Transcript Barcelona Summer School Lecture 2015

Macroeconomic Experiments
Charles Noussair
Tilburg University
Barcelona, June 14, 2015
Outline of this lecture
• This talk is divided into three segments:
– (1) Institutions and Growth
– (2) Asset Market Bubbles
– (3) DSGE Economies
• I will describe three lines of research. The idea
is that they may stimulate research ideas on
your part.
The experimental approach to studying models
Economic Model
Structure of Economy
(objectives, constraints, rules)
Behavior of Agents
Outcomes
Theory
Experiment
A
Assume
structure
Asssumption on
Rationality,
Expectations
Equilibria,
simulated
Outcomes
Compare
Create
structure
Some
features
may not be
feasible,
some may
not be
sensible
?
Data
The experimenter specifies the structure of the economy, and observes behavior and
outcomes.
Main source
Theoretical models specify structure and behavior, and study outcomes
of
hypotheses
Piece “A” is sometimes easy (desirable), and sometimes not
Part I: Growth and Institutions
•
•
THE STRUCTURE OF THE MODEL
A representative consumer in the economy has a lifetime utility given by:

 (1   )
t 0
•
t
U (C t )
ρ is the discount rate, Ct is the quantity of consumption at time t, and U(Ct) is the
utility of consumption. The economy faces the resource constraint:
Ct  Kt 1  A  F ( Kt )  (1   ) K t
•
δ is the depreciation rate, Kt is the economy’s aggregate capital stock at the
beginning of period t, and A is an efficiency parameter on the production
technology. The production function is concave.
•
•
•
THE ASSUMPTION ON BEHAVIOR: The agent maximizes lifetime utility
•
The optimal steady state given by the solution to: C* = F(K*) – δK* and K* = ρ + δ
OUTCOME OF MODEL: the principal result of the model is that Ct and Kt converge
asymptotically to optimal steady state levels.
If individuals are given incentives to solve the dynamic optimization
problem on their own, it is very difficult.
Social Planners
Figure 6: Time Series of Consumption: Social Planners High Endowment
25
Consumption
20
A1
A2
C1
C2
E1
E2
C*
15
10
5
0
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34
Time
Suppose a team of five people is
making the decision instead. They do
better but still have a lot of trouble
Implement model as a Decentralized Economy.
• There are five agents in the economy
• The economy’s production capability and utility function is divided up
among the five agents.
• Agents are made asymmetric.
• A market is available to exchange capital (using double auction rules,
because a competitive model is being tested).
• There is money, an experimental currency, in the economy, which agents
use for purchases and sales of capital.
Timing within a period t
• At the beginning of period t, production occurs mapping kt
into output (ct + kt+1)
• A double auction market for output is open for two minutes in
which they can exchange output.
• Agents have one minute to allocate any portion of their
output to consumption ct
• At the beginning of period t+1, production occurs mapping
kt+1 into output (ct+1 + kt+2).
Timing within a period
Timing of sessions (ending a session)
• A horizon refers to the entire life of an economy.
• Implementation of infinite horizon with discounting: In each period,
there was a /(1+ ) probability that the horizon would end.
• If a horizon ended with more than one hour to go in the
experimental session, a new horizon was started.
• If a horizon still had not ended at the scheduled end of the session,
the horizon would be continued on another evening.
• Subjects would have the option of continuing in their roles in the
continued session.
• If they chose not to continue, a substitute would be recruited to
take her place. The original subject would also receive the money
earned by the substitute.
Results: Consumption patterns in the
decentralized economy
C*=12
18
K*=10
16
Ko =20
14
Consumption
12
C2
E2
F3
G4
C*
10
8
6
4
2
0
1
2
3
4
5
6
7
8
9
10
11
Time
12
13
14
15
16
17
18
19
20
An environment with multiple
equilibria
• Now suppose that there exist two stable equilibria,
which are Pareto-ranked so that the inferior
equilibrium represents a poverty trap.
