee. can never fall. 1996) who made diﬀerent assumptions to overcome the “indexation lag problem” and estimated the yield curve using diﬀerent methods than Evans used. using the properties of stochastic discount factors. Evans’ (1998) estimates of the term structure of real interest rates in the UK from January 1984 until August 1995 reveal the following stylized facts: 1.12 However. Evans also tests for the versions of expectations hypothesis used by Kandel. and Sarig (1996) and Barr and Campbell (1997) assume that diﬀerent versions of the expectations hypothesis hold. Ofer.13 He also rejects the Fisher hypothesis. Finally. He then estimates this term using a VAR model and derives the real interest rates. the long-term of the real term structure appears to be highly stable. These data also produce on average downward-sloping term structures. Deacon and Derry (1994) and Brown and Schaefer (1994) impose the Fisher hypothesis: nominal yields move one for one with changes in inﬂation. Kandel. this means that the long-term of the term structure should converge. such as Cox.” On the other hand. the average real term structure is downward sloping. In aﬃne-yield models. While the average term structure is upward sloping. which means that there is no “inﬂation risk premia.” Woodard (1990). 2. Roll. Shapes of the real term structure are relatively simple compared with the nominal term structure. More details are available upon request. 4. 1996) estimated the term structure of real interest rates to be upward-sloping on average. 3. and these models have been criticized on the grounds that the long-term of the nominal term structure does not appear to be stable. 13 One interesting observation should be mentioned. and Ross (1985). Diﬀerent authors have made diﬀerent assumptions in order to overcome the “indexation lag problem. If one calculates the average term structure using only the 12 11 6 .accrued interest can always be calculated. Ingersoll. While the long-term of nominal term structure is quite volatile. the methodology of Evans seemed best suited for my purposes. and Sarig (1996) and Barr and Campbell (1997) and rejects both versions at the 1 percent signiﬁcance level. This observation is very interesting since Dybvig. With the exception of the average shape of real term structure. The data discussed below was provided by him. Both the short-term and and long-term of the real term structure seem to be less autocorrelated than the corresponding maturities in the real term structure. Ofer. the results in Brown and Schaefer are consistent with results presented here. the reverse is true for the short-term: the short-term of the real term structure is highly volatile compared with that of the nominal term structure. if it exists. and Ross (1996) showed that under very general conditions the limiting forward rate.

{c} = {ct (z t ) : ∀t ≥ 0. . zt determines the aggregate endowment et . u(c) = 1−γ where γ is the agents’ constant coeﬃcient of relative risk-aversion. Since everyone knows this and there is no private information. Therefore.14 Agents have identical preferences represented by time-separable expected discounted utility. Let i = 1. The matrix Π determines the conditional probabilities for all histories π(z t |z0 ). . U (c)(z t ) = ∞ j=0 z t+j ∈Z t+j β j u(ct+j (z t+j ))π(z t+j |z t ). . In the planning problem. i = 1. The agents have an incentive to default on their contracts if honoring their contracts would leave them worse oﬀ than they would be in autarky. the term structure has been upward-sloping on average regardless whether one uses Evans data or real rates estimated from nominal data. where λ(zt+1 ) is the growth rate of aggregate endowment between t and t + 1 when the state in period t + 1 is zt+1 and αi (zt ) determines agent i’s share of the aggregate endowment when the state in period t is zt . 2000. z t ∈ Z t . and time-dependent solvency constraints. and rank them by discounted expected utility. In the planning problem. The diﬀerence between the AlvarezJermann economy and the Lucas economy is that in the former agents cannot commit to their contracts. z t ∈ Z t }. . . i = 1. From now on. 1) is a constant discount factor and the one-period utility function is of the constant relative risk aversion type c1−γ . . zt ) denote the history of z up to time t. Households care only about their consumption streams. 2. 2. . the Lucas economy being a special case of the Alvarez-Jermann economy where participation or solvency constraints do not ever bind. zN } with transition matrix Π. the possibility of default is prevented by participation constraints. The economy can be decentralized through complete asset markets where the positions that the agents can take are endogenously restricted by person. . Their endowments follow a ﬁnite-state ﬁrst-order Markov process. in equilibrium nobody defaults but risk-sharing is limited by the risk of default. 1 < γ < ∞. ei . 14 Alvarez and Jermann (2000) consider the more general case with I ≥ 2 agents. state. 2 denote each agent and {zt } denote a ﬁnite-state Markov process z ∈ Z = {z1 . the discussion concentrates on the Alvarez-Jermann economy. 7 . nobody is willing to lend more than the debtor is willing to pay back. 2001) consider a pure exchange economy with two agents. 2 as follows: t et+1 = λ(zt+1 )et and ei = αi (zt )et t for i = 1. where β ∈ (0.3 3.1 The Lucas and Alvarez-Jermann Models The Environment Alvarez and Jermann (1996. and the individual endowments. (1) data after the UK left the European Exchange Rate Mechanism in October 1992. Let z t = (z1 . the participation constraints force the allocations to be such that under no history will the expected utility be lower than that in autarky U (ci )(z t ) ≥ U (ei )(z t ) ∀t ≥ 0.

The motivation for autarky constraints of the form (1) is quite natural. wz is agent 2’s promised utility when the next-period state of the world is z .3 shows that adding more agents need not change the asset pricing implications of the model. that maximize agent 1’s expected utility subject to feasibility and participation constraints at every date and every history. The second welfare theorem holds for the economy. The model is easiest to solve when there are only two agents. One can argue that the model is unrealistic in two respects. since punishment for default would not be so serious.2 Constrained Optimal Allocations The constrained optimal allocations are deﬁned as processes.In other words. Note that the model abstracts completely from game-theoretic issues related to bargaining and renegotiation. in no time period and in no state of the world can either agent’s expected discounted utility be less than what the agent would obtain in autarky. In the real world. there are actually many more agents than just one and each agent has a very small weight in the collective bargaining problem. The recursive formulation of the constrained optimal allocations is given by the functional equation T V (w. Relaxing the ﬁrst assumption would reduce risk-sharing. (2) (3) where primes denote next-period values. Section 3. respectively. 8 . there is no conﬁscation of assets as punishment for default. On the one hand. e) = subject to c1 + c2 ≤ e u(c2 ) + β z ∈Z c1 . If this were the case. z . w is agent 2’s promised utility this period. it is sometimes diﬃcult to make a debtor pay. and (2) and (3) are the participation constraints for agent 1 and agent 2. 2. e ) ≥ U 1 (z . e) is agent 1’s value function.c2 . Trying to relax both of these unrealistic features of the model probably would not change the asset pricing implications too much. z. e ) ∀z ∈ Z wz ≥ U 2 (z . The punishment for default would be permanent exclusion from the risk-sharing arrangement. the agent would not commit to his contract and would choose to default. Punishing one agent by forever excluding him from the risk-sharing arrangement makes the punishing agent much worse oﬀ than with ﬁnite punishment. given agent 2’s initial promised expected utility.{wz } max u(c1 ) + β z ∈Z V (wz . Hence. since punishment for default would be more serious. one can solve for the allocations by solving the planning problem and read the prices oﬀ the ﬁrst-order conditions of the competitive equilibrium. e )π(z |z) wz π(z |z) ≥ w V (wz . z. The justiﬁcation is that as in Alvarez and Jermann (2000). the agent who reverts to autarky can keep his endowment stream. typically an agent who declares bankruptcy can return to the risk-sharing arrangement after a ﬁnite number of periods. e ) ∀z ∈ Z. 3. People do not always keep their promises and debt collection can be costly and possibly useless. {ci }. z . V (w. i = 1. the punishment is too severe. Relaxing the second assumption would increase risk-sharing. On the other hand.

Hence. and allocations {ci . z t+1 ) = U i (ei )(z t+1 ) ∀t ≥ 0. z) be the price of an Arrow security that pays one unit of consumption good at the beginning of the next period when the next-period state is z and the current-period state is z. the continuation utility equals autarchy utility: i H i (Bt+1 (z t+1 ). z )π(z |z) H i (a. ∀z t+1 ∈ Z t+1 . Agent i’s holdings of this asset are denoted by ai . z) = max u(c) + β c. The solvency constraints aﬀect each state diﬀerently. Taking the constraints and prices as given. 2. When the solvency constraints are binding. The household’s problem given the current state (a. ∀z t+1 ∈ Z t+1 . Hence. Let q(z . The equilibrium is a set of solvency constraints {Bt }. This ensures that the solvency constraints prevent default and it allows as much insurance as possible.3. ai } such that t t+1 1. z) is to maximize the expected utility H i (az . short-selling. since the relative value of autarky compared to honoring the contract is diﬀerent in every state. prices {qt }. so that in every state z there are one-period Arrow securities for each next period state of nature. z) be the minimum position z agent i can take in the asset that pays oﬀ when the next-period state is z and the current-period state is z. A binding solvency constraint means that the agent is indiﬀerent between defaulting and staying in the risksharing regime. the solvency constraints prevent agents from holding so much debt in any state that they would like to default on their debt contracts. z ∈Z ∀z ∈ Z The equilibrium can now be deﬁned. the economy must be decentralized.{az }z ∈Z z ∈Z subject to solvency and budget constraints az ≥ B i (z . The markets are complete. Markets clear: a1 (z t+1 ) + a2 (z t+1 ) = 0 ∀t ≥ 0. z . Deﬁnition 1. Finally. i Deﬁnition 2. A slack solvency constraint means that the agent’s expected discounted utility is strictly higher than what he would obtain in autarky. 9 . let B i (z . and other exogenous constraints that are commonly used in the literature on incomplete markets. the model provides endogenous justiﬁcation for debt. z) q(z . However. z)az + c ≤ a + ei (z). 3.3 Asset Pricing To analyze asset prices. (4) Condition (4) means that solvency constraints are endogenously generated. the allocations solve both households’ optimization problems. solvency.

