MFE Macro Slides

MFE (Macro)Economics
Lecture 1: Economic Growth

in
*
exams, expect us to know functions in the box
Artur Doshchyn
University of Oxford
Hilary Term 2023

Introduction Solow model and exogenous growth Endogenous growth Romer model Social infrastructure
Questions that Macroeconomics helps to answer
What determines economic growth in the long run?

Why are there booms and recessions, and how to deal with the latter?
What determines the level of unemployment?
How to conduct monetary and fiscal policy?
Why financial crises happen, and how to prevent them?
How quickly will the economy recover after the Covid-19 pandemic?
Do issues in the financial sector matter for the macroeconomy?
etc. etc.
1 / 49
This course
How can we answer macroeconomic questions?
We need models and theories!
In this course, you will

learn some of the most important models in modern macroeconomics
thus obtain a toolbox that will enable you to think about
macroeconomic issues in a systematic manner.
2 / 49
How to do well in the exam

Necessary and sufficient:
1. Attend and study the lectures
2. Work through the problem sets
3. Attend the classes.
Core textbook:
God level: read all the papers and blogs on the reading list, then read top
100 cited papers in econ, then prove all theorems in Bob Lucas’s papers.
3 / 49
Plan of action
Week 1 Economic Growth – TODAY!

Week 2 Economic Fluctuations
Week 3 Monetary Policy
Week 4 Labour Market and Unemployment
Week 5 Financial Frictions and Financial Crises
Week 6 Budget Deficits and Sovereign Debt Crises
4 / 49
Why there are such huge cross-country income di↵erences?
5 / 49
What are the reasons for growth miracles and disasters?
6 / 49
How did rich countries maintain such steady growth?
7 / 49
Small di↵erences in growth matter. How increase it?
8 / 49
Let Bob Lucas say it
“The consequences for human welfare

involved in questions like these are simply
staggering: Once one starts to think about
them, it is hard to think about anything
else.”
Robert E. Lucas Jr.

Nobel Prize in Economics 1995
9 / 49
Kaldor’s facts about economic growth
In 1957 paper, Nicholas Kaldor summarized statistical properties of

economic growth in the long run with the following stylized facts:
1. Output per worker grows at a roughly constant rate

2. Capital per worker grows over time
3. Capital/output ratio is roughly constant
4. Rate of return to capital is constant
5. Shares of capital and labour in net income are nearly constant
6. Real wage grows over time.
We would expect a good model of growth to fit and explain these facts.
10 / 49
Solow model overview
Simple dynamic model of growth many of you already know

Features production and capital accumulation
Assumes diminishing marginal returns to capital (accumulable factor)
There are no optimizing agents (e.g. saving rate is fixed)
May not answer central questions of growth...
but surprisingly powerful framework we will use extensively
P.S. This lecturer takes Robert Solow seriously!
11 / 49
Inputs and outputs

Neoclassical production function: The only two inputs are capital and labour
Yt = F( Kt , At ⇥ Lt ) (1)
|{z} |{z} |{z} |{z}
output capital knowledge labour
At Lt – e↵ective labour, technology is labour-augmenting.

Constant return to scale (CRS): F (cK, cAL) = cF(K, AL) 8c > 0.
K 1
Set c = 1/AL to get F( AL , 1) = AL F(K, AL).
Effective labour
K Y
Let k ⌘ AL , y ⌘ ALbe capital and output per unit of e↵ective
labour, and f ⌘ F(k, 1).
Can write production function in its intensive form:
y = f(k) (2)
12 / 49
Inada conditions
We assume f(k) to satisfy Inada conditions:
f(0) = 0, f 0 (k) > 0, f 00 (k) < 0, lim f 0 (k) = 1, lim f 0 (k) = 0
k!0 k!1
Concave function
f(k)
k
13 / 49
Evolution of inputs
dXt
Time is continuous. We will typically denote dt as Ẋt .
Labour and knowledge grow at constant rates n and g:
L̇t = nLt () Lt = L0 ent (3)

g dxt1d<- Ȧt = gAt
= () At = A0 egt (4)
continuous compounding factor
E↵ective labour At Lt then growth at the rate n + g.
Accumulation of capital:
K̇t = It - Kt , 2 (0, 1) (5)

|{z} |{z}
investment depreciation
14 / 49
Saving rate and investment
Solow model assumes constant saving rate s 2 (0, 1), i.e. at any time
t fraction 1 - s of output is consumed, and fraction s is invested:
Ct = (1 - s)Yt
It = sYt
Capital thus accumulates according to:

There’s no model to determine
consumers’ preference of saving, so
the assumption of people always K̇t = sYt - Kt (6)
saves s and invest sY is strong
This is a strong assumption: we do not specify preferences that yield
such saving behaviour, and hence cannot say anything about welfare.
But we still get a bang for the buck with the Solow model!
15 / 49
Dynamics of capital per unit of e↵ective labour
Take derivative of kt with respect to time:

✓ ◆
d Kt K̇t At Lt - Kt [At L̇t + Ȧt Lt ]
k̇t = =
dt At Lt [At Lt ]2
K̇t L̇t Ȧt

Use At Lt = syt - kt , Lt = n, At = g, and yt = f(kt ) to get
k̇t = sf(kt ) - (n + g + )kt (7)
sf(kt ) is the actual (gross) investment per unit of e↵ective labour.

(n + g + )kt is break-even investment necessary to keep k at its
existing level, i.e. overcome depreciation and growing e↵ective labour.
k grows over time (k̇t > 0) when actual inv. > break-even inv.
16 / 49
Dynamics of k graphically
k̇ = 0
k̇ > 0 (n + g + )k
sf(k)
k̇ < 0
k⇤ k
17 / 49
Steady state
Regardless of the starting point, k converges to steady state level k⇤

k̇ > 0 when k < k⇤ , and
k̇ < 0 when k > k⇤ .
k⇤ solves k̇ = 0, i.e.
sf(k⇤ ) - (n + g + )k⇤ = 0. (8)
Output, consumption, and investment per unit of e↵ective labour also

converge to their respective steady state levels:
y⇤ = f(k⇤ )
c⇤ = (1 - s)f(k⇤ )
i⇤ = sf(k⇤ )
18 / 49
Steady state graphically
f(k)
y⇤
(n + g + )k
c⇤
sf(k)
i⇤
k⇤ k
19 / 49
Balanced growth path (BGP)
How does the economy behave in the steady sate (k = k⇤ )?

Labour and knowledge grow at the rates n and g by assumption.
E↵ective labour At Lt and capital stock Kt = At Lt k⇤ both grow at
the rate n + g.
Output Yt also grows at the rage n + g given const. return to scale.
Kt Yt
Capital per worker Lt and output per worker Lt grow at the rate g.
Thus, regardless of the starting point, the economy converges to a
balanced growth path on which all variables grow at a constant rate.
This fits well with Kaldor’s stylized facts.
20 / 49
Change in the saving rate s changes the steady state

When saving rate increases
ST: consume less, because save and invest more
LT: still consume less as shown in graph
-> in Solow, saving more will not make people better off f(k)
y⇤new
y⇤old
cnew)
[
(n + g + )k
snew f(k)
Cold
sold f(k)
&
k⇤old k⇤new k
21 / 49
The economy converges to the new balanced growth path

knew
When saving increases, capital then
converges to new steady state
kold
t0 t
Growth
Growth is boosted temporarily
rate of
Y/L
g
t0 t
ln(Y/L)
ln(Y/L) increases in the short run, then

converges back to the growth rate of g
t0 t
22 / 49
The e↵ect of an increase in s: in words
Higher saving rate shifts the investment curve up, so the steady state
level of capital per unit of e↵ective labour increases.
k does not jump to k⇤new instantaneously, but grows gradually.
Growth rate of output per worker initially accelerates above g, since
Y/L = Af(k), and both A and k are increasing.
But as k reaches k⇤new , growth rate of Y/L returns to g.
Output per worker settles on a new path, parallel to the first.
Increase in the saving rate thus has a level e↵ect, but not growth
e↵ect on output.
23 / 49
How big is the level e↵ect quantitatively?
Suppose prod. func. is Cobb-Douglas: Y = K↵ (AL)1-↵ , or y = k↵ .

Can solve closed form for y⇤ (practice, use slide 18):
✓ ◆ 1-↵
↵ ✓ ◆ 1-↵
↵
⇤ s y⇤new snew
y = ) =
n+g+ y⇤old sold
If markets are competitive and there’re no externalities, ↵ corresponds

to the share of income that goes to capital, which is about 1/3.
Thus, for example, doubling saving rate s results in only about 41%
increase in output in the long run.
Not nearly enough to explain vast income di↵erences across the world!
24 / 49
E↵ect of saving rate on consumption
It is consumption that matters for people’s welfare, not output!

Increase in the saving rate always implies a fall in consumption in the
short run, because it takes time to increase capital and output.
In the long run, the e↵ect is ambiguous.
Steady state c⇤ = f(k⇤ ) - (n + g + )k⇤ .
If s (and hence k⇤ ) increases, c⇤ increases only if the marginal
product of capital f 0 (k⇤ ) exceeds gross depreciation rate n + g + .
Otherwise extra output from additional capital is not enough to
maintain higher k⇤ , and resources are diverted from consumption.
25 / 49
The Golden Rule
Golden-rule level of k⇤ is one that maximises consumption in the

steady state, i.e. k⇤G satisfies the FOC: f 0 (k⇤G ) = n + g + .
Never socially optimal to accumulate capital above this level.
But can be optimal to remain below it. Why?
) While consumption is lower in the long run when k⇤ < k⇤G , it is
higher in the short run, and people are impatient.
Not maximise consumption, but total welfare here
26 / 49
Other modern exogenous growth models
How is the saving rate determined in the economy?

1 Ramsey model:
add infinitely lived representative agents with dynamic preferences
optimal consumption/saving decisions (remember maths lectures)
No externalities, socially optimal level of saving, k⇤ < k⇤G .
2 Overlapping-generations model:
agents also make privately optimal consumption/saving choices
but live limited amount of time: old agents die, new agents are born
Possibility of dynamic inefficiency: agents save too much for old age
and k⇤ > k⇤G .
Redistribution from young to old can make every generation better o↵.
Similarly to the Solow model, both above model economies converge to a

balanced growth path, on which Y/L grows at the exogenous rate g.
27 / 49
What have we achieved so far?
An elegant growth model:

characterizes dynamics of key economic variables such as capital,
output, and consumption.
features balanced growth path that fits Kaldor’s facts.
models that endogenize consumption/saving behaviour exist.
Central prediction:
Capital accumulation cannot account for long-run growth in output per
worker, or cross-country income di↵erences.
Only changes in the rate of tech. progress a↵ect long-run growth.
But g, the growth rate of A is exogenous here – hence the name of

this class of models.
To quote Romer textbook: “it is only a small exaggeration to say that
we have been modeling growth by assuming it.”
28 / 49
Endogenous growth: introduction
Endogenous growth models seek to explain growth as the result of

purposeful human activity.
One of the main aims is to identify forces behind the rate of
technological progress g, such that it can be a↵ected by policy.
Some famous examples include:
knowledge spillovers and learning by doing (Frankel 1962, Arrow 1962)
human capital accumulation (Lucas 1988)
research and development (Romer 1990).
We will begin with a very simple extension of Solow model by adding

production of knowledge.
29 / 49
Simple model: production of goods and knowledge

No capital for simplicity.
Fraction aL of people work in the R&D sector, and 1 - aL in the
goods-producing sector.
Production functions for output and knowledge:
Yt = At (1 - aL )Lt (9)
Ȧt = B(aL Lt ) A✓
t (10)
The growth rate of knowledge is thus
Ȧt
gt = = B(aL Lt ) A✓-1
t (11)
At
Change in growth rate gt over time:
ġt = ngt + (✓ - 1)g2t (12)
30 / 49
Case 1: Fishing out ideas (✓ < 1)
✓ < 1: decreasing return to knowledge in generation of new ideas.

There is a balanced growth path, as gt converges to
n
g⇤ = (13)
1-✓
Easy to verify that ġt > 0 when gt < g⇤ , and ġt < 0 when gt > g⇤ .
Only higher population growth leads to higher g: when there are
more people to make discoveries, more discoveries are made!
Such important parameters of the model as aL do not a↵ect long-run
growth. Like s in Solow model, aL only has a level e↵ect here.
This case is thus known as “semi-endogenous” growth model.
The purpose of endogenous growth model is for policy makers to set rules to stimulate growth. Because the
only thing that drives growth is n(population growth), but it’s not useful for policy making. From this
equation, policy makers cannot set the policy to boost growth, so we name it semi-endogenous
31 / 49
Cases 2 and 3: Standing on the shoulders of giants (✓ > 1)
✓ > 1: increasing return to scale, ever increasing growth (ġt > 0).
✓ = 1, constant return to scale in knowledge production:
gt = B(aL Lt ) (14)
ġt = ngt . (15)
✓ = 1 and n > 0: ever increasing growth.

Scale e↵ect: more population ) more R&D ) more growth.
There are ways to remove scale e↵ect, e.g. growing # of sectors.
✓ = 1 and n = 0: simple fully endogenous linear growth, or “AK”

model. Growth depends on the fraction of resources devoted to R&D.
Fully endogenous growth: overcoming diminishing returns is key.
32 / 49
Romer (1990) model overview

Fully endogenous growth model with purposeful R&D activity.
Knowledge is nonrival - many firms can benefit from the same idea.
One more computer makes one worker more productive.
One more idea makes all workers more productive!
Fundamental idea: knowledge is different from other inputs, it generates increasing rate of return because of the feature of non -rivalry
Non-rivalry of ideas ) growth in income per person is tied to growth

in the total stock of ideas, not to growth in ideas per person!
R&D requires payment of fixed costs upfront.
Must introduce monopolistic competition framework: innovators
become monopolists and make profits, otherwise no innovation.
Knowledge in the model is thus excludable (think patents).
Deviation from perfect competition creates inefficiencies.
Landed Paul Romer a Nobel prize in 2018 - must be worth learning.
33 / 49
Ideas and input goods
First consider a static framework (drop t’s), then add dynamics.

No capital for simplicity.
There is a continuum of ideas: i 2 [0, A], A - total stock of ideas.
Each idea is held by a patent holder with exclusive rights to use it.
Patent holder of idea i hires Li labour to produce xi input goods
based on his idea.
Simple one-for-one technology:
xi = Li .
Patent holder is a monopolist and sells inputs to final good producers

at price pi .
34 / 49
Final good producers

Competitive final good producers combine input goods using
Dixit-Stiglitz-Ethier CES technology to produce final output:
Constant elasticity substitution
Z !1/
A
Y= xi di , 2 (0, 1). (16)
i=0
Marginal product of input i is

ZA ! 1- ✓ ◆1-
@Y 1 -1 Y
= xi di xi =
@xi i=0 xi
Equalize with (marginal) cost pi , yielding a downward-sloping

demand function for input i:
- 1-1
xi = Y pi . (17)
High substitutability \phi means: if any goods becomes cheaper, then the demand for the
good will increase as much as possible
35 / 49
Patent holder profit within one period

Profit maximization of patent holder i: monopolistic competition
- 1-1
max ⇡i = pi xi - wLi s.t. xi = Y pi , (18)
where w is the wage on the competitive labour market.
FOC yields:
w
p= , (19)
1
i.e. price is the same for all monopolists, who set constant markup .
Number of labour employed per monopolist for each unit of good
Quantity of each input is thus also the same, x = LAY , where LY is the
aggregate labour employed in production of inputs.
Therefore, a monopolist earns the following profit in equilibrium:
1 - LY
⇡= w. (20)
A
36 / 49
Development of new ideas: adding dynamics

LY
Substitute all xi with x = A in production function (16) to get:
h i1/ 1-
Y = A (LY /A) =A LY (21)
Constant return to labour LY , but increasing return to scale to labour

and knowledge A: New ideas are valuable!
->
constant thatrepresents
A new idea can be developed with 1/(BA) units of labour. productivity
R&D externality: A raises productivity in development of new ideas.
Free entry: anyone can pay cost w/(BA) and become a patent holder.
Evolution of knowledge over time is, then:
totallaborate
LA.t
labour new idea IBA
Ȧt = BLA,t At , (22)
where LA,t is the aggregate labour employed in R&D at time t.

37 / 49
Individual consumers: determining the interest rate

Representative individual’s lifetime utility:
Z1
U= e-⇢t ln Ct dt (23)
t=0
where Ct is consumption at t, ⇢ is the discount rate.
Similar to the maths lectures, but in continuous time!
For an optimizing agent, marginal cost of saving equals marginal
value of saving for a small period t, given the interest rate rt :
1 1
= e(rt -⇢) t
Ct Ct+ t
⇣ ⌘
Rearrange and use Ct+ t = Ct exp Ċ t
Ct t to get Euler equation:
need to know this
Ċt
rt = ⇢ + in discrete times (24)
Ct
38 / 49
Solving the model: growth rates and interest rate

Population is constant L̄ > 0, and labour is supplied inelastically.
Suppose labour employed in production LY , and labour employed in
R&D LA are constant in equilibrium (which is, in fact, unique), so
LA + LY = L̄. (25)
Knowledge At then grows at the constant rate g = BLA , using (22).

1-
Output Yt grows at the rate gY = BLA , using (21).
Yt
Because all output is consumed, Ct = L̄
also grows at the rate gY .
Euler equation (24) implies that the interest rate r is constant:
1-
r=⇢+ BLA . (26)
39 / 49
Solving the model: the value of a new idea
Competition and CRS imply that final good producers earn zero profit
) all revenues (Yt ) paid to the monopolist suppliers of inputs.
Wage wt must grow at the same rate as output gY , since the wage
bill wt LY is a constant fraction of monopolists’ revenues – see (19).
The growth rate of a monopolist’s profit ⇡t given in (20) is then the
growth rate of wage gY minus the growth rate of knowledge g:
1-2
g⇡ = gY - g = BLA (27)
The present value of the profits from discovering a new idea is, then:
⇡t 1- L̄ - LA wt
Rt = = . (28)
r - g⇡ ⇢ + BLA At
40 / 49
Solving the model: final step
When R&D is positive, free entry into idea creation ensures that PV
of profits Rt is equal to the cost of generating a new idea at any t:
1- L̄ - LA wt wt
= . (29)
⇢ + BLA At BAt
Solve for LA and impose LA > 0 to get:
⇢
LA = max (1 - )L̄ - ,0 . (30)
B
This, finally, yields the equilibrium growth rate of output:
* (1 - )2
gY = max BL̄ - (1 - )⇢, 0 (31)
41 / 49
Implications
Successfully derived growth from the microeconomic environment:

rational agents purposefully devote resources to R&D.
Growth gY is higher when
individuals are more patient (⇢ is lower),
substitutability between inputs is lower,
productivity of R&D sector B is higher, and when
population L̄ is larger (scale e↵ect again).
Ability to capture profit from the new idea is key, hence the
monopolistic competition structure is important.
But since the economy is not perfectly competitive, welfare theorems
do not apply, and the equilibrium is generally not socially optimal.
42 / 49
The gap between equilibrium and socially optimal R&D

Given LA , we can compute the welfare of a representative household:
Z1  z A
}|t { 1-
-⇢t L̄ - LA
U= e ln A0 eBLA t dt
t=0 | {z L̄ }
Ct =Yt /L̄
(32)
✓ ◆
1 L̄ - LA 1 - 1- BLA
= ln + ln A0 +
⇢ L̄ ⇢
Socially optimal Lopt

A > 0 maximizes the above. FOC yields:
⇢
Lopt
A = max L̄ - ,0 (33)
1- B
Easy to see that equilibrium LA in (30) is a fraction (1 - ) of Lopt

A .
Growth is inefficiently low in the Romer model! Why?

43 / 49
What have we achieved?
Knowledge accumulation is the driving force behind global growth.

Endogenous growth models identify many determinants of R&D.
Policy-relevant, e.g. Romer model makes a case for R&D subsidies.
But less successful in explaining cross-country income di↵erences:
Knowledge is non-rival: why don’t producers in poor countries use it?
Flow of knowledge would have to be unrealistically slow (homework).
Lack of intellectual property rights etc. could be a factor, but difficult
to believe this alone accounts for vast di↵erences in incomes.
What factors allow richer countries to take better advantage of

technology?
44 / 49
Social infrastructure
One of the leading hypotheses is that di↵erences in productivity and

income stem from di↵erences in social infrastructure (SI).
Hall and Jones (1999) define SI as institutions and policies that align
private and social returns to activities.
Productive activities: investment, education, R&D etc. may earn
individuals lower returns than their social value.
Diversion activities: theft, corruption, and rent-seeking that can earn
individuals high private returns, but have zero or negative social value.
SI has many dimensions:
Fiscal policy: tax rates directly a↵ect relationship between private and
social returns.
Environment for private decisions: crime prevention, enforcement of
contracts, limits to monopoly power etc.
Extent of rent-seeking by the government: lobbying, bribes,
expropriation etc.
45 / 49
Social infrastructure strongly correlates with income
Source: Hall and Jones (1999).

The measure of SI is based on an index of government anti-diversion policies from Political
Risk Services and an index of openness by Sachs and Warner (1995).
46 / 49
E↵ect of SI on income: IV regressions

But correlation is not necessarily causation: would improvements in
social infrastructure lead to higher income?
Yi
Regression of the form ln L i
= a + b · SIi + ei is unlikely to give a
reliable estimate of impact of SI on income due to endogeneity:
) SIi is likely to correlate with the error term ei due to omitted variables,
measurement errors and simultaneity, resulting in a biased estimate of b
Need valid instrumental variables (IV) that correlate with social
infrastructure, but not with other potential determinants of income.
) For example, Hall and Jones use the country’s distance from the
equator, the extent to which European languages are spoken as native
today, and a measure of influence of geography on trade.
2-stage estimation: first regress SIi on instruments, and then

Yi
regress ln L ˆ i from the first stage.
on the fitted values SI
i
If instruments are valid, we get a valid estimate of b.

47 / 49
E↵ect of SI on income: results
Hall and Jones (1999) find strong evidence in favour of the social
infrastructure hypothesis.
First, the estimated impact of SI on income is both quantitatively
large and statistically significant:
b̂ = 5.143 with s.e. (0.508).
I.e. a 0.01 increase in SI index is associated with a 5.14% increase in
income per worker.
Second, variations in SI account for a large fraction of cross-country
income di↵erences:
) Countries with the highest SIi predicted to have 25 to 38 times higher
income than countries with lowest SIi based on the regressions.
These results are generally representative of the broader empirical

literature on the role of SI in cross-country income di↵erences.
48 / 49
Where are we?
The work on social infrastructure highlights that countries’

institutions and policies are important determinants of income.
BUT! SI is a broad concept, and many questions remain.
What aspects of SI are the most important?
) e.g. Glaeser et al (2004), Jones and Olken (2005).
What determines SI in the first place?
) e.g. Shleifer and Vishny (1993), Acemoglu and Robinson (2000, 2006).
What are the other important sources of cross-country income
di↵erences?
) e.g. Nunn and Puga (2012), Hendricks and Schoellman (2018).
Fascinating, growing, and important research!
49 / 49
Lecture 2: Economic Fluctuations
Artur Doshchyn
Hilary Term 2023

Introduction Real Business Cycles Monetary Theories New Keynesian Model
US real GDP per capita: trend and fluctuations
1 / 50
Short run properties of macroeconomic data
Variable Sd% Cross-correlation of output with:

t-3 t-2 t-1 t t+1 t+2 t+3
Output (GNP) 1.72 0.38 0.63 0.85 1.00 0.85 0.63 0.38
Cons., non-dur. 0.86 0.55 0.68 0.78 0.77 0.64 0.47 0.27
Cons., durable 4.96 0.49 0.65 0.75 0.78 0.61 0.38 0.11
Investment 8.24 0.38 0.59 0.79 0.91 0.76 0.50 0.22
Hours 1.59 0.30 0.53 0.74 0.86 0.82 0.69 0.52
Avg Hours 0.63 0.34 0.48 0.63 0.62 0.52 0.37 0.23
Employment 1.14 0.23 0.46 0.69 0.85 0.86 0.76 0.59
GNP/Hours 0.90 0.20 0.30 0.33 0.41 0.19 0.00 -0.18
Avg wage 0.55 0.21 0.14 0.09 0.03 -0.07 -0.09 -0.09
Source: Cooley and Prescott (1995)
Output, consume durable, employment is relatively volatile in business cycle;

Average wage barely changes over business cycle
2 / 50
Some stylized business cycles facts
No simple regular or cyclical pattern: ‘cycles’ is rather a misnomer.

Relatively long periods with output slightly above trend, interrupted
by recessions – brief periods when it falls relatively far below trend.
Non-durable consumption is smooth, much less volatile than output.
Investment and durable consumption much more volatile than output.
Volatility of output and total hours is similar.
Employment is more volatile than average hours.
) labour market adjusts on extensive rather than intensive margin.
Productivity (GNP/Hours) is slightly procyclical.

Wages are acyclical, and less volatile than productivity.
A good model of fluctuations should at least fit these patterns.
3 / 50
DSGE models
Modern macroeconomic models of fluctuations are DSGE:
Dynamic
) Agents’ decisions take expectations of the future into account, while
future outcomes are a↵ected by current agents’ decisions.
Stochastic
) The economy is hit by random shocks.
Agents’ decisions take uncertainty into account.
General Equilibrium
) All markets are interconnected – need to analyse them together.
As opposed to partial equilibrium analysis of a single market.
4 / 50
Most DSGE models follow the Frisch-Slutsky paradigm
5 / 50
Part A
Real Business Cycles 4
Real Business Cycle (RBC) models: an overview

Walrasian framework: perfect competition and no imperfections.
Real disturbances:
Shocks to production technology in the original RBC models
Modern RBC models consider a range of other shocks, e.g. to gvt.
spending, news about future productivity, investment technology etc.
No role for nominal disturbances or monetary policy:
Remember GE micro: only relative prices matter!
No involuntary unemployment:
Employment reflects changes in the amount people want to work.
No role for stabilization policy:
First Welfare Theorem: equilibrium is Pareto efficient!
RBC attracted fierce criticism.

But it is a great first step in learning about DSGE models.
6 / 50
Production and technology shocks

The production function is Cobb-Douglas:
1-↵
Yt = K↵
t (At Lt ) ↵ 2 (0, 1). (1)
Stochastic growth in technology At :
ln At = ln A0 + gt +Ãt , where Ãt = ⇢Ãt-1 + "t

| {z } | {z }
trend persistent deviation
Our focus is on fluctuations around the trend, so, for simplicity,

abstract from growth and set ln A0 + gt = 0 to get an AR(1) process:
ln At = ⇢ ln At-1 + "t , (2)
where "t is a mean-zero technology shock at time t.

Textbook keeps the growth, but this does not a↵ect conclusions.
7 / 50
Firms
There is a large number of identical firms.

A static process because firm can re-
A representative firm maximises profits: optimise in every period
static problem
1-↵
max K↵
t (At Lt ) - wt Lt - rt Kt (3)
Kt ,Lt
Constant return to scale + perfect competition = zero profits.

FOCs yield interest rate and wage:
✓ ◆
At Lt 1-↵
rt = ↵ ⌘ Marginal Product of Capital (4)
Kt
✓ ◆
Kt ↵
wt = (1 - ↵) At ⌘ Marginal Product of Labour (5)
At Lt
8 / 50
Households
A representative household solves (note the expectation E):

dynamic problem 1
X
t
max E [U(Ct ) - V(Lt )] (6)
{Ct ,Lt ,Kt+1 }1
t=0
Technology shock affects Kt 0
1) there’s uncertainty in the utility s.t. Ct + It = wt Lt + rt Kt (7)
function, so take expectation
2) cannot decide Kt, can only plan for the Kt+1 = (1 - )Kt + It (8)
future K(t+1)
A dynamic optimization problem, as we studied in maths classes!
U(Ct ) is instantaneous utility from consumption, U 0 > 0, U 00 < 0.
V(Lt ) is the disutility from working, V 0 > 0, V 00 > 0.
< 1 is the subjective discount factor.
Ct and Lt are controls, Kt – stock (aka state) variable.
9 / 50
Household optimality conditions
Substitute It from the capital evolution equation (8) into the budget
constraint (7), then Ct from (7) into the objective function (6) to get:
1
X
t
max 1 E [U(wt Lt + (1 + rt - )Kt - Kt+1 ) - V(Lt )] (9)
{Lt ,Kt+1 }t=0 | {z }
0
Ct
=...
+
p+ [u(WtLt RtKt+
-
kt 1)
+
-
V(Lt)] + ptH(u(Wt+ (t + 1 +
~
FOCs at any arbitrary time t are: Rt

+t+ k+ 2) V(Lt +1)] ...
-
+ +
-
⇥ ⇤
w.r.t. Kt+1 : U 0 (Ct ) = Et U 0 (Ct+1 )Rt+1
w.r.t. Lt : V 0 (Lt ) = U 0 (Ct )wt ,
has uncertainty, so I
where we denoted Rt ⌘ 1 + rt - .
Lets analyse these FOCs in turn.
10 / 50
Euler equation and consumption

need to know how to derive this
Stochastic Euler equation: utility

↓ for any given
⇥ ⇤
U 0 (Ct ) = Et U 0 (Ct+1 )Rt+1 (10)
Interpretation: Marginal disutility of saving today should equal

discounted expected marginal utility gain next period.
How does Ct behave when interest rate increases?
Substitution e↵ect (from Euler equation): Rt+1 " )
U 0 (Ct+1 ) Ct+1
U 0 (Ct ) # ) Ct ", since t + 1 consumption is relatively cheaper.
Income e↵ect: (from budget constraint): Rt+1 " increases interest

rate income, so consumption increases in all periods, including today.
The net e↵ect is ambiguous, though empirically income > substitution.
The two e↵ects exactly cancel each other when U(Ct ) = ln(Ct ).
11 / 50
Labour supply
Optimal labour supply equalizes marginal disutility from work with

marginal utility gain from consumption:
V 0 (Lt ) = U 0 (Ct )wt (11)
Substitute U 0 (Ct ) = V 0 (Lt )/wt into the Euler equation to get:

u'CCt)Wt PlEt[n'(Ct+1) Rt+]
=>
=

wt
V 0 (Lt ) = Et V 0 (Lt+1 ) Rt+1 (12)
wt+1
Intertemporal substitution in labour supply:

) It is more attractive to work today rather than next period when
relative wage (wt /wt+1 ) is high, or interest rate Rt+1 is high.
Crucial to generate employment fluctuations in RBC.
12 / 50
No-Ponzi scheme constraint and Transversality condition
Recalling maths classes, the solution must also satisfy:
1. No-Ponzi constraint: not possible to finance ever increasing

consumption with ever increasing debt.
TVC
2. Transversality condition: not optimal to let assets (Kt ) diverge
relative to consumption in the long run.
Formally, this requires
T
lim K
T T +1 = 0,
↓ T !1
It means there’s no additional capital left behind after time T
where T = U 0 (CT ) is the value of an extra unit of resources at T .
Interpretation: Agents do not leave unconsumed resources on the table
(especially clear when the planning horizon T is finite).
13 / 50
Solving the model
Solve for Walrasian equilibrium in which all markets clear at all times.
To solve the model means to find all endogenous variables (Ct , Lt ,
Kt+1 etc) as functions of state variables (Kt and At in our case).
The solution can then be used to study the model’s behaviour.
In general, analytical solution is not possible.
Instead, RBC and other DSGE models are usually solved and analysed
using approximations and numerical techniques.
Homework guides you through the steps of solving a special case of
the RBC model which can be solved analytically.
14 / 50
Model calibration
First, pick functional forms, e.g. following Romer textbook:
U(Ct ) = ln(Ct ) and V(Lt ) = -b ln(1 - Lt ). (13)
Then calibrate: choose parameter values based on empirical evidence.

Examples of evidence used: factor shares, capital-output ratio,
investment-output ratio, division of time between work and leisure etc.
Typical parameter values at quarterly frequency: ↵ = 0.36, = 2.5%,
b = 2, = 0.99.
Based on the behaviour of the Solow residual in the data, set
iid
ln At = 0.95 ln At-1 + "t , "t ⇠ N(0, 0.0112 ). (14)
Calibration is not a formal statistical test of the model, but allows us

to identify the model’s major successes and failures.
15 / 50
Impulse response functions (IRFs) and simulations
IRFs are very useful in tracing out how various macroeconomic

variables move in response to shocks to the economy.
Suppose the economy starts in its steady state, and is hit by a single
1% positive technology shock, i.e. set "1 = 0.01.
Compute the path of At using (14)
) “switch o↵” any future shocks, i.e. set "2 = "3 = ... = 0.
Compute the paths of other variables using your solution of the model
(often as log or % deviation from the steady state).
Plot your results and enjoy
Alternatively, we can also simulate the model: hit it by a sequence of
randomly drawn shocks, then compute variances, correlations etc.
!
16 / 50
Example IRFs for an RBC model
17 / 50
Short-run properties of simulated RBC data
Variable Sd% Cross-correlation of output with:

t-3 t-2 t-1 t t+1 t+2 t+3
Y 1.35 0.23 0.44 0.70 1.00 0.70 0.44 0.23
C 0.33 0.46 0.59 0.73 0.84 0.50 0.23 0.02
I 5.95 0.17 0.39 0.66 0.99 0.71 0.47 0.27
L 0.77 0.15 0.37 0.65 0.99 0.72 0.48 0.28
Y/L 0.61 0.33 0.51 0.73 0.98 0.65 0.38 -0.16
Source: Cooley and Prescott (1995)
18 / 50
Evaluation
Some successes, e.g. the model produces significant volatility in

output and investment.
Unfortunately, the model fails along several important dimensions:
predicts only about half of the volatility in hours (L) relative to data.
wage strongly procyclical in the simulations, but acyclical in the data.
productivity (Y/L) highly procyclical in the model but not in the data.
consumption is too smooth in the simulations.
Some of these can be addressed by:
adding other shocks, e.g. government purchases, tastes etc.
adding extensions such as indivisible labour, adjustment costs,
distortionary taxes etc.
But there are deeper issues with the RBC )
19 / 50
Difficulties with RBC models

No meaningful propagation mechanism to productivity shocks:
) dynamics of output closely follows the dynamics of the shocks.
i.e. we can generate volatile and persistent output fluctuations only by
putting in volatile and persistent shocks (WYGIWYPI).
Little independent evidence for the large technological disturbances:
Variations in the Solow residual reflect factors other than pure
productivity shocks, e.g. aggregate demand.
Great Depression was surely not caused by a technological regress.
Strong direct evidence that goods, labour, and financial markets are
non-Walrasian in ways that are relevant to aggregate fluctuations.
There is more to (un)employment fluctuations than just willingness of
people of substitute labour between periods – more in week 4!
Overwhelming evidence that monetary shocks have real e↵ects:
Significant deviations from Walrasian model needed to account for this.
That’s what we are going to pursue for the rest of this lecture.
20 / 50
Part B
Monetary Theories of the Business Cycle
The estimated e↵ect of a monetary policy shock on output
Source: Romer and Romer (2004)
21 / 50
Nominal rigidities and the way forward 5
For monetary disturbances to have real e↵ects, there must be some

nominal rigidities.
Otherwise, monetary changes would result in proportional changes in
all nominal prices with no impact on real prices and quantities.
Thus modern macroeconomic models introduce some barriers in
adjustment of nominal prices or wages.
We will shortly introduce a New Keynesian model – a modern

DSGE framework with nominal rigidities.
But to set the stage, let us first have a brief overview of the major
developments in the monetary economics since the mid XX century.
We will trace the evolution of understanding of the relationship between
inflation and output (or unemployment), known as the Phillips curve.
22 / 50
Original Phillips curve: unemployment-inflation tradeo↵?

William Phillips (1958) showed that there was a strong and negative
empirical relation between unemployment and inflation in the UK.
The basis for macroeconomic policy discussions of the time:
) Permanently lower unemployment at the cost of higher inflation?
Existence of this tradeo↵ also had theoretical grounding in the
dominant at the time Keynesian school of macroeconomics.
Keynes’s original theory was based on the assumption that nominal
wages Wt are fixed in the short run (an example of nominal rigidity)
-
-
E.g. assume Wt is set on the basis of past prices: Wt = APt-1 .

Wt
Firms equalize marginal product of labour with the real wage Pt , so
Wt APt-1 A
F 0 (Lt ) =
= = , F 0 > 0, F 00 < 0. (15)
Pt Pt 1 + ⇡t
) Higher inflation rate ⇡t implies higher output and employment.
23 / 50
A. W. Phillips’s original graph
Source: Phillips (1958)

24 / 50
The natural rate hypothesis
Friedman and Phelps argued in 1968 that the idea that nominal
variables could permanently a↵ect real variables was unreasonable.
Suppose expansionary policies try to maintain permanently high
inflation, low real wage, and hence high employment and output.
This would sooner or later lead workers and firms to learn to expect
high inflation and account for it when setting nominal wages.
In the long run, real wage and employment will return to their
natural levels, determined by the real rather than nominal forces.
Their arguments anticipated the failure of the Phillips curve during
the 1970s stagflation (high inflation and unemployment).
25 / 50
Empirical downfall of the Phillips curve
Unemployment and inflation in the United States, 1961-1995

Source: Romer (2018) textbook
26 / 50
Expectations-augmented Phillips curve
Friedman’s ideas could be roughly formulated as follows:
⇡t = ⇡et + (ln Yt - ln Ytn ), (16)
where ⇡et is expected inflation and Ytn is the natural level of output.
This is known as the expectations-augmented Phillips curve.
Positive relationship between inflation rate and output/employment,
but no permanent tradeo↵ that could be exploited by the policy.
Permanently high inflation cannot fool the public in the long run:
eventually ⇡et ⇡ ⇡t and output returns to its natural level, Yt ⇡ Ytn .
27 / 50
The accelerationist Phillips curve
If expectations are adaptive, e.g. ⇡et = ⇡t-1 , we get:
⇡t = ⇡t-1 + (ln Yt - ln Ytn ). (17)
This is the accelerationist Phillips curve that relates change in

inflation ⇡t - ⇡t-1 and output.
There is still a tradeo↵, but now output can be kept above the natural
rate only by permanently accelerating inflation (Friedman, 1968)
Predicts inflation inertia: for inflation to fall, there must be a period
with output below its natural level.
Hard to get inflation down quickly without inflicting unemployment.
Consistent with empirical evidence and intuition of many economists.
Still used in policy analysis and textbooks today.
28 / 50
Lucas critique as motivation for rational expectations
Motivating example:
A heavily guarded bank has never been robbed. Does it mean it can
safely eliminate the guards and save money on security?
No! It is the presence of security that prevents robbery.
Lucas (1976) Critique:
Changes in policy regimes will change peoples’ behaviour,
and so invalidate previously observed statistical regularities.
For useful policy analysis need ‘micro-founded’ models with ‘deep’
policy-invariant parameters that govern people’s behaviour.
Lucas critique led to a paradigm shift in macroeconomic theory
was targeted at the way Economics was done at the time, e.g. simple,
policy-invariant decision rules of mainstream Keynesian models.
justified the widespread use of rational expectations (RE) in
macroeconomic theory.
29 / 50
Lucas supply and policy ine↵ectiveness
Lucas (1972) derived an aggregate supply curve with RE:
⇡t = E[⇡t ] + (ln Yt - ln Ytn ). (18)
Can have Yt 6= Ytn only when there is an inflation surprise, ⇡t 6= E[⇡t ]

But since expectations are rational, these surprises must be random
and unpredictable based on publicly available information. Eshocks) 0 =
Thus, there is no room for systematic and predictable (Keynesian)

stabilization – this is the policy-ine↵ectiveness proposition.
But the Keynesians weren’t quite down and out yet.
30 / 50
New Keynesian (NK) economics: an overview
Agenda: develop micro-founded theory in which people are rational,

yet monetary policy still has systematic e↵ects.
) i.e. rebuild Keynesian economics, but robust to the Lucas critique.
Start with RBC, then add two key ingredients:
1. Nominal price and/or wage rigidities
2. Monopolistic competition (as in the Romer model last week)
Monopolistic competition is crucial:
It allows to model prices as purposefully set by firms.
Firms’ prices are above their marginal costs, so they will meet an
increase in nominal demand with more output at a given price.
The workhorse framework in modern macro and central banking.

We now turn to a basic NK model. We will not derive it in detail
(that’s ), but you must understand the big picture.
Need to understand the intuition
31 / 50
Households’ problem
no capital for simplicity
As in RBC, the representative household maximizes expected utility:
1
X ✓ ◆
t C1-
t L1+'
max E - t (19)
Ct ,Lt ,Bt
t=0 |1 {z
- } |1 +
{z'}
U(Ct ) V(Lt )
s.t. Pt Ct + Bt /(1 + it ) = Bt-1 + Wt Lt + Tt (20)
We assume CRRA utility with relative risk aversion > 0.

Ct is consumption index consisting of many di↵erent goods.
Pt is aggregate price level, i.e. weighted average price of Ct .
Wt is nominal wage., Tt is any other income and transfers.
it is the nominal interest rate, Bt is nominal bond holdings.
32 / 50
Demand for individual goods
Ct is given by a Dixit-Stiglitz CES aggregator over a continuum of

di↵erentiated goods i 2 [0, 1] (like in Romer model last week):
✓Z 1 ◆1/
Ct = Ct (i) di (21)
0
Optimal budget allocation yields the following demand for good i:

price for each individual
✓ ◆- 1-
1
Pt (i)
Yt (i) = Yt , where (22)
Pt
aggregate price index
Ct = Yt , assuming there is no capital and all output is consumed.

Pt (i) is the nominal price of good i, e.g. in dollars.
Pt (i)
Demand for good i negatively depends on its relative price Pt .
33 / 50
Firms, production, and marginal costs
Each good i produced by a separate firm, using only labour as input:
Yt (i) = At Lt (i)1-↵ ↵ 2 (0, 1), (23)
where At is exogenous aggregate productivity.

Decreasing returns to scale ) increasing marginal costs.
Wt
Given that the real wage is Pt , the real marginal cost is:
↵
Wt @Lt (i) Wt Yt (i) 1-↵
MCt (i) = = 1 (24)
Pt @Yt (i) Pt (1 - ↵)A 1-↵
t
MCt (i) increases in output Yt (i) and falls in productivity At .

Note: if we know firm i’s MCt (i), we can infer its level of output!
34 / 50
The benchmark: flexible prices

If nominal prices are flexible, firms can freely set them every period.
Monopolistic competition: di↵erentiation gives firms market power.
We saw last week that each firm facing demand in (22) would set a
constant markup 1/ over their (nominal) marginal cost Pt · MCt (i):
1
Pt⇤ (i) = Pt · MCt (i). (25)
Taking logarithm (we’ll use lowercase letters for logs of variables):
p⇤t (i) = µ + (pt + mct (i)), (26)
where µ ⌘ ln(1/ ) is the log desired markup.

All firms would set the same price, employment and output in a
symmetric equilibrium, so pt = p⇤t and mct = -µ at all times t.
35 / 50
Flexible prices and the natural level of output
Since the total amount of firms normalised to 1, aggregate output is

the same as that of an individual firm in this symmetric equilibrium.
This output is pinned down by firms’ real marginal cost mct , which
we know is always equal to -µ in equilibrium, i.e. is fixed.
The nominal price level pt is irrelevant for output determination!
We will refer to the level of output that would obtain under the
flexible-prices benchmark as the natural level of output ynt.
We could solve for ynt (although we won’t ) and show that it is the
function only of the ‘real’ forces in the model, similarly to RBC.
Here, only changes in technology at move the natural level of output.
) Having monopolistic competition is not enough for monetary shocks

to have real e↵ects in the model. Also need some nominal rigidities.
36 / 50
Introducing simple price rigidity: the Calvo (1983) model

Every period, each firm is able to reset its price with probability 1 - ✓.
Thus, only a fraction 1 - ✓ of firms get to update their prices in
period t, while fraction ✓ keep their old prices (think contracts etc.).
This assumption is neither realistic nor robust to Lucas critique:
) Fraction of firms that revise prices likely depends on conditions.
But it is a simple and tractable way to introduce price rigidity.
There are more convincing ways to do it, but they’re also a lot harder.
Then, the (log) aggregate price level is:

pt = (1 - ✓)p⇤t + ✓pt-1 , (27)
where p⇤t is the (log) optimal price set by firms able to change prices.
Pt
Aggregate inflation rate ⇡t ⌘ Pt-1 - 1 ⇡ pt - pt-1 is thus:
⇡t = (1 - ✓)(p⇤t - pt-1 ). (28)
37 / 50
Optimal price setting
With Calvo pricing, a firm’s price-setting problem becomes dynamic.

Indeed, the price p⇤t it sets at t may persist for a while, i.e. it may
still be active s periods in the future with probability ✓s .
But if the aggregate price level changes significantly by t + s, then
firm’s (log) relative price p⇤t - pt+s will di↵er considerably from 0.
This would significantly distort the demand for the firm’s good at
t + s, which, using (22) and taking logs, would be:
yt+s|t = yt+s - (p⇤t - pt+s )/(1 - ). (29)
But distorted demand = lost profits.

) Firms care about expected future inflation when setting prices today.
38 / 50
Optimal price setting (cont.)
While the firms’ price-setting problem is conceptually straightforward,

it is death by maths.
It can be shown, however, that firms optimally set:
In (Ptts.MCtfsit) Pt+s+mettsIt
=
1
X ⇥ ⇤
p⇤t = µ + (1 - ✓) ( ✓)s Et pt+s + mct+s|t (30)
s=0
Simple interpretation: desired markup µ over the weighted

average of current and expected nominal marginal costs.
Weights reflect discounting ( s ) and probabilities (✓s ) that the price
set today will remain e↵ective in period t + s.
39 / 50
New Keynesian Phillips Curve in terms of marginal cost
After A LOT of tedious algebra, the firms’ optimal price-setting

equation (30) can be aggregated to yield:
⇡t = Et [⇡t+1 ] + (mct + µ) (31)

comes from (30), but don't need to know how
where > 0 is a function of the underlying parameters, and mct is
the average real marginal cost in the economy.
(31) is the New Keynesian Phillips Curve in terms of marginal cost.
Remember: this was derived from firms’ forward-looking behaviour!
Interpretation: firms set higher prices today if
1 they expect high inflation next period when some of them will be
unable to reset prices.
2 Their marginal costs increase.
40 / 50
Output gap
mct + µ in (31) measures the deviation of firms’ average real

marginal cost mct from its level in the flexible-prices benchmark, -µ.
But if average marginal cost di↵ers, so does the aggregate output!
One can show that the deviation of the aggregate output yt from its
natural level is proportional to the deviation of mct from -µ:
yt - yn
t / mct + µ (32)
Makes sense: when mct = -µ, output is at its natural level, yt = yn

t.
ỹt ⌘ yt - yn
t is known as the output gap.
41 / 50
Key equation #1/3: New Keynesian Phillips Curve
Can use (32) to rewrite (31) in terms of the output gap to get the
New Keynesian Phillips Curve (NKPC) in its canonical form:
⇡t = Et [⇡t+1 ] + ỹt (33)
where  > 0 is a function of the underlying parameters.

One of the key building blocks of a basic modern NK model.
Results from purposeful price-setting decisions by firms, which adjust
prices in light of current and anticipated cost conditions (homework).
) Hence, NKPC captures the supply side of the economy.
42 / 50
Euler equation and aggregate demand

Households’ Euler equation for consumption takes the familiar form:

Pt
C- t = E t C- t+1 (1 + it ) (34)
|{z} | {z } Pt+1
U 0 (Ct ) U 0 (C )
| {z }
t+1
Rt+1
Replace Ct = Yt (market-clearing), and take log-linear approximation:

1
yt = Et [yt+1 ] - (it - Et [⇡t+1 ] -⇢). (35)
| {z }
rt
Here ⇡t+1 = PPt+1

t
- 1 is inflation rate, ⇢ ⌘ - ln is discount rate,
and 1/ measures the intertemporal elasticity of substitution.
Note the slightly di↵erent timing convention compared to RBC: rt
here is the expected real interest rate between t and t + 1.
43 / 50
Key equation #2/3: New Keynesian IS curve
Condition (35) would also hold in the flexible-price benchmark:
1
yn n
t = Et [yt+1 ] - (rn
t - ⇢). (36)
rn
t is the natural (real) rate of interest consistent with the
expected path of the natural level of output, driven by technology.
Note: rn n
t = ⇢ when there are no technology shocks and yt is constant.
Subtract (36) from (35) to get the NK IS curve in terms output gap:
1
ỹt = Et [ỹt+1 ] - (it - Et [⇡t+1 ] - rn
t) (37)
Remember: IS here comes from optimal intertemporal substitution

behaviour of households, so reflects the demand side of the economy.
44 / 50
Key equation #3/3: Taylor rule
To close the model, we need to know how the nominal interest rate it
is determined.
John Taylor suggested a simple rule in his 1993 paper:
it = 1% + 1.5 ⇡t + 0.5 ỹt . (38)
“Simple enough to put on the back of a business card!”

Taylor principle: stabilising monetary policy requires that the
nominal rate is raised more than one-for-one with inflation.
We will look into this in detail next week!
45 / 50
Taylor rule vs. Fed Funds Rate
Source: Bernanke (2015)
46 / 50
The simple 3 equations model

These are the key equations, need to know the intuition, where they come from and how to use them;
At exam, you’ll never be asked to derive these equations
⇡t = Et [⇡t+1 ] + ỹt + ut Philips Curve

1
ỹt = Et [ỹt+1 ] - (it - Et [⇡t+1 ] - rn t ) + "t IS Curve (39)
it = rn
t + ⇡ ⇡t + y ỹt + vt Taylor Rule
Here we added (i) a cost-push shock ut , (ii) a demand shock "t ,

and (iii) a monetary policy shock vt .
Trick question: where is the technology shock here?
Captured by the natural real rate rn n
t , related to yt and technology.
Can solve for equilibrium ⇡t , ỹt and it as functions of the exogenous

shocks once we know their distributions (homework).
47 / 50
Success: the e↵ects of a monetary policy shock
Source: Gali (2008)
48 / 50
Some of the difficulties of the NK model
1. The standard assumption of time-dependent Calvo pricing is

contradicted by microeconomic evidence on price adjustments.
2. The basic NK model makes some extreme predictions, e.g. about the
power of forward guidance by the Central Bank (homework).
3. Empirical evidence provides little support for NKPC, and points in the
direction of inflation inertia, i.e. it being costly to reduce (see sl. 28)
Yet NKPC predicts that anticipated disinflation is associated with an
output boom.
Indeed, rearrange it to get:
1- 
Et [⇡t+1 ] - ⇡t = ⇡t - (yt - yn
t) (40)
1-
is close to 1, so ⇡ 0 and Et [⇡t+1 ] < ⇡t () yt > yn
t.
49 / 50
We have successfully derived a micro-founded New Keynesian model:

Nominal disturbances have real e↵ects.
Inflation is the result of forward-looking price setting behaviour of firms.
Demand comes from forward looking optimising households.
While the basic NK model has important shortfalls, there is a very

active ongoing work to extend the model and fix them.
Of course, it may well happen that the paradigm will shift away from
NK in the coming years.
Still, if you go to a research conference at a central bank nowadays,
95% of presentations will use this framework.
Next week: we use this to understand monetary policy.
50 / 50
Lecture 3: Monetary Policy
Artur Doshchyn
Hilary Term 2023

Introduction Monetary policy in the NK model Dynamic inconsistency ZLB and the liquidity trap
Why we need stabilization monetary policy: inflation
There is a wide agreement that the goal of monetary policy is to keep

inflation low and stable, while also minimizing fluctuations of output.
First, why stabilize inflation?
Since individual prices are not adjusted continuously, high inflation may
lead to relative price variability and misallocation.
Individuals may have trouble accounting for high inflation and make
mistakes when making long-run investments and saving for retirement.
High inflation is typically also volatile, creating uncertainty and
discouraging investments.
Symptom of a government functioning badly.
Simply, people don’t like it!
Okun (1975): a policy that reduces the length of a mile by a fixed

amount each year might have few e↵ects on real decisions, but can
cause considerable unhappiness.
1 / 45
Why we need stabilization monetary policy: output
Lucas (1987) showed that in a representative agent setting, the

welfare gain from stabilizing consumption around its trend is small:
⇡ 0.06% of average consumption in $ terms.
There are, however, important reasons why economists believe that
there is an important role for output stabilization:
Individuals may dislike risk much more than Lucas assumes,
substantially rising the welfare costs of short-run fluctuations.
Big welfare losses may be stemming not from consumption, but
employment variability.
Macroeconomic stability promotes investment, including R&D, and
thus rises long-run growth.
Output losses in recessions may not be fully o↵set by comparable
periods of above-normal output.
2 / 45
Your favourite New Keynesian (NK) model

The New Keynesian Phillips Curve (NKPC):
⇡t = Et ⇡t+1 + ỹt (1)

forward looking
price setting is
The New Keynesian IS (‘intertemporal substitution’) curve:
1
ỹt = Et [ỹt+1 ] - (it - Et [⇡t+1 ] -rn ). (2)
| {z } t
rt
ỹt = yt - yn
t is the output gap, i.e. deviation of the actual output
from its natural level yn
t , due to nominal rigidities.
rn
t is the natural real interest rate.
yn n
t , rt would obtain in the flexible-price benchmark, so are only
a↵ected by real forces (e.g. technology), but not any monetary shocks
Fluctuations in yt = fluctuations in ỹt + fluctuations in yn
t.
3 / 45
First attempt at stabilizing monetary policy
For now assume for simplicity that there are only real technology
shocks to the economy, i.e. shocks that move yn n
t and rt .
Suppose central bank (CB) objective is ỹt = ⇡t = 0 at all times.

The IS curve implies that this requires CB to set
it = rn
t 8t. (3)
But is it enough?
No! While ỹt = ⇡t = 0 is an equilibrium, there are in fact infinitely
many other equilibria that satisfy NKPC and IS when CB uses (3) )
4 / 45
Self-fulfilling movements in output and inflation

3e5uxt
2x
Suppose inflation and output gap spontaneously jump up and agents

expect them to gradually return back to normal.
-> assume policy always sets a down
Consistent with IS: with nominal rate it set to rn t , positive expected
inflation (Et [⇡t+1 ] > 0) lowers the real rate rt below rn t , and thus
implies falling output gap (Et [ỹt+1 ] < ỹt ), as assumed.
Consistent with NKPC: positive and falling inflation (⇡t > Et [⇡t+1 ])
requires above-normal output (ỹt > 0), as assumed.
Can derive the appropriate speed of return to zero and the relationship
between output and inflation such that equations are satisfied exactly.
Generally, can show that there are infinitely many equilibria of the
form ⇡t = t Z and ỹt = c t Z, < 1, that di↵er only by constant Z.
This is known as indeterminacy, and can lead to spontaneous,
self-fulfilling (‘sunspot’) fluctuations in output and inflation.
5 / 45
Taylor rule and Taylor principle
Consider a variation of the Taylor rule we have seen last week:
it = rn
t + ⇡ ⇡t + y ỹt (4)
When y = 0, ⇡ > 1 ensures that sunspot equilibria are ruled out:

Try to repeat the experiment: suppose both ⇡t and ỹt spontaneously
jump up, and are expected to gradually return back to normal.
CB now raises the nominal interest it more than one-to-one with
inflation, so the real interest rate increases.
But then the IS curve implies that the output gap ỹt should be rising
rather than falling, a contradiction, so such deviations can’t occur.
Taylor principle: Monetary policy must react sufficiently strongly

such that the real interest rate rises in the face of increasing inflation.
6 / 45
Taylor principle in the NK model formally

Can use the Taylor rule and write the NKPC-IS system in matrix form:
 
ỹt ỹ
= A Et t+1 , (5)
⇡t ↓ ⇡t+1
will not be examined to derive A
where A is a matrix of coefficients that are functions of parameters.
By forward substitution s - 1 times, this implies:
Xt a
=
X++ s =>x+es
Xt
=  
as
ỹt ỹ
a>1 => UXt, XttS90 as 5-y = A Et t+s
s
(6)
all => ftts ->
as 5- unless X++S
⇡t =0
⇡t+s
Can show that if policy responds to deviations sufficiently strongly, i.e.
1-
⇡ + y >1 (7)

then eigenvalues of A are < 1 in absolute value and lims!1 As = 0.
But then the economy explodes unless ỹt = ⇡t = 0 8t, which is thus
the unique equilibrium – CB eliminates self-fulfilling fluctuations.
7 / 45
Taylor principle in the NK model: connecting the dots
What is crucial for stabilization is that by strongly raising the nominal

rate in the face of inflation, CB can increase the real interest rate.
As we have seen, this rules out multiple equilibria, and hence
eliminates self-fulfilling fluctuations.
no inflation & output gap
The actual interest rate in equilibrium never departs from it = rn
t.
But the threat to raise the interest rate in response to increases in

inflation stops them from occurring, so never needs to be carried out.
8 / 45
Aside: Taylor principle in backward-looking models

Consider a super simple old-style backward-looking PC and IS curves:
⇡t = ⇡t-1 + ỹt (8)

ỹt = -⌘(it - ⇡t ) + "t (9)
| {z }
rt
Consider positive demand shock "t :

"t " ) ỹt " ) ⇡t " ) ⇡t+1 " ) ⇡t+2 " ...
In each period reinforced by falling rt : ⇡t " ) rt # ) ỹt " ) ⇡t ".
Inflation spirals up!
Taylor principle arises again: to o↵set the inflationary pressure and

stabilise the economy, CB should increase nominal rate it more than
1-to-1 with inflation, and thus increase the real interest rate.
This is the familiar ‘lean-against-the-wind’ logic one often hears.
But it is very di↵erent from how modern NK models work.
9 / 45
Estimating interest-rate rules: Clarida, Gali, Gertler (2000)
US experienced a period of high and volatile inflation and

macroeconomic instability from the late 1960s to early 1980s.
To see if monetary policy was to blame, Clarida, Gali, and Gertler
(2000) estimate the interest-rate rule similar to (4).
Specifically they focus on two periods: the ‘pre-Volcker’ period
1960-1979, and ‘Volcker-Greenspan’ period 1979-1996.
They estimate ⇡ = 0.83(0.07) in the first period, but
⇡ = 2.15(0.40) in the second period.
That is, the Federal Reserve on average cut real interest rate when
inflation rose in the first period!
) highly accommodative monetary policy can account for high inflation
and instability in the 70s, which ended with Paul Volcker’s disinflation.
10 / 45
Inflation and the real rate before and after Volcker

When inflation is increasing, real interest rate is decreasing, even to the negative territory.
Taylor principal does work. Eventually, interest rate brings down the inflation rate
Source: Clarida, Gali, and Gertler (2000)

11 / 45
The divine coincidence
So far: only real shocks to technology, can achieve ⇡t = ỹt = 0 8t.

We can also add demand shocks "t to the IS equation and monetary
policy shocks vt to the Taylor rule:
1
ỹt = Et [ỹt+1 ] - (it - Et [⇡t+1 ] - rn
t ) + "t (10)
it = rn
t + ⇡ ⇡t + y ỹt + vt (11)
Both demand side and monetary shock enters the IS equation
As all these shocks enter on the demand side of the economy via the
IS curve, they move output gap and inflation in the same direction.
) Appropriate monetary policy can stabilize both inflation and output!
Just set ⇡ ! 1 to perfectly stabilise both (homework last week).
This is theoretical result when there is no conflict between output and

inflation stabilisation is known as the divine coincidence.
12 / 45
Breaking the divine coincidence: backward-looking forces
However, in practice there appears to be a tradeo↵ between the two

objectives, and there are several reasons why divine coincidence fails.
Firstly, the divine coincidence arises partly due to the pure
forward-looking nature of the New-Keynesian model
) there are no backward-looking forces to keep inflation away from zero.
As we have seen last week, however, there is evidence for inflation

inertia, namely it being hard to reduce inflation w/o a fall in output.
Once we accept some degree backward-looking behaviour, e.g. a
hybrid Phillips curve of the form
⇡t = b ⇡t-1 + f Et [⇡t+1 ] + ỹt , (12)
the divine coincidence no longer holds.
13 / 45
Breaking the divine coincidence: cost-push shocks
Second, consider cost-push (supply) shocks in the NKPC:
⇡t = Et [⇡t+1 ] + ỹt + ut (13)
ut captures shocks that a↵ect firms’ marginal costs & markups, e.g.
movements in wages due to frictions in the wage contracting process,
or changes in the market structure that a↵ect firms’ market power.
They move output gap and inflation in the opposite directions.

Policymaker now faces an inflation-output volatility tradeo↵.
) Larger ⇡ reduces inflation, but at the cost of deeper recession!
(again, last week’s homework)
This is precisely the dilemma many central banks are facing now...
as the recent surge in inflation is oft attributed to rising energy prices.
14 / 45
Breaking the divine coincidence: market imperfections
Lastly, we’ve been assuming that the central bank’s objective is to

achieve yt = ynt , i.e. the natural level of output under flexible prices.
However, the yn
t is almost surely not Pareto efficient:
Remember, that there is monopolistic competition in the NK model;
There are also likely many other non-Walrasian features in reality.
⇤
Welfare Theorem does not apply, so yn t 6= yt , where
⇤
yt is the potential, or welfare-maximizing level of output that
would obtain if all market imperfections were removed.
But then CB’s output objective should be to achieve yt = y⇤ , not yn

t.
Conflicts with inflation objective, since ⇡t = 0 only when yt = yn

t.
Due to market imperfections, in reality y⇤t > yn

t almost surely.
As we will see, this can lead to the inflation bias in CB’s policy.
15 / 45
Connecting market imperfections and cost-push shocks
The divine coincidence fails even if we were to assume that yn

t equals
⇤ ⇤
yt on average, and only allow random deviations yt - yt .n
To see this formally, rewrite NKPC in terms of the welfare-relevant

output gap yt - y⇤t :
⇡t = Et [⇡t+1 ] + (yt - y⇤t ) + (y⇤t - yn ) (14)

| {z t }
ut
But ut ⌘ (y⇤t - yn
t ) is nothing else than a cost-push shock!
The oil crisis in the 1970s is often cited as a classic cost-push shock,
since the creation of OPEC (change in the market structure) allowed
oil producers to restrict supply and charge higher prices (markups ").
16 / 45
Policy optimality
Given that there is a conflict in objectives, the central bank should
find an optimum trade-o↵ between stabilizing output and inflation.
Woodford (2003) derived a quadratic approximation of consumer
welfare losses in the NK model that CB should strive to minimize:
1
1X ⇥ ⇤
Lt = t
Et ⇡2t+s + ↵y ỹ2t+s , (15)
2
s=0
where ỹt is redefined to be the welfare-relevant output gap yt - y⇤t ,

and ↵y is the relative weight of output in the loss function.
When all the shocks are i.i.d., this simplifies to var(⇡t ) + ↵y var(ỹt ).
There is a large literature on the optimal monetary policy.
As a start, one can assume Taylor rule and derive optimal ⇡ that
minimizes the objective function (homework).
17 / 45
We analysed the principles of stabilization monetary policy in the New

Keynesian framework.
Central bank should react strongly in the face of inflation in order to
stabilize the economy.
In addition, CB faces a trade-o↵ between stabilizing output and
inflation, so should optimally calibrate its responses.
But so far we assumed that CB has no problem committing to
optimal policy.
However, as we shall now see, CB may have incentives to deviate
from the set policy, a problem known as dynamic inconsistency.
18 / 45
Dynamic inconsistency and inflation bias: introduction

INOT EXAMINABLE]
Actual policy often appears to be far from optimal, e.g.

policymakers’ failure to respond to shocks during the Great Depression
high inflation in many industrialized countries in the 70s.
Moreover, departures from optimal policy do not seem random:
) ‘Too high’ inflation is far more common than ‘too low’ inflation.
Kydland and Prescott (1977) show that the inability of policymakers
to commit to low-inflation policy can lead to an inflation bias:
When inflation expectations are low, the policymaker would prefer to
pursue expansionary policy to boost output and welfare.
But the public anticipates policymaker’s actions.
In equilibrium, this results in a high inflation and no change in output.
Ex ante, the policymaker would prefer to commit to low inflation, but
ex post finds it optimal to deviate – he is not dynamically consistent.
19 / 45
Kydland and Prescott (1977) model assumptions

A very simple model.
Aggregate supply is given by the Lucas supply curve:
y = yn + b(⇡ - ⇡e ), (16)
where ⇡e = E[⇡] is inflation expected by rational agents.

NB: Similar results can be obtained with NKPC and hybrid Phillips curves.
Due to market imperfections, the natural level level of output yn is
below the Walrasian (welfare-maximizing) level of output y⇤ .
The policymaker chooses inflation ⇡ to minimize the welfare losses:
1 1
L = (y - y⇤ )2 + a(⇡ - ⇡⇤ )2 , y⇤ > yn , a > 0, (17)
2 2
where a is the relative weight on inflation, and ⇡⇤ is target inflation.
20 / 45
Benchmark case: commitment
Suppose the central banker can make a binding commitment about

what inflation will be before expected inflation is determined.
Since the commitment is binding, and expectations are rational,
expected inflation equals actual inflation: ⇡e = ⇡.
Supply curve (16) then implies that the actual output equals its
natural level: y = yn .
Policy maker’s problem simplifies to choosing inflation ⇡ to minimize
1 n 1
(y - y⇤ )2 + a(⇡ - ⇡⇤ )2 (18)
2 2
The solution is simply ⇡ = ⇡⇤ .
21 / 45
Discretionary policy
Now suppose that instead the policymaker has discretion and can
choose inflation after agents’ form their expectations ⇡e .
Taking ⇡e as given, the policymaker solves (substitute (16) into (17)):
1 n 1
min (y + b(⇡ - ⇡e ) -y⇤ )2 + a(⇡ - ⇡⇤ )2 (19)
⇡ 2 | {z } 2
y
Taking the first order condition and solving it for ⇡ yields:

b b2
⇡ = ⇡⇤ + (y⇤
- yn
) + (⇡e - ⇡⇤ ) (20)
a + b2 a + b2
Inflation bias: policymaker chooses inflation greater than ⇡⇤ !
Intuition. Suppose the public expects ⇡e = ⇡⇤ . Then
the marginal cost of slightly higher inflation is zero,
but the marginal benefit of the resulting higher output is positive.
22 / 45
Equilibrium
Since expectations are rational and there is no uncertainty, in
equilibrium the expected inflations equals actual inflation, ⇡e = ⇡EQ .
Substituting this into the policymaker’s reaction function (20) and
solving for the equilibrium inflation rate, we get:
b ⇤
⇡EQ = ⇡⇤ + (y - yn ) > ⇡⇤ . (21)
a
Because expectations are correct in equilibrium, we have yEQ = yn .
Policymaker’s discretion increases inflation, without a↵ecting output!
The problem is dynamic inconsistency:
Ex ante, policymaker knows that it is best to commit to ⇡⇤ .
But ex post, if ⇡e = ⇡⇤ , reneging and raising inflation raises welfare.
The key to the result above is that the public knows that the
policymaker has discretion, and anticipates its actions.
23 / 45
Addressing dynamic inconsistency
Large literature on dynamic inconsistency and the ways to address it.

One way out is to simply legislate that monetary policy has to be
determined by the rules rather than discretion.
However, no rule can account for unprecedented events like the 2008
Financial Crisis or 2020 Covid recession.
Many countries where the policy is not made according to fixed rules
have low inflation – there must be other ways to alleviate the problem.
There several major approaches, the most prominent two being:
Delegation (Rogo↵, 1985): appoint a conservative central banker who
is tough on inflation – today!
Reputation building (Backus and Driffill, 1985; Barro, 1986): central
banker is concerned with their reputation as being tough on inflation.
These explanations are not only theoretically sound, but also realistic.
24 / 45
Delegation
Idea: appoint conservative central bankers who are known to

particularly dislike inflation, like Paul Volcker!
Suppose policy is determined by an individual whose loss function is:
1 1
L = (y - y⇤ )2 + a 0 (⇡ - ⇡⇤ )2 , y⇤ > yn , a 0 > 0, (22)
2 2
where a 0 di↵ers from the weight a that the society places on inflation.
Solving for the equilibrium as before, we get:
b ⇤
⇡EQ = ⇡⇤ + (y - yn ), yEQ = yn . (23)
a0
Equilibrium inflation is lower when the central banker dislikes inflation
more, i.e. a 0 is higher!
25 / 45
Social welfare consequences of delegation
Because output is unchanged, but the equilibrium inflation is lower

and closer to ⇡⇤ when a 0 increases, social welfare improves.
Formally, the term a(⇡ - ⇡⇤ )2 in the social welfare loss function falls.
However, so far we assumed no uncertainty, and thus abstracted from
the stabilization role of monetary policy.
In fact, Rogo↵ shows that when monetary policy is controlled by
someone whose preferences between output and inflation stabilization
di↵er from the society’s, they may not respond optimally to shocks.
Therefore, there is a trade-o↵: more conservative policymaker
stabilizes average inflation, but responds worse to disturbances.
One of the questions in the homework develops this idea formally.
26 / 45
Dynamic inconsistency: conclusion
Dynamic inconsistency arises in many other contexts, including

monetary policy, fiscal policy, law enforcement etc.
For example, in the NK model, the fact that y⇤t > ynt creates
incentives for the policymaker to boost output above its natural level.
NKPC implies that if CB can induce agents to expect negative inflation
next period, Et [⇡t+1 ] < 0, then yt rises at t.
But at t + 1, CB would prefer to surprise agents and deliver zero
inflation instead, thus reaching both its output and inflation targets.
With rational agents, CB cannot do this consistently.
But the fact that inflation rate CB wants agents to expect di↵ers from
the rate it wants to deliver is another example of dynamic inconsistency
A simpler example: policymaker may want agents to expect that

individuals who break the law will be punished harshly, yet there are
only costs and no benefits of delivering harsh punishments ex post.
27 / 45
The Zero Lower Bound (ZLB) on the nominal interest rate
So far we have presumed that CB can set any nominal interest rate.
But because money (e.g. cash) earns a nominal return of zero, no-one
would buy an asset o↵ering a negative nominal return.
As a result, the nominal rate cannot fall (far) below zero.
) This is known as the Zero Lower Bound or ZLB.
Huge problem during and after the 2008 Financial Crisis:
Many CBs lowered interest rate targets to zero for extended periods.
By Taylor rule, target fed funds rate should have been -4% and lower.
We will now consider a simple model by Krugman (1998).

It shows how the ZLB constraint can lead to a liquidity trap.
⌘ situation in which increases in the current money supply have no e↵ect
on the economy, rendering current monetary interventions ine↵ective.
28 / 45
The 2008 Crisis: ZLB bites
Source: Rudebusch (2009)

29 / 45
Krugman (1998) model setup

A very simple, yet micro-founded general equilibrium model.
) most of its main implications apply to the NK models as well!
No uncertainty or production, competitive markets, flexible prices.
) By construction, monetary policy cannot influence output here.
But the model still allows us to study the ability of monetary policy to
a↵ect the nominal aggregate demand (AD).
Can introduce price stickiness, s.t. AD a↵ects quantities (homework).
A large number (unit mass) of identical household who maximize:

1
X
t
U= at ln Ct , 2 (0, 1), at > 0 8t. (24)
t=0
Deterministic process at drives changes in the demand for goods.

In every period, each household receives an endowment of y units of
non-storable consumption good.
30 / 45
Krugman (1998) model setup (cont.)

Two assets: money and bonds that pay nominal interest it > 0.
Households and government trade money and bonds at the beginning
of each period t (households can also issue bonds, but not money).
Two critical assumptions to generate the demand for money:
1. Households cannot consume their own endowments or barter: they
must use money to purchase consumption goods from others.
2. Cash-in-advance (CIA) constraint: to buy goods in period t, agents
can only use their existing money holdings Mt , but not new money
received from selling their period-t endowments (e.g. trade takes time).
Formally, if Pt is the nominal price of goods, the CIA constraint is:

< total money stock
Mt
Ct 6 . (25)
Pt
31 / 45
Timing of events within one period t
1 2 3
1. Households start with their 1. Households purchase 1. Interest it on bond

holdings of money from goods, subject to the holdings is paid.
t - 1. CIA constraint 2. Government
2. Households and government implements any
Mt
trade bonds for money. Ct 6 . taxes or transfers
Pt necessary to satisfy
3. This is where any monetary
policy interventions occur its budget constraint.
2. Households sell their
(e.g. gvt sells bonds ) endowments.
reduces money supply).
4. Let Mt be a representative
household’s holdings of
money after these trades.
32 / 45
Household optimality conditions

If it > 0, a representative household will hold just enough money to
pay for planned good purchases, and otherwise invest in bonds. Thus:
Mt
Ct = if it > 0 (CIA binds). (26)
Pt
But when it = 0, the household is indi↵erent between money and
bonds, and will hold at least enough money to pay for goods. Thus:
Mt
Ct 6 if it = 0 (CIA slack). (27)
Pt
The household will also satisfy the usual Euler equation:
at at+1
= (1 + it )Pt /Pt+1 . (28)
Ct Ct+1 | {z }
1+rt Investors care about real, not nominal
interest rate
Because in symmetric equilibrium Ct = y in all periods, this yields:
at Pt+1
it = - 1. (29)
at+1 Pt
33 / 45
W
The steady state from period 2 onwards

Consider economy with fixed money stock M⇤ and demand shifter a⇤
If the price level is also constant at P⇤ , then (29) simplifies to
1
i⇤ = - 1 > 0. (30)
In this case, by (26), households satisfy their CIA constraint with

equality, so the price level is indeed constant in equilibrium:
M⇤
P⇤ = . (31)
y
ECt
=
Thus there is a conventional steady state equilibrium with a positive

nominal interest rate and price level proportional to money stock M⇤ .
To greatly simplify the analysis, we assume that the economy is this
steady state from period 2 onwards, and then focus on period 1.
34 / 45
Equilibrium conditions in period 1

1. In period 1, (29) becomes
a1 P ⇤
i1 = - 1, (32)
a⇤ P1
which gives a downward-sloping CC curve in the (P1 , i1 ) space.
Intuitively, higher P1 implies lower inflation in period 2, so the nominal
interest rate i1 must fall to keep the real interest rate r1 unchanged
) then households willingly consume y in both periods in equilibrium.
2. Combine CIA conditions (26)-(27) with market clearing Ct = y to get
M1
= y if i1 > 0
P1 (33)
6 M1
y if i1 = 0,
which is an MM curve: horizontal at i1 = 0, and vertical when i1 > 0
Intuition: the opportunity cost of holding money is 0 when i1 = 0.
35 / 45
Normal equilibrium in the Krugman model with i1 > 0

Equilibrium occurs when both conditions are satisfied, i.e. where the
CC (consumption choice) and MM (money market) curves intersect.
i1
MM
0
CC
P1
36 / 45
E↵ect of a monetary expansion when i1 > 0
Consider the e↵ect of an increase in the money stock, M1 .

M1 enters the equation for the MM curve (33): an increase in M1
raises M
y and hence moves the vertical portion of MM to the right.
1
The CC curve is not a↵ected, as M1 does not enter its equation (32).
When i1 > 0, monetary expansion has a conventional e↵ect of
raising aggregate demand (AD).
M1
Specifically, P1 = y rises proportionally with M1 , and i1 falls.
Because prices are flexible, consumption and real interest rate are
unchanged, so the fall in i1 reflects the fall in expected inflation.
) But in an extension with sticky prices, the raise in aggregate demand
would also a↵ect real quantities (homework).
37 / 45
Illustration: monetary expansion when i1 > 0

i1 Conventional view of monetary policy:
MS increases -> lower interest rate ->
MM higher price level
E
E’
0
CC
P1
38 / 45
Liquidity trap
Now consider the e↵ect of a monetary expansion, i.e. an increase in

M1 , when the initial nominal interest rate is already zero, i1 = 0.
The change has no e↵ect on aggregate demand at all!
The intersection of CC and MM curves happens at the same point as
before the expansion - on the horizontal part of the MM.
With i1 = 0, the opportunity cost of holding money is zero, so once
agents meet their liquidity needs, i.e. satisfy CIA, money and bonds
are perfect substitutes, so exchanging one for another has no e↵ect.
Such situation is therefore known as a liquidity trap.
In extensions with sticky prices, monetary expansions would fail to
increase output (homework).
39 / 45
Illustration: liquidity trap

i1
MM
E, E’
0
CC
P1
40 / 45
Addressing a liquidity trap
It would be imprecise to say that monetary policy is powerless in a

liquidity trap: rather, changes in current money supply are ine↵ective.
Consider an increase in M⇤ – the expected money supply once the
economy is out of the liquidity trap in period 2.
This raises the steady-state price level P⇤ , which enters the expression
for the CC curve (32), which therefore shifts up/to the right.
The price level P1 rises, and if the shift is large enough, i1 too!
Intuitively, with i1 stuck at 0, increasing P⇤ without a change in P1
would raise inflation and decrease the real interest rate r1 .
) Demand for goods would exceed supply at time 1, which cannot happen
) P1 must also increase in equilibrium, so that the real rate is unchanged.
41 / 45
E↵ect of anticipated future monetary expansion at ZLB

credible policy
i1
MM
QE let CC shifts upward.

Only expectation changes, so CC
(demand) increases while MM remains
stable.
If nominal interest is at ZLB, QE is still

effective since it will boost CC, nominal ir
may not increase, but P increases, acting
as a demand side stimulus. In the case of
price rigidity, output will increase to the
same level of price change.
E’
0
E
CC
P1
42 / 45
Policies to raise expected inflation in a liquidity trap
Way out of a liquidity trap: change expectations about future policy.

Krugman proposed adopting a permanently higher inflation target to
boost expectations of future price levels and stimulate the economy.
Eggertsson and Woodford (2003): even better would be not to aim
for higher inflation always, but only following periods when ZLB binds.
A variation on this theme is price-level targetting:
Target not the inflation rate, but the path of the aggregate price level.
Since inflation is likely to fall during a ZLB episode, CB would be
expected to o↵set this shortfall by a period of above-normal inflation in
order to return to the target price level path, stimulating the economy.
While this policy is not optimal, it is simple to communicate.
Lastly, during the 2008 Crisis many CBs adopted forward guidance:
promise to keep interest rates low for a while even after the recovery.
43 / 45
Other policies to stimulate the economy in a liquidity trap
All the measures to raise inflation expectations face an important

challenge: dynamic inconsistency (yes, again!)
Optimal to promise high future inflation in a liquidity trap, but once
the economy is out of it, high inflation is no longer desirable.
CB’s promises thus may not be credible.
Other potential policies include possible fiscal interventions

(targetting a1 directly) and quantitative easing (QE).
QE used to simply mean conventional open market operations of
buying short-term government debt and increasing money supply.
Today QE involves central banks buying assets (e.g. mortgage-backed
securities) for which interest rates are still positive, thus lowering
rates on those assets and producing a rightward shift in the CC curve.
QE increases supply and thus lower real interest rate
44 / 45
ZLB: conclusions
We have seen how a ZLB constraint can lead to a liquidity trap - a

situation in which current monetary expansions are impotent.
Increasing inflation expectations helps boost the economy at ZLB.
But raising them may be challenging.
ZLB is no longer a theoretical curiosity: estimated costs of the ZLB

during 2009-2012 were about 3% of cumulative GDP, or $1.7 tr.
In addition, there is considerable evidence that normal real interest
rates are significantly lower today than in the past.
) For a given inflation target, normal nominal interest rates are also lower
) increasing the likelihood of hitting ZLB more often going forward.
How to make liquidity traps occur less often, and how to make their
consequences less severe is a hugely important area of research.
45 / 45
Lecture 4: Unemployment
Artur Doshchyn
Hilary Term 2023

Introduction Efficiency wages Shapiro-Stiglitz model Search and matching
A photo from the Great Depression in the 1930s

US unemployment reached 25%
1 / 44
Introduction
We say there is unemployment when there are people who are not
working, yet actively want to work in jobs like those held by
individuals similar to them at the wages those individuals are earning.
One of the central subjects in macro, yet the models we have
considered so far had little to say about it.
There are two main groups of issues we are interested in:
1. The determinants and consequences of average unemployment, e.g.
Why does the labour market not clear, i.e. why do wages not fall in the
face of significant unemployment?
Why does unemployment vary across countries and over time?
What are the welfare consequences of normal unemployment?
2. Cyclical behaviour of the labour market, e.g.

Why does employment fluctuate a lot more than real wage?
Why do firms lay o↵ workers in downturns rather than rely on
work-sharing sharing arrangements?
2 / 44
Failure of Walrasian models of the labour market
Recall that in the RBC and New Keynesian models we have

considered so far the labour markets are Walrasian:
Wage always clears the market.
Individuals are always on their optimal labour supply.
There is no unemployment, only choices to consume more leisure.
Fluctuations in employment in these models reflect willingness of
RBC/NKaggregatein reality
people to substitute labour and leisure across periods.
xW L
-
Yet empirical studies find little evidence of significant intertemporal
:F1cy
substitution, and point to inelastic individual labour supply.
This mechanism predicts counterfactually large fluctuations in the real
wage, but much less volatility in employment than in the data.
Key question: Why do shifts in labour demand lead to large

movements in employment, and only small changes in the real wage?
3 / 44
The game plan
We will consider three highly influential models of the labour market:

1. Efficiency wage theory (warm-up)
2. Shapiro-Stiglitz model, which formally explores the deeper reasons
for efficiency wages.
3. Search and matching model.
The first two are examples of the traditional approach of modelling

the labour market within the standard supply and demand framework.
The last one is an example of the modern approach.
) Focus on the heterogeneity among workers and jobs, and the costly
process of job search and recruitment.
4 / 44
Efficiency wage theories: overview
The key idea of efficiency wage theories is that there are benefits of
paying higher wages to employees.
Among suggested reasons, the following received the most attention:
1. Better nourishment, and thus productivity.
2. Incentive to exert high e↵ort when firms cannot monitor workers
perfectly, as in the Shapiro-Stiglitz (1984) model – later today.
3. Higher wages can attract workers of higher ability.
4. The fair wage-e↵ort hypothesis due to Akerlof and Yellen (1990):
high wage can build loyalty and hence induce e↵ort – homework.
) extensive evidence that workers’ e↵ort is a↵ected by such feelings as
anger, jealousy, and gratitude.
We begin with a simple efficiency wage model due to Solow (1979).
5 / 44
Setup
No capital for simplicity, labour is the only factor of production.
There is a large number N of firms.
A representative firms maximizes profits:
⇡ = F(eL) - wL, F 0 (·) > 0, F 00 (·) < 0, (1)
where e is workers’ e↵ort, so eL is e↵ective labour.

Assume e↵ort e is an increasing function of the wage:
e = e(w), e 0 (·) > 0. (2)
For now, we are interested in the implications, and not precise reasons.
There is L̄ workers, each supplying 1 unit of labour inelastically
) i.e prepared to work at any wage.
6 / 44
V
The firm’s problem

A representative firm solves
max F(e(w)L) - wL. (3)

L,w
There can be two cases:

1. There are unemployed workers and the firm can choose its wage freely.
2. There is zero unemployment, so the firm must pay at least the wage
paid by other firms to attract any workers.
In either case, the firm is free to choose its employment level.

FOC w.r.t. L yields:
w
F 0 (e(w)L) = . (4)
e(w)
I.e. marginal product of e↵ective labour equals its unit cost, w/e(w).
Note: When a firm hires a worker, it gets e(w) units of e↵ective labour.
7 / 44
The efficiency wage
When the firm is unconstrained and sets wage freely, FOC w.r.t. w is
F 0 (e(w)L)Le 0 (w) - L = 0. (5)
Substitute for F 0 (e(w)L) from (4) and rearrange to get:
decu)/ecw) we 0 (w)
= =1 (6)
du, w e(w)
I.e. at the optimum, elasticity of e↵ort w.r.t. wage is 1. Intuition:

The firm wants to hire e↵ective labour, eL, as cheaply as possible.
(6) defines the efficiency wage that minimizes the unit cost of
e↵ective labour, i.e. solves minw w/e(w).
Put di↵erently, optimal w maximizes e↵ort per dollar spent, e(w)/w
8 / 44
The determination of the efficiency wage: illustration
ecul= an => max

a elu)
=
9 / 44
Equilibrium
Let L⇤ and w⇤ denote the values that satisfy conditions (4) and (6).
Since all firms are identical, the total labour demand at w⇤ is NL⇤ .
If NL⇤ < L̄, then there is positive unemployment in equilibrium:
Firms are free to set wages, so the equilibrium wage is simply w⇤ .
At this wage, employment is indeed given by the labour demand, NL⇤ .
L̄ - NL⇤ workers are unemployed.
If NL⇤ > L̄, there is full employment in equilibrium:
At w⇤ , labour demand would exceed supply,
so the wage is bid up above w⇤ in equilibrium until NL(w) = L̄.
Firms are constrained, and unable to reduce the wage to w⇤ .
10 / 44
Implications
1. The model implies the possibility of involuntary unemployment:

When NL⇤ < L̄ and w = w⇤ , there are workers who want to work at
the prevailing wage, yet cannot find employment.
The wage does not fall to equilibrate labour supply and demand, since
it reflects efficiency considerations of maximizing workers’ e↵ort.
2. The model also sheds light on the labour market dynamics over the
business cycle:
There is no reason for the firms to adjust real wages in response to,
say, a negative demand or productivity shock.
The model thus predicts that shifts in labour demand will lead to large
movements in employment, with little changes in the wages.
11 / 44
a
Shapiro-Stiglitz (1984) model: an overview
So far we have simply been assuming that workers’ e↵ort is an

increasing function of the wage in (2).
In the highly influential paper, Shapiro and Stiglitz (1984) provide a
microeconomic rationale for this assumption.
Idea: if firms have a limited ability to monitor their workers, they are
forced to provide them with enough incentives to exert high e↵ort.
In the model such incentive arises from the risk of being fired and
loosing a well-paid job if not working hard (shirking).
Not only does the model provide a logical justification for the
efficiency wage, but it also does it in a spectacularly elegant fashion.
12 / 44
Setup
There is a large number of workers, L̄, and a large number of firms, N.

Time is continuous.
A representative worker maximizes their lifetime utility:
Z1
U= e-⇢t ut dt, (7)
t=0
where ⇢ > 0 is their discount rate.

Given wage and e↵ort level, a worker’s instantaneous utility is given by
wt - et if employed
ut = (8)
0 if unemployed.
2 possible e↵ort levels: e = ē > 0 (exerting e↵ort) & e = 0 (shirking).

13 / 44
v
Worker states and transitions

At any moment in time, a worker can be in one of three states:
E: employed (i.e. employed and exerting e↵ort);
S: shirking (i.e. employed and not exerting e↵ort);
U: unemployed.
Transitions between states follow simple Poisson processes.

First, there is an exogenous rate b > 0 at which jobs randomly end.
) I.e. the probability that an employed worker’s job ends for exogenous
reasons in the next dt units of time is b dt, when dt ! 0.
Second, there is an exogenous rate q at which firms detect shirking:
Imperfect monitoring, but a worker caught shirking is fired.
) The total probability that a worker in state S (employed and shirking)
loses his job in the next dt ! 0 units of time is (b + q) dt.
Third, unemployed workers find jobs at the endogenous rate a:
Firms choose workers at random, so a is determined by the rate at
which firms are hiring and the number of unemployed workers.
14 / 44
The value of being employed

Let Vi denote the ‘value’ to a worker of being in state i 2 {E, S, U}.
e.g. VE is the expected discounted lifetime utility of a worker who is
currently employed and exerting e↵ort.
We focus on a steady state in which Vi ’s are constant over time.

How can we compute the value of, say, being employed VE ?
·
Intuitively, think of employment like holding an asset.

This asset pays ‘dividend’ w - ē per unit time.
VE is then the fair price of such an asset, priced by a risk-neutral
worker with the required rate of return ⇢.
The expected return must thus be ⇢VE per unit time.
Lastly, there is a probability b per unit time of a ‘capital loss’
(VE - VU ) if the worker gets unemployed, i.e. his ‘asset’ loses value.
Thus, we can write an asset equation:
⇢VE = (w - ē) - b(VE - VU ) (9)
15 / 44
The values of shirking and being unemployed
We can extend the asset analogy to the other two states.

When a worker is shirking, the ‘dividend’ is w per unit time, but the
rate at which he looses the job is also higher at b + q. Thus:
⇢VS = w - (b + q)(VS - VU ). (10)
When the worker is unemployed, he receives no ‘dividend’, but he gets

a job (‘capital gain’) at the rate a, so
⇢VU = a(VE - VU ). (11)
We assumed above that if an unemployed worker gets a job, he exerts

e↵ort, as will indeed be the case in equilibrium.
16 / 44
Firms’ problem
A representative firm’s profit per unit time is given by:
⇡ = F(ēL) - w [L + S], F 0 (·) > 0, F 00 (·) < 0. (12)
L, S are numbers of workers who exert e↵ort and shirk, respectively.

The problem facing the firm is to incentivise employees not to shirk:
It must pay enough that VE > VS , otherwise workers prefer shirking.
The optimizing firm will not overpay more then necessary and will
choose w so the incentive constraints is just satisfied, i.e. VE = VS .
Use VS = VE in (10), and then subtract (9) and rearrange to get:

ē
VE - VU = > 0. (13)
q
The workers thus must strictly prefer employment to unemployment.
) to induce e↵ort, firms pay a premium over&above the cost of e↵ort ē.
17 / 44
Wage level that induces e↵ort
Subtract ⇢VU in (11) from ⇢VE in (9) to get:
⇢(VE - VU ) = (w - ē) - (a + b)(VE - VU ) (14)

ē
Substitute q for VE - VU from incentive cond. (13) and solve for w:
ē
w = ē + (a + b + ⇢) . (15)
q
This is the wage level needed to induce e↵ort, which
exceeds the cost of e↵ort ē by a positive amount.
increases in the cost of e↵ort ē, the ease of finding jobs a, the rate of
job breakup b, and the discount rate ⇢.
decreases in the rate at which shirkers are detected q.
The firms will pay this wage so there is no shirking in equilibrium.
18 / 44
The aggregate no-shirking condition (NSC)

Since the economy is in a steady state, the number of unemployed
workers is constant, so flows into and out of unemployment balance:
NLb workers become unemployed per unit time (where L is
employment per firm and so NL is aggregate employment).
a(L̄ - NL) unemployed workers find jobs per unit time.
Equating the two, we get the equilibrium job-finding rate:
NLb
a= . (16)
L̄ - NL
Substitute this into (15) to get the no-shirking condition (NSC):
✓ ◆
L̄ ē
w = ē + ⇢ + b (17)
L̄ - NL q
No-shirking wage is an increasing function of aggregate employment:

When NL ", it is easier for unemployed workers to find jobs, see (16).
The cost of being fired falls, so wage has to rise to prevent shirking.
19 / 44
Equilibrium
The FOC of a firm’s profit function (12) w.r.t. L yields
ēF 0 (ēL⇤ ) = w, (18)
so firms hire workers until marginal product of labour equals the wage.
This implies downward sloping aggregate labour demand LD = NL⇤ .
In the absence of any monitoring issues, Walrasian equilibrium would
occur at point EW where LD crosses the inelastic labour supply L̄.
I.e. there would be full employment in equilibrium (assuming that the
marginal product of labour at full employment exceeds cost of e↵ort ē).
However, with imperfect monitoring and possible shirking, equilibrium
occurs at the intersection E of LD and the no-shirking condition.
Wage is above the Walrasian level and there is positive unemployment.
Unemployed workers strictly prefer to be employed and exert e↵ort,
but the wage does not fall, because then workers would start shirking.
20 / 44
W
Equilibrium in the Shapiro-Stiglitz model: illustration
21 / 44
Some comparative statics
To clarify the workings of the model, its helpful to do comparative

statics, i.e. see how the equilibrium depends on parameters.
An increase in the shirking detection rate q shifts the NSC down.
The equilibrium wage falls and employment rises.
Intuition: monitoring becomes better, so there is less need to
incentivise workers by high wage.
As q ! 1, the economy approached the Walrasian equilibrium.
If the job separation rate b falls to 0, there is no turnover, and
unemployed workers are never hired.
) The no-shirking wage in this case is simply ē + ⇢ē/q, see (17)
i.e. the NSC becomes flat and independent from employment.
Intuitively, workers now only consider the cost of e↵ort and the risk of
permanently losing employment when contemplating shirking.
Homework invites you to conduct some additional exercises.

22 / 44
v
The e↵ect of a rise in q
a -> W/ -> NSCU
23 / 44
The Shapiro-Stiglitz model without turnover (b = 0)
24 / 44

Like any efficiency wage theory, Shapiro-Stiglitz model implies:
1. There is involuntary unemployment.
2. Although individual labour supply is inelastic when w > ē, shifts in
labour demand result in movements along the relatively flat NSC curve.
) wages respond less to demand fluctuations, and employment more.
But the formal model also yields additional insights:
The theory implies that decentralized equilibrium is inefficient, since
the marginal product of labour exceeds the cost of e↵ort.
) wage subsidies financed by lump-sum taxes improve welfare.
The model (modified to allow for flexible hours) also explains why
firms lay o↵ workers during downturns, rather than reduce hours.
) Reductions in hours make jobs less valuable, and hence workers
would be more inclined to shirk.
The view that workers are ‘rational cheaters’ is not uncontroversial –

see homework for a more ‘humane’ theory due to Akerlof and Yellen.
25 / 44
Search and matching models: an overview clip 11
A modern approach to model labour markets.

Starting point: workers and jobs are highly heterogeneous.
As the result, one should not think about the market for labour as a
single market with traditional supply and demand.
) Instead, matching of workers and jobs occurs through a complex
process of search and matching.
Workers and firms meet in one-on-one fashion and engage in costly
process of trying to match idiosyncratic preferences, skills, and needs.
Wages are determined by bargaining between workers and firms, rather
than by a Walrasian auctioneer picking a market-clearing wage.
As the process is not instantaneous, some unemployment is inevitable.
P.Diamond, D.Mortensen & C.Pissarides got Nobel prize for this work.
Search and matching models are relatively complicated, so we will be
looking only at the basic framework and the main issues.
26 / 44
Setup
Time is continuous. The economy consists of workers and firms/jobs.
There is a continuum of workers of mass 1.
Agents are risk-neutral and their discount rate is r > 0.
Each worker can be either employed (E) or unemployed (U):
An employed worker produces y per unit time at his job.
An unemployed worker receives ‘benefit’ b per unit time.
A job can be either filled (F) or vacant (V):
A filled job generates y, and pays the worker wage wt per unit time.
If a jobs is vacant, there is neither output nor labour costs.
Both filled and vacant jobs involve maintenance cost c per unit time.
Assume y > b + c, so filled jobs generate positive value.
Free entry: jobs can be created freely, but incur cost c once created.
Absent search frictions, there would be full employment in equilibrium
) There would be exactly a unit mass of filled jobs, and no vacant jobs.
27 / 44
Search frictions and matching
The central feature of the model is that there are search frictions.
) Unemployed workers and vacant jobs cannot find each other costlessly.
Stocks of unemployed workers Ut and vacancies Vt yield a flow of

matches of workers with firms according to the matching function:
Mt = M(Ut , Vt ), MU > 0, MV > 0. (19)
It proxies the process of recruitment, worker search & evaluation etc.

We assume that all matches lead to hirings.
This is reasonable, because filling a job creates a positive surplus.
This surplus is divided between the worker and the firm according to
their relative bargaining powers, as we shall see shortly.
Positive turnover: there is an exogenous separation rate at which

jobs end, similarly to the Shapiro-Stiglitz model.
28 / 44
Job finding and vacancy filling rates

In line with empirical studies, we assume that the matching function
M has constant returns to scale (CRS), allowing us to write:
M(Ut , Vt ) = Ut M(1, Vt /Ut ) = Ut m(✓t ). (20)

Vt
Here ✓ ⌘ Ut is the labour marker tightness, and m(✓) ⌘ M(1, ✓).
Then the endogenous job finding rate, i.e. the probability per unit
time that an unemployed worker finds a job is:
M(Ut , Vt )
at = = m(✓t ) " in ✓. (21)
Ut
Similarly, the endogenous vacancy-filling rate is:
M(Ut , Vt ) m(✓t )
↵t = = # in ✓. (22)
Vt ✓t
29 / 44
The Beveridge Curve

For simplicity, we are again going to focus on the steady state.
U is constant, and flows into and out of unemployment balance:
E = (1 - U) workers become unemployed per unit time, and
aU workers find employment, allowing us to write:
(1 - U) = aU. (23)
Using that a = m(✓), where ✓ ⌘ V/U, and rearranging yields:
U= (24)
+ m(V/U)
(24) defines a negative relationship between unemployment and

vacancies, known the Beveridge curve, or UV curve.
It reflects job matching efficiency and the rate at which jobs end, .
Hence shifts in the Beveridge curve are often thought to reflect
structural changes in the labour market.
30 / 44
The Beveridge Curve

V
Beveridge Curve
U
31 / 44
Values of employment, unemployment, filled & vacant jobs

Alas, we must do more work and focus now on agents’ decisions to
determine where along the Beveridge Curve equilibrium takes place.
Value VE that a worker attaches to being employed satisfies:
rVE = w - (VE - VU ) (25)
Here wage w is the ‘dividend’, reflects the probability of a losing
the job, and VE - VU is the ‘capital loss’ if realised.
Similar reasoning allows us to write the equations for values of being
unemployed VU , firm having a filled job VF , and a vacant job VV :
rVU = b + a(VE - VU ) (26)
rVF = (y - w - c) - (VF - VV ) (27)
rVV = -c + ↵(VF - VV ). (28)
To compute these values, we must know how w is determined.
32 / 44
Wage determination: Nash bargaining

The wage is determined by Nash bargaining:
Fraction of the total surplus generated by a match goes to the
worker, and fraction 1 - goes to the firm, where
2 (0, 1) is the workers’ relative bargaining power.
Subtract (26) from (25) and solve for the worker’s surplus VE - VU :
w-b
VE - VU = . (29)
a+ +r
Similarly, (27) and (28) yield the firm’s surplus from a match:
y-w
VF - VV = . (30)
↵+ +r
Since the total surplus is the sum of the two, Nash bargaining implies:
VE - VU = (VE - VU + VF - VV ). (31)
33 / 44
Wage determination: Nash bargaining (cont.)

Combining these three equations and solving for the wage, we get:
(a + + r)
w=b+ (y - b). (32)
a + (1 - )↵ + + r
It looks complicated, but we can actually interpret this.

When stocks of vacancies and unemployment are equal and ✓ = 1,
job finding rate a equals vacancy-filling rate ↵, and (32) simplifies to:
w = b + (y - b). (33)
I.e. fraction of the generated value y - b goes to the worker.

When a exceeds ↵, workers can find new jobs more rapidly than firms
can find new employees, so their bargaining position improves and w "
Vice versa, w # when ↵ exceeds a.
34 / 44
Free entry and the value of a vacancy
Lastly, we need to examine the firms’ decisions to post vacancies.

Because entry is free, in equilibrium the value of a vacancy must be
VV = 0, since otherwise firms would create or remove vacancies.
Substituting for VF - VV in (28) from (30), we get:
y-w
rVV = -c + ↵ . (34)
↵+ +r
Substitute in wage from (32) and simplify to get:
(1 - )(y - b)↵
rVV = -c + = 0, (35)
a + (1 - )↵ + + r
the last equality by the free entry condition.

This now helps us determine the supply of vacancies in equilibrium )
35 / 44
Vacancies supply
Recall that a = m(✓) and ↵ = m(✓)/✓, where ✓ ⌘ V/U.

Substituting these into the condition (35) and rearranging, we get the
equilibrium condition for the supply of vacancies:
✓ ◆
+r 1-
✓ + = (y - b - c) (36)
m(✓) c
While this looks complicated, note that the only endogenous variable
in this equation is the labour market tightness ✓.
Condition (36) thus implicitly determines the unique ratio ✓ between
vacancies and unemployment that is consistent with firms’ free entry.
In the (U, V) space, it is represented by a straight line with slope ✓,
known as the Vacancies Supply (VS), or Job Creation Curve.
36 / 44
The Job Creation Curve

V
Job Creation
✓
U
37 / 44
Equilibrium in the search and matching model
We now have derived:

1. The Beveridge Curve, that describes the structural relationship between
vacancies and unemployment in the steady state
2. The Vacancies Supply Curve that describes firms’ decisions to post
vacancies given the level of unemployment.
The equilibrium occurs at point E where the two conditions intersect.
The levels of unemployment and vacancies are uniquely determined.
Can solve mathematically: substitute for ✓ in the vacancies supply
condition (36) from the Beveridge Curve (24) to get equation for UEQ .
Then can solve for other variables like vacancies V EQ , employment
EEQ = 1 - UEQ , wage wEQ , etc. (though usually not in closed form).
There is positive unemployment in equilibrium due to matching

frictions.
38 / 44
Equilibrium vacancies and unemployment

V
Job Creation
E
V EQ
Beveridge Curve
✓
UEQ U
39 / 44
Welfare
Firms’ entry decisions have externalities for workers and other firms:
1. Entry makes it easier for unemployed workers to find jobs, and
improves their bargaining position when they do.
2. It also makes it harder for other firms to find workers, and also worsens
their bargaining position when they do.
As a result, the decentralised equilibrium is generally not efficient.
I.e. social welfare Ey + (1 - E)b - (E + V)c is not necessarily maximized.
However, depending on which e↵ect dominates, equilibrium

employment can be either inefficiently high, or inefficiently low.
Determining which of these cases is correct is an important
policy-relevant question.
40 / 44
Comparative statics: a shift in labour demand
We are interested in how matching frictions a↵ect the cyclical

behaviour of the labour market.
I.e. does a shift in labour demand have a larger e↵ect on employment and
smaller e↵ect on the wage compared to the Walrasian case?
Shortcut: model cyclical change as a shift in y in the steady state.
An increase in y to y 0 does not a↵ect the Beveridge Curve.
But the vacancies supply condition (36) implies that the equilibrium
labour market tightness increases from ✓ to ✓ 0 .
The Vacancies Supply curve thus rotates up.
The new equilibrium at E’ has more vacancies and lower unemployment
Alas, the model does not imply much wage rigidity, however.
When ✓ increases, a rises and ↵ falls.
From equation (32), wage thus increases substantially with y.
Large increase in the wage reduces incentives to create new vacancies.
Thus employment e↵ects from shifts in demand are typically small.
41 / 44
A shift in labour demand: illustration

V
VS’
VS
E’
E
Beveridge Curve
✓0
✓
U
42 / 44
Extensions
1. Worker heterogeneity
We motivated the model by heterogeneity, yet so far focused on a
simple case with homogenous workers and rather mechanical matching.
Can add heterogeneity by assuming that when a worker and a firm
meet, the worker’s productivity is drawn from a random distribution.
If productivity is too low, the match does not form.
Stochastic worker productivity can also cause endogenous job break-up.
2. On the job search
Workers can continue searching even when they are employed.
Change jobs when they find a more productive match.
3. Competitive search
Directed search: people don’t search randomly but gather .
Posted wages: bargaining does not take place from scratch, but is
within firms’ wage policies. Posted wages thus a↵ect directed search.
43 / 44
Where are we?

Successfully solved a modern search and matching model.
Explains normal unemployment as the result of continually matching
workers and jobs in the complex real world.
) This is known as frictional unemployment.
Evidence indeed points to a considerable role for matching frictions.
Yet even if unemployment appears to result from search and matching
frictions, it may have deeper underlying causes.
E.g. Difficulty of finding parking in NY may look like searching for a spot,
yet ultimately the search is long because there are more cars than spots.
Evidence suggests that significant part of unemployment isn’t frictional.
Thus considerations like efficiency wages likely play an important role.
The basic model also does not generate substantial wage rigidity to
explain the cyclical behaviour of the labour market.
For these reasons, introducing wage rigidity into search and matching
models (e.g. via efficiency wages) is an important research agenda.
44 / 44
Lecture 5: Financial Frictions and Crises
Artur Doshchyn
Hilary Term 2023

Introduction Agency costs and fin. accelerator Diamond-Dybvig model D-D and the 2008 Crisis Contagion
Introduction
In the models that we’ve seen so far the financial markets are perfect:
Households save or borrow to satisfy their Euler equations.
Firms borrow until the marginal product of capital = interest rate.
There are no defaults or borrowing constraints.
The Crisis of 2007-8 demonstrated that even in the advanced modern

economies this description could not be further away from the truth.
There has been a tremendous increase in macro-finance research.
Major issues this research is interested in:
Do financial markets amplify and propagate macroeconomic shocks?
Are financial markets an independent source of shocks to the economy?
Why do financial crises happen, and what should policymakers do? etc.
1 / 45
Plan of action
There are potentially many di↵erent imperfections in the financial

sector with important macroeconomic consequences.
The literature is huge, and warrants a separate course.
Today we will focus on three classic models that illustrate some of the
most important issues in macro-finance:
1. ‘Real’ financial frictions and financial accelerator
2. Liquidity provision by banks and bank runs
3. Moral hazard created by government guarantees.
We will discuss how these models fit in the modern macro-finance

literature and help us shed light on complex events and policy choices.
2 / 45
Asymmetric information and agency costs in finance
The financial sector appears to be ridden with agency problems:

Firms are much better informed about their prospects than investors.
It may be difficult to tell good and bad entrepreneurs apart.
Managers that have limited liability may invest in too risky projects.
Firms’ performance may be difficult to verify by external investors.
Fund managers may not exert e↵ort to make the best investments etc.
We therefore have institutions like banks, mutual funds, and credit

rating agencies that specialise in acquiring & transmitting information.
But as became obvious during the 2008 Crisis, such institutions
themselves may not be immune to agency problems.
Agency costs and asymmetric information can distort investment
choices, as well as magnify e↵ects of shocks to the economy.
3 / 45
A simple model of moral hazard in financial intermediaries
Several ways agency problems are modelled in the literature:

) Costly state verification, adverse selection, moral hazard etc.
They all, however, have broadly similar implications:
It is important that agents (e.g. banks or entrepreneurs) have enough
‘skin in the game’, i.e. their own resources they contribute to projects
The wealth, or ‘net worth’ of key agents in the economy can thus
have important macroeconomic consequences.
Today we will focus on a particularly simple way to model moral

hazard introduced by Gertler-Karadi (2011), Gertler-Kiyotaki (2011).
) Banks can default and do a runner with their depositors’ funds.
The model predicts that the net worth of the banking sector becomes
crucial to the efficiency of financial intermediation.
4 / 45
Setup
There are two periods {1, 2}.

There are households, banks, and firms in the economy.
Each household has a unit mass of members, including a banker.
A banker runs a bank that provides services to other households.
Generated profits shared with his own household.
In period 1, a representative household has endowment y.

Household deposits d with a bank in period 1.
Deposits pay return R in period 2.
Household receives profit ⇡ from the own banker in period 2.
All members within a household consume the same amounts:
c1 in period 1, and c2 in period 2.
5 / 45
Household problem
A representative household maximizes utility:
p p
max U(c1 , c2 ) = c1 + c2 (1)
c1 ,c2 ,d
Subject to period 1 and 2 budget constraints:
c1 + d = y (2)
c2 = Rd + ⇡. (3)
Note that R and (lump-sum) ⇡ are treated as given by the household,

but will be endogenously determined in financial markets.
Substitute constraints in and take the FOC to get the Euler equation:
1 1
p = p R () c2 = c1 ( R)2 (4)
c1 c2
6 / 45
Solution to household optimisation problem
Three equations (2 BCs + Euler) and three unknowns: c1 , c2 , d.

The solution is:
y+ ⇡R
c1 = (5)
1 + 2R
d = y - c1 (6)
c2 = Rd + ⇡ (7)
Substitution e↵ect dominates: c1 is a decreasing function of R .

Therefore, the supply of funds d by households increases in R.
7 / 45
Firms
Produce goods for consumption in period 2.

Constant return to scale technology: each unit of good invested in
period 1 generates Rk goods in period 2 (Rk exogenously fixed).
A representative firm borrows s from banks in period 1, and so
generates sRk in period 2.
Firms are perfectly competitive, so entire return sRk goes to banks.
Note: Firms are unable to borrow directly from households.
) Think about banks as relationship lenders that are able to screen and
monitor firms they lend to, while households are not.
8 / 45
Bankers
A representative banker is endowed with N goods in period 1.

This is in addition to their household’s endowment y.
We will refer to N as the banker’s net worth.
The bank accepts deposits d from households and lends s to firms.
Takes market rates of returns on deposits R and loans Rk as given.
Chooses d, s to solve
max ⇡ = sRk - Rd.

d,s
Always maxes out on s, as loans pay positive return with certainly:
s = N + d.
9 / 45
Overview
10 / 45
Benchmark: equilibrium with no financial frictions

Equilibrium requires
1. Household maximises utility given R, ⇡.
2. Banker maximises profit given R, Rk .
3. Markets clear.
With no financial frictions, there is a unique equilibrium in which

R = Rk . (8)
Contradiction otherwise:
If R > Rk then bank chooses d = 0, no financial intermediation.
If R < Rk then bank chooses d ! 1, unbounded fin. intermediation
With R = Rk , ⇡ = NRk : banks earn market return on endowment N,

but otherwise earn zero profit from intermediating deposits d.
Can compute equilibrium c1 , c2 , d using (5)-(7).
Equilibrium is first best, as there are no inefficiencies.
11 / 45
Introducing moral hazard
Moral hazard à la Gertler-Karadi (2011), Gertler-Kiyotaki (2011).

The bank can default after receiving sRk from firms in period 2.
If the banker defaults, he absconds with fraction ✓ of resources:
) ✓Rk (N + d) goes to the banker, (1 - ✓)Rk (N + d) goes to depositors.
The banker will not default if:
(N + d)Rk - Rd > ✓Rk (N + d) (9)

| {z } | {z }
Profit when no default Profit when default
N " increases LHS more than RHS, so makes default less likely.
12 / 45
Equilibrium with no defaults

Without loss of generality (WLOG), we can restrict attention to
equilibria in which banks never default:
If there is an equilibrium in which banks default, the e↵ective return on
deposits that households get (after default) is:
(1 - ✓)Rk (N + d)
Re↵ective = . (10)
d
Households anticipate the default, so use Re↵ective rather than the
promised return R when making their saving choices d.
But then there is also an equivalent equilibrium in which R = Re↵ective :
households supply the same amount of savings d.
banks do not default, since (9) holds with equality (easy to verify).
Thus, we look for a symmetric equilibrium with no default:

1. Household maximises utility given R, ⇡.
2. Banker maximises profit given R, Rk , and no default condition (9).
3. Markets clear
13 / 45
The banker’s problem
The representative banker solves:
max ⇡ = (N + d)Rk - Rd (11)

d
s.t. (N + d)Rk - Rd > ✓Rk (N + d) (12)
There can be only two possibilities:

1. The no-default constraint (12) is slack (does not bind) in equilibrium
2. The no-default constraint (12) binds in equilibrium
14 / 45
Case 1: The financial constraint is slack
If the no-default constraint (12) is slack, then it is irrelevant, and

analysis in the benchmark case applies.
Thus R = Rk and we again have the first-best allocation.
This is indeed the equilibrium i↵ (12) is satisfied at R = Rk .
Substituting for R and rearranging, yields:
1-✓ ✓
d6 N or N> d (13)
✓ 1-✓
That is, the net worth of the banking sector has to be sufficiently
high relative to the supply of funds by households.
Intuition: high N implies bankers have enough ‘skin in the game’.
15 / 45
First-best equilibrium when N is high
R
Household supply of d
First best equilibrium

E
Rk ·
bank prefers to default
Bank demand for d
(1 - ✓)Rk
1-✓ d
✓ N
16 / 45
Case 2: the financial constraint binds
The financial constraint (12) binds in equilibrium when N is low, i.e.
✓
N< d. (14)
1-✓
Banks’ demand for deposits is now a decreasing function of R.
Indeed, binding (12) yields a downward-sloping curve in (d, R) space:
✓ ◆
k N
R = (1 - ✓)R +1 . (15)
d
Intuition: as d increases relative to N, banks have a stronger

incentive to default, so R must fall to reduce their liabilities, Rd.
Equilibrium obtains where the household supply of d intersects (15).
17 / 45
Equilibrium when banks’ net worth is low
R
Household supply of d
Rk
E’
(1 - ✓)Rk
Equilibrium with binding
financial constraint
1-✓ 0 d
✓ N
18 / 45
Role of banker net worth
Note that R < Rk when the financial constraint binds in equilibrium.

Thus there is a positive spread between the exogenous return Rk on
technology and the deposit rate R.
) R and d are less than socially optimal.
Can verify by rearranging the binding constraint (12) and using (14):
Rk - R N
= ✓ - (1 - ✓) > 0. (16)
Rk d
N " raises R nearer to Rk .
N " thus brings financial intermediation closer to the first best.
Explains the desire to repair bank balance sheets after 2008-9 crisis.
19 / 45
The financial accelerator
Our simple model only has two periods.

With more periods, we would almost certainly get a financial
accelerator, when negative shocks propagate through the fall in N.
As we have seen, the net worth of banks (and agents more generally)
is crucial to their ability to borrow an invest.
) when N falls, there are less investments made, leading to output losses.
But agents’ resources depend on output in the first place!
Financial accelerator then appears naturally:
Suppose some shock reduces output
Due to lower output, the net worth of banks (and other agents) falls
Agency costs rise and investment falls
The initial fall in output is thus magnified
Leading to further fall in net worth etc. etc.
20 / 45
The financial accelerator and asset prices
Most firms hold a variety of assets they use as collateral for loans.
Their net worth thus depends on the market value of their assets.
) Financial accelerator can magnify shocks via the value of collateral.
Kiyotaki & Moore (1997) model the vicious amplification cycle:
A bad shock reduces net worth and increases agency costs of lending.
The ability of constrained firms to borrow and purchase assets falls.
More assets are held by unconstrained firms, which are less productive
on the margin (since they already hold lots of assets and produce a lot!)
Asset prices and the value of collateral fall.
The net worth of constrained firms further falls.
Agency problems become even worse.
The ability of constrained firms to borrow and invest falls further.
etc etc.
21 / 45

We have seen a simple model in which a fall in net worth of banks
can lead to a fall in lending, and distort investment.
Theory not limited to banks, and applies also to firms, households, etc.
Alternative formulations of agency problems lead to similar conclusions.
Because the model does not feature money or any role for liquidity,
these mechanisms are known as real financial frictions.
Financial frictions are present and cause welfare losses all the time:
Significant empirical evidence that there’re financially constrained firms,
implying that average investment may be inefficiently low, while
business cycles are amplified and propagated by financial accelerator.
But agency problems can become particularly bad during crises:
Baron et al (2020) document that large declines in bank equity are
associated with substantial credit contractions and output losses.
Financial accelerator can make the situation a lot worse.
We now turn to the issues of liquidity.

22 / 45
Diamond-Dybvig (1983) model: an overview
Financial markets are subject to sudden, convulsive changes.

A bank that appears to function normally can suddenly be in trouble.
Financial crises typically unfold very rapidly.
In the seminal paper, Diamond and Dybvig (1983) model bank runs
as a spontaneous switch between multiple equilibria.
Both earned a Nobel prize last year for this work!
The model makes two fundamental predictions:
1. Banks improve social welfare because they provide liquidity;
2. But bank runs are possible exactly because banks’ liabilities are liquid.
Key: maturity mismatch of banks’ liquid liabilities & illiquid assets.
23 / 45
Setup
There are three periods, denoted {0, 1, 2}.

There is a continuum of agents of mass 1.
Each agent is endowed with 1 unit of consumption good in period 0.
Investment technology: each unit invested in period 0 yields
R > 1 if held to maturity in period 2;
only 1 if the project is liquidated in period 1.
Agents can also freely store the good between periods.

Because the investment technology dominates storage (since it pays
out at least 1), all endowments are invested in period 0.
Agents decide whether to liquidate projects in period 1, and how
much to consume in periods 1 and 2, c1 and c2 , respectively.
24 / 45
Types of agents and liquidity shocks

2 types of agents in the economy:
‘type-a’, or ‘impatient’ individuals only value consumption in period 1:
Ua = ln ca
1. (17)
‘type-b’, or ‘patient’ agents are willing to consume in period 1 or 2:
Ub = ⇢ ln(cb b
1 + c2 ), ⇢ 2 (0, 1), ⇢R > 1. (18)
⇢ < 1 implies that impatient agents value consumption more.

All agents are identical ex ante: type is unknown in period 0.
In period 1, a fraction ✓ 2 (0, 1) of agents learn that they are type-a,
and a fraction 1 - ✓ learn that they are type-b.
Each agent thus faces a probability ✓ of being impatient ex ante.
Once realised, an individual’s type is not observable by others.
‘Impatience’ here is a metaphor for liquidity needs more generally.

25 / 45
Benchmark 1: competitive equilibrium
Agents hold investments directly and every period there is a

competitive market for claims on future consumption goods.
Because agents’ types are not observable, such claims cannot be
contingent on buyers’ or sellers’ types.
Period-1 price of a claim on a unit of period-2 consumption is 1/R.
If it was above, type-a agents would keep their investments and sell
such claims to finance their consumption, but there would be no buyers.
If it was below, type-b agents would liquidate their investments and
buy such claims, but there would be no sellers.
Similarly, equilibrium in period 0 requires that
price of period-1 consumption is simply 1;
price of period-2 consumption is simply 1/R.
Otherwise everybody would want to buy ‘underpriced’ claims or sell
‘overpriced’ ones, which cannot be an equilibrium.
26 / 45
Benchmark 1: competitive equilibrium (cont.)
Agents are identical in period 0, so there is no trade at these prices.

Each individual invests their own unit of endowment.
In period 1, impatient agents liquidate their projects and consume 1.
Patient agents wait until period 2 and thus consume R.
Allocation is therefore the same as under autarky, since markets

serve no useful purpose here.
Expected utility is
E[Uautarky ] = ✓ ln 1 + (1 - ✓)⇢ ln R = (1 - ✓)⇢ ln R. (19)
Is this equilibrium efficient?
27 / 45
Benchmark 2: first-best social optimum
Suppose there is a social planner that observes individual types.

Social planner chooses consumption levels of each type (ca b
1 and c2 )
to maximize the expected utility of a representative agent:
E[U] = ✓ ln ca b
1 + (1 - ✓)⇢ ln c2 . (20)
Note that the planner always sets ca b

2 = c1 = 0,
) type-a’s do not value consumption in period 2, while consumption and
utility of type-b’s can be increased by waiting until period 2.
Social planner must respect the resource constraint:
(1 - ✓)cb a
2 = (1 - ✓c1 )R, (21)
where ✓ca
1 is the fraction of investments liquidated in period 1.
28 / 45
Benchmark 2: first-best social optimum (cont.)

Substituting for cb2 in the objective function from the constraint,
taking the first-order condition w.r.t. ca a
1 , and solving for c1 yields:
1
ca⇤
1 = > 1. (22)
✓ + (1 - ✓)⇢
) Impatient agents’ ca⇤
1 is higher than in the competitive/autarky case.
The constraint (21) then implies:

⇢R
cb⇤
2 = < R. (23)
✓ + (1 - ✓)⇢
) Patient agents’ cb⇤
2 is lower than in the competitive/autarky case.
Although still higher than ca⇤
1 , since ⇢R > 1.
The competitive outcome is not socially efficient! Why?

Social optimality requires greater liquidation of projects in period 1,
i.e. patient agents insuring individuals with liquidity shocks.
But because types are unobservable, such insurance market is missing.
29 / 45
Introducing a bank
Key result: we do not need a social planner or observability of

agents’ types to achieve the first best, all we need is to set up a bank!
The banks o↵ers deposit contracts on the following terms:
agents deposit their unit endowments to the bank in period 0.
anyone, regardless of their type, can withdraw ca⇤
1 units in period 1.
Whatever funds the bank has in period 2 are divided equally between
the remaining depositors who did not withdraw in period 1.
The bank invests all deposits in the projects, and liquidates

investments as required to meet the demand for early withdrawals.
30 / 45
Supporting the social optimum
Suppose only impatient agents withdraw in period 1 and get ca⇤

1 .
The amount that patient depositors receive in period 2 is then cb⇤

2 :
(1 - ✓ca⇤
1 )R
c2 = = cb⇤
2 . (24)
1-✓
This is exactly the socially optimal allocation we saw before!
It is also a Nash equilibrium, since no-one has incentives to deviate:
Impatient agents always prefer to withdraw in period 1, not 2.
Because cb⇤ a⇤
2 > c1 , patient agents indeed prefer to wait and withdraw
in period 2, not 1.
All agents indeed have incentives to deposit their endowments in
period 0, since their expected utility is greater than under autarky.
Banks are socially useful: they provide liquidity.
31 / 45
Bank runs
What happens, however, if all agents try to withdraw in period 1?
The bank cannot pay ca⇤
1 > 1 to everyone.
The sequential-service constraint (aka first-come, first-served):

Pays ca⇤
1 to as many agents as possible, until runs out of resources.
Since types are not observable, those who are paid are chosen randomly.
E.g. who happens to reach the queue first.
The remaining depositors get nothing.
Unfortunately, such a bank run is also a Nash equilibrium:
If all agents withdraw early, there is nothing left in period 2.
Thus, if all patient agents believe that other patient agents will try to
withdraw in period 1, they prefer trying to withdraw in period 1 too.
Better a probability of something than certainty of nothing!
Why would anyone deposit in the bank in period 0 then?
Can construct cases when it’s still optimal if a run probability is small.
Possibility of a bank run is thus another key result of the model.

32 / 45
A run on American Union Bank, New York. April 26, 1932
33 / 45
Taking stock
We saw that banks support two types of equilibria:

1. Socially optimal equilibrium competitive equilibrium/autarky
2. Bank run equilibrium competitive equilibrium/autarky.
The second equilibrium exists because of maturity mismatch:

bank has illiquid assets, but liquid liabilities.
In the model, a run is a pure liquidity crisis for the bank:
Bank is solvent in the sense that it would have enough funds to make
promised payments in period 2 (which is not always the case!).
The problem is that all depositors panic and demand funds immediately.
Diamond & Dybvig consider several policies to prevent liquidity runs.
34 / 45
Policy #1: Suspension of convertibility
Modify the deposit contract: pay out ca⇤

1 to at most fraction ✓ of
depositors, and then suspend payments until period 2.
Success: type-b agents no longer have incentives to run.
This is what XIX and early XX century banks actually did.
However, if aggregate liquidity needs are uncertain (e.g. fraction ✓ of
impatient agents is random), this is does not maximize social welfare.
Moreover, in the real world, there is a risk that insolvent banks may
suspend payments, even though liquidation is preferable.
35 / 45
Policy #2: A lender of last resort

Government (central bank) can also act as a lender of last resort.
I.e. stand ready to lend to the bank at a gross interest rate cb⇤ a⇤
2 /c1 .
Suppose fraction > ✓ of agents withdraw.
Bank pays ✓ of them by liquidating projects, and - ✓ by borrowing.
Bank thus has (1 - ✓ca⇤ b⇤
2 )R = (1 - ✓)c2 resources in period 2.
Uses ( - ✓)cb⇤2 to repay the central bank and (1 - )cb⇤2 to pay
remaining depositors, who thus get cb⇤
2 each.
But now all type-b depositors prefer to withdraw in period 2.
How does the central bank finance its lending to the a↵ected banks?
1. Backed by fiscal authority: tax early withdrawers.
2. CB can supply new money (if we introduce it into the model):
Assume that deposits are denominated in nominal terms.
CB lending to the bank increases money supply and nominal prices of
goods in period 1 ) real consumption of early withdrawers falls.
Government power to tax and create inflation is crucial.

36 / 45
Lender of last resort: Diamond-Dybvig vs. Bagehot dictum
Note that because cb⇤ a⇤

2 /c1 = ⇢R < R, the CB lends at the discount
compared to the market rate R.
This can generate a moral hazard:
Bank has incentives to borrow rather than liquidate assets in period 1.
This leaves the bank with a profit at the expense of the central bank
and taxpayers, since projects generate more than it has to repay.
The solution is a reserve requirement: the bank must meets first ✓
withdrawers by liquidating own assets before it can borrow from CB.
W. Bagehot (1873) argued that the lender of last resort should:
1. lend freely against a good collateral
2. but at a penalty interest rate to discourage unnecessary draws.
The Diamond-Dybvig model supports #1, but not #2.
37 / 45
Policy #3: Deposit insurance
Government guarantees that all who wait until period 2 get cb⇤
2 .
Government finances these payments by levying a tax on those who

withdraw in period 1, if more than fraction ✓ withdraw.
:
Again, type-b agents no longer have incentives to run.

Run equilibrium is eliminated by the mere presence of deposit
insurance, the government never needs to make actual payments.
Policy widely introduced around the world after the banking crises
that accompanied the Great Depression in 1930s.
) Arguably, traditional runs on commercial banks are left in the past.
But, as we will see now, it does not mean that there can be no panics
in the modern financial system.
38 / 45
Panics during the 2008 Financial Crisis?
Modern ‘bank runs’ can take various forms.

Gorton and Metrick (2012) argue that the panic of 2007-2008 was
e↵ectively a run on the repo (sale and repurchase) market:
repos = very short-term, collateralized transactions widely used as a
primary source of financing by institutions like investment banks.
Not commercial banks, not subject to deposit insurance.
Repo run = many investors simultaneously refusing to roll over their
loans, or demanding much tougher terms.
The bank is thus forced to liquidate investments early, and may fail.
Therefore, a repo run can be a self-fulfilling equilibrium: once many
investors refuse to roll over loans, everybody refuses.
just as in the Diamond-Dybvig model!
Similarly, Covitz et al. (2013) document runs on asset-backed

commercial paper (ABCP) in 2007.
39 / 45
The rise of shadow banking

Repos and ABCP are prominent examples of shadow banks’ liabilities.
Shadow banks are non-bank financial institutions that perform many
banks’ functions: credit, maturity and liquidity transformation.
E.g. investment banks, money market funds, mortgage companies etc.
Mostly outside normal banking regulations ) subject to runs.
Shadow banking grew dramatically since 1990s (Pozsar et al. 2013):
40 / 45
Runnable liabilities in the US before and after the Crisis
Notes. The data on aggregate runnable liabilities comes from Bao et al. (2015). It includes
uninsured deposits, repurchase agreements, securities lending, commercial paper, money market
mutual funds shares, variable-rate demand obligations, federal funds borrowed, and funding
agreement backed securities. The data on total financial sector liabilities is from the Board of
Governors of the Federal Reserve System.
41 / 45
Macroeconomic contagion
Financial crises are characterised by simultaneous difficulties faced by

many financial institutions, firms, and individuals.
But if e.g. we introduce many banks in the Diamond-Dybvig model,
there is no reason why a run on one bank would spread to others.
The last issue we are going to discuss only briefly is contagion: how
do financial troubles faced by individual agents spread at macro level?
There are several major sources of contagion:
1. Counterparty contagion
2. Confidence contagion
3. Fire-sale contagion
4. Macroeconomic contagion.
42 / 45
Sources of contagion
1. Counterparty contagion
In practice, financial institutions often hold claims on one another.
When one fin. institution faces a run and a risk of failure, institutions
that are exposed to it may see the value of their claims fall,
and ultimately su↵er similar fate.
2. Confidence contagion
Recall that the Diamond-Dybvig model is that of a pure liquidity run.
But a run could also happen when investors suspect that there is a risk
that the bank may be insolvent.
In fact, bank runs tend to happen when fundamentals are weak.
If a large bank faces a run, it could provide a negative signal about its
solvency and the value of its assets.
But this could lead investors of other banks that hold similar assets to
question their solvency and run as well.
43 / 45
Sources of contagion (cont.)
3. Fire-sale contagion
An institution facing a run or a large increase in borrowing costs is
likely to have to sell assets.
But if many institutions do it simultaneously in an imperfect financial
market, this leads to a fall in asset prices.
Decrease in the value of collateral leads to a further round of
insolvencies, runs, and fire sales.
4. Macroeconomic contagion
Difficulties faced by borrowers are likely to reduce economic activity
Which in turn reduces asset prices and agents’ net worth
Leading to further increases in borrowing costs etc.
This is the financial accelerator again.
44 / 45
Conclusion
Financial markets are complex: we have seen various mechanisms that

can generate financial frictions and trigger financial crises.
These issues are all inter-related:
For example, in practice it is often difficult, if not impossible, to
distinguish between illiquid and insolvent banks.
A bank may be solvent under normal macro conditions and asset prices.
But insolvent once there are widespread runs on its counterparties and
a collapse in asset prices.
Modern macroeconomic models often combine di↵erent mechanisms.

Gertler and Kiyotaki (2015) is a great example:
Their model features the conventional financial accelerator.
But systemic bank runs are also possible.
Run equilibria exist only in recessions when the macroeconomic
fundamentals are sufficiently bad, which is realistic.
45 / 45
Lecture 6: Budget Deficits and Sovereign Debt Crises
Artur Doshchyn
Hilary Term 2023

Introduction Tax smoothing Costs of deficits Strategic debt accumulation Delayed stabilization Sovereign debt crises
Introduction
We have already studied some key issues related to monetary policy -

one of the two major branches of macroeconomic policy.
Today we will focus on the second branch – fiscal policy, which is
concerned with levels and composition of taxes and gvt. spending.
Consensus: macro stabilization is best left to monetary policy.
) Central banks are able to react quickly and decisively, whereas fiscal
changes typically face significant political barriers.
Though massive fiscal stimulus during 2008-9 proves that there are
important exceptions to any rule.
Nevertheless, the features of a country’s tax system and government

fiscal policy can have significant implications for the economy.
For example, large budget deficits can be inefficient, and could also
lead to defaults and sovereign debt crises.
1 / 37
Game plan
Fiscal policy is a broad topic.

We will thus restrict our attention to the issues surrounding the
sources and costs of government budget deficits.
The outline of today’s lecture:
1. Tax smoothing: how should the governments conduct fiscal policy?
2. Persistent budget deficits and their costs.
3. Sources of deficits: strategic debt accumulation & delayed stabilization.
4. Sovereign debt crises.
2 / 37
How does the government finance itself?

The government needs to finance a large number of di↵erent public
goods and services: military, healthcare, education etc etc.
The main source of income for the government is taxation.
In any given period, the gvt can finance expenditures either through
1. current tax income, or
2. by borrowing against future tax incomes.
perfectly competitive markets
Ricardian equivalence: in ideal theoretical circumstances, the gvt
choice between taxes and borrowing does not a↵ect the economy.
Specifically, if taxes are lump-sum and households can freely
lend/borrow at the same interest rate as the gvt,
timing of taxes is irrelevant for households’ consumption choices.
at the same interest rate as government
Intuitively, households can borrow/lend to o↵set periods of high/low
taxes, and thus smooth consumption.
) Only the present value of all taxes and transfers matters to households.
3 / 37
Distortionary taxes and tax smoothing
But the real world is not ideal, and gvt financing choices matter.
In practice, very few taxes could be seen as lump-sum.
Instead, taxes often distort behaviour:
E.g. labour income tax reduces people’s incentives to work, capital tax
discourages investment, and VAT distorts consumption choices.
taxes create deadweight loss
This causes inefficiencies, and therefore is costly.
We will now consider a problem of a government that strives to
minimize the costs associated with tax distortions:
The government must ultimately finance its expenditures by taxes.
But can borrow/lend, and so decide on the timing of taxes.
We will see that under plausible assumptions, the government wants to
smooth taxes over time.
4 / 37
Setup
A simple 2-period economy, t = {1, 2}

The path of output Y1 , Y2 is exogenously given.
The government needs to finance expenditures G1 and G2 .
It can levy taxes T1 and T2 in periods 1 and 2 respectively.
It can also borrow/lend at the interest rate r between periods.
The governments taxes cause distortion costs given by:
distortionary costs of✓ taxes
◆
Tt
Ct = Yt f , f(0) = 0, f 0 (0) = 0, f 00 (·) > 0 (1)
Yt
I.e. distortion costs relative to output are a function of taxes relative

to output, and increase more than proportionally with Tt /Yt .
5 / 37
The government’s problem

In period 1, the government chooses the path of taxes to minimize
the present value of distortion costs:
✓ ◆ ✓ ◆
T1 Y2 T2
min Y1 f + f (2)
T1 ,T2 Y1 1+r Y2
The government’s budget constraints in periods 1 and 2 are:
G1 = T1 + D (3)
G2 = T2 - (1 + r)D, (4)
where D is the government debt (or budget deficit) in period 1.
Substitute for D from (4) into (3) to get the gvt’s intertemporal BC:
G2 T2
= T1 +G1 + . (5)
1+r 1+r
) Thus, the present value of current and future government spending
must equal the present value of current and future tax revenues.
6 / 37
Optimal taxes
Substitute for T2 from constr. (5) into the obj. function (2) to get:
✓ ◆ ✓ ◆
T1 Y2 (1 + r)(G1 - T1 ) + G2
min Y1 f + f (6)
T1 Y1 1+r Y2
The first order condition yields:
✓ ◆ ✓ ◆
0 T1 0 T2
f =f , (7)
Y1 Y2
i.e. the gvt. equalizes relative marginal distortion costs in two periods.
Since f is strictly convex, this can only hold if the tax rate (i.e. tax
as a share of output) is constant across periods:
T1 T2
= (8)
Y1 Y2
7 / 37
Implications
To minimize welfare losses associated with distortions, the

government thus strives to keep tax rates constant.
But then how does the government finance abnormally high but
temporary expenditures, such as wars?
) Borrowing!
Also, how does the government finance its (roughly acyclical)
expenditures during recessions when tax revenues are low?
) Borrowing!
The model thus predicts that governments run deficits in such times.
This prediction is consistent with empirical evidence.
8 / 37
Persistent deficits
The tax-smoothing model provides an explanation for variations in

budget deficits over time.
However, many countries run persistent deficits over time.
Moreover, fiscal policies of many countries appear to be unsustainable.
For example, the US government has run large deficits over the
majority of last four decades.
Auerbach & Gale (2017) estimate that if the U.S. fiscal policy remains
unchanged, taxes would need to be raised by 7.7% of GDP (relative to
the current level of 18% of GDP) for gvt to meet its budget constraint.
This is an extraordinarily high imbalance!
Both theory and practice suggest that excessive budget deficits can
be very costly.
9 / 37
Costs of sustainable deficits
Even if a country’s fiscal policy is sustainable in the long run,

extended periods of excessive deficits can have big welfare costs.
Most obviously, running excessive deficits over a prolonged time
involves departure from tax-smoothing:
If current tax rates are below those necessary to satisfy the gvt budget
constraint, then tax rates must be expected to rise in the future.
But this means larger expected future distortions.
Costs of short periods of modestly excessive deficits are likely small.

But persistent and large deficits can lead to substantial distortions:
If the US fiscal policy is unchanged in the next two decades, satisfying
the government budget constraint may require tax rates over 50%.
The distortions from such policy would surely be very large.
10 / 37
Costs of unsustainable deficits

Unsustainable fiscal policies (e.g. an ever rising ratio of debt-to-GDP)
cannot continue indefinitely.
The stop, however, can be sudden and unexpected: it is likely to take
the form of a sovereign debt crisis rather than a smooth transition.
) sharp contractions in fiscal policy, large decline in aggregate demand,
major repercussions in capital markets, government default...
It can trigger a financial crisis, as falling output and asset prices +

disrupted capital markets can hit many firms and fin. institutions.
Low foreign demand for country’s assets can also lead to a large
depreciation of the exchange rate.
) Increases prices of foreign goods, and leads to high inflation.
) Hits firms and banks whose debt is denominated is foreign currency,
making fin. crisis even more likely (aka twin crisis).
After the sharp contraction, the recovery is likely to be only gradual.

11 / 37
Deficit bias: an overview
Many countries run large and persistent deficits despite these costs.
What is the source of this deficit bias?
The answer to a large extend must lie within the political process.
We will consider two simple political-economic models that
illustrate how political process can produce inefficient outcomes.
We begin with a highly simplified version of Tabellini and Alesina’s
(1990) strategic debt accumulation model.
The idea is an elected policymaker may accumulate an inefficient
amount of debt to restrain his (opposition) successor’s spending.
For simplicity, we will focus on the case of diametrically opposed
preferences, and leave out the election process.
12 / 37
Setup
The economy lasts two periods: 1 and 2.
Gvt spending can be devoted to two types of public goods: M or N.
Think, for example, ‘military’ and ‘healthcare’.
Each period, the government receives endowment W.
The period-1 policymaker chooses spending levels M1 and N1 , as well
as debt D subject to the period-1 budget constraint:
M1 + N1 = W + D (9)
The period-2 policymaker chooses M2 and N2 s.t.
M2 + N2 = W - D (10)
Therefore, D > 0 increases period-1 gvt purchases, but reduces

resources available in period 2.
13 / 37
Extreme preferences
There are two types of policymakers:
Type-1 only cares about military, and has per-period utility U(M).
Type-2 only cares about healthcare, and has per-period utility U(N).
Standard assumptions about U: U 0 > 0, U 00 < 0.

Each type only spends resources on public goods they care about, and
is unable to commit to a di↵erent policy.
Suppose that a type-1 policymaker is in office in period 1.
Naturally, he maxes out on military spending:
M1 = W + D and N1 = 0. (11)
Suppose the policymaker elected in period 2 is either type-1 with
probability ⇡, or type-2 with probability 1 - ⇡.
Thus, with probability ⇡, we have M2 = W - D and N2 = 0,
with probability 1 - ⇡, we have M2 = 0 and N2 = W - D.
14 / 37
The choice of government debt

The period-1 type-1 policymaker chooses government debt D to
maximize his expected utility over both periods:
max U(W + D) + ⇡U(W - D) + (1 - ⇡)U(0) (12)
D
The FOC implies:

U 0 (W + D)
=⇡ (13)
U 0 (W - D)
When ⇡ = 1 and the period-2 policymaker is also type-1 for sure, the
period-1 policymaker chooses D = 0 to smooth military expenditure.
However, when ⇡ < 1, U 0 (W + D) < U 0 (W - D), and so D > 0.
This is because there is a chance 1 - ⇡ that the period-2 policymaker
will not share the period-1 policymaker’s preferences.
Moreover, D " as ⇡ #.
15 / 37
Intuition and implications
The intuition is straightforward:

From the point of view of the period-1 policymaker, there is a positive
risk of period-2 policymaker devoting resources to a ‘wasteful’ activity.
Period-1 policymaker thus borrows in order to transfer resources from
period 2 to period 1, when he can devote them to the ‘useful’ activity.
Strategic debt accumulation: period-1 policymaker accumulates

debt strategically to constrain the period-2 policymaker.
The resulting budget deficit is inefficient: all parties would benefit
from a smoother expenditures of M and N.
The mechanism is realistic: for example, a desire to restrain future
spending is often cited in the debates over U.S. fiscal policy.
The model thus provides one plausible explanation of why
governments accumulate sizeable debt and budget deficits.
16 / 37
Delayed stabilization: an overview
A related issue: why do sizeable deficits persist over time?

We saw that deficits can cause large costs to the society both through
distorting taxation and, at the extreme, causing sovereign debt crises.
Thus, a timely fiscal reform to avoid these could benefit most people.
Yet in practice reforms are often delayed as interest groups struggle
over how to divide the burden of the reform.
E.g. even during very costly post-WW1 hyperinflation, policymakers
could not agree if higher taxes should be levied on capital or labour.
In modern times, disagreements are usually over whether the deficit
should be closed by tax increases, or by reductions in gvt spending.
Alesina and Drazen (1991) model: the reform may fail and inefficient
deficits may persist because each party tries to get a better deal.
We now consider a simple model of bargaining based on this idea.
17 / 37
Setup
There are two groups: capitalists and workers, all risk-neutral.

Must agree to reform fiscal policy and how to divide tax burden T > 0.
Otherwise, deep fiscal crisis and both groups receive zero payo↵.
If there is a reform, workers receive pre-tax income W (assume
W > T ) and capitalists receive pre-tax income R > 0.
Capitalists’ pre-tax income R is random, distributed uniformly on
[A, B], with A > 0, and its realisation is known only to capitalists.
If capitalists pay X out of the total tax T necessary for the reform, the
after-tax incomes are R - X for capitalists and W - T + X for workers.
Note that since W > T , the reform is strictly desirable, and any X
between 0 and A makes both groups better o↵.
18 / 37
Bargaining
Simple bargaining: workers make a take-it-or-leave-it proposal X.
If capitalists agree, they pay tax X and reform is implemented.
If capitalists reject the proposal, the reform fails, and payo↵s are 0.
Capitalists accept the proposal as long as R > X.

Thus, the probability P(X) that the proposal is accepted is the
probability that R > X, which is
8
>
<1 if X 6 A
P(X) = B-AB-X (14)
if A < X < B
>
:
0 if X > B,
using the uniform distribution of R on [A, B].

Workers do not know R when they make the proposal, but they can
figure out the probability of acceptance P(X).
19 / 37
Workers’ problem
Workers only receive W - T + X if their proposal is accepted.

Their expected payo↵ is V(X) = P(X)(W - T + X), or, using (14):
8
>
<W - T + X if X 6 A
(B-X)[W-T +X]
V(X) = B-A if A < X < B (15)
>
:
0 if X > B
This is a continuous function: there are no jumps at e.g. X = A or B.

Clearly, the workers will never o↵er X > B.
The workers e↵ectively have two options:
1. O↵er the largest X that is accepted for sure, namely X = A
) in which case their payo↵ is V(A) = W - T + A.
2. O↵er X 2 (A, B) and accept a risk of rejection.
20 / 37
Workers’ problem (cont.)

Workers would pick option #2 i↵ V(X) > V(A) for some X 2 (A, B)
The derivative of V(X) on (A, B) is
B - (W - T ) - 2X
V 0 (X) = (16)
B-A
Note that V 0 (X) is decreasing in X, so if it is negative at X = A (i.e.
B - (W - T ) - 2A 6 0), it is negative on the entire range X 2 (A, B).
But then V(X) is decreasing on (A, B), and workers pick X = A.
Otherwise, workers pick X 2 (A, B) that satisfies the FOC: V 0 (X) = 0.
Putting this all together, the workers optimally choose:
A if B - (W - T ) - 2A 6 0
X⇤ = B-(W-T ) (17)
2 if B - (W - T ) - 2A > 0.
21 / 37
Examples: workers’ expected payo↵ as a function of X
· A the key difference between
the two cases happens - *

⑧
after x passes A
0
intercept is not very meaningful
under this set up
0
Workers pick X⇤ = A Workers pick X⇤ > A
22 / 37
Equilibrium probability of reform
Combining workers’ optimal choice of X in (17) with the probability

that capitalists accept in (14), the equilibrium probability of reform is:
1 if B - (W - T ) - 2A 6 0
P(X⇤ ) = B+(W-T ) (18)
2(B-A) if B - (W - T ) - 2A > 0.
Thus P(X⇤ ) < 1 when B - (W - T ) - 2A > 0.

The reform is less likely to pass when B - A is large, i.e. there is a
larger uncertainty about capitalists (pre-tax) income.
23 / 37
Implications
Key implication: the two sides can fail to agree on the necessary
reform, even though there are packages that make everyone better o↵.
Specifically, when B - (W - T ) - 2A > 0, workers make a proposal
less generous than the one that would be accepted for sure.
Their motive is improve their expected outcome at the expense of the
opposition.
The model predicts that countries with weak governments, where
no single interest party is setting policy, are the least likely to
implement reforms, and so are more likely to run large deficits.
This is intuitive, although the empirical evidence is mixed.
Another implication is that crises (when failure to agree is very costly)

can sometimes be beneficial by spurring essential reforms (homework)
24 / 37
Sovereign debt crises
Our last topic for today (and for the course!) is sovereign debt crises.
A fully fledged theory of such crises is beyond the scope of this course.
We will instead focus on a simple model of a government attempting
to issue debt that sheds light on some of the key issues:
1. Why investors refuse to buy debt at any interest rate?
2. Why can crises occur suddenly and unexpectedly?
We will see that modest changes in a country’s fundamentals can

cause dramatic shifts in outcomes.
And that self-fulfilling crises are likely to be a big part of the story.
25 / 37
Setup
A government has quantity of debt D coming due.

It has no funds immediately available ) wants to roll over the debt.
The government needs investors to hold the debt until the following
period, when it will be obtaining tax revenues.
Investors are risk neutral, with required rate of return r̄, or R̄ ⌘ 1 + r̄.
Government o↵ers them (endogenous) interest rate r, or R ⌘ 1 + r.
The following period’s tax revenue T is a continuous random variable,
with cumulative distribution function F(·) (i.e. Pr(T 6 x) = F(x)).
If T > R D, the government pays the debtholders.
If T < R D, the government defaults ) debt crisis.
Default is all or nothing: if defaults, the gvt. repudiates debt entirely.
26 / 37
The condition for investors to hold government debt
Let ⇡ be the (endogenous) probability of default.

For investors to be willing to hold debt given the prob. of default ⇡,
interest R must be such that they break-even in equilibrium, i.e.
(1 - ⇡)R = R̄ (19)
When the government is certain to repay and ⇡ = 0, we have R = R̄.

As ⇡ ! 1, R ! 1.
Rearrange to get an expression for ⇡ as an (increasing) function of R:
R̄
⇡=1- (20)
R
‘Debt demand’ curve: # 1 of the 2 key equilibrium conditions.
27 / 37
The debt demand curve
28 / 37
The probability of default
The probability of default = probability that tax revenues are

insufficient to pay o↵ debt given the interest rate.
That is, ⇡ equals the probability that T is less than R D, implying
⇡ = Pr(T < R D)
(21)
= F(R D)
‘Default probability’ curve: # 2 of the 2 key equilibrium conditions.

If the probability density function of T is bell-shaped (e.g. normal
distribution), then the default probability curve is S-shaped.
There may be a minimum and maximum values of T , T and T .
In which case ⇡ = 0 for R < T /D and ⇡ = 1 for R > T /D
29 / 37
The default probability curve
30 / 37
Equilibria
An equilibrium is a pair (⇡, R) s.t. both (20) and (21) are satisfied.
) I.e. where the debt demand and the default probability curves intersect.
Due to the S-shape of the default probability curve, there can be
multiple equilibria:
E.g. point A on the graph next slide is an equilibrium with low
probability of default ⇡, and interest rate R only slightly above R̄.
Equilibrium at point B features large probability of default and high R.
In addition, there is always an equilibrium in which default is certain:
Investors are unwilling to purchase debt at any interest rate: R ! 1.
Which implies that the probability of default is indeed ⇡ = 1 by (21).
31 / 37
Equilibria
32 / 37
Stable and unstable equilibria
Are any of these equilibria more plausible then others?

Turns out, equilibrium at B is unstable with respect to small changes
in beliefs, and hence is unlikely to materialise:
) If investors believe that ⇡ is slightly below ⇡B , lower interest R is
needed to induce them to hold debt, given their belief.
But at lower R the actual ⇡ is even less than what investors conjecture
(since now the default prob. curve is below the debt demand curve).
Investors’ estimate of ⇡ is thus likely to fall lower, further reducing R.
The process continues, until point A is reached.
Similarly, if investors believe ⇡ is slightly above ⇡B , R diverges to 1.
Therefore, the two stable equilibria more likely to materialise are:
1. Eq-m at point A with low interest rate and low likelihood of default
2. Eq-m with certain default and investors not buying debt at any R.
We are thus unable to get rid of the multiplicity completely.
33 / 37
Implications
1. Default and debt crises can be self-fulfilling

Which equilibrium materialises depends on investors’ expectations.
If they do not expect default, R is low, and so ⇡ is indeed low.
If they expect default, R is very high, and default is certain.
Similar in spirit to the Diamond-Dybvig model that we saw last week.
2. Though fundamentals matter too:
A rise in R̄ shifts the debt demand curve to the right
A rise in D or a fall in the expected tax income T shift the default
probability curve to the left.
Such developments increase equilibrium ⇡.
Thus high debt, high required rate of return, and low future revenues
all make default more likely.
34 / 37
Implications (cont.)
3. Modest changes in fundamentals can have dramatic consequences:

Suppose a country is in a low R/low ⇡ equilibrium (point A).
A small increase in the safe rate from R̄0 to R̄1 leads to a smooth
increase in equilibrium R and ⇡ (point A 0 on the next slide).
Yet after another similar increase from R̄1 to R̄2 , the two curves no
longer intersect at low levels of ⇡ and R at all.
The only equilibrium now is the certain default!
4. When default occurs, it may often be quite unexpected:
As we saw above, an unexpected increase in interest rates can quickly
tip a country with apparently well-managed sovereign debt into a crisis.
The economy starts at point A with a low probability of default ⇡.
Yet the increase in R̄ completely eliminates that equilibrium, leading to
a discontinuous jump.
Thus, the resulting default is unexpected: it is not preceded by high ⇡.
35 / 37
The e↵ect of an increase in the safe interest rate R̄
36 / 37
Conclusions (for the course)
This is the end of the course.

We have studied a large number of topics, ranging from issues of
economic growth to financial and sovereign debt crises.
Modern macroeconomics does not end here: there are many more
important and interesting issues, and a lot more depth to each topic.
) The researchers never sleep (literally)!
Still, I hope this course has equipped you with a useful toolbox of
classic models to help you think about the complex world.
) Next time you read an FT article about central banks’ limited options
to respond to a crisis due to low nominal rates, think liquidity trap.
Or when unemployment raises and wages appear to fail to adjust down,
remember that there may be good reasons for that.
Thank you!
37 / 37
Final Slides for
Paul Klemperer’s 1 lecture
st
NOT FOR CIRCULATION

Free book on my website
Objectives
• Introduce main types of auctions
• Discuss main pitfalls of auction design
• Illustrate broader economic principles,

especially industrial economics and business strategy
This week
1. Introduction
2. Standard single-unit auctions
3. Standard multi-unit auctions

- oligopsony behavior: collusion
- complements, and monopoly bundling
Next week
4. Standard multi-unit auctions for homogeneous goods

- monopsony behaviour: demand reduction
- more on collusion
5. Common values
- almost common values
- entry deterrence
6. Examples
- auctions of Treasury Bills & Bonds, etc.
- the 3G auctions
- the Bank of England’s new auction
- some recent developments
My teaching
• My exposition will (deliberately) not be too technical
(Responding to previous surveys, and noting we are at the
end of a long term)
• In conjunction with the notes and the classes (which

are more technical), you will have enough information
to do well on the exam
– The problem set is already posted on Canvas.
– Preliminary slides will be posted the night before each lecture;
the final versions of all slides will be posted after each lecture.
• Please feel free to Ask Questions in Class

Examples of auctions
• Oil-drilling rights
• Government debt
• Mobile-phone licenses
• Online Ad Auctions
• Permits for CO2 emissions (and other pollutants)
• Central Bank funds in Financial Crisis
• B2B Exchanges
• Takeover battles
• Initial Public offerings
• Electricity pools
Basic Model: “Independent Private Values”
• 1 unit for sale; n bidders
• Bidder i ’s value, v i ,is known only to him/herself.
• Each bidder’s value is drawn independently, and from

a common distribution, F (); F ( v ) = 0
• Each bidder wants to maximise her/his expected profits

– i.e., Bidders are risk-neutral
Basic Model: “Independent Private Values”
• 1 unit for sale; n bidders
• Bidder i ’s value, v i ,is known only to him/herself.

– e.g., Last 4 digits of your mobile phone number* (in pennies)
• Each bidder’s value is drawn independently, and from

a common distribution, F (); F ( v ) = 0
– F uniform [0, £99 − 99]
• Each bidder wants to maximise her/his expected profits

– i.e., Bidders are risk-neutral
*If you have a British and a foreign mobile phone, use your British number. If you have more than one
British mobile phone number, please use the number which is larger (looking at all the digits)
Common Auction Formats
• First-price sealed-bid Auction:

– Bidders simultaneously write down their bids
– Highest bidder wins the object and pays the bid she wrote
– e.g., oil, mineral rights, real estate, construction,
procurement contracts
• Please write down your bid!

– In the event of a tie among k players, each will pay
me (1/k) of their bid and receive (1/k) of the prize
Common Auction Formats
• First-price sealed-bid Auction:

– Highest bidder wins the object and pays the bid she wrote
– e.g., oil, mineral rights, real estate, construction,
procurement contracts
• Dutch Auction:
– Price starts high and gradually falls until one bidder agrees
to buy the object at the price
– e.g., flowers in Holland, some fish and agricultural products
Dutch Auction
-
First-price sealed-bid Auction I I

I
t
n −1
0 D
 
if i bids b she earns ( vi − b ) with probability  b 
 n −1 
 V 
Prewinning one agent) Pr)maxi "n' vis< b)
=
 n 
n
=Pr(max[ViS b)
( n −1)
<
 
-b  b 
to maximise ( vi − b )  n − 1 
  
  n  V 
=n
i  

set (...............) = 0 (cf., problem set)
b
 n −1
 b=  vi
 n 
• How should you bid in Dutch Auction?

Dutch Auction & Sealed-Bid Auction
• How should the price at which you jump in, in a Dutch
auction compare with the price you bid in a (first-price)
sealed-bid auction?
• In any auction for a single prize:

– The price you jump in, in a Dutch auction = the price you bid in a
(first-price) sealed-bid auction
• They are the same game!

– Bidder’s strategy set = set of all prices
– Bidder who chooses highest price wins & pays bid
• But do people play the same way in these games?

– If not, then game-theorists’ descriptions are incomplete
• Optional reading about an example of when this mattered:
pp139-142 of my book (section C4, pp17-20, of version on web)
English Auction
• English (or “ascending”) Auction:
– Price starts low and is gradually raised until only one
bidder remains
– That bidder wins the object and pays the price at which the
runner-up dropped out
– e.g., art, antiques, etc. at Christie’s, Sotheby’s etc.
• How should you bid in an English “mobile#” auction?

English Auction and Japanese Auction
• English (or “ascending”) Auction:
– Price starts low and is gradually raised until only one
bidder remains
– That bidder wins the object and pays the price at which the
runner-up dropped out
– e.g., art, antiques, etc. at Christie’s, Sotheby’s etc.
– We will model this as a Japanese Auction

Second-price sealed-bid Auctions
• Second-price sealed-bid Auction:
– Highest bidder wins object and pays the second-highest bid
– e.g., stamps, some internet auctions
• How should you bid in a Second-price sealed-bid

“mobile#” Auction?
Proxy Auction and Japanese Auction
• Proxy Auction:
– A Japanese auction except that each bidder, instead of
bidding itself, instructs a “proxy” to bid on its behalf up to
some pre-specified maximum
– Similar to how eBay, etc., work in practice
– We will ignore the final bid increment
• In an ordinary Japanese "mobile#" auction, you

would lose nothing by instructing a friend (or proxy)
to bid for you up to your actual value
 the proxy “mobile#” auction has the same outcome

as an ordinary Japanese “mobile#” auction
Proxy Auctions & Second-price Auctions
• It is a dominant strategy to bid your actual value in a proxy
auction with independent private values
• A proxy auction is identical to a 2nd-price auction

– every bidder chooses a number (the instruction to its proxy);
the bidder who chooses the highest number wins, and pays the
number chosen by the runner-up
• So it is a dominant strategy to bid your actual value in a

2nd-price auction with independent private values.
– See the notes for an alternative proof of this
Expected Revenue for Uniform Distribution on [0, V ]
 n − 1  
• E(revenue from F or D) = E   (highest value) 
 n  
• E(revenue from S or J) = E (2nd highest value)

Order Statistics for Uniform Distribution on [0, V ]
• If n independent draws from uniform on [0,V ]
• Useful facts:
– (1) Conditional on kth highest draw = v*, the lowest n – k draws
are independent draws from uniform on [0, v*]
v*
0
Order Statistics for Uniform Distribution on [0, V ]
• If n independent draws from uniform on [0,V ]
• Useful facts:
 n +1− k 
– (2) E(k th highest) =
 V
 n +1 
V
e.g., n =3:
0
 n − 1  
(1) E(revenue from F or D) = E  (highest value) 
 n  
 n − 1    
=   V =  V
 n    
(2) E(revenue from S or J) = E (2nd highest value)
 
= V
 
 n − 1  
(1) E(revenue from F or D) = E  (highest value) 
 n  
 n − 1  n   n −1 
=   V =  V
 n  n + 1   n +1
(2) E(revenue from S or J) = E (2nd highest value)
 n −1 
= V
 n +1 
• So F, D, S and J, are all equally profitable in expectation!
• The formulae depend on the Uniform Distribution, but the

“equally profitable” result doesn’t!
Revenue Equivalence Theorem
Absent entry and collusion problems, under natural
assumptions, when selling a single object:
• All auctions with no reserve price yield the same expected

revenue for the seller
– Formal statement:
“Assume each of n risk-neutral potential buyers has a
privately-known value (or signal in the common-values case)
independently drawn from a common distribution F(z) that is
strictly increasing and atomless on  z , z 
Then any auction mechanism in which
(i) the object always goes to the buyer with the highest value (or
signal), and
(ii) any bidder with value (or signal) z expects zero surplus,
yields the same expected revenue, and results in a buyer
with value (or signal) z making the same expected payment.”
Comparing Auction Designs
Absent entry and collusion problems, under natural
assumptions, when selling a single object:
• All auctions with no reserve price yield the same expected

revenue for the seller
• All auctions with a suitable (public) reserve price yield the

maximum possible expected revenue for the seller
– So modifications such as entry fees, minimum bid increments,
matching rights, discounts to favoured bidders, etc., cannot help,
and may hurt under these assumptions
– Moreover, absent collusion problems, using a reserve price
typically only makes much difference if you have very few bidders
Are all Auctions Equally Good?
• “Revenue Equivalence Theorem” ignores:
– Bidders’ risk-aversion
• See problem set
– Auctioneer’s risk-aversion may overpay to avoid losing
• F (or D) preferred
– Budget constraints
• F (or D) often most profitable
– Correlation of information means they can see others' bidding behaviour
• E (J) often most profitable ascending auction pushes price up

=>
– Asymmetries efficiency also likely to

=>
• F (or D) probably most profitable, but E (J) more efficient

Are all Auctions Equally Good?
• “Revenue Equivalence Theorem” also ignores:
– Multi-unit issues
• See Later
• And especially:
since an auction is just a market, so standard economics
applies …
Key Problems in Auction Design
• Same problems as faced by competition agencies

– e.g., UK Competition and Markets Authority
or US Federal Trade Commission
1. Collusion
2. Entry deterrence
An auction is just a market

and standard economics applies
What must you know
to be an auctioneer?
Lots
not much
an auction is just a market
and standard economics applies,
especially in multi-unit contexts LN
Simple Auctions for Multiple Units of Multiple Products
problem:hard choice between thr's second preferred product
or next week's first preferred product
inefficiency of allocation
->
Sequential Ascending (e.g., Christie’s & Sotheby’s, Ebay)
S Sequential Sealed-Bid (e.g., Oil)
Simultaneous Sealed-Bid (e.g., TV Franchises)

may need and round auction for unallocated
goods
problem:winner's curse,
Simultaneous Ascending (e.g., Mobile Phones)

Simultaneous Ascending Auction: the UK 3G Auction
• UK 3G Auction March-April, 2000
• 5 licenses: A, B, C, D, E
--2 large, 3 small, but all different from each other
• bidders permitted to win at most 1 license each
• 13 bidders:
Vodafone Airtouch, BT Cellnet, Orange,
One2One, TIW UMTS (UK) [Winners]
NTL Mobile, WorldCom Wireless (UK), Crescent Wireless, 3G(UK),
Epsilon, Spectrum, One.Tel Global Wireless, Telefonica UK [Losers]
Simultaneous Ascending Auction: the UK 3G Auction
larger license small licenses
leading in: A B C D E
round 1 TIW Or - Cres EPS
other 9 now must bid on C, or overbid by 5% (or more), or quit
round 2 (no quits) NTL Tel WC Cres EPS
other 8 must now overbid by 5% (or more), or quit
round 3 (no quits) NTL TIW WC 3GUK 1-2-1
other 8 must overbid by 5% (or more), or now quit
round 94 (1st quit) Tel BT TIW NTL Or
Increments reduced by 1½% as bidders quit
round 150 (8th quit) TIW Vod BT 1-2-1 Or

Simultaneous Ascending Auction
Proposition: "straightforward" myopic bidding is optimal

(i.e., it is optimal to bid in each round as if:
this round is the final round & no one else will bid in this round
i.e., choose object for which (value – current minimum required bid)
is largest, and bid the minimum required bid on that)
Intuition:
- if your next bid wins, it was clearly optimal
- if your next bid is topped, you can bid again,
and you could not have won any object cheaper than its current price
(within a bid increment)
Assumptions: - others bid this way

- private values
- no bidder wants to win more than 1 unit
Proposition: The outcome is efficient
Assumptions: - "straightforward" bidding

- private values & no bidder wants to win more than 1 unit
- no budget constraints
Proof: let final price of object 𝑖 be 𝑃𝑖

let valuation of winner of 𝑖 for object 𝑗 be 𝑉𝑖𝑗
if winner of 𝑖 did not win 𝑗 :
𝑉𝑖𝑖 − 𝑃𝑖 ≥ 𝑉𝑖𝑗 − 𝑃𝑗 (or winner of 𝑖 would have bid differently)
𝑉𝑗𝑗 − 𝑃𝑗 ≥ 𝑉𝑗𝑖 − 𝑃𝑖 (or winner of 𝑗 would have bid differently)
⇒ 𝑉𝑖𝑖 + 𝑉𝑗𝑗 ≥ 𝑉𝑖𝑗 + 𝑉𝑗𝑖
⇒ inefficient to swap allocation of 𝑖 and 𝑗
(and inefficient to swap with a non-winner, since all its values below prices)
• SAA is an efficient mechanism

under the assumptions above
and yields the same result as a benevolent central planner
(& central planner would have to know every bidder's value for every object,
and then solve an optimisation problem)
• and a mechanism for price discovery:

if bidders are uncertain about their values multiple-round bidding
allows them to work out how much they value different objects
(with private values, and even more so with common values)
BUT:
• With multi-unit demand, there can be problems (see later)
• SAA is not necessarily most profitable
Bundling (with complements or substitutes)
Kids movie Romance Both

Family 20 10 30
Couple 10 20 30
↓second price auction
simultaneous ascending auction

(or independent ascending auctions) efficient,
but low revenue (10+10=20)
ascending auction for bundle inefficient,

but more profitable (30)
Austrian 3G Auction
• 6 bidders for 12 lots; ascending auction
– Asking price rises until bidders ask for 12 units in total
• Telekom Austria said beforehand:

– "would be satisfied with just 2 of the 12 lots ..if the 5 other
bidders behave similarly, it should be possible to get the lots on
sensible terms ...[but it] would bid for a 3rd block if one of its
rivals did ”
• Result: Bidding ended almost immediately

Austrian 3G Auction

rivals did ”
• cf. Dominant firm in “normal” market says:

– "would be satisfied with market share of just 1/6
...if the other five firms also stick to 1/6 of the market
each, it should be possible to sell at high prices
...[but it] would compete aggressively for more than
1/6 of the market, if one of its rivals did "
Abilities firms require to collude
1) agree division of market
2) detect defection from agreement
3) credibly punish defection
4) deter new entry
• See, e.g., Waterson’s text book

A US Spectrum Auction
Marshalltown, IA Rochester, MN Waterloo, IA
(lot # 283) (lot # 378) (lot # 452)
McLeod USWest McLeod USWest McLeod USWest
56,000 287,000
... ...
568,000
689,000
723,000
795,000
875,000 313,378
345,000
963,000
62,378 1,059,000
69,000
What are the advantages of this kind of auction?

Problems with Complements
Hotel Hotel Both

room A room B rooms
Couple 25 25 25
Family 0 0 30
C just bids on lower price room,

so F can’t win either room for less than 25;
if F bids at all it loses money.
The exposure problem.
Problems with Simultaneous Ascending Auctions
1. Standard Oligopsony Behaviour
e.g., mobile-phone licenses
2. Complements
e.g., airport takeoff- and landing-slots
3. Auctioneer (unlike bidders) can’t vary what to

sell (or buy) based on prices
e.g., bond sales
4. Speed (and cost)

e.g., financial markets
1. Oligopsony Behaviour
2. Complements
3. Auctioneer (in)flexibility
4. Speed (and cost)
Simple Alternative: Sealed-bid Combinatorial Auction
Bidders offer prices for alternative combinations
Seller accepts revenue-maximising set
(accepting at most one package per bidder)
Addresses problems in rough-and-ready way
BUT often hard to know how (and on what) to bid
(also often disadvantages “small” bidders)
so may not be very efficient (or profitable)
1. Oligopsony Behaviour
2. Complements
3. Auctioneer (in)flexibility
4. Speed (and cost)
New Auctions
(a) Clock-Combinatorial Package auctions address 1 & 2:
e.g., UK’s 4G auction
(b) Product-Mix auction addresses 1, 3 & 4:
e.g., Bank of England's Long Term Repo auctions
[see films on front page of www.paulklemperer.org
and next lecture]
Many-unit Auctions with Homogenous Units
1. (Multi-unit) Ascending
e.g., Spectrum,
early auction of Greenhouse Gas Emissions Permits
2. Uniform Price
e.g., UK Electricity until 2001, UK Gold, some IPOs,
Greenhouse Gas Emissions Permits,
US, some UK, and many other countries’ Treasury Bills &
Bonds
3. Discriminatory, or “Pay-Your-Bid”
e.g., French, German, some UK, and many other countries’
Treasury Bills & Bonds
Please make sure
you understand
these Payment Rules
before next time
Bidding in a Many-Unit Auction
Price Price
0 0
Bidder A’s Bidder B’s

Quantity Quantity
LN
Bidding in a Discriminatory (Pay-as-bid) Auction
Price Price
A’s bid
B’s bid
A’s
payment B’s
payment
0 0
Bidder A’s Quantity Bidder B’s Quantity
Bidding in a Uniform Price Auction
Price Price
A’s bid
B’s bid
A’s
B’s
payment
payment
0 0
Please make sure
you understand
these Payment Rules
before next time
Coins Game
Coin Game
1. Write down your estimate of amount in bag

– I will (later) give a prize for the best estimate
– Do not let anyone else see what you write down
2. Write down your sealed-bid for amount in bag

– The person whose bid is highest will pay me their bid
– I will pay that person the amount of money in the bag
• In the event of a tie among k players, each will pay me (1/k) of their
bid and receive (1/k) of the prize
• NOW I will find out who won the 2nd game

– i.e., I will find out highest sealed bid
– Do not tell me your estimate
Coin Game
1. Write down your estimate of amount in bag

– I will (later) give a prize for the best estimate
2. Write down your sealed-bid for amount in bag

– The person whose bid is highest will pay me their bid
3. Ascending (Japanese) Auction for amount in bag

– The winner will pay me the final price
• The same person may win more than once

How Should You Bid in Coins Game?
• Suppose bidders act as if the value of the coins is their

estimate of the value before any information is revealed
by other bidders’ behaviour
• Then the winner is likely to have paid too much

This is the “winner’s curse”
• If you win, you probably had the highest estimate:

– Knowing that everyone else has lower estimates,
you probably want to revise your estimate downwards
How Should You Bid in Coins Game?
• You should drastically shade your bid below your
initial estimate in a sealed-bid auction
• In an ascending auction you can learn a lot about

rivals’ estimates as you go along, but you must still
shade your bid down
Polar Cases of Bidder Values
• “Common values” (e.g., Coins game)
– The eventual value of the object will be the same for all bidders,
but different bidders may have different estimates of this value
– Each bidder’s estimate of the value would be altered by
knowledge of other bidders’ estimates
• e.g., an oil-field (when bidders all have the same production costs)
• “Private values”
– Each bidder has some value for the object (i.e., maximum price
she is willing to pay) which does not depend on the information
or values of other bidders
• e.g., wine of known characteristics, which the winner will drink on her own
(but if it was for resale, it might be common values)
Final Slides for
Paul Klemperer’s 2 lecture
nd
NOT FOR CIRCULATION

Free book on my website
Objectives
• Introduce main types of auctions
• Discuss main pitfalls of auction design
• Illustrate broader economic principles,

especially industrial economics and business strategy
Last week
1. Introduction
2. Standard single-unit auctions
3. Standard multi-unit auctions

- oligopsony behavior: collusion
- complements, and monopoly bundling
This week
4. Standard multi-unit auctions for homogeneous goods

- monopsony behaviour: demand reduction
- more on collusion
- auctions of Treasury Bills & Bonds, etc.
5. Common values
- almost common values
- entry deterrence
6. More Examples
- the 3G auctions
- the Bank of England’s new auction
- some recent developments
My teaching
• My exposition will (deliberately) not be too technical
(Responding to previous surveys, and noting we are at the
end of a long term)
• In conjunction with the notes and the classes (which

are more technical), you will have enough information
to do well on the exam
– The problem set is already posted on Canvas.
– The final versions of the slides will be posted after the lecture.
• Please feel free to Ask Questions in Class

Many-unit Auctions with Homogenous Units
1. (Multi-unit) Ascending
e.g., Spectrum,
early auction of Greenhouse Gas Emissions Permits
2. Uniform Price
e.g., UK Electricity until 2001, UK Gold, some IPOs,
Greenhouse Gas Emissions Permits,
US, some UK, and many other countries’ Treasury Bills &
Bonds
3. Discriminatory, or “Pay-Your-Bid”
e.g., French, German, some UK, and many other countries’
Treasury Bills & Bonds
A German Spectrum Auction (1999)
• 10 lots; minimum raises of 10%
• 2 credible bidders: Mannesman and T-Mobil
• Round 1
– Mannesman bids DM 18.18m/MHz for lots 1,...,5
– Bids DM 20.00m/MHz for lots 6,...,10
– T-Mobil bids even lower on all lots
• T-Mobil’s manager later says

– “There were no agreements with Mannesman.
But Mannesman’s first bid was a clear offer.”
• Round 2
– T-Mobil bids DM 20m/MHz for lots 1,...,5
– Doesn’t bid again for lots 6,...,10
– And no subsequent bids, so auction closes.
Bidding in a Discriminatory (Pay-as-bid) Auction
Price Price
A’s bid
B’s bid
A’s
payment B’s
payment
0 0
Price Price
A’s bid
B’s bid
A’s
B’s
payment
payment
0 0
Demand Reduction
Price Price
A’s true demand
B’s bid
~lossduring
gain from ->
price
lower
A’s B’s payment
payment
0 0

With more than 1 unit, a bidder does not “tell the truth”
--uniform price auction are not “just like 2nd price auctions”
Shading the bid loses red area of profit; gains banded green area
(The bidder does always bid full value for the 1st unit.)
Demand Reduction
Price Price
A’s true demand
B’s bid
A’s B’s payment

payment
0 0

The “smaller” the bidder relative to the rest of the market (B),
the less the price responds to a proportional shading of its bid
(the thinner the banded green area), so the less the bidder shades
Optimal Bidding in a Uniform Price Auction
Price uniform-price bid schedule

uniform-price bid schedule for
bidder who does not much
affect market price*
Marginal
Value
(= “true
demand”)
0
Bidder’s Quantity
*a very small bidder bids his true demand in a uniform-price auction
Optimal Bidding in a DiscriminatoryAuction
Price
discriminatory
bid schedule Marginal
Value
(= “true
demand”)
0
Bidder’s Quantity
Bidders in discriminatory auctions reduce bids for

high-value units a long way below their true demands, because they
don’t want to pay (much) more than the lowest winning price.
Optimal Bidding in Multi-Unit Auctions
Price uniform-price bid schedule

uniform-price bid schedule for
bidder who does not much
affect market price
discriminatory
bid schedule Marginal
Value
(= “true
demand”)
0
Bidder’s Quantity
Bidders reduce bids below their true demands, in different ways
in different auctions → Ambiguous which auction is more efficient
Uniform auction is more efficient if all bidders small
& uniform price auction is easier for bidders to bid sensibly in
Standard non-cooperative oligopsonistic
Demand Reduction in a Uniform Price Auction
Price Price
A’s bid
B’s bid
A’s B’s payment

payment
0 0
Bidder A’s Bidder B’s

Quantity Quantity
(Implicitly) Collusive Bidding in a Uniform
Price Auction
Price Price
0 A’s payment B’s payment 0

Price Auction
Price Price
A’s payment
B’sB’s
payment
payment
0 0
A German Spectrum Auction (1999)
• 10 lots; minimum raises of 10%
• 2 credible bidders: Mannesman and T-Mobil
• Round 1
– Mannesman bids DM 18.18m/MHz for lots 1,...,5
– Bids DM 20.00m/MHz for lots 6,...,10
– T-Mobil bids even lower on all lots
• T-Mobil’s manager later says

– “There were no agreements with Mannesman.
But Mannesman’s first bid was a clear offer.”
• Round 2
– T-Mobil bids DM 20m/MHz for lots 1,...,5
– Doesn’t bid again for lots 6,...,10
– And no subsequent bids, so auction closes.
Price Auction
Price Price

Price Auction--Fishing Quotas in the Faroe Islands
(Marszalec, Teytelboym, and Laksá, 2020)
Source: Map data @2020 Google

• Three incumbent firms bid for quota of groundfish (2018)
• Bidders allowed to bid demand curves

(at most ﬁve price-quantity bids)
• Supply of quota was pre-announced and ﬁxed
• “Crank-handle”/“Hockey-stick” bids (almost) coincide

with firms’ grandfathered shares prior to the auction
Price
14 Bid schedule
Grandfathered supply
Quantity
Bidder A’s demand and grandfathered share

held prior to the auction
Price
14 Bid schedule
Quantity
Bidder B’s demand and grandfathered share

Price
16
Bid schedule
Quantity
Bidder C’s demand and grandfathered share

Price
Incumbents’ demand
Supply
Quantity (tonnes)
Incumbents’ aggregate demand function (bids)

and pre-announced and fixed supply of quota
Difficulties with (Implicitly) Collusive Bidding
What if a 3rd buyer enters? (4th buyer in the Faroes example)
Price Price
0
A’s payment B’s payment 0
C’s
Bidder A’s Quantity Bidder B’s Quantity Quantity
Price Price
0
A’s payment B’s payment 0
Bidder C’s C’s
Bidder A’s Quantity Quantity Bidder B’s Quantity Quantity
Price C’s Payment Price
A’s payment B’s payment
0 0
Bidder C’s C’s
Bidder A’s Quantity Quantity Bidder B’s Quantity Quantity
What if seller reduces supply?
Price Price
A’s payment B’s payment

0 0
0
Price Auction
Price Price

usually difficult for bidders to achieve

and usually easy for auctioneer to guard against
Price Auction
Price Price


electricity may be an exception
Price Auction: “Hockey Sticks”
Price Price
A’s receipts B’s receipts
0 0

electricity may be an exception
UK Electricity Auction
Electricity “Pool” was a uniform price auction until 2001
the “New Electricity Trading Arrangements” (NETA)

changed it to a Exchange + Discriminatory Auction
(probably due to concern about excessive prices)
Was this a good idea?

