Bounded Rationality and Cycles in Real

Estate Markets
Bernhard Borges
Max Plank Institute
Winslow Farrell
Coopers & Lybrand
Todd Kaplan
Ben-Gurion University
Charles Tansey
Commonwealth Capital Partners, Inc.
October 18, 1996

Abstract
We present a model of commercial real estate. We solve for the
steady state equilibrium and show the response to a shock to the
system under rational expectations. We then introduce bounded ra-
tionality into the system in two ways. First, we replace the perfectly-
rational, perfectly-predicting agents with adaptive agents. Second,
we replace the perfect markets with adaptive markets agents. The
response to the bounded rationality approach is more indicative of
actual real estate markets.
 This research is part of a project by the Emergent Systems Group at Coopers &
Lybrand. E-mail correspondence about this paper to todd@bgumail.bgu.ac.il

1
 1 Introduction
Sargent (1993) states that rational expectations has two environmen-
tal conditions. First, that agents are individually rational; that is,
given their beliefs, information, and knowledge,1 they make decisions
that maximize their expected utility. Second, that the beliefs of agents
are consistent. Their beliefs about strategies of other agents must be
consistent with the actions of those agents and their beliefs about the
environment must be consistent with the information they possess.
In macroeconomic models with a continuum of agents, beliefs are
formed about future prices and aggregate stock variables (such as
capital). These conditions force all agents' predictions about future
prices to be accurate, eliminating any arbitrage possibilities.2 In such
models, there is a third condition, market clearing. Not only must the
beliefs about future prices be consistent, but today's price along with
agents' beliefs would be such that demand is equal to supply for each
good.
Some research has been done only relaxing these assumptions((Bray
1982)). LeBaron (1995) summarizes results of the Santa Fe Articial
Stock Market. In this market, the second condition of rational expec-
tations is relaxed. Agents are allowed to have dierent beliefs about
future prices. Using these beliefs, the market program nds today's
price that results in market clearing. This is not necessarily easy, since
these beliefs are based in part on today's price.
We propose that while the agents in such simulations are bound-
edly rational, if we think of the market itself as an agent, then the
market is omniscient in that it not only knows the beliefs of the agents,
but it knows how today's price aects these beliefs and decisions. By
relaxing the market clearing assumption, we hope to keep the market's
rationality on par with the agents' rationality.
Although such an experiment can be applied to a large range of
models, we wish to analyze a model of commercial real estate. A key
feature of this model is that investors in the real estate market are
collaterally constrained. This feature is based upon the Kiyotaki and
Moore (1995) model of credit cycles. We structure our paper around
They implicitly have common knowledge.
1

Prices are a function of both the actions of agents and events of nature. Since agents
2

must accurately predict both of these, they must also accurately predict price. Agents
may of course have a probability distribution as a prediction.

2
 our simulation experiment. In section 2, we present a description of
the model. In section 3, we describe a equilibrium concept and solve
for the steady state equilibrium. In section 4, we present the results of
an unanticipated shock to the steady state and show the eects caused
by the shock under rational expectations. In section 5, we introduce
bounded rationality into the model. Finally, in our last section, we
discuss results and discuss paths of future research.

2 Model
There are four types of innitely-lived agents: developers, investors,
renters, and banks. There are three goods in the model: buildings,
housing services, and food (the numeraire good). The interaction of
agents in the model is displayed in gure 1. The developers construct
xt buildings. They sell these buildings to the investors at price qt .
The investors borrow bt from the banks to nance their purchase of
buildings. The investors use the buildings to sell housing services to
the renters at price pt.
Renters are endowed with a certain amount of food each period
and have utility for both housing services s and food f . There is no
market clearing constraint for food and interest rates are exogenous.3
The renters have utility function:
u(s; f ) = s + f
where 0 <  < 1. They exhibit decreasing returns in utility to
housing service and constant returns in utility to food.
Developers have the unique ability to convert food into buildings.
This technology has increasing per period costs; the cost of producing
xt buildings is xt (where  > 1). The production from this technol-
ogy is used to produce both new buildings and replace depreciation
of old buildings, which depreciate at rate  . The supply of buildings
is competitive: developers construct buildings whenever the price of
buildings is above their marginal cost, and developers are not collat-
erally constrained.4
Investors have constant returns to scale technology in converting
buildings into housing services. Unlike developers, investors are col-
3
This is a partial equilibrium model.
4
This may change in future models.

