OTT Steven

A Generalized Real Options Model of Land:
Valuation, Expected Returns and Development

Richard Buttimer
Belk College of Business
University of North Carolina Charlotte
Charlotte, North Carolina 28223-0001
(704) 687-7695
buttimer@uncc.edu

Steven P. Clark
University of North Carolina Charlotte
Charlotte, North Carolina 28223-0001
(704) 687-7689
spclark@uncc.edu

Steven H. Ott
University of North Carolina-Charlotte
Charlotte, North Carolina 28223
704-687-7630
shott@uncc.edu

January, 2009

We thank the seminar participants for suggestions and comments made at The 2007 UNC
Charlotte-Cambridge University Symposium on Real Estate Risk Management and
Property Derivatives, The 2007 Maastricht-Cambridge-MIT Real Estate Finance &
Investment Symposium, and the 2008 Homer Hoyt Institute International Conference.

A Generalized Real Options Model of Land:
Valuation, Expected Returns and Development

Abstract

We develop an equilibrium model of land valuation, expected returns and development.
The model extends the standard real options framework by combining cash flow
uncertainty with uncertain buyer reservation prices. These reservation prices also serve as
a proxy for general economic conditions. The model allows for the realism of lumpy
uncertain demand and absorption of space over time where renters/buyers arrive and
depart according to a Poisson process with a stochastic intensity that is a function of the
uncertain variables. Buyers and renters can arrive and depart before construction has
commenced, while additional buyers/renters can arrive during and after construction.
This structure allows for presales/preleasing of space to mitigate developers risk. Given
various parameter sets, we numerically solve the model and derive numerous outputs
including: the equilibrium value of undeveloped land, the equilibrium value of developed
space, the optimal construction commencement trigger as a function of the presales, price
and expected absorption levels, the expected time to construction, and the expected return
on land investment.

1

1. Introduction

Researchers have long recognized holding vacant land as an option granting the right to
develop a building in the future. Titman (1985), Capozza and Helsley (1990), Williams
(1991), Grenadier (1995) and others have used this analogy to develop theoretical real-
option models revealing the importance of uncertainty in the valuation of land and
optimal timing of the development decision. Williams (1993) and Grenadier (2005) have
extended the models employing a partial equilibrium framework by incorporating supply
competition among developers and uncertain demand in a continuous-time Nash
equilibrium. Williams (1993) shows how land values and development decisions are
affected in the presence of less than perfect competition and downward sloping demand.
Grenadier (2005) uses a similar setting to derive the stochastic properties of the
equilibrium cash flow levels resulting in the derivation of the term structure of lease
rates.
Several empirical studies have tested and validated theoretical real-option models
for land valuation and development decision-making. Holland, Ott and Riddiough (2000)
specify and empirically estimate a structural model of the commercial real estate asset
market equilibrium. They find the real-options model in generaland the uncertainty
variable in particularhas significant power in explaining aggregate commercial real
estate investment and development cycles.
1
Other studies by Quigg (1993) and
Cunningham (2006) provide empirical evidence of real option values embedded in land
prices in the residential housing market.

1
Other related empirical work includes a study by Sivitanidou and Sivitanides (2000). Examining
investment over time for the U. S office market, they find results consistent with those of Holland, Ott and
Riddiough (2000).

2

This paper builds upon previous theoretical research by employing a methodology
similar to that of Williams (1993) and Grenadier (2005) to model the real estate market in
equilibrium. We modify their structure of real estate markets and land development by
relaxing the assumption that tenants or buyers are continuously available in the market.
Uncertain demand and absorption of space are alternatively modeled as a lumpy process.
Specifically, we treat demand as tenants/buyers arriving and departing the market
according to a Poisson process. The stochastic intensity of this demand is a function of
the ratio of the market service/cash flow to the mean buyers reservation service/cash
flow, both of which are stochastic.
2
The reservation flow also serves as a proxy for
general economic conditions. This approach allows us to model lumpiness and
uncertainty in space absorption, which lets us address real risks faced by developers, such
as the risk of unsold inventory and/or declining prices after construction has begun. We
provide for the ability of the developer to prelease or pre-sell space before, during and
after construction to mitigate these risks.
3,4
Optimal development decisions are shown to
be a function of presale/prelease levels, the intensity function described above, and
expected absorption levels.
Similar to Grenadier (2005), we find equilibrium building cash flows at the point
where additional supply is added to the market. Equilibrium cash flow in our model is the

2
If we are modeling the residential housing market we would use a service flow. For the commercial real
estate markets we would use cash flow.
3
The basic option-based model has been extended to examine strategic real estate development. Grenadier
(1996) shows that investment timing options for projects held by competitors, combined with the
strategically optimal exercise of these options, can explain both simultaneous investment in competing
projects and long lag times between investments. Grenadier (1999) and Childs, Ott and Riddiough (2002)
extend this strategic stetting, to one where there is either asymmetric or incomplete information among
competing developers. In these models actual development reveals information to competitors. This work
provides further explanation as to why supply boom and bust patterns are observed in many localized real
estate markets.
4
See Buttimer and Ott (2007) for a similar set-up for already built properties, but one without the
determination of a market equilibrium, land valuation, time to build and optimal development timing.

3

level where land developers are willing to presell/prelease space preceding an ultimate
development decision. The equilibrium cash flow level can then be mapped into an
expected absorption of space over time.
We model the cash flow level and the tenants/buyers reservation cash flow as
mean-reverting stochastic processes reverting to the endogenously determined long-run
equilibrium. These processes reflect empirically observed economic cycles and slow real
estate supply adjustments due to time to build. Specifying the correlation between the two
uncertainties along with the elasticity of demand allows us to model development in
varying local economic environments.
We solve the model numerically for a given set of input parameters to determine
the real estate markets equilibrium cash flow, which in then used to compute, in closed-
form, the markets equilibrium price per completed development unit. Based on this
market equilibrium, we determine undeveloped land values, the preleasing/presales
necessary to initiate development, and expected time to development for properties
without the requisite presales. We are also able to determine the expected return to
investing in undeveloped land. Perturbing the input parameters allows us to isolate how
the change affects each of these values and determine those parameters that most affect
the results.
Empirical evidence and investor surveys support the outputs of our model. For
example, in the preponderance of the cases where parameters are varied, we find that
expected returns on speculative land investment generally range from 12% to 25%. Risk
and expected returns are affected primarily by completed space demand and absorption
sensitivity, development timing risk and the degree of leverage inherent in the land

4

option. Absorption times to sell or lease any remaining units that are not presold/leased
run 3 to 4 years on average. Absorption levels and times are relatively constant across
different parameter sets as completed project prices adjust in equilibrium to smooth
changes in absorption. Land values range from 12% to 25% of completed project value
and presales needed to commence construction are generally 33% to 60% and increase
with the level of risks inherent in the land.
In the following sections, we first develop the model and present the general
economic environment in which it is placed. We demonstrate that we can solve this
model using standard contingent-claims techniques. Solutions to the model for a variety
of economic environments are then presented and the implications of the model are
discussed. The analysis concludes with a discussion of our findings.

2. An Economic Model of Real Estate Markets
We develop a model applicable to land development and space markets in both
commercial and residential settings. Each of these settings exhibits clear differences in
the rate at which potential buyers arrive, per-unit construction costs, and the level of
uncertainty regarding future sales prices.
5
However, these differences are primarily
parametric and the fundamental economic process of land development is the same for
both settings. The commonality allows us to build a general development model.

2.1. Basic Features of Land Development
We begin by describing the general land development process. A landowner sells to a

5
We will use the term buyer from this point forward to refer to either buyers or renters (tenants) of space.
Price will be used to refer to the price of space for sale or the rental price of space for rent, where rent flow
maps directly into a space value function.

5

developer a piece of undeveloped land at time 0. The developer will, either at time 0 or at
some point in the future, begin the process of developing N units on that land.
6
The
developer chooses when to begin construction. Once started, construction takes T
C

months to complete.
7
We assume that the developer will develop all N units
simultaneously to maintain economies of scale.
The developer will seek to pre-sell units prior to and during the construction
period with delivery of all pre-sold units upon completion.
8
Buyers in the pre-
construction phase place deposits on their units, but have the option to abandon their
purchase (by giving up their deposit) should they chose to do so at any time prior to the
beginning of construction. Once construction begins, however, they are committed to
purchasing the unit. Buyers that purchase units after construction has begun do not get
the option to terminate their purchase. Note that these purchasers will pay the then-
prevailing market price for the unit at closing. A developer unable to pre-sell their entire
inventory will close upon the remaining units at the end of construction at the spot market
price, subject to stochastic buyer arrivals.
Before developing the model to determine land value, we must first develop
several of the models sub-components. These include features such as the property
service flow, which maps into the space price function, and the developer cash flows. We
define these components in the following sub-sections.

6
We use the somewhat vague term units intentionally. The model is flexible enough to handle cases
where the units are developed lots which are sold to homebuilders or commercial developers, and also
cases where the units are residential and commercial buildings that are rented and/or sold.
7
We do have to place a constraint on the developers option to begin development: they must do so prior to
reaching some development horizon, T
H
. We set this development horizon far enough into the future where
discounting effects will render it moot. Normally we set this to 15 years in our numerical solutions.
8
We use the term pre-sold for convenience, if the developer is creating rental space they would be pre-
leasing units.

6

2.1.1. The Service Flow Processes for Built Space
The developers return on investment is based primarily upon the price received for the
units created. Ultimately the units values are derived from the economic benefits they
produce, i.e. the present value of the service flows (cash flows) they provide.
9
In our
model, the real estate market for built space is perfectly competitive where all market
participants are price takers with respect to the equilibrium spot service flow rates.
First, we let
{ } ( )
0
, , ,
t
t
F F P be a filtered probability space, where the { }
0
t
t
F
(the information structure) is the P -completion of a filtration generated by the vector
stochastic process that we construct to model the evolution of the state of the system
through time. Let
t
M denote the price of the traded built property market index and
assume that it follows,
( ) = +
t M t M t M
dM M dt M dz t (1)
Where
M
is the expected growth in cash flows (the capital gain component) of the
index, 0
M
> , and
M
z is a standard Brownian motion under P . Thus the market price
of risk for the property index is
M
M
r

= where
M
is the total expected return.
The service flow provided by a specific building, however, is due to a
combination of a local factor and the general property index. The local factor for an
individual building is measured in units of
t
M , which when multiplied by
t
M gives the

9
The type of service flow will depend upon the context of the specific development we model. For
example, in a leased building the service flow will be the net rental flow. In the case of an owner-occupied
commercial building, the service flow is the stream of economic benefits which accrue to the owner from
physically occupying the building. The service flow for owner-occupied residential housing is the implicit
rent on the property.

