The Shale Gas Phenomenon: Utilizing the Power of System Dynamics to
Quantify Uncertainty
La Tonya N. Walker, Sandia National Laboratories', Inwalke@ sandia.gov
Leonard A. Malczynski, Sandia National Laboratories
Peter H. Kobos, Sandia National Laboratories
Garrett Barter, Sandia National Laboratories
Abstract
Abundance of shale gas and less expensive extraction techniques led to a boom of natural ges (Nt G) supply i in U. S.
with a corresponding drop in prices. This i igation captures a multitude of
factors that impact production. A few of the key findings include the ability to more accurately model the shale gas
behavior on top of the conventional and coalbed methane-based systems within the system dynamics environment.
This is especially noteworthy given the recent rapid increase in production within the U.S. The objective is to
quantify the key technical and economic drivers in the United States’ (U.S.) Natural Gas exploration markets. The
analysis does this by quantifying conditions in the NG exploration system that can lead to innovations and
transitions in U.S. NG supplies.
Importance
The low prices spurred increased use of natural gas (NG) in electric power generation, industrial and
commercial uses, as well as heavy-duty transportation. With its intrinsic thermal efficiency, and low carbon-
hydrogen ratio, there are also engineering and environmental benefits to using NG. However, growing the market
share of NG is d dent on d and infrastructure stakeholders having confidence that low
prices will continue in the future. If the U.S. economy becomes more dependent on NG, it also becomes more
dependent and more vulnerable to its price or supply volatility as illustrated in Figure 1. Also, reducing reliance
upon petroleum imports from countries that view the U.S. unfavourably would extricate foreign and military policy
from energy dependencies. The increase in known, domestic NG reserves, coupled with extraction innovations,
could realize such a future.
' Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a
wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear
Security Administration under contract DE-AC04-94AL85000. SAND2014-16613 C
Long-term Energy
Security
+ Supply Changes
+ Changing NG
Demand
Imports/Exports
Short-term Conditions
J) NG prices Eas Sornly
t Unproven Reserve
tT Technological Progress
Figure 1. Natural Gas Exploration Market Factors involved in Short-term C onditions and Long-term Energy Security
Dynamic Modelling Effort
This project therefore is developing capabilities to identify the propagation pathways of NG supply shocks
through the economy. By doing so, we can highlight the risks, vulnerabilities, and mitigation strategies across the
NG value-chain. Hence, we could begin to assess the feasibility of using NG heavily in all economic sectors and/or
of becoming a net NG exporter. This project has three primary tasks; quantifying the uncertainty in the supply,
infrastructure constraints, and demand dynamics. The supply-side investigation captures the i of
technological, geoscience factors that impact production. This analysis’ effort utilizes a system dynamics
fy to forecast NG production, and incorporates shale gas extraction and discovery developments, which
can be leveraged to conduct sensitivity studies of future technological and economic changes.
Methodology
This quantitative modelling project began through a literature review of the NG exploration process,
regulations, and history, which led to the discovery of Roger F. Naill’s “The Discovery Life Cycle of a Finite
Resource” model. Naill’s work provides the basic understanding of NG market dynamics. The translation of Naill’s
model from DY NAMO syntax to Powersim® Studio syntax was an iterative, intensive effort. Once the conversion of
the model was deemed accurate, a model calibration phase began to test parameters and restructure the model to
reflect the current NG regulatory environment. A dditional salient modeling efforts in the research community
focusing on natural gas systems include Moniz et al. (2011), Medlock (2012), NPC (2011), Abada et al. (2013), Chi
et al. (2009), Managi et al. (2005), and Sterman and Richardson (1985).
Natural Gas Exploration Process
Natural gas exploration is a process of characterizing sites for its potential gas output through geoscience
factors (e.g. thermal maturity, TOC, reserves’ areal and thickness distributions, etc.) as well as technological and
economic evaluations. A site is initially refined by its geoscience characteristics, and then the assessment of
technological feasibility of recovery is performed. This helps determine the breakeven price for profitable
production. Then regulatory and economic environments are assessed for any additional constraints and added costs
to determine if the reserve is proven to be economically producible.