3
 Buildings Housing
Services

Developer Investor Renter

$ $
$ $

Bank

Figure 1: Diagram of exchanges in the model.

laterally constrained. The investors can borrow up to of the value
of their buildings. In addition, investors have utility such that they
consume ct which is a constant fraction of their wealth.5

3 Rational-Expectations Equilibrium
A market equilibrium is dened as a sequence of fpt; qt; ct; kt; xt; btg
such that (1) given pt , renters choose an amount of housing services
to purchase to maximize their expected discounted utility. (2) Given
qt , developers choose to build amount xt . (3) Given qt, pt and the
collateral constraint, investors purchase kt buildings to rent, consume
ct and borrow bt from the banks. (4) Markets clear.
The investors' wealth at time t, wt , is equal to the prots they get
from renting housing services plus the value of their property (minus
P
An example of this is  t u(ct ) where u(ct) = log(ct ). With constant-returns-to-scale
5

technology, investors can substitute between today's and tomorrow's consumption at a

constant rate. Given this, investors consume 1+w where w is wealth. When utility is
ct 
(1 )

1 
, investors also consume a fraction of their wealth, however, this fraction depends
upon marginal rate of substitution which may vary out of equilibrium.

4
 depreciation) minus both their debts to the banks and consumption:
wt = ptkt 1 + (1  )qt kt 1 bt 1 (1 + r) ct

3.1 Collateral constraint

The investors are limited to how much they can borrow. The banks
are only willing to let them borrow up to fraction of the value of their
collateral, their buildings. The key feature of this constraint is that
the bank looks at what it thinks tomorrow's value will be. Naturally,
a bank would not loan an investor a hundred thousand dollars on a
home worth hundred twenty thousand dollars if the bank thinks the
value of the home will drop to eight thousand dollars.
An investor needs to borrow bt = qt kt wt . The bank is willing to
lend him bt = qt+1kt .6
By combining these two equations, we can solve for kt .
kt = q w qt
t t+1
This can be thought of as the aggregate demand for buildings.
Notice the higher the bank expects future prices, the higher amount
of buildings the investor will be able to buy. Thus, in a boom the
bank lends out more, requiring less current collateral, and in a bust a
bank lends less, requiring more current collateral.
By looking at the developers' problem, we can solve for the supply
of buildings.
kt = (1  )kt 1 + ( qt )1= 1

3.2 Steady State

A Steady State equilibrium is prices ps ; q s and amounts cs ; ks; bs; xs
such that:
 Given ps, renters rent ks .
 Given ps and qs, investors hold ks buildings.
 The collateral constraint is binding (bs = qsks).
6
For simplicity, we assume re ects interest rates and depreciation. Otherwise, the
equation would be bt = (1+
(1  )
r) qt+1kt

5
  Investors consume cs equal to the net prots they earn.
 Given qs developers build enough buildings to replenish the de-
preciation on old buildings (xs = ks )
Renters want to maximize their utility. They do this by choosing a
k that maximizes k pk. The developer wants to maximize his period
prots. He chooses x buildings to construct to maximize qx x .
The steady state equilibrium conditions are
ps =  (ks) 1 (1)
bs = q sks (2)
qs = (xs ) 1 (3)
xs = ks (4)
cs = [ps q s ( + r )]ks (5)
cs = (1  ) q sks (1 ) (6)
In this model, there are two stocks: the number of buildings and
the wealth of the investors. We can reduce the number of equations
to two.
cs = (1 )  1k (7)

cs = k   1 ( + r )k (8)
The two variables in these equations are consumption and capital.
Consumption is a proxy for wealth since consumption is a constant
fraction of wealth. The more wealth investors have the more capi-
tal they are able to buy. This credit-wealth curve is represented by
equation 7. The amount of prot investors make on capital depend
upon how much capital exists. There is a negative externality between
investors; higher amounts of capital lowers rent prices and increases
costs of replacing capital depreciation. Without the collateral con-
straint, consumers will borrow until they reach zero prots. With the
collateral constraint, they will still earn prots. The maximum point
of the prot-capital curve is the point that a monopolist investor would
choose. Figure 2 shows the steady state equilibrium as the intersection
of these two equations.
6
 3
credit-wealth
2.5 prot-capital

