Monetary Policy Tradeoffs:
Discretion vs. Commitment
Takeki Sunakawa
Advanced Macroeconomics at Tohoku University
Introduction
The efficient allocation is obtained under fully flexible prices. The optimal policy fully stabilizes the price level.
In practice, central banks face short-run tradeoffs: inflation vs. real variables such as output and employment.
We need a monetary policy design in environment in which the central bank faces a nontrivial tradeoff.
The case of an efficient steady state
Consider a situation in which the flexible price equilibrium allocation is inefficient. The natural level of output ytn deviates from its efficient counterpart yet in the short run. On the other hand, yn= yeholds in the long run.
Some real imperfections generates a time-varying gap ut≡ ytn− yet even in the absence of price rigidities.
The CB’s problem
The central bank (CB hereafter) will minimize the welfare loss function:
E0
X∞ t=0
βt π2t+ ϑx2t
,
where xt≡ yt− yteis the welfare-relevant output gap with ytedenoting the efficient level of output. ϑ = κ/ǫ is the weight of output gap fluctuations. Minimization is subject to
πt= βEtπt+1+ κxt+ ut, where ut≡ κ(yet− ytn).
[Note that ˜xt= yt− ynt = (yt− yte)
| {z }
xt
+ (yet− ytn)
| {z }
ut
.]
Cost-push shock
ut follows the exogenous AR(1) process
ut= ρuut−1+ εut,
where ρu∈ [0, 1) and εut is white noise with variance σ2u. ut generates a tradeoff between stabilizing πt and stabilizing xt.
Optimal Discretionary Policy
Each period the CB minimizes the period losses πt2+ ϑx2t, subject to the constraint
πt= κxt+ νt,
where νt≡ βEtπt+1+ utis taken as given by the central bank, as there are no endogenous state variables.
Optimal Discretionary Policy, cont’d
The optimality condition is given by xt= −κ
ϑπt, for t = 0, 1, 2, ....
Substituting it into the NKPC and after some manipulation (e.g., undetermined coefficient method), we have
πt = ϑ
κ2+ ϑ(1 − βρ)ut,
πt = − κ
κ2+ ϑ(1 − βρ)ut.
Optimal Commitment Policy
Now we assume that the CB is able to commit to future policies. The CB will choose a state-contingent sequence {xt, πt}∞t=0so as to minimize
E0
X∞ t=0
βt π2t+ ϑx2t
,
subject to
πt= βEtπt+1+ κxt+ ut, and utfollows the AR(1) process:
ut= ρuut−1+ εut.
Optimal Commitment Policy, cont’d
It is useful to set up the Lagrangian as
L = E0
X∞ t=0
βt 1 2 π
2 t + ϑx
2 t
+ ξt(πt− βEtπt+1− κxt)
,
where {ξt} is a sequence of Lagrange multipliers. The FONCs are
ϑxt− κξt= 0, πt+ ξt− ξt−1= 0,
for t = 0, 1, 2, ...,given ξ−1= 0, because there is no commitment in period 0.
Optimal Commitment Policy, cont’d
Combining the two, we have
x0= −κ ϑπ0, and
xt− xt−1= −κ ϑπt for t = 1, 2, 3, ....
Solving for the policy function under commitment
We assume: xt= axut+ bxxt−1and πt= aπut+ bπxt−1with the initial condition x−1= 0.
Substitute them into the NKPC and the tradeoff equation, aπut+ bπxt−1 = βEt(aπut+1+ bπxt) + κxt+ ut,
= (1 + βρuaπ)ut+ (βbπ+ κ)(axut+ bxxt−1),
= (1 + (βbπ+ κ)ax+ βρuaπ)ut+ (βbπ+ κ)bxxt−1.
aπut+ bπxt−1 = −(ϑ/κ) (xt− xt−1) ,
= −(ϑ/κ) (axut+ (bx− 1)xt−1) .
Solving for the policy function under commitment, cont’d
Then we have
aπ = −(ϑ/κ)ax
bπ = (ϑ/κ)(1 − bx)
aπ = 1 + (βbπ+ κ)ax+ βρuaπ, bπ = (βbπ+ κ)bx
These can be solved for
ax= −(κ/ϑ)/[β(δ+− ρu)], aπ= 1/[β(δ+− ρu)], bx= δ−∈ (0, 1), bπ= (ϑ/κ)(1 − δ−). where δ± =1 ±p1 − 4βγ2/ (2βγ) is the solution of a quadratic equation βγδ2− δ + γ = 0 where γ = (1 + β + κ2/ϑ)−1.
Discretion vs. Commitment: ρ
u= 0.0 (when α = 0)
0 5 10
-2 -1.5 -1 -0.5
0 Output gap
0 5 10
-0.2 -0.1 0 0.1
0.2 Inflation
0.1 0.15
0.2 Price level
0.5 1
1.5 u
Discretion vs. Commitment: ρ
u= 0.8 (when α = 0)
0 5 10
-2 -1.5 -1 -0.5
0 Output gap
0 5 10
-0.1 0 0.1 0.2
0.3 Inflation
0.2 0.4 0.6 0.8
1 Price level
0.5 1
1.5 u
The case of a distorted steady state
Consider the case in which there is a permanent gap x ≡ yn− ye. Specifically,
−Un Uc
= (1 − Φ)M P N,
where Φ ≥ 0 measures the wedge between the marginal product of labor and marginal rate of substitution.
For example, monopolistic competition and associated markup is a source of the distortion, Φ ≡ 1 − [(1 − τ )M]−1≥ 0.
Second-order approximation to the household utility
Note that −Un/Uc= (1 − Φ)M P N = (1 − Φ)C/N implies (1 − Φ)UcC = −UnN . Then we have
Ut− U UcC ≃
ˆ
yt(1 + zt) +1 − σ 2 yˆ
2 t
−(1 − Φ)
ˆ
yt(1 + zt) + dt+1 + ϕ
2 (ˆyt− at)
2
+ t.i.p.,
= Φ
ˆ
yt(1 + zt) + dt+1 + ϕ
2 (ˆyt− at)
2
−
dt+σ + ϕ 2 yˆ
2
t + (σ + ϕ)ˆytat
+ t.i.p.,
= Φˆyt−
dt+σ + ϕ 2 yˆ
2
t + (σ + ϕ)ˆytat
+ t.i.p.,
Under the small distortion assumption, the product of Φ with second-order terms is negligible. Also, Φˆytcan be considered as a second-order term.
The CB’s problem: The case of small SS distortions
Under the small distortion assumption, the welfare loss function is given by
E0
X∞ t=0
βt 1 2 π
2 t + ϑˆx2t
− Λˆxt
,
where Λ ≡ Φλ/ǫ > 0 and ˆxt≡ xt− x with x ≡ yn− ye. Similarly, the NKPC can be written as
πt= βEtπt+1+ κˆxt+ ut.
Optimal Discretionary Policy
Each period the CB minimizes the period losses 1
2 π
2 t + ϑˆx
2 t
− Λˆxt,
subject to the constraint
πt= κˆxt+ νt,
where νt≡ βEtπt+1+ utis taken as given by the central bank.
Optimal Discretionary Policy, cont’d
The optimality condition is given by ˆ xt=Λ
ϑ − κ ϑπt, for t = 0, 1, 2, ....
Substituting it into the NKPC and we have
πt = Λκ
κ2+ ϑ(1 − β) +
ϑ
κ2+ ϑ(1 − βρ)ut, ˆ
xt = Λ(1 − β) κ2+ ϑ(1 − β) −
κ
κ2+ ϑ(1 − βρ)ut. The constant terms are known asthe inflation bias.
Optimal Commitment Policy
The Lagrangian is set up as
L = E0
X∞ t=0
βt 1 2 π
2 t+ ϑˆx
2 t
− Λˆxt+ ξt(πt− βEtπt+1− κˆxt)
,
where {ξt} is a sequence of Lagrange multipliers. The FONCs are
ϑxt− κξt− Λ = 0, πt+ ξt− ξt−1= 0, for t = 0, 1, 2, ... and ξ−1= 0.
Optimal Commitment Policy, cont’d
Combining the two, we have
ˆ x0= −κ
ϑπ0+ Λ ϑ, and
ˆ
xt− ˆxt−1= −κ ϑπt for t = 1, 2, 3, ....
Note that the equilibrium conditions has the same form in the case of the efficient steady state except for t = 0.
ˆ
x−1 = Λ/ϑ is given as the initial condition.
Initial dynamics
0 5 10
0 0.5 1 1.5
2 Output gap
0 5 10
0 0.05 0.1 0.15 0.2
0.25 Inflation
1 2
3 Price level
-0.5 0 0.5
1 u
The case of large SS distortions
When the steady state distortions are large, the naive LQ approximation is no longer correct.
There are two (three?) alternative approaches to overcome the problem.
1 Use a second-order approximation of the structural equations to replace the linear terms in the welfare function to obtain the correct LQ approximation (Benigno and Woodford, 2005).
2 Solve the original nonlinear Ramsey problem for the equilibrium conditions. Then log-linearize the resulting equilibrium conditions (Khan, King and Wolman, 2003).
The zero lower bound on nominal interest rates
Money has no nominal payoffs but otherwise is identical to short-term nominal debt. This fact may impose the zero lower bound (ZLB) on the nominal return of such debt:
it≥ 0. Then the CB minimizes
E0
X∞ t=0
βt π2t+ ϑx2t
,
subject to
πt = βEtπt+1+ κxt,
xt ≤ xt+1+ σ−1(πt+1+ rtn) .
The last equation is anoccasionally binding constraint. The equality holds when it= 0.
The natural rate and the ZLB
We assume the natural rate rnt follows an exogenous deterministic path: In the steady state, rtn= ρ. Itunexpectedlydrops to and remains at
rnt = −ǫ < 0 for t = 0, 1, ..., tZ. From period tZ+ 1 onward, it reverts to rnt = ρ again.
We assumethe perfect foresight(get rid of expectational operators
hereafter). Agents know the subsequent path of the natural rate in period 0.
Whenever rnt < 0, the efficient allocation, implied by it= rnt, is no longer attainable.
Optimal Discretionary Policy
Each period the CB minimizes the period losses πt2+ ϑx2t, subject to the constraint
πt = κxt+ ν0,t, xt ≤ ν1,t,
where ν0,t≡ βEtπt+1 and ν1,t≡ xt+1+ σ−1(πt+1+ rtn) are taken as given.
Optimal Discretionary Policy, cont’d
The Lagrangian is given by L = 1
2 π
2 t+ ϑx2t
+ ξ1,t(πt− κxt− ν1,t) + ξ2,t(xt− ν2,t) .
The FONCs are
xt+ ξ1,t= 0, ϑxt− κξ1,t+ ξ2,t= 0, and the slackness conditions
ξ2,t≥ 0, it≥ 0, ξ2,tit= 0. for t = 0, 1, 2, ....
The solution
From t = tZ+ 1 onward, it= ρ > 0, ξ2,t= 0, and xt= πt= 0 hold. For t = 0, 1, ..., tZ, it= 0, ξ2,t> 0, and ϑxt= −κπt− ξ2,thold. Then we can solve the two equations backward
πt = βπt+1+ κxt,
xt = xt+1+ σ−1(πt+1+ rtn) , from t = tZ, tZ− 1, ..., 0, given xtZ+1 = πtZ+1 = 0.
Optimal Commitment Policy
The CB will choose a state-contingent sequence {xt, πt}∞t=0so as to minimize
E0
X∞ t=0
βt π2t+ ϑx2t,
subject to
πt = βπt+1+ κxt,
xt ≤ xt+1+ σ−1(πt+1+ rtn) .
Optimal Commitment Policy, cont’d
It is useful to set up the Lagrangian as
L = E0
X∞ t=0
βt 1 2 π
2 t + ϑx
2 t
+ ξ1,t(πt− βπt+1− κxt)
+ξ2,t xt− xt+1− σ−1(πt+1+ rnt). The FONCs are
πt+ ξ1,t− ξ1,t−1− (βσ)−1ξ2,t−1 = 0, ϑxt− κξ1,t+ ξ2,t− β−1ξ2,t−1 = 0, and the slackness conditions
ξ2,t≥ 0, it≥ 0, ξ2,tit= 0.
for t = 0, 1, 2, ..., given the initial conditions ξ1,−1= ξ2,−1= 0.