Electricity—What auction would small new entrant prefer?
In uniform price (or ascending),

small bidders protected since don’t set price
But in discriminatory, bidders are playing “guess the price”

→disproportionately costly for small bidders*
Small bidders prefers uniform (or ascending).
→ Worry more about entry in discriminatory

but worry more about collusion in uniform
*can sometimes resolve problem by allowing non-competitive bids

Treasury Bills & Bonds, etc.
Many potential bidders & Low entry costs.
So standard collusion and entry arguments =>

uniform-price auctions probably fine,
especially with variable supply,
&
discriminatory auctions probably also fine,
if non-competitive bids allowed
and/or if there is a good resale market.
Probably have cap on any bidder's winnings,

to not permit monopoly power (after auction).
Treasuries--More Arguments for Uniform
"Fairer" and Less Regret
"More like" Pre- and Post- Markets (Wilson)
Harder for any Bidder to Win Very Large Share
Information Less Valuable
=> Less Resources Spent Collecting Information
& Better Pre-Market (Less Information Withheld)
& Maybe Less (Explicit) Collusion (Friedman)
& Bidding Definitely Easier for Smaller Bidders
(unless discriminatory allows non-competitive bids)
Also, anyway Favours Smaller Weaker Bidders
=> More Entry (and so Less Collusion)
At Least with Many Bidders:
More Efficient, More Informative & Bidding Easier
Investment Bankers Dislike
Treasuries--More Arguments for Discriminatory
Stops "Free-Riding on Information"
Less (Implicit) Collusion
May Sometimes Be More Efficient
and Sometimes Be More Profitable

Treasury Bills & Bonds, etc.
1/3 of OECD countries (inc. Fr, Ger) use discriminatory

1/3 of OECD countries (inc. US) use uniform
1/3 of OECD countries of countries (inc. UK) use both
(as of 2018 -- 30 years ago was mostly discriminatory)
Central Banks OMOs: Fed, BoE, ECB use both

Canada, Japan, Aust, NZ, Brazil use discriminatory
In most cases, no compelling evidence either way

-- reasons for switching often arbitrary
Wallet Games*
1. I will designate two players to each write down
Z = last 3 digits of her/his (British) mobile phone**
*I use the terminology in the literature—we will play with phone numbers
**If you have more than one, choose the higher one based on all the digits
(not just the last 3)
Pure Common Values: Ascending Auction
v1 = v2 = v = z1 + z2
v1 = v2 = v = z1 + z2
• Type 𝑧1 of player 1 quits at 𝑏 𝑧1 when she is just
indifferent about whether or not opponent quits
– If strictly happy to win at this price, she would quit later
– If unhappy to win at this price, she would quit earlier
• Assume a symmetric equilibrium

⇒ Type 𝑧1 quits at same price as opponent with type 𝑧2 = 𝑧1
⇒ If type 𝑧1 wins at (close to) her quitting price, 𝑏 𝑧1 ,
the type she will have beaten will have type 𝑧2 (close to) 𝑧1
⇒ At the price she quits, 𝑏 𝑧1 ,she is just indifferent about winning

on the assumption that 𝑧2 = 𝑧1 , i.e., that 𝑣1 = 𝑧1 + 𝑧1 = 2𝑧1
⇒ Type 𝑧1 of player 1 quits at price 𝑏 𝑧1 = 2𝑧1
v1 = v2 = v = z1 + z2
• Type 𝑧1 of player 1 quits at price 𝑧1 + 𝑧1 = 2𝑧1
– e.g. 𝑧1 = 900 bids up to, and then quits at, price 1800p (=£18)
• Check:
– If wait and win at 1800 + 𝜖 ,
1800+𝜖
𝑣1 = 900 + < 1800 + 𝜖, so lose money
2
– If quit at 1800 − 𝜖 ,
1800−𝜖
𝑣1 ≥ 900 + > 1800 − 𝜖, so may gain by waiting
2
• This is independent of distributions of 𝑧1 and 𝑧2

v1 = v2 = v = z1 + z2
• Type 𝑧1 of player 1 quits at price 𝑧1 + 𝑧1 = 2𝑧1
– e.g. 𝑧1 = 900 bids up to, and then quits at, price 1800p (=£18)
– e.g. 𝑧1 = 030 bids up to, and then quits at, price 60p
• Question: At price 60p, 𝑧2 is at least 030

— So why doesn’t 𝑧1 = 030 bid further?
— 𝑣1 is at least 60p ⇒ On average 𝑣1 > 60p
• Winner’s Curse!
61 1
— If 𝑧1 = 030 wins at 61p, then 𝑣1 ≤ 30 + = 60
2 2
— So 𝑧1 should have quit earlier!
Pure Common Values: Sealed-bid Auction
v1 = v2 = v = z1 + z2
• How should you bid?
• Those with no information should bid zero
• For those with information, it depends on the distributions

of 𝑧1 & 𝑧2
– e.g. if 𝑧1 & 𝑧2 are independently and uniformly distributed
on [0,K], for any K>0 (here K=999), then, in equilibrium,
bidder i bids 𝑧𝑖
(Slightly) Asymmetric Case –
Ascending Auction
• Prize is 𝑣1 = 𝑧1 + 𝑧2 + £1 if Bidder 1 wins
• Prize is 𝑣2 = 𝑧1 + 𝑧2 if Bidder 2 wins
• Q: Up to what prices should they bid?

• A: if bidders bid as before (up to 2 x signal)
Bidder 1 gets £1 larger prize whenever she wins, so can
bid 2 x £1 = £2 more than before (for any given signal)
𝑝
– If she wins at price 𝑝, she infers 𝑧2 = ,
2
𝑝
and so her value is 𝑣1 = 𝑧1 + + 1
2
– She bids until she is indifferent between winning and losing, so bids until 𝑝 = 𝑣1 ,
𝑝
i.e. until 𝑝 = 𝑧1 + + 1, so until 𝑝 = 2𝑧1 + 2
2
i.e. £2 more than she bid originally
Ascending Auction

⇒ Bidder 2 will then receive £2 ÷ 2 = £1 less if she wins, so
should quit 2 x £1 = £2 earlier than before (for any signal)
𝑝−2
– Now if bidder 2 wins at 𝑝, she infers 𝑧1 = .
2
𝑝−2
So her value is now 𝑣2 = + 𝑧2 , that is, £1 less than before. Bidder 2 bids
2
until she is indifferent, so bids £2 less than before.
Ascending Auction

⇒ Bidder 2 will then receive £2 ÷ 2 = £1 less if she wins, so
should quit 2 x £1 = £2 earlier than before (for any signal)
⇒ Bidder 1 gets additional £2 ÷ 2 = £1 whenever she wins,
so should bid £4 more than before (for any signal)
⇒ Bidder 2 will then receive £4 ÷ 2 = £2 less if she wins,
so should quit £4 earlier than before (for any signal)
⇒ Bidder 1 gets additional £4 ÷ 2 = £2 whenever she wins…
Ascending Auction

so should quit 2 x £1 = £2 earlier than before (for any signal)
⇒ Bidder 1 gets additional £2 ÷ 2 = £1 whenever she wins,
so should bid £4 more than before (for any signal)
so should quit £4 earlier than before (for any signal)
⇒ Bidder 1 gets additional £4 ÷ 2 = £2 whenever she wins…
⇒… … …
• Bidder 2 quits bidding at price ≤ 𝑧2 , in equilibrium
– and Bidder 1 never quits, in equilibrium
Sealed-bid Auction vs Ascending Auction
• revenue-equivalence fails in asymmetric case

Example: An early U.S. Airwaves
(Ascending) Auction
• Consider the (only) Los Angeles license
• Value very uncertain

– mobile telephony was very new
• Value similar to all bidders
• But Pacific Telephone had a slightly higher actual value

than the other bidders
– Pacific Telephone had
• Well-known brand name
• Data base about potential customers
• Executives with local knowledge
(Ascending) Auction
• Prices (per head of population; 3 largest regions)
Chicago $31
Los Angeles $26

– Pacific Telephone advantaged
New York $17

– Sprint advantaged
• But most commentators believed Los Angeles’

demographics much better than Chicago’s or New York’s
 Los Angeles had highest value
Example: Takeover Battles
• When one bidder has a “toehold”
– i.e., a small ownership stake
• Larger toeholders win more often (good evidence)

and at lower prices (weak evidence)
• i.e., if target is “common values”, then toehold has

strategic value, not just direct value
Entry Deterrence
• Disadvantaged bidder wins rarely and earns very little
when he does win
• Will disadvantaged bidder bid at all?

⇒ Becoming a stronger bidder can scare off rivals
– an extreme case of strategic substitutes*
(downward sloping, almost coincident, reaction curves)
ß2 ß1(ß2)
• ß1 ~ % of own signal that 1
bids up to in symmetric case
• ß2 ~ % of own signal that 2 ß2(ß1)
bids up to in symmetric case
ß1
*in terminology of Bulow, Geanakoplos and Klemperer (JPE, 1985)
Entry Deterrence
• Disadvantaged bidder wins rarely and earns very little
when he does win
• Will disadvantaged bidder bid at all?

⇒ Becoming a stronger bidder can scare off rivals
– an extreme case of strategic substitutes
(downward sloping, almost coincident, reaction curves)
• So entry deterrence is very easy
ß2 ß1(ß2)
• ß1 ~ % of own signal that 1
bids up to in symmetric case ß1’(ß2)
• ß1’~ % of own signal that 1 ß2(ß1)
bids up to when has a “toehold”
– see strictly optional problem A12
on website for precise argument ß 1
cf. Dixit EJ (1980)--the classic article on entry deterrence in oligopoly
Example: Takeover of Wellcome
• Glaxo (who had particular synergies) offered £9 billion
• Wellcome sought other bidders including Roche, Zeneca
• Zeneca (at least) willing to pay > £10 billion
• But Zeneca refuse to enter ascending auction, which it

would expect to lose
• Result: No auction ⇒ Glaxo’s first bid wins

– Wellcome’s shareholders forgo £ billions
• What should Wellcome’s Board have done?
·
Example: Takeover of Wellcome
• Glaxo said beforehand:

“would almost certainly top a rival bid”

“would almost certainly undercut any new entrant’s price”
(Ascending) Auction
• Prices (per head of population; 3 largest regions)
Chicago $31
Los Angeles $26

– Pacific Telephone advantaged
New York $17

– Sprint advantaged
• But most commentators believed Los Angeles’

demographics much better than Chicago’s or New York’s
 Los Angeles had highest value
(Ascending) Auction
• Pactel said beforehand:
“If somebody takes [the licence] away from us,
they’ll never make any money”

“If anyone takes business from us, we’ll make the price
so low that they’ll never make any money”
(Ascending) Auction
• Pactel said beforehand:
“If somebody takes [the licence] away from us,
they’ll never make any money”

“If anyone takes business from us, we’ll make the price
so low that they’ll never make any money”
• Pactel hired auction theorist to explain "winner’s curse”

to potential entrants
• cf. Dominant firm in “normal” market hiring industrial

economist to explain incumbent’s advantages to potential
entrants
Entry problems
• Often arise with common values
• But do not require common values
• e.g., if a “strong” bidder almost certainly has highest

value for a single unit and there are bidding costs,
then other bidders unwilling to compete especially
in an ascending auction
Entry problems
• With a single unit for sale, or if bidders are permitted

to win at most one each of a small number of units,
weaker bidders prefer first-price sealed-bid auctions to
ascending auctions.
• With many units, small bidders prefer uniform price or

ascending auctions (where they are protected by not
setting the price) to discriminatory auctions (where
bidding intelligently is disproportionately costly for them)
• These preferences are especially strong in common-

values contexts.
Case Study:
3G Mobile-Phone License Auctions
UK 3G Auction
• “... the Government’s objectives are to
(i) utilise the available UMTS spectrum with

optimum efficiency;
(ii) promote effective and sustainable competition

for the provision of UMTS services; and
(iii) subject to the above objectives, design an auction

which is best judged to realise the full economic value
to consumers, industry and the taxpayer of the spectrum.”
UK 3G auction (very simplified description)
• 5 mobile-phone licenses on sale
• Bidders each allowed to win at most 1 license
• Ran (simultaneous) ascending auction

– Prices rise on all 5 licences until only 5 bidders left
• 13 bidders entered
– 4 incumbent operators & 9 potential entrants
• Raised £22.5 billion = 2.5% of GNP

But Auctions can go wrong…
• Revenues from first five 3G Auctions

(€/capita)
UK 650
Netherlands 170
Germany 615
Italy 240
Austria 100
– But estimated Value 300 – 700 €/capita in every case

Netherlands 3G Auction
• Copied UK design
• Earned 26% of revenues it “should” have earned
• Netherlands Parliamentary Enquiry into how its

government “lost” €7 billion
– If UK government had done as badly, it would be £
16,598,695,384-61p poorer
Netherlands 3G Auction
• Before Netherlands auction I wrote:
“Netherlands should not copy UK design”
“auction design is not ‘one size fits all’’
and “the devil lies in the details”
What was the crucial detail?
• UK auctioned 5 licences and UK had 4 strong bidders

(incumbent operators)
– So at least one license had to go to an entrant
→ 9 new entrants bid, and pushed up the prices of all 5 licences
• Netherlands auctioned 5 licences but had 5 strong

bidders (incumbent operators)
– So why would any new entrant bother bidding ?
• In an ascending auction, their bids would always be

topped by an incumbent
Entry problems
• e.g., if a “strong” bidder almost certainly has highest

value for a single unit and there are bidding costs,
then other bidders unwilling to compete especially
in an ascending auction (for a single unit)
Entry problems
• e.g., if 5 “strong” bidders almost certainly have highest

values for a single unit each and there are bidding
costs, then other bidders unwilling to compete
especially in an ascending auction for 5 units
But Auctions can go wrong…
• Revenues from first five 3G Auctions

(€/capita)
UK 650
Netherlands 170
Germany 615
Italy 240
Austria 100
– But estimated Value 300 – 700 €/capita in every case

Austrian 3G Auction

rivals did ”
• Result: Bidding ended almost immediately

Are auctions a good idea?
Administrative Allocation
Auctions
(“Beauty Contests”)
• Efficient: winners are bidders • Often inefficient
with highest values
• Transparent • Hard to specify criteria
• Speedy • Time-consuming
• Fair • Outcome often contested
• Seller gets most of value • Seller gets little or nothing
(without deadweight losses) 3G beauty contests: €2bn
3G auctions: €100 billion (7 EU countries, pop. 130m)
(8 EU countries, pop. 250m)
• Even a bad auction is usually better than the alternative

Additional Reading on Auctions vs.
Administrative Allocation
• Editorials on Auctions vs Beauty Contests, Wall Street
Journal, 05-10-10, etc.
– Optional & Suggest reading either first page only, or pages 5-10
• Editorial on Effects of 3G Auctions on Telecoms

Industry, Financial Times, 26-11-02, etc.
– Definitely optional
• Both articles can be found under “Telecom Auctions” via

“Recent Press Articles” on my website:
http://www.paulklemperer.org
Why did the UK choose an Ascending Auction
(rather than a Sealed-bid Auction)?
Why did the UK choose an Ascending Auction
(rather than a Sealed-bid Auction)?
• Bidders allowed to win 1 license only, so no scope for

dividing spoils
 Collusion/demand reduction not a major worry
• 5 licenses and 4 strong (incumbent) bidders, so at least

one license must go to an entrant
 Entry not a major worry
• In UK case, ascending bidding easy and efficient, so

outcome likely to be more efficient than with sealed bids
Northern Rock Bank Run, Sept. 2007, &
(full) Global Financial Crisis, Sept. 2008
• Bank of England wanted to “sell” multiple “types” of loans

to commercial banks, building societies, etc.,
“Type” = quality of collateral used by borrower
Why not run a separate auction for each variety?
1. Auctioneer has to decide: how much to sell in each auction?
– e.g. Bank of England might prefer this if just two qualities:
Interest-rate
premium for
poor collateral
% poor collateral accepted
2. Bidders have to decide: which auction to enter?

– e.g. potential borrower (commercial bank) might prefer to say
“I would like to borrow £200m, and would pay up to 60 basis points to use
my poor collateral, or up to 50bp to use my good collateral (and at lower
interest rates I prefer to use my poor collateral if rate difference < 10bp)”
3. Market power: too little competition in each auction if

run separate auctions
Why not run Simultaneous Multiple Round
Auction?
• As pioneered by Paul Milgrom and Bob Wilson
• e.g., I ran SMRA for UK’s 3G mobile-phone licences (with Ken

Binmore)
– 2 large + 3 small licences; raised $34 billion = 2½% of GNP
• but SMRA works less well in other contexts

1. May take too long
2. May aid collusion &/or predation
3. Hard to allow the mix of varieties
sold to depend upon the bids
“An auction is just a market and standard
economics applies”
• Bidders for spectrum licences are oligopoly

⇒ Use game theory models/insights
• But many potential bidders for loans and easy entry

⇒ Use competitive models/insights
– Also, we would like to achieve competitive outcome
• Can focus on making bidding easy and efficient

Why not run a separate auction for each variety?
1. Auctioneer has to decide: how much to sell in each auction?
– e.g. Bank of England might prefer this if just two qualities:
Interest-rate
premium for
poor collateral


Product-Mix Auction
1. Auctioneer expresses preferences as a supply function
– e.g. Bank of England can “bid” this:
supply function
Interest-rate
premium for
poor collateral


Product-Mix Auction
supply function
Interest-rate
premium for
poor collateral
2. Bidders express preferences between goods

– e.g. potential borrower can bid “(60bp,50bp; £200m)” to mean

Product-Mix Auction
supply function
Interest-rate
premium for
poor collateral
2. Bidders express preferences between goods

– e.g. potential borrower can bid “(60bp,50bp; £200m)” to mean
3. Market power: low when all goods in same auction

“An auction is just a market and standard
economics applies”
• Bidders for spectrum licences are oligopoly

⇒ Use game theory models/insights
• But many potential bidders for loans and easy entry

⇒ Use competitive models/insights
– Also, we would like to achieve competitive outcome
• Can focus on making bidding easy and efficient

Product-Mix Auction
1. Each participant simultaneously states its preferences

– as a set of bids, or as a supply function
2. Implement competitive equilibrium allocation

consistent with stated preferences; basic version
of auction uses uniform prices
= (lowest) competitive equilibrium prices
Product-Mix Auction


⇒ Bidders bid (approx.) true values provided they are not

“too large”
Supply and Demand for, e.g., Wheat
Price
Supply
v
p
Demand
Quantity
• In an ordinary competitive market for a single variety

of good, a consumer who reveals she has value v
pays price p, like all other purchasers
Product-Mix Auction


⇒ Bidders bid (approx.) true values provided they are not

“too large”
⇒ Bidding is efficient, informative, and easy

(participants can express their preferences easily and accurately)
Bank of England’s Product-Mix Auction
(1st Version)
• Two types of goods auctioned by Bank of England
– (i) 3-month loans against “poor” collateral
e.g., mortgage-backed securities
– (ii) 3-month loans against “good”collateral
e.g., UK government bonds
• Bidders (commercial banks, etc.) can bid for one or both

goods, and/or express preferences between them
• Total allocation = £2.5 billion or £5 billion

• Bank of England expresses its preferences as supply

curves
• Bidders make simultaneous (sets of) bids
• Auction infers demand from bids (assuming “truth-telling”),

and determines competitive equilibrium allocations
• Winners pay (lowest) competitive equilibrium price

= lowest price they could have bid and still won
cf. “second-price” (or ascending) auction
Supply and Demand for, e.g., Wheat
Price
Supply
v
p
Demand
Quantity
• In an ordinary competitive market for a single variety

of good, a consumer who reveals she has value v
pays price p, like all other purchasers
• Bank of England expresses its preferences as supply

curves
• Bidders make simultaneous (sets of) bids
• Auction infers demand from bids (assuming “truth-telling”),

and determines competitive equilibrium allocations
• Winners pay (lowest) competitive equilibrium price

= lowest price they could have bid and still won
cf. “second-price” (or ascending) auction
• Example of all the bids of all the bidders
‒ numbers are £million bid for at that price vector
250
345
375
Price of B, PB
680
275 165
55
255
500 350
315
65
470
310 330 300 460

0 Price of A, PA
• All Pairs of Cut-off Rates that allocate £2,500m
250
345
375
Price of B, PB
680
275 165
55
255
500 350
315
65
470
310 330 300 460

0 Price of A, PA
• Can go from bids to a “relative demand curve”
‒ see film Auction Design in the Financial Crisis on
Price Difference, PA – PB
my website or see Klemperer 2008, 2018)
(Relative) Demand
0 Quantity allocated to A (as fraction of all goods allocated)

• and Bank of England (pre-)determines its
supply curve
Supply

“Screen Shot” from Bank of England’s
Product-Mix Auction (1st Version)
“Demand” Supply

“Demand” Supply
Theorem: Under some conditions (that roughly

applied in the Bank of England’s auction)
the Product Mix-Auction achieves an “efficient”
allocation. All bidders, and the auctioneer, get exactly
what they would have chosen at the final prices.

“Stressed
“Normal Demand”
Demand”
Supply
Greater proportion of
A (= “weak” collateral)
in stressed conditions
Product-Mix Auction
“A marvellous application of
theoretical economics to a
practical problem of vital importance"
Mervyn King, then-Governor
“A major step forward in practical

policies to support financial stability"
Paul Fisher, Executive Director
Extended to more varieties,

and to variable total supply
by Mark Carney, next-Governor
Another potential application of the
Product-Mix Auction
• IMF staff suggesting variant of Arctic PMA may help debt
restructuring
• Give creditors’ “budgets” proportional to their current claims;

they bid their budgets for different alternative claims:
– early vs. late repayment, repayment in $ vs. local currency,
fixed repayment vs. GDP linked, etc.
• And the Product-Mix Auction has many other applications

beyond finance …
Product-Mix Auction: More Details, and Software
• see film Art of Simple Multiproduct Auction Design
and free, open-source, Auction Software
on front page of website
Other New Developments:
Post-Brexit Agricultural Policy
• Auctions allow Simultaneous consideration of bids,

hence landscape-level optimisation
– Joint work with Elizabeth Baldwin, Alex Teytelboym
Conclusions
• Auction design is not ‘one size fits all’
– “De duivel zit in de details”
• Collusion and entry deterrence are key concerns

– Standard economics applies
• Auctions usually dominate administrative allocation
• Auction techniques can help analyse “ordinary markets”
• For more information see

“Auctions: Theory and Practice”
– Download free from www.paulklemperer.org
and “Product-Mix Auctions”
– Also on www.paulklemperer.org
Thank you for listening
I welcome feedback
about whether or not you enjoyed the lectures, and
about what you think I should have done differently
Please fill in my Questionnaire

MFE Macro Slides

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MFE Macro Slides

Uploaded by

Copyright:

Available Formats

MFE (Macro)Economics

Lecture 1: Economic Growth

Hilary Term 2023

Questions that Macroeconomics helps to answer

What determines economic growth in the long run?

In this course, you will

How to do well in the exam

Week 1 Economic Growth – TODAY!

Why there are such huge cross-country income di↵erences?

What are the reasons for growth miracles and disasters?

How did rich countries maintain such steady growth?

Small di↵erences in growth matter. How increase it?

Let Bob Lucas say it

“The consequences for human welfare

Robert E. Lucas Jr.

Kaldor’s facts about economic growth

In 1957 paper, Nicholas Kaldor summarized statistical properties of

1. Output per worker grows at a roughly constant rate

Solow model overview

Simple dynamic model of growth many of you already know

P.S. This lecturer takes Robert Solow seriously!

Inputs and outputs

At Lt – e↵ective labour, technology is labour-augmenting.

L̇t = nLt () Lt = L0 ent (3)

K̇t = It - Kt , 2 (0, 1) (5)

Saving rate and investment

Capital thus accumulates according to:

Dynamics of capital per unit of e↵ective labour

Take derivative of kt with respect to time:

K̇t L̇t Ȧt

k̇t = sf(kt ) - (n + g + )kt (7)

sf(kt ) is the actual (gross) investment per unit of e↵ective labour.

Regardless of the starting point, k converges to steady state level k⇤

sf(k⇤ ) - (n + g + )k⇤ = 0. (8)

Output, consumption, and investment per unit of e↵ective labour also

Steady state graphically

Balanced growth path (BGP)

How does the economy behave in the steady sate (k = k⇤ )?

Change in the saving rate s changes the steady state

The economy converges to the new balanced growth path

ln(Y/L) increases in the short run, then

The e↵ect of an increase in s: in words

How big is the level e↵ect quantitatively?

Suppose prod. func. is Cobb-Douglas: Y = K↵ (AL)1-↵ , or y = k↵ .

If markets are competitive and there’re no externalities, ↵ corresponds

E↵ect of saving rate on consumption

It is consumption that matters for people’s welfare, not output!

The Golden Rule

Golden-rule level of k⇤ is one that maximises consumption in the

Other modern exogenous growth models

How is the saving rate determined in the economy?

Similarly to the Solow model, both above model economies converge to a

What have we achieved so far?

An elegant growth model:

But g, the growth rate of A is exogenous here – hence the name of

Endogenous growth: introduction

Endogenous growth models seek to explain growth as the result of

We will begin with a very simple extension of Solow model by adding

Simple model: production of goods and knowledge

Case 1: Fishing out ideas (✓ < 1)

✓ < 1: decreasing return to knowledge in generation of new ideas.

Cases 2 and 3: Standing on the shoulders of giants (✓ > 1)

✓ = 1 and n > 0: ever increasing growth.

✓ = 1 and n = 0: simple fully endogenous linear growth, or “AK”