Understanding the Nomenclature of NG Industry
The naming convention of natural gas quantities among the different data sources has inconsistencies.
Throughout the last 50 years there have been many attempts to categorize the natural resource stocks including
natural gas. A prominent attempt includes the joint effort of the Society of Petroleum Engineers and the World
Petroleum Council (SPE and WPC, 2005). One early attempt is the USGS Principles of the Mineral Resource
Classification System of the U.S. Bureau of Mines and U.S. Geological Survey (1976). All classification systems
examine two characteristic axes of a resource, knowledge of the quantity and cost of extraction. A recent attempt
from the Massachusetts Institute of Technology (Moniz et al., 2011) supports the basic idea (Figure 2). This work
classifies NG into Unproven Reserves (UPR) and Proven Reserves (PR). This classification preserves the general
ideas expressed in more formal classification work by simplifying the supply space.
Discovered/Identified
Confirmed Unconfirmed
TAN Cumulative
a Production
g 5 Inferred Undiscovered zx
z s Reserves/ Technically =
= R Reserve Recoverable a
z ' Growth Resources =
2 ee ee H_------
8 1
a 1
2 1
€ x
He -
z| || 8 ™
z $
2
a 8
<=
Increasing Geologic Knowledge
Figure 2. Adopted from Moniz et al. 2011 Figure 2.3 Modified McKelvey Diagram, Showing the Interdependencies
between Geology, Technology and Economics and Their Impacts on Resource Classes
Data Collection of Publicly Available Sources
After Naill’s model was converted from DYNAMO, the simulated results for price and Proven Reserves
were compared to historical cumulative data from U.S. Energy Information Administration (EIA) (Walker et al.,
2014). Data collection on NG industry is no small task. Collection is further complicated by the naming convention
and aggregation approaches used by different sources. Therefore, the primary data sources utilized were EIA and the
United States Geological Survey (USGS), and data from other sources such as Oil & Gas Journal and Potential Gas
Committee were used. Acquiring shale gas time series datasets on proven reserves, production, and consumption on
national and shale play aggregated levels has been a challenging task.
Basis for Model
Since the focus of the effort is to identify the potential for supply volatility, the model assumes the United
States’ NG industry has three main source —_ Shale gas, Coalbed Methane, and Conventional with Tight gas, and
all these sources are assumed to be ble, undif iated product that require a similar
production process and can be used intorchonecbiy. Each source type is assumed to be discovered and produced
based on the rate of retum on investments and cumulative quantity demanded. The supply-demand dynamics within
the model determines the NG price at the wellhead as a weighted average of the total costs of all three sources. This
assumption of undifferentiated product is similar to one used by Naill (1973) and Sterman and Richardson (1985).
The model considers effects of recognizing increases in the potential reserve based of technically feasible
NG unlike the models of Naill (1973) and Sterman and Richardson (1985), which have a single fixed initial value
for technically-recoverable reserve. The technologically-recoverable reserve increases due to the recognition of
gaseous hydrocarbons developed from source rocks (i.e., shale gas and coalbed methane) (USGS 2013 and NPC
2011). The recognition of these formation types as a lly-recoverable, icall itabl ibility was
caused by the technological breakthroughs of using horizontal drilling with hydrologic fracturing. The reserve
increase estimates were acquired from the USGS. Currently, the model simulates this as a discrete and exogenous
input process, but technology improvements are not explicitly modelled. A future direction of the model is to have
hnology imp! be an end process based on some internal thresholds.
Additional core model assumptions are:
e No interdependencies between gas and oil.
e Total production cost for each source type is a proportional to its cost of exploration.
e The cost of exploration is assumed to rise as resources are depleted.
e Quantity demanded is a function of current price and exponential growth in usage over time.
e — Investments in exploration are determined by sales revenue generated.
