Risk Aversion in intertemporal contexts Comparative Risk Aversion Application to choice under uncertainty Summary
Comparative Risk Aversion:
A Formal Approach with Applications to
Saving Behaviors
A. Bommier1, A. Chassagnon2 F. Le Grand3
1GREMAQ, Université Toulouse I and TSE,2LEDa-SDFi, Université Paris-Dauphine and PSE3EMLyon Business School
Séminaire "Economie du Risque", 13 octobre 2009
Chaire de la Fondation du Risque, Dauphine-ENSAE-Groupama: «Les Particuliers Face aux Risques»
Risk Aversion in intertemporal contexts Comparative Risk Aversion Application to choice under uncertainty Summary
Outline of the talk
1 Risk Aversion in intertemporal contexts
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
2 Comparative Risk Aversion
A simple model
Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
3 Application to choice under uncertainty
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Robustness of the impact of risk aversion
The idea is that conclusions about the impact of risk aversion radically depend on the setup which is chosen. This is
particularly true, for example, when discussing the relation between risk aversion and precautionary savings.
In a simple two-period model,
working with Kihlstrom and Mirman (1974) ’s preferences or with Quiggin (1982)’s preferences leads to conclude that precautionary savings increase with risk aversion (also, Drèze and Modigliani (1972), Yaari (1987), Bleichrodt and Eeckhoudt, (2005)).
On the contrary the relation is ambiguous when using Epstein and Zin’s (1989) preferences (also Kimball and Weil (2009))
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Risk aversion Vs other indicators
Sometime, ambiguous results have been interpreted as the indication that other indicators matters. In Kimball (1990) precautionary saving relates to a well identified aspect of preferences other than risk aversion. . . and gave birth to other notions like prudence
Gollier (2001, p236):
Definition (Kimball, 1990): An agent is prudent if adding an insurable zero mean risk to his future wealth raises his optimal saving
Gollier (2001, p237)
Local prudence and local risk aversion are independent concepts in the sense that, at least locally, one could be risk averse and prudent, risk averse and imprudent, risk lover and prudent, or, risk-lover and imprudent.
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Risk aversion in multidimensional models
Risk aversion definition is not immediate when uncertainty is over multi-dimensional profiles. In particular, hard to separate “risk aversion” from the other dimensions.
We restrict to preferences over “certain × uncertain pairs (c1, ˜c2).
Kihlstrom and Mirman(1974) convincingly explain that comparing agents’ risk aversions is possible if and only if agents have identical preferences over certain prospects. We shall therefore focus on utility classes that allows us to consider different attitudes towards risk, while keeping preferences over certain consumption paths unchanged.
This means that risk aversion is built upon an ordinal utility function U(C1, C2).
This rules out the standard class of expected utility models assuming additively separable utility functions. Indeed, once additive separability is assumed, it is impossible to change risk preferences, without altering ordinal preferences.
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Some popular classes of preferences
A common approach to discuss the role of risk aversion involves considering particular class of preferences.
Most of the classes are built assuming additively separable utility functions over certain prospects: U(C1, C2) = U1(C1) + U2(C2). A utility function U(c1, ˜c2)is called:
a Kihlstrom and Mirman utility function if there exist increasing u1, u2, k:
UKMk (c1, ˜c2) = k−1(E [k (u1(c1) + u2(˜c2))])
a Selden utility function if there exist increasing u1, u2, v:
UvS(c1, ˜c2) = u1(c1) + u2
“
v−1(E[v(˜c2)])
”
a Quiggin utility function if there exist increasing u1, u2, and φ : [0, 1] → [0, 1]:
UφQ(c1, ˜c2) = u1(c1) + Eφ[u2(˜c2)]
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Epstein and Zin classes of preferences
A very popular specification is obtained when choosing
isoelastic utility function in Selden representation. Indeed, one obtains a class of utility function assuming assuming constant intertemporal elasticity of substitution and homothetic
preferences. Such utility functions are often called “Epstein and Zin utility functions”, though it is imperfectly indicative of what can be found in Epstein and Zin (1989) who considered preferences over infinitely long consumption paths.
A utility function U(c1, ˜c2)is called an Epstein and Zin utility function if there
exist positive scalars ρ 6= 1 and γ 6= 1 such that:
UγEZ(c1, ˜c2) = c1−ρ1 1 − ρ+ 1 1 − ρ “ E h ˜ c1−γ2 i”1−ρ1−γ
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Lower certainty equivalents representation of RA
One way to quantify and to compare risk aversion is to focus on how agents compare lotteries to certain outcomes.