if the current state is z t . z t ) = Et n mt+j .t+1 (8) The bond prices are invariant with respect to time. In other words. provided that the new agents’ marginal valuations are always less than or equal to market valuations. n. the crucial point of the model is that the prices of Arrow securities are given by the maximum of the marginal rates of substitution for agents 1 and 2. Let mt+1 denote the real stochastic discount factor mt+1 ≡ max β i=1. A corollary of this is that each new agent whose income process is perfectly correlated with aggregate income has no eﬀect on asset prices. the constrained agent has an internal interest rate that is higher than the market rate. In the full risk-sharing regime (Lucas economy).t 15 This result was also derived by Cochrane and Hansen (1992) and Luttmer (1996). and hence equation (8) gives a recursive formula for pricing zero-coupon bonds of any maturity. Forward prices are deﬁned by pb n+1.t pn.t z t+1 ∈Z t+1 max β i=1. j=1 (7) Using (5). the price of an Arrow security that pays one unit of consumption good at the beginning of the next period if the next-period state is z t+1 is given by q(z t+1 .From the perspective of asset pricing. z t ) be the price of an Arrow security from state z t to state z t+j . In addition. Notice that one can introduce as many new agents into the economy as desired without changing the asset pricing implications. u (ci. let q(z t+j . and the equation above.z ) = k=t t t+j−1 q(z k+1 . he would like to borrow more than is feasible.2 u (ci.2 u (ci.t pf = b . z t ) = max β i=1.t z t+n ∈Z t+n q(z t+n . (6) and let q(z k+1 . (6).t ) The price of an n-period zero-coupon bond is given by pb = n. u (ci. u (ci ) t (5) The economic intuition is that the unconstrained agent in the economy does the pricing.2 u (ci ) t+1 π(z t+1 |zt ). the constrained agent would like to sell that asset and hence his marginal valuation of the asset is lower. z k ). Therefore.t+1 ) . to keep the autarky constraints satisﬁed. z k ) be given by (5).t+1 ]. which is given by q(z t+j . the two marginal rates of substitution are equalized.15 That is. Since B(z t+1 ) gives the minimum amount of an asset one can buy.t+1 ) b p π(z t+1 |z t ) = Et [mt+1 pb n−1. 10 . note that pb = n.t ) n−1.

t+1 ]. 16 Without income heterogeneity.t+1 ] + cov t mt+1 . and Zin (1989).t ≡ cov t mt+1 . Gregory. p1.t+n ]. I will call a diﬀerence between the one-period forward rate and the expected value of one-period interest rate next period the term premium for the one-period forward contract.t − Et [r1.t = − log(pf ) and rn.and the above prices are related to interest rates (or yields) by fn.t+1 pb 1. Similar techniques can be applied for more complicated cases. pb 1.t+n pb 1. I will solve the model for the simplest possible case that produces nonconstant interest rates and has income heterogeneity.1 Calibration Free Parameters In the spirit of Backus.t = Et [pb 1.t − Et [r1.t In addition. n.t+1 ] b = Et [mt+1 ]Et [pb 1. 4 4. p1.t pb 1.t = −(1/n) log(pb ). given by rp 1.t 1.t ≡ f1.t pb 2.t and similarly rp n.t ≡ cov t j=1 mt+j .t is the risk premium for the n-period forward contract: n rp n.t 1.t ≡ fn.t : tp 1. I will call it the risk premium for the one-period forward contract.t+1 ] b = pb Et [pb 1.t = pf − Et [pb 1.t+1 ] + cov t [mt+1 .t+1 ] and similarly tp n.t+n ].t+1 ]. n. 11 .t+1 ] + cov t [mt+1 .t . the Alvarez-Jermann and Lucas models would be identical. pb 1. (10) Since the conditional covariance term is zero for risk-neutral investors. rp 1. tp 1.16 This is obtained by introducing uncertainty in the growth rate of aggregate endowment while treating the agents in a symmetric fashion. 1.t+1 pb 1. write (8) for a two-period bond using the conditional expectation operator and its properties: pb = Et [mt+1 pb 2.t = pf − Et [pb 1. pb 1.t .t is the term premium for the n-period forward contract: tp n.t (9) To deﬁne the risk premium as in Sargent (1987).t n. which implies that pf = 1.

Finally. The states are ordered so that α1 (zhl ) = α2 (zlh ) ≥ α1 (ze ) = α2 (ze ) ≥ α1 (zlh ) = α2 (zhl ) and the transition matrix Π preserves symmetry between the agents. See Barton. Π = (1 − πe )/2 πr 0 1 − πr The order of calibration is as follows. 12 . An alternative would be to allow individual incomes to take diﬀerent values during expansions. Thus. λr λ = λe .S. data. α1 determines the standard hl deviation of individual income. (1 − θ)π (1 − θ)(1 − π) + θ Next. 1 − α1 α1 hl hl λe and λr for the growth rates of aggregate endowment. First. π. Telmer. I discuss in Section 4. the transition matrix for the aggregate states can be expressed as a function of the fraction of time spent in the expansion state.17 During the expansion both agents have the same endowment ze . ze . 17 18 Storesletten.2. π= 1 − πr 2 − (πe + πr ) θ = πe + πr − 1. λe and λr determine the average growth rate of aggregate consumption and its standard deviation. The states zhl and zlh are associated with a recession.In particular. there are ﬁve free parameters associated with aggregate and individual endowment. and Fix (1962). λr and πr and πe for the transition matrix 1 − πr 0 πr πe (1 − πe )/2 . the recessions are associated with a widening of inequality in earnings. given the transition matrix for the aggregate states.5 and α2 = 0. David. and the ﬁrst-order autocorrelation of aggregate consumption. and zlh . and Yaron (1999) document this for U.2 why I chose not to do so. hl 1 αhl 1 − α1 hl α1 = 0. and following Mankiw (1986) and Constantinides and Duﬃe (1996). Note that with this parameterization one cannot pin down the persistence of individual income. θ:18 Π= Clearly. (1 − θ)π + θ (1 − θ)(1 − π) . there will be three exogenous states: zhl . These are α1 for individual incomes.5 .

I chose to use the BHPS. The quarterly population series were constructed by assuming that population grows at constant rate within the year and that the original annual data are as at December 31 of each year. there are two free parameters. Campbell (1998) reports that in the annual UK data for 1891– 1995. Since the UK does not publish oﬃcial deﬁnitions of expansions and recessions. An alternative would be to allow individual incomes to take diﬀerent values also during expansions. which calibrate asset 19 The growth (rate) is the diﬀerence in the logarithm of the series. and Section 4. I followed Chapman (1997). Since I am interested in household income heterogeneity for particular households and the FES is a survey.443%. Section 4. aggregate endowment equals aggregate consumption) and business cycles. A peak is the last quarter prior to the beginning of a contraction. in the UK there are only two data sets that provide information on household income. the standard deviation of the growth rate of aggregate consumption is 2. 13 . The quarterly observations on the GDP at 1990 prices and annual observations on the population were obtained from the CD-ROM July 1999 version of the International Monetary Fund’s International Financial Statistics. Business cycle expansions are deﬁned as at least two consecutive quarters of positive growth19 in a three quarter equally-weighted. This approach is similar to that taken in previous studies. The ﬁrst is the Family Expenditure Survey (FES). and Yaron (1999). and Appendix B of Sepp¨l¨ (2000) shows that the main results are highly robust with respect to measurement error aa in moment conditions. Cycle contractions are at least two consecutive quarters of negative growth in the moving average of output. associated with preferences. 4. As mentioned above.2 explains the ﬁve equations that determine endowment parameters. real term structure.1 Aggregate and Individual Endowment Aggregate Consumption and Business Cycles The ﬁrst step is to calibrate the law of motion for the aggregate endowment so that it matches a few facts about aggregate consumption (in the model. the current speciﬁcation only allows one to match the standard deviation of individual income.In addition.2 Individual Endowment The next step is to calibrate the process for the individual endowments. β and γ. who used the following deﬁnitions in his study of the cyclical properties of U. centered moving average of the real GDP/capita (“output”).8824 times more likely to occur than recessions in the UK from the ﬁrst quarter of 1957 until the last quarter of 1997.898%. which covers the period September 1990 to January 1998.2. Telmer. and a trough is the last quarter prior to the beginning of a expansion.2. the average growth rate of aggregate consumption is 1.S. According to the deﬁnitions. which covers the years 1968–1992 and the second is the British Household Panel Survey (BHPS). expansions are 3.281.3 shows how the preference parameters are pinned down to match the ﬁrst and second moments of the risk-free rate in the UK Appendix A gives details on how the model is solved numerically. and the ﬁrst-order autocorrelation is 0. such as Heaton and Lucas (1996) and Storesletten. 4. Unfortunately.2 4. Business cycles thusly deﬁned are presented in Figure 1 for the UK from the ﬁrst quarter of 1957 until the last quarter of 1997.