• The value of the productivity parameter A depends
on the economy’s capital stock. There exists a
threshold level of capital stock, above which A has a
higher value.
 A,
A
 A,
ˆ
if K  K
ˆ
if K  K
Production function includes threshold
externality
Aggregate Production Function
220
200
180
O utput
160
140
120
100
80
60
40
20
0
0
25
50
75
U nits of Input
100
125
150
Theoretical Predictions for data in the
following slides
•
•
•
•
There is an optimal steady state in which (C*, K*) = (70,45)
From any initial level of capital stock, optimal decisions (of a benelovent social
planner) at each point in time imply monotonic convergence to (C*, K*).
However, if the economy is decentralized, there are two stationary rational
expectations competitive equilibria at (CH, KH, pH) = (70,45,118) and (CL, KL, pL)
= (16,9,334)
RESULT: The decentralized economy converges to the poverty trap.
Results: Observed and Equilibrium Aggregate Consumption (Five Sessions)
C* optimal = 70, C* poverty trap = 16
Emory B1
60
60
40
40
20
20
0
0
Dat a
Horizontal axes: time
Time
Time
C* opt imal
Data
C* inferior
Emory B3
Consumption
60
40
20
0
C* optimal
C* inferior
Breaks in series: New
horizon beginning
Caltech B1
80
80
Consumption
Vertical axes:
aggregate
consumption
Consumption
Consumption
Emory B2
80
80
60
40
20
0
Time
Dat a
Time
C* opt imal
Dat a
C* inferior
C* opt imal
C* inferior
Results:
Caltech B2
Consumption
80
No economy
surpasses the
capital stock
threshold.
60
40
20
0
Time
Dat a
C* opt imal
C* inferior
§= Each data point represents a period in a horizon. Horizons are separated by spaces
Convergence to
near poverty trap is
typical outcome
The Communication treatment
– Identical to the baseline treatment, except that
before the market opened, subjects were allowed
to communicate with each other.
– Each agent’s screen displayed a chat-room, which
they could use to send and receive messages in
real time.
– Communication was unrestricted and all agents
could observe all messages.
Observed and Equilibrium Aggregate Consumption, Communication
Treatment; C* optimal = 70, C* inferior = 16
Emory C1
80
Consumption
Vertical axes:
aggregate
consumption
60
40
20
However, which equilibrium it
converges to varies between
sessions.
0
Horizontal axes: time
Time
Dat a
Consumption
C* opt imal
C* inferior
Emory C2
80
60
40
Example of how institutional
structure affects mean and
variance of income.
20
0
Time
Dat a
C* opt imal
C* inferior
Caltech C1
Consumption
Consumption
Emory C3
80
60
40
20
80
60
40
20
0
0
Time
Dat a
C* o p t im al
Time
C* in ferio r
Dat a
Caltech C2
Consumption
60
40
20
C* o p t im al
C* in ferio r
Caltec h C3
80
80
Consumption
Breaks in series: New
horizon beginning
Results: Individual sessions
converge to near one of the
equilibria.
60
40
20
0
0
Time
Dat a
C* o p t im al
Time
C* in ferio r
Dat a
C* o p t im al
C* in ferio r
The Voting treatment
–
–
–
–
–
Identical to the baseline treatment except that consumption and
investment decisions were determined in the following manner:
Two agents were randomly chosen in each period to make proposals
on how much each agent in the economy should consume.
Before submitting proposals, proposers received information
indicating the current stock of capital held by each agent.
Proposals were followed by majority voting. All agents were
required to vote in favor of exactly one of the two proposals.
The proposal that gained at least 3 (of the 5 total) votes became
binding. Each agent consumed the quantity of output specified
under the winning proposal, and began next period with the
amount of capital allotted to her under the winning proposal.