Preceding class preferences seem well ranked in terms of risk aversion in the sense that “more risk averse agents” always need less certainty equivalents. Formally, if one compare U and UA(UAbeing more risk averse):
U(c1, ˜c2) = U(C1, C2) ⇒ UA(c1, ˜c2) < UA(C1, C2).
Following this criteria, agents are more risk averse in Kihlstrom and Mirman, for k2more concave that k1
in Selden utility function, for v2more concave that v1
in Epstein and Zin, for γ2> γ1,
Risk Aversion in intertemporal contexts
Comparative Risk Aversion Application to choice under uncertainty Summary
Are there good reasons to focus on risk aversion ? Disentangling risk aversion and intertemporal substitution Comparative statics: certainty equivalents Vs dispersion criteria
Aversion for dispersion and RA
Risk aversion represents the aversion for increases in risk. Naturally then, another way to compare risk aversion is to focus on how agents accepts risky profiles.
Preceding class preferences are not necessarily well ranked in terms of risk acceptance in the sense that “more risk averse agents” always accept “riskier profiles”. Formally, if one compare U and UA(UAbeing more risk averse), for
any profile (c1, ˜c2)that is “riskier” than the profile (C1, ˜C2):
U(c1, ˜c2) = U(C1, ˜C2) ⇒ UA(c1, ˜c2) < UA(C1, ˜C2).
Following this criteria, when considering spreads in Head and Tails, agents are more risk averse, in Kihlstrom and Mirman, for k2more concave that k1
and in Quiggin, for φ2more convex that φ1
but,
there is no simple definition of “riskier profile” that could be render compatible a ranking of Epstein and Zin preferences following that criteria. a γ2agent
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
General setup
X is an abstract set of consequences, without any
particular structure (not necessarily convex, etc.)
L(X)the set of lotteries with outcomes in X.
Economy populated by two agents A and B with
preferences A and Bover L(X).
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
Acceptance sets and comparative risk aversion
Consider two rational agents Aand B .
Standard approach (initiated by Yaari, 1969):
1 Take as given (for all agents) a partial universal order
`"riskier than" on M(X).
2 For all l ∈ M(X) define agent i0s acceptance set of l by:
Ωi(l) = {m ∈ M(X)|m i land m `i l}
3 Comparative risk aversion (definition)
Ais more risk averse than B if and only if
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
Meaning of "Riskier than"
1 Yaari’s minimalist definition (1969): it is assumed that
l`i
mif and only if m is degenerate (m = δxfor some x)
2 This definition is the one used by Epstein and Zin, to argue the
preferences they consider are ordered in terms of risk aversion.
3 In this paper a less partial order (inspired from Jewitt, 1987).
l ` i
mif and only if there exist an x0∈ X such that
For all x idx0 we have Pr(l idx) ≥ Pr(m
i d x) For all x idx0 we have Pr(l idx) ≤ Pr(m id x)
4 The more restrictive `, the stronger the corresponding
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
Heads or tails gambles: Definition
First a focus on extremely simple random elements =Heads or
tails gambles:
(xl; xh)is the lottery that gives xl∈ X with probability 0.5
and xh ∈ X with probability 0.5. We suppose xh xl.
H(X) =set of such gambles.
Assume that H(X) is endowed with a relation of preferences .
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
A marginal increase in risk
Definition (Gamble spread)
The gamble (xl; xh)is a spread of (yl; yh), which is noted
(xl; xh) ` (yl; yh), if the following relationship holds:
(xl; xh) ` (yl; yh) ⇐⇒ xh yh yl xl
Example, x is a spread of y (x ` y):
x 100 0 y 19 18
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
1 u F(u) xl yl yh xh 1 2
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
Spread compatibility enough to define a unique
ordering in terms of risk aversion
A partial order “riskier than” R is spread compatible, iff: (xl; xh) ` (yl; yh) ⇒ (xl; xh)R(yl; yh)
Moreover, if (xl; xh) ` (yl; yh)and xh yhor yl xlthen it cannot be the case that (yl; yh)R(xl; xh).
Preferences over gambles i fulfills ordinal dominance iff:
(xh yh and xl yl) ⇒ (xl; xh) i (yl; yh)
Result
If preferences of A and B fulfill ordinal dominance and if A is more risk averse than B with respect to a spread compatible R,
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
Preferences over certain × uncertain consumption pairs
1 An agent with utility function UKM
kA is more risk averse than
an agent with utility function UKM
kB with respect to any
spread compatible relation R iff kA is more concave than kB.
2 An agent with utility function UQ
φA is more risk averse than
another one with utility UφQB with respect to any spread
compatible relation R iif φA(1 2) ≤ φ
B(1 2).