I concentrate on the simplest possible speciﬁcation (the least free parameters) that would provide diﬀerent results from the Lucas model.000 individuals. Therefore.S. given the length of data it seems unlikely that one could test whether this relationship is reasonable.20 The BHPS provides a panel of monthly observations of individual and household income and other variables from September 1990 until January 1998 for more than 5. Following Heaton and Lucas (1996). This provides a close connection for the standard deviation of i in the t t i=1 t data with the standard deviation of α’s in the model. This gives a total of 2. Thus. one needs to know how the cross-sectional variance in household income varies over business cycles. t i where ηt = ei / n ei . 20 Heaton and Lucas (1996) and Storesletten. 14 .Growth Rate of Real GDP/Capita 5 4 3 2 1 0 -1 -2 -3 1957 1962 1967 1972 1977 1982 1987 1992 1997 Figure 1: The growth rate of real GDP/capita in the UK 1957:1–1997:4 and its moving average (quarterly observations). see Taylor (1998). pricing models to the U. Note that it also imposes a cointegration relationship between aggregate and individual income. the individual annual income dynamics are assumed to follow an AR(1)-process: i i ¯ (11) log(ηt ) = η i + ρi log(ηt−1 ) + i . Telmer. Unfortunately.391 individuals in the panel. data on households. and Yaron method. and the depression ended in the ﬁrst quarter of 1992. The BHPS was started only in September 1990 when the UK was in depression. For households with more than one member with reported income.21 I use the subset of the panel that has reported positive income in every year since 1991. Telmer. 21 For more details on the BHPS. the data does not cover a full cycle and it has only one data point for the recession. and Yaron (1999) also provide examples on how to estimate the standard deviation of individual income. giving a total of approximately 10.000 households. I constructed the individual income as the total household income divided by the number of the members of the household with reported income. To implement the Storesletten.

They show strong growth in transitory inequality toward the end of this period. which are currently unavailable. 0. Table 2 reports cross-sectional means and standard deviations of the coeﬃcient estimates obtained by Heaton and Lucas. the standard deviation of the error term in the BHPS.2830 Cross-Sectional Standard Deviation 5. 15 . Finally.2830. which is closer to exercise here.S. 0. and assume is that the process is composed of a pure transitory component and a random walk. Blundell and Preston (1998) impose the presence of a random walk component. 0. Storesletten. is signiﬁcantly smaller than that in the PSID.251 Cross-Sectional Standard Deviation 2.6315 0. Coeﬃcient ηi ¯ ρi σi Cross-Sectional Mean −6. Baker (1997) provides analysis that questions the assumption of a unit root.2035. 22 For previous estimates of the standard deviation of individual income using the UK data. 0.22 The question of the persistence of individual income is. Telmer.4331 0. and the standard deviation of the error term. On the other hand. their estimates are close to mine. and Yaron (1999) obtain estimates much closer to one. σ i = E[( i )2 ]. but at the end their estimates are higher. see Meghir and Whitehouse (1996) and Blundell and Preston (1998). Heaton and Lucas (1996) estimate a relatively low number while Storesletten. Blundell and Preston (1998). Telmer. while young cohorts are shown to face signiﬁcantly higher levels of permanent inequality.2853 Table 2: Cross-sectional means and standard deviations of the coeﬃcient estimates in the regression (11) obtained by Heaton and Lucas (1996). Telmer.2376 0.413 0. Heaton and Lucas (1996) estimate the same t coeﬃcients using a sample of 860 U.529 0. households in the Panel Study of Income Dynamics (PSID) that have annual incomes for 1969 to 1984. unfortunately.131 Table 1 reports cross-sectional means and standard deviations of the coeﬃcient estimates in the regression (11).2035 0. and Yaron (1999) obtain a higher number for the standard deviation of individual income. At the beginning of the sample. as in Storesletten.354 0. and Yaron (1999). The relevant numbers are the ﬁrst-order autocorrelation coeﬃcient. Both studies use the FES data.529.251. is roughly the same as that in the PSID. decompose the variance in income into permanent and transitory components. Data: Panel Study of Income Dynamics Coeﬃcient ηi ¯ ρi σi Cross-Sectional Mean −3.Table 1: Cross-sectional means and standard deviations of the coeﬃcient estimates in the regression (11). Data: British Household Panel Survey.332 0. ρi . The estimated autocorrelation coeﬃcient in the BHPS. an unresolved issue. In order to reasonably pin down the persistence of individual income one would need to have very long panels of data.

γ.9573 α1 = 0.Standard Deviation of (Individual Consumption/Aggregate Consumption) 35 30 25 20 % 15 10 5 0 0. and the coeﬃcient of relative risk aversion. The next step is to match agents’ preference parameters.2 0.6 0.9573 and the transition matrix 0. 16 .2294 0. I will report all the parameter values with four decimal precision.8 Discount Factor 1 2 3 4 5 6 7 Risk aversion Figure 2: Standard deviation of (individual consumption/aggregate consumption) as a function of the discount factor and coeﬃcient of relative risk aversion.5 .0736 .7706 0.2. 4. Figure 2 plots the standard deviation of individual consumption.4 0.5717 0 Π = 0. the results are quite sensitive to the parameter values. Therefore. α2 = 0.7706 0.0736 0. as j 23 Due to highly nonlinear nature of the model.0291 . and λ = 1.4283 4.5 .8527 0.3 Calibrated Values Solving the system of ﬁve unknowns in ﬁve equations leads to the endowment vectors and growth rates23 0. to the key statistics of the asset market data.3 Preferences a function of β and γ. 0.5717 0.4283 0. std(log( c cj )). the discount factor β. 0 0. To illustrate the features i of the model.2294 0.

In Section 5. Note that in the Alvarez-Jermann economy in autarky no trade is allowed because one of the agents would default on his contract. or the standard deviation of individual income moves the results in the direction of more risk-sharing (the Lucas model).2 0.8 Discount Factor 1 2 3 4 5 6 7 0. The ﬂat segment in the upper-left corner of the Figure corresponds to autarky and the ﬂat segment in the lower-right corner to perfect risk-sharing. Figure 4: Average risk-free rate as a function of the discount rate and risk aversion in a Lucas economy. I report numerical results only for one pair of coeﬃcients of risk aversion and discount factor. I show how increasing the coeﬃcient of risk-aversion. The parameter values which are interesting for the purpose of asset pricing are those that generate allocations between these two extremes. To accomplish this. and individual income. aggregate consumption. the discount factor. and hence all the statistics have been set to zero. as was done in Section 2.6 0. Figures 3–6 present the average and standard deviations of the risk-free rate in the AlvarezJermann and Lucas economies.4 0. Recall that the price of an n-period zero-coupon bond in a Lucas economy is given by pb = β n Et n.Expected Riskfree Rate Expected Riskfree Rate 300 250 200 % % 300 250 200 150 100 50 0 −50 0. 17 .t et+n et −γ .2 Risk aversion Risk aversion Figure 3: Average risk-free rate as a function of the discount rate and risk aversion in an Alvarez-Jermann economy. business cycles.6 0. and in Appendix B of Sepp¨l¨ (2000) I show that these results are very robust to measurement error aa in risk-free rates.4 0. The shapes in the Lucas economy are relative easy to understand using log-linear approximations. In addition.8 Discount Factor 1 2 3 4 5 6 7 150 100 50 0 −50 0. one has to choose either relatively low risk aversion and a relatively high discount factor or relatively high risk aversion and a relatively low discount factor.

2 0. In the Alvarez-Jermann model.2 ci ˆt+1 ci ˆt −γ .3378 and γ = 3. This leads to an increase in variability in consumption shares and hence an increase in the “max” operator.8 Risk aversion Discount Factor 1 2 3 4 5 6 7 % 40 Risk aversion Figure 5: Standard deviation of the riskfree rate as a function of the discount rate and risk aversion in an Alvarez-Jermann economy.3378 is considerably less than what either 24 Again. 2n et et+n γ2 var log − .8 Discount Factor 1 2 3 4 5 6 7 0 0. matching these values leads to β = 0.Standard Deviation of Riskfree Rate Standard Deviation of Riskfree Rate 100 100 80 80 60 % 60 40 20 20 0 0. the ˆt incentive to participate in the risk-sharing arrangement is reduced. In the Alvarez-Jermann model. Figure 6: Standard deviation of the riskfree rate as a function of the discount rate and risk aversion in a Lucas economy.24 one obtains γ et+n Et log n et et+n γ E[rn. the relationship between parameter values and interest rates is more complicated because the interest rates depend not only on aggregate endowment but also on the consumption share of the unconstrained agent: − exp(rt ) = pb = βEt t et+1 et −γ max i=1. See Campbell (1986).198 and its standard deviation was 5. (14) where ci is agent i’s consumption share in period t. 18 . the approximation is exact only if consumption growth has a log-normal distribution.514.4 0. β = 0.446.2 0.4 0.6 0. Taking a log-linear approximation of the n-period interest rate.t = − log(β) + γ2 et+n vart log . thereby increasing bond prices and reducing one-period interest rates. When the discount factor is lowered. 2n et − (12) (13) Therefore. the average real risk-free rate was 1. the interest rate decreases exponentially in the discount factor and increases slowly in the risk-aversion coeﬃcient.6 0.t ] = − log(β) + E log n et rn. Campbell (1998) reports that in annual UK data for 1891–1995.