Observed and Equilibrium Aggregate Consumption, Voting Treatment
C* optimal = 70, C* inferior = 16
Results:
Emory V1
-In most sessions,
economy escapes
poverty trap
40
20
0
Time
Dat a
C* o p t imal
- High variance from one
period to the next within
sessions.
C* inferior
Emory V2
Horizontal axis: time
Consumption
80
60
40
- Convergence toward
equilibrium typically
does not occur
20
0
Time
Dat a
C* o p t imal
C* inferior
Emory V3
Caltech V1
Consumption
80
80
60
40
20
60
40
20
0
0
Time
Time
Dat a
C* o p t imal
C* in ferio r
Caltech V2
80
Consumption
Breaks in series: New
horizon beginning
60
Consumption
Vertical axis:
aggregate
consumption
Consumption
80
60
40
20
0
Time
Dat a
C* o p t im al
C* in ferio r
Dat a
C* o p t imal
C* inferior
The hybrid treatment: Both communication and
voting are present
Timing in the hybrid treatment
Observed and Equilibrium Aggregate Consumption, Hybrid
Treatment; C* optimal = 70, C* inferior = 16
Hybrid (Emory se ssion 1)
Data
C* optimal
90
90
80
Aggregate consumption
Aggregate consumpti on
80
70
60
50
40
30
20
70
60
50
Horizontal axis: time
40
30
20
Breaks in series: New
horizon beginning
10
10
0
0
Time
Time
Hybrid (Emory session 3)
H ybrid (C alte ch session 1)
D ata
90
C* op tim al
C* infe rio r
90
Ag greg ate Con su mpt ion
80
Aggregate consumption
Vertical axis:
aggregate
consumption
Hybrid (Emory session 2)
C* inferior
70
60
50
40
30
20
10
80
70
60
50
40
30
20
10
0
0
Time
Time
H ybrid (Calte ch se ssion 2)
90
Ag g reg at e Co ns ump tio n
80
70
60
50
40
30
20
10
0
Time
Result; The addition of voting
and communication allows the
economy to escape poverty
trap in all sessions.
Results
• Baseline: The economies of the baseline treatment converge
to near the poverty trap. Does not escape poverty trap in any
session.
• Communciation: The economies of the communication
treatment converges to close to one of the stationary
equilibria. However, the one it converges toward varies
between sessions. Probability of avoiding the poverty trap
greater than under baseline.
• Voting: The voting treatment exhibits variable behavior from
one period to the next. Probability of avoiding the poverty
trap greater than under baseline.
• Hybrid: Also shows variable behavior from one period to the
next. Escapes the poverty trap in all sessions.
Part 2: Bubbles and Crashes
• I will discuss work on experimental markets
for long-lived assets.
• In such markets, the bubble and crash pattern
in pervasive.
• I will first describe the effect of different
market institutions and parameters on bubble
formation
• Then I will concentrate on differences
between sessions but within treatments.
The type of asset considered: the type first studied by (Smith,
Suchanek Williams, 1988)
•
The asset has a life of 15 periods. At the end of each period, the
asset pays a dividend which is equally likely to be 0, 8, 28 and 60
cents, determined independently for each draw.
•
After the last dividend is paid, the asset has no value.
•
Assuming traders are risk neutral, the fundamental value of this
asset can be calculated at any point in time. It is equal to the
expected total future flow of dividends.
•
Since the expected dividend in any dividend draw is equal to 24,
the fundamental value is equal to the number of dividend draws
remaining times 24.
•
Therefore, the fundamental value is 360 at the beginning of the
life of the asset, and declines by 24 cents every period.
The fundamental value of this asset over
time
What happens in such a market?
b) Benchmark
Rather than tracking the fundamental value, a bubble and crash pattern is
typically observed.