The initial insights into the basic dynamics of NG industry were derived from Naill’s work (1973). The
objective of Naill’s NG discovery model is to “represent and analyse the implications of the factors that control the
supply of non-renewable resources, in order to determine the nature of the turing point in supply, and examine the
effectiveness of various policies in alleviating the [shortage] problem” (Naill 1973). The main focus of the model is
to determine trends that would answer the question of what are the effects of governmental policies such as ceiling
price regulations or tax incentives on the short-term and long-term supply of NG resources. The motivating
processes in the model are the economic processes that cause the transfer of NG from unknown resource to proven
reserve category and subsequent exploitation of those reserves. There are two feedback loops in the Naill’s NG
discovery model, which are:
Discovery) Negative feedback loop, that symbolizes the long-term effects of unproven reserves
depletion on the exploration cost and discovery rate, which relates unproven reserves, cost of
exploration, discovery rate together.
Market Demand) _Goal-seeking loop, that dictates the need for new discoveries though proven reserves, usage
rate, and price, and is the system’s equilibrium mechanism via Reserve-Production ratios
(R/P) to desired R/P.
Though Naill’s model provided valuable insights, it does not address supply volatility issues without
modifications. The substantial increase in the technically-recoverable reserves (unproven reserves) caused by
technological advance is not possible in this model structure. Also, Naill’s model does not track different NG source
types and their associated parameter differences.
In addition, Naill’s model contained price policies for regulations that were no longer applicable for the time
period of interest (shale gas boom of early 2000’s and beyond), which complicated understanding the system’s
behavior. Therefore, price ceiling and control structure was removed from the model. Prior to 1992, price controls
regulation existed, which caused hindrances via suppressed production and restricted real demand (Joskow 2013).
The model begins in 1993 to eliminate those complications.
Arraying Model by Source Types
Then the model was arrayed to include producing and potential shale gas plays, cumulative coalbed
methane, and cumulative conventional gas. The model produced results that were difficult to explain given the
understood dynamic behavior. Thus, the modelers decided to go back to the pre-arrayed model to make the problem
more tractable. The model was arrayed using three categories: Conventional, Shale, and Coalbed Methane. The
national datasets became more important at this stage of the research to calibrate the model for shale gas and coalbed
methane.
With this change in model structure, the modelers recognized that the different NG source types have
inherent differences in normal cost of exploration and discovery delays. This alteration was deemed necessary
because the production behavior for shale gas and coalbed methane differ substantially from conventional natural
gas.
The considerations identified in the decision to array by source types:
¢ How NG price is determined. The NG price is a singular value as a weighted average function of all three
natural gas types’ costs and could be comparable to Henry Hub Price. The current structure may allow the
proven reserves of a specific type to decrease while another increases due to the individual prices being
different. The price and total cost influence investments in explorations which directly influences discovery
rate. The individual total costs were maintained to influence i in exploration for specific types
(e.g., percent invested in exploration (PIIE) and sales revenue (SR)).
e How usage rate is handled. The total usage rate is the cumulative NG quantity demanded. The total usage
rate is distributed among the source types based on their exploration costs (i.e., the highest portion of total
usage rate taken from the source with lowest exploration cost). One important aspect is the restructure to
redistribute total usage when a source type’s proven reserve is depleted. The quantity demanded is
determined by price and exponential growth in potential usage.
e How initial technically-recoverable (unproven) reserves is determined. The U.S. Unproven Reserves
estimate (2203.3 tcf) was obtained from the EIA for 2009, but three historical values for end of 1992 were
better initiating parameters. This process took two phases: estimate 1992 cumulative production prior to
arraying the model, and then estimate individual initial unproven reserve parameters.
= — Inthe first stage for non-arrayed model, a value for the end of 1992 was approximated based on
Naill’s Gas-in-Place? estimate (1040 tcf) assumed be accurate in 1962 (Naill 1973) subtracted by
end of 1992 proven reserve and estimate cumulative production from 1900 to 1992. The initial
cumulative Unproven Reserves amount was incrementally increased to estimate suitable initial
values and optimized using Powersim Solver, an evolutionary search algorithm built-in the
Powersim® Studio software.
= Inthe second stage for the arrayed model, initial unproven reserves for each source type were
calculated. The 1992 cumulative Gas-in-Place value was distributed between the three source
types based on their approximate percent contribution to the EIA 1992 cumulative NG production.
It was assumed the production distribution reflects the unknown, but assumed distribution in
unproven reserves. Then individual unproven reserves subtracted by three source type 1992
proven reserve estimates and their entire, historical individual production from 1900 to 1992
reported by EIA.