7

service flow in dollars per unit time. We assume that this local process, denoted by
L
t
P ,
satisfies the following mean-reverting process

1
( ) ( )
P
L L L L
t L L L
t t t t
P P P
dP Pe M P dt P dz t
= + (2)
where , , 0
L L L
P P
P > ,
L
P
z is a standard Brownian motion under P with 0
L
M
P
dz dz = .
Then the service flow process (measured in dollars) is
L
t t t
P M P = and
1
( )
( ) ( ) ( )
( ) ( ) ( )
( ) ( )
P
L L L L
P
L L L L
P
L L
L L L
t t t t t t t
t L L L L
M t t t t t t t t M t t M
P P P
t
t M t M t
P P P
t
t P t P
P
dP d M P dM P MdP
M P dt Pe M M M P dt M P dz t M P dz t
P e P dt Pdz t Pdz t
P e P dt Pdz t
= = +
= + + +

= + +

= +
(3)
where,
L L
M
P P

= ,
L L
L
L
P
P
P
P

,
2 2
L
P M
P
= + , and
2 2
( ) ( )
( )
L L
L
M M
P P
P
M
P
z t z t
z t

+
=
+

is a standard Brownian motion. We assume that the local spot service-flow factor follows
a mean-reverting stochastic process to reflect a real estate market where time to build for
new development creates an inelastic supply of built space in the short-term and service
flows adjust (revert) to a long-run equilibrium over time.
Since there are two sources of risk in the service flow process, we must specify
risk adjustments for both. Under the risk-neutral measure, the discounted price process
for the market index must be a martingale since it is a traded asset. However, the local
service flow process is not the price of a traded asset. Let
*
P be a probability measure
equivalent to P under which

*
( ) ( )
M M
dz t dt dz t = + , (4)
and

8

( )
( )
*
( ) ( )
P P M
L L
L L
L
t r t
P
L P P
t
P
P e e
dz t dt dz t
P

+

= + (5)

Under
*
P ,

( ) ( )
( ) * *
*
( ) ( )
( )
L
P P M
L L L
t r t
M M L P P P
t
P
P
P
dz t dt dz t e e dt
P
dz t

+
= (6)
and

( ) *
1 ( )
P M
L L
L
r t M
t t P t P
P
P
dP P e P dt Pdz t

+

= + +

(7)
If we set
L
P M
P

= + , and
L L
P
P
P
P

= , then

( ) *
( ) ( )
P M
r t
t P t P t P
dP Pe P dt Pdz t

+
= + (8)
We note that
* * M
P M
P
dz dz dt
= .

2.1.2 Completed Unit Values
A completed building unit, whether it is a lot, a house, an apartment, an office, or a retail
facility, will be worth the present value of its expected future service flows over its
economic life,
B
T . The units value, denoted as B , at time t is given by:

*
( , , )
B
T
B r
t t
t
B t T P E e Pd
(9)

where
*
t
E denotes conditional expectation with respect to
*
P
.
We can find the value of
(9) by noting that Fubinis Theorem applies here allowing us to interchange the order of
the integral and expectation. Thus,

9

[ ]
( ) ( )
( )( ) ( )( )
( )
( )( )
*
( )( )
( )( )
( , , )
.
B B
B
B
P
P M
T T
B r r
t t t
t t
r T t
T t
P t P M P P M t P P
P M P P P M P P M P P M
B t T P E e P d e E P d
P P r e P P Pe
r r r r

+

= =

+ +
= + +
+ + + + + +

(10)

Therefore, given a current spot service flow rate
t
P , an economic life of the unit
B
T , and
some parameters from the market implicit rent process, we can fully determine the value
of the unit in closed-form.

2.1.3 The Reservation Service Flow Process
Although market participants are price takers with respect to the spot service flows,
purchasers are heterogeneous with respect to the reservation service flow at which they
are willing to buy a unit from the developer. The distribution of buyers has an average
reservation service flow ( ) R t , and each individual potential buyer/lessor has an
idiosyncratic reservation service flow
( ) R t . When the market service flow rate ( ) P t is

above an individual buyers reservation service flow rate,
( ) R t , it is not economically
viable for the buyer to purchase the unit. The average reservation service flow also
follows a mean-reverting stochastic process to reflect a general economy that is cyclical
in nature but adjusts to a long-run equilibrium over time. As with the market service
flow, we map general economic conditions into a buyer service flow process measured in
units of
t
M , denoted by
L
t
R , which then determines the buyer reservation price for a
particular unit. That is, as economic conditions strengthen,
L
t
R buyer service flows and
reservation prices for the property are increased, and vice versa for weaker economic
conditions. We assume that
L
t
R evolves according to

10

1
( ) ( )
R
L L L L
t L L L
t t t t
R R R
dR R e M R dt R dz t
= + (11)
where , , 0
L L L
R R
R > ,
L
R
z is a standard Brownian motion under P with 0
L
M
R
dz dz = ,
and
L L
L
P R
dz dz dt = . As above, the average service flow process (measured in dollars) is
L
t t t
R M R = and
( ) ( )
R
L L
t
t t R t R
R
dR R e R dt Rdz t

= + (12)
where,
L L
M
R R

= ,
L L
L
L
R
R
R
R

,
2 2
L
R M
R
= + , and
2 2
( ) ( )
( )
L L
L
M M
R R
R
M
R
z t z t
z t

+
=
+

is a standard Brownian motion.
Under the risk-neutral measure,
*
P , we have
( ) ( )
( ) * *
*
( ) ( )
( )
R R M
L L L
L
t r t
M M L R R R
t
R
R
R
dz t dt dz t e e dt
R
dz t

+
= (13)
and

( ) *
1 ( )
R M
L
L
r t M
t L t R t R
R
R
dR R e R dt R dz t

+

= + +

(14)
If we set
L
R M
R

= + , and
L L
R
R
R
R

= , then

( ) *
( ) ( )
R M
r t
t R t R t R
dR Re R dt Rdz t

+
= + (15)
We note that
* * M
R M
R
dz dz dt
= , and that the average market service flow will be positively

correlated with reservation service flows where
2
L L
M L
P R
P R
P R
dz dz dt dt

+
= = . This
reflects increased (decreased) demand for units with higher (lower) reservation service

11

flows. In the long-term, reservation service flow shocks will adjust to equilibrium over
time reflecting general economic cycles.

2.1.4 Buyer Arrival and Departure Processes
One of the key elements of this model is that it incorporates the lumpiness of buyer
availability in the market for finished space. The model assumes that willing and able
buyers arrive in the market stochastically, where the probability of a buyer arriving is a
function of both the current market price and the reservation price of the average buyer in
the market. Similarly, in the pre-construction period, buyers will also depart the market,
i.e. give up their deposits and void their sales contracts, stochastically. Ultimately,
our model requires a function which can calculate the conditional probability of there
being m units available for sale at the end of the period, conditional upon having n units
available at the beginning of the period. This function must incorporate the probability of
potential buyers arriving and departing, the number of units available for sale, and the
construction status of the building. We build this complex function from simpler
functions below.

12

We begin with the notion that individual buyers who are willing and able to
purchase units are those with individual reservation service flows,
t R , greater than or
equal to the current market service flow
t
P . They are assumed to arrive randomly
according to a doubly stochastic Poisson process ( )
A
q t . We model the intensity, ( )
A
t , of
this process at time t as a function of the average reservation flow
t
R , and the market spot
flow
t
P :
( )
A
t
A A
r
P
t
R

=

(16)
The scaling function
A
allows us to simulate the different levels of responsiveness of
the arrival process to market conditions. The constant elasticity demand parameter,
A
,
represents the responsiveness of arrivals to changes in
t
P and
t
R .
The probability of 0 k willing and able potential buyers arriving in a time
interval t s , for 0 s t , is

{ }
( )
( )
*
( )
( ) ( ) |
!
t
A
s
k
t
A
u du
s
A A s
u du
q t q s k e
k
= =

F P . (17)
During the pre-construction period we assume that buyers place deposits on to-be-
constructed units, and that they have the option to void their contracts by giving up their
deposits. We assume that the only circumstances under which this would happen is when
the market service flow level rises above the purchasers individual reservation service
flow level, i.e. when
t
t
P R > . We assume that for a given set of market conditions,
t
P and
t
R , this will occur according to a doubly stochastic Poisson process ( )
D
q t . We model the

13

mean rate of departures for the process at time t also as a function of the average spot
price
t
R and the market spot price
t
P :
( )
D
t
D D
t
P
t
R

=

. (18)
The variable
D
is a scaling variable that allows us to vary the responsiveness of the
departure process to the market conditions.
10
The probability of 0 k buyers departing in
a time interval t s , for 0 s t , is

{ }
( )
( )
*
( )
( ) ( ) |
!
t
D
s
k
t
D
u du
s
D D s
u du
q t q s k e
k
= =

F P . (19)
We assume that once construction starts the buyer is obligated to purchase the
unit at the end of construction. In essence this means that the developer only faces
purchaser departures in the pre-development phase of the project. We note that arrivals
and departures are correlated through equations (16) and (18), however we do not posit
causality between arrivals and departures.
Individually ( )
A
q t and ( )
D
q t may create degenerate results because they do not
incorporate project-specific information which is available. For example, they will
indicate positive probabilities of arrivals when there are no units remaining to be sold, or
positive probability of buyer departures when no units have yet been pre-sold. As a result,
it is more useful to incorporate these functions into a more general function which does
incorporate project-specific information such as the number of units available for sale,
and whether construction has begun on the project. This general function will be denoted

10
As with the arrival process, we assume that this jump process also represents non-systematic risk. We
also suppress the R(t) and P(t) parameters and simply denote
D
(R(t),P(t)) as
D
(t).