Effect of Increasing Cash: Raising
prices
Effect of allowing short selling:
lowering prices
Effect of adding a futures market,
improving price discovery
The effect of experience: pricing closer
to fundamentals
The role of the fundamental value
time path: Increasing FV trajectory
Decreasing fundamental value
trajectory
Decreasing
treatment
tracks
fundamentals
more closely
than
increasing
Modeling trader types
• Can we explain what is observed in the data with a model in
which there are multiple trader types?
• Consider the following model (based on DeLong, Shleifer,
Summers, and Waldmann ,1990)
• Assume three types of trader: (1) Fundamental value traders,
(2) Rational Speculators, and (3) Momentum traders.
– Fundamental value traders purchase if prices are below fundamentals
and sell if they are above.
– Rational speculators anticipate future price movements. They buy if
the price is going to increase in the next period, and sell if it is going to
decrease.
– Momentum traders follow the current trend. They buy if the price has
been going up and sell if it has been going down.
Demand functions of the three types
• The three types have net demand functions of the
following form:
– Fundamental value trader: -α(pt – ft), where pt is the
price in period t, and ft is the fundamental value in
period t.
– Rational speculator: γ(E(pt+1) - pt),
– Momentum trader: -δ + β(pt-1 – pt-2)
• Six parameters: α, β, γ, δ, and the proportion of traders
that is of each type.
• This structure generates bubbles and crashes, is
consistent with treatment differences, and the types
correlate with other behaviors.
• Now I will explore the relationship between trader
characteristics and market behavior.
Loss aversion measurement
Cognitive reflection test (Frederick, 2005)
•
This consists of three questions, in which the first answer that comes to mind is incorrect,
but the correct answer is simple after some reflection.
•
Example: If it takes 5 people 5 minutes to make 5 units, how much time does it take 100
people to make 100 units?
– First answer that comes to mind is 100 minutes
– With reflection realize the answer is 5 minutes
•
•
Other two questions:
A bat and a ball cost 1.10 Euro in total. The bat costs 1 Euro more than the ball. How much
does the ball cost?
In a lake, there is a patch of lily pads. Every day, the patch doubles in size. If it takes 48
days for the patch to cover the entire lake, how long would it take for the patch to cover
half of the lake?
•
Risk aversion measurement (Holt and
Laury, 2002)
Correlation between price level and average risk
aversion in the market: Bullmarket treatment
Correlation between number of trades in the market and
average loss aversion: Bearmarket treatment
Average CRT score and departure
from fundamental value
Final Individual Holdings and Risk
Aversion Level
Number of trades at individual level
and loss aversion
Individual CRT score and earnings
Determinants of bubble magnitude: measured
with average dispersion (AD) and average bias (AB)
Correlation between trader
characteristics and strategies
Emotional correlates of asset price
movements
• Positive emotions have typically been
associated with higher prices
– Irrational exuberance (Greenspan, 1996)
– Speculative euphoria (Galbraith, 1984)
• Empirical support
– Weather affects returns (Hirshleifer and Shumway, 2003; Kamstra et al, 2003)
– Outcomes of sporting events also affect prices (Edmans et al., 2007)
– Twitter mood (Bollen et al., 2010) and anxiety level in blog postings (Gilbert
and Karahalios, 2009) predict stock price movements.
• Fear has been associated with low prices and
volatility
Emotions and markets in the
laboratory
• We consider whether these and other connections between emotions and
market behavior appear in the laboratory
• We construct an experimental asset market with a monotonically
decreasing fundamental value.
• All traders are videotaped throughout the session
• The videotapes are analysed with Noldus Facereader at a later date.
• The Facereader reads a facial expression and classifies it along seven
dimensions.
– These correspond to the basic universal emotions identified by Ekman
(1978, 1999)
Facereading software
• We use facereading software to track traders’ emotional state while
the market is operating.
• These are
– Happiness
– Anger
– Sadness
– Disgust
– Fear
– Surprise
– Neutrality
• Facereader also reports the overall valence of emotion.