γ)] − 1. Obviously. but rather how the fraction of population is aﬀected by solvency constraints over the business cycle. 25 19 . In the Alvarez-Jermann endowment economy. I show aa that this result is also very robust. one has to reduce risk-sharing considerably from the full risk-sharing benchmark. and Yaron (1999). given the above discount factor and risk aversion values. the more tempting is default (the agents would like to accumulate state-contingent debt during bad times and default during good times) and hence the risk-sharing is reduced for more patient agents.99. The quantitative results in Alvarez and Jermann (2001) indicate that increasing the persistence does not change the asset pricing implications. In addition. Brav. the agent who got the lower share during the recession has the higher consumption growth and would like to default. the allocations are close to autarky allocations. Telmer.100) {|E[r(β. Constantinides. in the Alvarez-Jermann economy. Table 3 summarizes the main statistics for the models. In Appendix B of Sepp¨l¨. the agent who got the lower share during the recession has the higher consumption growth rate. The average duration of expansion is about six years and the average duration of depression is about two years. In other words. and Geczy (1999) conclude that the observation error in the consumption data makes it impossible to test the complete consumption insurance assumption against the assumption of incomplete consumption insurance. std[r(β. note that the persistence of individual income is lower than the values estimated by either Heaton and Lucas (1996) or Storesletten. one cannot interpret the agents literally. 6.1 Results The Term Structure of Interest Rates Figures 7–10 present the interest rates for maturities of 1 to 30 years and the forward rates of 1 to 30-year forward contracts during expansions and recessions in the Alvarez-Jermann and Lucas Clearly. Notice that. When recession changes to expansion. 5 5.446} . It is diﬃcult to say how reasonable this result is: In a recent paper. with only two agents.25 The next section shows how this mechanism translates to the behavior of the term structure of interest rates. and in the data. Since the standard deviation is more important for explaining the behavior of the term premium.γ∈(1. the standard deviation of individual consumption is close to the standard deviation of individual income.0. When expansion changes to recession. in order to be able to match asset market data. he would like to default. Only when expansion changes to recession or vice versa is somebody constrained. The reason is that. so it is not possible to match both the average risk-free rate and the standard deviation simultaneously. in the Lucas economy risk-sharing is never limited.t. I chose the discount factor and risk-aversion coeﬃcients in the Lucas economy (0. an incentive to participate in risk-sharing is very high so that only by lowering the discount factor can autarky become tempting. In order to be able to match the basic asset pricing data. The mechanism for limited risk-sharing works as follows. Hence. Both agents are oﬀ the solvency constraint when the current state is the same as the previous state.complete markets or incomplete markets literature typically use.1149) = arg min β∈(0. one has to reduce risk-sharing. the more persistent the individual income. γ)] = 5. provided that one is allowed to increase the discount factor.99].198| s. That is.

.Table 3: Selected statistics for the Lucas model when β = 0.446 1. In the Alvarez-Jermann model.3 0.3378 and γ = 3.443 2.) [I]n every business cycle of the 1952–1988 period the one-year spot rate is lower at the business trough than at the preceding or following peak. and for the data. both models produce both upward and downward-sloping term structures.) [I]n every business cycle of the 1952–1988 period the ﬁve-year yield spread (the ﬁve-year yield less the one-year spot rate) is higher at the business trough than at the preceding or following peak.882 — — 28.198 5.1149. 20 .) Another stylized fact is that long rates rise less than short rates during business expansions and fall less during contractions.0 28.514. A few things are worth noting from the Figures. Fama (1990) reports that “A stylized fact about the term structure is that interest rates are pro-cyclical. However.4193 Alvarez-Jermann Model 1. On the other hand. at the top of the cycle the term structure lies uniformly below the term structure at the bottom of the cycle.93 5.4193 Data 1. . Donaldson. .898 0. (. Lucas Model 7.3 0.443 2.4336 28.281 3. In the Lucas model.443 2.71 0.898 0. the term structure of interest rates is downwardsloping in recessions and upward-sloping during expansions. .281 3. (. . the models’ cyclical behavior is exactly the opposite. Johnsen.)/ Pr(rec. the term structure of interest rates is upward-sloping in recessions and downward-sloping during expansions. Also.882 26.99 and γ = 6. First.882 0.0 1.2035 E[r] (%) std[r] (%) E(∆c) (%) std[∆c] (%) corr[∆c] Pr(exp. and Mehra (1990) report that in a stochastic growth model with full depreciation the term structure of (ex-ante) real interest rates is rising at the top of the cycle and falling at the bottom of the cycle.446 1. Thus spreads of long-term over short-term yields are counter-cyclical.26 The cyclical behavior of the term structure is of particular interest since empirical and theoretical results from previous studies have been contradictory.198 5.” Notice that this statement applies to the term structure of nominal interest rates.446 1.281 3. in both models upward-sloping term structures are always uniformly below downward-sloping term structures.3 0.898 0.) std(log(ci / j (cj ))) (%) corr(log(ci / j (cj )) std(log(ei / j (ej ))) (%) corr(log(ei / j (ej )) economies. In addition. . for the AlvarezJermann model when β = 0. (. 26 See Proposition 1 below for the explanation.

5 −2 0 6 0 5 10 15 Years 20 25 30 5 10 15 Years 20 25 30 Figure 9: Interest rates (solid line) and forward rates in the Alvarez-Jermann economy during expansions. Figure 10: Interest rates (solid line) and forward rates in the Lucas economy during expansions.5 8 2 7.Interest Rates and Forward Rates (state 1) 12 7 6 10 5 8 4 3 % Interest Rates and Forward Rates (state 1) 6 % 2 1 4 2 0 −1 0 −2 −2 0 −3 0 5 10 15 Years 20 25 30 5 10 15 Years 20 25 30 Figure 7: Interest rates (solid line) and forward rates in the Alvarez-Jermann economy during recessions. 21 .5 8 10 9.5 7 0 6. Figure 8: Interest rates (solid line) and forward rates in the Lucas economy during recessions.5 6 9 % Interest Rates and Forward Rates (state 3) 4 % 8. Interest Rates and Forward Rates (state 3) 10 11 10.

See den Haan (1995) and Vigneron (1999). Therefore. the cyclical behavior of the term structure will depend on the autocorrelation of consumption growth. will converge and the limiting risk premium is the diﬀerence between the limiting forward price and the limiting expected spot price. Proposition 1. In general. Similarly. to study the slope of the term structure it is suﬃcient to know whether interest rates are procyclical or countercyclical. To see this.t ) = pb = β n Et n. however. the Perron-Frobenius theorem guarantees that the dominant eigenvalue of B is positive and that any positive vector operated on by powers of B will eventually approach the associated eigenvector and grow at the rate of this eigenvalue.In the Lucas economy. Recall equation (12): rn. Recall that the n-period forward price is the ratio of the price of an (n + 1)-period bond to the price of an n-period bond.t et+n et −γ × max i=1. the price of a one-period bond is j πij mij . where bij denotes the (i. the prices of multiple-period bonds are determined by −n exp(rn.2 ci ˆt+n i ct+n−1 ˆ −γ . and Ross (1996) holds in the Alvarez-Jermann economy. An immediate corollary of Proposition 1 is that the risk premia will converge. Roll. Since the transition matrix is ergodic. It implies that if consumption growth is positively autocorrelated.t Proof. Johnsen. and Mehra (1990) report. where bij deﬁnes a matrix B. the autocorrelation decreases and the maturity term in the denominator starts to kick in reducing the interest rates.27 In the Alvarez-Jermann economy. converges to a constant. In state i.t+n ]. note that the following version of Dybvig. The interest rates move one-for-one with the business cycle.2 ci ˆt+1 ci ˆt −γ × · · · × max i=1. then a good shock today will forecast good shocks in the future and consequently high interest rates for the near future. pf . the expected spot price. the price of a two-period bond is πij mij πjk mjk j k or (2) j bij . the price of an n-period bond is (n) (2) given by j bij . Since the transition matrix is ergodic. As n gets large. the question of the cyclical behavior of the term structure is more complicated for production economies. it is enough to study one-period bonds that are determined by equation (14) − exp(rt ) = pb = βEt t 27 et+1 et −γ max i=1. If the transition matrix for the economy is ergodic.2 ci ˆt+1 ci ˆt −γ . However. Let mij denote the pricing kernel between states i and j.t = − log(β) + et+n γ Et log n et − γ2 et+n vart log 2n et . However. Et [pb 1. the ratio converges to the dominant eigenvalue of B regardless of the current state. 22 . n. exactly as Donaldson. This can be expressed as j bij . then as n approaches inﬁnity. the forward price. As the maturity increases. j) element of B 2 .