2 Gas-in-Place is the NG original quantity regard as technically-recoverable before production started in 1900. Naill
obtained this Gas-in-Place estimate from Hubbert (1969).
Current Model
Our model structure allows understanding the motivations that caused shale gas boom in the early 2000’s,
which will lead to providing forecasts and variability in the unconventional resource estimation. The model
simulates from 1993 to 2035. The casual loop diagram (Figure 3) is the aim of the model structure.
Extraction
Technology
Effectiveness
Discovery
Exploration
Technology ;
Effectiveness dnvestment ma
Extraction
Technology
‘Average Investment in
Grade Exploration
“ha Reserve-Production Technolo
Usage Rate——__ rotal Cost my
( Market dant Rate> R&D
Investment
Pr Demand Return on
: > : = fnvestment__-
(Time) ( OF
Substitution Extraction Tech.
Fraction Improvements Percent Invested
x . Sales inR&D
Potential Revenue a
Substitution Fraction ZP
NG Sources <Usage Rate>
Substitution
Figure 3. Causal Loop Diagram illustrating the conceptual intersections of Naill’s natural gas discovery model and
Behrens’s natural resource Utilization Model (Meadows and Meadows, 1973) with the addition of exploration and
extraction technologies for NG exploration market. (Note: An influence arrow with a minus sign at the end refers to the
underlying inverse nature of the relationship between the variable at the beginning of the arrow and the end. Thus, if the
beginning variable increased, the minus sign suggests the variable at the end of the arrow will decrease by some
relationship. The dashed boxes highlight the major stocks in the model.)
Several loops in Figure 3 offer topic-specific information that is salient to the underlying operations of the model.
The ‘Exploration Technology Improvements’ loop (shown using red lines) was developed to motivate the proven
hnology’s ability to p i increase proven
reserves by quantifying the relationship between new
reserves. From there, if the ‘Proven Reserves’ increase, this increases the ‘Fraction Remaining’ of the available
proven reserves of natural gas. If the ‘Fraction Remaining’ increases, there is a potential for the ‘Average Grade’ of
natural gas available to increase, but this part of the model is largely available as a placeholder when expanding the
model to site-specific gas fields. From the ‘Average Grade’, if this increases then the ability to maintain a similar
output (of BtUs, for example) could translate into a reduced ‘Total Cost’ for that amount. When ‘Total Cost’
increases, the ‘Return on Investment’ would decrease which would decrease the amount of ‘R&D Investment’
available to the system. If ‘R&D Investment’ increases, this might manifest itself in terms of increasing the
in Exploration Technology’ and thereby increase the potential to increase ‘Exploration Technology
Effectiveness’. When the ‘E: ion T gy Effecti ” increases, so might the ‘Discovery Rate’ of new
natural gas which then affects (increases) the ‘Proven Reserves’ to complete the ‘Exploration Technology
Improvements’ loop.
Similarly, the influential and subject-specific loops exist in the model shown in Figure 3 that include ‘Extraction
Technology Improvements’, ‘Reserve Limits’, ‘Natural Gas Sources Substitution’, as well as the two previously
describe including the ‘Market Demand’ and ‘Discovery’. It is worth mentioning the ‘Natural Gas Sources
Substitution’ represents the trade-offs between the three reserves types within the model including shale gas,
conventional and tight gas and coalbed methane.
Confidence Building
‘Did you build the RIGHT Model? Did you build the MODEL Right?’
To address these questions, the confidence building process was performed using five testing levels. These levels,
illustrated in Figure 4, include System Mapping (correct structure & actors), Quantitative Modeling (observed
behavior modes), Hypothesis Testing (feasible decision tules & boundaries), Uncertainty Analysis (realistic
ities), and F ing & Optimi & predi
System
Mapping
Forecasting & Quantitative
Optimization Modeling
Uncertainty
Analysis Hypothesis
Testing
Figure 4. Levels of C onfidence in the Pragmatic A pproach (Zagonal and C orbet 2006), Built Confidence in the Model’s
Structure and Parameters.