• Unlike other neuroeconomic methodologies, Facereader translates
physiological responses directly into of psychological models
“Smiling” for a picture
Absence of negative emotions
…not so happy
Neutrality, anger and disgust
Neutrality with some anger
Very happy
Market Prices
Initial valence and average prices
We hypothesize that:
(1) More
positive
valence is
associated
with higher
prices.
 = .708,
p < .01
Valence is typically negative: Experiments are not fun for subjects
Initial fear and average prices
We hypothesize that:
(2) Fear is
associated with
lower prices
= -.549
p < .06
People who are more neutral during crash period earn more
money
Crash period: The period with the largest price decrease
Hypothesis 4:
Neutrality during
market turbulence is
associated with
greater earnings
Correlations Between Loss Aversion
Measure and Emotions
Loss
aversion
Fear
Valence
Happiness
Anger
Surprise
Disgust
Sadness
Neutral
0.3427***
-0.3012**
-0.0459
-0.0680
-0.0851
0.2098
0.1096
-0.1989
(0.025)
(0.759)
(0.649)
(0.569)
(0.157)
(0.463)
(0.180)
(0.018)
Correlations between initial emotional
state and likelihood of being a FV trader
Initial Emotional State
Fundamental Value Trader
Neutral
.202**
Happy
.012
Sad
-.050
Angry
Surprised
-.195**
-.091
Scared
-.207**
Disgusted
-.182*
Valence
.097
More positive valence predicts more purchases
Buy t
Buy t
Sell t
Sell t
Model 1
Model 2
Model 3
Model 4
valence t-1
.237*
.238*
fear t-1
2.151
2.690
money t-1
7.29e-06
4.95e-06
money t-1
5.79e-06
6.25e-06
units t-1
-.021**
-.015
units t-1
.050***
.046***
P level t-1
-.00007
-.00008
P level t-1
-.00012
-.00012
Buy t-1
.355***
Sell t-1
.362***
Prob>chi2
=0.000
Prob>chi2
=0.0586
Prob>chi2 =0.000
Prob> chi2
=0.0000
9770 obs
9770 obs
9971 obs
9971 obs
49 groups
49 groups
50 groups
50 groups
Momentum traders buy more when they are in a
more positive emotional state
Buy t
Fundamental Value
Trader
Momentum Trader
Rational Speculator
Trader
Valence t-1
.280
.521**
-.025
Money t-1
.00003
-.00004
-.00007**
Units t-1
.015
-.119***
-.068***
Price level t-1
-.0004*
.0007***
3.15e-06
Buy t-1
.435***
.141*
.370***
Obs: 3762
Obs: 3396
Obs: 2612
Groups: 19
Groups: 17
Groups: 13
Prob>F =.0000
Prob>F =.0000
Prob>F =.0000
More fear correlates with more sales
Buy t
Buy t
Sell t
Sell t
Model 1
Model 2
Model 3
Model 4
valence t
-.004
.003
fear t
4.995***
4.861***
money t-1
9.00e-06
money t-1
3.42e-06
units t-1
-.020**
units t-1
.048***
P level t-1
-.00011
P level t-1
.00012
Buy t-1
.355***
Sell t-1
.363***
6.50e-06
-.015
-.00011
3.87e-06
.044***
-.00012
Prob>chi2=0.000
Prob>chi2 =.1950
Prob>chi2 =0.000
Prob>chi2 =0.000
9769 obs
9769 obs
9970 obs
9970 obs
49 groups
49 groups
50 groups
50 groups
More neutral traders are less active submitting offers
bids t
asks t
Bids&asks t
neutrality t-1
-.247**
-.031
-.126*
money t-1
-.00008***
.00005***
9.81e-06
units t-1
-.058***
.054***
.020***
price level t-1
-.0002
.0001*
-.0001
bids t-1
.128***
asks t-1
.097***
Bids&askst-1
.103***
Obs: 9770
Obs: 9971
Obs: 9971
Groups: 49
Groups: 50
Groups: 50
Prob>F =.000
Prob>F =.000
Prob>F =.000
Better overall financial position improves traders’ emotional state
Valence t
Valence t
money t-1
2.51e-06*
4.01e-06**
units t-1
.0013*
.0022***
P level t-1
-.000044***
-.000087***
valence t-1
.480***
const.
