Econ 2010d: Heterogeneity in Macroeconomics

Lecture 1: Unemployment Facts

Adrien Bilal
Spring 2024
Harvard University
Introduction
Welcome to Econ 2010d, part 1!

• Adrien Bilal
I Littauer 212, [email protected]
I Lectures: MW, 1:30pm-2:45pm, Littauer M-15
I My part: Jan 22-Feb 29
I Office hours: Tuesday afternoon, 4pm-5pm, email before to set appointment

• Amazing Teaching Fellow: Pierfrancesco Mei

I [email protected]
I Sections: 9am Sever Hall 303
I Office Hours: TBD

1 / 27
We What (Are Going To) Do
and Why
What Do We Do in Macro?
• We study a lot of models
I Last 15 years also bring in a lot of data

• Why? Typical macro questions: very hard to measure all pieces of it

I Example: Effect of employment protection programs on employment?
I Can measure partial equilibrium effect using quasi-experiment
F E.g. a discontinuity design based on firm size
F Then also need to worry about SUTVA
I Missing intercept
I How to go from PE to GE? Use a model!
F GE can be drastically different from PE (e.g. China import competition shock)
F If had credible variation at the country-level would not need the model
F But hard to come by...

• A model is a way to fill in missing pieces in an empirically plausible way

I Structured “best guess”
I Also useful to analyze mechanisms, organize thoughts, other counterfactuals...
I Most useful when can be estimated or at least calibrated
2 / 27
Gearing up for Research

• Doing research is different

• Two essential ingredients

1. An interesting research question
2. Familiarity with models and data
F So that can juggle with them fast & efficiently when writing papers

• For 1, we will cover important branches of macro

• For 2, no better way than to practice, practice and practice—and practice!

I So we will practice, practice and practice—and practice!

3 / 27
Organization & Requirements
• Slides have main material
I Slides online ahead of class
I Lecture notes have additional details of derivations
I Will update them periodically!
I Syllabus has further readings, overall plan for class

• Requirements
I 5 problem sets (12% of grade each)
F Posted on Thursday, due following Friday 9am
F Can work in groups ≤ 3 (encouraged!) but individual writeup
F Will have empirical, theory and quantitative/coding components
F Submit on time to avoid a grade tax if late (see syllabus)
F Use of previous solutions keys is not allowed
I Final exam (40% of grade)
F 3h not in class, week of March 4-8
F Covering my part

• Please ask questions!

4 / 27
Overview

• In Marty’s part learned about

1. Efficient starting point of modern macro
F The Real Business Cycle model: no role for policy
2. A first deviation: Price stickiness
F The New Keynesian model ⇒ Monetary policy

• In this part will cover two more core frictions in macro

1. Unemployment and labor market frictions
F The DMP model ⇒ Labor market policy
2. Borrowing frictions
F The Incomplete Market model ⇒ Fiscal policy
I Both frictions related to heterogeneity in the labor market/income

• Xavier’s part will expand on financial frictions & other topics

5 / 27
Recap

• So far studied a frictionless labor market in RBC/NK

I Conditional on observables everyone earns the same

• Clearly a simplification
I Observationally identical workers can be employed and earn a wage...
I ...Or be unemployed and receive unemployment insurance

• In 1st half studying income heterogeneity

I What happens when workers transition in and out of unemployment
I Will abstract away from consumption insurance
I DMP model

• In 2nd half study interaction b/w consumption & income heterogeneity

I Incomplete market model

6 / 27
Unemployment in the Data
Today: Unemployment Facts
• Unemployment is one of the main motivation behind business cycle analysis
I But no unemployment in RBC model! Full employment + hours fluctuate
• Keynes’ General Theory (1936) largely shaped by Great Depression
I Unemployment rate as high as 25%
• 3 main reasons why care a priori
I Individual: lower emotional well-being (Krueger Mueller 2012)
I Individual: lower income and consumption (Ganong Noel 2019)
I Aggregate: under-utilization of resources
• Natural questions
1. Is unemployment inefficient?
2. If so, what policies should we implement?
I Need a theory that clarifies causes & consequences
• But before theorizing, first need to
1. Define unemployment
2. Measure unemployment
I Today’s lecture
7 / 27
 Total US population
330 million

Non-institutional civilian population < 16 y/o,

260 million army, prisons

Civilian labor force Not in labor force Jobless but

not looking
160 million 100 million for work

Jobless and actively

Employed Unemployed looking for work
150 million 10 million Unemployment rate
Unemployed
= Employed+Unemployed

8 / 27
Out of Labor Force vs. Unemployed

• Definitions from the Bureau of Labor Statistics

• Unemployed: currently jobless but looked for work in past 4 weeks

I Includes temporary layoffs

• Not in labor force: jobless but not looking in past 4 weeks

I Retirees
I Students
I At home taking care of family members
I Discouraged workers: available for a job but not looking

• Researchers sometimes use own variation of def. depending on data

9 / 27
Long-Run Flows Across Labor Market States

Current Population Survey

Not in LF
1996-2014
100m
Monthly averages

4.3m 2.3m

4m 2.2m

2.1m
Employed Unemployed
150m 10m
2.2m

10 / 27
Time Series of US Participation Rate
Labor Force Participation Rate
Labor Force Participation Rate - Women
Labor Force Participation Rate - Men
90

80

70
Percent

60

50

40

30
1950 1960 1970 1980 1990 2000 2010 2020

Shaded areas indicate U.S. recessions. Source: U.S. Bureau of Labor Statistics fred.stlouisfed.org

• Decline in male LFP: aging and longer education

• Rise in female LFP: social norms, better education, rise of clerical jobs

11 / 27
Time Series of US Participation Rate: ages 25-54
Labor Force Participation Rate - 25-54 Yrs.
Activity Rate: Aged 25-54: Females for the United States
Activity Rate: Aged 25-54: Males for the United States
100

90

80
Percent

70

60

50

40
1950 1960 1970 1980 1990 2000 2010 2020

Sources: BLS; OECD fred.stlouisfed.org

12 / 27
Time Series of US Unemployment Rate
Unemployment Rate

15.0

12.5

10.0
Percent

7.5

5.0

2.5

0.0
1950 1960 1970 1980 1990 2000 2010 2020

Shaded areas indicate U.S. recessions. Source: U.S. Bureau of Labor Statistics fred.stlouisfed.org

13 / 27
Time Series of US Unemployment Rate: By Sex
Unemployment Rate
Unemployment Rate - Women
Unemployment Rate - Men
17.5

15.0

12.5

10.0
Percent

7.5

5.0

2.5

0.0
1950 1960 1970 1980 1990 2000 2010 2020

Shaded areas indicate U.S. recessions. Source: U.S. Bureau of Labor Statistics fred.stlouisfed.org

14 / 27
Stock-Flow Model
Stock-Flow Accounting Model
• Unemployment represents a stock of workers
I Balance between inflows and outflows
I Useful to break down role of inflows vs. outflows
F For modeling choices & mechanisms
F Modeling choices ultimately determine policy implications

• To get started abstract from individuals not in labor force

I Normalize unemployment + employment to 1

• Basic stock-flow model

u − u = st (1 − ut ) − ft ut
| t+1{z }t | {z } |{z}
Change in U-rate Inflow Outflow
sep. into U hires from U

I ut : unemployment rate in month t

F Typically ∼ 5% in data
I st : separation rate into unemployment in period t
F Typically ∼ 1% monthly in data
I ft : job finding rate from unemployment in period t
F Typically ∼ 20% monthly in data
15 / 27
Steady-State and Transitions

• In steady-state,

s SS
u SS =
s SS + f SS
• Out of steady-state, no such simple formula

• But if transitions towards steady-state are “fast enough”, using

st
ut ≈ ≡ ût
st + ft
will deliver a good approximation.
I Unemployment close to “as-if” steady-state with contemporaneous flows
I Can use this approximate formula to unpack role of inflows vs. outflows

16 / 27
Transitions Are Fast in Stock-Flow Model

.12
.1
unemployment rate (p.p.)
.08
.06
.04
.02
0

1980m1 1990m1 2000m1 2010m1 2020m1

time

Data Fitted
17 / 27
Transitions Are Fast in Stock-Flow Model

.12
.1
unemployment rate (p.p.)
.08
.06
.04
.02
0

1980m1 1990m1 2000m1 2010m1 2020m1

time

Data Fitted & mean-adjusted

17 / 27
Inflows vs. Outflows
Inflows vs. Outflows in Stock-Flow Model

• Two related ways to evaluate role of separations st vs. job-finding ft

1. Hold either one fixed at steady-state and plot implied unemployment rate

st s SS st
and vs. ≡ ût
st + f SS s SS + ft st + ft

2. Exact variance decomposition:

I Use
ût st
=
1 − ût ft
I To obtain
     
ût ût ût
Var log = Cov log , log st + Cov log , − log ft
1 − ût 1 − ût 1 − ût

18 / 27
Eyeball Test of Inflows vs. Outflows

.12
.1
unemployment rate (p.p.)
.08
.06
.04
.02
0

1980m1 1990m1 2000m1 2010m1 2020m1

time

Data
Only job finding varying (time-aggregated)
Only job separation varying (time-aggregated)
19 / 27
Visualizing the Variance Decomposition