Figure 12: Risk premium (solid line) and term premium in the Lucas economy during recessions. as the Britain had time to go through only one business cycle in the sample. 28 23 .600) is −0. the Alvarez-Jermann model seems to be consistent with the real data and the Lucas model with the nominal data. Alvarez-Jermann economies. The greatest variability inside the “max” operator occurs when one moves from the expansion state to the recession state. The correlations between nominal data and the cyclical component are 0. unfortunately.17. A positive sign on the term premium means that forward rates tend to overpredict future interest rates and a negative sign means that the term premium underpredicts.28 Next. Recall that the variability of consumption shares increases in recessions and that aggregate income growth is positively autocorrelated. In other words.5 −1 −3 −4 −5 0 −1.Risk and Term Premium (state 1) 5 4 3 1 2 1 % Risk and Term Premium (state 1) 2 1. Figures 15–20 present the mean and the standard deviation of the interest rates. The correlation between yield spread (ﬁve-year yield minus one-year yield) and the cyclical component is 0. the correlation between one-year real interest rate and the cyclical component of real GDP/capita (obtained using a Hodrick-Prescott ﬁlter with a smoothing parameter of 1. forward rates. respectively.S. although the term premium is also stable over the cycle for long maturities. the two terms inside the conditional expectation work opposite to each other. These estimates. Figures 11–14 present the risk and term premia for maturities of 1 to 30 years during expansions and recessions in the Alvarez-Jermann and Lucas economies.5 5 10 15 Years 20 25 30 −2 0 5 10 15 Years 20 25 30 Figure 11: Risk premium (solid line) and term premium in the Alvarez-Jermann economy during recessions. the dominant term is consumption heterogeneity. In the British data presented in Section 2. This means that bond prices increase during expansions or that interest rates decline.48.43. It is interesting to note that King and Watson (1996) obtained the same result for the cyclical behavior of nominal and real interest rates in U. and in the Alvarez-Jermann model.19 and −0. They obtained real interest rates by estimating expected inﬂation using VAR. are not very reliable. Interest rates are countercyclical even though consumption growth is positively autocorrelated. In the current parameterization. Hence. it varies considerably and negatively with the level of interest rates for short maturities. In addition. data.5 0. This result indicates that the Alvarez-Jermann model may be useful in accounting for rejections of the expectations hypothesis. in the Lucas model the term premium is very stable.5 0 −1 −2 % 0 −0. Note how the signs of risk and term premia are opposite in the Lucas vs.

24 . In the Lucas economy.t ] − E[r1.5 % 0 % 0 −2 −0. The average term structure is upward-sloping in the Alvarez-Jermann economy and downward-sloping in the Lucas economy.t+j ]] = n n−1 tp j.t .5 −4 −1 −6 −1. the variance term grows faster than the maturity since shocks in the growth rate are persistent. The relationship between the term premium and the shape of the term structure is as follows.5 −8 0 5 10 15 Years 20 25 30 −2 0 5 10 15 Years 20 25 30 Figure 13: Risk premium (solid line) and term premium in the Alvarez-Jermann economy during expansions. the standard deviations decrease with the maturity as in the UK nominal and real data. the average shape of the term structure is easy to explain.t ] = − log(β) + If the E log et+n et et+n γ E log n et − et+n γ2 var log 2n et . the average term structure can be expressed as 1 E[rn. Recall equation (13): E[rn.t = 1 n n−1 fj. risk. but it is suﬃcient to note that the term structure is upward-sloping during the expansions and the economy is growing most of the time. Yields can be expressed as averages of forward rates: rn.5 4 1 2 0. then the average shape of the term structure is determined by the e ratio of the variance term to maturity. Figure 14: Risk premium (solid line) and term premium in the Lucas economy during expansions. j=0 Therefore. j=0 See den Haan (1995).t − Et [r1. unconditional expectations are more diﬃcult to obtain. ≈ n¯.Risk and Term Premium (state 3) 8 2 Risk and Term Premium (state 3) 6 1. In both economies.t ] = n 29 n−1 j=0 1 E[fj.t. and term premia.29 In the Alvarez-Jermann economy. If consumption growth is positively autocorrelated.

Figure 16: Average interest rates (solid line) and forward rates in the Lucas economy.2 1 0 6 0 5 10 15 Years 20 25 30 5 10 15 Years 20 25 30 Figure 15: Average interest rates (solid line) and forward rates in the AlvarezJermann economy.5 −2 −1 −4 −1.Average Interest Rates and Forward Rates 7 8 7.4 2 6.2 % Average Interest Rates and Forward Rates 4 % 7 6.5 4 1 2 0.4 5 7. Average Risk and Term Premium 6 2 Average Risk and Term Premium 1.5 −6 0 5 10 15 Years 20 25 30 −2 0 5 10 15 Years 20 25 30 Figure 17: Average risk premium (solid line) and term premium in the AlvarezJermann economy.5 % 0 % 0 −0. Figure 18: Average risk premium (solid line) and term premium in the Lucas economy.8 6 7.8 3 6.6 7. 25 .6 6.

et+n Suppose that ∆ log(et ) = ρ∆ log(et−1 ) + t .281)2 1 1 + = −1%. σ 2 ). this reduces to tp n. 2 2 ρ which is less that zero. Using the log-linear approximation and assuming homoscedastic errors. and Premia 5 5 4 4 % 3 % 3 2 2 1 1 0 0 5 10 15 Years 20 25 30 0 0 5 10 15 Years 20 25 30 Figure 19: Standard deviation of interest rates (solid line). in the British data tp 1. where tp 1.02898)2 (0. Forward Rates. 2 In the Lucas economy. Figure 20: Standard deviation of interest rates (solid line). forward rates (dashed line). and term premium (dotted line) in the Lucas economy. if ρ > 0. and Premia 6 6 Standard Deviation of Interest Rates. Then γ2 2 2 1 1 ρ σ − γ 2 ρσ 2 = −γ 2 σ 2 ρ2 + . risk premium (dash-dot line). Forward Rates.t = −(6.t = − ∼ N (0. 2 0. For example. and term premium (dotted line) in the Alvarez-Jermann economy.t = − vart [log(pb 1. log(et+n ) . conditional on current aggregate state or “heteroscedasticity ex-ante”: std[log(αi (zt+1 ))|λt = λr ] σr .Standard Deviation of Interest Rates. log(mt+n )].281 Alvarez and Jermann (2001) show that in the Alvarez-Jermann economy the sign of the term premium depends on one-period ahead individual income variance.t+n )] − cov t [log(p1.1149)2 (0.t = − γ2 et+n+1 vart Et+n log 2 et+n − γ 2 cov t Et+n log t et+n+1 . forward rates (dashed line). ≡ σe std[log(αi (zt+1 ))|λt = λe ] 26 .t+n ). the term premium can be expressed as 1 b tp n. risk premium (dash-dot line).

t+1 ] − p1. as in the empirical term structure of real interest rates.t − p1. Therefore.t+1 pb 1. 5.12 0.18 0. Gregory.t = a. assuming that the risk premium was constant. Figures 21 and 22 present autocorrelations for interest rates in both economies.Autocorrelation of Interest Rates 0.08 and 0.t ) (15) 27 . i. The next section studies whether this term premium is volatile enough to account for rejections of the expectations hypothesis.14 0.2 0.2. From (14) it follows that the max-term in the stochastic discount factor becomes more volatile and hence bond prices are higher (interest rates lower) in expansions.22 Autocorrelation of Interest Rates 0.t pb 1.281).1 0. Finally.t = a + b(p1. the term premium is negative and the average term structure is downward-sloping. autocorrelations are U-shaped between 0.2 0.e.t+1 ] + cov t mt+1 . a positive term premium is required to compensate bond holders.t+1 − p1. σe when σr < 1.1 0. Figure 22: Autocorrelation of interest rates in the Lucas Economy. and regressing f f b pb 1.5 0. While most of the empirical literature has concentrated on the expectations hypothesis in rates. Backus.4 0. 1.16 0. the term premium is positive and the average term structure is upward-sloping.08 0 5 10 15 Years 20 25 30 0 0 5 10 15 Years 20 25 30 Figure 21: Autocorrelation of interest rates in the Alvarez-Jermann Economy..2 The Expectations Hypothesis There are two main versions of the expectations hypothesis. it is pedagogical to start with the expectations hypothesis in prices. However.3 0. and Zin (1989) tested the expectations model in the Lucas economy by starting with (10). f Et [pb 1. in the Lucas economy the autocorrelation is constant and determined by the autocorrelation of consumption growth (0. In the Alvarez-Jermann economy.t . σe σr σe < 1 means that in expansions one expects more idiosyncratic risk in the future. When σr > 1. Recall equation (10) pf = Et [pb 1.