Calibration and Sensitivity Analysis
The calibration and sensitivity analysis are iterative processes. The calibration was implemented using
literature review to define a probable range of suitable values for the constant in the model (e.g. discovery delay and
normal cost of exploration), and then using optimization to reduce the squared residuals between simulated and
historical EIA proven reserve data.
The singular sensitivity analyses were performed using two approaches to assess the potential risks:
Tomado diagram (Figure 5) and a triangular distribution to develop the probability distribution plot (Figure 6). The
model tested for its sensitivity to 50% increase and decreases to all the initial and constant parameters. In the risk
assessment approach, estimates of the specific parameters were tested by using a combination of optimization and
tisk analysis via evolutionary search and Latin Hypercube methods, tools within Studio software.
50% increase = 50% decrease
Average Usage Rate, initial
Cost of Exploration, normal
Desired Reserve to Production Ratio
Discovery Delay for Coalbed Methane Gas
Discovery Delay for Conventional Gas
Discovery Delay for Shale Gas
Marginal Cost Margin a
Proven Reserves, initial ———
Unproven Reserves, initial
Usage Growth Constant for 1993 to 2007
Usage Growth Constant from 2007
Usage Rate in 1993 =
Usage Rate in 2007
-60% -50% -40% -30% -20% -10% 0% 10% 20% 30% 40%
Change in 2035 Forecasted Proven Reserve (% )
Figure 5. itivity Analysis il ‘ing a 50% increase and decrease from the initial values of select core variables.
Total Proven Reserves
ber
500,0007 Triangular Distribution Assumptions-
Initial Unproven Reserve Parameter
400,0004 Min: 1,101,650 bef
Peak: 2,203,300 bef
Max: 3,304,950 bef
Ba9,900;7 Bend due
Unproven Reserve meal
200,000-4 Recognitio: —————eee
100,000 SSS
jan 01, 1993 Jan 01, 2007 Jan 01, 2021 Jan 01, 2035
bet Total Proven Reserves
500,000+,
Triangular Distribution Assumptions-
Desired Reserve to Production Ratio
Min: 19.075 yr.
Peak: 38.15 yr.
300,000+ Max: 57.225 yr.
400,000+-
200,0004
100,000+ ~Total Reserves (High) ~Total Reserves (90 Percentile) ~Total Reserves (75 Percentile)
-Total Reserves (Average) ~Total Reserves (25 Percentile) ~Total Reserves (10 Percentile)
exe baa se
Preliminary Results
The simulated results of this model were compared to historical estimates of NG reserves and extraction rates, in
order to build confidence in the model. Figure 7 illustrates the simulated and historical natural gas reserves.
Total Proven Reserves Shale Gas
=Total Reserves =Historical Dry Proven Reserves reported =PR Proven Reserves[Shale Gas}
tof =Historical Shale Gas Proven Reserves reported
400. tef
300- 200.
200-
100. 100:
Jan o, 1993 Jan 01, 2007 Jan 01, 2021 Jan 01, 2035 Jan 01, 1993 Jan 01, 2007 Jan 01, 2021 Jan 01, 2035
Comparison of Simulated Results to EIA Historical Results for Comparison of Simulated Results to EIA Historical Results for
Total Proven Reserves. Shale Proven Reserves.
Conventional Gas Coalbed Methane
=PR Proven Reserves{Conventional Gas] =PR Proven Reserves{Coalbed Methane Gas]
Historical Conventional Gas Proven Reserves estimated ‘Historical Coalbed Methane Proven Reserves reported
tef tof
2a 20.
150 15
100 10:
50 5
0 0.
Jjan 01,1993 Jan01,2007 _ Jan0i, 2021 Jan0i,2035] Jan 01,1993 JanOl, 2007 JanOi, 2021 Jan 01, 2035
Comparison of Simulated Results to EIA Historical Values for Comparison of Simulated Results to EIA Historical Values for
Conventional Gas Proven Reserves. Coalbed Methane Gas Proven Reserves.
Figure 7. Simulated and EIA Historical Proven Natural Gas Reserves.
Figure 7 illustrates the close correspondence between the general shapes of the simulated results to the historical
data. This is encouraging due to the fact the underlying model framework is reflective of the underlying dynamics.