14

as ( | ) q m n , and will tell us the probability of having m units remaining to be sold at the
end of the period conditional upon having had n units available at the beginning of the
period.
From a practical standpoint, we actually create two versions of ( | ) q m n , one for
use when the project is in the pre-construction period, and one for use once construction
has begun or has been completed. For ease of exposition we first consider the simpler of
the two versions, that which is used in the post-construction state of the project.
Since all sales after the beginning of construction are final, the model can ignore
buyer departures in this version of ( | ) q m n . This means that by definition, the
( | ) 0 q m n = for m n < . Further, we note that the developer can never sell more than the
N units which constitute the project. Since ( )
A
q t is not limited to N , we truncate its
distribution and include in the case where the aggregate number of arrivals exactly equals
the number of available all of the remaining probability. Given this we can write a
compound function for ( | ) q m n for use in the post-construction state as:

0
0 for
1 ( ) for
( | )
( ) for
0 for .
m
A
i
A
m N
q i m N
q m n
q m n n m N
m n
=
>
<= <
<
(20)

The pre-construction version of

( | ) q m n is complicated by the fact that buyer
departures are allowed. The model must thus calculate the joint probability of a given
number of purchasers arriving and departing, and so we must construct the joint

15

probability density function of the two Poisson processes. We can determine the joint
density function from the individual marginal densities. That is, we can determine the
joint probability that A potential purchasers will arrive and D purchasers will depart,
resulting in a net change of A D purchasers, by first determining the probability of A
arrivals using equation (17) and D departures using equation (19), and then multiplying
those probabilities together to get the joint probability of their occurring together. Adding
together the joint probabilities of all potential arrival and departure levels which result in
a net change of A D purchasers determines the cumulative probability.
We can once again impose some additional structure on the function by appealing
to the practicalities of the development process. Specifically, we note that the aggregate
number of departures, D, cannot exceed the number of units pre-sold at the beginning of
the period, n. Similarly, regardless of how many new potential purchasers arrive the
developer cannot sell more than the N units that will constitute the project. Since the
Poisson probability equations (17) and (19) can generate probabilities for any number of
arrivals or departures, we truncate those distributions to conform to these realities. Thus
we include in the case where the aggregate number of departures exactly equals the initial
number of pre-sales the aggregate probability from all cases where the aggregate number
of departures would be greater than or equal ton. Similarly we include in the case where
the aggregate number of arrivals would exactly equal the total number of remaining
unsold units in the development the aggregated probability from all cases where the
number of arrivals would exceed that capacity. We denote the aggregated probability of
having m units pre-sold at the end of a pre-construction period, conditional on having n

16

sales at the beginning of the period as
*
( | ) q m n . The formula for determining
*
( | ) q m n
is:

( ) ( ) ( )
( ) ( )
( 1) ( )
0 0 0 0
*
|min(0, )| 0
0 for
1 1 1 ( ) for
( | )
( ) 1 ( ) ( ) for
0 for 0
i m n n n m
D A D A
i j i i
n
D A D A
i m n i
m N
q i q j q i q i m N
q m n
q i q i m n q i q m m N
m
+
= = = =
= =
>
+ =

=

+ + <

<

(21)
where
A
q and
D
q are defined in equations (17) and (19), respectively.

2.2. Development Project Cash Flows
Having specified dynamics for the evolution of the market service flow,
t
P , average
reservation service flow
t
R , and buyer arrival and departure, we can begin to discuss the
cash flows that the developer will receive. These cash flows will vary depending upon
where the project is in the development cycle. We can consider the cash-flows in four
distinct regimes: pre-construction, during construction, at construction completion, and
post-construction.

2.2.1 Pre-Construction Cash Flows
We assume that prior to construction the developer faces some minimal level of fixed
costs each period. These costs represent basic ownership costs such as property taxes,
hazard insurance, and basic property-rights enforcement. We assume that these costs are
a function of the size of the property and as a result represent them as being a function of
the number, N , of units that can be built on the property. That is, we actually specify a

17

per-unit cost, ( )
PC
C t
,
growing deterministically at the rate
V
, and then assume that the
total cost to the developer is the product of that cost and the number of potential units
they could build. Thus the costs to the developer of holding the undeveloped land each
period as:

Predevelopment
* ( )
PC
Cost N C t = (22)

2.2.2 Construction Cash Flows
During construction the developer will face a different set of costs. We denote the
construction period as T
C
months. We assume that in each period the developer must pay
1/T
C
of the total construction costs ( ) K t , where K grows deterministically at the rate
K
. The developer must also continue to pay monthly holding costs for the property. The
developers cost in a given construction period is given by:

( )
* ( )
In-development PC
C
K t
Cost N C t
T
= + . (23)

2.2.3 Cash Flows upon Construction Completion
Since development takes T
C
time steps to complete, at T
C
+1 time steps after the
commencement of construction the developer and purchasers close on all pre-sold units.
With that closing, the developer no longer bears any costs for the sold units. The
developer does, however, continue to pay holding costs on any unsold units. Again we
recognize that these per-unit holding costs may differ from those borne in the pre-

18

construction and construction periods, and we denote them as ( )
CC
C t .
11
Thus, the
developers cost in the period when construction is completed is given by:

( ) * ( )
Construction Completion presold CC
Cost N n C t = , (24)
The developers net income for the period in which construction is complete and in which
pre-sold units are closed is given by

( )
* ( ) * ( )
Construction Completion presold presold CC
n
n B t N n C t = , (25)
where B(t) is determined by equation (10).

2.2.4 Post-Construction Cash Flows
In the periods after construction, the developer continues to pay holding costs on any
units that are unsold. If a sale is made the developer receives the value of the unit, ( ) B t ,
as determined by equation (10). Thus, given
spot
n sales in a given period and
closed
n units
closed (including pre-sales) in all previous periods, the developers net-revenues are
given by:

( )
* ( ) * ( )
Post development spot closed CC
n
n B t N n C t

=

. (26)
We now have in place all of the pieces needed to model the developers options,
and ultimately to determine the value of the undeveloped land to the developer. In the
next section we develop the formal mathematics of the model.

2.3. The Stochastic Control Model
The problem facing the developer is choosing the optimal construction start time; once
started we assume that construction must continue for a deterministic length of time. Due

11
One can easily justify this by noting that property taxes rates and insurance rates may differ for vacant
land, property that is pre-completion, and completed properties.

19

to this feature, the problem is not one of classical stochastic control since the optimal
constrained control would not be Markovian (once turned on, it cannot be turned off for a
deterministic length of time, even if it is optimal to do so). The problem can, however, be
formulated as an optimal switching problem as in Brekke and ksendal (1994), which is
in the class of optimal stopping problems.
To do this we augment the state process to include a state that is 0 if construction
has not yet started and 1 if construction has started (including any time during
construction and after completion). Since the construction time and costs (given the start
time) are deterministic, the present value of construction costs at the start time is the
switching cost, i.e. the cost of using the impulse control. Once formulated as a stochastic
impulse control problem, it can be solved (with some minor modifications) using the
methods of stochastic impulse control for jump diffusions as detailed in ksendal and
Sulem (2005). We present a formal derivation of the stochastic control model in
Appendix A.
Unfortunately there is not a closed-form solution due to the highly path-dependent
nature of the problem. As a result, we use a numerical method to determine the solution,
which we discuss in the following section. In Appendix B we show that the numerical
method provides a good discrete approximation to the value function.

2.4 The Numerical Method
Our numerical method must model the evolution of
t
P and
t
R through time. We chose to
do this through a bivariate-binomial lattice using the method of Nelson and Ramaswamy
(1990). We provide full details of the lattice construction in Appendix B. We use this
lattice to solve the valuation equation subject to a set of terminal and interior boundary

20

conditions. We discuss those conditions, as well as various interior boundary conditions
in the sub-sections that follow.
Given the stochastic control model, we can now write the valuation equations
used within the lattice that satisfies those conditions. These equations also allow us to
find the free-boundaries within the lattice of the various options held by the developer. In
some sense the free-boundary that we are most interested in is the one associated with the
developers decision to begin construction. The existence of this boundary illustrates a
path-dependency problem for the model.
Since the model works backwards in time, at an arbitrary point in time t we
cannot know for certain whether the developer has already exercised their option to begin
construction.
12
This means that we cannot know whether the project is in its pre-
construction period, during one of the
C
T months of its construction, at the completion
of its construction, or in post-construction. Thus, there are a total of 4
C
T +

potential
states that the project could be in at time

t.
Fortunately, there is a numerical technique we can use to solve this path-
dependency problem. The key is to recognize that the uncertainty with respect to the
building state and entitlement state at time t resolves as the model moves backwards in
time. For example, at time t -1 we know whether the builder would elect to begin
construction at that time period. Once we know this we also know whether it is possible
for the project to be in its first month of construction at time t, since if construction does
not begin at

t -1, it cannot be one month along at time t. In fact, all uncertainty regarding
the building state resolves in a similar manner as the model moves backwards in time.

12
We will demonstrate shortly that for t near the development horizon T
H
, we sometimes do know whether
the developer has exercised that option.

21

We can exploit this resolution of uncertainty to solve the path-dependency
problem. We do this by artificially expanding the state space to include new temporary
state variables: construction status. That is, at an arbitrary point in time t we determine
what the value of the project would be for each of the possible 4
C
T + construction states
in which the project could be at that time. As the model moves further back in time the
uncertainty regarding the time t construction state resolves and we can begin to discard
some of these temporary state variables. Ultimately we will reach a point, potentially the
initial point of the lattice, in which all building-state uncertainty within the lattice is
resolved, and we retain only the correct time t state variable.
13

We note that at the beginning and end of the model the path-dependency problem
is somewhat lessened. For example, at time 0 we know with certainty that the project is in
a pre-construction state. At time step 1, there are only two potential states of the project:
it is still in pre-construction, or it is in the first month of construction. Only at time step
2
C
T + and beyond are all 4
C
T + construction states possible. Similarly, at time
H
T , the
end of the economic horizon of the developer (and the end of the model), and only three
construction states are possible: the project could have been completed at some point in
the past, it could have been completed during this time period, and it could still be in a
pre-construction status. If the project is tagged as being in pre-construction this means
that the developer has previously abandoned the project. The model does not allow the
developer to begin construction on the project if completion would occur after the

13
We note that the problem is actually more complex because there is also uncertainty regarding how many
units remain to be sold (or pre-sold.) That is, for each of the T
C
+4 building states, there are N+1 potential
sales states of the project. These states include the situation where there are 0 units remaining, 1 unit
remaining, 2 units remaining, etc. Thus, at most nodes in the lattice there are actually (N+1)(T
C
+4)
conditions (and hence temporary state variables) to be considered. Although we suppress the sales-states in
this discussion for expositional clarity, the model does fully consider these conditions.

22

economic horizon of the developer,
H
T . Thus, at time
H C
T T the developer is forced to
make a decision, they must either begin construction or abandon the project.
In the sections that follow we develop the valuation equations that apply at each
of the potential states of the project. We subsequently then combine those valuation rules
to develop the decision rules that the developer uses to determine the projects value.