Proposition (Epstein and Zin utility functions and risk aversion)
Consider two agents A and B with respective utility functions UγEZ
A and U
EZ
γB, with γA> γB. There does not exist any spread
compatible relation R, such that A is more risk averse than B with respect to R.
Risk Aversion in intertemporal contexts
Comparative Risk Aversion
Application to choice under uncertainty Summary
A simple model Heads or tails gambles
Comparative Risk Aversion for popular classes of preferences
Elements of proof
We build two heads or tails gambles Gaand Gbwith Gabeing a spread of Gb,
such that the agent A is indifferent between both gambles, and the agent B strictly prefers Ga. With 0 < ε << 1, c
1, c2> 0, Gaand Gbare defined as
follows: Ga = (ca; 31−ρ1 (1 − ε)( 1 2), 3 1 1−ρ(1 + ε)( 1 2)) Gb = (cb; 1 − 2ε( 1 2), 1 + 2ε( 12)) where: c1−ρa − c1−ρb = (1 − 2ε) 1− γA + (1 + 2ε)1−γA 2 1−ρ 1−γA − 3 (1 − ε) 1−γA + (1 + ε)1−γA 2 1−γA1−ρ
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
A model free set-up
Agents chose an action t ∈ I ⊂ R modifying the payoff of a gamble Gt ≡ (xl(t), xh(t)) ∈ Z.
Typical example: saving choice. A couple assumptions over lotteries:
non constant: If for s = h, l, xs(t1) ∼ xs(t2)then t1 = t2.
actions do not modify outcome orders: if xh(t) xl(t)for all
t∈ I.
lotteries are single-peaked. For s = h, l:
∃ts∈ I, ∀t ∈ I, xs(ts) xs(t)
t1≤ t2≤ ts≤ t3≤ t4(∈ I),
xs(ts) xs(t2) xs(t1)
xs(ts) xs(t3) xs(t4) Best actions are ordered as states of the world: th≥ tl.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
A model free result
Proposition
We consider two agents A and B choosing an action t providing them a gamble satisfying preceding assumptions. The
preferences of agents A and B are supposed to fulfill the ordinal
dominance and to define respective single optimal actions tA
and tB. Then:
If A is more comparative risk averse than B, then tA ≤ tB.
Comment.General result. Few assumptions. Direct relationship between risk aversion and optimal choice.
Intuition.tl≤ th. A larger action increases the (ex post) welfare in the good
state and diminishes it in the bad one. Decreasing t = redistribution from good states of the world to bad ones to diminish the ex-post utility dispersion.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Idea of the proof
Proof. Suppose that tA> tB.
tl≤ tB< tA ≤ th. Otherwise if (for example) tB < tl< th, then xl(tl) xl(tB)and xh(tl) xh(tB)(single peakedness) with one strict relation. Then from ordinal dominance: B strictly prefers G(tl)to G(tB)
Single peakedness implies xl(tA) xl(tB) xh(tB) xh(tA), with one strict inequality. G(tA)is a spread of G(tB)
⇒ A prefers G(tA)to G(tB), which is a contradiction with A
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Three applications
Three types of uncertainty:
(i) the second period income is random.
(ii) the saving interest rate is uncertain
(iii) the agent faces a mortality risk, i.e. a risk of dying at the end of the first period.
=⇒We use our model free result to draw unambiguous conclusions about the role of risk aversion in these three frameworks.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Precautionary saving
Precautionary saving = saving when the 2nd period income is uncertain. Addressed in many papers: Drèze an Modigliani (1972), Kimball (1990), Kimball and Weil (2009). Notion of prudence whose link with risk aversion could not be clarified.
X is the set of 2 period consumption profiles, endowed with
a preference relation : (c1, c2) ∈ X.
H(X)is the set of lotteries and Y ⊂ H(X) is the set of
gambles with deterministic first consumption: (c1; ch2, c
l 2) ∈ Y.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Precautionary saving
Two assumptions on ordinal preferences :
is convex on X.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Precautionary saving
Proposition (Precautionary saving)
A and B have to chose the first period consumption c1providing
them (c1,ey2+ (1 + R)(y1− c1)), where y1> 0, andey2 = y h 2, y
l 2. fulfills both previous assumptions and that risk preferences verify the ordinal dominance and define respective optimal first
period consumptions cA
1 and c
B
1 for both agents.
agent A is more risk averse than agent B =⇒ cA1 ≤ cB1
Intuition. cl 1≤ c
A
1 ≤ cB1 ≤ ch1(normality of c1). Due to the the single
peakedness, the closer cA
1 to cl1, the larger the welfare when the bad state
realizes and the closer cA
1 to ch1, the larger the welfare when the good state
realizes. Increasing cA
1 increases the welafre in h and decreases it in l ⇒
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Precautionary saving when the interest rate is
uncertain
Same framework as the preceding one, except: y2: certain eR= Rh, Rl: random.
s(y1, y2, R) = arg maxs(y1− s, y2+ (1 + R)s).
