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

# Risk Aversion and the Labor Margin

**in Dynamic Equilibrium Models
**

Eric T. Swanson

Federal Reserve Bank of San Francisco

eric.swanson

@

sf.frb.org

http://www.ericswanson.us

Abstract

The household’s labor margin has a substantial eﬀect on risk aversion, and

hence asset prices, in dynamic equilibrium models even when utility is addi-

tively separable between consumption and labor. This paper derives simple,

closed-form expressions for risk aversion that take into account the house-

hold’s labor margin. Ignoring this margin can wildly overstate the house-

hold’s true aversion to risk. Risk premia on assets priced with the stochastic

discount factor increase essentially linearly with risk aversion, so measuring

risk aversion correctly is crucial for asset pricing in the model. Closed-form

expressions for risk aversion in models with generalized recursive preferences

and internal and external habits are also derived.

JEL Classiﬁcation: E44, D81

Version 1.7

April 30, 2010

I thank Ivan Jaccard, Martin Schneider, Harald Uhlig, Elmar Mertens, Marcelo Ferman,

Jonas Fisher, Edward Nelson, Glenn Rudebusch, John Williams, and seminar participants

at the Federal Reserve Bank of San Francisco and Universit`a Bocconi for helpful dis-

cussions, comments, and suggestions. The views expressed in this paper, and all errors

and omissions, should be regarded as those solely of the author, and are not necessarily

those of the individuals listed above, the management of the Federal Reserve Bank of San

Francisco, or any other individual in the Federal Reserve System.

1

1. Introduction

In a static, one-period model with household utility u(·) deﬁned over a single consumption

good, Arrow (1964) and Pratt (1965) deﬁned the coeﬃcients of absolute and relative risk

aversion, −u

(c)/u

(c) and −c u

(c)/u

**(c). Diﬃculties immediately arise, however, when
**

one attempts to generalize these concepts to the case of many periods or many goods (e.g.,

Kihlstrom and Mirman, 1974). These diﬃculties are particularly pronounced in a dynamic

equilibrium model with labor, in which there is a double inﬁnity of goods to consider—

consumption and labor in every future period and state of nature—all of which may vary in

response to a typical shock to household income or wealth.

The present paper shows how to compute risk aversion in dynamic equilibrium models

in general. First, we verify that risk aversion depends on the partial derivatives of the

household’s value function V with respect to wealth a—that is, the coeﬃcients of absolute

and relative risk aversion are essentially −V

aa

/V

a

and −aV

aa

/V

a

, respectively. Even though

closed-form solutions for the value function do not exist in general, we nevertheless can derive

simple, closed-form expressions for risk aversion because derivatives of the value function are

much easier to compute than the value function itself. For example, in many DSGE models

the derivative of the value function with respect to wealth equals the current-period marginal

utility of consumption (Benveniste and Scheinkman, 1979).

The main result of the paper is that the household’s labor margin has substantial

eﬀects on risk aversion, and hence asset prices. Even when labor and consumption are ad-

ditively separable in utility, they remain connected by the household’s budget constraint: in

particular, the household can absorb income shocks either through changes in consumption,

changes in hours worked, or some combination of the two. This ability to absorb shocks

along either or both margins greatly alters the household’s attitudes toward risk. For exam-

ple, if the household’s utility kernel is given by u(c

t

, l

t

) = c

1−γ

t

/(1 − γ) − χl

t

, the quantity

−c u

11

/u

1

= γ is often referred to as the household’s coeﬃcient of relative risk aversion, but

in fact the household is risk neutral with respect to gambles over income or wealth—the

proper measure of risk aversion for asset pricing, as we show in Section 2. Intuitively, the

household is indiﬀerent at the margin between using labor or consumption to absorb a shock

to income or wealth, and the household in this example is clearly risk neutral with respect to

gambles over hours. More generally, when u(c

t

, l

t

) = c

1−γ

t

/(1−γ)−χ

0

l

1+χ

t

/(1+χ), risk aver-

2

sion equals (γ

−1

+χ

−1

)

−1

, a combination of the parameters on the household’s consumption

and labor margins, reﬂecting that the household absorbs shocks using both margins.

1

While modeling risk neutrality is not a main goal of the present paper, risk neutrality

nevertheless can be a desirable feature for some applications, such as labor market search or

ﬁnancial frictions, since it allows for closed-form solutions to key features of the model.

2

A

contribution of the present paper is to show ways to model risk neutrality that do not require

utility to be linear in consumption, which has undesirable implications for interest rates and

consumption growth. Instead, any utility kernel with zero discriminant can be used.

A ﬁnal result of the paper is that risk premia computed using the Lucas-Breeden

stochastic discounting framework are essentially linear in risk aversion. That is, measuring

risk aversion correctly—taking into account the household’s labor margin—is necessary for

understanding asset prices in the model. Since much recent research has focused on bringing

dynamic stochastic general equilibrium (DSGE) models into closer agreement with asset

prices,

3

it is surprising that so little attention has been paid to measuring risk aversion

correctly in these models. The present paper aims to ﬁll that void.

There are a few previous studies that extend the Arrow-Pratt deﬁnition beyond the

one-good, one-period case. In a static, multiple-good setting, Stiglitz (1969) measures risk

aversion using the household’s indirect utility function rather than utility itself, essentially

a special case of Proposition 1 of the present paper. Constantinides (1990) measures risk

aversion in a dynamic, consumption-only (endowment) economy using the household’s value

function, another special case of Proposition 1. Boldrin, Christiano, and Fisher (1997) apply

Constantinides’ deﬁnition to some very simple endowment economy models for which they

can compute closed-form expressions for the value function, and hence risk aversion. The

present paper builds on these studies by deriving closed-form solutions for risk aversion in

dynamic equilibrium models in general, demonstrating the importance of the labor margin,

and showing the tight link between risk aversion and asset prices in these models.

1

Note that the intertemporal elasticity of substitution in this example is still 1/γ, so a corollary of this

result is that risk aversion and the intertemporal elasticity of substitution are nonreciprocal when labor

supply can vary.

2

See, e.g., Mortensen and Pissarides (1994), and Bernanke, Gertler, and Gilchrist (1999).

3

See, e.g., Boldrin, Christiano, and Fisher (2001), Tallarini (2000), Rudebusch and Swanson (2008, 2009),

Van Binsbergen, Fernandez-Villaverde, Koijen, and Rubio-Ramirez (2008), and Backus, Routledge, and Zin

(2009).

3

The remainder of the paper proceeds as follows. Section 2 presents the main ideas of

the paper, deriving Arrow-Pratt risk aversion in dynamic equilibrium models for the time-

separable expected utility case and demonstrating the importance of risk aversion for asset

pricing. Section 3 extends the analysis to the case of generalized recursive preferences (Ep-

stein and Zin, 1989), which have been the focus of much recent research at the boundary

between macroeconomics and ﬁnance. Section 4 extends the analysis to the case of internal

and external habits, two of the most common intertemporal nonseperabilities in preferences

in both the macroeconomics and ﬁnance literatures. Section 5 discusses some general im-

plications and concludes. An Appendix provides details of derivations and proofs that are

outlined in the main text.

2. Time-Separable Expected Utility Preferences

To highlight the intuition and methods of the paper, we consider ﬁrst the case where the

household has additively time-separable expected utility preferences.

2.1 The Household’s Optimization Problem and Value Function

Time is discrete and continues forever. At each time t, the household seeks to maximize the

expected present discounted value of utility ﬂows:

E

t

∞

¸

τ=t

β

τ−t

u(c

τ

, l

τ

), (1)

subject to the sequence of asset accumulation equations:

a

τ+1

= (1 + r

τ

)a

τ

+ w

τ

l

τ

+ d

τ

−c

τ

, τ = t, t + 1, . . . (2)

and transversality condition:

lim

T→∞

T

¸

τ=t

(1 + r

τ+1

)

−1

a

T+1

≥ 0, (3)

where E

t

denotes the mathematical expectation conditional on the household’s information

set at the beginning of period t, β ∈ (0, 1) is the household’s discount factor, c

t

≥ 0 and

l

t

∈ [0,

¯

l] are the household’s choice of consumption and labor in period t, a

t

is the household’s

beginning-of-period assets, and w

t

, r

t

, and d

t

denote the real wage, interest rate, and net

4

transfer payments at time t. There is a ﬁnite-dimensional Markovian state vector θ

t

that is

exogenous to the household and governs the processes for w

t

, r

t

, and d

t

. Conditional on θ

t

,

the household knows the time-t values for w

t

, r

t

, and d

t

. The state vector and information

set of the household’s optimization problem at each date t is thus (a

t

; θ

t

).

We make the following regularity assumptions regarding the utility kernel u:

Assumption 1. The function u : R

+

× [0,

¯

l) → R,

¯

l ∈ (0, ∞], is increasing in its ﬁrst

argument, decreasing in its second, twice-diﬀerentiable, and concave.

Assumption 1 guarantees the existence of a unique optimal choice for (c

t

, l

t

) at each point in

time, given (a

t

; θ

t

). Note that since u is increasing in consumption (i.e., there is no satiation),

condition (3) holds with equality at the optimum.

Let V (a

t

; θ

t

) denote the value function for the household’s optimization problem. Then

V satisﬁes the Bellman equation:

V (a

t

; θ

t

) = max

c

t

,l

t

u(c

t

, l

t

) + βE

t

V (a

t+1

; θ

t+1

), (4)

where a

t+1

is given by (2). Letting c

∗

t

≡ c

∗

(a

t

; θ

t

) and l

∗

t

≡ l

∗

(a

t

; θ

t

) denote the household’s

optimal choices of c

t

and l

t

as functions of the state (a

t

; θ

t

), V can be written as:

V (a

t

; θ

t

) = u(c

∗

t

, l

∗

t

) + βE

t

V (a

∗

t+1

; θ

t+1

), (5)

where a

∗

t+1

≡ (1 + r

t

)a

t

+ w

t

l

∗

t

+ d

t

−c

∗

t

.

To avoid having to consider boundary solutions, we make the following standard as-

sumption:

Assumption 2. For any feasible state (a

t

; θ

t

), the household’s optimal choice (c

∗

t

, l

∗

t

) lies in

the interior of R

+

×[0,

¯

l).

Intuitively, Assumption 2 requires the partial derivatives of u to grow suﬃciently large toward

the boundary that only interior solutions for c

∗

t

and l

∗

t

are optimal for the set of possible

(a

t

; θ

t

) that the household may face.

Assumptions 1–2 also guarantee that V is continuously diﬀerentiable and satisﬁes the

Benveniste-Scheinkman equation, but we will require slightly more than this below:

Assumption 3. The value function V (·; ·) is twice-diﬀerentiable.

Assumption 3 also implies diﬀerentiability of the optimal policy functions, c

∗

and l

∗

. San-

tos (1991) provides relatively mild suﬃcient conditions for Assumption 3 to be satisﬁed;

intuitively, u must be strongly concave.

5

2.2 Representative Household and Steady State Assumptions

Up to this point, we have considered the case of a single household, leaving the other house-

holds of the model and the production side of the economy unspeciﬁed. Implicitly, the other

households and production sector jointly determine the process for θ

t

(and hence w

t

, r

t

,

and d

t

), and much of the analysis below does not need to be any more speciﬁc about these

processes than this. However, to move from general expressions for household risk aversion to

more concrete, closed-form expressions, we adopt two standard assumptions from the DSGE

literature.

4

Assumption 4. The household is atomistic and representative.

Assumption 4 implies that the individual household’s choices for c

t

and l

t

have no eﬀect on

the aggregate quantities w

t

, r

t

, d

t

, and θ

t

. It also implies that, when the economy is at the

nonstochastic steady state (described shortly), any individual household ﬁnds it optimal to

choose the steady-state values of c and l given a and θ.

Assumption 5. The model has a nonstochastic steady state, or a balanced growth path that

can be renormalized to a nonstochastic steady state after a suitable change of variables. At

the nonstochastic steady state, x

t

= x

t+1

= x

t+k

for k = 1, 2, . . . , and x ∈ {c, l, a, w, r, d, θ},

and we drop the subscript t to denote the steady-state value.

It is important to note that Assumptions 4–5 do not prohibit us from oﬀering an

individual household a hypothetical gamble of the type described below. The steady state

of the model serves only as a reference point around which the aggregate variables w, r,

d, and θ and the other households’ choices of c, l, and a can be predicted with certainty.

This reference point is important because it makes it much easier to compute closed-form

expressions for many features of the model.

2.3 The Coeﬃcient of Absolute Risk Aversion

The household’s risk aversion at time t generally depends on the household’s state vector

at time t, (a

t

; θ

t

). Given this state, we consider the household’s aversion to a hypothetical

4

Alternative assumptions about the nature of the other households in the model or the production sector

may also allow for closed-form expressions for risk aversion. However, the assumptions used here are standard

and thus the most natural to pursue.

6

one-shot gamble in period t of the form:

a

t+1

= (1 + r

t

)a

t

+ w

t

l

t

+ d

t

−c

t

+ σε

t+1

, (6)

where ε

t+1

is a random variable with mean zero and unit variance that represents the gam-

ble.

5

A few words about (6) are in order: First, the gamble is dated t + 1 to clarify that its

outcome is not in the household’s information set at time t. Second, neither a

t

nor c

t

can be

the subject of the gamble: a

t

is a state variable known with certainty at t, and c

t

is a choice

variable under control of the household at time t. However, (6) is clearly equivalent to a

gamble over net transfers d

t

or asset returns r

t

, both of which are exogenous to the house-

hold at time t. Indeed, thinking of the gamble as being over r

t

helps clarify the connection

between (6) and asset prices, to which we will return in Section 2.6, below. As shown there,

the gamble in (6) is exactly the right framework for thinking about asset prices.

Following Arrow (1964) and Pratt (1965), we can ask what one-time fee μ the household

would be willing to pay in period t to avoid the gamble in (6):

a

t+1

= (1 + r

t

)a

t

+ w

t

l

t

+ d

t

−c

t

−μ. (7)

The quantity 2dμ/dσ

2

, for inﬁtesimal dμ and dσ that make the household just indiﬀerent

between (6) and (7), is the household’s coeﬃcient of absolute risk aversion.

6

Proposition 1. The household’s coeﬃcient of absolute risk aversion with respect to the

gamble described in (6) is given by:

−E

t

V

11

(a

∗

t+1

; θ

t+1

)

E

t

V

1

(a

∗

t+1

; θ

t+1

)

, (8)

where V

1

and V

11

denote the ﬁrst and second partial derivatives of V with respect to its ﬁrst

argument. Evaluated at the steady state, (8) simpliﬁes to:

−V

11

(a; θ)

V

1

(a; θ)

. (9)

Proof: See Appendix.

5

The gamble ε

t+1

is assumed to be independent of the exogenous state variables θ

τ

for all times τ, and

independent of the household’s variables a

τ

, c

τ

, and l

τ

for all τ ≤ t.

6

We defer discussion of relative risk aversion until the next subsection because deﬁning total household

wealth is complicated by the presence of human capital—that is, the household’s labor income.

7

Equations (8)–(9) are essentially Constantinides’ (1990) deﬁnition of risk aversion, and

have obvious similarities to Arrow (1964) and Pratt (1965). Here, of course, it is the curvature

of the value function V with respect to assets that matters, rather than the curvature of the

utility kernel u with respect to consumption.

7

Deriving the coeﬃcient of absolute risk aversion in Proposition 1 is simple enough, but

the problem with (8)–(9) is that closed-form expressions for V (and hence V

1

and V

11

) do

not exist in general, even for the simplest DSGE models. This diﬃculty may help to explain

the popularity of “shortcut” approaches to measuring risk aversion, notably −u

11

(c

∗

t

, l

∗

t

)/

u

1

(c

∗

t

, l

∗

t

), which has no clear relationship to (8)–(9) except in the one-good one-period case.

Boldrin, Christiano, and Fisher (1997) derive closed-form solutions for V —and hence risk

aversion—for some very simple, consumption-only endowment economy models. This ap-

proach is a nonstarter for even the simplest DSGE models that include labor.

We solve this problem by observing that V

1

and V

11

often can be computed even when

closed-form solutions for V cannot be. For example, the Benveniste-Scheinkman equation:

V

1

(a

t

; θ

t

) = (1 + r

t

) u

1

(c

∗

t

, l

∗

t

), (10)

states that the marginal value of a dollar of assets equals the marginal utility of consumption

times 1 + r

t

(the interest rate appears here because beginning-of-period assets in the model

generate income in period t). In (10), u

1

is a known function. Although closed-form solutions

for the functions c

∗

and l

∗

are not known in general, the points c

∗

t

and l

∗

t

often are known—

for example, when they are evaluated at the nonstochastic steady state, c and l. Thus, we

can compute V

1

at the nonstochastic steady state by evaluating (10) at that point.

We compute V

11

by noting that (10) holds for general a

t

; hence we can diﬀerentiate

(10) to yield:

V

11

(a

t

; θ

t

) = (1 + r

t

)

¸

u

11

(c

∗

t

, l

∗

t

)

∂c

∗

t

∂a

t

+ u

12

(c

∗

t

, l

∗

t

)

∂l

∗

t

∂a

t

. (11)

All that remains is to ﬁnd the derivatives ∂c

∗

t

/∂a

t

and ∂l

∗

t

/∂a

t

.

We solve for ∂l

∗

t

/∂a

t

by diﬀerentiating the household’s intratemporal optimality con-

dition:

−u

2

(c

∗

t

, l

∗

t

) = w

t

u

1

(c

∗

t

, l

∗

t

), (12)

7

Arrow (1964) and Pratt (1965) occasionally refer to utility as being deﬁned over “money”, so one could

argue that they always intended for risk aversion to be measured using indirect utility or the value function.

8

with respect to a

t

, and rearranging terms to yield:

∂l

∗

t

∂a

t

= −λ

t

∂c

∗

t

∂a

t

, (13)

where

λ

t

≡

w

t

u

11

(c

∗

t

, l

∗

t

) + u

12

(c

∗

t

, l

∗

t

)

u

22

(c

∗

t

, l

∗

t

) + w

t

u

12

(c

∗

t

, l

∗

t

)

=

u

1

(c

∗

t

, l

∗

t

)u

12

(c

∗

t

, l

∗

t

) −u

2

(c

∗

t

, l

∗

t

)u

11

(c

∗

t

, l

∗

t

)

u

1

(c

∗

t

, l

∗

t

)u

22

(c

∗

t

, l

∗

t

) −u

2

(c

∗

t

, l

∗

t

)u

12

(c

∗

t

, l

∗

t

)

. (14)

Note that, if consumption and leisure in period t are normal goods, then λ

t

must be positive.

8

It now only remains to solve for the derivative ∂c

∗

t

/∂a

t

.

Intuitively, ∂c

∗

t

/∂a

t

should not be too diﬃcult to compute: it is just the household’s

marginal propensity to consume today out of a change in assets, which we can deduce from

the household’s Euler equation and budget constraint. Diﬀerentiating the Euler equation:

u

1

(c

∗

t

, l

∗

t

) = βE

t

(1 + r

t+1

) u

1

(c

∗

t+1

, l

∗

t+1

), (15)

with respect to a

t

yields:

9

u

11

(c

∗

t

, l

∗

t

)

∂c

∗

t

∂a

t

+ u

12

(c

∗

t

, l

∗

t

)

∂l

∗

t

∂a

t

= βE

t

(1 + r

t+1

)

¸

u

11

(c

∗

t+1

, l

∗

t+1

)

∂c

∗

t+1

∂a

t

+ u

12

(c

∗

t+1

, l

∗

t+1

)

∂l

∗

t+1

∂a

t

(16)

Substituting in for ∂l

∗

t

/∂a

t

gives:

(u

11

(c

∗

t

, l

∗

t

) −λ

t

u

12

(c

∗

t

, l

∗

t

))

∂c

∗

t

∂a

t

= βE

t

(1 +r

t+1

) (u

11

(c

∗

t+1

, l

∗

t+1

) −λ

t+1

u

12

(c

∗

t+1

, l

∗

t+1

))

∂c

∗

t+1

∂a

t

.

(17)

Evaluating (17) at steady state, β = (1 + r)

−1

, λ

t

= λ

t+1

= λ, and the u

ij

cancel, giving:

∂c

∗

t

∂a

t

= E

t

∂c

∗

t+1

∂a

t

= E

t

∂c

∗

t+k

∂a

t

, k = 1, 2, . . . (18)

∂l

∗

t

∂a

t

= E

t

∂l

∗

t+1

∂a

t

= E

t

∂l

∗

t+k

∂a

t

, k = 1, 2, . . . (19)

In other words, whatever the change in the household’s consumption today, it must be

the same as the expected change in consumption tomorrow, and the expected change in

consumption at each future date t + k.

10

8

We do not require this restriction in the analysis below, but intuitively we will think of λ > 0.

9

By ∂c

∗

t+1

/∂a

t

we mean:

∂c

∗

t+1

∂a

t

=

∂c

∗

t+1

∂a

t+1

da

∗

t+1

da

t

=

∂c

∗

t+1

∂a

t+1

¸

1 + r

t+1

+ w

t

∂l

∗

t

∂a

t

−

∂c

∗

t

∂a

t

,

and analogously for ∂l

∗

t+1

/∂a

t

, ∂c

∗

t+2

/∂a

t

, ∂l

∗

t+2

/∂a

t

, etc.

10

Note that this equality does not follow from the steady state assumption. For example, in a model with

internal habits, which we will consider in Section 4, the individual household’s optimal consumption response

to a change in assets increases with time, even starting from steady state.

9

The household’s budget constraint is implied by asset accumulation equation (2) and

transversality condition (3). Diﬀerentiating (2) with respect to a

t

, evaluating at steady state,

and applying (3), (18), and (19) gives:

1 + r

r

∂c

∗

t

∂a

t

= (1 + r) +

1 + r

r

w

∂l

∗

t

∂a

t

. (20)

That is, the expected present value of changes in household consumption must equal the

change in assets (times 1 + r) plus the expected present value of changes in labor income.

Combining (20) with (13), we can solve for ∂c

∗

t

/∂a

t

evaluated at the steady state:

∂c

∗

t

∂a

t

=

r

1 + wλ

. (21)

In response to a unit increase in assets, the household raises consumption in every period by

the extra asset income, r, adjusted downward by the amount 1 +wλ that takes into account

the household’s decrease in hours worked.

We can now compute the household’s coeﬃcient of absolute risk aversion. Substituting

(10), (11), (13)–(14), and (21) into (9), we have proved:

Proposition 2. The household’s coeﬃcient of absolute risk aversion in Proposition 1, eval-

uated at steady state, satisﬁes:

−V

11

(a; θ)

V

1

(a; θ)

=

−u

11

+ λu

12

u

1

r

1 + wλ

, (22)

where u

1

, u

11

, and u

12

denote the corresponding partial derivatives of u evaluated at the

steady state (c, l), and λ is given by (14) evaluated at steady state.

When there is no labor margin in the model, Proposition 2 has the following corollary:

Corollary 3. Suppose that l

t

is ﬁxed exogenously at l ≥ 0 for all t and the household chooses

c

t

optimally at each t given this constraint. Then the household’s coeﬃcient of absolute risk

aversion (22), evaluated at steady state, is given by:

−V

11

(a; θ)

V

1

(a; θ)

=

−u

11

u

1

r. (23)

Proof: The assumptions and steps leading up to Proposition 2, adjusted to the one-

dimensional case, are essentially the same as the above with λ

t

= 0.

Proposition 2 and Corollary 3 are remarkable. First, the household’s coeﬃcient of ab-

solute risk aversion in (23) is just the traditional measure, −u

11

/u

1

, times r, which translates

10

assets into current-period consumption. In other words, for any utility kernel u, the tradi-

tional, static measure of risk aversion is also the correct measure in the dynamic context,

regardless of whether u or the rest of the model is homothetic, whether or not we can solve

for V , and no matter what the functional forms of u and V .

More generally, when households have a labor margin, Proposition 2 shows that risk

aversion is less than the traditional measure by the factor 1 + wλ, even when consumption

and labor are additively separable in u (i.e. u

12

= 0). Even in the additively separable

case, households can partially absorb shocks to income through changes in hours worked.

As a result, c

∗

t

depends on household labor supply, so labor and consumption are indirectly

connected through the budget constraint. When u

12

= 0, risk aversion in Proposition 2 is

further attenuated or ampliﬁed by the direct interaction between consumption and labor in

utility, u

12

.