the forward premium is still useful in forecasting changes in spot prices. for all values of b except −1.t 54 65 1000 f pb 1.t 651 105 1000 f pb 1. It is worth noting how close the values of the regression coeﬃcients are for the Alvarez-Jermann model and the real UK data. yt+1 xt Wald(a = b = 0) Wald(b = 0) Wald(b = −1) f pb 1.t − rp 1.t pb − pf 1.t+1 − p1. Table 5 presents the same tests for the AlvarezJermann model.t 648 109 1000 f pb 1.t . They generated 200 observations 1000 times and used the Wald test with White (1980) standard errors to check if b = 0 at the 5% signiﬁcance level.t pb − pf 1.t ) for n = 2.t pf − pb 1. b is equal to zero with 5% signiﬁcance level.t 1.t 1. Unlike in the Lucas model.t 1000 1000 1000 f pb 1.t 1. Recent empirical literature has concentrated on the Log Pure Expectations Hypothesis. The model can generate enough variation in the risk premia to account for rejections of the expectations hypothesis.t 1.t+1 − p1.t = Et [r1.t+1 − p1. On the other hand. To test the hypothesis.t 1.t pb − pf 1.t pf − pb 1.t − r1. the n-period forward rate should equal the expected one-period interest rate n periods ahead: fn.t+1 − p1.t+1 − p1. 28 (16) .t 1000 1000 1000 f pb 1.t 1.t pb − pf 1.t+n ]. According to the hypothesis. the results with the Alvarez-Jermann model are consistent with empirical evidence on the expectations hypothesis. which is what one would expect from chance alone. They could reject the hypothesis only roughly 50 times out of 1000 regressions.t pf − pb 1.t 68 62 1000 Table 5: The number of rejects with regressions in the Alvarez-Jermann economy. 5.t+1 − p1. In Table 6.t+1 − p1. When the risk f premium is subtracted from pb 1.t 1.t+1 − p1.t − rp 1. one can run the regression (n − 1) ∗ (rn−1. 6.t − rp 1.t 1. 4.t − rp 1. Table 4 presents the number of rejections of diﬀerent Wald tests in the regressions yt+1 = a + bxt in the Lucas economy calibrated to UK data.t+1 − rn.t pf − pb 1. Wald rows refer to the marginal signiﬁcance level of the corresponding Wald test. 11.t 62 62 1000 to see if b = 0. In Table 6 the results of regression (15) are presented for one realization of 200 observations for the Lucas model and for the Alvarez-Jermann model. The hypothesis b = −1 was rejected every time. 3. yt+1 xt Wald(a = b = 0) Wald(b = 0) Wald(b = −1) f pb 1.t+1 − p1.Table 4: The number of rejects with regressions in the Lucas economy. and for UK real interest rate data.t 71 67 1000 f pb 1.t ) = a + b(rn.

0118 0. one should ﬁnd that b = 1.5%. and hence rejecting the expectations hypothesis becomes more and more diﬃcult.0032 −0.3457 3e−7 2e−6 2e−6 According to the Log Pure Expectations Hypothesis. the term premium changes sign and becomes less and less volatile. Ravikumar. Campbell. and the coeﬃcient of risk aversion. 1939–1945 (WWII).0067 0. Sensitivity analysis are provided with respect to estimated parameters can be done in two diﬀerent ways.0481 0. and the average consumption growth varies between 0. (Thatcherism).15. There is some evidence in the U. Appendix B) provides extensive sensitivity analysis on the Alvarez-Jermann model aa with respect to estimated and calibrated parameters of the model.Table 6: Tests of the expectations hypothesis with a single regression. 1980–1990.0047 0. the calibrated parameters can be either recalibrated or not..g. γ. autocorrelation of consumption varies between 0. Similarly.002 −0. Without recalibration. 1900–1914. the unconditional mean and standard deviation of consumption growth during those periods varied between 31 30 29 . The expectations hypothesis is clearly rejected in all cases except for the Lucas model. and 1991-2001. e.0279 0. the standard deviation of consumption growth varies between 1 and 7%. In addition. Once one parameter has been changed. and McKinley (1997).S. so that rejecting the expectations hypothesis becomes more and more diﬃcult. Variable/Test a se(a) b se(b) R2 Wald(a = b = 0) Wald(b = 0) Wald(b = −1) Lucas −0.4970 0.4982 0. Lo. 6 Sensitivity Analysis Sepp¨l¨ (2000. it is possible to explain the rejections of the expectations hypothesis when the standard deviation of individual income (std(log(ei / j (ej )))) is as high as 65%.0013 0. The results are not surprising. when the preference parameters are recalibrated. expansions and recessions are equally likely to take place. the allocations move closer to full sharing as the coeﬃcient of risk aversion is increased.0070 0. On the other hand.1046 0. expansions are four times more likely to occur than recessions. Following a suggestion by an anonymous referee.0041 0. the allocations move closer to full sharing.4036 2e−104 4e−19 1e−19 Data −0. When the discount factor is increased.2711 0. data that average consumption growth has become less volatile after the WWII. 1970–1979 (the British Disease). 1946–1969 (postWWII).30 Table 7 summarizes the results from this regression for the models and data. With the expectation of WWII period. I divided the UK data into several subperiods.45 and −0. see Chapman (2001) and Otrok. 1915–1938. Both results are reported in Sepp¨l¨ (2000).31 See. the alloaa cations move closer to full sharing when the standard deviation of individual income is increased. The calibrate parameters are the discount factor. β. and Whiteman (2002).5 and 3.9569 5e−15 Alvarez-Jermann 0.

2931 0.2752 0.3135 0.5663 0.0010 0.2910 0.3232 −1.0001 0.4059 0.0009 0.3135 0.0001 0.0001 0.0292 (n = 2) (n = 3) (n = 4) (n = 5) (n = 6) (n = 11) 30 .1029 0.8223 −1.3470 0.1808 0.0010 0.6003 −1.2543 0.2131 2.4567 0.0586 0.0001 0. Regression Lucas (n = 2) Lucas (n = 3) Lucas (n = 4) Lucas (n = 5) Lucas (n = 6) Lucas (n = 11) Alvarez-Jermann Alvarez-Jermann Alvarez-Jermann Alvarez-Jermann Alvarez-Jermann Alvarez-Jermann Data (n = 2) Data (n = 3) Data (n = 4) Data (n = 5) Data (n = 6) Data (n = 11) a 1.1448 0.0007 0.0007 0.1 0.1449 1.0024 0.3056 0.3891 se(a) 0.0002 0.1103 −0.4168 0.4199 0.0009 0.3841 1.0010 0.2336 0.0942 0.0001 0.1121 0.0001 0.1336 R2 0.3786 0.2522 0.Table 7: Expectations hypothesis regressions in rates.1113 0.2056 0.2265 −2.0006 0.0003 0.1609 1.1152 0.0005 0.2820 0.0001 0.2055 0.0365 0.5198 b 1.1060 0.5629 0.1219 2.9277 2.8688 −1.4714 0.2120 0.1443 0.2316 0.3983 0.0006 0.3891 se(b) 0.1830 1.4033 0.3517 0.2370 0.2225 0.0002 0.5077 0.2564 0.1034 0.0001 0.0001 0.1339 1.1896 2.5150 −0.0004 0.1264 1.0010 0.

autarky values of utility. since the British data are still relatively short and neither the Lucas model nor the Alvarez-Jermann model were built to confront this question.63 and 3. calibrated to UK aggregate and household data. While both models produce time-varying term premia. and even ex post data is of limited use since the stochastic discount factor is not observable. but they had to use real interest rates that they constructed using a VAR framework. V (·). both in data and in theory. C 2 (·). but the CRRA utility function implies that the value function. and the policies. However. as in Alvarez and Jermann (2000.7 Conclusions and Further Research With risk-averse agents. the average consumption growth was negative (−0. the model produces strange results with negative growth. 2001). we can get correct estimates of these inﬂation expectations.38 (the standard deviation). it is crucial to understand what are the inﬂation expectations that drive the market. W (·)}. only the model with limited risk-sharing can generate enough variation in the term premia to account for the rejections of expectations hypothesis. Unfortunately. A Algorithm This section explains how the Alvarez-Jermann model can be solved numerically. as in Lucas (1978) and in the second the risk-sharing was limited by the risk of default. King and Watson (1996) provide an example of how to do this.89). It has unfortunately proved to be very hard to explain (U. In the ﬁrst economy there were no frictions.12 and 4. the term structure contains expectations plus term premia. they need to have a general idea about the sign and magnitude of the term premium.S. 31 . But as S¨derlind and o Svensson (1997) note in their review “We have no direct measurement of this (potentially) time-varying covariance [term premium]. Since it is usually recognized that monetary policy can only have eﬀect with “long and variable lags” as Friedman (1968) put it. In my opinion. U i (·). but that is hardly typical for technologically advanced countries during the last 150 years. Once we understand the behavior of both nominal and real interest rates. An interesting topic for further research is whether a nominal version of the Alvarez-Jermann model is consistent with the nominal data. the British data would provide a better approximation for ex-ante real interest rates. What is needed now is a theory to explain the behavior of the term structure of nominal interest rates. {C 1 (·). this topic is left for further research. During the WWII. ex post) term premia by either utility based asset pricing models or various proxies for risk.45 (the mean) and 1. 0. satisfy the following homogeneity property. Another interesting topic would be to analyze the cyclical behavior of nominal and real term structures. I conclude that the Alvarez-Jermann model provides one plausible explanation for the BackusGregory-Zin term premium puzzle in real term structure data.” This paper studied the behavior of the default-risk free real term structure and term premia in two general equilibrium endowment economies with complete markets but without money. The aggregate endowment is growing over time. Both models were solved numerically. and the predictions were compared to the data on real interest rate constructed from UK index-linked data. In order for policy makers to extract information about market expectations from the term structure.