2.4.2 Post-Construction Interior Boundary Conditions
We begin our discussion of interior boundary conditions by introducing some additional
notation. The value of the project at any node in the lattice (i, j) at time step t will depend
upon the number of units which have already been sold, , and we denote this value as
( )
, , t i j
V .
14
We also need to reference the time t expected value of the unit at time t +1
under the risk-neutral probability measure. This value is also conditional upon the
number of units sold after time t +1, and we denote it as
( ) , , t i j V . Note that the t in the

subscript indicates that this is the discounted expected value at time t of the project at
time t +1. The boundary condition given in equation (23) requires that
( ) , ,
H
T i j V =0 for
all values of n since
H
T is the terminal point of the lattice.
In any post-construction period, the actions available to the developer will depend
upon the number of unsold units still available. If all units have been sold, the remaining
value of the project V
t,i,j
will be zero, since there are neither costs to be born nor
additional revenues to be earned. Should there be units available, however, the developer
must decide whether they are willing to sell those units this period should potential
purchasers arrive. If the developer sells those units today they give up the discounted

14
When referencing specific points in our lattice, we must use a notation with three coordinates. This is
because the lattice contains three primary state variables: P, R and time.

23

expected present value (under the risk-neutral probability measure) of the project at the
next time step. The developer also has the option to abandon any unsold units. Again, the
developer must weigh the benefit of doing this again the cost of forgoing any future sales
of those units. If the developer does abandon, however, they gain the benefit of not
paying holding costs. This is given by:
( )

( )
( )
, , , 1
*
, , , 1
0 , , 1
,
( | ) *max
, 0
C
C
C
post development
n
t i j T
m
t i j T
post development
m t i j T
n
V m
V n q m n
V n
= +
= +

= = +

+

=

+

(27)
Note that this equation determines the value of the project across all potential
number of unit sales and allows the developer to determine exactly how many units they
are willing to sell. These different values are weighted by the probability of the different
numbers of purchasers (m-n) arriving in that period.

2.4.3 Construction Completion Boundary Conditions
The builder faces a very similar decision if construction is completed in the current time
step. They will receive, with certainty, revenues for units that have been pre-sold. They
will also receive revenues for any units that they sell in the current period. They have to
weigh selling those units in the current period against the possible benefits of delaying
sales, i.e. the expected value of those future sales under the risk-neutral measure.
Similarly, the developer could elect to abandon the unsold units at time t. Thus, there are
three options for the developer to consider: selling units in this period, refusing to sell
units in this period but continuing to operate the project, and abandoning the project
entirely. The developer will elect the option that maximizes their wealth. We can
incorporate all three options into a single equation as:

24

( )

( )
( )
, , , 1
*
, , ,

0 , , , 1
,
( | ) *max
, 0
C
C
C
Construction Complete
n
t i j T
m
t i j T
m t i j T
n
V m
V n q m n
V n
= +
=
= = +

+

=

+

(28)
The three terms within the max function correspond to the sell, no-sell, and abandon
options. Also, recall from equation (20) that the net income function at completion
(
) requires two inputs the number of units pre-sold and the number sold
in the current period. Since the total number of units in the development are N, and if
there are n remaining at the beginning of the time step, this implies that (N-n) are pre-
sold, and so the m that are sold during the period are spot market sales.

2.4.4 During Construction Boundary Conditions
During the construction period the developer has the option of selling units if potential
purchasers arrive.
15
Once again, the developer must weigh the value of pre-selling
immediately versus the value that they give up of being able to sell in the future. The
developer also continues to have the option to abandon the project. If they abandon, they
not only avoid future holding costs, but they also avoid the remaining construction costs
that they have yet to bear. The developer will once again take the actions, conditional on
the number of arrivals of potential purchasers that maximizes their position. We represent
this through the following equation.
( )

( )
( )
, , , 1
*
, , ,
0 , , , 1
,
( | ) *max
, 0
In Development
n
t i j k
m
t i j k
In Development
m t i j k
n
V m
V n q m n
V n
= +
=

= = +

+

=

+

(29)
Again, the three terms in the max function correspond to the three options held by the
developer. It is worth noting that the
( ) , , , 1 t i j k V n = + and
( ) , , , 1 t i j k V m = + values referred to in
equation (34) are based upon the assumption that at time t+1 the construction is one-

15
Technically these are pre-sales since they will not close until construction is completed.

25

period further along, so 0
C
k T < < .
16
If we are in the last time period of construction,
these expected values will be based upon the completed construction cash flows, i.e.
equation (33), at the next time step.

2.4.5 Pre-Construction Boundary Conditions
During the pre-construction period, the developer holds multiple options. These options
interact to create a number of unique boundaries. Specifically we note that the developer
has five specific actions that they can take:
1. They can delay construction but pre-sell to arriving potential buyers;
2. They can delay construction and refuse to pre-sell to arriving potential buyers;
3. They can begin construction and pre-sell to arriving potential buyers;
4. They can begin construction and refuse to pre-sell to arriving potential buyers;
5. They can abandon the project.

If the developer delays construction, then the expected present values used in the
valuation equations are conditional upon the assumption that the next period will also be
a pre-development period. If the developer begins construction, then the expected present
values are conditional upon the assumption that the next period will be month 1 of the
construction process, and that the project will be complete in T
C
months. The developer
will choose the option that maximizes their position. We can express this in the following
equation:

16
An implication of this is that we do not allow the developer to halt development and the recommence it.
Once development has started the developer can only continue to develop or abandon the project.

26

( )
( )
( ) ( )
( ) ( )
( ) ( )
( ) ( )
0
0
, ,
, ,
, ,
, ,
, ,
| delay ,
| delay ,
| *max | start ,
| start ,
0
t i j
Pre Development
m t i j
Pre Development
n t i j
n
Pre Development
m t i j
m
Pre Development
t i j
V n
V m
V n
q m n V m
V n
=
+
+
+
+
(30)
Note that equation (30) allows us to determine the value of the land. To see this,
consider that every other equation ultimately feeds into this equation. At time step 0, with
0 units pre-sold (i.e. with N units yet to be constructed and sold) this is the most that the
rational developer would pay for the land in order to begin the development process.

2.5 Finding Equilibrium Values
We can also solve the system numerically to determine the markets equilibrium service
flow value, which in turn is used to compute the market price per completed unit. We
derive the equilibrium service flow as the level where land developers are willing to
presell/prelease space preceding their ultimate development decision. Although it is
assumed that there is finite price elasticity for the real estate market at the aggregate
level, it is also assumed that individual land parcels are perfect substitutes, so that prices
above the markets equilibrium of
0
L
L
P P = are not obtainable by individual properties
due to competing land supply.

27

The equilibrium service flow value is determined by finding the
0
L
L
P P = value,
for properties with no endowed pre-sales as of time 0, where the value of the land is
maximized
17

0
0 , , ,
0
0,
0
L
L
t i j
L
n P P
V
P
=
= =
=
(31)
Properties without presales maximize land value at the lowest
0
L
L
P P = relative to
land value with any number of presales. Properties with presales will not be allowed to
raise prices beyond this point since land owners without presales can maximize their land
value and presell units at a lower price; thus, these properties are the price setters in the
market and this is where the market price will settle.

2.6 Expected Returns and Time to Development
As shown in Appendix B, it is possible to recover the risky transition probabilities at each
node in the lattice. This allows us to examine both the price of the property, which is
governed by the risk-neutral probabilities, as well as the actual expected returns and time
until development to the property, which are governed by the risky probabilities.
To compute these, we first solve for the price of the land working backwards in
time. As we do so, however, we store the cash flows and actions of the developer under
the risk-neutral measure for every state at every node.
18
After determining the value of

17
As
0
L
L
P P = is varied there are two opposing effects: for increases (decreases) in
0
L
L
P P = , the market
value of land increases (decreases) because of the increased (decreased) value of the competed unit; while,
at the same time increases (decreases) in
0
L
L
P P = reduce (increase) the number of arrivals due to finite
supply and demand elasticity for the market as a whole. Given these competing effects, there is a
0
L
L
P P =
value for properties with no presales where land value is maximized.
18
This does require a tremendous amount of memory even with 8 gigabytes of memory, at best we can
only value a 16 unit building with at most a 36 month construction period when using 4 time steps per
month.

28

the project using the risk-neutral probabilities, we work our way forward through the
lattice using the risky-probability measure. At each node we examine the cash flows and
actions taken by the developer. We then use the risky probabilities to determine the actual
expected cash flow from the project. From this, we can calculate the expected return as
the IRR of the project, given the price determined by the risk-neutral valuation. Similarly,
by noting where in the lattice the developer exercises their development option, we can
then use the risky probabilities to determine the expected time until development in the
risky world.
By calculating price, expected return, and time to development, we are able to
demonstrate the degree to which the model is able to replicate behaviors seen in the real
world. In the next section we present the basic results of the model.

3. Results from the Model Simulations and Analysis
3.1 Base Case Parameters
Using a set of parameter value inputs and the specified numerical procedures described,
we solve the model and compute a number of output values: equilibrium local service
flow
0
L
L
P P = (which then maps into the equilibrium completed building value using
equation (10)), the value of the speculative land (land without presales), the expected
time until speculative land will begin construction, the expected return (IRR) on
speculative land until construction commences, the value of land ripe for development
(i.e., the land with the presales necessary to justify immediate commencement of
construction), land value as a percentage of the competed project, the number of presales
needed to promote construction commencement (which then maps into the equilibrium
level of absorption of units per year and the expected years to fully absorb the remaining

29

space).
Table 1a presents the parameters for our base-case scenario. First we specify the
initial market price of the general property market index
0
M as the numeraire and set its
total return, drift rate (capital gain component) and volatility to be 10%, 3.5% and 12.5%
respectively.
19
The parameters that relate to the local factors of the service flow and the
reservation service flow are those that populate the stochastic processes for service
L
t
P
and
L
t
R .
0
L
L
R R = is initially set to be the numeraire, while the equilibrium
0
L
L
P P = is
found endogenously. Since we would expect the market and reservation service flow to
behave similarly over the long term, we set both processes drift terms, reversion speeds,
and volatilities of to be 3.5%, 12.5% and 15% respectively. We set the correlation
coefficient
L
between the local the market and reservation service flow processes to be
0.25. The risk-free rate (r) is set to 5%.
Table 1a:
Base Case Model Parameters with respect to the general economic conditions
Unless otherwise noted, parameters are quoted in an annualized format.
Parameter Base Case Value
M is the expected total return on the market portfolio M = 0.10
M is the capital gain component of the market portfolios return M = 0.035
M is the constant volatility rate on the market portfolio return M = 0.125
P is the drift of the service flow

P =.035
p is the speed of reversion of the service flow p =.15
p is the constant volatility rate of the service flow p = .125
R is the drift of reservation service flow R = .035
R is the speed of reversion to the reservation service flow R = .15
R is the constant volatility rate of the reservation service flow R =.125
L
Correlation between the service flow price and the reservation service flow
L
= .25
0
L
L
R R = is the initial reservation service flow value
0
1
L
L
R R = =
r is the risk-free rate r =.05

19
The level of these inputs is based on common real estate return indices such as the National Council of
Real Estate Investment Fiduciaries the National Association of Real Estate investment Trusts, among
others. Also see The Benefits of Real Estate (2006) authored by the Center for International Securities and
Derivatives Markets.

