Monetary Theory University of Bern

1/38

The Rational Expectation Hypothesis

“You can fool some of the people all of the time, and all of the people some of the time, but you cannot fool all of the people all of the time.” (Abraham Lincoln)

2/38

The Rational Expectation Hypothesis

• Robert Lucas (University of Chicago) • Most common interpretation: Individuals do not make systematic errors in forming their expectations; expectations errors are corrected immediately, so that — on average — expectations are correct. • More subtle than that! • At least 3 different deﬁnitions

3/38

The Rational Expectation Hypothesis

Deﬁnition (Broad deﬁnition) Rational expectations are such that individuals formulate their expectations in an optimal way, which is actually comparable to economic optimization. • Meaning: Individuals collect information about the economic environment and use it in an optimal way to specify their expectations. • Unspeciﬁed problems: (i) the cost of collecting information (ii) the deﬁnition of the objective function.

=⇒ despite its general formulation, this deﬁnition remains weakly operative.

4/38

The Rational Expectation Hypothesis

Deﬁnition (mid–deﬁnition) Agents do not waste any available piece of information and use it to make the best possible ﬁt of the real world. • Advantage: Avoids the problem of the cost of collecting information — we only need to know that agents do not waste information • Problem: Does not deal with the objective function =⇒ this deﬁnition remains inoperative.

5/38

The Rational Expectation Hypothesis

Deﬁnition (weak deﬁnition) Agents formulate expectations in such a way that their subjective probability distribution of economic variables (conditional on the available information) coincides with the objective probability distribution of the same variable (the state of Nature) in an equilibrium: xet = E(xt |Ω) where Ω denote the information set

6/38

The Rational Expectation Hypothesis When the model is markovian, Ω essentially consists of (i) Past realizations of the state variables from t=0 on; (ii) The structure of the model and (iii) The probability distributions of the shocks that hit the economy. Implication Expectations should be consistent with the model =⇒ Solving the model amounts to solving for expectations. Notation: Et−i (xt ) = E(xt |Ωt−i ) where Ωt−i = {xk ; k = 0 . . . t − i}. 7/38

Meaning

Outcomes do not differ systematically (i.e., regularly or predictably) from what people expect them to be.

8/38

The Rational Expectation Hypothesis

Proposition (Unbiased Expectations) Rational Expectations are not biased: Let b xt = xt − xte denote the expectation error: Et−1 (b xt ) = 0

9/38

The Rational Expectation Hypothesis

Proposition (Uncorrelated Errors) Expectation errors do not exhibit any serial correlation: Covt−1 (b xt , b xt−1 ) = Et−1 (b xt b xt−1 ) − Et−1 (b xt )Et−1 (b xt−1 ) = 0

10/38

The Rational Expectation Hypothesis

Proposition (Law of Iterated Projection) Let’s consider two information sets Ωt and Ωt−1 , such that Ωt ⊃ Ωt−1 , then E(xt |Ωt−1 ) = E(E(xt |Ωt )|Ωt−1 )

11/38

The Rational Expectation Hypothesis xt

ωt

Px = E(xt | Ωt−1 , ωt ) | {z } Ωt E(Px |Ωt−1 )

Ωt−1

12/38

Example: An AR(2) process

xt = φ1 xt−1 + φ2 xt−2 + εt such that the roots lies outside the unit circle and εt is the innovation of the process. 1. Let’s now specify Ω = {xk ; k = 0, . . . , t − 1}, then E(xt |Ω) = φ1 xt−1 + φ2 xt−2 2. Let’s now specify Ω = {xk ; k = 0, . . . , t − 2}, then E(xt |Ω) = (φ21 + φ2 )xt−2 + φ1 φ2 xt−3

13/38

What do we Really Assume?

• Agents optimize; • They are clever enough to compute conditional expectations (!); • They do not make systematic mistakes (the only way to fool them is to surprise them).

14/38

What we DO NOT Assume!

• They have full information • They perfectly know the model • They have unbounded memory • They observe the world perfectly • They have perfect information about the distribution of shocks

15/38

A prototypical RE model

A First Simple RE Model

Characterize the behavior of an endogenous variable y that obeys the following expectational difference equation yt = aEt yt+1 + bxt where Et yt+1 ≡ E(yt+1 |Ω) where Ω = {yt−i , xt−i , i = 0, . . . , ∞}.