-.028**
-.046***
Obs: 9927
Obs: 9970
Groups: 50
Groups: 50
Prob>F =.000
Prob>F =.000
Profitable purchases lead to higher valence
Valence t
Valence t
Valence t
Model 1
Model 2
Model 3
const
-.031***
-.016***
-.005*
buy t-1
.001
.001
.002
sell t-1
.004
.003
.002
.483***
.472***
valence t-1
(FV-Price)*buy t-1
.00002**
(Price-FV)*sell t-1
-.00001*
Obs: 9970
Obs: 9927
Obs: 8990
Groups: 50
Groups: 50
Groups: 49
Prob>F =.6734
Prob>F =.000
Prob>F =.000
R2=0.0001
R2=0.355
R2=0.351
Excessive prices
associated with
negative valence
At the market level, fear increases the probability of prices decreasing, while
the rest of the emotions appear to help sustain bubbles
Random Effects
Fixed Effects
Fear
354.45**
98.08
Neutral
-6.35**
-14.27***
Happiness
-6.14**
-15.19***
Anger
-5.40*
-14.30***
Disgust
-6.17
-20.25***
Sad
-2.96
-10.68**
constant
6.46**
Conclusions
• Bubbles and crashes in experimental markets are a complex phenomenon,
with many determinants of their magnitude.
• Cash balances and quantity of shares available for sale influence price
levels.
• Some time paths of fundamentals are more conducive to mispricing than
others.
• The preference parameters of risk and loss aversion predict prices and
quantity traded.
• Cognitive ability/motivation predicts mispricing.
• There are consistent emotional substrates to market bubbles and crashes.
• Emotional state of traders responds to and can predict subsequent market
activity.
Part 3: An experimental DSGE
economy
• Construct an experimental macroeconomy,
similar in structure to a New Keynesian DSGE
macroeconomy. Close enough so that hypothesis
from NK DSGE model can be evaluated.
• Three types of (infinitely lived) agents
– Consumers: supply labor, purchase (3) products, and save for
the future
– Producers: purchase labor, produce one of the (3) products, sell
output
– Central bank: sets interest rates
• Preferences and productivity subject to shocks
What we want to model
Producer incentives
• Maximize profit:
Пit = pityit – wtLit
yit = AtLit
At = A0 + γAt-1 + δεt
Where
Пit = profit of firm i in period t
pit = price of good i in period t
yit = production of good i in t
wt = wage in t
Lit = labor bought by i in t
At = productivity parameter in t
εt = productivity shock in t
γ = 0.8, δ = 0.2, A0 = 0.7
Consumer incentives
•
•
•
•
•
Payoff in period t of consumer j = βt[Ujt(Cjt) – Dj(Ljt)]
Ujt(Cjt)=∑ihijt[cijt(1-σ)/(1- σ)]
hijt = μij + τhijt-1 + δεjt
D(Ljt) = d*Ljt1+η/ (1+η)
Where
Cjt= consumption at time t of consumer j
Ljt = labor supplied at t
Dj(Ljt) = disutility to j of labor he supplies at t
cijt = consumption of good i by consumer j at t
ε jt = preference shock for consumer j in period t
β = .99, μij = 120, τ = 0.8, d = 15, η = 2, n = 3.
Consumer incentives
• Faces a budget constraint:
wtLjt + 1/n∑iΠi,t-1 + (1 + rt)sj,t-1 = ∑ipitcijt + sjt
• sjt can be thought of as savings or bonds
• Create monopolistic competition with
different preference shocks for each good.