30

In Table 1b, we specify the parameters for the particular development project. The
completed development project will be a total of N=16 units.
20
The developer is endowed
with pre-sold units at time 0 anywhere from n=0 up to n=16. For the buyer arrival
function, we specify the base case scalar of arrivals
A
to be 7. The elasticity of arrivals

for new unit buyers
A
is set to be 6.
21
For the departure function, we specify the base
case scalar of departures
D
to be 1 and the elasticity of departures

D
to be 3. The
elasticity of departure is set to one-half of arrivals to reflect a level of hysteresis for
exiting a contract to purchase a unit. With these parameters, when
t
P and
t
R are equal the
average number of arrivals of buyers to the project per year is 7 and the number of
departures is 1.
22

The initial annual costs on a per unit basis of holding the land until construction is
complete is $0.0425
V
C = . This represents approximately 3% per year of preconstruction
speculative land value (land with zero presales). The annual cost of holding an unsold
unit after construction is complete is $0.50
V
C = . This again represents approximately
3% per year of the value of the unsold but completed unit. Both of these costs will set to
increase at a rate of 3% per year. Also shown in Table 1b are the base-case total
construction costs of $12.50 per unit and the time to complete construction of 18 =
C
T

20
We choose 16 due to computing capacity.
21
Although the elasticity may seem high, it has been found that elasticities for new residential construction
are historically between 4 and 13. See Malpezzi and Maclennan (2001) and Green, Malpezzi and May
(2005).
22
See equation (16) and equation (18) where the arrival and departure Poisson process are defined and
explained. With a 16 unit project, if the P/R ratio is one, the base case parameters provide for an expected
absorption for the project of between 2 and 3 years (16 units/7 net arrivals per year), which is realistic in
todays market for many residential and commercial projects.

31

months. Construction costs are assumed to increase at 3% per year. Once a unit is
constructed it is assumed that it has a useful life of 50 =
B
T years.
Table 1b:
Base case model parameters with respect to the development project
Unless otherwise noted, parameters are quoted in an annualized format.
Parameter Base Case Value
N, number of lots/units in the development project N = 16
n, number of presold units endowed at time 0 n = 0, 116
A
is the arrival scalar
A
= 7
D
is the departure scalar
D
= 1
A
is the arrival elasticity
A
= 6

D
is the departure elasticity
D
= 3
C
T time in months to complete construction once development begins
18
B
T is the useful life of a completed building
50
K is the total construction cost for unit $12.50
K
is the growth rate for construction costs over time
K
=.03
is the initial variable cost per unit upon construction completion =$0.50
is the initial variable cost per unit before construction completion =$0.0425
V
is the drift rate for variable costs over time
V
=.03

3.2 Numerical Results and Comparative Statics
Given the inputs shown in Tables (1a) and (1b), we run the model to determine the
outputs. In Table 2 we show the results for the base-case parameters (shown in bold)
along with comparative statics for varying the elasticity of arrival parameter
A
. We find
that speculative land value is maximized at a local service flow price of
0
1.180
L
L
P P = =

which maps into an equilibrium market price of each of the completed building units of
$17.45. The value of speculative land, on which 16 total building units could be
constructed, is $11.43 and it is expected to begin construction in 66 months. The
expected return (IRR) on an investment in speculative land until the uncertain time of
construction is computed to be 18.17%. Land that is ripe for development, i.e., with
presales of 38% of the units, land value has increased to $35.07, representing 12.56% of
(0)
v
C
v
C
(0)
v
C
v
C

32

the completed project value, and expected absorption of the remaining units will occur
over the next 3.86 years (this includes any absorption during the 18 month construction
period). All of these results are in the range of reasonableness based on a number of
empirical observations and sources.
23

Table 2 also reports results where the arrival price elasticity,
A
in the demand
function Equation (16), is varied from the base case. The elasticity of arrival parameter is
instrumental not only in determining the arrival volatilities of the demand function, but
also in finding the equilibrium local service flow
0
L
L
P P = , which is endogenously
determined using the speculative land supply optimization equation (31). Where the price
elasticity is less than the base case, equilibrium service flows, completed unit prices and
land values all increase. Decreased elasticity affects the equilibrium service flows
through equation (31) in the sense that it reflects markets that are less competitive due to
low land supply elasticity, e.g., in markets where land is constrained by public policy
and/or geographic reasons. In these less competitive land markets, demand is less
affected by economic changes, completed project and land values are higher,
development on available land occurs sooner, land values are a larger percentage of
completed project value, and expected returns from land investment are lower due to
decreased market risk from less demand/absorption uncertainty. These results reverse for
more competitive markets where land development is less constrained and supply
elasticities are higher.

23
See Davis and Heathcote (2007), Davis and Palumbo (2008), Korpacz Real Estate Investor Survey
(2
nd

quarter, 2008) and Zelman, McGill, Speer, and Ratner, 2006.

33

Table 2
Summary of Model Outputs as a Function of Arrival Elasticities
Column A B C D E F G H I J
Arrival
Elasticity
AP AR
=
Equilibrium
Service
Flow
0
L
L
P P =
Current
Value of a
Completed
Building
Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
2
2.269 33.55 115.61 9 7.05 197.66 36.82%
25% 1.36 8.83
3

1.524
22.54
45.40 11 9.65 76.91
21.33%
19% 1.98 6.57
4

1.299
19.21
24.14 29 13.56 50.16
16.32%
25% 2.46 4.88
5

1.217
18.00
15.59 50 16.36 39.75
13.80%
31% 2.62 4.20
6

1.180
17.45
11.43 66 18.17 35.07
12.56%
38% 2.59 3.86
7

1.160
17.15
9.13 78 19.34 32.96
12.01%
44% 2.48 3.63
9

1.150
17.01
7.77 86 20.03 32.48
11.94%
44% 2.29 3.93
9

1.142
16.89
7.03 93 20.52 33.08
12.24%
50% 2.12 3.78
10

1.140
16.86
6.89 98 20.65 28.68
10.63%
56% 1.89 3.71
11

1.134
16.77
5.95 100 20.95 31.59
11.77%
63% 1.76 3.49
12

1.132
16.74
5.70 103 21.01 33.65
12.56%
69% 1.58 3.16
Base Case outputs are shown horizontally in bold. Note that departure elasticities are set at 50% of arrival elasticities.
Columns A and B: The equilibrium local service flow maps into the equilibrium completed building value using equation (15)).
Column C: The value of speculative land represents the value of land without presales.
Columns D and E: The expected time to development is computed as the expected number of months before construction will begin on speculative land.
Expected return (IRR) is computed based on the expected cash flows from an investment in speculative land from time zero until construction commences.
Column F and G: The value of land ripe for development represents land with the presales necessary to justify commencement of construction. Land value that
is ripe for development is shown as a percentage of the completed project value.
Columns H, I and J: The percentage of units to be presold is the amount needed to commence construction, which then maps into equilibrium level of
absorption of units per year and the expected years to fully absorb the remaining space post construction.

34

Table (2) indicates that most results are extremely sensitive to changes in
elasticity with exception that changes in the equilibrium completed unit prices tend to
smooth absorption levels across different levels of elasticity. An exception occurs for
markets with very low elasticities, where very high prices are available allowing for
tolerance of slower absorption levels.
Risk and expected returns are affected in two major ways as elasticity is varied.
First, with lower (higher) elasticity, demand is less (more) sensitive to changes in
reservation and price service flow levels, mitigating (exacerbating) demand/absorption
volatility and risk. Secondly, because land is an option to develop completed units by
paying the construction costs (the exercise price), changes in the equilibrium completed
unit values directly affect the level of leverage inherent in the land option, i.e., the extent
the option (land) is in- or out-of-the -money. In less (more) competitive markets with
lower (higher) demand and supply elasticities, the construction costs represent a smaller
(larger) proportion of the completed project value and the inherent option leverage and
risk is decreased (increased). These two sources of risk will be a common theme
throughout the presentation of the results.
The sensitivity of many of the outputs to elasticity in Table (2) shows the
importance of determining demand and supply elasticity parameters more accurately
through additional econometric studies and/or the through the calibration of other more
transparent parameters within the model so as to imply the markets elasticity.
We now examine the effects of increases in the duration and cost of construction.
Panel A of Table 3 shows that increasing construction costs increase the necessary
equilibrium market price of completed units and at the same time decrease land values

35

quite significantly. The percentage effect of construction cost on land values is leveraged
upward since land values typically represent only 12-25% of the completed development
project value. Additionally, increases in construction cost extend development timing risk
and increase inherent option leverage causing expected speculative land returns to
increase. This increase in risk leads to an increase in the level of presales needed to begin
construction.
Panel B of Table 3 shows similar results when construction time is varied. As
construction time increases, it indirectly increases the overall net cost of the project, as
completed presold units cannot be sold and revenues collected until the project is
completed. As the cost of this delay increases, the necessary market price of completed
units increases. Unlike the construction cost results where the leverage of the option is
affected by the direct increase in construction costs (exercise price), expected returns
decrease with increased construction time since exercise price is constant and completed
unit market prices increase. All of the results from Table (3) show that increased costs in
construction are partially passed forward to the price of completed units and partially
passed backward to decrease land values.

36

Table 3
Summary of Model Outputs as a Function of Constriction Costs and Construction Time
Panel A: Construction Costs
Construction
Costs

Equilibrium
Service
Flow
0
L
L
P P =
Current Value
of a Completed
Building Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
180 1.113 16.46 24.42 21 13.28 42.72 16.22% 25% 3.68 3.26
200

1.180 17.45 11.43 66 18.17 35.07 12.56% 38% 2.59 3.86
220 1.286 19.02 3.82 123 22.47 38.78 12.74% 56% 1.55 4.52
Panel B: Construction Time
Construction
Time

Equilibrium
Service
Flow
0
L
L
P P =
Current Value
of a Completed
Building Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
12 1.163 17.20 13.43 64 17.94 37.25 13.54% 38% 2.83 1.163
18

1.180 17.45 11.43 66 18.17 35.07 12.56% 38% 2.59 1.180
24

1.201 17.76 9.31 72 18.44 32.42 11.41% 38% 2.33 1.201
Base Case outputs are shown horizontally in bold.
Columns D and E: The expected time to development is computed as the expected number of months before construction will begin on speculative land. Expected return
(IRR) is computed based on the expected cash flows from an investment in speculative land from time zero until construction commences.
Column F and G: The value of land ripe for development represents land with the presales necessary to justify commencement of construction. Land value that is ripe for
development is shown as a percentage of the completed project value.
Columns H, I and J: The percentage of units to be presold is the amount needed to commence construction, which then maps into equilibrium level of absorption of units per
year and the expected years to fully absorb the remaining space post construction.