Specifically, as shown in Figure 8, the Reserve-Production (R\P) ratio helps drive the underlying behaviour of the
NG investments which then influences the available supply. These in tu influence the ultimate price of NG within
the representative market.
In the shale NG graphic within Figure 7, it is noteworthy to point out that using the original Naill model with several
modifications, his underlying framework proved to be useful. The shale component of Figure 8 illustrates the large
increase in the R/P ratio accurately reflecting the substantial increase in proven reserves seen in the U.S. in recent
years.
Reserve-Production Ratios:
All Reserves
-DRPR DESIRED RESERVE PRODUCTION RATIO ~RPR reserve production ratio[Coalbed Methane Gas]
—RPR reserve production ratio[C onventional Gas] RPR reserve production ratio[Shale Gas]
yr
504 gp
40 ee
30
20
10 =
0,
Jan 01, 1993 Jan 01, 2007 Jan 01, 2021 Jan 01, 2035
Figure 8. Comparison of Simulated Results among three source types and desired state for Reserve-Production Ratios.
Discussion and Future Research
Much of the research showed successful matching of the underlying model’ s results to historical shale gas
data. This is a unique and novel approach to capturing the geologic, technol 1 and
the NG market during recent surges in shale gas production. The next stage of this project is to incorporate policy
design and evaluations.
= Shale gas model:
= Explore the behavior, possibly introduce an exogenous variable based on threshold analysis (based
on reserve to production ratio) and related items.
= Delivery Constraint:
= How do current infrastructure limitations constrain available supply to meet demand in regions of
the U.S.?
= — Introduce more technology:
= — Using fewer resources to extract shale gas
= Potentially use leaning curve calculations to lower normal cost of exploration
= Introduce competition among source types in reserve exploration investments
= Policy considerations:
= Environmental costs or benefits of increased NG extraction and consumption
= Potential exporting of NG
10
Cited References
Abada, Ibrahim, Vincent Briat, and Olivier Massol. “Construction of a Fuel Demand Function Portraying Interfuel
Substitution, a System Dynamics Approach.” Energy 49 (2013): 240-51.
Behrens III, William W. “The Dynamics of Natural Resource Utilization.” Toward Global Equilibrium: Collected
Papers. By Dennis L. Meadows and Donella H. Meadows. Cambridge, MA: Wright-Allen, 1973, pp.141-64.
Chi, Kong Chyong, William J. Nuttall, and David M. Reiner. “Dynamics of the UK Natural Gas Industry: System
Dynamics Modelling and Long-term Energy Policy Analysis.” Technological Forecasting and Social Change 76.3
(2009): 339-57.
Energy Information Administration (EIA). “Dry Natural Gas Proved Reserves as of Dec. 31.” EIA, U.S. Department
of Energy, 2013. Web. Last accessed on 14 Jan. 2013.
Energy Information Administration (EIA). “Natural Gas Gross Withdrawals from Gas Wells.” EIA, U.S.
Department of Energy, 2013. Web. Last accessed on 14 Jan. 2013.
Hubbert, M.K. , “Energy Resources.” Resources and Man. By Committee on Resources and Man. San Francisco,
CA: W.H. Freeman and Company, 1969, pp. 157-242.
Joskow, Paul L. “Natural Gas: From Shortages to Abundance in the United States.” American Economic Review:
Papers and Proceedings 2013, 103(3), pp. 338-343.
Managi, Shunsuke, James J. Opaluch, Di Jin, and Thomas A. Grigalunas. “Technological Change and Petroleum.
Exploration in the Gulf of Mexico.” Energy Policy 33.5 (2005): 619-32.
Medlock, Kenneth Barry. “Modeling the Implications of Expanded US Shale Gas Production.” Energy Strategy
Reviews 1.1 (2012): 33-41.
Moniz, Emest J. et al. The Future of Natural Gas. Tech. no. ISBN (978-0-9828008-5-0). Massachusetts Institute of
Technology, 2011.
Naill, Roger F. “The Discovery Life Cycle of a Finite Resource: A Case Study of U.S. Natural Gas.” Toward Global
Equilibrium: Collected Papers. By Dennis L. Meadows and Donella H. Meadows. Cambridge, MA: Wright-Allen,
1973, pp. 213-56.