Experimental Design
• Timing within a period
• Stage 1: Labor market
– There is a shock to productivity at the beginning of each
period.
– A double auction market operates for labor.
– Cost of supplying labor and productivity is (privately)
known at the time of trade.
– Sales take place in terms of (fiat) experimental currency.
Costs of labor supply are incurred in terms of utility
(Euros).
• Production occurs automatically
– Each producer has available a quantity of his product to
sell for stage 2
Labor market: Consumer
Labor Market: Producer
Stage 2 of a period:
Product market
• There is a shock to consumer preferences.
• Sellers post prices
• Buyers purchase units of each of the three products
at their own pace
– Product transactions take place in terms of (fiat)
experimental currency
– Valuations are in terms of utility (Euro paid to the subjects)
– It is possible that some units will go unsold, or that stock
will have been depleted at the time a consumer wants to
buy.
Product market: Producer
Product Market: Consumer
Savings, producer profit, discounting, and
ending the experiment
• Consumers’ unspent cash is saved for later periods, and earns interest.
• Producers’ unspent cash (profit) is awarded to the consumers in equal
shares.
– However, the agents acting as producers received a payment in Euro equal
proportionally to their profits. The payment was corrected for inflation.
• The game goes at least 50 periods, randomly stopping between periods 50
- 70.
• Utility (euro earnings) from consumption and labor supply exhibit a
decreasing trend of 1% per period.
• The final cash balance of consumers is “bought out” by the experimenter.
• Interest rate set by an instrumental rule:
rt = π* + 1.5(πt-1 - π*), π* = .03
where, πt = inflation in period t, π* = inflation target
Timing of a session
•
•
•
•
A session took 3 ¾ – 4 ¾ hours.
Instructions read (~30 minutes)
5 period practice economy (~30 minutes)
> 50 period economy that counted toward
earnings.
• Placed bounds on wages and prices for the
first two periods.
The treatments
• (1) Baseline
– The conditions described above
• (2) Human Central Banker:
In each period, three agents each chose an interest rate. The
group’s decision (and thus the rate in effect) was the
median of the three choices.
The agents had an incentive to minimize the loss function
Losst = (πt – π*)2
Central bankers were paid an amount equal to max{0, a –
b*Loss}
• (3) Menu Cost:
– To change the price from one period to the next, producers
had to pay a cost equal to: 0.025*pi,t-1*yit
– Otherwise identical to Baseline
• (4) Low Friction
– Valuations are the same for each good (μij = μ0), though
differ by individual and by time period (εjt > 0).
– Otherwise identical to Baseline
– Parameters set to equate welfare to Baseline under a
simulation we conducted.
Treatments
Monopolistic
Competition
Human central
banker
Menu cost for
product price
change
(= .025[pi,t-1*yit])
Baseline
Y
N
N
Menu cost
Y
N
Y
Human central
banker
Y
Y
N
Low friction
N
N
N
Hypothesis
• Persistence of shocks (effect beyond the current
period):
– Treatment differences
– In treatments Baseline, Human Central Banker, and Low Friction no
persistence, in treatment Menu Cost, shocks are persistent (both
Menu Cost and market power are needed for persistence in New
Keynesian DSGE model).
• Empirical stylized fact is that a shock to interest rates, output, or inflation,
has persistent effects on itself and on some of the other two variables.