ut ut
(a) − log ft vs. log 1−ut (b) log st vs. log 1−ut
1

1
.75

.75
.5

.5
- log UE (de-meaned)

log EU (de-meaned)
.25

.25
0

0
-.25

-.25
-.5

-.5
-.75

-.75
-1

-1
-1 -.75 -.5 -.25 0 .25 .5 .75 1 -1 -.75 -.5 -.25 0 .25 .5 .75 1

log u/(1-u) (de-meaned) log u/(1-u) (de-meaned)

Data Data
45 degree line 45 degree line
- log UE = 0.59 * log u/(1-u) log EU = 0.41 * log u/(1-u)

• Regression coefficient = variance share

20 / 27
Variance Decomposition

• Exact variance decomposition yields variance shares

I Job finding rate: 59%
I Job separation rate: 41%

• Using different data/methodology, Shimer (2012) finds

I Job finding rate: 90%
I Job separation rate: 10%

• Ongoing debate what the exact split is

• Bottomline: job finding rate is largest contributor

21 / 27
Taking Stock

• St.dev. of unemployment rate ≈ 1.5 p.p. over the business cycle

• Job finding rate accounts for ≈ 60 − 90% of that variation

• So far: relationship between unemployment, job finding & separation rates

I Mechanical to some extent
I What determines the job finding rate?

• Two main actors in labor market

I Unemployed workers search for jobs (by definition!)
I Firms advertise “vacancies”: open positions

• Job finding rate determined by meetings b/w unemployed workers & firms

• Measure vacancies with Job Openings and Labor Turnover Survey

22 / 27
 Vacancies
and the
Beveridge curve
The Beveridge Curve: Definition

• Beveridge curve is the empirical relationship between

I Vacancies
I Unemployment
over the business cycle

• Summarize time-series with cross-sectional graph (log ut , log vt )

23 / 27
The Beveridge Curve: Data

9
8.5
log vacancies
8
7.5

1 1.5 2 2.5 3

log unemployment rate

2001Q1-2008Q4
24 / 27
The Beveridge Curve: Data

9
8.5
log vacancies
8
7.5

1 1.5 2 2.5 3

log unemployment rate

2001Q1-2008Q4 2009Q1-2020Q1
24 / 27
The Beveridge Curve: Data

9
8.5
log vacancies
8
7.5

1 1.5 2 2.5 3

log unemployment rate

2001Q1-2008Q4 2009Q1-2020Q1 2020Q2-2020Q4
24 / 27
The Beveridge Curve: Data

.75
.5
log vacancies (de-meaned)
.25
0
-.25
-.5
-.75

-.75 -.5 -.25 0 .25 .5 .75

log unemployment rate (de-meaned)

2001Q1-2008Q4 2009Q1-2020Q1 log v = - 0.96 * log u

24 / 27
Vacancies and the Job Finding Rate

• Have just seen that log vt ≈ cste − log ut over the cycle (Beveridge curve)
I In booms, vacancy creation increases
I In recessions, vacancy creation decreases

• Now relate vacancy creation to the job finding rate

• Natural starting point: new meetings depend on ratio between

I Number of vacancies
I Number of unemployed

• So investigate relationship between log ft and log vt

ut

25 / 27
Vacancies and the Job Finding Rate

1
.75
.5
log UE (de-meaned)
.25
0
-.25
-.5
-.75
-1

-1 -.75 -.5 -.25 0 .25 .5 .75 1

log v/u (de-meaned)

2001Q1-2008Q4 2009Q1-2020Q1
log UE = 0.38 * log v/u
26 / 27
Taking Stock

• St.dev. of unemployment rate ≈ 1.5 p.p. over the cycle

• Job finding rate accounts for ≈ 60 − 90% of that variation

• Vacancy creation strongly procyclical (Beveridge curve, elasticity - 1)

• Job finding rate positively related to V/U ratio (elasticity 0.4)