The household’s labor margin can have dramatic eﬀects on risk aversion. For example,

no matter how large the traditional measure −u

11

/u

1

, the household can still be risk neutral:

Corollary 4. The household’s coeﬃcient of absolute risk aversion (22) vanishes as the

discriminant u

11

u

22

− u

2

12

vanishes, so long as either u

1

= −u

2

or u

12

< max{|u

11

|, |u

22

|}

in the limit.

Proof: The corollary is stated as a limiting result to respect concavity in Assumption 1.

Substituting out λ and w, (22) vanishes as u

11

u

22

−u

2

12

vanishes except for the special case

u

1

= −u

2

and u

11

= −u

12

= u

22

—that is, the special case u(c, l) = ˜ u(c −l) to second order

for some function ˜ u. The corollary rules out that case by assumption.

In other words, risk aversion depends on the concavity of u in all dimensions rather

than just in one dimension. Even when u

11

is very large, the household still can be risk

neutral if u

22

is small or the cross-eﬀect u

12

is suﬃciently large. Geometrically, if there

exists any direction in (c, l)-space along which u is ﬂat, the household will optimally choose

to absorb shocks to income along that line, resulting in risk-neutral behavior.

We provide some more concrete examples of risk aversion calculations in Section 2.5,

below, after ﬁrst deﬁning relative risk aversion.

2.4 The Coeﬃcient of Relative Risk Aversion

The diﬀerence between absolute and relative risk aversion is the size of the hypothetical

gamble faced by the household. If the household faces a one-shot gamble of size A

t

in

11

period t, that is:

a

t+1

= (1 + r

t

)a

t

+ w

t

l

t

+ d

t

−c

t

+ A

t

σε

t+1

, (24)

or the household can pay a one-time fee A

t

μ in period t to avoid this gamble, then it follows

from Proposition 1 that the household’s coeﬃcient of risk aversion, 2dμ/dσ

2

, for this gamble

is given by:

−A

t

E

t

V

11

(a

∗

t+1

; θ

t+1

)

E

t

V

1

(a

∗

t+1

; θ

t+1

)

. (25)

The natural deﬁnition of A

t

, considered by Arrow (1964) and Pratt (1965), is the household’s

wealth at time t. The gamble in (24) is then over a fraction of the household’s wealth and

(25) is referred to as the household’s coeﬃcient of relative risk aversion.

In DSGE models, however, household wealth can be more diﬃcult to deﬁne because of

the presence of human capital. In these models, there are two natural deﬁnitions of human

capital, so we consequently deﬁne two measures of household wealth A

t

and two coeﬃcients

of relative risk aversion (25).

First, when the household’s time endowment is not well-deﬁned—as when u(c

t

, l

t

) =

c

1−γ

t

/(1−γ)−l

1+χ

t

and no upper bound on l

t

is speciﬁed—it is most natural to deﬁne human

capital as the present discounted value of labor income, w

t

l

∗

t

. Equivalently, total household

wealth A

t

equals the present discounted value of consumption, which follows from the budget

constraint (2)–(3). We state this formally as:

Deﬁnition 1. The household’s consumption-based coeﬃcient of relative risk aversion is

given by (25), with A

t

≡ (1+r

t

)

−1

E

t

¸

∞

τ=t

m

t,τ

c

∗

τ

, the present discounted value of household

consumption, and where m

t,τ

denotes the stochastic discount factor β

τ−t

u

1

(c

∗

τ

, l

∗

τ

)/u

1

(c

∗

t

, l

∗

t

).

The factor (1 + r

t

)

−1

in the deﬁnition expresses wealth A

t

in beginning- rather than end-

of-period-t units, so that in steady state A = c/r and the consumption-based coeﬃcient of

relative risk aversion is given by:

−AV

11

(a; θ)

V

1

(a; θ)

=

−u

11

+ λu

12

u

1

c

1 + wλ

. (26)

Alternatively, when the household’s time endowment

¯

l is well speciﬁed, we can deﬁne

human capital to be the present discounted value of the household’s time endowment, w

t

¯

l. In

thise case, total household wealth

˜

A

t

equals the present discounted value of leisure w

t

(

¯

l −l

∗

t

)

plus consumption c

∗

t

, from (2)–(3). We thus have:

12

Deﬁnition 2. The household’s leisure-and-consumption-based coeﬃcient of relative risk

aversion is given by (25), with A

t

=

˜

A

t

≡ (1 + r

t

)

−1

E

t

¸

∞

τ=t

m

t,τ

c

∗

τ

+ w

τ

(

¯

l −l

∗

τ

)

.

In steady state,

˜

A =

c + w(

¯

l − l)

**/r, and the leisure-and-consumption-based coeﬃcient of
**

relative risk aversion is given by:

−

˜

AV

11

(a; θ)

V

1

(a; θ)

=

−u

11

+ λu

12

u

1

c + w(

¯

l −l)

1 + wλ

. (27)

Of course, (26) and (27) are related by the ratio of the two gambles, (c + w(

¯

l −l))/c.

Other deﬁnitions of relative risk aversion, corresponding to alternative deﬁnitions of

wealth and the size of the gamble A

t

, are also possible, but Deﬁnitions 1–2 are the most

natural for several reasons. First, both deﬁnitions reduce to the usual present discounted

value of income or consumption when there is no human capital in the model. Second, both

measures of risk aversion reduce to the traditional −c u

11

/u

1

when there is no labor margin

in the model—that is, when λ = 0. Third, in steady state the household consumes exactly

the ﬂow of income from its wealth, rA, consistent with standard permanent income theory

(where one must include the value of leisure w(

¯

l −l) as part of consumption when the value

of leisure is included in wealth).

We close this section by noting that neither measure of relative risk aversion is recip-

rocal to the intertemporal elasticity of substitution:

Corollary 5. Evaluated at steady state: i) the consumption-based coeﬃcient of relative

risk aversion and intertemporal elasticity of substitution are reciprocal if and only if λ = 0;

ii) the leisure-and-consumption-based coeﬃcient of relative risk aversion and intertemporal

elasticity of substitution are reciprocal if and only if λ = (

¯

l −l)/c.

Proof: Note that the case w = 0 is ruled out by Assumptions 1–2. The household’s

intertemporal elasticity of substitution, evaluated at steady state, is given by

dc

∗

t+1

−

dc

∗

t

/d log(1 + r

t+1

), which equals −u

1

/

c(u

11

− λu

12

)

**by a calculation along the lines
**

of (17), holding w

t

ﬁxed but allowing l

∗

t

and l

∗

t+1

to vary endogenously. The corollary follows

from comparing this expression to (26) and (27).

2.5 Examples

Some simple examples illustrate how ignoring the household’s labor margin can lead to wildly

inaccurate measures of the household’s true attitudes toward risk.

13

Example 2.1. Consider the additively separable utility kernel:

u(c

t

, l

t

) =

c

1−γ

t

1 −γ

−χ

0

l

1+χ

t

1 + χ

, (28)

where γ, χ, χ

0

> 0. The traditional measure of risk aversion for this utility kernel is

−c u

11

/u

1

= γ, but the household’s consumption-based coeﬃcient of relative risk aversion is

given by (26):

−AV

11

V

1

=

−cu

11

u

1

1

1 + w

wu

11

u

22

=

γ

1 +

γ

χ

wl

c

. (29)

The household’s leisure-and-consumption-based coeﬃcient of relative risk aversion (27) is

not well deﬁned in this example (the household’s risk aversion can be made arbitrarily

large or small just by varying the household’s time endowment

¯

l), so we focus only on

the consumption-based measure (29).

In steady state, c ≈ wl,

11

so (29) can be written as:

−AV

11

V

1

≈

1

1

γ

+

1

χ

. (30)

Note that (30) is less than the traditional measure of risk aversion by a factor of 1 + γ/χ.

Thus, if γ = 2 and χ = 1—parameter values that are well within the range of estimates in

the literature—then the household’s true risk aversion is less than the traditional measure by

a factor of about three. This point is illustrated in Figure 1, which graphs the coeﬃcient of

relative risk aversion for this example as a function of the traditional measure, γ, for several

diﬀerent values of χ. If χ is very large, then the bias from using the traditional measure is

small because household labor supply is essentially ﬁxed.

12

However, as χ approaches 0, a

common benchmark in the literature, the bias explodes and true risk aversion approaches

zero—the household becomes risk neutral. Intuitively, households with linear disutility of

work are risk neutral with respect to gambles over wealth because they can completely oﬀset

those gambles at the margin by working more or fewer hours, and households with linear

disutility of work are clearly risk neutral with respect to gambles over hours.

Expression (30) also helps to clarify several points. First, risk aversion in the model

is a combination of both parameters γ and χ, reﬂecting that the household absorbs income

11

In steady state, c = ra + wl + d, so c = wl holds exactly if there is neither capital nor transfers in the

model. In any case, ra + d is typically small for standard calibrations in the literature.

12

Similarly, if γ is very small, the bias from using the traditional measure is small because the household

chooses to absorb income shocks almost entirely along its consumption margin. As a result, the labor margin

is again almost inoperative.

14

4

5

6

7

8

9

10

c

i

e

n

t

o

f

r

e

l

a

t

i

v

e

r

i

s

k

a

v

e

r

s

i

o

n

Ȥ = 5

Ȥ =

0

1

2

3

0 1 2 3 4 5 6 7 8 9 10

C

c

o

e

f

f

i

c

Ȗ

Ȥ =5

Ȥ =0

Ȥ =1

Ȥ =2

Ȥ =3

Ȥ =4

Figure 1. Consumption-based coeﬃcient of relative risk aversion for the utility kernel u(c

t

, l

t

) =

c

1−γ

t

/(1−γ)−χ

0

l

1+χ

t

/(1+χ) in Example 2.1, as a function of the traditional measure γ, for diﬀerent

values of χ. See text for details.

gambles along both of its margins, consumption and labor. Second, for any given γ, actual

risk aversion in the model can lie anywhere between 0 and γ, depending on χ. That is,

having an additional margin with which to absorb income gambles reduces the household’s

aversion to risk. Third, (30) is symmetric in γ and χ, reﬂecting that labor and consumption

enter symmetrically into u in this example and play an essentially equal role in absorbing

income shocks. Put another way, ignoring the labor margin in this example would be just

as erroneous as ignoring the consumption margin.

Example 2.2. Consider the King-Plosser-Rebelo-type (1988) utility kernel:

u(c

t

, l

t

) =

c

1−γ

t

(1 −l

t

)

χ(1−γ)

1 −γ

, (31)

where γ > 0, γ = 1, χ > 0,

¯

l = 1, and χ(1 − γ) < γ for concavity. The traditional

measure of risk aversion for (31) is γ, but the household’s actual leisure-and-consumption-

15

based coeﬃcient of relative risk aversion is given by:

−

˜

AV

11

V

1

=

−u

11

+ λu

12

u

1

c + w(1 −l)

1 + wλ

= γ −χ(1 −γ). (32)

Note that concavity of (31) implies that (32) is positive. As in the previous example, (32)

depends on both γ and χ, and can lie anywhere between 0 and the traditional measure γ,

depending on χ. In this example, risk aversion is less than the traditional measure by the

amount χ(1−γ). As χ approaches γ/(1−γ)—that is, as utility approaches Cobb-Douglas—

the household becomes risk neutral; in this case, household utility along the line c

t

= w

t

(1−l

t

)

is linear, so the household ﬁnds it optimal to absorb shocks to wealth along that line.

The household’s consumption-based coeﬃcient of relative risk aversion is a bit more

complicated than (32):

−AV

11

V

1

=

−u

11

+ λu

12

u

1

c

1 + wλ

=

γ −χ(1 −γ)

1 + χ

. (33)

Again, (33) is a combination of the parameters γ and χ, and can lie anywhere between 0

and γ, depending on χ. Neither (32) nor (33) equals the traditional measure γ, except for

the special case χ = 0.

2.6 Risk Aversion and Asset Pricing

In the preceding sections, we showed that the labor margin has important implications for

Arrow-Pratt risk aversion with respect to gambles over income or wealth. We now show that

risk aversion with respect to these gambles is also the right concept for asset pricing.

2.6.1 Measuring Risk Aversion with V As Opposed to u

Some comparison of the expressions −V

11

/V

1

and −u

11

/u

1

helps to clarify why the former

measure is the relevant one for pricing assets, such as stocks or bonds, in the model. From

Proposition 1, −V

11

/V

1

is the Arrow-Pratt coeﬃcient of absolute risk aversion for gambles

over income or wealth in period t. In contrast, the expression −u

11

/u

1

is the risk aversion

coeﬃcient for a hypothetical gamble in which the household is forced to consume immediately

the outcome of the gamble. Clearly, it is the former concept that corresponds to the stochastic

payoﬀs of a standard asset, such as a stock or bond, in a DSGE model. In order for −u

11

/u

1

to be the relevant measure for pricing a security, it is not enough that the security pay oﬀ in

16

units of consumption in period t +1. The household would additionally have to be prevented

from adjusting its consumption and labor choices in period t +1 in response to the security’s

payoﬀs, so that the household is forced to absorb those payoﬀs into period t+1 consumption.

It is diﬃcult to imagine such a security in the real world—all standard securities in ﬁnancial

markets correspond to gambles over income or wealth, for which the −V

11

/V

1

measure of

risk aversion is the appropriate one.

2.6.2 Risk Aversion, the Stochastic Discount Factor, and Risk Premia

Arrow-Pratt risk aversion, and hence the labor margin, is also closely tied to asset prices in

the standard Lucas-Breeden stochastic discounting framework.

Let m

t+1

= βu

1

(c

∗

t+1

, l

∗

t+1

)/u

1

(c

∗

t

, l

∗

t

) denote the household’s stochastic discount factor

and let p

t

denote the cum-dividend price of a risky asset at time t, with E

t

p

t+1

normalized

to unity. The percentage diﬀerence between the risk-neutral price of the asset and its actual

price—the risk premium on the asset—is given by:

E

t

m

t+1

E

t

p

t+1

−E

t

m

t+1

p

t+1

/E

t

m

t+1

= −Cov

t

(dm

t+1

, dp

t+1

)/E

t

m

t+1

(34)

where Cov

t

denotes the covariance conditional on information at time t, and dx ≡ x

t+1

−

E

t

x

t+1

, x ∈ {m, p}. For small changes dc

∗

t+1

and dl

∗

t+1

, we have, to ﬁrst order:

dm

t+1

=

β

u

1

(c

∗

t

, l

∗

t

)

u

11

(c

∗

t+1

, l

∗

t+1

)dc

∗

t+1

+ u

12

(c

∗

t+1

, l

∗

t+1

)dl

∗

t+1

, (35)

conditional on information at time t. In (35), the household’s labor margin aﬀects m

t+1

and hence asset prices for two reasons: First, if u

12

= 0, changes in l

t+1

directly aﬀect the

household’s marginal utility of consumption. Second, even if u

12

= 0, the presence of the

labor margin aﬀects how the household responds to shocks and hence aﬀects dc

∗

t+1

.

Intuitively, one can already see the relationship between risk aversion and dm

t+1

in (35):

if dl

∗

t+1

= −λdc

∗

t+1

and dc

∗

t+1

= rda

t+1

/(1 + wλ), as in Section 2.3, then dm

t+1

equals the

coeﬃcient of absolute risk aversion times da

t+1

. In actuality, the relationship is slightly

more complicated than this because θ (and hence w, r, and d) may change as well as a. For

example, diﬀerentiating (12) and evaluating at steady state, we have, to ﬁrst order:

dl

∗

t+1

= −λdc

∗

t+1

−

u

1

u

22

+ wu

12

dw

t+1

, (36)

17

Similarly, combining (2), (3), and (15), diﬀerentiating, and evaluating at steady state, we

show in the Appendix that:

dc

∗

t+1

=

r

1 + wλ

¸

da

t+1

+ E

t+1

∞

¸

k=1

1

(1 + r)

k

(l dw

t+k

+ dd

t+k

+ adr

t+k

)

(37)

+

u

1

u

11

u

22

−u

2

12

dw

t+1

+

−ru

1

u

11

−λu

12

E

t+1

∞

¸

k=1

1

(1 + r)

k

λ

1 + wλ

dw

t+k

−d log R

t+1,t+k

,

where R

t+1,t+k

≡

¸

k

i=2

(1+r

t+i

). Note that for the Arrow-Pratt one-shot gamble considered

in Section 2.3, the aggregate variables w, r, and d were all held constant, so (36)–(37) reduce

to (13) and (21) in that case. The term in square brackets in (37) describes the change in

the present value of household income, and thus the ﬁrst line of (37) describes the income

eﬀect on consumption. The last line of (37) describes the substitution eﬀect: changes in

consumption due to changes in current and future interest rates and wages. (Recall that

−u

1

/

c(u

11

−λu

12

)

**is the intertemporal elasticity of substitution.)
**

Substituting (36)–(37) into (35) yields:

dm

t+1

= β

u

11

−λu

12

u

1

r

1 + wλ

¸

da

t+1

+ E

t+1

∞

¸

k=1

1

(1 + r)

k

(l dw

t+k

+ dd

t+k

+ adr

t+k

)

−βr E

t+1

∞

¸

k=1

1

(1 + r)

k

λ

1 + wλ

dw

t+k

−d log R

t+1,t+k

. (38)

The risk premium (34), evaluated at steady state, is then given by:

−u

11

+ λu

12

u

1

r

1 + wλ

Cov

t

(dp

t+1

, dA

t+1

) + r Cov

t

(dp

t+1

, dΨ

t+1

), (39)

where dA

t+1

denotes the quantity in square brackets in (38)—the change in household

wealth—and dΨ

t+1

denotes the summation on the second line of (38)—the change in cur-

rent and future wages and interest rates. Equations (38)–(39) are essentially Merton’s (1973)

ICAPM, generalized to include labor. In (39), the ﬁrst term is the covariance of the asset

price with household wealth, multiplied by the coeﬃcient of absolute risk aversion, while the

second term captures the asset’s ability to hedge what Merton calls “changes in investment

opportunities.” Intuitively, even for households that are Arrow-Pratt risk neutral, an asset

that pays oﬀ well when future interest rates are high or wages are low—and hence future

consumption is low—is preferable to an asset that has no correlation with future r or w.

Equation (39) shows the importance of risk aversion—and hence the labor margin—for

asset pricing in the model. Risk premia are essentially linear in the coeﬃcient of absolute risk

18

aversion, a relationship which also holds for the more general cases of Epstein-Zin preferences

and habits, considered below.

13

This link between risk aversion and risk premia should not

be too surprising: Arrow-Pratt risk aversion describes the risk premium for the most basic

gambles over household income or wealth. Here we have shown that the same coeﬃcient

also appears for completely general gambles that may be correlated with aggregate variables

such as interest rates, wages, and net transfers.

14

The risk premia on these gambles are

determined by the household’s stochastic discount factor, but the stochastic discount factor

is itself directly linked to risk aversion and the household’s labor margin.

3. Generalized Recursive Preferences

We now turn to the case of generalized recursive preferences, as in Epstein and Zin (1989) and

Weil (1989). The household’s asset accumulation equation (2) and transversality condition

(3) are the same as in Section 2, but now instead of maximizing (1), the household chooses

c

t

and l

t

to maximize the recursive expression:

15

V (a

t

; θ

t

) = max

c

t

,l

t

u(c

t

, l

t

) + β

E

t

V (a

t+1

; θ

t+1

)

1−α

1/(1−α)

, (40)

where α ∈ R, α = 1.

16

Note that (40) is the same as (4), but with the value function

“twisted” and “untwisted” by the coeﬃcient 1 − α. When α = 0, the preferences given by

(40) reduce to the special case of expected utility.

If u ≥ 0 everywhere, then the proof of Theorem 3.1 in Epstein and Zin (1989) shows

that there exists a solution V to (40) with V ≥ 0. If u ≤ 0 everywhere, then it is natural to

let V ≤ 0 and reformulate the recursion as:

V (a

t

; θ

t

) = max

c

t

,l

t

u(c

t

, l

t

) −β

E

t

(−V (a

t+1

; θ

t+1

))

1−α

1/(1−α)

. (41)

13

See the Appendix. For an example of this linearity, see Figure 1 of Rudebusch and Swanson (2009).

14

Boldrin, Christiano, and Fisher (1997) argue that it is u

11

/u

1

rather than V

11

/V

1

that matters for the

equity premium in their Figure 2. As shown above, it is in fact V

11

/V

1

that is crucial. What explains Boldrin

et al.’s Figure 2 is that the covariance of equity prices with the short-term interest rate is not being held

constant in their model—in particular, the variance of the risk-free rate in their model changes tremendously

over the points in their Figure 2.

15

Note that, traditionally, Epstein-Zin preferences over consumption streams have been written as:

¯

V (a

t

; θ

t

) = max

c

t

¸

c

ρ

t

+ β

E

t

¯

V (a

t+1

; θ

t+1

)

¯ α

ρ/¯ α

1/ρ

,

but by setting V =

¯

V

ρ

and α = 1 − ¯ α/ρ, this can be seen to correspond to (40).

16

We exclude the case α = 1 here for simplicity.

19

The proof in Epstein and Zin (1989) also demonstrates the existence of a solution V to (41)

with V ≤ 0 in this case.

To avoid the possibility of complex numbers arising in the maximand of (40) or (41),

we restrict the range of u to be either R

+

or R

−

:

17

Assumption 6. Either u : R

+

×[0,

¯

l) →R

+

, or u : R

+

×[0,

¯

l) →R

−

.

The main advantage of generalized recursive preferences (40) is that they allow for

greater ﬂexibility in modeling risk aversion and the intertemporal elasticity of substitution.

In (40), the intertemporal elasticity of substitution over deterministic consumption paths is

exactly the same as in (4), but the household’s risk aversion to gambles can be ampliﬁed (or

attenuated) by the additional parameter α.

3.1 Coeﬃcients of Absolute and Relative Risk Aversion

We consider the household’s aversion to the same hypothetical gamble as in (6):

Proposition 6. With generalized recursive preferences (40) or (41), the household’s coeﬃ-

cient of absolute risk aversion with respect to the gamble described by (6) is given by:

−E

t

V (a

∗

t+1

; θ

t+1

)

−α

¸

V

11

(a

∗

t+1

; θ

t+1

) − α

V

1

(a

∗

t+1

; θ

t+1

)

2

V (a

∗

t+1

; θ

t+1

)

E

t

V (a

∗

t+1

; θ

t+1

)

−α

V

1

(a

∗

t+1

; θ

t+1

)

. (42)

Evaluated at steady state, (42) simpliﬁes to:

−V

11

(a; θ)

V

1

(a; θ)

+ α

V

1

(a; θ)

V (a; θ)

. (43)

Proof: See Appendix.

The ﬁrst term in (43) is the same as the expected utility case (9), while the second

term in (43) reﬂects the ampliﬁcation or attenuation of risk aversion from the additional

curvature parameter α. When α = 0, (42)–(43) reduce to (8)–(9). When u ≥ 0 and hence

V ≥ 0, higher values of α correspond to greater degrees of risk aversion; when u and V ≤ 0,

the opposite is true: higher values of α correspond to lesser degrees of risk aversion.

17

Alternatively, one can restrict the domain of u to ensure u ≥ 0 or u ≤ 0; e.g., by requiring c ≥ 1 for

u(c, l) = log c + χ(

¯

l −l).

20

The household’s coeﬃcient of relative risk aversion is given by A

t

times (42), which,

evaluated at steady state, simpliﬁes to:

−AV

11

(a; θ)

V

1

(a; θ)

+ α

AV

1

(a; θ)

V (a; θ)

. (44)

We deﬁne the household’s total wealth A

t

based on the present discounted value of its lifetime

consumption or lifetime leisure and consumption, as in Section 2.4, and we refer to (44) as

the consumption-based or leisure-and-consumption-based coeﬀcient of relative risk aversion,

depending on the deﬁnition of A.

18

Expressions (43) and (44) highlight an important feature of risk aversion with general-

ized recursive preferences: it is not invariant with respect to level shifts of the utility kernel,

except for the special case of expected utility (α = 0). That is, the utility kernels u(·, ·) and

u(·, ·)+k, where k is a constant, lead to diﬀerent household attitudes toward risk. The house-

hold’s preferences are invariant, however, with respect to multiplicative transformations of

the utility kernel.