Naill, Roger F. Managing the Energy Transition: A System Dynamics Search for Alternatives to Oil and Gas.
Cambridge, MA: Ballinger Pub., 1977. Print.
National Petroleum Council (NPC). “Prudent Development: Realizing the Potential of North America’s Abundant
Natural Gas and Oil Resources” NPC 2011.
Society of Petroleum Engineers (SPE) and World Petroleum Council (WPC). “Glossary of Terms Used in Petroleum
Reserves/Resources Definitions.” SPE 2005.
Sterman, John D. and George P. Richardson. “An experiment to evaluate methods for estimating fossil fuels
resources.” Journal of Forecasting 4.2 (1985): 197-226.
11
US Government Printing Office. “Principles of the Mineral Resource Classification System of the U.S. Bureau of
Mines and U.S. Geological Survey”, U.S. Bureau of Mines and U.S. Geological Survey, Geological Survey Bulletin
1450-A, 1976.
US Geological Survey (USGS). “National Oil and Gas Assessment 2013 Assessment Updates.” National Oil and
Gas Assessment- Assessment Updates, USGS: Energy Resources Program. U.S. Department of the Interior. 2013.
US Geological Survey (USGS). “Reserve Growth Effects on Estimates of Oil and Natural Gas Resources.” USGS
Fact Sheet FS-119-00, USGS: Energy Resources Program. U.S. Department of the Interior. 2000.
Walker, L.T.N., L.A. Malczynski., “Converting DY NAMO simulations to Powersim Studio simulations,” SAND
2014-1343. 2014.
Zagonal, Aldo A. and Corbet Jr., Thomas F. “Levels of Confidence in System Dynamics Modeling: A Pragmatic
Approach to Assessment of Dynamic Models.” ISDC 2006: Submission #374.
12
increased profit margin. Thus, the varying "Interest rates' influence the ‘Minimum accepted price' of the market,
which in turn affects the actual 'Price'.
'Price' is determined by both, the 'Minimum accepted price' and the ‘Supply demand ratio’. A decri
ing non-
linear table function generates an 'Effect of supply demand ratio on price’. It captures that people are willing to
pay higher prices when supply is short, while oversupply reduces prices (Mankiw, 2010, p. 9). ‘Sensitivity of
price to supply demand ratio' adjusts the degree of how sensitive the 'Price' responds to changes in the 'Supply
demand ratio'. Accordingly, the price for housing is affected by the minimum epted price, the linked
construction costs and interest rates as well as the balance between supply and demand at any point of time.
‘Price! affects both, supply and demand. Both react according to their expectations of price developments
(Deutsche Bundesbank, 2013). Thus, the model structure to capture price and price expectation is similar to the
structure that captures “Construction costs’ and ‘Expected construction costs’. As 'Price' varies according to the
availability of supplied housing, it differs from the price perceived by both, suppliers and potential buyers.
Hence, suppliers and buyers form their expectations on price by adjusting the 'Expected price' to eliminate the
'Price variation' between the current and perceived value. Since real prices are not reported on a regular basis, the
‘Change in expected price' adjusts with an information time delay. There is an inflow to the stock when 'Price' is
higher than 'Expected price’. Vice versa, an outflow exists when ‘Price’ drops under 'Expected price’. Because of
its structural formulation, the 'Time to adjust expected price' results into a smoother and lagging development of
‘Expected price’ as reaction to variations in 'Price'.
For simplicity, we assume demand and supply determine prices adjustments with the same time delay. Yet, we
are aware that in reality, real estate experts, such as developers and consultants, might form their expectations
differently compared to potential customers that plan to purchase an apartment.
Eventually, market players on the demand side respond to changes in price by as: ing ‘Expected price to price
ratio’, i.e. the change in 'Expected price' relative to the current 'Price'. When 'Expected price’ over 'Price' changes,
there needs to be an opposing effect on demand. Hence, the non-linear, increasing S-shaped table function
determines the 'Effect of price on demand’, and regulates the 'Demand increase rate': Demand decreases when
the 'Expected price to Price Ratio! increases; vice versa, demand rises when a decline in prices is perceived. The
parameter Sensitivity of demand to price' determines the degree of changes in the behavior of demand. Muth
(1988) mes demand to be rather inelastic to changes in real estate prices.