• Also can compare between treatments
– GDP, inflation, welfare, employment, etc…
Results: GDP
GDP is highest under Low Friction
GDP is lowest in the late periods under Human Central Banker
Menu Costs do not affect GDP
Results: Inflation
-20
Inflation rate
0
20
40
Inflation - across treatments
0
10
20
30
40
50
ppp
HCB
menu_cost
baseline
low_friction
Inflation rate is similar on average in all four treatments,
including Human Central Bankers
Volatility is lowest under Menu costs
Volatility is highest under Human Central Banker
A degree of heterogeneity exists
within each treatment
600
400
0
200
Real GDP
800
1000
Real GDP - baseline treatment
0
10
20
30
40
ppp
session2
session11
session3
session12
50
Impulse Responses: Baseline
treatment
Shock to
GDP/Produc
tivity
Productivity
shock
persistent
Inflation
Shock
Persistent
Interest Rate
Shock
persistent
Row 1-3: Effect on
output, inflation,
interest rates
Shock to
demand
Inflation
Shock to
Interest
Rate
Price
Puzzle
Inflation
shock has
persistent
effect on
policy
Impulse Responses: Baseline
treatment
Shock to
GDP/Produc
tivity
Productivity
shock
persistent
Inflation
Shock
Persistent
Interest Rate
Shock
persistent
Row 1-3: Effect on
output, inflation,
interest rates
Shock to
demand
Inflation
Shock to
Interest
Rate
Price
Puzzle
Inflation
shock has
persistent
effect on
policy
Impulse Responses: Menu Cost
treatment
Shock to
GDP/Produc
tivity
Shock to
demand
Inflation
Shock to
Interest
Rate
Productivity
shock
persistent
Price
Puzzle
Interest Rate
Shock
persistent
Row 1 - 3: Effect on
output, inflation,
interest rates
Impulse Responses: Low friction
Shock to
GDP/Produc
tivity
Productivity
shock
persistent
Productivity
shock
lowers
prices
Row 1: Effect on
output
Row 2: Effect on
inflation
Row 3: Effect on
interest rates
Shock to
demand
Inflation
Shock to
Interest
Rate
Impulse Responses: Human Central
Banker
Shock to
GDP/Produc
tivity
Productivity
shock
persistent
Productivity
shock
lowers
prices
Interest Rate
Shock
persistent
Row s1-3: Effect on
output, inflation,
interest rates
Shock to
demand
Inflation
Shock to
Interest
Rate
GDP responds
positively to
interest rate
shock. Income
effect
Price
Puzzle
Inflation
shock has
persistent
effect on
policy
Evaluating the hypothesis
• Very little persistence in the Low Friction treatment.
• Somewhat more persistence in Menu Cost than in Baseline.
• More persistence of policy (interest rate) shocks in Human
Central Banker treatment
• Mixed evidence with regard to hypothesis. Monopolistic
competition alone generates shock persistence, and menu
costs add to it.
Persistence of shocks in each session,
all treatments
How do the human central bankers set interest
rates?
• Consider the following regression
• rt = β0 + β1rt-1 + β2πt-1 + β3(GDP – GDP*)
• System GMM Dynamic Panel Estimator (Blundell and
Bond, 1998)
• Taylor rule coefficient (relationship between interest rate and
inflation) = β2/(1- β1) [Bullard and Mitra, 2007]
Interest rate at t
rt
Estimate
(std err)
Constant
.4212
(.2609)
Interest rate at t1, rt-1
.9395
(.0146)
Inflation at t-1,
πt-1
.1538
(.0114)
Output gap
GDP-GDP*
.0184
(.0074)
• Implied coefficient is close to 2, consistent with the Taylor
principle since it is greater than 1.
Hazard Rate of Price Changes
Average duration of price spells
similar except for Menu Cost, in
which it is longer.
In Menu Cost, the hazard
function is upward sloping.
Price is more likely to change
the longer it has been
unchanged
Conclusions
• Methodology
– It is feasible to construct a DSGE model in the laboratory. It is possible
to verify stylized facts and consider the effect of some assumptions
from a behavioral standpoint.
• Persistence
– Monopolistic competition, in conjunction with multiple agents and
bounded rationality, is sufficient to generate some persistence
– Menu costs add somewhat to this persistence.
– Discretionary central banking affects shock persistence.
– Negligible persistence in Low Friction. Biases in decision making of
producers and consumers alone do not generate persistence.