16/38

Example 1: CAPM

Let pt be the price of a stock, dt be the dividend, and r be the rate of return on a riskless asset, we have (for risk neutral agents) Et pt+1 − pt dt + =r p pt {zt } | |{z} Capital gains Dividend gains or equivalently pt = aEt pt+1 + adt where a ≡

1 1 Et yt+1

s

45◦

y′0 y′1 y′2

y⋆

′′ y′′ 2 y1

y′′ 0

yt 19/38

Backward Looking Solutions: |a| > 1 • yt = aEt yt+1 + bxt • deﬁne the expectation error yt+1 = Et (yt+1 ) + ζt+1 with Et ζt+1 = 0 • Rewrite the model as : yt = a(yt+1 − ζt+1 ) + bxt • this can be be restated as

1 b yt − xt + ζt+1 a a Since |a| > 1 this equation is stable and the system is fundamentally backward looking. yt+1 =

• We will not deal with such situations! 20/38

Forward Looking Solution: |a| < 1 Et yt+1

s

45◦

y′2 y′1 y′0

y⋆

′′ ′′ y′′ 0 y1 y2

yt 21/38

Forward Looking Solution: |a| < 1

Several ways of approaching the problem • Forward Substitution • Method of Undetermined Coefﬁcients • Factorization …

22/38

Forward Substitution Iterate forward on the system, making use of the law of iterated projection. yt = aEt yt+1 + bxt yt = aEt (Et+1 (ayt+2 + bxt+1 )) + bxt yt = a2 Et (yt+2 ) + abEt (xt+1 ) + bxt yt = a2 Et (Et+2 (ayt+3 + bxt+2 )) + abEt (xt+1 ) + bxt yt = a3 Et (yt+3 ) + a2 bEt (xt+2 ) + abEt (xt+1 ) + bxt .. . yt = b lim

k−→∞

k ∑ i=0

ai Et (xt+i ) + lim ak+1 Et (yt+k+1 ) k−→∞

23/38

Forward Substitution

• Bounded solution if: 1. the expectation explodes at a rate lower than |1/a| 2. we impose that lim |yt | < ∞ t−→∞

• The solution is given by yt = b

∞ ∑

ai Et (xt+i )

i=0

• yt is given by the discounted sum of all future expected values of xt .

24/38

Method of Undetermined Coefficients

• Make an initial guess on the form of the solution: yt =

∞ ∑

αi Et xt+i

i=0

• Plugging the guess in yt = aEt yt+1 + bxt leads to (∞ ) ∞ ∑ ∑ αi Et xt+i = aEt αi Et+1 xt+1+i + bxt i=0

i=0

25/38

Method of Undetermined Coefficients • Solving the model then amounts to ﬁnd the sequence of αi , i = 0, . . . , ∞ such that the guess satisﬁes the equation. i = 0 α0 = b i = 1 α1 = aα0 .. . • Solution: αi = aαi−1 , with α0 = b. Since |a| < 1, the α sequence converges toward 0 as i tends toward inﬁnity. • The solution writes yt = b

∞ ∑

ai Et xt+i

i=0

26/38

A step toward multivariate Models

A step toward multivariate Models

We are now interested in solving a more complicated problem involving one lag of the endogenous variable: yt = aEt yt+1 + byt−1 + cxt

27/38

Method of Undetermined Coefficients

• An educated guess in this case is given by yt = µyt−1 +

∞ ∑

αi Et xt+i

i=0

• Plug this guess in equation yt = aEt yt+1 + byt−1 + cxt • Identify terms.