37

Table (4) shows comparative statics for changes in the local reservation service
flow. Panel A of Table (4) assumes that the initial reservation service flow and the long-
run reservation service flow are equal, i.e.,
0
L
L
R R = .

For cases where
0
L
L
R R = is less
(more) the base case,
0
1
L
L
R R = = , this reflects an economy that permanently has less
(greater) demand strength relative to the base case, because cash flow/service flow
continually and permanently reverts to the decreased (increased) the long-run
equilibrium. Not surprisingly, as
0
L
L
R R = increases, completed unit prices and land
values increase due to the stronger economic demand. Expected time to development also
decreases with a stronger economy, reducing speculative land development timing risk
and expected returns.
Panel B of Table (4) shows the results from varying the initial reservation service
flow, but it keeps the long-run reservation service flow at the base case value of 1
L
R = .
For cases where
0
L
R

is less (more) the base case longrun value, this reflects an economy
that has less (greater) demand strength relative to the base case, but this is temporary
(cyclical) because service flows will revert back to the long-run equilibrium. As
0
L
R
increases, equilibrium completed unit prices stay the same, since the economy is only
temporarily out of equilibrium, and land values increase with a stronger economy, albeit
not as much as if the strength were permanent. Expected time to development also
decreases sharply as developers construct while the economy is strong. With a stronger
economy, time to development and the risk of this timing decreases and this is shown is
lower expected returns.

38

We now turn our attention to varying the volatility of the local and reservation
service flow parameters and the correlation between them. Panel A of Table (5) shows
that as volatility increases, values for completed units increase as the option to wait for
higher prices before development increases; thus, expected times to development also
increase. Because the option to wait is embedded in land values, these also increase with
volatility. Increased volatility translates into higher presales needed to commence
development to mitigate an increase in absorption risk. Interestingly, expected returns on
speculative land decreases with increases in volatility. This effect is due to the decrease
in leverage in the land option from the higher completed project values needed to induce
construction.
Panel B of Table (5) shows that as correlation increases, the volatility of the
service flow ratio decreases and the option to wait to develop land decreases. This causes
development to occur sooner and at lower prices. The leverage effect from decreased
completed unit values increases expected returns as correlation increases. It is interesting
to note that speculative land values are not greatly affected across the range of
correlations and volatilities shown due to offsetting effects of leverage and volatility of
the land option.

39

Table 4
Summary of Model Outputs as a Function of
0
L
L
R R =
Panel A: Permanent Economic Adjustment
0
L
L
R R =
Equilibrium
Service
Flow
0
L
L
P P =
Current
Value of a
Completed
Building
Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
0.90 1.169 17.29 2.65 131 23.17 38.40 13.88% 63% 1.46 4.12
0.95 1.170 17.30 6.20 98 20.73 30.79 11.12% 44% 2.01 4.49
1.00 1.180 17.45 11.43 66 18.17 35.07 12.56% 38% 2.59 3.86
1.05 1.199 17.73 18.16 39 15.73 40.13 14.14% 31% 3.16 3.48
1.10 1.230 18.19 26.08 21 13.50 46.12 15.85% 31% 3.58 3.07
1.15 1.271 18.80 34.87 11 11.62 56.49 18.78% 25% 3.84 3.12
1.20 1.320 19.52 44.25 7 10.22 67.32 21.56%
25%
3.95 3.04
1.25 1.372 20.29 53.93 5 9.22 78.37 24.14%
25%
4.00 3.00
Panel B: Cyclical Economic Adjustment
Adjust
0
L
R
1
L
R =
Equilibrium
Service
Flow
0
L
L
P P =
Current
Value of a
Completed
Building
Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
0.75 1.18 17.45 0.17 166 26.51 29.64 10.62% 63%
2.59 3.86
0.80 1.18 17.45 0.96 149 24.81 34.28 12.28% 63%
2.59 3.86
0.85 1.18 17.45 2.36 132 23.06 33.20 11.89% 56%
2.59 3.86
0.90 1.18 17.45 4.53 112 21.31 38.09 13.64% 56%
2.59 3.86
0.95 1.18 17.45 7.55 90 19.68 33.67 12.06% 44% 2.59 3.86
1.00 1.18 17.45 11.43 66 18.17 35.07 12.56% 38% 2.59 3.86
1.05 1.18 17.45 16.06 43 16.72 37.04 13.27% 31%
2.59 3.86
1.10 1.18 17.45 21.15 23 15.18 39.40 14.11% 25%
2.59 3.86
1.15 1.18 17.45 26.24 8 13.06 41.93 0.00% 19%
2.59 3.86
1.20 1.18 17.45 30.84 0 Immediate 39.73 14.23% 0%
2.59 3.86
1.25 1.18 17.45 34.58 0 Construction 45.08 16.15% 0%
2.59 3.86
Column F and G: The value of land ripe for development represents land with the presales necessary to justify commencement of construction. Land value that is
ripe for development is shown as a percentage of the completed project value.
Columns H, I and J: The percentage of units to be presold is the amount needed to commence construction, which then maps into equilibrium level of absorption
of units per year and the expected years to fully absorb the remaining space post construction.

40

Table 5
Summary of Model Outputs as a Function of Volatility and Correlation
Panel A: Volatility in Local Service Flows
L L
P R
=
Equilibrium
Service
Flow
0
L
L
P P =
Current
Value of a
Completed
Building
Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
.10 1.140 16.86 11.04 47 20.08 30.31 11.24% 31% 3.19 3.45
.125 1.180 17.45 11.43 66 18.17 35.07 12.56% 38% 2.59 3.86
.15 1.234 18.25 12.45 81 16.18 41.93 14.36% 44% 1.98 4.54
Panel B: Correlation in Local Service Flows

L

Equilibrium
Service
Flow
0
L
L
P P =
Current
Value of a
Completed
Building
Unit
Value of
Speculative
Land
Expected
Time to
Development
(in months)
Expected
Return
(IRR)
Value of
Land that is
Ripe for
Development
Land as a
Percentage
of
Completed
Value
Percentage
Units to be
Presold
Before
Construction
Unit
Absorption
Rate Post
Construction
Expected
Absorption
(in years)
0.00 1.222 18.07 12.12 78 16.62 41.41 14.32% 44% 2.10 4.28
0.25 1.180 17.45 11.43 66 18.17 35.07 12.56% 38% 2.59 3.86
0.50 1.143 16.90 11.13 49 19.83 30.35 11.22% 31% 3.14 3.50
Column F and G: The value of land ripe for development represents land with the presales necessary to justify commencement of construction. Land value that
is ripe for development is shown as a percentage of the completed project value.
Columns H, I and J: The percentage of units to be presold is the amount needed to commence construction, which then maps into equilibrium level of
absorption of units per year and the expected years to fully absorb the remaining space post construction.

41

As the parameters of the model are varied, we find that expected returns on
speculative land investment generally range from 12% to 25%, and the level of return is
directly related to the degree of leverage inherent in the land option, the
elasticity/volatility of demand and the expected time to development. Absorption times to
sell or lease any remaining units that are not presold run 3 to 4 years on average,
including the construction time. Absorption levels and times are not drastically affected
across different parameter sets as completed project prices adjust in equilibrium to
mitigate changes in absorption. Land values range from 12% to 25% of completed project
value. Finally, presales needed to commence construction are 33% to 60% and increase
with the level of risk inherent in the land. These results are all in line with empirical
results and investor surveys.

4. Conclusion
We have developed a model that generalizes and extends the standard real options
framework by combining project cash flow uncertainty with a general economic variable
that determines tenants/buyers demand for space through an aggregate reservation price.
This economic uncertainty variable reflects the actualities of uncertain absorption and
lumpy demand for space over time, and avoids the assumption that tenants are
continuously present to lease (buy) space. Tenants arrive and depart according to a
Poisson process with a stochastic intensity that is a function of both the current market
lease price and the aggregate reservation price.
We formulate the model as an optimal switching problem from stochastic impulse
control where switching corresponds to starting construction, and succeed in
implementing a numerical scheme to solve for the value function and the optimal time to

42

begin construction. We are also able to determine the equilibrium land value including
the developers embedded options, and the expected return to the development project,
among other results. Results from the model with respect to land and completed project
values, absorption rates, presales needed to commence construction, and expected returns
on land investment are generally in line with empirical results and investor surveys.

43

References
Ambrose, Brent W., Richard J. Buttimer, and Charles A. Capone, 1997. Pricing
Mortgage Default and Foreclosure Delay. Journal of Money, Credit and Banking, 29(3):
314-323.

Brekke, K.A. and B ksendal, 1994. Optimal Switching in an Economic Activity under
Uncertainty. SIAM Journal on Control and Optimization, 32(4): 1021 1036.

Buttimer, Richard J. and Steven H. Ott, 2007. Commercial Real Estate Valuation,
Development, and Occupancy Under Leasing Uncertainty, Real Estate Economics, 35
(1), 21-56.

Capozza, Dennis and Robert W. Helsley 1990. The Stochastic City, Journal of Urban
Economics 28, 187-203.

Capozza, Dennis and Y. Li, 1994. The Timing and Intensity of Investment: The Case of
Land, American Economic Review, 84, 889-904.

Capozza, Dennis and Gordon Sick, 1991. Valuing Long Term Leases: The Option to
Redevelop. Journal of Real Estate Finance and Economics, 4, 209-223.

Childs, Paul D., Steven H. Ott and Timothy J. Riddiough, 2002. Optimal Valuation of
Noisy Real Assets, Real Estate Economics, 30 (3), 385-414.

Childs, Paul D., Timothy J. Riddiough and Alexander J. Triantis, 1996, Mixed Uses and
the Redevelopment Option, Real Estate Economics 24, 317-339.

Cunningham, Christopher R., House Price Uncertainty: Timing of Development, and
Vacant Land Prices: Evidence for Real Options in Seattle. Journal of Urban Economics..
59. 1-31.

Davis, Morris A. and Michael G. Palumbo, 2008. The Price of Residential Land in Large
US Cities, Journal of Urban Economics 63, 352-384.

Davis, Morris A. and Jonathan Heathcote, 2007. The Price and Quantity of Residential
Land in the United States, Journal of Monetary Economics 54, 2595-2620.