When it comes to computing the risk aversion coeﬃcients (43)–(44), expressions (10)–

(21) for V

1

, V

11

, ∂l

∗

t

/∂a

t

, and ∂c

∗

t

/∂a

t

continue to apply in the current context. Moreover,

V = u(c, l)

**(1 −β) at the steady state. Substituting these into (43)–(44) gives:
**

Proposition 7. The household’s coeﬃcient of absolute risk aversion in Proposition 6, eval-

uated at steady state, is given by:

−V

11

V

1

+ α

V

1

V

=

−u

11

+ λu

12

u

1

r

1 + wλ

+ α

r u

1

u

. (45)

The household’s consumption-based coeﬃcient of relative risk aversion, evaluated at steady

state, is given by:

−AV

11

V

1

+ α

AV

1

V

=

−u

11

+ λu

12

u

1

c

1 + wλ

+ α

c u

1

u

. (46)

The household’s leisure-and-consumption-based coeﬃcient of relative risk aversion, evaluated

at steady state, is given by (c + w(

¯

l −l))/c times (46).

18

Note that, with generalized recursive preferences, the household’s stochastic discount factor is given by:

βu

1

(c

∗

t+1

, l

∗

t+1

)

u

1

(c

∗

t

, l

∗

t

)

⎛

⎝

V (a

∗

t+1

; θ

t+1

)

(E

t

V (a

∗

t+1

; θ

t+1

)

1−α

)

1/(1−α)

⎞

⎠

−α

,

which must be used to compute household wealth. At steady state, however, this simpliﬁes to the usual β.

21

Proposition 7 is important because risk aversion for Epstein-Zin preferences has only

been computed previously in homothetic, isoelastic, consumption-only models, where the

value function can be computed in closed form. Proposition 7 does not require homotheticity,

is valid for general functional forms u, unknown functional forms V , and allows for the

presence of labor.

3.2 Examples

Example 3.1. Consider the additively separable utility kernel:

u(c

t

, l

t

) =

c

1−γ

t

1 −γ

−χ

0

l

1+χ

t

1 + χ

, (47)

with generalized recursive preferences (41) and χ > 0, χ

0

> 0, and γ > 1, which was used by

Rudebusch and Swanson (2009).

19

In this case, where u(·, ·) < 0, risk aversion is decreasing

in α, and α < 0 corresponds to preferences that are more risk averse than expected utility.

In models without labor, period utility u(c

t

, l

t

) = c

1−γ

t

/(1 − γ) implies a coeﬃcient

of relative risk aversion of γ + α(1 − γ), which we will refer to as the traditional mea-

sure.

20

Taking into account both the consumption and labor margins of (47), the household’s

consumption-based coeﬃcient of relative risk aversion (46) is given by:

−AV

11

V

1

+ α

AV

1

V

=

γ

1 +

γ

χ

wl

c

+

α(1 −γ)

1 +

γ−1

1+χ

wl

c

,

≈

γ

1 +

γ

χ

+

α(1 −γ)

1 +

γ−1

1+χ

, (48)

using c ≈ wl. As in Example 2.1, the household’s leisure-and-consumption-based coeﬃcient

of relative risk aversion is not well deﬁned in this example, so we restrict attention to the

consumption-based measure (48).

As χ becomes large, household labor becomes less ﬂexible and the bias from ignoring

the labor margin shrinks to zero ((48) approaches γ +α(1 −γ)). As χ approaches zero, (48)

decreases to α(1−γ)/γ, which is close to zero if we think of γ as being small (close to unity).

19

We restrict attention here to the case γ > 1, consistent with Assumption 6. The case γ ≤ 1 can be

considered if we place restrictions on the domain of c

t

and l

t

such that u(·, ·) < 0; one can always choose

units for c

t

and l

t

such that this doesn’t represent much of a constraint in practice. Of course, one can also

consider alternative utility kernels with γ ≤ 1 for which u(·, ·) > 0.

20

Set χ

0

= 0 and λ = 0 and substitute (47) into (46). This is the case, for example, in Epstein and Zin

(1989) and Boldrin, Christiano, and Fisher (1997), which do not have labor. In models with variable labor,

Rudebusch and Swanson (2009) refer to γ + α(1 −γ) as the quasi coeﬃcient of relative risk aversion.

22

Thus, for given values of γ and α, actual household risk aversion can lie anywhere between

about zero and γ + α(1 −γ), depending on the value of χ.

Example 3.2. Van Binsbergen et al. (2008) and Backus, Routledge, and Zin (2008) consider

generalized recursive preferences with:

u(c

t

, l

t

) =

c

ν

t

(1 −l

t

)

1−ν

1−γ

1 −γ

, (49)

where γ > 0, γ = 1, and ν ∈ (0, 1). Van Binsbergen et al. call γ + α(1 − γ) the coeﬃcient

of relative risk aversion, while Backus et al. use γν + α(1 − γ)ν + (1 − ν), after mapping

each study’s notation over to the present paper’s. The former measure eﬀectively treats

consumption and leisure as a single composite commodity, while the latter measure allows ν—

the importance of the labor margin—to aﬀect the household’s attitudes toward risk.

Substituting (49) into (46), the household’s consumption-based coeﬃcient of relative

risk aversion is:

−AV

11

V

1

+ α

AV

1

V

= γν + α(1 −γ)ν, (50)

while the leisure-and-consumption-based coeﬃcient of relative risk aversion is:

−

˜

AV

11

V

1

+ α

˜

AV

1

V

= γ + α(1 −γ). (51)

The latter agrees with the Van Binsbergen et al. (2008) measure of risk aversion, while the

former is similar to (though not quite the same as) the Backus et al. (2008) measure. In this

paper, we have provided the formal justiﬁcation for both measures, (50) and (51).

21

Example 3.3. Tallarini (2000) considers an alternative Epstein-Zin speciﬁcation:

˜

V

t

(a

t

; θ

t

) ≡ u(c

∗

t

, l

∗

t

) +

β(1 + θ)

(1 −β)(1 −χ)

log E

t

exp

¸

(1 −β)(1 −χ)

1 + θ

˜

V

t+1

(a

∗

t+1

; θ

t+1

)

, (52)

with utility kernel:

u(c

t

, l

t

) = log c

t

+ θ log(

¯

l −l

t

). (53)

We can compute the coeﬃcient of absolute risk aversion for (52) by following along the steps

in the proof of Proposition 6, which yields:

−V

11

(a; θ)

V

1

(a; θ)

−

(1 −β)(1 −χ)

1 + θ

V

1

(a; θ). (54)

21

As ν → 0, w/c → ∞, so consumption becomes trivial to insure with variations in labor supply. This

explains why the consumption-based coeﬃcient of relative risk aversion in (50) vanishes as ν →0.

23

The other steps leading up to Proposition 7 are all the same, so substituting in for V

1

and

V

11

in (54) yields a consumption-based coeﬃcient of relative risk aversion of:

−u

11

+ λu

12

u

1

c

1 + wλ

−

1 −χ

1 + θ

cu

1

=

1

1 + θc

−

1 −χ

1 + θ

. (55)

The leisure-and-consumption-based coeﬃcient of relative risk aversion is (1 + θ) times (55).

Both coeﬃcients of relative risk aversion diﬀer from the value (θ +χ)/(1 +θ) reported

by Tallarini (2000). Tallarini applies the traditional measure of risk aversion, derived in a

consumption-only model under the assumption θ = 0, to the case where θ > 0. However,

simply setting θ > 0 in the traditional measure,

−cu

11

u

1

−

1−χ

1+θ

cu

1

, ignores the fact that

households vary their labor in response to shocks. As a result, Tallarini’s traditional measure

overstates the household’s true aversion to risk by a factor of 1 +θ/χ, if we normalize c to 1

in (55). As θ/χ approaches zero, the labor margin becomes unimportant and this bias

disappears, but the bias can be arbitrarily large as the ratio of θ to χ increases.

4. Internal and External Habits

Many studies in macroeconomics and ﬁnance assume that households derive utility not from

consumption itself, but from consumption relative to some reference level, or habit stock.

Habits, in turn, can have substantial eﬀects on the household’s attitudes toward risk (e.g.,

Campbell and Cochrane, 1999, Boldrin, Christiano, and Fisher, 1997). In this section, we

investigate how habits aﬀect risk aversion in the DSGE framework.

We generalize the household’s utility kernel in this section to u(c

t

− h

t

, l

t

), where h

t

denotes the household’s reference level of consumption, or habits. We focus on an additive

rather than multiplicative speciﬁcation for habits because the implications for risk aversion

are typically more interesting in the additive case. We adjust the feasible choice set for c

t

and Assumptions 1, 2, and 6 accordingly, replacing them with:

Assumption 1

**. The function u is deﬁned over (−
**

¯

h, ∞) × [0,

¯

l),

¯

h ≥ 0. For every t,

c

t

> h

t

−

¯

h. The rest of Assumption 1 applies.

Assumption 2

**. For any feasible state, the household’s optimal choice (c
**

∗

t

, l

∗

t

) lies in the

interior of (−

¯

h, ∞) ×[0,

¯

l).

Assumption 6

. Either u : (−

¯

h, ∞) ×[0,

¯

l) →R

+

, or u : (−

¯

h, ∞) ×[0,

¯

l) →R

−

.

24

If the habit stock h

t

is external to the household (“keeping up with the Joneses” utility),

then the parameters that govern the process for h

t

can be incorporated into the exogenous

state vector θ

t

, and the analysis proceeds much as in the previous sections. However, if the

habit stock h

t

is a function of the household’s own past levels of consumption, then the state

variables of the household’s optimization problem must be augmented to include the state

variables that govern h

t

. We consider each of these cases in turn.

4.1 External Habits

When the reference consumption level h

t

in the utility kernel u(c

t

− h

t

, l

t

) is external to

the household, then the parameters that govern h

t

can be incorporated into the exogenous

state vector θ

t

and the analysis of the previous sections carries over essentially as before. In

particular, the coeﬃcient of absolute risk aversion continues to be given by (9) in the case

of expected utility and (43) in the case of generalized recursive preferences. The household’s

intratemporal optimality condition (12) still implies:

∂l

∗

t

∂a

t

= −λ

t

∂c

∗

t

∂a

t

, (56)

where λ

t

is given by (14), and the household’s Euler equation (15) still implies:

∂c

∗

t

∂a

t

= E

t

∂c

∗

t+1

∂a

t

= E

t

∂c

∗

t+k

∂a

t

, k = 1, 2, . . . (57)

∂l

∗

t

∂a

t

= E

t

∂l

∗

t+1

∂a

t

= E

t

∂l

∗

t+k

∂a

t

, k = 1, 2, . . . (58)

evaluated at steady state. Together with the budget constraint (2)–(3), (56)–(58) imply:

∂c

∗

t

∂a

t

=

r

1 + wλ

. (59)

The only real diﬀerences that arise relative to the case without habits is, ﬁrst, that the

steady-state point at which the derivatives of u(·, ·) are evaluated is (c − h, l) rather than

(c, l), and second, that relative risk aversion confronts the household with a hypothetical

gamble over c rather than c − h, which has a tendency to make the household more risk

averse for a given functional form u(·, ·), because the stakes are eﬀectively larger.

Example 4.1. Consider the case of expected utility with additively separable utility kernel:

u(c

t

−h

t

, l

t

) =

(c

t

−h

t

)

1−γ

1 −γ

− χ

0

l

1+χ

t

1 + χ

, (60)

25

where γ, χ, χ

0

> 0. The traditional measure of risk aversion for this example is −cu

11

/u

1

=

γc/(c −h), which exceeds γ by a factor that depends on the importance of habits relative to

consumption. The consumption-based coeﬃcient of relative risk aversion is:

−AV

11

V

1

=

−cu

11

u

1

1

1 + w

wu

11

u

22

,

=

γc

(c −h)

1

1 +

γc

χ(c−h)

wl

c

. (61)

When there is no labor margin in the model (λ = 0), the consumption-based measure agrees

with the traditional measure. When there is a labor margin, the household’s consumption-

based coeﬃcient of relative risk aversion (61) is less than the traditional measure by the

factor 1 +

γc

χ(c−h)

, using wl ≈ c. Ignoring the labor margin in (61) thus leads to an even

greater bias in the model with habits (h > 0) than in the model without habits (h = 0).

If γ = 2, χ = 1, and h = .8c, then the household’s true risk aversion is smaller than the

traditional measure by a factor of more than ten.

When the household has generalized recursive preferences rather than expected utility

preferences, the consumption-based coeﬃcient of relative risk aversion for (60) is:

γc

(c −h)

1

1 +

γc

χ(c−h)

wl

c

+

α(1 −γ)c

(c −h)

1

1 +

c

(c−h)

γ−1

1+χ

wl

c

. (62)

Again, the bias from ignoring the labor margin in (62) is even greater in the model with

habits (h > 0) than without habits (h = 0).

4.2 Internal Habits

When habits are internal to the household, we must specify how the household’s actions

aﬀect its future habits. In order to minimize notation and emphasize intuition, in the present

section we focus on the case where habits are proportional to last period’s consumption:

h

t

= bc

t−1

, (63)

b ∈ (0, 1), and we assume the household has expected utility preferences. In the Appendix,

we derive the corresopnding closed-form expressions for the more complicated case where the

habit stock evolves according to the longer-memory process:

h

t

= ρh

t−1

+ bc

t−1

, (64)

26

with ρ ∈ (−1, 1).

With internal habits, the value of h

t+1

depends on the household’s choices in period t,

so we write out the dependence of the household’s value function on h

t

explicitly:

V (a

t

, h

t

; θ

t

) = u(c

∗

t

−h

t

, l

∗

t

) + β

E

t

V (a

∗

t+1

, h

∗

t+1

; θ

t+1

)

1−α

1/(1−α)

, (65)

where c

∗

t

≡ c

∗

(a

t

, h

t

; θ

t

) and l

∗

t

≡ l

∗

(a

t

, h

t

; θ

t

) denote the household’s optimal choices for

consumption and labor in period t as functions of the household’s state vector, and a

∗

t+1

and

h

∗

t+1

denote the optimal stocks of assets and habits in period t + 1 that are implied by c

∗

t

and l

∗

t

; that is, a

∗

t+1

≡ (1 + r

t

)a

t

+ w

t

l

∗

t

+ d

t

−c

∗

t

and h

∗

t+1

≡ bc

∗

t

.

The household’s coeﬃcient of absolute risk aversion can be derived in the same manner

as in Propositions 1 and 6:

Proposition 8. With generalized recursive preferences (40) or (41), utility kernel u(c

t

−

h

t

, l

t

), and internal habits h

t

given by (63), the household’s coeﬃcient of absolute risk aver-

sion with respect to the gamble (6) is given by:

−E

t

V (a

∗

t+1

, h

∗

t+1

; θ

t+1

)

−α

¸

V

11

(a

∗

t+1

, h

∗

t+1

; θ

t+1

) − α

V

1

(a

∗

t+1

, h

∗

t+1

; θ

t+1

)

2

V (a

∗

t+1

, h

∗

t+1

; θ

t+1

)

E

t

V (a

∗

t+1

, h

∗

t+1

; θ

t+1

)

−α

V

1

(a

∗

t+1

, h

∗

t+1

; θ

t+1

)

. (66)

Evaluated at steady state, (66) simpliﬁes to:

−V

11

(a, h; θ)

V

1

(a, h; θ)

+ α

V

1

(a, h; θ)

V (a, h; θ)

. (67)

Proof: Essentially identical to the proof of Proposition 6.

Computing closed-form expressions for V

1

and V

11

in (67) is substantially more compli-

cated for the case of internal habits, however, because of the dynamic relationship between

the household’s current consumption and its future habits. In order to minimize notation

and simplify this derivation as much as possible, we restrict attention in the main text to the

case of expected utility preferences (α = 0). In the Appendix, we derive the corresponding

closed-form expressions for the more complicated case of generalized recursive preferences.

The household’s ﬁrst-order conditions for (65) with respect to consumption and labor

(and imposing α = 0) are given by:

u

1

(c

∗

t

−h

t

, l

∗

t

) = βE

t

V

1

(a

∗

t+1

, h

∗

t+1

; θ

t+1

) − βbE

t

V

2

(a

∗

t+1

, h

∗

t+1

; θ

t+1

), (68)

u

2

(c

∗

t

−h

t

, l

∗

t

) = −βw

t

E

t

V

1

(a

∗

t+1

, h

∗

t+1

; θ

t+1

). (69)

27

Equation (69) is essentially the same as in the case without habits. The ﬁrst-order condi-

tion (68), however, includes the future eﬀect of consumption on habits in the second term

on the right-hand side.

Diﬀerentiating (65) with respect to its ﬁrst two arguments and applying the envelope

theorem yields:

V

1

(a

t

, h

t

; θ

t

) = β(1 + r

t

) E

t

V

1

(a

∗

t+1

, h

∗

t+1

; θ

t+1

), (70)

V

2

(a

t

, h

t

; θ

t

) = −u

1

(c

∗

t

−h

t

, l

∗

t

). (71)

Equations (69) and (70) can be used to solve for V

1

in terms of current-period utility:

V

1

(a

t

, h

t

; θ

t

) = −

(1 + r

t

)

w

t

u

2

(c

∗

t

−h

t

, l

∗

t

), (72)

which states that the marginal value of wealth equals the marginal utility of working fewer

hours.

22

This solves for V

1

.

To solve for V

11

, diﬀerentiate (72) with respect to a

t

to yield:

V

11

(a

t

, h

t

; θ

t

) = −

(1 + r

t

)

w

t

u

12

∂c

∗

t

∂a

t

+ u

22

∂l

∗

t

∂a

t

, (73)

where we drop the arguments of the u

ij

to reduce notation. It now remains to solve for

∂c

∗

t

/∂a

t

and ∂l

∗

t

/∂a

t

, which we do in the same manner as before, except that the dynamics

of internal habits require us to solve for ∂c

∗

τ

/∂a

t

and ∂l

∗

τ

/∂a

t

for all dates τ ≥ t at the

same time. To better keep track of these dynamics, we henceforth let a time subscript τ ≥ t

denote a generic future date and reserve the subscript t to denote the date of the current

period—the period in which the household faces the hypothetical one-shot gamble.

We solve for ∂l

∗

τ

/∂a

t

in terms of ∂c

∗

τ

/∂a

t

in much the same way as without habits.

The household’s intratemporal optimality condition ((68) combined with (69)) implies:

−u

2

(c

∗

τ

−h

∗

τ

, l

∗

τ

) = w

τ

u

1

(c

∗

τ

−h

∗

τ

, l

∗

τ

) + bβE

τ

V

2

(a

∗

τ+1

, h

∗

τ+1

; θ

τ+1

)

, (74)

= w

τ

(1 −βbF) u

1

(c

∗

τ

−h

∗

τ

, l

∗

τ

), (75)

where F denotes the forward operator, that is Fx

τ

≡ E

τ

x

τ+1

for any expression x dated τ.

Diﬀerentiating (75) with respect to a

t

yields:

−u

12

∂c

∗

τ

∂a

t

−

∂h

∗

τ

∂a

t

−u

22

∂l

∗

τ

∂a

t

= w

τ

(1 −βbF)

u

11

∂c

∗

τ

∂a

t

−

∂h

∗

τ

∂a

t

+ u

12

∂l

∗

τ

∂a

t

, (76)

22

Using the marginal utility of labor is simpler than the marginal utility of consumption in (72) because it

avoids having to keep track of future habits and the value function next period. However, in steady state it

is also true that V

1

= u

1

(1 −βb)/β, which we will use to express risk aversion in terms of u

1

and u

11

below.

28

where Fu

11

∂c

∗

τ

/∂a

t

denotes E

τ

u

11

(c

∗

τ+1

−h

∗

τ+1

, l

∗

τ+1

) ∂c

∗

τ+1

/∂a

t

, and ∂h

∗

τ

/∂a

t

= 0 for τ = t

since h

t

is given. Evaluating (76) at steady state and solving for ∂l

∗

τ

/∂a

t

yields:

∂l

∗

τ

∂a

t

= −

u

12

+ wu

11

−βbwu

11

F

u

22

+ wu

12

1 −

βbwu

12

u

22

+ wu

12

F

−1

(1 −bL)

∂c

∗

τ

∂a

t

. (77)

where the u

ij

are evaluated at steady state, L denotes the lag operator—that is, Lx

τ

≡ x

τ−1

for any expression x dated τ—and we assume |βbwu

12

/(u

22

+ wu

12

)| < 1 in order to ensure

convergence. Note that when b = 0, (77) reduces to −

wu

11

+u

12

u

22

+wu

12

∂c

∗

τ

∂a

t

, as in Section 2. This

solves for ∂l

∗

τ

/∂a

t

in terms of (current and future) ∂c

∗

τ

/∂a

t

.

As before, we solve for ∂c

∗

τ

/∂a

t

using the household’s Euler equation and budget con-

straint. Diﬀerentiating the household’s Euler equation:

1

w

τ

u

2

(c

∗

τ

−h

∗

τ

, l

∗

τ

) = βE

τ

1 + r

τ+1

w

τ+1

u

2

(c

∗

τ+1

−h

∗

τ+1

, l

∗

τ+1

), (78)

with respect to a

t

and evaluating at steady state yields:

u

12

(1 + b) −F −bL

∂c

∗

τ

∂a

t

= −u

22

(1 −F)

∂l

∗

τ

∂a

t

. (79)

Substituting (77) into (79) yields the following diﬀerence equation for c

τ

:

u

12

u

22

+ wu

12

−βbwu

12

F

(1 + b) −F −bL

−

u

22

(1 −F)

u

12

+ wu

11

−βbwu

11

F

(1 −bL)

∂c

∗

τ

∂a

t

= 0. (80)

Since FL = 1,

23

equation (80) simpliﬁes to:

(1 −βbF)(1 −F)(1 −bL)

∂c

∗

τ

∂a

t

= 0, (81)

which, from (79), also implies:

(1 −βbF)(1 −F)

∂l

∗

τ

∂a

t

= 0. (82)

Equations (81) and (82) hold for all τ ≥ t, hence we can invert the (1 − βbF) operator

forward to get:

(1 −F)(1 −bL)

∂c

∗

τ

∂a

t

= 0, (83)

(1 −F)

∂l

∗

τ

∂a

t

= 0. (84)

23

To be precise, FLx

τ

= E

τ−1

x

τ

, but since the household evaluates these expressions from the perspective

of the initial period t, E

t

FLx

τ

= E

t

x

τ

. Formally, take the expectation of (80) at time t and then apply

E

t

FL = E

t

to get (81).

29

In other words, whatever the initial responses ∂c

∗

t

/∂a

t

and ∂l

∗

t

/∂a

t

are, we must have:

E

t

∂c

∗

t+1

∂a

t

= (1 + b)

∂c

∗

t

∂a

t

,

E

t

∂c

∗

t+k

∂a

t

= (1 + b +· · · + b

k

)

∂c

∗

t

∂a

t

, (85)

and E

t

∂l

∗

t+k

∂a

t

=

∂l

∗

t

∂a

t

, k = 1, 2, . . . (86)

evaluated at steady state. Because of habits, consumption responds only gradually to a

surprise change in wealth, asymptoting over time to its new steady-state level, but labor

moves immediately to its new steady-state level in response to surprises in wealth.

From (85), we can now solve (79) to get:

∂l

∗

t

∂a

t

= −λ

∂c

∗

t

∂a

t

, (87)

where

λ ≡

w(1 −βb)u

11

+ u

12

u

22

+ w(1 −βb)u

12

=

u

1

u

12

−u

2

u

11

u

1

u

22

−u

2

u

12

, (88)

and where the latter equality follows because w = −(1 −βb)

−1

u

2

/u

1

in steady state. Thus,

that (87)–(88) are essentially identical to (13)–(14).

24

Again, λ must be positive if leisure

and consumption are normal goods.

It now remains to solve for ∂c

∗

t

/∂a

t

. From the household’s budget constraint and

condition (86), we have:

E

t

∞

¸

τ=t

(1 + r)

−(τ−t)

∂c

∗

τ

∂a

t

= (1 + r) + w

1 + r

r

∂l

∗

t

∂a

t

. (89)

Substituting (85)–(87) into (89) and solving for ∂c

∗

t

/∂a

t

yields:

∂c

∗

t

∂a

t

=

(1 −βb)r

1 + (1 −βb)wλ

. (90)

Without habits or labor, an increase in assets would cause consumption to rise by the amount

of the income ﬂow from the change in assets, r. The presence of habits attenuates this change

by the amount βb in the numerator of (90), and the consumption response is further attenu-

ated by the household’s change in hours worked, which is accounted for by the denominator.

24

However, unlike the model without habits, (87)–(88) only hold here in steady state.