I Next time: basic DMP model that speaks to those facts

27 / 27
Bonus slides
Time series of US unemployment rate: by race
Unemployment Rate
Unemployment Rate - Black or African American
Unemployment Rate - Hispanic or Latino
Unemployment Rate - Asian
22.5

20.0

17.5

15.0

12.5
Percent

10.0

7.5

5.0

2.5

0.0
1950 1960 1970 1980 1990 2000 2010 2020

Shaded areas indicate U.S. recessions. Source: U.S. Bureau of Labor Statistics fred.stlouisfed.org

28 / 27
FAQ

• Is there a minimum amount of hours that an individual has to work to be

counted as employed?
I BLS: any amount of time worked is enough to qualify as employed.

• What is the proportion of retirees/other groups in individuals out of the LF?

I Use BLS
I 1/2 are retirees
I Students, disabled, and home responsibilities make up 1/6 each

• How much does aging contribute to male’s LF participation rate decline?

I Use FRED
I Male part. rate of aged 25-54 ↓ 9pp b/w 1960 and 2020 (97 to 89pp)
I Male part. rate for all ↓ 14pp b/w 1960 and 2020 (83pp to 69pp)
I Interpreting double diff. as aging: 5pp decrease in LFP (out of 14pp)

29 / 27
Bonus: Unemployment in
RBC
Hours in RBC model

• RBC provides a useful first pass

• RBC model has a representative agent choosing consumption and hours:

"∞ #
X n o
t
max E0 β u(Ct ) − v (Nt ) s.t. Ct + Kt+1 = Rt Kt + wt Nt
Ct ,Nt
t=0

• Simplest interpretation
I Everyone employed and identical
I Households only adjust hours

• This interpretation at odds with data (more on this later)

• But can interpret RBC model differently

30 / 27
Indivisible labor and employment lotteries in RBC (1/3)
• Rogerson (1985)

• Now posit continuum of households i ∈ [0, 1]

• Labor fully indivisible nt ∈ {0, 1} −→ only v̄ ≡ v (1) − v (0) matters

• Arrow-Debreu markets where households trade their employment risk

I States of the world within period: vectors s ∈ {0, 1}I of employment statuses
I p(s): price consumption claim when i’s employment status is si ∈ {0, 1}
I Security is traded by all households

• Goal: show that household’s within-period problem boils down to

max πu(ce ) + (1 − π)u(cu ) − π v̄ s.t. πce + (1 − π)cu = πw + T

π,ce ,cu

I Households choose probability of work π

I Subject to a budget constraint with full insurance
I T = RK − K 0 is capital income
31 / 27
Indivisible labor and employment lotteries in RBC (2/3)
• Introduce AD formalism to derive result (just on this slide!)

• Within-period problem of household i and drop time subscripts

I Choose probability of work, consumption in each state
X n o
max π(s−i ) πi u(ci (1, s−i )) + (1 − πi )u(ci (0, s−i )) − πi v̄
πi ,ci
s−i ∈{0,1}I −1
X X

s.t. p(s)ci (s) ≤ p(s) 1{si =1} w + T , π(si = 1) = πi
s∈{0,1}I s∈{0,1}I

• FOC when i contemplates trading AD asset for j’s employment status:

p(1, s−j ) u 0 (ci (1, s−j )) πj πj

= 0 =
p(0, s−j ) u (ci (0, s−j )) 1 − πj 1 − πj

I In equilibrium, ci independent from sj because every agent is infinitesimal

• Substitute into AD problem to recover simpler formulation of previous slide

32 / 27
Indivisible labor and employment lotteries in RBC (3/3)

• We have reduced the problem to

max πu(ce ) + (1 − π)u(cu ) − π v̄ s.t. πce + (1 − π)cu = πw + T

π,ce ,cu

• From Lagrangian’s FOCs, ce = cu and so

max u(c) − π v̄ s.t. c = πw + T

π,c

• Obtained a representative agent with linear disutility of work

I Generates a very large (infinite!) aggregate labor supply elasticity
I Despite zero intensive-margin individual labor supply elasticity
I Good news for the RBC model

33 / 27
Empirical challenges

• Two challenges for employment lotteries & complete markets

I We do not observe those contracts in practice
I We do not observe risk-sharing either: consumption drops upon unemployment

• Two challenges for a frictionless labor market

I Co-existence of unemployed workers and vacancies
I Dispersion in wages

34 / 27