1) = ˆ subject to ˆ c1 + c2 ≤ 1 ˆ u(ˆ2 ) + β c z ∈Z c1 . z .Proposition 2. e) U i (z. ye) = y 1−γ V (w. 1) ≥ U 1 (z . z. However. z. For any y > 0 and any (w. 1) = U i (z . 1) = (λ(z )e)1−γ V (w. (19) w(z )λ(z )1−γ π(z |z) ≥ w ˆ ˆ ˆ V (w(z ). z. e) = V ( the functional equation can be rewritten with stationary variables as T V (w. e ) = (e )1−γ V ( (e )1−γ V (w. z. 1)λ(z )1−γ π(z |z) ˆ (17) (18) ∀z ∈ Z. The results in Alvarez and Jermann (1996) indicate that during recession the allocations do not depend on past history.9 in Alvarez and Jermann (1996). See the proof of Proposition 3. Deﬁning a new set of “hat” variables as c c u(c) = e1−γ u( ) = e1−γ u(ˆ) e ˆ U i (z . during the expansion there are two endogenous states: one 32 . 1) ˆ ˆ w(z ) ≥ U 2 (z . z. e ) = (e )1−γ U i (z . z. ˆ w = (e )1−γ w= and using the above proposition in the following way: e e1−γ w. z. ye) = y 1−γ i W (w. 1).{w(z )} ˆ c ˆ max u(ˆ1 ) + β c z ∈Z V (w(z ). z. 2 for i = 1. e). ∀z ∈ Z. z . e). e) w. z. z . z. 1) ˆ 1−γ e e w . 1) for i = 1. it is needed to assume that max β z∈Z z ∈Z λ(z )1−γ π(z |z) < 1. 2 e1−γ w = e1−γ w ˆ e1−γ (e )1−γ w = (e )1−γ w .ˆ2 . Proof. V (y 1−γ w. ˆ V (w . e) = e1−γ V (w. ye) = yC (w. 1) ˆ In order to guarantee the above maximization problem is well-deﬁned. z . z. 2 w. ye) = y 1−γ U i (z. z. e) C (y W (y i 1−γ 1−γ for i = 1.

“Eﬃciency. From now on. 775–797. 14. “Risk Premiums in the Term Structure: Evidence from Artiﬁcial Economies”. 24. ﬁve equations are given by (18). Jermann (2001). it is possible to solve for the allocations and prices simply by solving for at most two systems of nonlinear equations. z4 (zt = ze . there is a set of inequalities that determines which case is valid. S.E. (17) and symmetry imply that c1 + c5 = 1. and S. in the constrained case. and U. Alvarez. and another where zt = ze and zt−1 = zlh . Aiyagari. c2 = c1 and c4 = c5 . one needs only to solve for the unconstrained system and check whether w(z2 ) ≥ U 2 (z2 ). The two missing equations depend on whether the agents are constrained when entering the expansion state.J. zt−1 = ze ) . Jermann (1996). 659–684. during the expansion neither agent has a reason to trade: c3 = 0. F. zt−1 = zhl ) z3 = (zt = ze . Backus..B. (1993). and U. and U. 33 . Journal of Monetary Economics. D. 1117–1151. Alvarez. “Inﬂation and the Yield Curve”. 68. The ﬁrst system corresponds to the case in which the participation constraints are not binding.where zt = ze and zt−1 = zhl . In addition to the nonlinear equation. F.K. Federal Reserve Bank of Atlanta Economic Review. If this is not the case.J. Econometrica. May/June. From now on. To solve for the allocations. Quarterly Journal of Economics.J. Review of Financial Studies. Alvarez. Journal of Monetary Economics. ˆ cn will denote c2 (zn ) and “hats” will be dropped from other variables as well. 3–33. Abken. the solution is given by the constrained case. “Asset Pricing when Risk Sharing is Limited by Default”.R.W. it follows that agent 1’s consumption is always the aggregate endowment less agent 2’s consumption. In both the unconstrained case and the constrained case. From feasibility (17) and non-satiation. and. The Wharton School. zt−1 = zlh ) z5 (zt = zlh ) Moreover. 371–399. (1994). “Quantitative Asset Pricing Implications of Endogenous Solvency Constraints”. P.5. “Uninsured Idiosyncratic Risk and Aggregate Saving”. A. (1999).A. and the second system corresponds to the case in which the agents are constrained when entering the boom period. 109. Equilibrium. 13–31. Hence. and the participation constraint (19) holds with equality when agent 2 receives the most favorable shock: w(z5 ) = U 2 (z5 ). F. “Risk Premia and Term Premia in General Equilibrium”. Zin (1989). A. In the unconstrained case. References Abel. Gregory. w(z2 ) = U 2 (z2 ) and c2 + c4 = 1. Mimeo. the states are ordered as follows: (zt = zhl ) z1 z2 (zt = ze . and Asset Pricing with Risk of Default”. both systems have 10 equations in 10 unknowns: agent 2’s consumption and continuation utility in each of the ﬁve states. Jermann (2000). 43.

Preston (1998). 4088. “Some Lessons from the Yield Curve”. 338–375. (1986). Buraschi. and L. Review of Economic Studies. (1991). “Yield Curve Forecasts of Inﬂation: A Cautionary Tale”. 7.H. Mimeo. “A Monetary Explanation of the Equity Premium. Lo. (1997). “By Force of Habit: A Consumption-Based Explanation of Aggregate Stock Market Behavior”. Journal of Labor Economics. Barton. (1995). Constantinides. R. R. New Jersey. “Bond and Stock Returns in a Simple Exchange Model”. University of Chicago. Cochrane. A. National Bureau of Economic Research.Y. R. 104. “Does Intrinsic Habit Formation Actually Resolve the Equity Premium Puzzle?”. “Asset Pricing Explorations for Macroeconomics”. Brown.-F. MacKinlay (1997). D. 603–640. Campbell. “Asset Pricing with Heterogeneous Consumers and Limited Participation: Empirical Evidence”. “Financial Investment Opportunities and the Macroeconomy”.Y. Bansal. 265–296. Journal of Finance. “An Intertemporal Asset Pricing Model with Stochastic Consumption and Investment Opportunities”. A. Term Premium. Y. S. 101. Quarterly Journal of Economics. Harvard University.Y. Princeton University Press. (1979).G. and J. “The Term Structure of Real Interest Rates and the Cox. D. Campbell. The Econometrics of Financial Markets. Cochrane. 3–42. (1962).Baker. 1135–1171. 15. J.H. and E. 39. 113.A. “The Cyclical Properties of Consumption Growth and the Real Term Structure”. D.W. 107. and the Bond Market: A Study of UK Nominal and Index-Linked Government Bond Prices”. Campbell (1997). A. “Inﬂation. University of Texas. 145–172. Hansen (1992). Jiltsov (2003). 46. Chapman.H. J. Campbell. “Growth-Rate Heterogeneity and the Covariance Structure of Life-Cycle Earnings”. Blundell R. and I. 58. Coleman II (1996). and the Business Cycle”. (1997). “Consumption Inequality and Income Uncertainty” Quarterly Journal of Economics. Journal of Political Economy.C. Chen. Geczy (1999). 351–357.G. Campbell. 35. and Ross Model”. and S. National Bureau of Economic Research. New England Economic Review. March. 6–22. “New Facts in Finance”. 34 . N. 785–803. Mimeo. 49.. J. and R. “Persistence in a Chain of Multiple Events when There Is Simple Persistence”. 3–15.. J.Y. “Ten Years of the Real Term Structure: 1984–1994”. David. and Risk-Free Rate Puzzles”. 39. and S. London Business School.Y. “Yield Spreads and Interest Rate Movements: A Bird’s Eye View”. and A. J. Campbell. Ingersoll. and J. Journal of Financial Economics. Brown. and C. May/June.M. Mimeo. Real Rates. and W. Cochrane (1999). J. (1999). “Habit Persistence and the Nominal Term Structure of Interest Rates ”. Princeton. 529–554. (2001). Working Paper No. and A. Campbell.H.T. Schaefer (1996). F.M. M. University of Chicago. Journal of Financial Economics. Breeden. Working Paper No. Blough. 205–251. Journal of Fixed Income Research. 361–383. Schaefer (1994). (1994).P.H. Barr. Mimeo. Shiller (1991). 5031. E. Chapman. “Asset Prices. Brav. J.R. Journal of Political Economy. Y.. Fix. G.M. Journal of Monetary Economics. (1998). D. Mimeo. 495–514.J. Consumption.J. Biometrika. J.A. Journal of Monetary Economics.