28/38

Method of Undetermined Coefficients

• Plug the guess µyt−1 +

∞ ∑

[ αi Et xt+i

=

i=0

aEt µyt +

aµ µyt−1 +

+aEt

] αi Et+1 xt+1+i + byt−1 + cxt

i=0

( =

∞ ∑

[∞ ∑

∞ ∑ i=0

) αi Et xt+i ]

αi Et+1 xt+1+i + byt−1 + cxt

i=0

29/38

Method of Undetermined Coefficients

• Rewrites as µyt−1 +

∞ ∑

αi Et xt+i

=

(aµ2 + b)yt−1 + aµ

i=0

∞ ∑

αi Et xt+i

i=0

+a

∞ ∑

αi Et xt+1+i + cxt

i=0

• Everything is then a matter of identiﬁcation: µ

= aµ2 + b

α0

= aµα0 + c

αi

= aµαi + aαi−1

∀i ⩾ 1

30/38

Method of Undetermined Coefficients • Finding µ amounts to solve the second order polynomial µ2 −

1 b µ+ =0 a a

which admits two solutions such that { µ1 + µ2 = µ1 µ2 = ba

1 a

• Three conﬁgurations may emerge from the above equation 1. A source, (|µi | > 1, i = 1, 2) 2. A sink, (|µi | < 1, i = 1, 2) 3. A saddle path (|µ1 | < 1, |µ2 | > 1).

31/38

Method of Undetermined Coefficients

• Once µ1 known α0 =

c a and αi = αi−1 = 1 − aµ1 1 − aµ1

1 a

1 αi−1 − µ1

• Since µ1 + µ2 = 1/a =⇒ αi = µ−1 2 αi−1 where |µ2 | > 1: this sequence converges toward zero. • Therefore the solution is given by yt = µ1 yt−1 +

∞ ∑ c µ−i 2 Et xt+i 1 − aµ1 i=0

32/38

Rational Expectations: Why so much emphasis on all this?

• The Lucas critique (1976): The parameters of reduced forms are not invariant to economic policy • Old approach to policy analysis • Estimate behavioral equations (consumption, investment, …) • Introduce some changes in the policy • Study the implications of these changes

• Lucas argues this is wrong: With RE the form of expectations and therefore behavior will also change • This should change in turn our policy evaluation. • Critical for policy analysis!

33/38

The Lucas Critique • Consider the following simple AS/AD Model (AS) πt = κyt + βEt πt+1 (AD) yt = Et yt+1 − σ(it − Et πt+1 ) with the policy rule it = γπt + vt where vt = ρvt−1 + εt • Conjectured solution: yt = αy vt and πt = απ vt 34/38

The Lucas Critique

• We have σ(1 − βρ) vt (1 − ρ)(1 − βρ) + σκ(γ − ρ) σκ πt = − vt (1 − ρ)(1 − βρ) + σκ(γ − ρ) yt = −

• Function of γ and ρ! any change in the behavior of the central bank would affect the behavior of output and inﬂation. • Old approach would just estimate yt = βy vt and πt = βπ vt , and make any change to the policy.

35/38

The Lucas Critique: Change in aggressiveness vis–à–vis inflation, γ.

• Incorrect solution: No change in the behavior of yt and πt as it does not affect vt • Correct solution: αy and απ decrease! Output and inﬂation are less sensitive to monetary policy shocks, because agents take into account that the central bank will be more aggressive vis–à–vis inﬂation after a shock. Output

2.5 2.0 1.5 1.0 0.5 0.0

0

5

10 Periods

15

20

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Inflation Rate

0

Initial

5

10 Periods

15

Correct

20

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Nominal Interest Rate

0

5

10 Periods

15

20

Incorrect

36/38

The Lucas Critique: Change in persistence of shock, ρ

• Incorrect solution: No change in the initial behavior of yt and πt , just in persistence. • Correct solution: αy and απ increase! Output and inﬂation are more sensitive to monetary policy shocks, because agents know that inﬂation will persist longer. 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0

Output

0

5

10 Periods

15

20

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Inflation Rate

0

Initial

5

10 Periods

15

Correct

20

0.25 0.20 0.15 0.10 0.05 0.00 0.05 0.10 0.15

Nominal Interest Rate

0

5

10 Periods

15

20

Incorrect

37/38

The Lucas Critique

• Is this so important? • YES! • The second example shows that ignoring this would lead the central bank to underestimate the inﬂationary effects of the new policy • Can be very costly in terms of welfare! • To evaluate this issue a model is needed!

38/38