2
1.5
C

1
0.5
0
0 2 4 6 8 10 12 14
K
Figure 2: Prot-capital and credit-wealth curves with (; ; ; ; ; r) =
(1:5; :7; :5; :1; :8; :1)

4 Shocks to the system

We allow the model to converge to steady state. We then introduce an
unanticipated shock to the system.7 Several dierent types of shocks
can occur.
Debt If there is a sudden jump in income or rebate in taxes, then
investors may see a one-time reduction of debt. The immediate
result is that investors will have more income to spend.
Demand A sudden increase (or decrease) of the renters' demand for
housing services. The long term consequences should be an in-
crease in both the price and quantity of housing services.
Interest Rate A sudden increase in the rate of interest banks charge
investors to borrow money.
7
This can be justied as a low probability event. If probability is low enough, the
system would behave as in the no uncertainty case until the event happens. When the
event happens, it will be equivalent to an unanticipated shock.

7
 Collateral Requirement This is refers to when banks suddenly re-
quire less capital.
Capital A sudden reduction in the number of buildings. For example,
a hurricane suddenly wipes out some buildings.
Input Technology This corresponds to a shock in the cost of build-
ing houses. This could be a sudden shock to the price of building
materials or the discovery of a new ecient technology.
In gures 3, 4, and 5, we show the eects from a a sudden injection
of money into a system in steady state. This injection is done by
reducing the investors' debt by ten percent. The price of buildings
(see gure 3) increases dramatically and then settles to near steady
state. This is a result of the investors spending their extra money
(they are able to borrow more) on buildings. This increased price is
short-lived because it is a one-time injection. Also, the rental price
suddenly drops (see gure 4) due to the extra buildings (see gure 5).
This drop decays over time due to depreciation.

1.19
1.18
1.17
1.16
1.15
price

1.14
1.13
1.12
1.11
1.1
0 2 4 6 8 10 12 14 16 18 20
time
Figure 3: The price of buildings with rational expectations

8
 0.421
0.4208
0.4206
0.4204
0.4202
price

0.42
0.4198
0.4196
0.4194
0.4192
0.419
0 2 4 6 8 10 12 14 16 18 20
time
Figure 4: The rental price with rational expectations
5 Model with Adaptation
5.1 Adaptive markets and price adjustment
We introduce a tantonnement price adjustment mechanism to the
model.8
In this process, a price p adjusts by the following equation:
p0 = p
(1 + ( D S ))
D+S
The adjustment parameter  ( 1 >  > 0 ) determines how fast the
prices adjusts to dierences between supply and demand. We use this
adjustment mechanism for both building prices q and housing service
prices p.
8
The tatonnement price adjustment process was the initial focus of research on equilib-
rium stability (Jordan 1986), until Scarf [1960] constructed an example where this process
does not converge to a unique Walrasian Equilibrium. Jordan shows that a tantonnement
process would require the aggregate excess demand (demand minus supply) and informa-
tion about its derivative in order to converge.

9
 5.54
5.53
5.52
5.51
5.5
number

5.49
5.48
5.47
5.46
5.45
0 2 4 6 8 10 12 14 16 18 20
time
Figure 5: The number of buildings with rational expectations
Because of this adjustment process, there are times when demand
is not equal to supply. During such times, an even rationing rule is
used to determine who buys and who sells.9
We also create two agents to govern this adjustment rule, one for
each market. The only goal of these agents is to clear the market.
They accomplish this goal by changing the adjustment parameter .
Currently, the rental market agent chooses a value for . It then sees
the performance of this rule. It can judge the performance in the
following way. If after an increase the price demand is still greater
than supply, then the agent realizes that it should have increased it
more. Thus, the agent increases . Likewise, if after the price increase
demand is less than supply, then the agent realizes that the price was
increased too much. Thus, it lowers .
Kiyotaki and Moore [1996] have a probability  that someone can invest in depreciable
9

capital. This causes heterogeneity among their agents.