Journal of Political Economy. and Inﬂation Risk Premia”. Derry (1994).D. 22. and G. Bliss (1987). 405–443. and S. 35 .Constantinides. Singleton (2002). Froot. J. 385–408. Time-varying Risk Premia. G. (1990). P. and R. and A. Journal of Economic Dynamics and Control. Journal of Finance. “The Real Term Structure and Consumption Growth”. New York University. “The Term Structure of Interest Rates in Real and Monetary Economies”. 19. 219–240. 269–282.P.. (1984). Heaton. 443–487. 1–17. Expected Inﬂation. Journal of Economic Dynamics and Control. Journal of Financial Economics. and K. “Deriving Estimates of Inﬂation Expectations from the Prices of UK Government Bonds”. 25. J. 63. Fama. 415–441. 13. J. Journal of Economic Dynamics and Control. (2002). “The Information in the Term Structure”. 59–76. Ross (1996). Harvey. and D. and Aﬃne Models of the Term Structure”. J. 104. T. 46. 519–543. 283–304. “Habit Formation: A Resolution of the Equity Premium Puzzle”. Journal of Finance. (2003). Q.. Johnsen. Fama. 77. Mehra (1990). 187–218. Bank of England. K. “The Role of Monetary Policy”. Econometrica. E. “Expectations Puzzle. (1990). “Long Forward and Zero-Coupon Rates Can Never Fall”. G. Ingersoll. 25. Journal of Financial Economics. Journal of Monetary Economics. Q. E. 246–273. C. 305–333.A. (1995).R. Fama. Hardouvelis (1991). E. (1975).F. Constantinides. (1998). 104. Inﬂation. Duﬃe (1996). “Short Term Interest Rates as Predictors of Inﬂation”.F. “Business Conditions and Expected Returns on Stocks and Bonds”. “Term Structure Dynamics in a Model with Stochastic Internal Habit”. “Real Rates. Fama. 14. W. Jr. M.B. Journal of Finance. 57.. French (1989). Duﬀee. 571–596. G. Estrella A. 98. 53. 601–620. “On the Term Structure of Interest Rates”. Cox. Fama. Journal of Political Economy. E. Heaton. Lucas (1992). Journal of Finance.F. Mimeo. American Economic Review.J. Donaldson.D.A. 555–580. Journal of Financial Economics. Dai. 23. “The Term Structure as a Predictor of Real Economic Activity”. and S. and K. Friedman. Ross (1985). J. 63. 69. (1968).F. “The Information in Long-Maturity Forward Rates”. (1988).J. M. American Economic Review. and D. Dai. 1–25. 23–50. 53. “Asset Pricing with Heterogeneous Consumers”. Journal of Financial Economics. (1989).J.R. and R.. Deacon. Dybvig. American Economic Review. Journal of Business. M. “Evaluating the Eﬀects of Incomplete Markets on Risk Sharing and Asset Pricing”. and Real Returns”. “Term-Structure Forecasts of Interest Rates. den Haan. Lucas (1996). and D. 44.F. Working Paper No. J. Ingersoll. Journal of Political Economy. “Term premia and interest rate forecasts in aﬃne models”.R. 65. 909–940.R. 16. 509–528. “A Theory of the Term Structure of Interest Rates”. “New Hope for the Expectations Hypothesis of the Term Structure of Interest Rates”. “The Eﬀects of Incomplete Insurance Markets and Trading Costs in a Consumption-Based Asset Pricing Model”. E. Evans.

Kocherlakota. “Yield Curve”. University of Illinois. R. “Asset Pricing in Economies with Frictions”. C. Shiller. 14.Y. Mankiw.W. 407–425.R. G. Mishkin. 14. and M. Friedman and F. Review of Economic Studies. (1976). 195–215. Journal of Monetary Economics.H.F.. Working Paper No. “The Term Structure of Interest Rates” in B. 865–888.. and D. King. “The Real Interest Rate: An Empirical Investigation”.S. Journal of Finance. “Implications of Eﬃcient Risk Sharing without Commitment”. 59–80. Journal of Labor Economics.). Journal of Monetary Economics. B. 46. (1978). H. Shiller.J. Mishkin. Review of Economic Studies. Otrok. Schoenholtz (1983). 2. and E. (1987). 105. Watson (1996).S. (1981). 60. M. Mishkin (1991). 1190–1219. M. Kandel. Journal of Financial Economics. Dynamic Macroeconomic Theory. F.R. and C. LeRoy. 30. 64. 35–53. A.E. Kehoe. 29.. “Asset Prices in an Exchange Economy”. Melino. P. and K. Luttmer. 51. (1990a). Ofer. 145–161. 436–446.Jorion. F. C.. 151–200. 3550. “A Multicountry Comparison of Term-Structure Forecasts at Long Horizons”. “Debt-Constrained Asset Markets”. 15. T. “Forward Rates and Future Policy: In36 . Sarig (1996). “Is the Fisher Eﬀect for Real?”. C. R.G. Econometrica. Bell Journal of Economics and Management Science. Journal of Political Economy. Mishkin. (1979). and O.S. Sargent. Hahn (eds.I. R. “Money. Mishkin. R. T. (1996). United States of America. and the Business Cycle”. Mishkin. 49. 7. (1992). Levine (1993). (1990). “The Valuation of Uncertain Income Streams and the Pricing of Options”. (1991). and F. “The Term Structure of Real Interest Rates: Theory and Evidence from UK aa Index-Linked Bonds”. 205–225. Lucas. (2000). (1973). Netherlands. “The Equity Premium and the Concentration of Aggregate Shocks”.S. E. N. 87. “The Volatility of Long-Term Interest Rates and Expectations Models of the Term Structure”. 335–366. Prescott (1985) “The Equity Premium: A Puzzle”. North-Holland.I. 10. Carnagie-Rochester Conference Series on Public Policy. 17. (1996). Handbook of Monetary Economics. National Bureau of Economic Research. 1261–1288. Journal of International Money and Finance. F. Journal of Monetary Economics. “The Information in the Longer Maturity Term Structure about Future Inﬂation”. Shiller. A. Quarterly Journal of Economics. Journal of Financial Economics. 2–22. Ravikumar. Oﬃce of Research Working Paper 00–0120. R. “Risk Aversion and the Martingale Property of Stock Prices”. and E. 815–828. F.S. 15. (1990b). (1988) “The Term Structure of Interest Rates: Evidence and Theory”. 211–219. 1429–1446. Whitehouse (1996). 78. Review of Economics and Statistics. “Habit Formation: A Resolution of The Equity Premium Puzzle?”. 63. “A Multi-Country Study of the Information in the Shorter Maturity Term Structure about Future Inﬂation”.I. Whiteman (2002). Sepp¨l¨. Econometrica. 1–25. Harvard University Press.J. Meghir. 595–609. International Economic Review. F. R. Mehra. S.K. S. Jr. “Real Interest Rates and Inﬂation: An Ex-Ante Empirical Analysis”. Campbell. 1439–1467. Rubinstein. 1. Journal of Economic Surveys. Prices. J. J. “The Evolution of Wages in the United Kingdom: Evidence from Micro Data”.

M. Technical Report and Appendices. Colchester. 37 . Svensson (1997). Journal of Business. Svensson.O. 40. Journal of Monetary Economics. Storesletten. (1999). White. 2187–2209. “Asset Pricing with Idiosyncratic Risk and Overlapping Generations”. British Household Panel Survey User Manual Volume A: Introduction.. K. 63. 4871. Weil. 401–421.) with J. and E. “Estimating and Interpreting Forward Interest Rates: Sweden 1992– 1994”. Journal Monetary Economics.A. Svensson. (1990). University of Essex. Wachter. C. Prentice-Lane (1998). L. Econometrica. “Monetary Policy with Flexible Exchange Rates and Forward Interest Rates as Indicators”. Brookings Papers on Economic Activity. 48. “Habit Formation and Returns on Bonds and Stocks”. (2002). 24. Mimeo. Yaron (1999). Mimeo. Buck. Mimeo.E. 373–398. “The Real Term Structure of Interest Rates and Aggregate Economic Fluctuations”. S¨derlind. and A. O. “The Equity Premium Puzzle and the Risk-Free Rate Puzzle”. (1994b). Brice. Vigneron. (1980). H. University of Chicago. Telmer.E. “New Techniques to Extract Market Expectations from o Financial Instruments”. 383–429. (1989).E. Working Paper No. Zhang. P. Taylor.O. 941.I.O. Woodard. N.G.terpreting the Term Structure of Interest Rates”. Discussion Paper No. 173–225. “A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity”. New York University. National Bureau of Economic Research. and L. L. 817–838. “Endogenous Borrowing Constraints with Incomplete Markets”. (1994a).F. H.H. T. Journal of Finance. Carnegie-Mellon University. (ed. J. 2. 52. “The Real Thing: A Dynamic Proﬁle of the Term Structure of Real Interest Rates and Inﬂation Expectations in the United Kingdom 1982–1989. Centre for Economic Policy Research. (1997). P.

- DecomposingTheYieldCurve - Cochrane, Piazzesi (2008)
- optifin - bdt
- US Federal Reserve
- Lecture 2
- GW JEL Draft
- 19601819-Lehman-Brothers-a-Model-of-the-Yield-Curve-With-TimeVarying-Interest-Rate-Targets
- Viney - Financial Markets and Institutions
- Arbitrage-Free Binomial Models of the Term Structure
- Chap 013
- 68670_2000-2004
- David Aranson MTA Journal Patterns
- Vasicek Print
- Hw Answers Chapter 6 Buad 3040
- Tutorial Answers Week 2
- JPM Smooth Interpolation of Zero Curves
- Section 4
- Vasicek Model, Simple
- SSRN-id2403067
- 68302_1965-1969
- Done - FIS_10
- FRBSF Yield Curve
- Lecture 2 - Determining the Term Structure of Interest Rate Through Bootstrapping
- eao97-04
- 20060518
- Ceiops Calibration Paper QIS 5
- 7.0Term Structure of Interest Rates
- MiER72
- Yield Curve
- Risk Db 3 Solution
- Finance+NPV Nt
- Strucutre Term of Interest Rates

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd