Lecture 2: Central Banks, Monetary Policy and Risk

Harjoat S. Bhamra

Imperial College

2015

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 1 / 65
1 Quantitative Easing

2 Stochastic Control Theory

Hamilton-Jacobi-Bellman Equations and the Stochastic Maximum Principle
Bellman’s Principle of Optimality and the Hamilton-Jacobi-Bellman Equation
Stochastic Maximum Principle

3 Applying the Stochastic Maximum Principle to the CIR Model

Model Outline
Solution

4 Applying HJB Equation to Solve Stochastic Growth Model

Model
Solution

5 Basic New Keynesian Model with Risk

Model
Equilibrium

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 2 / 65
 Quantitative Easing

The Central Bank’s Balance Sheet

Central Banks do not just trade in short-term Government bond markets.

They also trade in longer term Government debt in an attempt to influence
the long end of the yield curve.
Central banks also hold risky assets, such as mortgage-backed securities and
in some cases even equity, which impacts risk premia
The money used to purchase securities can be created electronically by a
Central Bank
Quantitative Easing – when a central bank creates money electronically and uses
it to purchase securities from financial institutions

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 3 / 65
 Quantitative Easing

Central Bank’s Budget Constraint

N
X Mt
HtCB = WB,t
CB
+ CB
Wi,t − (1)
Pt
i=1

N
CB CB
X
CB Mt
Ht+dt = WB,t (1 + rt dt) + Wi,t (1 + dRi,t ) − (1 − πt dt) + (Tt − Gt )dt
Pt
i=1
(2)
N
X Mt
dHtCB = WB,t
CB
rt dt + CB
Wi,t dRi,t + πt dt + (Tt − Gt )dt (3)
Pt
i=1
N
X Mt
dHtCB = HtCB rt dt + CB
Wi,t (dRi,t − rt dt) + (rt + πt )dt + (Tt − Gt )dt (4)
Pt
i=1

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 4 / 65
 Stochastic Control Theory Hamilton-Jacobi-Bellman Equations and the Stochastic Maximum Principle

Typical Stochastic Optimal Control Problem I

t ∈ T = [0, ∞)
We have a 1-d state1 , s, which evolves over time according to the following stochastic law
of motion
ds(t) = f (s(t), c(t))dt + σ(s(t), c(t))dZ (t) (5)

Z (t) is a standard Brownian motion under physical probability measure P

The starting value of the state is given by s(0) = s0 . The future values of the state will
depend on the control variable u, which is also 1-d.
An agent chooses the path of the control, c(t)t∈T . Her objective is to maximize the
discounted value of some flow function. At time-t, the flow function is given by
u(s(t), c(t)) (6)

With a constant discount rate ρ, the agent’s objective is given by

Z ∞
J(s0 ) = sup E0 e −ρt u(s(t), c(t))dt (7)
c(t)t∈T 0

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 5 / 65
 Stochastic Control Theory Hamilton-Jacobi-Bellman Equations and the Stochastic Maximum Principle

Typical Stochastic Optimal Control Problem II

Date-t objective function

Z ∞
J(t) = J(s(t)) = sup Et e −ρ(u−t) u(s(u), c(u))du (8)
c(u)u≥t t

The maximized objective function is called the value function

What path should the agent choose?

1 later on we shall deal with multidimensional states

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 6 / 65
 Stochastic Control Theory Hamilton-Jacobi-Bellman Equations and the Stochastic Maximum Principle

Hamilton-Jacobi-Bellman Equation & Stochastic

Maximum Principle

HJB equation

1 00
0 = sup u(s(t), c(t)) − ρJ(s(t)) + J 0 (s(t))f (s(t), c(t)) + J (s(t))(σ(s(t), c(t)))2 (9)
c(t) 2

Stochastic Maximum Principle

1
H(s(t), c(t), λ(t), φ(t)) = u(s(t), c(t)) + λ(t)f (s(t), c(t)) + φ(t)σ(s(t), c(t))2 (10)
2

Hc (s(t), c(t), λ(t), φ(t)) = 0, given s0 (11)

Hs (s(t), c(t), λ(t), φ(t))dt + dλ(t) − ρλ(t)dt = φ(s(t))σ(s(t), c(t))dZ (t) (12)
−ρ(T −t)
Et [e λT ] = 0, (13)
∂λ(t)
where φ(s(t)) = ∂s(t)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 7 / 65
 Stochastic Control Theory Hamilton-Jacobi-Bellman Equations and the Stochastic Maximum Principle

Nonlinear differential equation from HJB Equation

The value function satisfies the Hamilton-Jacobi-Bellman (HJB) equation

1 00
0 = sup u(s(t), c(t)) − ρJ(s(t)) + J 0 (s(t))f (s(t), c(t)) + J (s(t))(σ(s(t), c(t)))2 (14)
c(t) 2

The FOC condition of the HJB is

uc (s(t), c(t)) = J 0 (s(t))fc (s(t), c(t)) + J 00 (s(t))σ(s(t), c(t))σs (s(t), c(t)) (15)
Solving the above equation gives the date-t value of the optimal control in terms of the
date-t state and the derivative of the value function.
To find the value function, substitute the optimal control, c ∗ (t) into the HJB to get a
second order nonlinear ordinary differential equation (no longer need the sup)

1 00
0 = u(s(t), c ∗ (t)) − ρJ(s(t)) + J 0 (s(t))f (s(t), c ∗ (t)) + J (s(t))(σ(s(t), c ∗ (t)))2 (16)
2
How do we solve the above nonlinear ode?
Does not generally have a closed-form solution – need numerical methods
Do not have boundary conditions – need a new concept of what a solution to a
differential equation is – viscosity solution – shall understand this informally when we
study finite difference methods
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 8 / 65
 Stochastic Control Theory Bellman’s Principle of Optimality and the Hamilton-Jacobi-Bellman Equation

Bellman’s Principle of Optimality

Principle of Optimality: An optimal policy has the property that whatever the
initial state and initial decision are, the remaining decisions must constitute an
optimal policy with regard to the state resulting from the first decision. (See
Bellman, 1957, Chap. III.3.)
Key conceptual difference relative to Pontryagin’s Maximum Principle
Think of the control a function of the state.

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 9 / 65
 Stochastic Control Theory Bellman’s Principle of Optimality and the Hamilton-Jacobi-Bellman Equation

Heuristic derivation of HJB I

Split the integral for the value function between the next small time interval, [t, t + dt) and the
rest of time [t + dt, ∞)
We use the infinitesimal dt instead of the small increment ∆t, so we can use Ito’s Lemma and
our usual results from Stochastic Calculus

Z t+dt
J(s(t)) = sup Et e −ρ(u−t) u(s(u), c(u))du (17)
c(u)u≥t t
 Z ∞ 
+ e −ρdt Et Et+dt e −ρ(u−[t+dt]) u(s(u), c(u))du (18)
t+dt

Apply Bellman’s Principle of Optimality

Suppose we choose an optimal path at date-t: (cut (s))u≥t
Suppose we choose an optimal path at date-τ > t: (cuτ (s))u≥τ
Principle of Optimality ⇒ (cut (s))u≥t = (cut (s))t≤u<τ ∪ (cuτ )u≥τ
If we choose an optimal path for the control at some future date, when considered as functions
of the state, the future optimal path is contained within today’s – time consistency

For our infinite horizon problem, where the law of motion for the state does not depend explicitly on time, we can go even further: thinking of the control
as a map from the state to the reals, it so happens that the map is time invariant, i.e. ∀t ∈ T , (cut (s))u≥t , for each u ≥ t, we have cut (s) = c(s)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 10 / 65
 Stochastic Control Theory Bellman’s Principle of Optimality and the Hamilton-Jacobi-Bellman Equation

Heuristic derivation of HJB II

Exploiting the Principle of Optimality reveals a recursive structure for the value function

 Z t+dt
J(s(t)) = sup Et Et e −ρ(u−t) u(s(u), c(u))du
c(u)u∈[t,t+dt) t


Z ∞ 
+e −ρdt sup Et+dt e −ρ(u−[t+dt]) u(s(u), c(u))du  (19)

c(u)u≥t+dt t+dt 

| {z }
J(s(t+dt))

To derive the HJB equation we just need to indulge in some Stochastic Calculus. First observe
that
Z t+dt
Et e −ρ(u−t) u(s(u), c(u))du = u(s(t), c(t))dt + o(dt) (20)
t
h i
h(t)
h(t) = o(j(t)) ⇐⇒ limt→0 j(t)
=0
The Et is important – it kills off dZ (t) terms!

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 11 / 65
 Stochastic Control Theory Bellman’s Principle of Optimality and the Hamilton-Jacobi-Bellman Equation

Heuristic derivation of HJB III

Hence

J(s(t)) = sup u(s(t), c(t))dt + e −ρdt Et J(s(t + dt)) + o(dt) (21)

c(u)u∈[t,t+dt)

Now

e −ρdt = 1 − ρdt + o(dt), (22)

and so

J(s(t)) = sup u(s(t), c(t))dt + Et J(s(t + dt)) − ρdtEt J(s(t + dt)) + o(dt) (23)
c(u)u∈[t,t+dt)

Furthermore dtEt J(s(t + dt)) = dtJ(s(t)) + o(dt), and so

J(s(t)) = sup u(s(t), c(t))dt + Et J(s(t + dt)) − ρdtJ(s(t)) + o(dt) (24)

c(u)u∈[t,t+dt)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 12 / 65
 Stochastic Control Theory Bellman’s Principle of Optimality and the Hamilton-Jacobi-Bellman Equation

Heuristic derivation of HJB IV

Now
J(s(t + dt)) = J(s(t)) + dJ(s(t)) (25)
1
= J(s(t)) + J (s(t))Et ds(t) + J 00 (s(t))Et [(ds(t))2 ] + o(dt)
0
(26)
2
1
= J(s(t)) + J 0 (s(t))f (s(t), c(t))dt + J 00 (s(t))(σ(s(t), c(t)))2 + o(dt) (27)
2
(28)
and so
 
1
0 = sup u(s(t), c(t)) − ρJ(s(t)) + J 0 (s(t))f (s(t), c(t)) + J 00 (s(t))(σ(s(t), c(t)))2 dt + o(dt)
c(t) 2
(29)
In the continuous time limit, we obtain
1 00
0 = sup u(s(t), c(t)) − ρJ(s(t)) + J 0 (s(t))f (s(t), c(t)) + J (s(t))(σ(s(t), c(t)))2 (30)
c(t) 2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 13 / 65
 Stochastic Control Theory Stochastic Maximum Principle

Stochastic Maximum Principle and Forward-Backward

Sde’s I

For the deterministic case, it was fairly straightforward to see the connection between
Pontrayagin’s Maximum Principle and the Hamilton-Jacobi-Bellman Equation. In the stochastic
case, the mathematics is more challenging, so I give a heuristic overview.

The main idea of interest to us is that the Hamiltonian needs to be adjusted for risk – this is
very intuitive for financial economists. In the deterministic case, the Hamiltonian included the
‘expected return’ given a particular control. In the stochastic case, we need to subtract the
‘variance’.

The Hamiltonian now takes the form

1
H(s(t), c(t), λ(t), φ(t)) = u(s(t), c(t)) + λ(t)f (s(t), c(t)) + φ(t)σ(s(t), c(t))2 (31)
2
As before I make the identification, λ(t) = J 0 (s(t)). I make the additional identification
φ(t) = J 00 (s(t)). For problems with a well defined maximum, J 00 (s(t)) < 0, so the term
1
2
φ(t)σ(s(t), c(t))2 is a downward risk adjustment, in accordance with our portfolio theory
intuition.

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 14 / 65
 Stochastic Control Theory Stochastic Maximum Principle

Stochastic Maximum Principle and Forward-Backward

Sde’s II

Hc (s(t), c(t), λ(t), φ(t)) = 0, given s0 (32)

∂λ(t)
Hs (s(t), c(t), λ(t), φ(t))dt + dλ(t) − ρλ(t)dt = σ(s(t), c(t))dZ (t), Et [e −ρ(T −t) λT ] = 0,
∂s(t)
(33)

Equation (32) is a forward sde, while (33) is a backward sde. If they cannot be uncoupled they
constitute a forward-backward sde (FBSDE) system.

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 15 / 65
 Stochastic Control Theory Stochastic Maximum Principle

Solving the FBSDE system

One way to try and get a solution is to take the conditional expectation of
the backward sde, treat λ as a function of the state and the use Ito’s Lemma
to derive a differential equation for λ.
The differential equation for λ can be used in conjunction with the
Feynman-Kac Theorem to obtain a probabilistic solution for λ
See Ma & Yong (2008) for a thorough treatment of FBSDE’s.

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 16 / 65
 Stochastic Control Theory Stochastic Maximum Principle

Connection between Stochastic Maximum Principle and

the Hamilton-Jacobi-Bellman Equation

Write HJB equation in terms of Hamiltonian

Identify λ(t) = J 0 (s(t)) (useful to observe that λ(t) = λ(s(t)))
Identify φ(t) = J 00 (s(t)) [useful to observe that φ(t) = ∂λ(s(t))/∂s(t)]

0 = sup H(s(t), c(t), J 0 (s(t)), J 00 (s(t))) − ρJ(s(t)) = 0 (34)

c(t)

FOC of HJB gives us one part of Maximum Principle

Hc (s(t), c(t), J 0 (s(t))) = 0 (35)

∂λ(t)
What about Hs (s(t), c(t), λ(t), φ(t))dt − dλ(t) − ρλ(t)dt + ∂s(t)
σ(s(t), c(t))dZ (t) = 0?
We can also derive this from the HJB!

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 17 / 65
 Stochastic Control Theory Stochastic Maximum Principle

Start by noting that at the optimum, where c(t) = c ∗ (s(t)), the HJB becomes the following ode
0 = H(s(t), c ∗ (s(t)), J 0 (s(t)), J 00 (s(t))) − ρJ(s(t)) (36)
Differentiate the ode wrt to s(t) and use the fact that Hc (s(t), c ∗ (s(t)), J 0 (s(t)), J 00 (s(t))) =0

=0
z }| { ∂c ∗ (s(t))
∗ 0 00
0 = Hs (s(t), c (s(t)), J (s(t)), J (s(t))) + Hc (s(t), c ∗ (s(t)), λ(t), φ(t)) (37)
∂s(t)
+ Hλ (s(t), c ∗ (s(t)), J 0 (s(t)), J 00 (s(t))) J 00 (s(t)) (38)
| {z }
=f (s(t),c ∗ (t))

+ Hφ (s(t), c ∗ (s(t)), J 0 (s(t)), J 00 (s(t))) J 000 (s(t)) − ρJ 0 (s(t)) (39)

| {z }
1 σ(s(t),c ∗ (t))2
2

∂λ(t)
Remember the identification λ(t) = J 0 (s(t)), which implies ∂s(t)
= J 00 (s(t))

∂λ(t) 1 ∂ 2 λ(t)
0 = Hs (s(t), c ∗ (s(t)), λ(t)) + f (s(t), c ∗ (t)) + σ(s(t), c ∗ (t))2 − ρλ(t) (40)
∂s(t) 2 ∂s(t)2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 18 / 65
 Stochastic Control Theory Stochastic Maximum Principle

Noting that
∂λ(t) 1 ∂ 2 λ(t)
dλ(t) = ds(t) + (ds(t))2 (41)
∂s(t) 2 ∂s(t)2
1 ∂ 2 λ(t)
 
∂λ(t)
= f (s(t), c ∗ (t)) + 2
σ(s(t), c ∗ (t))2 dt (42)
∂s(t) 2 ∂s(t)
∂λ(t) ∗
+ σ(s(t), c (t))dZ (t), (43)
∂s(t)
we obtain
dλ(t)
0 = Hs (s(t), c ∗ (s(t)), λ(t)) + Et − ρλ(t) (44)
dt
and hence by treating λ(t) as a function of s(t) and using Ito’s Lemma
∂λ(t)
σ(s(t), c ∗ (t))dZ (t) = Hs (s(t), c ∗ (s(t)), λ(t))dt + dλ(t) − ρλ(t)dt (45)
∂s(t)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 19 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Model Outline

Model

Look at a simple version of the Cox-Ingersoll-Ross production economy (see Cox,

Ingersoll & Ross (1985))
Stochastic optimal control problem
Z ∞
C 1−γ
sup Et e −ρ(u−t) u du (46)
(Cu )u≥t t 1−γ
s.t.

dKt = (AKt − Ct )dt + σKt dZt . (47)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 20 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Model Outline

Underlying economics I

Household – standard power utility

Capital accumulation equation

General capital accumulation equation

Kt+dt = (Kt + It dt)e −δdt + σt Kt dZt (48)

Current capital stock is augmented by an investment flow, giving Kt + It dt

There is depreciation at the rate δ
The is a zero-mean additive shock to the capital stock
In the continuous time limit, we obtain

dKt = (It − δKt )dt + σt Kt dZt (49)

Impose market clearing

Yt = Ct + It (50)

The rate of output is sum of the consumption rate and the investment flow

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 21 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Model Outline

Underlying economics II

Obtain capital accumulation equation in terms of capital stock level, consumption

rate and output flow

dKt = (Yt − δKt − Ct )dt + σt Kt dZt (51)

Assume a linear output function, Yt = αKt

dKt = (AKt − Ct )dt + σt Kt dZt , A = α − δ (52)

Assume constant σt , i.e. σt = σ

dKt = (AKt − Ct )dt + σKt dZt (53)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 22 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

Use log variables I

Stochastic optimal control problem

Z ∞
C 1−γ
sup Et e −ρ(u−t) u du (54)
(Cu )u≥t t 1−γ
s.t.

dKt = (AKt − Ct )dt + σKt dZt . (55)

Using log variables

∞
e (1−γ)cu
Z
sup Et e −ρ(u−t) du (56)
(cu )u≥t t 1−γ

dkt = (µ − e ct −kt )dt + σdZt (57)

where µ = A − 21 σ 2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 23 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

HJB Equation and Maximum Principle I

HJB equation

0 = sup H(kt , ct , ∂k Jt , ∂kk Jt ) − ρJt (58)

ct

where
e (1−γ)ct 1
H(kt , ct , Λt , Φt ) = + (µ − e ct −kt )Λt + σ 2 Φt (59)
1−γ 2
FOC for c

Hc = 0 (60)
(1−γ)c c−k
e =e Λ (61)
−γc −k
e =e Λ (62)
k−γc
Λ=e (63)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 24 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

HJB Equation and Maximum Principle II

1 1
e c = e γ k Λ− γ (64)
Differentiate HJB eqn wrt k

0 = H(kt , ct∗ , ∂k Jt , ∂kk Jt ) − ρJt (65)

∂c ∗ ∂Λ ∂Φ
0 = Hc + Hk + HΛ + HΦ − ρΛ (66)
∂k ∂k ∂k
∂Λ ∂Φ
0 = Hk + HΛ + HΦ − ρΛ (67)
∂k ∂k
dΛ
0 = H k + Et − ρΛ (68)
dt
dΛ = (ρΛ − Hk )dt + ΦσdZ (69)
(70)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 25 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

HJB Equation and Maximum Principle III

∂Λ
where Φ = ∂k . Last step – used Ito’s Lemma
 
dΛ = ρ − e −(1− γ )k Λ− γ Λdt + ΦσdZ
1 1
(71)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 26 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

Deriving FBSDE I
forward-backward sde (FBSDE) system
forward sde
   1

1 k
− 1− γ t
−γ
dkt = µ − e Λt dt + σdZt , given k0 (72)

backward sde

 1

∂Λt
 
1 k
− 1− γ −γ
dΛt = ρ−e t
Λt Λt dt +Φt σdZ , Φt = , lim e −ρT E0 [ΛT ] = 0
∂kt T →∞
(73)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 27 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

Deriving FBSDE II

Change of variables: λ = ln Λ
forward sde
 
−1
 
1 k
− 1− γ
dkt = µ−e t
Λt γ dt + σdZt , given k0 (74)

backward sde

 2 !
1 ∂λt ∂λt
 
1 k
− 1− γ 1λ
2 t −γ t
dλt = ρ− σ −e e dt + σdZ
2 ∂kt ∂kt
(75)
−ρT λT
lim e E0 [e ]=0 (76)
T →∞

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 28 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

Solving FBSDE system I

FBSDE
forward sde
   
1 k
− 1− γ 1λ
t −γ t
dkt = µ−e e dt + σdZt , given k0 (77)

backward sde

 2 !
1 ∂λt ∂λt
 
1 k
− 1− γ 1λ
−γ
dλt = ρ− 2
σ −e t
e t
dt+ σdZ , lim e −ρT E0 [e λT ]
2 ∂kt ∂kt T →∞

(78)
Ansatz: λt = a0 + a1 kt

 
dkt = µ − e −(1− γ )kt e − γ (a0 +a1 kt ) dt + σdZt
1 1
(79)
 
1 2 2 −(1− γ1 )kt − γ1 (a0 +a1 kt )
dλt = ρ − a1 σ − e e dt + a1 σdZ (80)
2
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 29 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

Solving FBSDE system II

Since dλ = a1 dk, we have

1
ρ − a12 σ 2 − a1 µ = (1 − a1 )e −(1− γ )k− γ a1 k e − γ a0
1 1 1
(81)
2
Two possibilities
RHS is function of k ⇒ no solution for most parameter values
RHS is constant ⇒ a1 = 1 − γ &
  
−1a 1 1 1
e γ 0 = ρ+ 1− A − γσ 2 (82)
γ γ 2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 30 / 65
 Applying the Stochastic Maximum Principle to the CIR Model Solution

Solving FBSDE system III

Solution

   
1 1 1
λt = −γ ln ρ+ 1− A − γσ 2 + (1 − γ)kt (83)
γ γ 2
   −γ
1 1 1 2
Λt = ρ+ 1− A − γσ Kt1−γ (84)
γ γ 2

and
   
dKt 1 1 1 2
= (A − ρ) + 1 − γσ dt + σdZt , given K0 (85)
Kt γ γ 2
2
K =K e γ [ 1
(A−ρ)+ (1− )
1 1
γ 2 γσ ]t e − 21 σ2 t+σZt (86)
t 0

Optimal consumption rate

   
1 1 1 2
Ct = ρ+ 1− A − γσ Kt (87)
γ γ 2
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 31 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Model

Stochastic Growth Model

Z ∞ 
Jt = sup Et f (Cu , Ju )du (88)
(Cu )u≥t t

where
 
c
f (c, v ) = ρ(h−1 (v ))1−γ u (89)
h−1 (v )
x 1−γ
h(x) = (90)
1−γ
1
1− ψ
x −1
u(x) = 1
(91)
1− ψ

s.t.
dKt = (It − δKt )dt + σKt dZt , (92)
where
It + C t = Y t (93)
and
Yt = AKtα (94)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 32 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

HJB Equation
HJB equation
1 2 2 00
sup f (Ct , Jt ) + J 0 (Kt )(AKtα − δKt − Ct ) + σ Kt J (Kt ) (95)
Ct 2
FOC (drop excessive notation, such as t-subscript and K -argument)
 
1
− γ− ψ 1
−ψ
ρ(h−1 (J)) C = J0 (96)
Make a transformation
e (1−γ)g (k)
J= (97)
1−γ
k = ln K (98)
Note that while J 0 = ∂K J, g 0 = ∂k g , so we obtain
J 0 = e (1−γ)g −k g 0 (99)
00 (1−γ)g −2k 00 0 2 0
J =e (g − (γ − 1)(g ) − g ) (100)
 
C
f (C , J) = ρe (1−γ)g u (101)
eg
The HJB reduces to the following nonlinear second order ode

  1
0 = ρu C ∗ e −g + Ae −(1−α)k − δ − C ∗ e −k g 0 (k) + σ 2 (g 00 − (γ − 1)(g 0 )2 − g 0 ), (102)
2
where C ∗ denotes the optimal consumption rate. From the FOC, we can see that
C ∗ = [ρe k (g 0 )−1 ]ψ e (1−ψ)g (103)
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 33 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

We want to isolate the nonlinear terms

From the FOC
C ∗ e −g = [ρe k e −g (g 0 )−1 ]ψ , (104)
Hence
1
1− ψ
(C ∗ e −g ) = [ρe k e −g (g 0 )−1 ]ψ−1 (105)
and so
ρψ [e k e −g (g 0 )−1 ]ψ−1 − ρ
ρu(C ∗ e −g ) − C ∗ e −k g = 1
− ρψ [e k e −g (g 0 )−1 ]ψ−1 (106)
1− ψ

ρψ [e k e −g (g 0 )−1 ]ψ−1 ρψ
= − (107)
ψ−1 ψ−1
C ∗ e −g g 0 ρψ
= − (108)
ψ−1 ψ−1
The ode can thus be rewritten as
      
1 1
0 = βψ − (ψ − 1) Ae −(1−α)k − δ + σ 2 + C ∗ e −g ∂k − (ψ − 1)σ 2 ∂kk g (109)
2 2
1 2 0 2
+ σ (γ − 1)(ψ − 1)(g ) (110)
2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 34 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

Approximate Loglinear Solution I

We now linearize the ode around the deterministic steady-state (dss).

k̂ = k − k|dss (111)
Observe that
e −(1−α)k = e −(1−α)(kdss +k̂) = e −(1−α)kdss [1 − (1 − α)k̂] + O(k̂ 2 ) (112)
and
∗
C ∗ e −g = e c −g
= e cdss −gdss + [c ∗ − g − (cdss − gdss )]e cdss −gdss + O([c ∗ − g − (cdss − gdss )]2 )
(113)
= [1 − (cdss − gdss )]e cdss −gdss + (c ∗ − g )e cdss −gdss + O([c ∗ − g − (cdss − gdss )]2 ) (114)
From
C ∗ e −g = [ρe k e −g (g 0 )−1 ]ψ , (115)
we have
c ∗ − g = ψ[ln ρ + k − g − ln ∂k̂ g ] (116)
and so
C ∗ e −g = [1 − (cdss − gdss )]e cdss −gdss + ψ[ln ρ + k − g − ln ∂k̂ g ]e cdss −gdss (117)
∗ 2
+ O([c − g − (cdss − gdss )] ) (118)
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 35 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

Approximate Loglinear Solution II

Now suppose that

g = gdss + a k̂ + O(k̂ 2 ) (119)
I can therefore reduce the nonlinear ode for g into an equation which is linear in k̂. By
comparing coefficients, I can obtain two nonlinear simultaneous equations in gdss and a. This is
best done in mathematica!

     
1
0= βψ − (ψ − 1) Ae −(1−α)kdss (1 − (1 − α)k̂) − δ + σ 2 + C ∗ e −g ∂k (120)
2

1
− (ψ − 1)σ 2 ∂kk g (121)
2
1
+ σ 2 (γ − 1)(ψ − 1)(g 0 )2 (122)
2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 36 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

Numerical Solution I

For the numerical solution, I shall solve for J, but use k = ln K as the underlying state variable.
(this turns out to be better than (g , k) and (J, K ))

∂K J = e −k ∂k J (123)
∂KK J = −e −2k ∂k J + e −2k ∂kk J (124)
The HJB equation thus becomes
   
1 1
sup f (C , J) + Ae −(1−α)k − δ + σ 2 − Ce −k ∂k J + σ 2 ∂kk J (125)
C 2 2
The FOC is
 
1
− γ− ψ 1
−ψ
ρ(h−1 (J)) C = e −k ∂k J (126)
Hence
 
1
−ψ γ− ψ
ρψ (h−1 (J)) C −1 = e −ψk (∂k J)ψ (127)
 
1
ψ γ− ψ
−1 −ψ −ψk ψ −1
C =ρ e (∂k J) (h (J)) (128)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 37 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

Numerical Solution II

1
1− ψ 1 −1
ρ[h−1 (J)]1−γ [C h−1 (J)] ψ − 1]
f (C , J) = 1
(129)
1− ψ
1 )
−(γ− ψ 1
1− ψ
ρ[h−1 (J)] C − ρ[h−1 (J)]1−γ
= 1
(130)
1− ψ

Ce −k ∂k J − ρ(1 − γ)J
= 1
(131)
1− ψ

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 38 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

Want to solve numerically

Ce −k ∂k J − ρ(1 − γ)J
   
1 1
1
+ Ae −(1−α)k − δ + σ 2 − Ce −k ∂k J + σ 2 ∂kk J (132)
1− ψ
2 2

We shall use a finite difference method.

Approximate the function J at N + 1 equispaced points k1 , . . . , kN+1 , where
kn+1 = kn + ∆k.
J(n) = J(kn ) (133)
We are approximating the continuous time stochastic process k, where the underlying
state space is continuous (it’s actually R) via a continuous time Markov chain, where the
underlying state space is discrete and bounded (it’s {k1 , . . . , kN+1 }).
We approximate the first derivative in a very special way, called an upwind approximation
common in fluid dynamics. Below the steady state, when Et [dkt ] > 0, k is increasing with
time if there are no shocks, so we use a forward first difference to approximate ∂k J
J(n + 1) − J(n)
∂kF J(kn ) = (134)
∆k
Above the steady state, when Et [dkt ] < 0, k is decreasing with time if there are no shocks,
so we use a backward first difference to approximate ∂k J
J(n) − J(n − 1)
∂kB J(kn ) = (135)
∆k
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 39 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

The optimal control C is a function of the first derivative of the value function, so using
the upwind approximation
C U (n) = C F (n)I{D F (n)>0} + C ss (n)I{D F (n)<0,D B (n)>0} + C B (n)I{D B (n)<0} (136)

C F (n) = ρψ (h−1 (J))−θψ e ψk [∂kF J(n)]−ψ (137)

C B (n) = ρψ (h−1 (J))−θψ e ψk [∂kB J(n)]−ψ ; (138)
where
 
1 2
D F (n) = Ae −(1−α)k(n) − σ − C F (n)e −kn
δ+ (139)
2
 
1
D B (n) = Ae −(1−α)k(n) − δ + σ 2 − C B (n)e −kn (140)
2

Here we are making use of the steady-state, which we are essentially using as a boundary
condition.
For the second derivative
J(n + 1) − 2J(n) − J(n − 1)
∂kk J = (141)
(∆k)2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 40 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

Discretizing as above, we obtain

J(n + 1) − J(n)
0 =f (n) + D F (n) · I{D F (n)>0} (142)
∆k
B J(n) − J(n − 1)
+ D (n) · I{D B (n)>0} (143)
∆k
1 2 J(n + 1) − 2J(n) − J(n − 1)
+ σ , (144)
2 (∆k)2
where
C U (n)e −k ∂kU J(n) − ρ(1 − γ)J(n)
f (n) = f (C U (n), J(n)) = 1
(145)
1− ψ
and
C U (n) = C F (n)I{D U (n)>0} + C ss (n)I{D U (n)<0,D B (n)>0} + C B (n)I{D B (n)<0} (146)

C F (n) = ρψ (h−1 (J))−θψ e ψk [∂kF J(n)]−ψ (147)

−1 −θψ ψk
B
C (n) = ρ (h ψ
(J)) e [∂kB J(n)]−ψ ; (148)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 41 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

For clarity, we can write the discretized ode as

0 = f (n) + x(n)J(n − 1) + y (n)J(n) + z(n)J(n + 1) (149)
where
D B (n) · I{D B (n)<0}
x(n) = − (150)
∆k
D B,− (n) · I{D B (n)<0} − D F ,+ (n) · I{D F (n)>0}
y (n) = (151)
∆k
D F ,+ (n) · I{D F (n)>0}
z(n) = . (152)
∆k
We can use an iterative scheme to find J. Suppose we have an approximate solution
J i = (J i (1), . . . , J i (I ))> . How can we obtain a new approximation,
J i+1 = (J i+1 (1), . . . , J i+1 (I ))> ?
There are two main approaches: an explicit method or an implicit method
The explicit method allows to obtain J i+1 from J i without any matrix inversion

J(n)i+1 − J i (n)
=f (n) + x(n)J(n − 1) + y (n)J(n) + z(n)J(n + 1) (153)
∆

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 42 / 65
 Applying HJB Equation to Solve Stochastic Growth Model Solution

In matrix form, it is clear that we can obtain J i+1 directly from J i without matrix inversion.

J i+1 = f ∆ + (I + ∆A)J i (154)

where f = (f (1), . . . , f (N + 1))> and A is an N + 1 by N + 1 matrix with (y1 , . . . , yN+1 )
along the diagonal (z1 , . . . , zN ) along the superdiagonal and (x2 , . . . , xN+1 ) along the
subdiagonal and 0’s everywhere else.
We have approximated the continuous time stochastic process for k via a continuous time
Markov chain with generator matrix A (under the physical measure P)
Pk (∆) = e ∆A ≈ I + ∆A is the physical transition matrix for the Markov chain used to
approximate k. [Pk (∆)]ij is the physical probability that kt = k(j) at date-(t + ∆),
conditional on kt = k(i) at date-t
We could also use an implicit method
J i+1 (n) − J i (n)
= f (n) + x(n)J i+1 (n − 1) + y (n)J i+1 (n)
∆
z(n)J i+1 (n + 1) (155)

In matrix form
(I − ∆A) J i+1 = ∆ f + J i (156)

We must invert the matrix I − ∆A to obtain tje updated solution – this is why the method
is implicit.
Which method to use? Implicit methods are more stable in the sense that they converge
more rapidly for a given step-size, ∆.
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 43 / 65
 Basic New Keynesian Model with Risk Model

Model I

Representative household
Z ∞
Nu1+ϕ
 
−δ(u−t)
Et e ln Cu − du (157)
t 1+ϕ

C , consumption rate of composite good

N, rate of labor supply
Continuum of firms i ∈ [0, 1] produces differentiated goods

Yt (i) = At Nt (i) (158)

At , common exogenous tech level

Composite good defined by basket
Z 1  1−1 1
1− 1 
Ct = Ct (i) di (159)
0

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 44 / 65
 Basic New Keynesian Model with Risk Model

Model II
Ct (i)dt is the quantity of good i consumed by the household over the interval
[t, t + dt).
Can show that:
nominal expenditure on consumption aggregates nicely
Z 1
Pt C t = Pt (i)Ct (i)di (160)
0
where
1
Z 1  1−
Pt = Pt (i)1− di (161)
0
 −
Pt (i)
∀i ∈ [0, 1], Ct (i) = Ct (162)
Pt

static intertemporal budget constraint

Z ∞  
Wt
W0 = Et Λt Ct − Nt dt (163)
0 Pt
Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 45 / 65
 Basic New Keynesian Model with Risk Model

Model III

Lagrangian

!
∞
N 1+ϕ ∞
Z Z  
Wt
L = Et e −δt
ln Ct − t dt − κ Λt Ct − Nt dt (164)
0 1+ϕ 0 Pt

FOC’s
consumption

e −δt Ct−1 = κΛt (165)

labor

Wt
e −δt Ntϕ = κΛt (166)
Pt
Implications of FOC’s

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 46 / 65
 Basic New Keynesian Model with Risk Model

Model IV

SDF process
 −1
Cu Λu
e −δ(u−t) = (167)
Ct Λt
consumption-labor

Wt
Ntϕ Ct = (168)
Pt

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 47 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem I

Objective of firm j is to set prices in order to maximize firm value net of

adjustment costs
∞
Λ$u
Z
sup Et [Πu (j) − Θu (j)] du (169)
dPt (j)/dt t Λ$t

real firm value is the above nominal value dividend by Pt , the aggregate price
index
Nominal profit flow function

Πt (j) = Pt (j)Yt (j) − Wt Nt (j) (170)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 48 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem II

Adjustment cost function
 2
1 dPt (j)/dt
Θt = θ Pt Yt (171)
2 Pt

" 1−   #
Pt (j) Wt Pt (i)
Πt (j) = − Pt Yt (172)
Pt At Pt Pt

Simpler notation: xt = Pt (j), µt = dPt (j)/dt

Stochastic optimal control problem

"  2 #
∞
Λ$u
Z 
Wu − 1 µu
sup Et Pu Yu xu1− − xu −1
Pt − θ du (173)
µt t Λ$t Au 2 xu

s.t. dxt = µt dt (174)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 49 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem III

x is the state variable, which is the price of the good produced by the firm
µ, the rate of change of the state variable, is the control
assuming price is locally deterministic, i.e. no volatility
Rewrite objective function

"  2 #
∞
Λ$u Pu Yu
Z 
Wu − 1 µu
Pt Yt sup Et xu1− − xu −1
Pt − θ du
µt t Λ$t Pt Yt Au 2 xu
(175)

"  2 #
Z ∞ 
Λu Yu Wu − 1 µu
Pt Yt sup Et xu1− − x −1
Pt − θ du (176)
µt t Λt Yt Au u 2 xu

Λu Yu
With log utility over intermediate consumption, Λt Yt = e −δ(u−t)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 50 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem IV

Real value of firm net of adjustment costs reduces to

"  2 #
Z ∞ 
−δ(u−t) Wu − 1 µu
Yt sup Et e xu1− − xu −1
Pt − θ du (177)
µt t Au 2 xu

Value function

"  2 #
Z ∞ 
−
Ru
ks ds Wu − 1 µu
Jt = sup Et e t xu1− − xu −1
Pu − θ du (178)
µt t Au 2 xu

Hamiltonian
   2  
Wt − 1 µt dJt
H= xt1− − x −1
Pt − θ + Et (179)
At t 2 xt dt

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 51 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem V

HJB equation

sup H − δJt (180)

µt

FOC wrt µ

Hµ = 0 (181)
µ
Jx = θ 2 (182)
x

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 52 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem VI

Derive Stochastic Maximum Principle by diff’ing HJB wrt x at optimum

∂µ
0 = Hµ + Hx − kJx (183)
∂x
0 = Hx − kJx (184)
"   2 #  
1− Wt − −1 1 µt d
0 = ∂x xt − xt Pt − θ − kJx − Et Jx (185)
At 2 xt dt

Define Φ = Jx

   
 W dΦ
0 = ( − 1)x − − 1 P −1 + θµ2 x −3 + Et − δΦ (186)
 − 1 Ax dt
µ
Φ=θ (187)
x2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 53 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem VII

Symmetric equilibrium, where x = P. Define π = µ/P

   
 W dΦ
0 = ( − 1) − 1 P −1 + θπ 2 P −1 + Et − δΦ (188)
 − 1 AP dt
Φ = θπP −1 (189)

Symmetric equilibrium, where x = P. Define π = µ/P

With no adjustment costs, i.e. θ = 0, then

 Wt
Pt = (190)
 − 1 At

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 54 / 65
 Basic New Keynesian Model with Risk Model

Individual Firm’s Stochastic Optimal Control Problem VIII

Use the second equation to eliminate Φ in the first equation.

π
dΦ = θd (191)
P(
 2 )
dP dP dP
= θP −1 dπ − π − dπ +π (192)
P P P

Since P has zero diffusion term, we obtain

dΦ = θP −1 dπ − π 2 dt


(193)

Hence obtain real wage rate in terms of technology level and inflation

   
−1  Wt dπt
− 1 + Et − δπt = 0, (194)
θ  − 1 Pt At dt

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 55 / 65
 Basic New Keynesian Model with Risk Model

Bond Pricing

Nominal interest rate

it = rt + πt (195)
Real interest rate

rt = δ + µY ,t − σY2 ,t , (196)

Conditional moments of output


dYt
Et = µY ,t dt (197)
Yt
" 2 #
dYt
Et = σY2 ,t dt (198)
Yt

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 56 / 65
 Basic New Keynesian Model with Risk Equilibrium

Summary
From household’s FOC’s

 −1
Cu Λu
e −δu = (199)
Ct Λt
W t
Ntϕ Ct = (200)
Pt
From firm’s FOC’s

   
−1  Wt dπt
− 1 + Et − δπt = 0, (201)
θ  − 1 Pt At dt
From bond pricing

it = rt + πt (202)
rt = δ + µY ,t − σY2 ,t (203)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 57 / 65
 Basic New Keynesian Model with Risk Equilibrium

Solving for Eqm

Relating output to inflation

W N ϕC N ϕ NA
= = = N 1+ϕ = e (1+ϕ)(y −a) (204)
PA A A
and so

   
−1  (1+ϕ)(yt −at ) dπt
e − 1 + Et − δπt = 0, (205)
θ −1 dt

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 58 / 65
 Basic New Keynesian Model with Risk Equilibrium

New Keynesian Phillips Curve

By imposing market clearing, we can show that

 
 − 1  (1+ϕ)xt  dπt
0= e − 1 + Et − δπt , (206)
θ dt

where
Y
x = ln X = ln
, (207)
Yn
where Y n is natural output flow, i.e. output flow in the economy with no
price adjustment costs (θ = 0)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 59 / 65
 Basic New Keynesian Model with Risk Equilibrium

Dynamic Investment Savings Equation

We can show that (see Assignment 2)

1 2
it = rtn + πt + µX ,t − σX2 ,t = rtn + πt + µx,t − σx,t , (208)
2
where
dxt = µx,t dt + σx dZx,t , dZx,t dZt = ρX ,t dt (209)
It follows that
1 2
dxt = (it − r n − πt + σx,t )dt + σx dZx,t (210)
2

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 60 / 65
 Basic New Keynesian Model with Risk Equilibrium

FBSDE

forward sde for output gap

1 2
dxt = (it − r n − πt + σx,t )dt + σx dZx,t , x0 given (211)
2
backward sde for inflation

 − 1  (1+ϕ)xt  ∂πt
dπt = δπt dt − e − 1 dt + dZx,t lim Et [e −δ(T −t) πT ] = 0
θ ∂xt T →∞
(212)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 61 / 65
 Basic New Keynesian Model with Risk Equilibrium

Nominal Interest Rate Rules I

Make nominal interest rate rule depend on current inflation and output gap

it = v (πt , xt ) (213)
Obtain a system of stochastic differential equations to pin down outgap and
inflation
forward sde for output gap
1 2
dxt = (v (πt , xt ) − r n − πt + σx,t )dt + σx dZx,t , x0 given (214)
2
backward sde for inflation

 − 1  (1+ϕ)xt  ∂πt
dπt = δπt dt − e − 1 dt + dZx,t lim Et [e −δ(T −t) πT ] = 0
θ ∂xt T →∞
(215)

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 62 / 65
 Basic New Keynesian Model with Risk Equilibrium

Nominal Interest Rate Rules II

Suppose that v (π, x) = a + φπ π + φx x. Then

1 2
dxt = (a + (φπ − 1)π + φx x − r n + σx,t )dt + σx dZx,t , x0 given (216)
2
 − 1  (1+ϕ)xt  ∂πt
dπt = δπt dt − e − 1 dt + dZx,t lim Et [e −δ(T −t) πT ] = 0
θ ∂xt T →∞
(217)

The nominal interest rate rule does not pin down σx .

Need to think about more than interest rates when setting policy. Need to
think about output gap volatility, which can be determined via policies
affecting risk prices/risk premia.

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 63 / 65
 Basic New Keynesian Model with Risk Equilibrium

Summary

Quantitative Easing
Some Stochastic Control Theory!
Derived Basic New Keynesian Model with Risk
Still need to apply Basic New Keynesian Model with Risk

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 64 / 65
 Basic New Keynesian Model with Risk Equilibrium

Next Steps

Monetary Policy in Liquidity Traps

Quantitative Easing

Harjoat S. Bhamra Lecture 2: Central Banks, Monetary Policy and Risk 2015 65 / 65