30

Substituting (72), (73), (87), (88), and (90) into (67), we have established:

25

Proposition 9. The household’s coeﬃcient of absolute risk aversion in Proposition 8, eval-

uated at steady state, is given by:

−V

11

V

1

=

−u

11

+ λu

12

u

1

(1 −βb)r

1 + (1 −βb)wλ

. (91)

The household’s consumption-based coeﬃcient of relative risk aversion, evaluated at steady

state, is given by:

−AV

11

V

1

=

−u

11

+ λu

12

u

1

(1 −βb)c

1 + (1 −βb)wλ

. (92)

The household’s leisure-and-consumption-based coeﬃcient of relative risk aversion, evaluated

at steady state, is given by (c + w(

¯

l −l))/c times (92).

Equations (91)–(92) have essentially the same form as the corresponding expressions

in the model without habits.

Example 4.2. Consider the utility kernel of example 4.1:

u(c

t

−h

t

, l

t

) =

(c

t

−h

t

)

1−γ

1 −γ

− χ

0

l

1+χ

t

1 + χ

, (93)

where γ, χ, χ

0

> 0, but now with habits h

t

= bc

t−1

internal to the household rather than

external. In thise case, the household’s consumption-based coeﬃcient of relative risk aversion

is given by:

−AV

11

V

1

=

−cu

11

u

1

1 −βb

1 + (1 −βb)wλ

,

= γ

1 −βb

1 −b

1

1 +

γ

χ

1−βb

1−b

wl

c

,

≈

γ

1 +

γ

χ

, (94)

where the last line uses β ≈ 1 and wl ≈ c.

The most striking feature of equation (94) is that it is independent of b, the importance

of habits. This is in sharp contrast to the case of external habits, where risk aversion is

strongly increasing in b (cf. equation (61)).

25

In order to express (91) in terms of u

1

and u

11

instead of u

2

and u

22

, we use V

1

= (1 − βb)u

1

/β and

diﬀerentiate the ﬁrst-order condition:

u

1

(c

∗

t

−h

t

, l

∗

t

) =

1

1 + r

t

V

1

(a

t

, h

t

; θ

t

) + βbE

t

u

1

(c

∗

t+1

−h

∗

t+1

, l

∗

t+1

),

with respect to a

t

to solve for V

11

using (85)–(88) and (90).

31

5. Discussion and Conclusions

The ability to vary labor supply has dramatic eﬀects on household risk aversion and asset

prices in dynamic equilibrium models. The traditional measure of risk aversion, −cu

11

/u

1

,

ignores the household’s ability to partially oﬀset shocks to income with changes in hours

worked. For reasonable parameterizations, the traditional measure can easily overstate risk

aversion by a factor of three or more. Indeed, households can even be risk neutral when the

traditional measure of risk aversion is far from zero. Many studies in the macroeconomics,

macro-ﬁnance, and international literatures thus may be overstating the actual degree of risk

aversion in their models by a substantial degree.

Risk aversion matters for asset pricing. Risk premia on assets computed using the

stochastic discount factor are essentially linear in the degree of risk aversion. As a result,

asset prices in DSGE models can be very diﬀerent and can behave very diﬀerently depending

on how the household’s labor margin is speciﬁed. Understanding how labor supply aﬀects

asset prices is thus important for bringing DSGE-type models closer to ﬁnancial market data.

If risk aversion is measured incorrectly because the labor margin is ignored, then risk premia

in the model are also more likely to be surprising or puzzling. An extreme example of this

is when household utility has a zero discriminant—implying risk neutrality—even when the

traditional measure of risk aversion is large.

Risk neutrality itself can be a desirable feature for some applications, such as labor mar-

ket search or ﬁnancial frictions. In these applications, risk neutrality allows for much simpler

or even closed-form solutions to key aspects of the model. The present paper suggests new

ways of modeling risk neutrality in a DSGE framework. The traditional approach—linearity

of utility in consumption—has undesirable implications for interest rates and consumption

growth, but the present paper shows that any utility kernel with a zero discriminant can be

used instead.

A related observation is that risk aversion and the intertemporal elasticity of substitu-

tion are nonreciprocal, even for expected utility preferences. There is a wedge between the

two concepts that depends on the household’s labor margin.

The simple, closed-form expressions for risk aversion that this paper derives, and the

methods of the paper more generally, should be useful to researchers interested in pricing any

asset—stocks, bonds, or futures, in foreign or domestic currency—within the framework of

32

dynamic equilibrium models. Since these models are a mainstay of research in academia, at

central banks, and international ﬁnancial institutions, the applicability of the results should

be widespread.

33

Appendix: Mathematical Derivations

Proof of Proposition 1

For an inﬁnitesimal fee dμ in (7), the change in household welfare (5) is given, to ﬁrst order, by:

−V

1

(a

t

; θ

t

)

1 +r

t

dμ = −βE

t

V

1

(a

∗

t+1

; θ

t+1

) dμ, (A1)

where the right-hand side of (A1) follows from the envelope theorem.

Turning now to the gamble in (6), note ﬁrst that the household’s optimal choices for con-

sumption and labor in period t, c

∗

t

and l

∗

t

, will generally depend on the size of the gamble σ—for

example, the household may undertake precautionary saving when faced with this gamble. Thus, in

this section we write c

∗

t

≡ c

∗

(a

t

; θ

t

; σ) and l

∗

t

≡ l

∗

(a

t

; θ

t

; σ) to emphasize this dependence on σ.

Because c

∗

t

and l

∗

t

depend on σ, the household’s value-to-go at time t also depends on σ. We

write this dependence out explicitly as well, so that:

¯

V (a

t

; θ

t

; σ) = u(c

∗

t

, l

∗

t

) +βE

t

V (a

∗

t+1

; θ

t+1

), (A2)

where a

∗

t+1

≡ (1 + r

t

)a

t

+ w

t

l

∗

t

+ d

t

− c

∗

t

. Because (6) describes a one-shot gamble in period t, it

aﬀects assets a

∗

t+1

in period t+1 but otherwise does not aﬀect the household’s optimization problem

from period t +1 onward; as a result, the household’s value-to-go at time t +1 is just V (a

∗

t+1

; θ

t+1

),

which does not depend on σ except through a

∗

t+1

. The tilde over the V on the left-hand side of (A2)

emphasizes that the form of the value function itself is diﬀerent in period t due to the presence of

the one-shot gamble in that period.

Diﬀerentiating (A2) with respect to σ, the ﬁrst-order eﬀect of the gamble on household welfare

is:

¸

u

1

∂c

∗

∂σ

+u

2

∂l

∗

∂σ

+βE

t

V

1

· (w

t

∂l

∗

∂σ

−

∂c

∗

∂σ

+ε

t+1

)

dσ, (A3)

where the arguments of u

1

, u

2

, and V

1

are suppressed to simplify notation. Optimality of c

∗

t

and

l

∗

t

implies that the terms involving ∂c

∗

/∂σ and ∂l

∗

/∂σ in (A3) cancel, as in the usual envelope

theorem (these derivatives vanish at σ = 0 anyway, for the reasons discussed below). Moreover,

E

t

V

1

(a

∗

t+1

; θ

t+1

)ε

t+1

= 0 because ε

t+1

is independent of θ

t+1

and a

∗

t+1

, evaluating the latter at

σ = 0. Thus, the ﬁrst-order cost of the gamble is zero, as in Arrow (1964) and Pratt (1965).

To second order, the eﬀect of the gamble on household welfare is:

¸

u

11

∂c

∗

∂σ

2

+ 2u

12

∂c

∗

∂σ

∂l

∗

∂σ

+u

22

∂l

∗

∂σ

2

+u

1

∂

2

c

∗

∂σ

2

+ u

2

∂

2

l

∗

∂σ

2

+βE

t

V

11

·

w

t

∂l

∗

∂σ

−

∂c

∗

∂σ

+ε

t+1

2

+ βE

t

V

1

·

w

t

∂

2

l

∗

∂σ

2

−

∂

2

c

∗

∂σ

2

¸

dσ

2

2

. (A4)

The terms involving ∂

2

c

∗

/∂σ

2

and ∂

2

l

∗

/∂σ

2

cancel due to the optimality of c

∗

t

and l

∗

t

. The derivatives

∂c

∗

/∂σ and ∂l

∗

/∂σ vanish at σ = 0 (there are two ways to see this: ﬁrst, the linearized version

of the model is certainty equivalent; alternatively, the gamble in (6) is isomorphic for positive and

negative σ, hence c

∗

and l

∗

must be symmetric about σ = 0, implying the derivatives vanish). Thus,

for inﬁnitesimal gambles, (A4) simpliﬁes to:

βE

t

V

11

(a

∗

t+1

; θ

t+1

) ε

2

t+1

dσ

2

2

. (A5)

Finally, ε

t+1

is independent of θ

t+1

and a

∗

t+1

, evaluating the latter at σ = 0. Since ε

t+1

has unit

variance, (A5) reduces to:

βE

t

V

11

(a

∗

t+1

; θ

t+1

)

dσ

2

2

. (A6)

34

Equating (A1) to (A6), the Arrow-Pratt coeﬃcient of absolute risk aversion, 2dμ/dσ

2

, is:

−E

t

V

11

(a

∗

t+1

; θ

t+1

)

E

t

V

1

(a

∗

t+1

; θ

t+1

)

. (A7)

Recall that (A7) is already evaluated at σ = 0, so to evaluate it at the nonstochastic steady

state, set a

t+1

= a and θ

t+1

= θ to get:

−V

11

(a; θ)

V

1

(a; θ)

. (A8)

Derivation of Risk Aversion, the Stochastic Discount Factor, and Risk Premia

Diﬀerentiating the household’s Euler equation (15) and evaluating at steady state yields:

u

11

(dc

∗

t

−E

t

dc

∗

t+1

) +u

12

(dl

∗

t

−E

t

dl

∗

t+1

) = βE

t

u

1

dr

t+1

, (A9)

which, applying (36), becomes:

(u

11

−λu

12

)(dc

∗

t

−E

t

dc

∗

t+1

) −

u

1

u

12

u

22

+wu

12

(dw

t

−E

t

dw

t+1

) = βE

t

u

1

dr

t+1

. (A10)

Note that (A10) implies, for each k = 1, 2, . . .,

E

t

dc

∗

t+k

= dc

∗

t

−

u

1

u

12

u

11

u

22

−u

2

12

(dw

t

−E

t

dw

t+k

) −

βu

1

u

11

−λu

12

E

t

k

¸

i=1

dr

t+i

. (A11)

Combining (2)–(3), diﬀerentiating, and evaluating at steady state yields:

E

t

∞

¸

k=0

1

(1 +r)

k

(dc

∗

t+k

−wdl

∗

t+k

−ldw

t+k

−dd

t+k

−adr

t+k

) = (1 + r) da

t

. (A12)

Substituting (36) and (A11) into (A12), and solving for dc

∗

t

, yields:

dc

∗

t

=

r

1 + r

1

1 + wλ

¸

(1 +r)da

t

+E

t

∞

¸

k=0

1

(1 +r)

k

(l dw

t+k

+ dd

t+k

+adr

t+k

)

¸

+

u

1

u

12

u

11

u

22

−u

2

12

dw

t

+

r

1 +r

−u

1

u

11

−λu

12

E

t

∞

¸

k=0

1

(1 +r)

k

¸

λ

1 +wλ

dw

t+k

−d log R

t,t+k

, (A13)

where R

t,t+k

≡

¸

k

i=1

(1 +r

t+i

). Combining (35), (36), and (A13) gives:

dm

t+1

= βr

u

11

−λu

12

u

1

1

1 +wλ

¸

da

t+1

+E

t+1

∞

¸

k=1

1

(1 +r)

k

(l dw

t+k

+dd

t+k

+adr

t+k

)

¸

−βr E

t+1

∞

¸

k=1

1

(1 +r)

k

¸

λ

1 + wλ

dw

t+k

−d log R

t+1,t+k

, (A14)

as in the main text. Equation (A14) also holds for the case of external habits (cf. Section 4.1).

For generalized recursive preferences, equations (A9)–(A13) still hold, but dm

t+1

has extra

terms related to dV

t+1

. In this case, we get the more general expression:

dm

t+1

= βr

u

11

−λu

12

u

1

1

1 +wλ

−

αu

1

u

¸

da

t+1

+E

t+1

∞

¸

k=1

1

(1 +r)

k

(l dw

t+k

+dd

t+k

+adr

t+k

)

¸

−βr E

t+1

∞

¸

k=1

1

(1 +r)

k

¸

λ

1 +wλ

dw

t+k

−d log R

t+1,t+k

. (A15)

35

Proof of Proposition 6

For generalized recursive preferences, the hypothetical one-shot gamble and one-time fee faced by

the household are the same as for the case of expected utility. However, the household’s optimality

conditions for c

∗

t

and l

∗

t

(and, implicitly, a

∗

t+1

) are slightly more complicated:

u

1

(c

∗

t

, l

∗

t

) = β(E

t

V (a

∗

t+1

; θ

t+1

)

1−α

)

α/(1−α)

E

t

V (a

∗

t+1

; θ

t+1

)

−α

V

1

(a

∗

t+1

; θ

t+1

), (A16)

u

2

(c

∗

t

, l

∗

t

) = −βw

t

(E

t

V (a

∗

t+1

; θ

t+1

)

1−α

)

α/(1−α)

E

t

V (a

∗

t+1

; θ

t+1

)

−α

V

1

(a

∗

t+1

; θ

t+1

). (A17)

Note that (A16) and (A17) are related by the usual u

2

(c

∗

t

, l

∗

t

) = −w

t

u

1

(c

∗

t

; l

∗

t

), and when α = 0,

(A16) and (A17) reduce to the standard optimality conditions for expected utility.

For an inﬁnitesimal fee dμ in (7), the change in welfare for the household with generalized

recursive preferences is:

−V

1

(a

t

; θ

t

)

dμ

1 +r

t

= −β(E

t

V (a

∗

t+1

; θ

t+1

)

1−α

)

α/(1−α)

E

t

V (a

∗

t+1

; θ

t+1

)

−α

V

1

(a

∗

t+1

; θ

t+1

) dμ, (A18)

where the right-hand side of (A18) follows from the envelope theorem.

Turning now to the gamble in (6), the ﬁrst-order eﬀect of the gamble on household welfare is:

¸

u

1

∂c

∗

∂σ

+u

2

∂l

∗

∂σ

+β(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

1

· (w

t

∂l

∗

∂σ

−

∂c

∗

∂σ

+ε

t+1

)

dσ, (A19)

where we have dropped the arguments of u

1

, u

2

, V , and V

1

to simplify notation. As before, optimality

of c

∗

t

and l

∗

t

implies that the terms involving ∂c

∗

/∂σ and ∂l

∗

/∂σ cancel, and E

t

V

−α

V

1

ε

t+1

= 0

because ε

t+1

is independent of θ

t+1

and a

∗

t+1

, evaluating the latter at σ = 0. Thus, the ﬁrst-order

cost of the gamble is zero.

To second order, the eﬀect of the gamble on household welfare is:

⎧

⎨

⎩

u

11

∂c

∗

∂σ

2

+ 2u

12

∂c

∗

∂σ

∂l

∗

∂σ

+u

22

∂l

∗

∂σ

2

+ u

1

∂

2

c

∗

∂σ

2

+u

2

∂

2

l

∗

∂σ

2

+αβ(E

t

V

1−α

)

(2α−1)/(1−α)

¸

E

t

V

−α

V

1

·

w

t

∂l

∗

∂σ

−

∂c

∗

∂σ

+ε

t+1

2

−αβ(E

t

V

1−α

)

α/(1−α)

E

t

V

−α−1

¸

V

1

·

w

t

∂l

∗

∂σ

−

∂c

∗

∂σ

+ε

t+1

2

+β(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

11

·

w

t

∂l

∗

∂σ

−

∂c

∗

∂σ

+ε

t+1

2

+β(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

1

·

w

t

∂

2

l

∗

∂σ

2

−

∂

2

c

∗

∂σ

2

⎫

⎬

⎭

dσ

2

2

. (A20)

The derivatives ∂c

∗

/∂σ and ∂l

∗

/∂σ vanish at σ = 0, the terms involving ∂

2

c

∗

/∂σ

2

and ∂

2

l

∗

/∂σ

2

cancel due to the optimality of c

∗

t

and l

∗

t

, and ε

t+1

is independent of θ

t+1

and a

∗

t+1

(evaluating the

latter at σ = 0). Thus, (A20) simpliﬁes to:

β(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

11

−αE

t

V

−α−1

V

2

1

dσ

2

2

. (A21)

Equating (A18) to (A21), the Arrow-Pratt coeﬃcient of absolute risk aversion is:

−E

t

V

−α

V

11

+αE

t

V

−α−1

V

2

1

E

t

V

−α

V

1

. (A22)

Since (A22) is already evaluated at σ = 0, to evaluate it at the nonstochastic steady state,

set a

t+1

= a, θ

t+1

= θ to get:

−V

11

(a; θ)

V

1

(a; θ)

+α

V

1

(a; θ)

V (a; θ)

. (A23)

36

Derivation of Risk Aversion with Long-Memory Internal Habits and EZ Preferences

We consider here the case of generalized recursive preferences:

V (a

t

, h

t

; θ

t

) = u(c

∗

t

−h

t

, l

∗

t

) +β

E

t

V (a

∗

t+1

, h

∗

t+1

; θ

t+1

)

1−α

1/(1−α)

, (A24)

and a longer-memory speciﬁcation for habits:

h

t

= ρh

t−1

+bc

t−1

, (A25)

with |ρ| < 1, and we assume ρ +b < 1 in order to ensure h < c.

We wish to compute V

1

and V

11

. The household’s ﬁrst-order conditions for (A24) with respect

to consumption and labor are given by:

u

1

= β(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

[V

1

−bV

2

], (A26)

u

2

= −βw

t

(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

1

, (A27)

where we drop the arguments of u and V to reduce notation. Equations (A26) and (A27) are the

same as in the main text except that the discounting of future periods involves the value function

V when α = 0.

Diﬀerentiating (A24) with respect to its ﬁrst two arguments and applying the envelope theorem

yields:

V

1

= β(1 +r

t

) (E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

1

, (A28)

V

2

= −u

1

+ ρβ(E

t

V

1−α

)

α/(1−α)

E

t

V

−α

V

2

. (A29)

As in the main text, (A27) and (A28) can be used to solve for V

1

in terms of current-period

utility:

V

1

(a

t

, h

t

; θ

t

) = −

(1 +r

t

)

w

t

u

2

(c

∗

t

−h

t

, l

∗

t

). (A30)

To solve for V

11

, diﬀerentiate (A30) with respect to a

t

to yield:

V

11

(a

t

, h

t

; θ

t

) = −

(1 +r

t

)

w

t

u

12

∂c

∗

t

∂a

t

+u

22

∂l

∗

t

∂a

t

, (A31)

It remains to solve for ∂c

∗

t

/∂a

t

and ∂l

∗

t

/∂a

t

. As in the main text, we solve for ∂c

∗

τ

/∂a

t

and ∂l

∗

τ

/∂a

t

for all dates τ ≥ t at the same time. We henceforth let a time subscript τ ≥ t denote a generic

future date and reserve the subscript t to denote the date of the current period—the period in which

the household faces the hypothetical one-shot gamble.

We solve for ∂l

∗

τ

/∂a

t

in terms of ∂c

∗

τ

/∂a

t

in the same manner as in the main text, except

that the expressions are more complicated due to the persistence of habits and the household’s more

complicated discounting of future periods. Note ﬁrst that (A29) can be used to solve for V

2

in terms

of current and future marginal utility:

V

2

= −(1 −ρβF)

−1

u

1

, (A32)

where now F denotes the “generalized recursive” forward operator; that is,

Fx

τ

≡ (E

τ

V

1−α

)

α/(1−α)

E

τ

V

−α

x

τ+1

. (A33)

The household’s intratemporal optimality condition ((A28) combined with (A29)) implies:

−u

2

(c

∗

τ

−h

∗

τ

, l

∗

τ

) = w

τ

[u

1

(c

∗

τ

−h

∗

τ

, l

∗

τ

) + bβE

τ

V

2

(a

∗

τ+1

, h

∗

τ+1

; θ

τ+1

)]. (A34)

= w

τ

(1 −βbF(1 −βρF)

−1

) u

1

(c

∗

τ

−h

∗

τ

, l

∗

τ

), (A35)

37

Diﬀerentiating (A35) with respect to a

t

and evaluating at steady state yields:

−u

12

∂c

∗

τ

∂a

t

−

∂h

∗

τ

∂a

t

−u

22

∂l

∗

τ

∂a

t

= w(1 −βbF(1 −βρF)

−1

)

¸

u

11

∂c

∗

τ

∂a

t

−

∂h

∗

τ

∂a

t

+u

12

∂l

∗

τ

∂a

t

, (A36)

where we have used the fact that:

∂

∂a

t

Fx

τ

= F

∂x

τ

∂a

t

, (A37)

when the derivative is evaluated at steady state. Solving (A36) for ∂l

∗

τ

/∂a

t

yields:

∂l

∗

τ

∂a

t

= −

u

12

+wu

11

−β(ρu

12

+ (ρ +b)wu

11

)F

u

22

+ wu

12

×

¸

1 −

β(ρu

22

+ (ρ +b)wu

12

)

u

22

+wu

12

F

−1

(1 −bL(1 −ρL)

−1

)

∂c

∗

τ

∂a

t

. (A38)

where we’ve used h

τ

= bL(1 − ρL)

−1

c

τ

and we assume

β(ρu

22

+ (ρ +b)wu

12

)/(u

22

+wu

12

)

< 1

to ensure convergence. This solves for ∂l

∗

t

/∂a

t

in terms of (current and future) ∂c

∗

τ

/∂a

t

.

We now turn to solving for ∂c

∗

τ

/∂a

t

. The household’s intertemporal optimality (Euler) con-

dition is given by:

1

w

τ

u

2

(c

∗

τ

−h

∗

τ

, l

∗

τ

) = βF

1 +r

τ

w

τ

u

2

(c

∗

τ

−h

∗

τ

, l

∗

τ

). (A39)

Diﬀerentiating (A39) with respect to a

t

and evaluating at steady state yields:

u

12

(1 −F) [1 −bL(1 −ρL)

−1

]

∂c

∗

τ

∂a

t

= −u

22

(1 −F)

∂l

∗

τ

∂a

t

. (A40)

Using (A38) and noting FL = 1 at steady state, (A40) simpliﬁes to:

[1 −β(ρ +b)F] (1 −F) [1 −bL(1 −ρL)

−1

]

∂c

∗

τ

∂a

t

= 0, (A41)

which, from (A40), also implies:

[1 −β(ρ +b)F] (1 −F)

∂l

∗

τ

∂a

t

= 0. (A42)

Equations (A41) and (A42) hold for all τ ≥ t, hence we can invert the [1 − β(ρ + b)F] operator

forward to get:

(1 −F) [1 −bL(1 −ρL)

−1

]

∂c

∗

τ

∂a

t

= 0, (A43)

(1 −F)

∂l

∗

τ

∂a

t

= 0. (A44)

Finally, we can apply (1 −ρL) to both sides of (A43) to get:

(1 −F) [1 −(ρ +b)L]

∂c

∗

τ

∂a

t

= 0, (A45)

which then holds for all τ ≥ t + 1. Thus, whatever the initial responses ∂c

∗

t

/∂a

t

and ∂l

∗

t

/∂a

t

, we

must have:

E

t

∂c

∗

t+1

∂a

t

= (1 +b)

∂c

∗

t

∂a

t

,

E

t

∂c

∗

t+k

∂a

t

= (1 +b(ρ + b)

k−1

)

∂c

∗

t

∂a

t

, (A46)

and E

t

∂l

∗

t+k

∂a

t

=

∂l

∗

t

∂a

t

, k = 1, 2, . . . (A47)

38

Consumption responds gradually to a surprise change in wealth, while labor moves immediately to

its new steady-state level.

From (A46), we can now solve (A40) to get:

∂l

∗

t

∂a

t

= −λ

∂c

∗

t

∂a

t

. (A48)

where

λ ≡

w(1 −β(ρ +b))u

11

+ (1 −βρ)u

12

(1 −βρ)u

22

+ w(1 −β(ρ +b))u

12

=

u

1

u

12

−u

2

u

11

u

1

u

22

−u

2

u

12

, (A49)

where the latter equaltiy follows because w = −

u

2

u

1

1−βρ

1−β(ρ+b)

in steady state.

It remains to solve for ∂c

∗

t

/∂a

t

. The household’s intertemporal budget constraint implies:

E

t

∞

¸

τ=t

(1 +r)

−(τ−t)

∂c

∗

τ

∂a

t

= (1 +r) + w

1 +r

r

∂l

∗

t

∂a

t

. (A50)

Substituting (A46) and (A48) into (A50) and solving for ∂c

∗

t

/∂a

t

yields:

∂c

∗

t

∂a

t

=

(1 −

βb

1−βρ

) r

1 + (1 −

βb

1−βρ

)wλ

. (A51)

Without habits or labor, an increase in assets would cause consumption to rise by the amount of the

income ﬂow from the change in assets—the ﬁrst term on the right-hand side of (A51). The presence

of habits attenuates this change by the amount βb/(1 − βρ) in the numerator of the second term,

and the consumption response is further attenuated by the household’s change in hours worked,

which is accounted for by the denominator of the second term in (A51).

Together, (A48) and (A51) allow us to compute the household’s coeﬃcient of absolute risk

aversion (63) in Proposition 7:

26

−V

11

V

1

+α

V

1

V

=

−u

11

+λu

12

u

1

(1 −

βb

1−βρ

) r

1 +(1 −

βb

1−βρ

)wλ

+ α

r u

1

u

1 −

βb

1 −βρ

. (A52)

The consumption-based coeﬃcient of relative risk aversion is given by:

−AV

11

V

1

+ α

AV

1

V

=

−u

11

+λu

12

u

1

(1 −

βb

1−βρ

) c

1 +(1 −

βb

1−βρ

)wλ

+ α

c u

1

u

1 −

βb

1 −βρ

. (A53)

Equations (A52) and (A53) have obvious similarities to the corresponding expressions without habits

and with expected utility preferences.

26

In order to express (A52) in terms of u

1

and u

11

instead of u

2

and u

22

, we use V

1

= (1−β(ρ+b))u

1

/(β(1−

βρ)) and diﬀerentiate the ﬁrst-order condition:

V

1

(a

t

, h

t

; θ

t

) = (1 + r

t

) (1 −βbF(1 −βρF)

−1

) u

1

(c

∗

τ

−h

τ

, l

∗

τ

),

with respect to a

t

to solve for V

11

.

39

References

Arrow, Kenneth. “Aspects of the Theory of Risk-Bearing,” Yrj¨ o Jahnsson Lectures (1965), in

Essays in the Theory of Risk Bearing, ed. Arrow, Kenneth (Chicago: Markum, 1971).

Backus, David, Bryan Routledge, and Stanley Zin. “Asset Prices in Business Cycle Analysis,”

unpublished manuscript (2008), Tepper School of Business, Carnegie Mellon University.

Benveniste, L.M., and J.A. Scheinkman. “On the Diﬀerentiability of the Value Function in

Dynamic Models of Economics,” Econometrica 47 (1979), 727–32.

Bernanke, Ben, Mark Gertler, and Simon Gilchrist. “The Financial Accelerator in a Quan-

titative Business Cycle Framework,” Handbook of Macroeconomics 1 (1999), 1341–93.

Boldrin, Michele, Lawrence Christiano, and Jonas Fisher. “Habit Persistence and Asset

Returns in an Exchange Economy,” Macroeconomic Dynamics 1 (1997), 312–32.

Boldrin, Michele, Lawrence Christiano, and Jonas Fisher. “Habit Persistence, Asset Re-

turns, and the Business Cycle,” American Economic Review 91 (2001), 149–66.

Campbell, John and John Cochrane. “By Force of Habit: A Consumption-Based Explanation

of Aggregate Stock Market Behavior,” Journal of Political Economy 107 (1999), 205–51.

Constantinides, George. “Habit Formation: A Resolution of the Equity Premium Puzzle,” Jour-

nal of Political Economy 98 (1990), 519–43.

Epstein, Larry and Stanley Zin. “Substitution, Risk Aversion, and the Temporal Behavior of

Consumption and Asset Returns: A Theoretical Framework,” Econometrica 57 (1989), 937–69.

Kihlstrom, Richard and Leonard Mirman. “Risk Aversion with Many Commodities,” Journal

of Economic Theory 8 (1974), 361–88.

King, Robert, Charles Plosser, and Sergio Rebelo. “Production, Growth, and Business

Cycles I: The Basic Neoclassical Model,” Journal of Monetary Economics 21 (1988), 195–232.

Merton, Robert. “An Intertemporal Capital Asset Pricing Model,” Econometrica 41 (1971), 867–

87.

Mortensen, Dale, and Christopher Pissarides. “Job Creation and Job Destruction in the

Theory of Unemployment,” Review of Economic Studies 61 (1994), 397–415.

Pratt, John. “Risk Aversion in the Small and in the Large,” Econometrica 32 (1964), 122–36.

Rudebusch, Glenn, and Eric Swanson. “Examining the Bond Premium Puzzle with a DSGE

Model,” Journal of Monetary Economics 55 (2008), 111–26.

Rudebusch, Glenn, and Eric Swanson. “The Bond Premium in a DSGE Model with Long-Run

Real and Nominal Risks,” Federal Reserve Bank of SF Working Paper 2008–31 (2009).

Santos, Manuel. “Smoothness of the Policy Function in Discrete Time Economic Models,” Econo-

metrica 59 (1991), 1365–82.

Stiglitz, Joseph. “Behavior Towards Risk with Many Commodities,” Econometrica 37 (1969),

660–7.

Tallarini, Thomas. “Risk-Sensitive Business Cycles,” Journal of Monetary Economics 45 (2000),

507–32.

Van Binsbergen, Jules, Jesus Fernandez-Villaverde, Ralph Koijen, and Juan Rubio-

Ramirez. “Working with Epstein-Zin Preferences: Computation and Likelihood Estimation of

DSGE Models with Recursive Preferences,” unpublished manuscript (2008).

Weil, Phillipe. “The Equity Premium Puzzle and the Risk-Free Rate Puzzle,” Journal of Monetary

Economics 24 (1989), 401–21.

1

1. Introduction

In a static, one-period model with household utility u(·) deﬁned over a single consumption good, Arrow (1964) and Pratt (1965) deﬁned the coeﬃcients of absolute and relative risk aversion, −u (c)/u (c) and −c u (c)/u (c). Diﬃculties immediately arise, however, when one attempts to generalize these concepts to the case of many periods or many goods (e.g., Kihlstrom and Mirman, 1974). These diﬃculties are particularly pronounced in a dynamic equilibrium model with labor, in which there is a double inﬁnity of goods to consider— consumption and labor in every future period and state of nature—all of which may vary in response to a typical shock to household income or wealth. The present paper shows how to compute risk aversion in dynamic equilibrium models in general. First, we verify that risk aversion depends on the partial derivatives of the household’s value function V with respect to wealth a—that is, the coeﬃcients of absolute and relative risk aversion are essentially −Vaa /Va and −aVaa /Va , respectively. Even though closed-form solutions for the value function do not exist in general, we nevertheless can derive simple, closed-form expressions for risk aversion because derivatives of the value function are much easier to compute than the value function itself. For example, in many DSGE models the derivative of the value function with respect to wealth equals the current-period marginal utility of consumption (Benveniste and Scheinkman, 1979). The main result of the paper is that the household’s labor margin has substantial eﬀects on risk aversion, and hence asset prices. Even when labor and consumption are additively separable in utility, they remain connected by the household’s budget constraint: in particular, the household can absorb income shocks either through changes in consumption, changes in hours worked, or some combination of the two. This ability to absorb shocks along either or both margins greatly alters the household’s attitudes toward risk. For example, if the household’s utility kernel is given by u(ct , lt ) = c1−γ /(1 − γ) − χlt , the quantity t −c u11 /u1 = γ is often referred to as the household’s coeﬃcient of relative risk aversion, but in fact the household is risk neutral with respect to gambles over income or wealth—the proper measure of risk aversion for asset pricing, as we show in Section 2. Intuitively, the household is indiﬀerent at the margin between using labor or consumption to absorb a shock to income or wealth, and the household in this example is clearly risk neutral with respect to

1+χ gambles over hours. More generally, when u(ct , lt ) = c1−γ /(1−γ)−χ0 lt /(1+χ), risk avert

2 sion equals (γ −1 + χ−1 )−1 , a combination of the parameters on the household’s consumption and labor margins, reﬂecting that the household absorbs shocks using both margins.1 While modeling risk neutrality is not a main goal of the present paper, risk neutrality nevertheless can be a desirable feature for some applications, such as labor market search or ﬁnancial frictions, since it allows for closed-form solutions to key features of the model.2 A contribution of the present paper is to show ways to model risk neutrality that do not require utility to be linear in consumption, which has undesirable implications for interest rates and consumption growth. Instead, any utility kernel with zero discriminant can be used. A ﬁnal result of the paper is that risk premia computed using the Lucas-Breeden stochastic discounting framework are essentially linear in risk aversion. That is, measuring risk aversion correctly—taking into account the household’s labor margin—is necessary for understanding asset prices in the model. Since much recent research has focused on bringing dynamic stochastic general equilibrium (DSGE) models into closer agreement with asset prices,3 it is surprising that so little attention has been paid to measuring risk aversion correctly in these models. The present paper aims to ﬁll that void. There are a few previous studies that extend the Arrow-Pratt deﬁnition beyond the one-good, one-period case. In a static, multiple-good setting, Stiglitz (1969) measures risk aversion using the household’s indirect utility function rather than utility itself, essentially a special case of Proposition 1 of the present paper. Constantinides (1990) measures risk aversion in a dynamic, consumption-only (endowment) economy using the household’s value function, another special case of Proposition 1. Boldrin, Christiano, and Fisher (1997) apply Constantinides’ deﬁnition to some very simple endowment economy models for which they can compute closed-form expressions for the value function, and hence risk aversion. The present paper builds on these studies by deriving closed-form solutions for risk aversion in dynamic equilibrium models in general, demonstrating the importance of the labor margin, and showing the tight link between risk aversion and asset prices in these models.

Note that the intertemporal elasticity of substitution in this example is still 1/γ, so a corollary of this result is that risk aversion and the intertemporal elasticity of substitution are nonreciprocal when labor supply can vary.

2 3 1

See, e.g., Mortensen and Pissarides (1994), and Bernanke, Gertler, and Gilchrist (1999).

See, e.g., Boldrin, Christiano, and Fisher (2001), Tallarini (2000), Rudebusch and Swanson (2008, 2009), Van Binsbergen, Fernandez-Villaverde, Koijen, and Rubio-Ramirez (2008), and Backus, Routledge, and Zin (2009).

3 The remainder of the paper proceeds as follows. Section 2 presents the main ideas of the paper, deriving Arrow-Pratt risk aversion in dynamic equilibrium models for the timeseparable expected utility case and demonstrating the importance of risk aversion for asset pricing. Section 3 extends the analysis to the case of generalized recursive preferences (Epstein and Zin, 1989), which have been the focus of much recent research at the boundary between macroeconomics and ﬁnance. Section 4 extends the analysis to the case of internal and external habits, two of the most common intertemporal nonseperabilities in preferences in both the macroeconomics and ﬁnance literatures. Section 5 discusses some general implications and concludes. An Appendix provides details of derivations and proofs that are outlined in the main text.

**2. Time-Separable Expected Utility Preferences
**

To highlight the intuition and methods of the paper, we consider ﬁrst the case where the household has additively time-separable expected utility preferences.

**2.1 The Household’s Optimization Problem and Value Function
**

Time is discrete and continues forever. At each time t, the household seeks to maximize the expected present discounted value of utility ﬂows:

∞

Et

τ =t

β τ −t u(cτ , lτ ),

(1)

**subject to the sequence of asset accumulation equations: aτ +1 = (1 + rτ )aτ + wτ lτ + dτ − cτ , and transversality condition:
**

T T →∞

τ = t, t + 1, . . .

(2)

lim

(1 + rτ +1 )−1 aT +1 ≥ 0,

(3)

τ =t

where Et denotes the mathematical expectation conditional on the household’s information set at the beginning of period t, β ∈ (0, 1) is the household’s discount factor, ct ≥ 0 and l] lt ∈ [0, ¯ are the household’s choice of consumption and labor in period t, at is the household’s beginning-of-period assets, and wt , rt , and dt denote the real wage, interest rate, and net

Assumption 1 guarantees the existence of a unique optimal choice for (ct . Santos (1991) provides relatively mild suﬃcient conditions for Assumption 3 to be satisﬁed. Conditional on θt . decreasing in its second. . given (at . V can be written as: ∗ V (at . Letting c∗ ≡ c∗ (at . lt ) at each point in time. and dt . The state vector and information set of the household’s optimization problem at each date t is thus (at .e. intuitively. is increasing in its ﬁrst argument. there is no satiation). θt ) that the household may face. t+1 t (5) To avoid having to consider boundary solutions. Assumptions 1–2 also guarantee that V is continuously diﬀerentiable and satisﬁes the Benveniste-Scheinkman equation. lt ) + βEt V (a∗ . ¯ ∈ (0. The value function V (·. Assumption 2 requires the partial derivatives of u to grow suﬃciently large toward ∗ the boundary that only interior solutions for c∗ and lt are optimal for the set of possible t (at . θt+1 ). ∞]. θt ) denote the value function for the household’s optimization problem. u must be strongly concave. The function u : R+ × [0. ¯ → R. lt ) + βEt V (at+1 . but we will require slightly more than this below: Assumption 3. θt ) = max u(ct . θt ) = u(c∗ . we make the following standard assumption: ∗ Assumption 2.lt (4) ∗ where at+1 is given by (2). θt ). θt ). and dt . θt+1 ). the household’s optimal choice (c∗ . ct . For any feasible state (at . Then V satisﬁes the Bellman equation: V (at . Let V (at . the household knows the time-t values for wt . Note that since u is increasing in consumption (i. condition (3) holds with equality at the optimum. rt . twice-diﬀerentiable. We make the following regularity assumptions regarding the utility kernel u: l) l Assumption 1. θt ) denote the household’s t optimal choices of ct and lt as functions of the state (at . rt . θt ). lt ) lies in t the interior of R+ × [0. ¯ l). t t+1 ∗ where a∗ ≡ (1 + rt )at + wt lt + dt − c∗ . θt ) and lt ≡ l∗ (at .4 transfer payments at time t. ·) is twice-diﬀerentiable. Intuitively. and concave. There is a ﬁnite-dimensional Markovian state vector θt that is exogenous to the household and governs the processes for wt . Assumption 3 also implies diﬀerentiability of the optimal policy functions. θt ). c∗ and l∗ ..

the other households and production sector jointly determine the process for θt (and hence wt . xt = xt+1 = xt+k for k = 1. and dt ). Assumption 4 implies that the individual household’s choices for ct and lt have no eﬀect on the aggregate quantities wt . . At the nonstochastic steady state. closed-form expressions. 2. and x ∈ {c. the assumptions used here are standard and thus the most natural to pursue. l.5 2. and θt . However. This reference point is important because it makes it much easier to compute closed-form expressions for many features of the model. It also implies that. r. However. r. The model has a nonstochastic steady state. and much of the analysis below does not need to be any more speciﬁc about these processes than this. l. θt ). It is important to note that Assumptions 4–5 do not prohibit us from oﬀering an individual household a hypothetical gamble of the type described below. . Implicitly. The steady state of the model serves only as a reference point around which the aggregate variables w. . 2. leaving the other households of the model and the production side of the economy unspeciﬁed. d. (at .4 Assumption 4. rt . rt . The household is atomistic and representative. d. Given this state. when the economy is at the nonstochastic steady state (described shortly). and a can be predicted with certainty. we adopt two standard assumptions from the DSGE literature. we have considered the case of a single household. we consider the household’s aversion to a hypothetical Alternative assumptions about the nature of the other households in the model or the production sector may also allow for closed-form expressions for risk aversion. and θ and the other households’ choices of c. or a balanced growth path that can be renormalized to a nonstochastic steady state after a suitable change of variables.2 Representative Household and Steady State Assumptions Up to this point. 4 . to move from general expressions for household risk aversion to more concrete. dt . w. .3 The Coeﬃcient of Absolute Risk Aversion The household’s risk aversion at time t generally depends on the household’s state vector at time t. Assumption 5. any individual household ﬁnds it optimal to choose the steady-state values of c and l given a and θ. and we drop the subscript t to denote the steady-state value. a. θ}.

is the household’s coeﬃcient of absolute risk aversion.θ Et V1 (at+1 t+1 ) (8) where V1 and V11 denote the ﬁrst and second partial derivatives of V with respect to its ﬁrst argument. V1 (a. and lτ for all τ ≤ t. the gamble in (6) is exactly the right framework for thinking about asset prices. cτ . θ) Proof: See Appendix. (6) where εt+1 is a random variable with mean zero and unit variance that represents the gamble. below. for inﬁtesimal dμ and dσ that make the household just indiﬀerent between (6) and (7). to which we will return in Section 2. 6 5 (9) .6. The household’s coeﬃcient of absolute risk aversion with respect to the gamble described in (6) is given by: − Et V11 (a∗ . We defer discussion of relative risk aversion until the next subsection because deﬁning total household wealth is complicated by the presence of human capital—that is.6 one-shot gamble in period t of the form: at+1 = (1 + rt )at + wt lt + dt − ct + σεt+1 . both of which are exogenous to the household at time t. Second. the gamble is dated t + 1 to clarify that its outcome is not in the household’s information set at time t. (6) is clearly equivalent to a gamble over net transfers dt or asset returns rt . and ct is a choice variable under control of the household at time t. (7) The quantity 2dμ/dσ 2 . ∗ . thinking of the gamble as being over rt helps clarify the connection between (6) and asset prices. θ) . (8) simpliﬁes to: −V11 (a. Following Arrow (1964) and Pratt (1965). Evaluated at the steady state. However.5 A few words about (6) are in order: First. As shown there. θt+1 ) t+1 . neither at nor ct can be the subject of the gamble: at is a state variable known with certainty at t. we can ask what one-time fee μ the household would be willing to pay in period t to avoid the gamble in (6): at+1 = (1 + rt )at + wt lt + dt − ct − μ.6 Proposition 1. Indeed. the household’s labor income. and independent of the household’s variables aτ . The gamble εt+1 is assumed to be independent of the exogenous state variables θτ for all times τ .

when they are evaluated at the nonstochastic steady state. For example. lt ). In (10). which has no clear relationship to (8)–(9) except in the one-good one-period case. θt ) = (1 + rt ) ∗ u11 (c∗ .7 Deriving the coeﬃcient of absolute risk aversion in Proposition 1 is simple enough. lt ). lt ) . lt ) t ∗ ∂c∗ t ∗ ∗ ∂lt + u12 (ct . of course. notably −u11 (c∗ . lt )/ t ∗ u1 (c∗ . Thus. t ∗ We solve for ∂lt /∂at by diﬀerentiating the household’s intratemporal optimality con- dition: ∗ ∗ −u2 (c∗ . lt ). c and l. We compute V11 by noting that (10) holds for general at . t Boldrin. This diﬃculty may help to explain ∗ the popularity of “shortcut” approaches to measuring risk aversion. t (10) states that the marginal value of a dollar of assets equals the marginal utility of consumption times 1 + rt (the interest rate appears here because beginning-of-period assets in the model generate income in period t). we can compute V1 at the nonstochastic steady state by evaluating (10) at that point. even for the simplest DSGE models. t t 7 (12) Arrow (1964) and Pratt (1965) occasionally refer to utility as being deﬁned over “money”. . rather than the curvature of the utility kernel u with respect to consumption. so one could argue that they always intended for risk aversion to be measured using indirect utility or the value function. Christiano. consumption-only endowment economy models. and have obvious similarities to Arrow (1964) and Pratt (1965). θt ) = (1 + rt ) u1 (c∗ . and Fisher (1997) derive closed-form solutions for V —and hence risk aversion—for some very simple. u1 is a known function. Here.7 Equations (8)–(9) are essentially Constantinides’ (1990) deﬁnition of risk aversion. it is the curvature of the value function V with respect to assets that matters. hence we can diﬀerentiate (10) to yield: V11 (at . ∂at ∂at (11) ∗ All that remains is to ﬁnd the derivatives ∂c∗ /∂at and ∂lt /∂at . the points c∗ and lt often are known— t for example. lt ) = wt u1 (c∗ . but the problem with (8)–(9) is that closed-form expressions for V (and hence V1 and V11 ) do not exist in general. This approach is a nonstarter for even the simplest DSGE models that include labor. Although closed-form solutions ∗ for the functions c∗ and l∗ are not known in general. the Benveniste-Scheinkman equation: ∗ V1 (at . We solve this problem by observing that V1 and V11 often can be computed even when closed-form solutions for V cannot be.

lt ) − u2 (c∗ . lt )u12 (c∗ . lt+1 ).10 8 9 We do not require this restriction in the analysis below. lt ) t t t t t t (14) Note that. lt+1 )) t+1 ∂at . λt = λt+1 = λ. and the expected change in consumption at each future date t + k. . . k = 1. whatever the change in the household’s consumption today. ∂c∗ /∂at should not be too diﬃcult to compute: it is just the household’s t marginal propensity to consume today out of a change in assets. lt ) − λt u12 (ct . t+2 ∗ ∂lt+2 /∂at . ∂at ∂at ∂at (18) (19) In other words. lt ) t t t t t t = ∗ ∗ ∗ ∗ ∗ ∗ . By ∂c∗ /∂at we mean: t+1 ∂at ∂c∗ t+1 ∂c∗ da∗ t+1 t+1 ∂at+1 dat ∂c∗ t+1 ∂at+1 ∗ ∂c∗ ∂lt − t . ∂c∗ /∂at . even starting from steady state.8 It now only remains to solve for the derivative ∂c∗ /∂at . ∂at ∂at (13) where λt ≡ ∗ ∗ ∗ ∗ ∗ ∗ wt u11 (c∗ . lt ) u1 (c∗ . lt )u12 (c∗ . in a model with internal habits. . which we can deduce from the household’s Euler equation and budget constraint. if consumption and leisure in period t are normal goods. Note that this equality does not follow from the steady state assumption. it must be the same as the expected change in consumption tomorrow. giving: ∂c∗ ∂c∗ ∂c∗ t = Et t+1 = Et t+k . lt ) t ∗ ∂c∗ ∂l∗ ∂c∗ t ∗ ∂l ∗ ∗ + u12 (c∗ . . which we will consider in Section 4.8 with respect to at . ∂at ∂at ∂at ∗ ∗ ∂l∗ ∂l ∂lt = Et t+1 = Et t+k . lt )) ∂at = ∂c∗ ∗ ∗ ∗ ∗ βEt (1 + rt+1 ) (u11 (ct+1 . lt+1 ) t+1 + u12 (c∗ . lt+1 ) t+1 t t+1 t+1 ∂at ∂at ∂at ∂at (16) ∗ Substituting in for ∂lt /∂at gives: ∗ ∗ ∗ ∗ ∗ ∂ct (u11 (ct . lt ) + u12 (c∗ . For example. β = (1 + r)−1 . lt ) u1 (c∗ . etc. then λt must be positive. lt ) t = βEt (1 + rt+1 ) u11 (c∗ . and rearranging terms to yield: ∗ ∂c∗ ∂lt = −λt t . lt ) − u2 (c∗ . Diﬀerentiating the Euler equation: ∗ ∗ u1 (c∗ . lt ) = βEt (1 + rt+1 ) u1 (c∗ . lt+1 ) − λt+1 u12 (ct+1 . t Intuitively. . the individual household’s optimal consumption response to a change in assets increases with time. 2. . lt )u11 (c∗ . . u22 (c∗ . k = 1. lt ) + wt u12 (c∗ . t t+1 (15) with respect to at yields:9 ∗ u11 (c∗ . (17) Evaluating (17) at steady state. 2. and the uij cancel. ∂at ∂at = = 1 + rt+1 + wt and analogously for 10 ∗ ∂lt+1 /∂at . but intuitively we will think of λ > 0. lt )u22 (c∗ .

We can now compute the household’s coeﬃcient of absolute risk aversion. the household raises consumption in every period by the extra asset income. ∂at 1 + wλ (21) In response to a unit increase in assets. θ) = . r ∂at r ∂at (20) That is. Proposition 2 has the following corollary: Corollary 3. (13)–(14). which translates . satisﬁes: −u11 + λu12 r −V11 (a. The household’s coeﬃcient of absolute risk aversion in Proposition 1. is given by: −V11 (a. l). the household’s coeﬃcient of absolute risk aversion in (23) is just the traditional measure. and (21) into (9). V1 (a. First. When there is no labor margin in the model. Diﬀerentiating (2) with respect to at . (18). r. and (19) gives: ∗ 1 + r ∂c∗ 1 + r ∂lt t w = (1 + r) + . and u12 denote the corresponding partial derivatives of u evaluated at the steady state (c. −u11 /u1 . times r. Proposition 2 and Corollary 3 are remarkable. adjusted downward by the amount 1 + wλ that takes into account the household’s decrease in hours worked. θ) −u11 = r. we have proved: Proposition 2. u11 . Suppose that lt is ﬁxed exogenously at l ≥ 0 for all t and the household chooses ct optimally at each t given this constraint. θ) u1 (23) Proof: The assumptions and steps leading up to Proposition 2. evaluated at steady state. we can solve for ∂c∗ /∂at evaluated at the steady state: t r ∂c∗ t = . evaluating at steady state. Then the household’s coeﬃcient of absolute risk aversion (22). Substituting (10). θ) u1 1 + wλ (22) where u1 . are essentially the same as the above with λt = 0. and λ is given by (14) evaluated at steady state. adjusted to the onedimensional case. the expected present value of changes in household consumption must equal the change in assets (times 1 + r) plus the expected present value of changes in labor income. evaluated at steady state.9 The household’s budget constraint is implied by asset accumulation equation (2) and transversality condition (3). Combining (20) with (13). V1 (a. (11). and applying (3).

u12 . l)-space along which u is ﬂat. the household can still be risk neutral: Corollary 4. households can partially absorb shocks to income through changes in hours worked. whether or not we can solve for V . l) = u(c − l) to second order for some function u. risk aversion depends on the concavity of u in all dimensions rather than just in one dimension. the traditional.e. resulting in risk-neutral behavior. ˜ In other words. If the household faces a one-shot gamble of size At in . even when consumption and labor are additively separable in u (i. Even when u11 is very large. Even in the additively separable case.5. The household’s coeﬃcient of absolute risk aversion (22) vanishes as the discriminant u11 u22 − u2 vanishes. We provide some more concrete examples of risk aversion calculations in Section 2. regardless of whether u or the rest of the model is homothetic. risk aversion in Proposition 2 is further attenuated or ampliﬁed by the direct interaction between consumption and labor in utility. The corollary rules out that case by assumption. the special case u(c. |u22 |} 12 in the limit. Proposition 2 shows that risk aversion is less than the traditional measure by the factor 1 + wλ. Proof: The corollary is stated as a limiting result to respect concavity in Assumption 1. Substituting out λ and w. As a result. In other words. For example. static measure of risk aversion is also the correct measure in the dynamic context. u12 = 0). if there exists any direction in (c. after ﬁrst deﬁning relative risk aversion.10 assets into current-period consumption. Geometrically. When u12 = 0. when households have a labor margin. (22) vanishes as u11 u22 − u2 vanishes except for the special case 12 ˜ u1 = −u2 and u11 = −u12 = u22 —that is. the household still can be risk neutral if u22 is small or the cross-eﬀect u12 is suﬃciently large. c∗ depends on household labor supply. The household’s labor margin can have dramatic eﬀects on risk aversion. below. the household will optimally choose to absorb shocks to income along that line. no matter how large the traditional measure −u11 /u1 . and no matter what the functional forms of u and V . for any utility kernel u.4 The Coeﬃcient of Relative Risk Aversion The diﬀerence between absolute and relative risk aversion is the size of the hypothetical gamble faced by the household. so long as either u1 = −u2 or u12 < max{|u11 |. 2. More generally. so labor and consumption are indirectly t connected through the budget constraint.

when the household’s time endowment is not well-deﬁned—as when u(ct . total household wealth At equals the present discounted value of leisure wt (¯− lt ) plus consumption c∗ . 2dμ/dσ 2 . then it follows from Proposition 1 that the household’s coeﬃcient of risk aversion. so we consequently deﬁne two measures of household wealth At and two coeﬃcients of relative risk aversion (25). τ t The factor (1 + rt )−1 in the deﬁnition expresses wealth At in beginning.11 period t. lτ )/u1 (c∗ . In these models. θ) = . Et V1 (a∗ . Equivalently. when the household’s time endowment ¯ is well speciﬁed. that is: at+1 = (1 + rt )at + wt lt + dt − ct + At σεt+1 . and where mt. we can deﬁne l human capital to be the present discounted value of the household’s time endowment. θt+1 ) t+1 (25) The natural deﬁnition of At . there are two natural deﬁnitions of human capital. is the household’s wealth at time t. with At ≡ (1 + rt )−1 Et τ =t mt. We state this formally as: Deﬁnition 1. θ) u1 1 + wλ (26) Alternatively.τ c∗ . for this gamble is given by: − At Et V11 (a∗ . considered by Arrow (1964) and Pratt (1965).rather than endof-period-t units.τ denotes the stochastic discount factor β τ −t u1 (c∗ . the present discounted value of household τ ∗ ∗ consumption. household wealth can be more diﬃcult to deﬁne because of the presence of human capital. wt lt . (24) or the household can pay a one-time fee At μ in period t to avoid this gamble. The household’s consumption-based coeﬃcient of relative risk aversion is ∞ given by (25). θt+1 ) t+1 . lt ) = 1+χ c1−γ /(1 −γ) −lt and no upper bound on lt is speciﬁed—it is most natural to deﬁne human t ∗ capital as the present discounted value of labor income. which follows from the budget constraint (2)–(3). however. wt ¯ In l. V1 (a. lt ). We thus have: t . First. from (2)–(3). ˜ l ∗ thise case. The gamble in (24) is then over a fraction of the household’s wealth and (25) is referred to as the household’s coeﬃcient of relative risk aversion. total household wealth At equals the present discounted value of consumption. In DSGE models. so that in steady state A = c/r and the consumption-based coeﬃcient of relative risk aversion is given by: −u11 + λu12 c −A V11 (a.

corresponding to alternative deﬁnitions of wealth and the size of the gamble At . A = c + w(¯ − l) /r. in steady state the household consumes exactly the ﬂow of income from its wealth.12 Deﬁnition 2. The household’s intertemporal elasticity of substitution. which equals −u1 / c(u11 − λu12 ) by a calculation along the lines t ∗ ∗ of (17). Third. when λ = 0. V1 (a. both measures of risk aversion reduce to the traditional −c u11 /u1 when there is no labor margin in the model—that is.5 Examples Some simple examples illustrate how ignoring the household’s labor margin can lead to wildly inaccurate measures of the household’s true attitudes toward risk. (c + w(¯ − l))/c. l Other deﬁnitions of relative risk aversion. is given by dc∗ − t+1 dc∗ /d log(1 + rt+1 ).τ c∗ + wτ (¯ − lτ ) . l Proof: Note that the case w = 0 is ruled out by Assumptions 1–2. l τ ˜ In steady state. Evaluated at steady state: i) the consumption-based coeﬃcient of relative risk aversion and intertemporal elasticity of substitution are reciprocal if and only if λ = 0. Second. First. (26) and (27) are related by the ratio of the two gambles. holding wt ﬁxed but allowing lt and lt+1 to vary endogenously. We close this section by noting that neither measure of relative risk aversion is reciprocal to the intertemporal elasticity of substitution: Corollary 5. (27) 2. θ) u1 1 + wλ Of course. are also possible. ii) the leisure-and-consumption-based coeﬃcient of relative risk aversion and intertemporal elasticity of substitution are reciprocal if and only if λ = (¯ − l)/c. . The household’s leisure-and-consumption-based coeﬃcient of relative risk ∞ ∗ ˜ aversion is given by (25). evaluated at steady state. The corollary follows from comparing this expression to (26) and (27). rA. θ) −u11 + λu12 c + w(¯ − l) l = . consistent with standard permanent income theory (where one must include the value of leisure w(¯− l) as part of consumption when the value l of leisure is included in wealth). and the leisure-and-consumption-based coeﬃcient of l relative risk aversion is given by: ˜ −A V11 (a. with At = At ≡ (1 + rt )−1 Et τ =t mt. both deﬁnitions reduce to the usual present discounted value of income or consumption when there is no human capital in the model. but Deﬁnitions 1–2 are the most natural for several reasons.

for several diﬀerent values of χ. the bias explodes and true risk aversion approaches zero—the household becomes risk neutral. χ0 > 0. the bias from using the traditional measure is small because the household chooses to absorb income shocks almost entirely along its consumption margin. Intuitively. but the household’s consumption-based coeﬃcient of relative risk aversion is given by (26): 1 −cu11 γ −AV11 = .11 so (29) can be written as: −AV11 1 . The traditional measure of risk aversion for this utility kernel is −c u11 /u1 = γ. reﬂecting that the household absorbs income In steady state. As a result. then the bias from using the traditional measure is small because household labor supply is essentially ﬁxed. so c = wl holds exactly if there is neither capital nor transfers in the model. and households with linear disutility of work are clearly risk neutral with respect to gambles over hours. In any case. if γ is very small. large or small just by varying the household’s time endowment ¯ so we focus only on the consumption-based measure (29). c = ra + wl + d. ≈ 1+ 1 V1 γ χ (30) Note that (30) is less than the traditional measure of risk aversion by a factor of 1 + γ/χ. If χ is very large. ra + d is typically small for standard calibrations in the literature. the labor margin is again almost inoperative. Expression (30) also helps to clarify several points.13 Example 2.12 However. if γ = 2 and χ = 1—parameter values that are well within the range of estimates in the literature—then the household’s true risk aversion is less than the traditional measure by a factor of about three. 11 12 Similarly. 1−γ 1+χ (28) where γ. . a common benchmark in the literature. Consider the additively separable utility kernel: u(ct . This point is illustrated in Figure 1.1. wu11 = γ V1 u1 1 + w u22 1 + χ wl c (29) The household’s leisure-and-consumption-based coeﬃcient of relative risk aversion (27) is not well deﬁned in this example (the household’s risk aversion can be made arbitrarily l). lt ) = l1+χ c1−γ t − χ0 t . Thus. households with linear disutility of work are risk neutral with respect to gambles over wealth because they can completely oﬀset those gambles at the margin by working more or fewer hours. First. In steady state. χ. risk aversion in the model is a combination of both parameters γ and χ. as χ approaches 0. which graphs the coeﬃcient of relative risk aversion for this example as a function of the traditional measure. γ. c ≈ wl.

Second. u(ct . ¯ = 1. χ > 0. lt ) = 1+χ c1−γ /(1− γ) − χ0lt /(1+ χ) in Example 2.1. having an additional margin with which to absorb income gambles reduces the household’s aversion to risk. Third. for any given γ.2. Consumption-based coeﬃcient of relative risk aversion for the utility kernel u(ct . Put another way. for diﬀerent t values of χ. gambles along both of its margins. The traditional l measure of risk aversion for (31) is γ. Consider the King-Plosser-Rebelo-type (1988) utility kernel: c1−γ (1 − lt )χ(1−γ) . reﬂecting that labor and consumption enter symmetrically into u in this example and play an essentially equal role in absorbing income shocks. (30) is symmetric in γ and χ. ignoring the labor margin in this example would be just as erroneous as ignoring the consumption margin.14 10 9 8 = Ccoeffic cient of relative risk aversion 7 6 5 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 =5 =4 =3 =2 =1 =0 10 Figure 1. lt ) = t 1−γ (31) where γ > 0. That is. as a function of the traditional measure γ. depending on χ. See text for details. actual risk aversion in the model can lie anywhere between 0 and γ. γ = 1. Example 2. but the household’s actual leisure-and-consumption- . and χ(1 − γ) < γ for concavity. consumption and labor.

1 Measuring Risk Aversion with V As Opposed to u Some comparison of the expressions −V11 /V1 and −u11 /u1 helps to clarify why the former measure is the relevant one for pricing assets. such as stocks or bonds. In this example. Clearly.15 based coeﬃcient of relative risk aversion is given by: ˜ −u11 + λu12 c + w(1 − l) −A V11 = = γ − χ(1 − γ). As in the previous example. except for the special case χ = 0. in the model. In order for −u11 /u1 to be the relevant measure for pricing a security. as utility approaches Cobb-Douglas— the household becomes risk neutral. −V11 /V1 is the Arrow-Pratt coeﬃcient of absolute risk aversion for gambles over income or wealth in period t. such as a stock or bond. it is the former concept that corresponds to the stochastic payoﬀs of a standard asset. (33) is a combination of the parameters γ and χ. we showed that the labor margin has important implications for Arrow-Pratt risk aversion with respect to gambles over income or wealth. (32) depends on both γ and χ. so the household ﬁnds it optimal to absorb shocks to wealth along that line. From Proposition 1. in a DSGE model. We now show that risk aversion with respect to these gambles is also the right concept for asset pricing.6 Risk Aversion and Asset Pricing In the preceding sections. the expression −u11 /u1 is the risk aversion coeﬃcient for a hypothetical gamble in which the household is forced to consume immediately the outcome of the gamble. Neither (32) nor (33) equals the traditional measure γ. risk aversion is less than the traditional measure by the amount χ(1 − γ). In contrast. depending on χ. depending on χ. 2. household utility along the line ct = wt (1−lt ) is linear. 2. V1 u1 1 + wλ (32) Note that concavity of (31) implies that (32) is positive. in this case. and can lie anywhere between 0 and γ. and can lie anywhere between 0 and the traditional measure γ. V1 u1 1 + wλ 1+χ (33) Again. The household’s consumption-based coeﬃcient of relative risk aversion is a bit more complicated than (32): −A V11 c −u11 + λu12 γ − χ(1 − γ) = = .6. it is not enough that the security pay oﬀ in . As χ approaches γ/(1 − γ)—that is.

16 units of consumption in period t + 1. and hence the labor margin. and Risk Premia Arrow-Pratt risk aversion. It is diﬃcult to imagine such a security in the real world—all standard securities in ﬁnancial markets correspond to gambles over income or wealth. The percentage diﬀerence between the risk-neutral price of the asset and its actual price—the risk premium on the asset—is given by: Et mt+1 Et pt+1 − Et mt+1 pt+1 /Et mt+1 = −Covt (dmt+1 . lt+1 )/u1 (c∗ . dpt+1 )/Et mt+1 (34) where Covt denotes the covariance conditional on information at time t. For example.2 Risk Aversion. lt+1 )dc∗ + u12 (c∗ . lt ) t (35) conditional on information at time t. t+1 t+1 t+1 ∗ u1 (c∗ . if u12 = 0. 2. and d) may change as well as a. we have. changes in lt+1 directly aﬀect the household’s marginal utility of consumption. r. For small changes dc∗ and dlt+1 . lt ) denote the household’s stochastic discount factor t+1 t and let pt denote the cum-dividend price of a risky asset at time t. to ﬁrst order: t+1 dmt+1 = β ∗ ∗ ∗ u11 (c∗ . for which the −V11 /V1 measure of risk aversion is the appropriate one. p}. diﬀerentiating (12) and evaluating at steady state. then dmt+1 equals the t+1 t+1 coeﬃcient of absolute risk aversion times dat+1 . one can already see the relationship between risk aversion and dmt+1 in (35): ∗ if dlt+1 = −λdc∗ and dc∗ = rdat+1 /(1 + wλ). In (35). is also closely tied to asset prices in the standard Lucas-Breeden stochastic discounting framework. the relationship is slightly more complicated than this because θ (and hence w. In actuality. lt+1 )dlt+1 . and dx ≡ xt+1 − ∗ Et xt+1 . t+1 Intuitively.6. ∗ ∗ Let mt+1 = βu1 (c∗ . as in Section 2. The household would additionally have to be prevented from adjusting its consumption and labor choices in period t + 1 in response to the security’s payoﬀs. the household’s labor margin aﬀects mt+1 and hence asset prices for two reasons: First. with Et pt+1 normalized to unity. the Stochastic Discount Factor. u22 + wu12 (36) . the presence of the labor margin aﬀects how the household responds to shocks and hence aﬀects dc∗ . Second. to ﬁrst order: ∗ dlt+1 = −λdc∗ − t+1 u1 dwt+1 .3. even if u12 = 0. we have. so that the household is forced to absorb those payoﬀs into period t+1 consumption. x ∈ {m.

t+k . is then given by: r −u11 + λu12 Covt (dpt+1 . and thus the ﬁrst line of (37) describes the income eﬀect on consumption. dAt+1 ) + r Covt (dpt+1 . combining (2). The last line of (37) describes the substitution eﬀect: changes in consumption due to changes in current and future interest rates and wages. 1 + wλ The risk premium (34). an asset that pays oﬀ well when future interest rates are high or wages are low—and hence future consumption is low—is preferable to an asset that has no correlation with future r or w. r. evaluated at steady state. while the second term captures the asset’s ability to hedge what Merton calls “changes in investment opportunities. In (39).” Intuitively. and d were all held constant. the aggregate variables w. we show in the Appendix that: dc∗ t+1 r = dat+1 + Et+1 1 + wλ ∞ k=1 1 (l dwt+k + ddt+k + adrt+k ) (1 + r)k ∞ (37) 1 u1 −ru1 + dwt+1 + Et+1 2 u11 − λu12 (1 + r)k u11 u22 − u12 k=1 where Rt+1. The term in square brackets in (37) describes the change in the present value of household income.t+k ≡ k i=2 (1 + rt+i ). generalized to include labor. u1 1 + wλ (39) where dAt+1 denotes the quantity in square brackets in (38)—the change in household wealth—and dΨt+1 denotes the summation on the second line of (38)—the change in current and future wages and interest rates. (3).) Substituting (36)–(37) into (35) yields: dmt+1 u11 − λu12 r = β dat+1 + Et+1 u1 1 + wλ ∞ ∞ k=1 1 (l dwt+k + ddt+k + adrt+k ) (1 + r)k (38) − βr Et+1 k=1 1 (1 + r)k λ dwt+k − d log Rt+1. and evaluating at steady state. Equation (39) shows the importance of risk aversion—and hence the labor margin—for asset pricing in the model.t+k .17 Similarly. even for households that are Arrow-Pratt risk neutral. λ dwt+k − d log Rt+1. (Recall that −u1 / c(u11 − λu12 ) is the intertemporal elasticity of substitution. and (15). 1 + wλ Note that for the Arrow-Pratt one-shot gamble considered in Section 2. multiplied by the coeﬃcient of absolute risk aversion. diﬀerentiating.3. dΨt+1 ). Risk premia are essentially linear in the coeﬃcient of absolute risk . Equations (38)–(39) are essentially Merton’s (1973) ICAPM. the ﬁrst term is the covariance of the asset price with household wealth. so (36)–(37) reduce to (13) and (21) in that case.

For an example of this linearity. α = 1. θt+1 )1−α ct . θt ) = max cρ + β Et V (at+1 . As shown above. . Christiano. (41) See the Appendix. 3. Epstein-Zin preferences over consumption streams have been written as: V (at . (40) where α ∈ R. a relationship which also holds for the more general cases of Epstein-Zin preferences and habits.1 in Epstein and Zin (1989) shows that there exists a solution V to (40) with V ≥ 0. What explains Boldrin et al.lt 1/(1−α) . The household’s asset accumulation equation (2) and transversality condition (3) are the same as in Section 2. θt+1 ))1−α ct . Boldrin. but the stochastic discount factor is itself directly linked to risk aversion and the household’s labor margin. the household chooses ct and lt to maximize the recursive expression:15 V (at . considered below. and net transfers. 15 Note that. see Figure 1 of Rudebusch and Swanson (2009). then the proof of Theorem 3. wages. θt ) = max u(ct . this can be seen to correspond to (40). Here we have shown that the same coeﬃcient also appears for completely general gambles that may be correlated with aggregate variables such as interest rates. the preferences given by (40) reduce to the special case of expected utility. and Fisher (1997) argue that it is u11 /u1 rather than V11 /V1 that matters for the equity premium in their Figure 2. θt+1 )α t ct ρ/α 1/ρ . If u ≤ 0 everywhere. θt ) = max u(ct . Generalized Recursive Preferences We now turn to the case of generalized recursive preferences. traditionally. lt ) + β Et V (at+1 .’s Figure 2 is that the covariance of equity prices with the short-term interest rate is not being held constant in their model—in particular. as in Epstein and Zin (1989) and Weil (1989). but now instead of maximizing (1). then it is natural to let V ≤ 0 and reformulate the recursion as: V (at . but with the value function “twisted” and “untwisted” by the coeﬃcient 1 − α.18 aversion. lt ) − β Et (−V (at+1 .lt 13 14 1/(1−α) . When α = 0. If u ≥ 0 everywhere.16 Note that (40) is the same as (4).13 This link between risk aversion and risk premia should not be too surprising: Arrow-Pratt risk aversion describes the risk premium for the most basic gambles over household income or wealth. but by setting V = V ρ and α = 1 − α/ρ. 16 We exclude the case α = 1 here for simplicity.14 The risk premia on these gambles are determined by the household’s stochastic discount factor. it is in fact V11 /V1 that is crucial. the variance of the risk-free rate in their model changes tremendously over the points in their Figure 2.

When α = 0. The ﬁrst term in (43) is the same as the expected utility case (9). When u ≥ 0 and hence V ≥ 0. l 17 (43) . or u : R+ × [0. θt+1 )−α V11 (a∗ . To avoid the possibility of complex numbers arising in the maximand of (40) or (41). θ) V (a. θ) Proof: See Appendix. θ) −V11 (a. when u and V ≤ 0. while the second term in (43) reﬂects the ampliﬁcation or attenuation of risk aversion from the additional curvature parameter α..g. higher values of α correspond to greater degrees of risk aversion. 3. V1 (a. e. With generalized recursive preferences (40) or (41). by requiring c ≥ 1 for u(c. the intertemporal elasticity of substitution over deterministic consumption paths is exactly the same as in (4). l) = log c + χ(¯ − l). the household’s coeﬃcient of absolute risk aversion with respect to the gamble described by (6) is given by: −Et V (a∗ . θt+1 )2 t+1 V (a∗ . (42)–(43) reduce to (8)–(9). θt+1 ) t+1 . ¯ → R− . (42) simpliﬁes to: V1 (a. θt+1 ) − α t+1 t+1 V1 (a∗ .1 Coeﬃcients of Absolute and Relative Risk Aversion We consider the household’s aversion to the same hypothetical gamble as in (6): Proposition 6. ∗ . l) l) The main advantage of generalized recursive preferences (40) is that they allow for greater ﬂexibility in modeling risk aversion and the intertemporal elasticity of substitution. In (40).θ −α V (a∗ . θ Et V (at+1 t+1 ) 1 t+1 t+1 ) (42) Evaluated at steady state.19 The proof in Epstein and Zin (1989) also demonstrates the existence of a solution V to (41) with V ≤ 0 in this case. one can restrict the domain of u to ensure u ≥ 0 or u ≤ 0. Alternatively. the opposite is true: higher values of α correspond to lesser degrees of risk aversion. ¯ → R+ . θ) +α . Either u : R+ × [0. but the household’s risk aversion to gambles can be ampliﬁed (or attenuated) by the additional parameter α. we restrict the range of u to be either R+ or R− :17 Assumption 6.

lt ) (Et V (a∗ . l 18 Note that. is given by (c + w(¯ − l))/c times (46).18 Expressions (43) and (44) highlight an important feature of risk aversion with generalized recursive preferences: it is not invariant with respect to level shifts of the utility kernel.20 The household’s coeﬃcient of relative risk aversion is given by At times (42). +α V1 V u1 1 + wλ u (45) The household’s consumption-based coeﬃcient of relative risk aversion. evaluated at steady state. At steady state. . expressions (10)– ∗ (21) for V1 . evaluated at steady state. The household’s preferences are invariant. and we refer to (44) as the consumption-based or leisure-and-consumption-based coeﬀcient of relative risk aversion. θt+1 )1−α )1/(1−α) t t+1 which must be used to compute household wealth. with respect to multiplicative transformations of the utility kernel.4. the household’s stochastic discount factor is given by: ⎞ −α ⎛ ∗ βu1 (c∗ . the utility kernels u(·. which. is given by: −AV11 −u11 + λu12 c u1 c AV1 = + α . simpliﬁes to: AV1 (a. θ) V (a. t V = u(c. V1 (a. is given by: −V11 −u11 + λu12 r u1 r V1 = + α . ·)+k. with generalized recursive preferences. θ) +α . Moreover. and ∂c∗ /∂at continue to apply in the current context. When it comes to computing the risk aversion coeﬃcients (43)–(44). as in Section 2. V11 . Substituting these into (43)–(44) gives: Proposition 7. this simpliﬁes to the usual β. except for the special case of expected utility (α = 0). evaluated at steady state. depending on the deﬁnition of A. ∂lt /∂at . however. θ) −AV11 (a. ·) and u(·. lead to diﬀerent household attitudes toward risk. l) (1 − β) at the steady state. where k is a constant. That is. evaluated at steady state. +α V1 V u1 1 + wλ u (46) The household’s leisure-and-consumption-based coeﬃcient of relative risk aversion. however. ∗ u1 (c∗ . θ) (44) We deﬁne the household’s total wealth At based on the present discounted value of its lifetime consumption or lifetime leisure and consumption. The household’s coeﬃcient of absolute risk aversion in Proposition 6. θt+1 ) t+1 t+1 ⎠ ⎝ . lt+1 ) V (a∗ .

is valid for general functional forms u. (48) decreases to α(1 − γ)/γ. Rudebusch and Swanson (2009) refer to γ + α(1 − γ) as the quasi coeﬃcient of relative risk aversion. ·) < 0.19 In this case. consumption-only models. ·) < 0. so we restrict attention to the consumption-based measure (48). and γ > 1. 19 . and Fisher (1997). In models without labor. the household’s consumption-based coeﬃcient of relative risk aversion (46) is given by: γ −AV11 AV1 α(1 − γ) = +α . In models with variable labor. 20 Set χ0 = 0 and λ = 0 and substitute (47) into (46). As χ approaches zero. where the value function can be computed in closed form. risk aversion is decreasing in α. which do not have labor. which we will refer to as the traditional measure. Christiano. in Epstein and Zin (1989) and Boldrin. Of course. 3.21 Proposition 7 is important because risk aversion for Epstein-Zin preferences has only been computed previously in homothetic. The case γ ≤ 1 can be considered if we place restrictions on the domain of ct and lt such that u(·. As in Example 2. Consider the additively separable utility kernel: u(ct . which was used by Rudebusch and Swanson (2009). the household’s leisure-and-consumption-based coeﬃcient of relative risk aversion is not well deﬁned in this example. for example. We restrict attention here to the case γ > 1. χ0 > 0. isoelastic. period utility u(ct . which is close to zero if we think of γ as being small (close to unity). 1−γ 1+χ (47) with generalized recursive preferences (41) and χ > 0. unknown functional forms V . consistent with Assumption 6. ·) > 0. lt ) = c1−γ /(1 − γ) implies a coeﬃcient t of relative risk aversion of γ + α(1 − γ). As χ becomes large.1.20 Taking into account both the consumption and labor margins of (47). 1 + 1+χ (48) using c ≈ wl. one can always choose units for ct and lt such that this doesn’t represent much of a constraint in practice. γ wl + γ−1 V1 V 1+ χ c 1 + 1+χ wl c ≈ γ 1+ γ χ + α(1 − γ) γ−1 . one can also consider alternative utility kernels with γ ≤ 1 for which u(·.2 Examples Example 3. where u(·.1. and allows for the presence of labor. This is the case. lt ) = l1+χ c1−γ t − χ0 t . Proposition 7 does not require homotheticity. household labor becomes less ﬂexible and the bias from ignoring the labor margin shrinks to zero ((48) approaches γ + α(1 − γ)). and α < 0 corresponds to preferences that are more risk averse than expected utility.

1). (52) t+1 (1 − β)(1 − χ) 1+θ l u(ct . after mapping each study’s notation over to the present paper’s. lt ) = t 1−γ 1−γ . call γ + α(1 − γ) the coeﬃcient of relative risk aversion. (53) with utility kernel: We can compute the coeﬃcient of absolute risk aversion for (52) by following along the steps in the proof of Proposition 6. so consumption becomes trivial to insure with variations in labor supply. Tallarini (2000) considers an alternative Epstein-Zin speciﬁcation: ∗ ˜ Vt (at .22 Thus. lt ) = log ct + θ log(¯ − lt ).21 Example 3. which yields: (1 − β)(1 − χ) −V11 (a. +α V1 V ˜ ˜ AV1 −AV11 = γ + α(1 − γ). w/c → ∞. +α V1 V (50) while the leisure-and-consumption-based coeﬃcient of relative risk aversion is: (51) The latter agrees with the Van Binsbergen et al. (2008) and Backus. γ = 1. Van Binsbergen et al. depending on the value of χ. Example 3. θt+1 ) . Substituting (49) into (46). θt ) ≡ u(c∗ . we have provided the formal justiﬁcation for both measures. actual household risk aversion can lie anywhere between about zero and γ + α(1 − γ).2. use γν + α(1 − γ)ν + (1 − ν). (2008) measure of risk aversion. The former measure eﬀectively treats consumption and leisure as a single composite commodity. In this paper. Van Binsbergen et al. V1 (a. and ν ∈ (0. This explains why the consumption-based coeﬃcient of relative risk aversion in (50) vanishes as ν → 0. (50) and (51). θ). lt ) + t β(1 + θ) (1 − β)(1 − χ) ˜ log Et exp Vt+1 (a∗ . for given values of γ and α. (2008) measure. . while the former is similar to (though not quite the same as) the Backus et al. Routledge. while the latter measure allows ν— the importance of the labor margin—to aﬀect the household’s attitudes toward risk. while Backus et al.3. θ) − V1 (a. (49) where γ > 0. and Zin (2008) consider generalized recursive preferences with: cν (1 − lt )1−ν u(ct . the household’s consumption-based coeﬃcient of relative risk aversion is: AV1 −AV11 = γν + α(1 − γ)ν. θ) 1+θ 21 (54) As ν → 0.

For any feasible state. lt ). As θ/χ approaches zero. or habits. but from consumption relative to some reference level. ∞) × [0. if we normalize c to 1 in (55). the household’s optimal choice (c∗ . can have substantial eﬀects on the household’s attitudes toward risk (e. ¯ ¯ ct > ht − h. so substituting in for V1 and V11 in (54) yields a consumption-based coeﬃcient of relative risk aversion of: 1−χ 1−χ c −u11 + λu12 1 − cu1 = − . and Fisher. ¯ → R− . 1997). u1 1 + wλ 1+θ 1 + θc 1+θ (55) The leisure-and-consumption-based coeﬃcient of relative risk aversion is (1 + θ) times (55).. ¯ h ≥ 0. −cu11 u1 − 1−χ 1+θ cu1 . ¯ ¯ Assumption 6 . The rest of Assumption 1 applies. As a result. Habits.g. In this section. l). We generalize the household’s utility kernel in this section to u(ct − ht . or habit stock. Both coeﬃcients of relative risk aversion diﬀer from the value (θ + χ)/(1 + θ) reported by Tallarini (2000). We focus on an additive rather than multiplicative speciﬁcation for habits because the implications for risk aversion are typically more interesting in the additive case. 1999. derived in a consumption-only model under the assumption θ = 0. Internal and External Habits Many studies in macroeconomics and ﬁnance assume that households derive utility not from consumption itself. However. l) l) . in turn. ∞) × [0. 2. Tallarini applies the traditional measure of risk aversion. simply setting θ > 0 in the traditional measure. ∗ Assumption 2 . 4. where ht denotes the household’s reference level of consumption. Christiano. or u : (−h. ∞) × [0. ¯ l). For every t. the labor margin becomes unimportant and this bias disappears. but the bias can be arbitrarily large as the ratio of θ to χ increases. Tallarini’s traditional measure overstates the household’s true aversion to risk by a factor of 1 + θ/χ. The function u is deﬁned over (−h. we investigate how habits aﬀect risk aversion in the DSGE framework. Campbell and Cochrane. We adjust the feasible choice set for ct and Assumptions 1. ¯ → R+ .23 The other steps leading up to Proposition 7 are all the same. to the case where θ > 0. Boldrin. and 6 accordingly. ∞) × [0. lt ) lies in the t ¯ interior of (−h. replacing them with: ¯ Assumption 1 . Either u : (−h. ignores the fact that households vary their labor in response to shocks.

1 External Habits When the reference consumption level ht in the utility kernel u(ct − ht . . the coeﬃcient of absolute risk aversion continues to be given by (9) in the case of expected utility and (43) in the case of generalized recursive preferences. Consider the case of expected utility with additively separable utility kernel: 1+χ lt (ct − ht )1−γ u(ct − ht . . . k = 1. Example 4. then the parameters that govern the process for ht can be incorporated into the exogenous state vector θt . l) rather than (c. ﬁrst. (56)–(58) imply: (59) The only real diﬀerences that arise relative to the case without habits is. ·). Together with the budget constraint (2)–(3). and the household’s Euler equation (15) still implies: ∂c∗ ∂c∗ ∂c∗ t+1 t = Et = Et t+k . l). 2. . that relative risk aversion confronts the household with a hypothetical gamble over c rather than c − h. and second.1. We consider each of these cases in turn. ·) are evaluated is (c − h.24 If the habit stock ht is external to the household (“keeping up with the Joneses” utility). (57) (58) evaluated at steady state. lt ) is external to the household. . because the stakes are eﬀectively larger. that the steady-state point at which the derivatives of u(·. ∂at ∂at ∂at r ∂c∗ t . 2. . ∂at ∂at (56) where λt is given by (14). if the habit stock ht is a function of the household’s own past levels of consumption. and the analysis proceeds much as in the previous sections. In particular. The household’s intratemporal optimality condition (12) still implies: ∗ ∂c∗ ∂lt = −λt t . then the parameters that govern ht can be incorporated into the exogenous state vector θt and the analysis of the previous sections carries over essentially as before. which has a tendency to make the household more risk averse for a given functional form u(·. then the state variables of the household’s optimization problem must be augmented to include the state variables that govern ht . However. lt ) = − χ0 . 4. = ∂at 1 + wλ k = 1. ∂at ∂at ∂at ∗ ∗ ∂l∗ ∂l ∂lt = Et t+1 = Et t+k . 1−γ 1+χ (60) .

When there is a labor margin. In the Appendix. the consumption-based measure agrees with the traditional measure. we must specify how the household’s actions aﬀect its future habits. (62) Again. (64) . Ignoring the labor margin in (61) thus leads to an even greater bias in the model with habits (h > 0) than in the model without habits (h = 0). If γ = 2.8c. When the household has generalized recursive preferences rather than expected utility preferences. the bias from ignoring the labor margin in (62) is even greater in the model with habits (h > 0) than without habits (h = 0). χ0 > 0. χ(c−h) using wl ≈ c. in the present section we focus on the case where habits are proportional to last period’s consumption: ht = bct−1 .2 Internal Habits When habits are internal to the household. 4. and h = . = γc (c − h) 1 + χ(c−h) wl c (61) When there is no labor margin in the model (λ = 0). then the household’s true risk aversion is smaller than the traditional measure by a factor of more than ten. The consumption-based coeﬃcient of relative risk aversion is: 1 −AV11 −cu11 = . we derive the corresopnding closed-form expressions for the more complicated case where the habit stock evolves according to the longer-memory process: ht = ρht−1 + bct−1 . (63) b ∈ (0. χ = 1. 1). which exceeds γ by a factor that depends on the importance of habits relative to consumption. and we assume the household has expected utility preferences. the consumption-based coeﬃcient of relative risk aversion for (60) is: γc (c − h) 1 + 1 γc wl χ(c−h) c + α(1 − γ)c (c − h) 1 + 1 γ−1 wl c (c−h) 1+χ c . In order to minimize notation and emphasize intuition.25 where γ. V1 u1 1 + w wu11 u22 1 γc . the household’s consumptionbased coeﬃcient of relative risk aversion (61) is less than the traditional measure by the factor 1 + γc . χ. The traditional measure of risk aversion for this example is −cu11 /u1 = γc/(c − h).

θt+1 )2 t+1 t+1 V (a∗ . h∗ . θt+1 )−α V11 (a∗ . lt ) = βEt V1 (a∗ . h∗ . θt ) and lt ≡ l∗ (at .26 with ρ ∈ (−1. lt ). h∗ . a∗ ≡ (1 + rt )at + wt lt + dt − c∗ and h∗ ≡ bc∗ . ht . (65) ∗ where c∗ ≡ c∗ (at . ∗ . θt+1 ). we derive the corresponding closed-form expressions for the more complicated case of generalized recursive preferences. θt ) = u(c∗ − ht . t t+1 t+1 (66) Evaluated at steady state. lt ) = −βwt Et V1 (a∗ . t+1 t t+1 t The household’s coeﬃcient of absolute risk aversion can be derived in the same manner as in Propositions 1 and 6: Proposition 8. lt ) + β Et V (a∗ . In order to minimize notation and simplify this derivation as much as possible. we restrict attention in the main text to the case of expected utility preferences (α = 0). t t+1 t+1 t+1 t+1 ∗ u2 (c∗ − ht . h. h∗ . θt+1 ) − α t+1 t+1 t+1 t+1 V1 (a∗ . because of the dynamic relationship between the household’s current consumption and its future habits. V1 (a. the value of ht+1 depends on the household’s choices in period t. θ −α V (a∗ . h∗ . θ) Proof: Essentially identical to the proof of Proposition 6. ht . and a∗ and t+1 h∗ denote the optimal stocks of assets and habits in period t + 1 that are implied by c∗ t t+1 ∗ ∗ and lt . ht . θ Et V (at+1 t+1 t+1 ) 1 t+1 t+1 t+1 ) −V11 (a. h∗ . θt+1 ) − βbEt V2 (a∗ . h. 1). θt+1 ). utility kernel u(ct − ht . however. θt+1 )1−α t t+1 t+1 1/(1−α) . h. With generalized recursive preferences (40) or (41). In the Appendix. and internal habits ht given by (63). The household’s ﬁrst-order conditions for (65) with respect to consumption and labor (and imposing α = 0) are given by: ∗ u1 (c∗ − ht . θ) V1 (a. θ) +α . θt+1 ) t+1 t+1 . h∗ . h. that is. h∗ . θ) V (a. θt ) denote the household’s optimal choices for t consumption and labor in period t as functions of the household’s state vector. Computing closed-form expressions for V1 and V11 in (67) is substantially more complicated for the case of internal habits. h∗ . With internal habits. (66) simpliﬁes to: (67) (68) (69) . the household’s coeﬃcient of absolute risk aversion with respect to the gamble (6) is given by: −Et V (a∗ . h∗ . so we write out the dependence of the household’s value function on ht explicitly: ∗ V (at .

t (70) (71) Equations (69) and (70) can be used to solve for V1 in terms of current-period utility: V1 (at . we henceforth let a time subscript τ ≥ t denote a generic future date and reserve the subscript t to denote the date of the current period—the period in which the household faces the hypothetical one-shot gamble. τ τ τ τ τ τ ∗ = wτ (1 − βbF ) u1 (c∗ − h∗ . lt ). t+1 t+1 ∗ V2 (at . lτ ) = wτ u1 (c∗ − h∗ . wt ∂at ∂at (73) where we drop the arguments of the uij to reduce notation. lt ). θt+1 ). in steady state it is also true that V1 = u1 (1 − βb)/β. θt ) = −u1 (c∗ − ht . lτ ). ∗ We solve for ∂lτ /∂at in terms of ∂c∗ /∂at in much the same way as without habits. θt ) = − (1 + rt ) ∗ u2 (c∗ − ht . Diﬀerentiating (75) with respect to at yields: −u12 ∂h∗ ∂c∗ τ τ − ∂at ∂at ∗ ∂h∗ ∂c∗ ∂lτ τ τ = wτ (1 − βbF ) u11 − − u22 ∂at ∂at ∂at ∗ ∂lτ + u12 . τ τ (74) (75) where F denotes the forward operator. h∗ +1 . except that the dynamics t ∗ of internal habits require us to solve for ∂c∗ /∂at and ∂lτ /∂at for all dates τ ≥ t at the τ same time. ht . . diﬀerentiate (72) with respect to at to yield: V11 (at . t wt (72) which states that the marginal value of wealth equals the marginal utility of working fewer hours. ∂at (76) 22 Using the marginal utility of labor is simpler than the marginal utility of consumption in (72) because it avoids having to keep track of future habits and the value function next period. lτ ) + bβEτ V2 (a∗ +1 . θτ +1 ) . which we do in the same manner as before. Diﬀerentiating (65) with respect to its ﬁrst two arguments and applying the envelope theorem yields: V1 (at . However. It now remains to solve for ∗ ∂c∗ /∂at and ∂lt /∂at . The ﬁrst-order condition (68). ht . To solve for V11 . ht . that is F xτ ≡ Eτ xτ +1 for any expression x dated τ . ht . however. which we will use to express risk aversion in terms of u1 and u11 below. θt ) = − ∂c∗ ∂l∗ (1 + rt ) u12 t + u22 t . h∗ . τ The household’s intratemporal optimality condition ((68) combined with (69)) implies: ∗ ∗ −u2 (c∗ − h∗ . θt ) = β(1 + rt ) Et V1 (a∗ .27 Equation (69) is essentially the same as in the case without habits. To better keep track of these dynamics.22 This solves for V1 . includes the future eﬀect of consumption on habits in the second term on the right-hand side.

Lxτ ≡ xτ −1 for any expression x dated τ —and we assume |βbwu12 /(u22 + wu12 )| < 1 in order to ensure convergence. This As before. also implies: (1 − βbF )(1 − F ) ∗ ∂lτ = 0. ∂at ∂l∗ (1 − F ) τ = 0. L denotes the lag operator—that is. we solve for ∂c∗ /∂at using the household’s Euler equation and budget conτ straint. Evaluating (76) at steady state and solving for ∂lτ /∂at yields: ∗ ∂lτ βbwu12 u12 + wu11 − βbwu11 F 1− = − F ∂at u22 + wu12 u22 + wu12 −1 (1 − bL) ∂c∗ τ . . ∂at (83) (84) 23 To be precise. ∂at ∂c∗ τ = 0. and ∂h∗ /∂at = 0 for τ = t τ τ τ τ τ ∗ since ht is given. Formally. ∂at (77) where the uij are evaluated at steady state. ∂at (80) u22 (1 − F ) u12 + wu11 − βbwu11 F (1 − bL) Since F L = 1. τ τ τ τ wτ wτ +1 with respect to at and evaluating at steady state yields: u12 (1 + b) − F − bL ∂l∗ ∂c∗ τ = −u22 (1 − F ) τ . τ ∂c∗ τ . F Lxτ = Eτ −1 xτ . ∂at as in Section 2. (77) reduces to − wu11 +u12 u22 +wu12 ∗ solves for ∂lτ /∂at in terms of (current and future) ∂c∗ /∂at . Diﬀerentiating the household’s Euler equation: 1 1 + rτ +1 ∗ ∗ u2 (c∗ − h∗ . from (79). hence we can invert the (1 − βbF ) operator forward to get: ∂c∗ (1 − F )(1 − bL) τ = 0. but since the household evaluates these expressions from the perspective of the initial period t. lτ +1 ) ∂c∗ +1 /∂at .28 ∗ where F u11 ∂c∗ /∂at denotes Eτ u11 (c∗ +1 − h∗ +1 .23 equation (80) simpliﬁes to: (1 − βbF )(1 − F )(1 − bL) which. Note that when b = 0. ∂at (81) (82) Equations (81) and (82) hold for all τ ≥ t. take the expectation of (80) at time t and then apply Et F L = Et to get (81). Et F Lxτ = Et xτ . ∂at ∂at (79) (78) Substituting (77) into (79) yields the following diﬀerence equation for cτ : u12 u22 + wu12 − βbwu12 F (1 + b) − F − bL − ∂c∗ τ = 0. lτ +1 ). lτ ) = βEτ u2 (c∗ +1 − h∗ +1 .

an increase in assets would cause consumption to rise by the amount of the income ﬂow from the change in assets. . we have: ∞ Et τ =t (1 + r)−(τ −t) ∗ ∂c∗ 1 + r ∂lt τ = (1 + r) + w . 24 However. ∂at ∂at ∂c∗ ∂c∗ Et t+k = (1 + b + · · · + bk ) t . ∂at r ∂at (89) Substituting (85)–(87) into (89) and solving for ∂c∗ /∂at yields: t (1 − βb)r ∂c∗ t = . . and the consumption response is further attenuated by the household’s change in hours worked. From (85). asymptoting over time to its new steady-state level. . Thus. ∂at ∂at ∗ ∗ ∂lt+k ∂lt = . but labor moves immediately to its new steady-state level in response to surprises in wealth. The presence of habits attenuates this change by the amount βb in the numerator of (90). consumption responds only gradually to a surprise change in wealth. r. It now remains to solve for ∂c∗ /∂at . ∂at 1 + (1 − βb)wλ (90) Without habits or labor. and Et ∂at ∂at (85) (86) evaluated at steady state. which is accounted for by the denominator.29 ∗ In other words. unlike the model without habits. whatever the initial responses ∂c∗ /∂at and ∂lt /∂at are. Because of habits. that (87)–(88) are essentially identical to (13)–(14). (87)–(88) only hold here in steady state. we must have: t ∂c∗ ∂c∗ Et t+1 = (1 + b) t . . ∂at ∂at (87) where λ ≡ w(1 − βb)u11 + u12 u1 u12 − u2 u11 = . 2. λ must be positive if leisure and consumption are normal goods. k = 1. From the household’s budget constraint and t condition (86). we can now solve (79) to get: ∗ ∂c∗ ∂lt = −λ t .24 Again. u22 + w(1 − βb)u12 u1 u22 − u2 u12 (88) and where the latter equality follows because w = −(1 − βb)−1 u2 /u1 in steady state.

(92) = V1 u1 1 + (1 − βb)wλ The household’s leisure-and-consumption-based coeﬃcient of relative risk aversion. Example 4. evaluated at steady state. 1+ χ where the last line uses β ≈ 1 and wl ≈ c. In thise case. = γ γ 1−βb wl 1 − b 1 + χ 1−b c γ ≈ γ . V1 u1 1 + (1 − βb)wλ 1 − βb 1 . This is in sharp contrast to the case of external habits. but now with habits ht = bct−1 internal to the household rather than external. t t+1 t+1 1 + rt with respect to at to solve for V11 using (85)–(88) and (90). equation (61)). lt+1 ). is given by: −u11 + λu12 (1 − βb)r −V11 = . V1 u1 1 + (1 − βb)wλ (91) The household’s consumption-based coeﬃcient of relative risk aversion. and (90) into (67). ht . In order to express (91) in terms of u1 and u11 instead of u2 and u22 . we have established:25 Proposition 9. χ. (88).2. evaluated at steady state. we use V1 = (1 − βb)u1 /β and diﬀerentiate the ﬁrst-order condition: 1 ∗ ∗ u1 (c∗ − ht . evaluated at steady state. The most striking feature of equation (94) is that it is independent of b.30 Substituting (72). the household’s consumption-based coeﬃcient of relative risk aversion is given by: −AV11 1 − βb −cu11 = . u(ct − ht . lt ) = 1−γ 1+χ (93) where γ. χ0 > 0. (87). is given by (c + w(¯ − l))/c times (92).1: 1+χ lt (ct − ht )1−γ − χ0 . is given by: (1 − βb)c −u11 + λu12 −AV11 . Consider the utility kernel of example 4. 25 (94) . the importance of habits. θt ) + βbEt u1 (c∗ − h∗ . where risk aversion is strongly increasing in b (cf. (73). The household’s coeﬃcient of absolute risk aversion in Proposition 8. l Equations (91)–(92) have essentially the same form as the corresponding expressions in the model without habits. lt ) = V1 (at .

Indeed. or futures. the traditional measure can easily overstate risk aversion by a factor of three or more. should be useful to researchers interested in pricing any asset—stocks. closed-form expressions for risk aversion that this paper derives. and the methods of the paper more generally. Understanding how labor supply aﬀects asset prices is thus important for bringing DSGE-type models closer to ﬁnancial market data. asset prices in DSGE models can be very diﬀerent and can behave very diﬀerently depending on how the household’s labor margin is speciﬁed. In these applications. even for expected utility preferences. households can even be risk neutral when the traditional measure of risk aversion is far from zero. bonds. The simple. The present paper suggests new ways of modeling risk neutrality in a DSGE framework. ignores the household’s ability to partially oﬀset shocks to income with changes in hours worked. As a result. There is a wedge between the two concepts that depends on the household’s labor margin. risk neutrality allows for much simpler or even closed-form solutions to key aspects of the model. Discussion and Conclusions The ability to vary labor supply has dramatic eﬀects on household risk aversion and asset prices in dynamic equilibrium models. Risk aversion matters for asset pricing. The traditional approach—linearity of utility in consumption—has undesirable implications for interest rates and consumption growth. −cu11 /u1 . in foreign or domestic currency—within the framework of . For reasonable parameterizations. A related observation is that risk aversion and the intertemporal elasticity of substitution are nonreciprocal. An extreme example of this is when household utility has a zero discriminant—implying risk neutrality—even when the traditional measure of risk aversion is large. but the present paper shows that any utility kernel with a zero discriminant can be used instead. and international literatures thus may be overstating the actual degree of risk aversion in their models by a substantial degree. macro-ﬁnance. The traditional measure of risk aversion. Many studies in the macroeconomics. such as labor market search or ﬁnancial frictions. If risk aversion is measured incorrectly because the labor margin is ignored. then risk premia in the model are also more likely to be surprising or puzzling. Risk neutrality itself can be a desirable feature for some applications.31 5. Risk premia on assets computed using the stochastic discount factor are essentially linear in the degree of risk aversion.