Green, Richard K., Stephen Malpezzi, and Stephen K. Mayo, (2005). Metropolitan-
Specific Estimates of the Price Elasticity of Supply of Housing, and Their Sources, The
American Economic Review, 95, 334-339.

Grenadier, Steven R., 1995. The Persistence of Real Estate Cycles, Journal of Real Estate
Finance and Economics 10, 95-119.

44

Grenadier, Steven R, 1996. The Strategic Exercise of Options: Development Cascades
and Overbuilding in Real Estate Markets, Journal of Finance 51, 1653-1679.

Grenadier, Steven R, 1999. Information Revelation Through Option Exercise, The
Review of Financial Studies 12(1), 95-129.

Grenadier, Steven R., 2005. An Equilibrium Analysis of Real Estate Leases, Journal of
Business 78, 1173-1213.

Hilliard, E. Jimmy, Adam L. Schwartz and Alan L. Tucker, 1996. Bivariate Binomial
Options Pricing with Generalized Interest Rate Processes. Journal of Financial Research,
19(4): 585-602.

Hilliard, E. Jimmy, James B. Kau and Carlos Slawson,1998. Valuing Prepayment and
Default in a Fixed-Rate Mortgage: A Bivariate Binomial Options Pricing Technique.
Real Estate Economics, 26(3): 431-468.

Holland, Steven, Steven Ott and Tim Riddiough, 2000. The Role of Uncertainty in
Investment: An Examination of Competing Investment Models Using Commercial Real
Estate Data, Real Estate Economics, 28 (1), 33-64

Korpacz Real Estate Investor Survey
, 2
nd
Quarter 2008.

Malpezzi, Stephen and Duncan Maclennan, (2001). The long-run price elasticity of
supply of new residential construction in the United States and the United Kingdom,
Journal of Housing Economics, 10, 278-306.

Merton, R. C., 1976. Option pricing when underlying stock returns are discontinuous,
Journal of Financial Economics, 3, 125-144.

Nelson, Daniel B., and K. Ramaswamy, 1990. Simple Binomial Processes as Diffusion
Approximations in Financial Models. Review of Financial Studies, 3(3): 393-430.

ksendal, B.K. and Agns Sulem, 2005. Applied Stochastic Control of Jump Diffusions.
Springer.

Titman, Sheridan, 1985. Urban Land Prices Under Uncertainty, American Economic
Review 75, 505-514.

Quigg, Laura, 1993. Empirical Testing of Real Option-Pricing Models, Journal of
Finance 48, 621-640.

Waddell, Paul and Terry Moore, 2001. Forecasting Demand for Urban Land, Land
Market Monitoring for Smart Urban Growth, 185-217.

45

Williams, Joseph T., 1991. Real Estate Development as an Option, Journal of Real Estate
Finance and Economics 4, 191-208.

Williams, Joseph T., 1993. Equilibrium and Options on Real Assets, The Review of
Financial Studies 6, 825-850.

Zelman, Ivy L., Dennis McGill, Justin Speer, and Alan Ratner, 2006. Wonder-Land,
Credit Suisse

1

Appendix A: Technical Results
In the first part of this appendix, we show that the model in this paper can be placed in the
framework of stochastic impulse control models and establish that the verification theorem for
general stochastic impulse control problems for It-Lvy processes proved in ksendal and
Sulem (2005) (Theorem 6.2, p. 83) may be applied to our problem providing sufficient
conditions for the optimality of a solution.
Let B
0
denote the collection of Borel sets U for which0 U .
Define the vector process for the prices

( )
( )
( ) 0
( ) ( )
0
( )
P M
R M
r t
P t t P t
r t
t R t
R t
Pe P P P
d t d dt d t
R R
Re R

+
+

= = +

X Z

where
( )
( )
( )
P
R
z t
t
z t

=

Z . Let ( ), ,
i
q t i A D = be doubly stochastic Poisson processes with
respective intensities as defined in equations (16) and (18). It can be shown that these intensities
are nonexplosive in that for , i A D = ,
0
( )
t
i
s ds <
P almost surely, 0 t .

Now we define random measures
A
N and
D
N by
( )
0
( , ) ( ) ( )
i U i i
s t
N t U I q s q s
<
=
,
for , i A D = . Let ( ) A t and ( ) D t be processes that count total actual arrivals and total actual
departures by time t . We combine ( ) A t and ( ) D t into a single process, ( ) ( ) ( ) N t A t D t = , which
represents total units under contract at time t . In terms of the marked point processes, we can
write

2

( )
*
(0, ]
( ) ( ) ( , )
A
t
A t I N s z N zN ds dz
+
= +

and
( )
(0, ]
( ) ( ) 0 ( ) ( , )
D
t
D t I N s z I s zN ds dz
+
= <

.
where is the switching time (construction start time). In the model of the paper, we only
consider sales of whole units, i.e. ( ) [0, ] N t N . However, here we start with the more
general case of ( ) [0, ] N t N , and later specialize to the case with ( ) [0, ] N t N .
We also need to add an additional state to keep track of construction progress. Let ( ) C t
be the proportion of construction completed by time t . That is,
0
1
( ) ( )
t
C
C
C t I s T ds
T
= +
.
We can write the state process in vector form as

( ) [ ] { }
3
*
( )
( )
( ) 0, 0,1 0,1
( )
( )
( )
t
P t
R t
t N
N t
C t
t
+

=

Y

where { } ( ) 0,1 t indicates whether or not construction has begun.
Define the infinitesimal generator, A, of ( ) t Y corresponding to =
1
(starting with
0 = and never switching) by

2 2 2
1 , 1
( )
,
1
( ) ( , ) ( ) ( , )
2
( ( )) ( ) ( , )
T
i ij
i i j
i i j
j
j
j A D
h h h
h a t bb t
t x x x
h z h t dz
= =
=

= + +

+ + +

y x x
y y
A

3

where ( , , , , ) t n c = y x , and

( )
0 0
0 0
( ) 0 0
0 0
j
z if j A and if j D
z z

= = =

.
The optimal switching problem for vector It processes is solved as a stochastic impulse control
problem in Brekke and ksendal (1994). Stochastic impulse control theory for jump diffusions
specified as It-Lvy processes is developed in ksendal and Sulem (2005). However, ( ) A t and
( ) D t are not Lvy processes as their time increments of are neither independent nor stationary.
Nevertheless, doubly stochastic Poisson processes are Markov processes and thus satisfy the
Dynkin formula, which is really the crucial requirement for stochastic control theory. In
particular, for a function h in the domain of A, and given a set
5
D such that the first exit
time
{ } inf 0| ( )
D
t Y t D = > ,
satisfies [ ]
y
D
E < , we have
[ ]
0
( ( )) ( ) ( ( ))
D
y y
E h Y h y E h Y s ds

= +

A .
In the context of our model, this is the version of the Dynkin formula that Brekke and ksendal
(1994) require to hold for the state process.
Define the switching cost function by

( )
0
( , )
C
K
T
s
H K e ds
+

=
x , (1)
the time present value of construction costs from to
C
T + . Let ( ) t denote the rate at which
carrying costs are incurred for holding unsold units. The expected present value of total profit

4

using switching time is given by
( ) *
0
( )
( ) * ( )
( ) ( ( ) ) ( , )
( )( ( ) ) ( ) ( ( ))
( ) ( ) ( ( ))
C
C
C
T
r s t
A
t
r T t
C C C C
T
r s t r t
C
t
J E w e I N s N zN ds dz
e N T P T w I t T V T
e N s I s T ds e H

+

+
+

= <
+ + + + + +

< +

y
y
Y
Y

where

*
0
0
( )
( )
0
( )
( ) *
( ) ( , )
( ( ))
( ( )) ( )
t N
r s t
A
t
t
r s t
t
e P s zN ds dz
V t E
e N N s s ds
+

+

=

y
y
y
Y

with

{ }
*
0
( ) inf 0; ( ) s N t s N = > + y .
That is, ( ( )) V t Y
is the expected present value of the unsold units minus carrying costs at the
time of construction completion given the current state at that time.
The optimal switching problem is to find
( ) sup ( ) V J
= y y
T

where T is the collection of admissible stopping (switching) times and, if possible, the optimal
switching time
1
for which

1
( ) ( ). V J
= y y
Immediately after switching, ( ) V y is given by
( ) * ( ) *
0
( ) ( ( ) ) ( , ) ( )
( ) ( ) ( ( ))
C C
C
t T t T
r s t r s t
A
t t
rT
C C C
V E w e I N s N zN ds dz e N s ds
e N t T P t T V t T
+ +

= <
+ + + + +

y
y
Y

and before the switching times ,

5

( ) * ( ) *
0
( ) ( )
( ) ( ( ) ) ( , ) ( )
( ( )) ( ) ( ( )) .
C
t T
r s t r s t
A
t t
r t r t
e H Y I t e V Y
<
< +

y
y

If an optimal switching time exists, then

( ) * ( ) *
0
( ) ( )
( ) sup ( ( ) ) ( , ) ( )
( ( )) ( ) ( ( )) .
C
t T
r s t r s t
A
t t
r t r t
e H I t e V
= <
< +

y
y
Y Y

Define
( ) ( ) [ ] { }
{ }
3
* *
, , , , , 0, 0,1 0,1 | , , 1 t p r n c N t K n N c
+
= < < <

S ,
and let
S
be the first exit time from S . Let : f S and suppose that ( ) f y is the profit per
time unit when in state y , so that
( ( ))
C
T
t
E f s ds
+

y
Y
( ) * ( ) *
0
( ( ) ) ( , ) ( ) ( )
C C
T T
r s t r s t
A C
t t
E w e I N s N zN ds dz e N s I s T ds

+ +

= < < +

y
.
Let : g S be the bequest function, so that
( ( )) ( ) ( ) ( ( ))
C
rT
g e N P V
= +

Y Y
S S S S
.
Define the switching operator M on B, the family of Borel measurable functions on
( ) [ ] { }
3
*
0, 0,1 0,1 N
+

, by
( , , , , ) ( , , , ,1) ( , ) Mh t n c h t n c H t = x x x .
Verification Theorem. Suppose there exists : S such that
(i)
1
( ) ( ) C C S S
(ii) M on S

6

Define the continuation region as { } | ( ) ( ) y y M y = > C S . Furthermore, if
(iii) ( )
0
( ) 0 E I t
y
Y
S
C for all y S
(iv) C is a Lipschitz surface (i.e. locally the graph of a Lipschitz continuous function)
(v)
2
( \ ) C S C with locally bounded derivatives near C
(vi) 0 + f A on \ S C
(vii) ( ( )) ( ( )) ( ) t g I < Y Y
S S
as t
S
almost surely, for all y S
(viii)
{ }
( ( ))

Y
T
is uniformly integrable for all y S
Then ( ) ( ) V y y . Moreover, if it is also the case that
(ix) 0 A f + = in C
we have that, if { } inf 0| ( ( )) ( ( )) t t M t = > = Y Y , then ( ) ( ) ( ) J V
= = y y y .