10
5.2 Adaptive Agents
The banks in this model are the adaptive agents. They use the past
history of the price of buildings to predict the future price of buildings.
This future price is crucial in determining how much to lend to the
investors. A predicted price that was under the actual price would
cause the investor to be overcollateralized. This implies a missed prot
opportunity. A predicted price that was above the actual price would
cause the investor to be under-collateralized. This would require the
investor to liquidate some of his buildings.
We use two methods of prediction. In both methods, we start with
a relationship. The future price is a function of the previous prices.
An example of such a function is
qt+1 = (qt qt 1) + qt + t
This function says that the future price will re ect a change from
today's price by an amount proportional to the change from last period
plus noise.
The rst method that we use is one of adjustment. It starts o an
estimate of . It predicts a future price. If the future price is dierent
from its prediction, it adjusts  by the following.
0 = 
 + (1 )[ qqt+1 q qt ]
t t 1
The new predictor 0 is equal to a linear combination of the old
predictor and what the predictor should have been to be correct last
period. This puts more emphasis on more recent events.
The second method is a least-squares regression with limited mem-
ory, that is, a regression with memory m would regress over the past
m periods and ignore the history before that.

5.3 Adaptive Model

The adaptive model is shown in gure 6. The renters determine the
demand for housing services. This indirectly aects the price for hous-
ing services through the market agent for housing services. The prot
from this market goes back to the investors. This changes the in-
vestors' net worth which in turn aects their ability to borrow.
The price for buildings aects the banks' prediction

11
 Developers
Banks q
Credit

q
Investors Dq Market S
q

profit

p Market p
D S
p

p
Renters

Figure 6: Adaptive Model.

12
 1.12
1.115
1.11
1.105
price

1.1
1.095
1.09
1.085
1.08
1000 1020 1040 1060 1080 1100
time
Figure 7: The price of buildings with a money injection

0.425
0.424
0.423
price

0.422
0.421
0.42
0.419
1000 1020 1040 1060 1080 1100
time
Figure 8: The rental price with a money injection

13
 5.5
5.4
5.3
5.2
amount

5.1
5
4.9
4.8
1000 1020 1040 1060 1080 1100
time
Figure 9: The number of buildings owned by the investors with a money
injection

2.2
2
1.8
1.6
1.4
1.2
amount

1
0.8
0.6
0.4
0.2
0
1000 1020 1040 1060 1080 1100
time
Figure 10: The cumulative loss by the banks with a money injection
14
 1
0.95
0.9
0.85
renter population

0.8
0.75
0.7
0.65
0.6
0.55
0.5
0 50 100 150 200 250 300 350 400 450 500
time
Figure 11: The demand shift

0.43
0.42
0.41
0.4
price

0.39
0.38
0.37
0.36
1000 1100 1200 1300 1400 1500
time
Figure 12: The price of buildings with a demand shift

15
 5.3.1 The asymmetry and changes in the collateral con-
straint.
In gure 14, we show the eects of both a tightening and loosening
of the collateral constraint. With rational expectations, the eect is
symmetric. When the constraint is loosened, there is an unexpected
increase in demand for buildings. The increase in demand increases
the price. This increases the investors' wealth which further increases
demand. When the constraint is tightened, the opposite eect hap-
pens. The asymmetry in the adaptive case is caused by the rationing
rule. When demand is greater than supply, the investor is hindered
from increasing his buildings. When supply is greater than demand,
the investor is not slowed from lowering his holdings.