Since these models are a mainstay of research in academia. at central banks. . the applicability of the results should be widespread.32 dynamic equilibrium models. and international ﬁnancial institutions.

θt+1 )εt+1 = 0 because εt+1 is independent of θt+1 and a∗ . so that: ∗ V (at . Because (6) describes a one-shot gamble in period t. (A4) simpliﬁes to: βEt V11 (a∗ . Thus. it t+1 t aﬀects assets a∗ in period t + 1 but otherwise does not aﬀect the household’s optimization problem t+1 from period t + 1 onward. note ﬁrst that the household’s optimal choices for con∗ sumption and labor in period t. will generally depend on the size of the gamble σ—for t example. εt+1 is independent of θt+1 and a∗ . the linearized version of the model is certainty equivalent. σ) to emphasize this dependence on σ. The tilde over the V on the left-hand side of (A2) t+1 emphasizes that the form of the value function itself is diﬀerent in period t due to the presence of the one-shot gamble in that period. alternatively. in ∗ this section we write c∗ ≡ c∗ (at . the eﬀect of the gamble on household welfare is: u11 ∂c∗ ∂σ 2 + 2u12 ∂c∗ ∂l∗ + u22 ∂σ ∂σ ∂l∗ ∂σ 2 + u1 2 ∂ 2 c∗ ∂ 2 l∗ + u2 2 ∂σ 2 ∂σ ∂ 2 c∗ ∂ 2 l∗ − ∂σ 2 ∂σ 2 dσ 2 . the ﬁrst-order cost of the gamble is zero. 2 (A4) + βEt V11 · wt ∂c∗ ∂l∗ − + εt+1 ∂σ ∂σ + βEt V1 · wt ∗ The terms involving ∂ 2 c∗ /∂σ 2 and ∂ 2 l∗ /∂σ 2 cancel due to the optimality of c∗ and lt . for inﬁnitesimal gambles. lt ) + βEt V (a∗ . Et V1 (a∗ . σ) = u(c∗ . σ) and lt ≡ l∗ (at . θt+1 ) ε2 t+1 t+1 dσ 2 . Turning now to the gamble in (6). 2 (A5) Finally. t+1 1 + rt (A1) where the right-hand side of (A1) follows from the envelope theorem. θt . hence c∗ and l∗ must be symmetric about σ = 0. t+1 which does not depend on σ except through a∗ . the household’s value-to-go at time t also depends on σ. θt . t t+1 (A2) ∗ where a∗ ≡ (1 + rt )at + wt lt + dt − c∗ . θt . The derivatives t ∂c∗ /∂σ and ∂l∗ /∂σ vanish at σ = 0 (there are two ways to see this: ﬁrst. Diﬀerentiating (A2) with respect to σ. θt+1 ). θt ) dμ = −βEt V1 (a∗ . Thus. the household’s value-to-go at time t + 1 is just V (a∗ . as in the usual envelope theorem (these derivatives vanish at σ = 0 anyway. the household may undertake precautionary saving when faced with this gamble.33 Appendix: Mathematical Derivations Proof of Proposition 1 For an inﬁnitesimal fee dμ in (7). u2 . To second order. c∗ and lt . for the reasons discussed below). Moreover. We write this dependence out explicitly as well. Thus. θt+1 ) dμ . the gamble in (6) is isomorphic for positive and negative σ. as in Arrow (1964) and Pratt (1965). to ﬁrst order. t ∗ ∗ Because ct and lt depend on σ. evaluating the latter at t+1 t+1 σ = 0. as a result. implying the derivatives vanish). evaluating the latter at σ = 0. the change in household welfare (5) is given. the ﬁrst-order eﬀect of the gamble on household welfare is: ∂c∗ ∂c∗ ∂l∗ ∂l∗ + u2 + βEt V1 · (wt − + εt+1 ) dσ. Since εt+1 has unit t+1 variance. Optimality of c∗ and t ∗ lt implies that the terms involving ∂c∗ /∂σ and ∂l∗ /∂σ in (A3) cancel. by: −V1 (at . (A6) t+1 2 . (A5) reduces to: dσ 2 βEt V11 (a∗ . θt+1 ) . and V1 are suppressed to simplify notation. θt+1 ). (A3) u1 ∂σ ∂σ ∂σ ∂σ where the arguments of u1 .

and Risk Premia Diﬀerentiating the household’s Euler equation (15) and evaluating at steady state yields: ∗ ∗ u11 (dc∗ − Et dc∗ ) + u12 (dlt − Et dlt+1 ) = βEt u1 drt+1 . θt+1 ) t+1 (A7) Recall that (A7) is already evaluated at σ = 0. becomes: (u11 − λu12 )(dc∗ − Et dc∗ ) − t t+1 u1 u12 (dwt − Et dwt+1 ) = βEt u1 drt+1 . . and solving for dc∗ .t+k . the Stochastic Discount Factor. Et dc∗ t+k = dc∗ t u1 u12 βu1 − (dwt − Et dwt+k ) − Et drt+i . we get the more general expression: dmt+1 = βr αu1 1 u11 − λu12 − u1 1 + wλ u ∞ ∞ dat+1 + Et+1 k=1 1 (l dwt+k + ddt+k + adrt+k ) (1 + r)k (A15) − βr Et+1 k=1 1 (1 + r)k λ dwt+k − d log Rt+1. 1 + wλ . for each k = 1. θt+1 ) t+1 . . is: − Et V11 (a∗ . Combining (35). set at+1 = a and θt+1 = θ to get: − V11 (a. (36).. . 2. θ) . applying (36). θ) Derivation of Risk Aversion. but dmt+1 has extra terms related to dVt+1 . t t+1 (A8) (A9) which.t+k . For generalized recursive preferences. u22 + wu12 k (A10) Note that (A10) implies. equations (A9)–(A13) still hold. Section 4. and evaluating at steady state yields: ∞ Et k=0 1 ∗ (dc∗ − wdlt+k − ldwt+k − ddt+k − adrt+k ) = (1 + r) dat . diﬀerentiating. In this case.1). Et V1 (a∗ . t+k k (1 + r) (A12) Substituting (36) and (A11) into (A12). 2 u11 − λu12 u11 u22 − u12 i=1 (A11) Combining (2)–(3). (A13) 1 + wλ where Rt. 2dμ/dσ 2 .t+k . so to evaluate it at the nonstochastic steady state. and (A13) gives: ∞ k=1 dmt+1 = βr u11 − λu12 1 dat+1 + Et+1 u1 1 + wλ ∞ 1 (l dwt+k + ddt+k + adrt+k ) (1 + r)k (A14) − βr Et+1 k=1 1 (1 + r)k λ dwt+k − d log Rt+1.34 Equating (A1) to (A6). 1 + wλ as in the main text. V1 (a.t+k ≡ + rt+i ). yields: t dc∗ = t + r 1 (1 + r)dat + Et 1 + r 1 + wλ ∞ k=0 1 (l dwt+k + ddt+k + adrt+k ) (1 + r)k ∞ r −u1 u1 u12 1 dwt + Et 2 1 + r u11 − λu12 (1 + r)k u11 u22 − u12 k=0 k i=1 (1 λ dwt+k − d log Rt. Equation (A14) also holds for the case of external habits (cf. the Arrow-Pratt coeﬃcient of absolute risk aversion.

a∗ ) are slightly more complicated: t t+1 ∗ u1 (c∗ . However. θ) V (a. the ﬁrst-order t+1 cost of the gamble is zero. θt+1 )1−α ) t t+1 ∗ u2 (c∗ . to evaluate it at the nonstochastic steady state. Et V −α V1 (A22) Since (A22) is already evaluated at σ = 0. To second order. the ﬁrst-order eﬀect of the gamble on household welfare is: ∂c∗ ∂c∗ ∂l∗ ∂l∗ α/(1−α) Et V −α V1 · (wt + u2 + β (Et V 1−α ) − + εt+1 ) dσ. the change in welfare for the household with generalized recursive preferences is: dμ α/(1−α) = −β (Et V (a∗ . (A23) V1 (a. t+1 t+1 Et V (a∗ . and V1 to simplify notation. the household’s optimality ∗ conditions for c∗ and lt (and. θt+1 ) t t+1 α/(1−α) Et V (a∗ . the terms involving ∂ 2 c∗ /∂σ 2 and ∂ 2 l∗ /∂σ 2 ∗ cancel due to the optimality of c∗ and lt . θ) +α . implicitly. (A20) simpliﬁes to: dσ 2 . θt+1 ). t+1 t+1 ∗ −wt u1 (c∗ .35 Proof of Proposition 6 For generalized recursive preferences. θt+1 = θ to get: V1 (a. θt+1 )−α V1 (a∗ . ∂σ 2 ∂σ 2 ⎭ 2 α/(1−α) Et V −α V11 · wt (A20) The derivatives ∂c∗ /∂σ and ∂l∗ /∂σ vanish at σ = 0. u2 . θt ) t+1 t+1 t+1 1 + rt where the right-hand side of (A18) follows from the envelope theorem. lt ). the hypothetical one-shot gamble and one-time fee faced by the household are the same as for the case of expected utility. evaluating the latter at σ = 0. V . Thus. θ) . and Et V −α V1 εt+1 = 0 t because εt+1 is independent of θt+1 and a∗ . optimality ∗ of c∗ and lt implies that the terms involving ∂c∗ /∂σ and ∂l∗ /∂σ cancel. θt+1 )−α V1 (a∗ . lt ) t (A16) (A17) 1−α α/(1−α) ) Note that (A16) and (A17) are related by the usual = and when α = 0. For an inﬁnitesimal fee dμ in (7). θt+1 ) dμ . 2 Equating (A18) to (A21). θt+1 )1−α ) Et V (a∗ . As before. (A18) −V1 (at . the Arrow-Pratt coeﬃcient of absolute risk aversion is: β (Et V 1−α ) α/(1−α) Et V −α V11 − αEt V −α−1 V12 (A21) −Et V −α V11 + αEt V −α−1 V12 . Turning now to the gamble in (6). lt ) = β (Et V (a∗ . set at+1 = a. and εt+1 is independent of θt+1 and a∗ (evaluating the t t+1 latter at σ = 0). θ) −V11 (a. (A16) and (A17) reduce to the standard optimality conditions for expected utility. θt+1 ). t ∗ u2 (c∗ . (A19) ∂σ ∂σ ∂σ ∂σ where we have dropped the arguments of u1 . the eﬀect of the gamble on household welfare is: ⎧ 2 2 ⎨ ∂c∗ ∂l∗ ∂ 2 c∗ ∂ 2 l∗ ∂c∗ ∂l∗ u11 + 2u12 + u1 + u2 + u22 ⎩ ∂σ ∂σ ∂σ ∂σ ∂σ 2 ∂σ 2 u1 + αβ (Et V 1−α ) − αβ (Et V 1−α ) + β (Et V 1−α ) (2α−1)/(1−α) Et V −α V1 · wt ∂c∗ ∂l∗ − + εt+1 ∂σ ∂σ 2 2 α/(1−α) Et V −α−1 V1 · wt ∂c∗ ∂l∗ − + εt+1 ∂σ ∂σ 2 ∂c∗ ∂l∗ − + εt+1 ∂σ ∂σ ⎫ 2 ∗ 2 ∗ ⎬ dσ 2 ∂ c ∂ l α/(1−α) + β (Et V 1−α ) Et V −α V1 · wt − . θt+1 )−α V1 (a∗ . lt ) = −βwt (Et V (a∗ . Thus.

(A30) t wt To solve for V11 . lτ ). diﬀerentiate (A30) with respect to at to yield: V11 (at . lτ ) = wτ [u1 (c∗ − h∗ . θt ) = − u2 (c∗ − ht . ∂at ∂at (A31) ∗ ∗ It remains to solve for ∂c∗ /∂at and ∂lt /∂at . τ τ τ τ τ τ (A34) (A35) = wτ (1 − βbF (1 − βρF ) −1 ) u1 (c∗ τ − ∗ h∗ . (A26) (A27) u2 = −βwt (Et V 1−α ) α/(1−α) where we drop the arguments of u and V to reduce notation. τ . lt ) + β Et V (a∗ . where now F denotes the “generalized recursive” forward operator. Note ﬁrst that (A29) can be used to solve for V2 in terms of current and future marginal utility: V2 = −(1 − ρβF )−1 u1 . ht . (A25) with |ρ| < 1. We henceforth let a time subscript τ ≥ t denote a generic future date and reserve the subscript t to denote the date of the current period—the period in which the household faces the hypothetical one-shot gamble. (A24) and a longer-memory speciﬁcation for habits: ht = ρht−1 + bct−1 . As in the main text. The household’s ﬁrst-order conditions for (A24) with respect to consumption and labor are given by: u1 = β (Et V 1−α ) α/(1−α) Et V −α [V1 − bV2 ]. and we assume ρ + b < 1 in order to ensure h < c. lt ). ht . h∗ +1 . θt ) = − (1 + rt ) wt u12 ∂c∗ ∂l∗ t + u22 t . h∗ . we solve for ∂c∗ /∂at and ∂lτ /∂at t τ for all dates τ ≥ t at the same time. As in the main text. Et V −α (A28) (A29) 1−α α/(1−α) ) V2 . except τ that the expressions are more complicated due to the persistence of habits and the household’s more complicated discounting of future periods. F xτ ≡ (Eτ V 1−α ) α/(1−α) (A32) Eτ V −α xτ +1 . (A33) The household’s intratemporal optimality condition ((A28) combined with (A29)) implies: ∗ ∗ −u2 (c∗ − h∗ . ∗ We solve for ∂lτ /∂at in terms of ∂c∗ /∂at in the same manner as in the main text. that is. ht . We wish to compute V1 and V11 . θt+1 )1−α t t+1 t+1 1/(1−α) . Diﬀerentiating (A24) with respect to its ﬁrst two arguments and applying the envelope theorem yields: V1 = β(1 + rt ) (Et V 1−α ) V2 = −u1 + ρβ (Et V α/(1−α) Et V −α V1 . lτ ) + bβEτ V2 (a∗ +1 . Equations (A26) and (A27) are the same as in the main text except that the discounting of future periods involves the value function V when α = 0. (A27) and (A28) can be used to solve for V1 in terms of current-period utility: (1 + rt ) ∗ V1 (at . θt ) = u(c∗ − ht .36 Derivation of Risk Aversion with Long-Memory Internal Habits and EZ Preferences We consider here the case of generalized recursive preferences: ∗ V (at . θτ +1 )]. Et V −α V1 .

we t must have: ∂c∗ ∂c∗ t+1 = (1 + b) t . ∂at (A45) ∗ which then holds for all τ ≥ t + 1. from (A40). The household’s intertemporal optimality (Euler) condition is given by: 1 + rτ 1 ∗ ∗ u2 (c∗ − h∗ . . ∂at ∂at Et (A46) (A47) . This solves for ∂lt /∂at in terms of (current and future) ∂c∗ /∂at . we can apply (1 − ρL) to both sides of (A43) to get: (1 − F ) [1 − (ρ + b)L] ∂c∗ τ = 0. whatever the initial responses ∂c∗ /∂at and ∂lt /∂at . k = 1. ∂at (A42) Equations (A41) and (A42) hold for all τ ≥ t. lτ ) = βF u2 (c∗ − h∗ . hence we can invert the [1 − β(ρ + b)F ] operator forward to get: ∂c∗ (A43) (1 − F ) [1 − bL(1 − ρL)−1 ] τ = 0. ∂at ∂l∗ (1 − F ) τ = 0.37 Diﬀerentiating (A35) with respect to at and evaluating at steady state yields: −u12 ∂h∗ ∂c∗ τ τ − ∂at ∂at − u22 ∗ ∂h∗ ∂lτ ∂c∗ τ τ = w (1 − βbF (1 − βρF )−1 ) u11 − ∂at ∂at ∂at + u12 ∗ ∂lτ . ∂at (A38) where we’ve used hτ = bL(1 − ρL)−1 cτ and we assume β (ρu22 + (ρ + b)wu12 )/(u22 + wu12 ) < 1 ∗ to ensure convergence. lτ ). ∂at ∂at ∗ ∗ ∂l ∂lt and Et t+k = . Thus. (A44) ∂at Finally. τ ∗ We now turn to solving for ∂cτ /∂at . ∂at ∂at ∂c∗ τ = 0. (A40) simpliﬁes to: [1 − β(ρ + b)F ] (1 − F ) [1 − bL(1 − ρL)−1 ] which. (A39) τ τ τ τ wτ wτ Diﬀerentiating (A39) with respect to at and evaluating at steady state yields: u12 (1 − F ) [1 − bL(1 − ρL)−1 ] ∂l∗ ∂c∗ τ = −u22 (1 − F ) τ . ∂at ∂at ∂c∗ ∂c∗ Et t+k = (1 + b(ρ + b)k−1 ) t . (A36) ∂at where we have used the fact that: ∂xτ ∂ F xτ = F . . ∂at ∂at (A37) ∗ when the derivative is evaluated at steady state. also implies: (A41) [1 − β(ρ + b)F ] (1 − F ) ∗ ∂lτ = 0. 2. ∂at (A40) Using (A38) and noting F L = 1 at steady state. . Solving (A36) for ∂lτ /∂at yields: ∗ u12 + wu11 − β (ρu12 + (ρ + b)wu11 )F ∂lτ = − × ∂at u22 + wu12 1− β(ρu22 + (ρ + b)wu12 ) F u22 + wu12 −1 (1 − bL(1 − ρL)−1 ) ∂c∗ τ .

In order to express (A52) in terms of u1 and u11 instead of u2 and u22 . we use V1 = (1−β(ρ+b))u1 /(β(1− βρ)) and diﬀerentiate the ﬁrst-order condition: ∗ V1 (at .38 Consumption responds gradually to a surprise change in wealth. = βb ∂at 1 + (1 − 1−βρ )wλ (A51) Without habits or labor. Together. ∂at ∂at (A48) where λ ≡ w(1 − β(ρ + b))u11 + (1 − βρ)u12 u1 u12 − u2 u11 = . (A53) Equations (A52) and (A53) have obvious similarities to the corresponding expressions without habits and with expected utility preferences. while labor moves immediately to its new steady-state level. ∗ It remains to solve for ∂ct /∂at . an increase in assets would cause consumption to rise by the amount of the income ﬂow from the change in assets—the ﬁrst term on the right-hand side of (A51). we can now solve (A40) to get: ∗ ∂c∗ ∂lt = −λ t . τ 26 with respect to at to solve for V11 . lτ ). which is accounted for by the denominator of the second term in (A51). The presence of habits attenuates this change by the amount βb/(1 − βρ) in the numerator of the second term. u1 u22 − u2 u12 (1 − βρ)u22 + w(1 − β(ρ + b))u12 (A49) u2 1−βρ where the latter equaltiy follows because w = − u1 1−β(ρ+b) in steady state. The household’s intertemporal budget constraint implies: ∞ Et τ =t (1 + r)−(τ −t) ∗ 1 + r ∂lt ∂c∗ τ = (1 + r) + w . (A48) and (A51) allow us to compute the household’s coeﬃcient of absolute risk aversion (63) in Proposition 7:26 βb (1 − 1−βρ ) r βb r u1 V1 −u11 + λu12 −V11 1− . and the consumption response is further attenuated by the household’s change in hours worked. . + α +α = βb V1 V u1 u 1 − βρ 1 + (1 − 1−βρ )wλ (A52) The consumption-based coeﬃcient of relative risk aversion is given by: βb (1 − 1−βρ ) c −AV11 c u1 AV1 −u11 + λu12 + α +α = βb V1 V u1 u 1 + (1 − 1−βρ )wλ 1− βb 1 − βρ . From (A46). θt ) = (1 + rt ) (1 − βbF (1 − βρF )−1 ) u1 (c∗ − hτ . ht . ∂at r ∂at (A50) Substituting (A46) and (A48) into (A50) and solving for ∂c∗ /∂at yields: t βb (1 − 1−βρ ) r ∂c∗ t .

Robert. Asset Returns. Weil. Bernanke. 149–66.” Econometrica 59 (1991). and Business Cycles I: The Basic Neoclassical Model.” American Economic Review 91 (2001). Rudebusch. and Eric Swanson.” Econometrica 47 (1979). L.” Journal of Political Economy 98 (1990). 1365–82. Carnegie Mellon University. Bryan Routledge. “Aspects of the Theory of Risk-Bearing.” Journal of Political Economy 107 (1999). “Habit Formation: A Resolution of the Equity Premium Puzzle. 361–88. Glenn. 397–415. “Risk-Sensitive Business Cycles. Ralph Koijen. Boldrin. Dale. 401–21. and J. “Examining the Bond Premium Puzzle with a DSGE Model.” Yrj¨ Jahnsson Lectures (1965). in o Essays in the Theory of Risk Bearing.” Journal of Economic Theory 8 (1974). Stiglitz. Mark Gertler. and Jonas Fisher. Arrow. Lawrence Christiano. George. and Eric Swanson. and Simon Gilchrist. 660–7. 727–32. Santos. Van Binsbergen. and the Temporal Behavior of Consumption and Asset Returns: A Theoretical Framework. 111–26. John. Merton. Scheinkman.” Econometrica 37 (1969). Charles Plosser. 195–232. Constantinides. and Jonas Fisher.” Journal of Monetary Economics 24 (1989). Manuel. “On the Diﬀerentiability of the Value Function in Dynamic Models of Economics. Epstein..” Journal of Monetary Economics 21 (1988). Jules. Pratt. Michele. “Habit Persistence and Asset Returns in an Exchange Economy.” Journal of Monetary Economics 45 (2000). and Christopher Pissarides.” Journal of Monetary Economics 55 (2008). “Job Creation and Job Destruction in the Theory of Unemployment. “Working with Epstein-Zin Preferences: Computation and Likelihood Estimation of DSGE Models with Recursive Preferences. Risk Aversion. 507–32.” Federal Reserve Bank of SF Working Paper 2008–31 (2009). Ben. Glenn. “Risk Aversion in the Small and in the Large. 312–32. “Habit Persistence. “Behavior Towards Risk with Many Commodities. 867– 87. Backus. and the Business Cycle.” Review of Economic Studies 61 (1994). Lawrence Christiano. “The Equity Premium Puzzle and the Risk-Free Rate Puzzle. Kenneth (Chicago: Markum. Tallarini. and Juan RubioRamirez. ed. and Sergio Rebelo. 1341–93. and Stanley Zin. Thomas. “An Intertemporal Capital Asset Pricing Model.” Econometrica 57 (1989).” Econometrica 41 (1971). David.” Macroeconomic Dynamics 1 (1997). 205–51. Michele. “Substitution. 1971). Mortensen. Joseph. “The Bond Premium in a DSGE Model with Long-Run Real and Nominal Risks. Rudebusch. John and John Cochrane. . Kenneth. “Asset Prices in Business Cycle Analysis. Kihlstrom. 122–36. Jesus Fernandez-Villaverde. Phillipe.M.” unpublished manuscript (2008). King. “By Force of Habit: A Consumption-Based Explanation of Aggregate Stock Market Behavior. 937–69. Boldrin. Larry and Stanley Zin. “Production. Benveniste. “Risk Aversion with Many Commodities. 519–43. Robert. “The Financial Accelerator in a Quantitative Business Cycle Framework. “Smoothness of the Policy Function in Discrete Time Economic Models.” Handbook of Macroeconomics 1 (1999).” unpublished manuscript (2008).39 References Arrow. Campbell. Growth.A. Tepper School of Business. Richard and Leonard Mirman.” Econometrica 32 (1964).