Note that (ii), (vi), and (ix) can be combined into the single condition
( ) max ( ) ( ), ( ) ( ) 0 A f M + = y y y y , for all y S
with boundary values
( ) ( ) g = y y , for all y S .
Now we can specialize to the case with ( ) [0, ] N t N . Assume that for , i A D = , the
intensity kernel ( )
i
t
dz is of the form ( ) ( )
i i
t t
dz dz = , where ( ) is a Gaussian distribution
function with mean 1 and standard deviation
. So changes in the number of units under

contract upon tenant arrival and departure are identically distributed as (1, )
N . As 0
,
the verification theorem remains valid after making the obvious changes to the continuity and
smoothness conditions.

7

In the remainder of this appendix, we present a sketch of the proof that the solution
obtained from our numerical scheme is consistent with the continuous-time solution of the
stochastic impulse control problem.
First, Nelson and Ramaswamy (1990) show that their approximation converges weakly to
the correct diffusion, i.e. ( ) ( ) X X
. Furthermore, it can be shown that
. Let [ ]
0, D K
denote the space of -values functions on [ ]
0, K that are right continuous and have left limits.
For [ ] [ ]
( , ) 0, 0, T D K K , the mapping
( ) ( ) ( ( ), ) ( ) ( )
T
t
T f s ds g s +

is continuous with probability 1 when the probability measure is that induced by the distribution
of ( ) ( ), X . Thus, by the continuous mapping theorem,
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
t t
f s ds g f s ds g

+ +

Y Y Y Y
S S
S S
.
Since
( ) ( )
{ }
( ) ( )
t
f s ds g
Y Y
S
S

are uniformly integrable random variables, we have that
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
t t
E f s ds g E f s ds g

+ +

y y
Y Y Y Y
S S
S S
, for all y S .
This suffices to show that the value functions, ( ) V
y , from the Nelson-Ramaswamy

approximation converge to a solution (at least in the viscosity sense) of
( ) max ( ) ( ), ( ) ( ) 0 A f M + = y y y y .

8

Appendix B: Derivation of the Numerical Method
From the paper, the risky dP process is:
( ) ( ),
P
I I
t
t t P t P
P
dP P e P dt Pdz t

= +
and the risk-neutral process is:
( )
( )
( ) ( ),
P M
I I
r t
t t t M P t P
P
dP P e P P r dt Pdz t

+

= +

or equivalently

( )
'
1 ( )
P M
I I
I
r t M
t t P t P
P
P
r
dP P e P dt Pdz t

= +

Similarly, examining the R processes the risky process is

( )
( ) ( ),
R M
I I
r t
t t R t R
R
dR R e R dt R dz t

+

= +
and the risk-neutral

( )
'
1 ( )
R M
I I
I
r t M
t t R t R
R
R
r
dR R e R dt Rdz t

= +

Now we have to set up the numerical transformations. We begin with the two risk-neutral
processes:

( )
'
1 ( )
P M
I I
I
r t M
t t P t P
P
P
r
dP P e P dt Pdz t

= +

and
( )
'
1 ( )
R M
I I
I
r t M
t t R t R
R
R
r
dR R e R dt Rdz t

= +

.
For convenience lets first work with the dP process and re-write it as:
( , ) ( , )
P
dP a P t dt b P t dz = + .

9

The Nelson-Ramaswamy transformation first removes non-constant volatility in the stochastic
portion. We begin this process by defining a new process G=ln(P) and noting that Itos lemma
states:

2
2
2
1
( , ) ( , ) ( , )
2
P
G G G G
dG a P t b P t dt b P t dz
P t P P

= + + +

.
The partials for dG are:

2
2 2
1 1
; 0;
G G G
P P t P P

= = =

,
and substituting into Itos lemma gives us:

2
2
1 1 1 1
( , ) 0 ( , ) ( , )
2
P
P P P

= + + +

.
So that we wind up with

( )
' 2
'
1
1
( ) 2
P M
I
I
I
r t
M
P p P
P
P
P e
r
dG dt dz
P t

= +

.
Similarly we can derive a similar process for R, dH:

( )
' 2
'
1
1
( ) 2
R M
I
I
I
r t
M
R R R
R
R
R e
r
dH dt dz
R t

= +

The correlation between dz
R
and dz
P
is a constant (defined elsewhere.) We need to remove that
correlation, so we define two new processes X
1
and X
2
which we define as:
X
1
=
R
G +
P
H
and
X
2
=
R
G -
P
H .
From Itos lemma these processes follow:

10

( ) ( )
' 2 ' 2
1 1 ' '
1 1
1 1 2(1 )
( ) 2 ( ) 2
P M R M
I I
I I
I I
r t r t
M M
R P P R P R
P R
P R
P e R e
r r
dX dt dz
P t R t

+ +

= + + +

and
( ) ( )
' 2 ' 2
2 2 ' '
1 1
1 1 2(1 )
( ) 2 ( ) 2
P M R M
I I
I I
I I
r t r t
M M
R P P R P R
P R
P R
P e R e
r r
dX dt dz
P t R t

+ +

= +

For ease of exposition, we rewrite these as

1 1
dX dt dZ
+ +
= +

and

2 2
dX dt dZ

= + .

It is straightforward to recover G and H from X
1
and X
2
at any point in time, simply note that:

1 2 1 2
and, .
2 2
R P
X X X X
G H

+
= =

The final step, therefore, is to calculate the required transition probabilities from a node at time t
to a node at time t+1. Hilliard, Schwartz and Tucker (1996) demonstrate one gets convergence in
the process if the probabilities are set equal to:

1 1
.5
2
dt
p k

+
+
= +
and

2 2
.5
2
dt
p k

= + .

11

At each node the values of k
1
and k
2
must be selected to insure that 0<=p
1
<=1 and
0<=p
2
<=1. Since there are two stochastic variables in the model, from any specific node in the
lattice, there are four possible nodes that can be attained at the next time step: where both X
1
and
X
2
increase, where X
1
increases but X
2
decreases, where X
1
decreases but X
2
increases, and
where both X
1
and X
2
decrease.
24
We denote these nodes as uu, ud, du, and dd, respectively. The
probability of reaching each of these nodes is given by:
1 2
1 2
1 2
1 2
*
*(1 )
(1 ) *
(1 )*(1 ).
uu
ud
du
dd
p p p
p p p
p p p
p p p
=
=
=
=

Of course we not only want to track the risk-neutral probabilities, but also we want to be able to
work with the real or risky probabilities. From above, we note that the risky processes follow:
( ) ( ),
P
I I
t
t t P t P
P
dP P e P dt Pdz t

= +
and
( ) ( ),
R
I I
t
t t R t R
R
dR R e R dt R dz t

= + .
We then have to go through the Ito transformations. We can once again re-write dP
t
as:
( , ) ( , )
P
dP P t dt b P t dz = +
Except that we use the upper-case alpha to denote the different drift term from the risk-neutral
process. If we define a function G=ln(P) (again in the risky context) we from Itos lemma:
2
2
2
1
( , ) ( , ) ( , )
2
P
G G G G
dG P t b P t dt b P t dz
P t P P

= + + +

.
The partials for dG are:

24
As Hilliard, Kau, and Slawson (1995) discuss, to insure well-behaved probabilities it is sometimes necessary for
k
1
and k
2
to take on values such that X
1
or X
2
will increase or decrease in both their up and down states. When
that happens, however, the up version of either X
1
or X
2
is still greater than the down version of X
1
or X
2
.

12

2
2 2
1 1
; 0;
G G G
P P t P P

= = =

,
and substituting into Itos lemma gives us:
2
2
1 1 1 1
( , ) 0 ( , ) ( , )
2
P
P P P

= + + +

.
So that we wind up with
' 2
1
1
( ) 2
p
I
I
t
P p P
P
P e
dG dt dz
P t

= +

.
Using the same logic for the risky R process, we define H=ln(R) and we wind up with

' 2
1
1
( ) 2
R
I
I
t
R R R
R
R e
dH dt dz
R t

= +

We can now go through a remove the instantaneous correlation between dG and dH by defining
two new processes Y
1
and Y
2

Y
1
=
R
G +
P
H
and
Y
2
=
R
G -
P
H .
From Itos lemma these processes follow:
' 2 ' 2
1 1
1 1
1 1 2(1 )
( ) 2 ( ) 2
p R
I I
I I
t t
R P P R P R
P R
P e R e
dY dt dz
P t R t

= + + +

and
' 2 ' 2
2 2
1 1
1 1 2(1 )
( ) 2 ( ) 2
p R
I I
I I
t t
R P P R P R
P R
P e R e
dY dt dz
P t R t

= +

For ease of exposition, we rewrite these as

13

1 1
dY dt dZ
+ +
= +

and

2 2
dY dt dZ

= + .

Recovery of G and H values is exactly analogous the process used in equation (15):

1 2 1 2
and, .
2 2
R P
Y Y Y Y
G H

+
= =

Similarly, we will get convergence in the process if the risky-probabilities are set equal to:

Real
1 1
.5
2
dt
p k

+
+
= +

and

Real
2 2
.5
2
dt
p k

= + .

Where once again the values of k
1
and k
2
are selected to insure valid values for
Real Real
1 2
and p p .
The risky transition probabilities, therefore, are given by:

Real Real Real
1 2
Real Real Real
1 2
Real Real Real
1 2
Real Real Real
1 2
*
*(1 )
(1 )*
(1 )*(1 ).
uu
ud
du
dd
P p p
P p p
P p p
P p p
=
=
=
=

These risk-neutral probabilities can then be used at an arbitrary node t,i,j to determine the

14

risk-neutral expected value of a variable at time t+1.

OTT Steven

Uploaded by

Copyright:

Available Formats

OTT Steven

Uploaded by

Document Information

Original Description:

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

OTT Steven

Uploaded by

Copyright:

Available Formats

A Generalized Real Options Model of Land:

Valuation, Expected Returns and Development

( ) R t . When the market service flow rate ( ) P t is

= , and that the average market service flow will be positively

( ) , , t i j V . Note that the t in the

to be 7. The elasticity of arrivals

to be 1 and the elasticity of departures

. So changes in the number of units under

. Furthermore, it can be shown that

y , from the Nelson-Ramaswamy

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.