Varying R ≡ varying the price of c2(c1=numeraire).
Two effects: (i) substitution effect and (ii) income effect.
⇒ smay increase or decrease as a function of R depending on which effect dominates (equivalent to EIS < or > 1).
Proposition (Optimal saving when interest rate is uncertain)
Under analogous assumptions, the following implication holds: agent A is more risk averse than agent B =⇒
8 < : sA1 ≤ sB1 if R 7→ s(y1, y2, R)is increasing or sB1 ≤ s A 1 if R 7→ s(y1, y2, R)is decreasing
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Optimal saving when the lifetime is uncertain
Impact of risk aversion on savings with lifetime uncertainty addressed without assuming expected utility and allowing for bequest.
Initial endowment W.
May live for one or two periods
ca2= (1 + Ra)(W − c1)is the second period consumption
when surviving and cd
2= (1 + Rd)(W − c1)is the bequest.
We assume that Ra> −1 and Rd ≥ −1 but make no
assumption on how Raand Rdcompare. If no annuity and
no tax on bequest, Ra = Rd, while with perfect annuities,
Ra > Rd= −1.
X = (R+)2× {a, d} and we note consumption profiles
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Optimal saving when the lifetime is uncertain
We make a set of assumptions regarding ordinal preferences:
The agent is always better off when surviving:
∀c1, (c1, (1 + Ra)(W − c1))a (c1, (1 + Rd)(W − c1))d. The saving when surviving is larger than when dying: sa> sd.
Ordinal preferences are convex on both (R+)2× {a} and
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Optimal saving when the lifetime is uncertain
The following proposition holds:
Proposition (Optimal saving with uncertain lifetime)
Agents A and B face an (identical) exogenous risk of dying after the first period. They chose a saving s providing them a
consumption profile (W − s, (1 + Ra)s)aif they survive and a
consumption/bequest profile (W − s, (1 + Rd)s)d when not.
Agents A and B preferences fulfill ordinal dominance and define optimal saving sAand sB. If ordinal preferences fulfill previous set of assumptions then:
agent A is more risk averse than agent B =⇒ sA ≤ sB
Positive correlation between risk aversion and impatience. Strong assumption: identical mortality risk for both agents.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Continuous distributions
Our previous results hold for continuous distributions, as long as we consider a proper definition for “riskier than”.
Previous results rely on an undisputable notion of “riskier than” for gambles.
However to gain generality, one needs to consider a broader class of lotteries.
No consensus regarding the notion of “riskier than” (Chateauneuf, Cohen, and Meilijson (2004)).
⇒ we chose a notion of spread, which generalizes the gamble case.
We suppose that the ordinal preference relationship over X can be represented by a function U : X → R.
The cdf for L ∈ L(X) is noted FLand defined over R.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions
Comparative riskiness
Definition (p−Spread)
Let p ∈]0, 1[. L is said to be a p−spread of L0 (L `pL0), if it exists
u0∈ R such that: 1 FL(u 0) = p 2 for all u ≤ u 0, FL(u) ≥ FL0(u), 3 for all u ≥ u 0, FL(u) ≤ FL0(u). Remarks:
The p−spread is an ordinal concept and not a cardinal one. States of the world can be split into “bad states" (measure 1 − p) and “bad states" (measure p), such that:
better outcomes for both lotteries in good states.
Conditional on being good, L first order dominates L0, and oppositely.
Risk Aversion in intertemporal contexts Comparative Risk Aversion
Application to choice under uncertainty
Summary
Model free set-up Model free result
Three applications to saving under uncertainty Continuous distributions 1 u F(u) L0 L u0 p Figure:p−spread L `pL0
Risk Aversion in intertemporal contexts Comparative Risk Aversion Application to choice under uncertainty
Summary
Conclusions
We develop a notion of risk aversion that reflects the willingness to reduce risks (marginally), and not only the willingness to avoid all sources of risk by choosing a certain outcome.
We focus on heads or tails gambles and prove that KM and Q utility functions are well ordered in terms of risk aversion. But this is not the case for Selden (while well ordered in terms of the willingness to opt for certain outcomes). We develop a model free approach to derive unambiguous results about the impact of risk aversion without assuming a particular form of rationality.
3 applications regarding precautionary saving,
precautionary saving when the interest rate is uncertain and optimal saving when the lifetime is uncertain.