A conjectured solution
The solution is conjectured and verified as follows;
1 From t = 0 to tC ≥ tZ , it= 0 and ξ2,t> 0.
2 For t = tC+ 1, it= 0, ξ2,t= 0 and ξ2,t> 0.
3 From t = tC+ 2 and onward, it> 0 and ξ2,t= ξ2,t−1 = 0,
t = t
C+ 2 and onward
Note that ξ2,t= ξ2,t−1 = 0, then
πt+ ξ1,t− ξ1,t−1= 0, ϑxt− κξ1,t= 0, πt= βπt+1+ κxt, hold, given the initial condition ξ1,tC+1.
These equations can be solved for
xtC+2+k = −κδ
k+1
ϑ ξ1,tC+1, πtC+2+k = (1 − δ)δkξ1,tC+1. for k = 0, 1, 2, .... [check by yourself]
t = t
C+ 1
Note that itC+1> 0 and ξ2,tC+1= 0, then
πtC+1+ ξ1,tC+1− ξ1,tC − (βσ)−1ξ2,tC = 0, ϑxtC+1− κξ1,tC+1− β−1ξ2,tC = 0,
πtC+1= β (1 − δ)ξ1,tC+1
| {z }
=πtC +2
+κxtC+1,
hold.
By substituting out ξ1,tC+1= [β(1 − δ)]−1(πtC+1− κxtC+1), we have [1 + β(1 − δ)]πtC+1− κxtC+1− β(1 − δ)ξ1,tC− [(1 − δ)/σ]ξ2,tC = 0, [β(1 − δ)ϑ + κ2]xtC+1− κπtC+1− (1 − δ)ξ2,tC = 0.
t = 0, 1, ..., t
Cand so on
Note that it= 0, then
πt = βπt+1+ κxt,
xt = xt+1+ σ−1(πt+1+ rtn) , hold, where
rnt =
(−ρ, when t = 0, 1, ..., tZ, ρ, when t = tZ+ 1, ..., tC. In addition, the original FONCs
πt+ ξ1,t− ξ1,t−1− (βσ)−1ξ2,t−1 = 0, ϑxt− κξ1,t+ ξ2,t− β−1ξ2,t−1= 0,
hold for t = 0, ..., tC+ 1, given the initial conditions ξ1,−1= ξ2,−1= 0.
The algorithm for the solution
The algorithm is as follows:
1 Fix tC≥ tZ. Solve 4 × (tC+ 2) equations for 4 × (tC+ 2) variables {xt, πt, ξ1,tξ2,t}tt=0C+1.
2 Obtain xt
C+2+k = −κδk+1ϑ ξ1,tC+1 and πtC+2+k= (1 − δ)δkξ1,tC+1 for k = 0, 1, ..., given ξ1,tC+1 at hand.
3 Check it= rn
t + πt+1+ σ(xt+1− xt) = 0 for t = 0, 1, ...tC and it> 0 t = tC+ 1, .... If not, increase tC and do 1-3 again.
The relevant equations
xt
πt
=
1 σ1 κ β +κσ
| {z }
A
xt+1
πt+1
+
1 σκ σ
rnt,
for t = 0, 1, ..., tC,
xtC+1
πtC+1
=
−κ 1 + β(1 − δ)
β(1 − δ) +κϑ2 −κϑ
−1
β(1 − δ) 1−δσ 0 1−δϑ
| {z }
M
×
ξ1,tC
ξ2,tC
,
ξ1,t
ξ2,t
=
1 (βσ)−1 κ β−1(1 + κσ−1)
| {z }
H
ξ1,t−1
ξ2,t−1
−
0 1 ϑ κ
| {z }
J
xt
πt
,
The relevant equations, cont’d
These equations can be stacked into
I2 −A
J I2
I2 −A
−H J I2
. .. . ..
−M I2
−H J I2
x0
π0
ξ1,0
ξ2,0
x1
π1
ξ1,1
ξ2,1
... xtC+1
πtC+1
ξ1,tC+1
=
rn0 κrσn0
σ
0 0
rn1 κrσn1
σ
0 0 ... 0 0 0
.
Discretion vs. Commitment with the ZLB (when α = 0)
0 5 10
-15 -10 -5 0
5 Output gap
0 5 10
-60 -40 -20 0
20 Inflation
0 2 4
6 Nominal rate
0
5 Natural rate
Caveats
We have assumed the perfect foresight.
Uncertaintyabout the natural rate matters. We need to solve the model with stochastic settings (e.g., Adam and Billi, 2006; 2007, Nakov, 2008).