6 RISDIC data
The Rhode Island Share and Deposit Indemnity Corporation (RIS-
DIC) collapse occured in December of 1990. We use building permit
data obtained from the U.S. Census Bureau to analyze construction
before and after the collapse. Figures 18, 17, and 16 show the esti-
mated value of the buildings in two way: cumulative and smoothed.
While the cumulative graph is simply the sum of all the values from
the start of the sample until the point plotted, the smoothed graph
is the sum of all values after multiplying them by (1  )t where t
is the number of months before the current date. This exponential
smoothing technique helps adjust for the uncertainty of construction
and turns the graph into an appropriate proxy for new buildings on
the market.
In the smoothed graphs, we see the single-family housing (see g-
ure 18) is pretty much at a steady state throughout the time period
analyzed.10 In contrast, the multi-family housing had a slight increase
in construction in the early part of 1989 followed by a crash in late
1989 bottoming out in 1993-94 (see gure 17). This is similar to what
happened in oce buildings except the acceleration is slightly later
and pronounced.
In the cumulative gures, an increase in slope corresponds to an
Single-family housing consists of townhouses and detached one-family homes excluding
10

mobile homes.

16
 1.12
1.1
1.08
1.06
price

1.04
1.02
1
0.98
0.96
1000 1100 1200 1300 1400 1500
time
Figure 13: The rental price with a demand shift

1.2

1.15

1.1
price

1.05

1 adaptive loosening
adaptive tightening
rational tightening
rational loosening
0.95
0 10 20 30 40 50 60
time
Figure 14: The price of buildings with a change in the collateral constraint

17
acceleration in the construction market and a decrease in slope cor-
responds to a deceleration in the construction market. In the single-
family housing market, the slope is constant. In the multi-family
housing market, the slope shows sharp decreases starting at the end
of 1989 and continuing until through 1990. The slope shows a modest
increase in 1994. In the oce market, there are several periods of
acceleration and deceleration; the most noticable is the acceleration
in the middle of 1989 and the deceleration starting in the end of 1990.
Our paper presents a model in which many shocks cause a general
pattern of a small boom followed by a large bust and a slow recovery.
We see a strong indication of this in oce permit data and a mild
indication of this in multi-family housing data. We see no indication
of this with single-family housing. This gives us a strong indication
that, as we assumed, the developer is not constrained. In addition
while some single-family housing owners would be constrained, their
ability to pay would be less vulnerable to price uctuations. Thus
banks would not require additional collateral.

7 Conclusion
We nd that the results of our model under adaptive expectations and
adaptive markets to be more realistic. Our future plans is to calibrate
this model and compare results to the RISDIC (Rhode Island Share
and Deposit Indemnity Corporation) collapse. We believe that such a
model can be extremely useful in helping lending institutions measure
systemic risk.

18
 1.2
adaptive
1.18 rational
1.16
1.14
price

1.12
1.1
1.08
1.06
0 10 20 30 40 50 60 70 80 90
time
Figure 15: The price of buildings with an injection of money.

100
90 cumulative
smoothed
80
70
price (in millions)

60
50
40
30
20
10
0
88 89 90 91 92year93 94 95 96 97

Figure 16: The estimated value of new oce buildings in Rhode Island

19
 200
180 cumulative
smoothed
160
140
price (in millions)

120
100
80
60
40
20
0
88 89 90 91 92year93 94 95 96 97

Figure 17: The estimated value of new multi-family housing in Rhode Island

2
1.8 cumulative
smoothed
1.6
1.4
price (in billions)

1.2
1
.8
.6
.4
.2
0
88 89 90 91 92years93 94 95 96 97

Figure 18: The estimated value of new single-family housing in Rhode Island

20
References
Bray, M. (1982): \Learning, Estimation, and Stability of Rational
Expectations," Journal of Economic Theory, 26, 318{339.
Gjerstad, S., and J. Dickhaut (1995): \Price Formation in Double
Auctions," Discussion Paper No. 284, Department of Economics,
University of Minnesota.
Jordan, J. (1986): \Instability in the Implementation of Walrasian
Allocations," Journal of Economic Theory, 39, 301{328.
Kiyotaki, N., and J. Moore (1995): \Credit Cycles," Working
paper, Financial Markets Group, LSE.
LeBaron, B. (1995): \Experiments in Evolutionary Finance," Work-
ing paper, University of Wisconsin.
Ortalo-Magne, F. (1996): \Measuring the eects of credit market
imperfections: A U.S. farmland application," Manuscript, LSE.
Sargent, T. J. (1993): Bounded Rationality in Macroeconomics.
Oxford University Press, Oxford, UK.

21