Chaos and Complexity in Economics

Cars Hommes

CeNDEF, School of Economics

University of Amsterdam

18e Nationale Wiskunde Dagen

Noordwijkerhout, 3-4 Februari 2012

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Main Themes

I nonlinear dynamics and chaos

I financial market model with heterogeneous traders
I estimation of nonlinear switching model
I laboratory experiments and nonlinear models

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Why is economics so difficult?

Isaac Newton, about 1700

“I can predict the motion of heavenly bodies,
but not the madness of crowds”

irrationality is difficult to predict and to model

expectations are difficult to predict and to model

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
The Traditional Rational View
Expectations are model-consistent
Milton Friedman, 1953: Irrational traders will be driven out of the
market by rational traders, who will earn higher profits.

Robert Lucas, 1971: economic policy should be based on rational

expectations models in macro-economics
Cars Hommes University of Amsterdam
Chaos and Complexity in Economics
alternative, complexity, agent-based modeling approach
in Economics

I bounded rationality and behavioral economics versus perfect

rationality (Simon (1957) versus Lucas (1971)
I heterogeneous agents versus representative agent
I market psychology, herding behavior (Keynes (1936)) versus
rationality
I markets as complex adaptive, nonlinear evolutionary systems
versus representative agent model
I computational versus analytical approach

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Chaos in Nonlinear Systems
quadratic example: xt+1 = 4xt (1 − xt )

1 1 13

0.8 0.8

0.6 0.6
xt

t
x
0.4 0.4

0.2 0.2

0 13 0
20 40 60 80 100 20 40 60 80 100
t t

x0 = 0.1 x0 = 0.1001

sensitive dependence on initial conditions

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Period-doubling Bifurcation Route to Chaos
for Quadratic Map xt+1 = λxt (1 − xt )
f (x)=λx(1−x) and x =0.5
λ 0
1

0.8

0.6
f (x )↑
λ t

0.4

0.2

λ
∞
3 3.2 3.4 3.6 3.8 4
λ→

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Lyapunov Exponent
measuring sensitive dependence

Lyapunov exponent ≡
average rate of expansion/contraction along orbit
n−1
1X
λ(x0 ) ≡ lim ln(| f 0 (f i (x0 )) |),
n→∞ n
i=0

I λ(x0 ) < 0: stable periodic behaviour

I λ(x0 ) > 0: chaos and sensitive dependence

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Chaos in Quadratic Map xt+1 = λxt (1 − xt )
(a) f(x)=λx(1−x), x0=0.4 and n=1000

0.6

0.4

λ(x )↑ 0.2
0

−0.2

−0.4

3.4 3.5 3.6 3.7 3.8 3.9 4

λ→

(b) f(x)=λx(1−x) and x0=0.4

1

0.8

0.6
f (x )↑
λ t

0.4

0.2

0
3.4 3.5 3.6 3.7 3.8 3.9 4
λ→
Cars Hommes University of Amsterdam
Chaos and Complexity in Economics
Financial Market Model

Investors can choose between risky asset and a risk free asset
I R = 1 + r > 1: gross return on risk free asset
I risky asset pays stochastic dividends
I yt : stochastic dividend process for risky asset
I pt : price (ex div.) per share of risky asset
I price of risky asset determined by market clearing:
H
X
Rpt = nht Eht (pt+1 + yt+1 )
h=1

I nht : fraction of agents of type h

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Rational Expectations (RE) fundamental benchmark

Rpt = Et (pt+1 + yt+1 )

common beliefs on future earnings and prices

unique bounded RE fundamental price p∗t :
Et (yt+1 ) Et (yt+2 )
p∗t = + + ....
R R2
For special case of IID dividends, with E(yt+1 ) = ȳ
ȳ ȳ
p∗ = =
R−1 r
pricing equation in deviations xt = pt − p∗t from fundamental:
Rxt = Et xt+1
Cars Hommes University of Amsterdam
Chaos and Complexity in Economics
Behavioral asset pricing model (Adaptive Belief Systems)
standard asset pricing model with heterogeneous beliefs
one risky asset, one risk free asset
I price of risky asset determined by market clearing
I beliefs about future prices given by simple, linear rule
I forecasting strategies updated according to discrete choice
model with realized profits
equilibrium price of risky asset
H
X
Rpt = nht Eht (pt+1 + yt+1 )
h=1
(in deviations xt = pt − p∗t from RE-fundamental)
H
X
Rxt = nht Eht xt+1
h=1

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Fractions of Strategy Type h
fractions of belief types are updated in each period:
discrete choice model (BH 1997,1998) with asynchronous
updating:

eβUh,t−1
nht = (1 − δ) + δnh,t−1 ,
Zt−1
where Zt−1 = eβUh,t−1 is normalization factor,
P
Uh,t−1 past strategy performance, e.g. (weighted average) past
profits
δ is probability of not updating
β is the intensity of choice.
β = 0: all types equal weight (in long run)
β = ∞: fraction 1 − δ switches to best predictor

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Example with 4 belief types
(zero costs; memory one lag)
e
xh,t+1 = gh xt−1 + bh

g1 = 0 b1 = 0 fundamentalists
g2 = 1.1 b2 = 0.2 trend + upward bias
g3 = 0.9 b3 = −0.2 trend + downward bias
g4 = 1.21 b4 = 0 trend chaser

4
P
Rxt = nh,t (gh xt−1 + bh )
h=1
β
(gh xt−2 +bh −Rxt−1 )(xt −Rxt−1 )
e aσ2
nh,t+1 = , h = 1, 2, 3, 4
Zt

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Chaos in Financial Market Model with
Fundamentalists versus Chartists
(a) (b)
0.75 1
0.5
0.5
0.25

0 0

-0.25
-0.5
-0.5

-0.75 -1

100 200 300 400 500 100 200 300 400 500

(c) (d)
0.1
0.75
0.5 0.05
0.25
0 0
-0.25
-0.5 -0.05

-0.75
-0.75 -0.5 -0.25 0 0.25 0.5 0.75 -0.05 0 0.05 0.1

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Chaos in Financial Market Model with
Fundamentalists versus Chartists

0.2

0.15

0.1

0.05

−0.05

−0.1

−0.15
40 50 60 70 80 90 100

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Model very sensitive to noise

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Nearest Neighbor Forecasting

0.8
Prediction Error

0.6

0.4

chaos
0.2 5% noise
10%
30%
40%
0
0 2 4 6 8 10 12 14 16 18 20
Prediction Horizon

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
 1+g
S&P 500, 1871-2003 + benchmark fundamental p∗t = 1+r yt
(g constant growth rate dividends)

100
7.5 S&P 500 PY Ratio
Fund. Value Fund. Ratio
90
7.0
80

6.5
70

6.0 60

50
5.5
40

5.0
30

4.5 20

10
4.0

1880 1900 1920 1940 1960 1980 2000 1880 1900 1920 1940 1960 1980 2000

log S&P 500 and fundamental PD-ratio and PD-fundamental

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Estimation 2-type model
R∗ xt = nt {0.762 xt−1 } + (1 − nt ){1.135 xt−1 } + b
t
(0.056) (0.036) (1)

nt = {1 + exp[−10.29(−0.373xt−3 )(xt−1 − R∗ xt−2 )]}−1

(2)
(6.94)

R2 = 0.82, AIC = 3.18, AICAR(1) = 3.24, σ = 4.77, QLB (4) = 0.44

1.0 1.1

0.8
1.0

0.6
φt
nt 0.9
0.4

0.8
0.2

0.0 0.7

1880 1900 1920 1940 1960 1980 2000 1880 1900 1920 1940 1960 1980 2000

fraction fundamentalists average market sentiment

Cars Hommes University of Amsterdam
Chaos and Complexity in Economics
How to model bounded rationality?

wilderness of bounded rationality:

"There is only one way you can be right, but there are many ways you
can be wrong"
I use laboratory experiments with human subjects to test a
behavioral theory of heterogeneous expectations
I fit simple complexity model to laboratory data
I test simple complexity model on real economic/financial data

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Example: Laboratory Experiments wit Human subjects
psychology – behavioral economics – computer science

Computer Screen Learning to Forecast Experiment

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Rational Expectations Benchmark

Rational expectations
70

If everybody predicts 65
rationally fundamental
60
price, Price
then 55

εt 50
pt = pf +
1+r
45

40
0 10 20 30 40 50

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Naive and Trend-following Expectations Benchmarks

Naive expectations AR(2) expectations

70 70

65 65

60 60
Price

Price
55 55

50 50

45 45

40 40
0 10 20 30 40 50 0 10 20 30 40 50

which expectations theory is correct?

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Prices in the Experiment with Humans
Group 2 Group 5
70 70
fundamental price experimental price fundamental price experimental price
65 65
60 60
Price

Price
55 55
50 50
45 45
40 40
0 10 20 30 40 50 0 10 20 30 40 50

Group 1 Group 6
70 70
fundamental price experimental price fundamental price experimental price
65 65
60 60
Price

Price
55 55
50 50
45 45
40 40
0 10 20 30 40 50 0 10 20 30 40 50

Group 4 Group 7
90 70
80 fundamental price experimental price fundamental price experimental price
65
70
60 60
Price

Price

50 55
40 50
30
20 45
10 40
0 10 20 30 40 50 0 10 20 30 40 50

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Individual Forecasts in Experiments with Humans
group 1 group 2
80 80
60 60
40 40
20 20
10 group 3 50 10 group 4 50
80 100
60
50
40
20 0
10 group 5 50 10 group 6 50
80 80
60 60
40 40
20 20
10 group 7 50 10 group 8 50
80 100
60
50
40
20 0
10 group 9 50 10 group 10 50
100 100

50 50

0 0
10 50 10 50

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Heterogeneous Expectations Hypothesis

Heuristics Switching Model

I there are a few simple heuristics
I adaptive expectations ADA
I trend extrapolating rule STR, WTR
I anchor and adjustment rule LAA
I impact of heuristics changes over time
I agents gradually switch to heuristics that have performed better
in the recent past

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Group 5 (Convergence)

Parameters: β = 0.4, η = 0.7, δ = 0.9

70 1
simulation experiment ADA WTR STR LAA

65
0.8

60
0.6
55
0.4
50

0.2
45

40 0
0 10 20 30 40 50 0 10 20 30 40 50

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Group 6 (Constant Oscillations)

Parameters: β = 0.4, η = 0.7, δ = 0.9

70 1
simulation experiment ADA WTR STR LAA

65
0.8

60
0.6
55
0.4
50

0.2
45

40 0
0 10 20 30 40 50 0 10 20 30 40 50

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Group 7 (Dampened Oscillations)

Parameters: β = 0.4, η = 0.7, δ = 0.9

70 1
simulation experiment ADA WTR STR LAA

65
0.8

60
0.6
55
0.4
50

0.2
45

40 0
0 10 20 30 40 50 0 10 20 30 40 50

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics
Concluding Remarks

I chaotic financial market model mimics bubble and crash

dynamics
I simple 2-type model with fundamentalists versus chartists fits
US stock market data
I nonlinear heuristic switching model with path dependence fits
experimental data
I theory of evolutionary selection of heterogeneous expectations
fits experimental data

Cars Hommes University of Amsterdam

Chaos and Complexity in Economics