EPPR = EP/P ay where EPTPR Expected price to price ratio
EP Expected price
P Price
EPD = (TEPD*EPPR)*SDP (12) where EPD Effect of price to demand
TEPD — Table for effect of price on demand
SDP Sensitivity of demand to price
TEPD = f(x); f 20; f'xex,.)203 f'xsx,0; (13)
The price setting mechanism of the STREM-model closes the first balancing feedback loop B1 - 'Demand
response’. On the one hand, assuming an increase in demand, initiated by a positive 'Households net growth rate!
and a subsequent increase in ‘Potential demand’, the 'Supply demand ratio! declines. The latter has a diverging
effect on 'Price', so that ‘Price’ increases when supply becomes scarce. On the other hand, a price increase creates
a 'Price variation’, so that 'Expected price' is adjusted respectively, yet with a time delay. Hence, an initial
increase in price translates into a declining 'Expected price to price ratio’ at first. Then, the effect of price
variation on demand is modeled with the variable 'Effect of price on demand’, which involves a lookup table and
a sensitivity parameter to determine the price elasticity of demand. In general, demand in real estate is found to
be not very elastic (Sterman, 2000; Muth, 1988). Further, the effect of price determines the 'Attractivity of real
ite purchase’. The latter variable affects demand by incorporating not only the demand side's reactions to
es
varying price levels of housing but also the effect of interest rates. 'Attractivity of real estate purchase’
increasing with lower prices, and vice versa, declining with an increasing price level. In addition, interest rates
is
determine real estate attractiveness as an investment option in contrast to other financial products, such as bank
account savings, bonds or stocks. Low interest rates make real estate a promising investment opportunity, thus,
raising its attractiveness — and vice versa. Consequently, demand declines because prices have increased,
contrary to the starting point. As already indicated in the CLD (Figure 6), demand is regulated by a balancing
feedback loop that adjusts demand and prices towards market equilibrium (Figure 9, B2 - 'Demand response').
Profit Generation
The price setting mechanism, and in particular 'Expected price’ determines the supply response to changes in
'Price' (Figure 10). For this reason, the supply side calculates its 'Expected profit' by taking into account
‘Expected price! as well as 'Expected construction costs'. Increasing prices boost profitability while construction
costs lead to a reduction. Profitability is the key driver of the supply side, as suppliers’ investment decisions are
determined by financially aspects. Therefore, when prices rise above construction costs, investors’ expected
profitability increases, too. Thus, higher 'Expected profit’ stimulates more new construction. We model this with
a table function that causes a change in 'Expected profit' to have a positive 'Effect of expected profitability’ on
‘Desired new construction’. 'Desired new construction' feeds into the inflow of the aging chain of the model, thus
pushing the 'Construction start rate’ to rise. Thereby, it closes the second balancing feedback loop B2 - 'Profit-
driven supply’.
pronning’ fim Avarage Average
janine’ me N Buildings under Constcian Time IN Bulings sales Tine yBulaings
male ec
Buildings Buildings
Completed Occupied
‘Gonstraction Sales Rate
Desired Now Start Rate: Completion Rate
2
erate a sma
eran gt eares Sart
‘ Sw opane cst cn
Ratio on Price seston Demand,
ATS, Segal
Construction Costs,
Expected Profit
Change in Expected
Construction Costs
Price Variation
Minimum iba
PF nccepted Price
x Expected Price to
+ Price Ratio
Accepted
Profit Margin
+ 4
Time to Adjust onstruction
Expocted Costs orga etime>
Future Construction Construction Cost Margin:
ost Assumption Index DATA
nang in
Expect Price
Figure 10: Substructure: Profit generation.
New construction in the stock ly, as i ingly more
available housing is supplied to the Stuttgart real estate market, the ‘Supply demand ratio’ increases as well. As
described above, the balance between demand and supply has a negative effect on 'Price'. Hence, more supply
results in decreasing prices, as people are not willing to pay as much as before. Here again, as the price
movement is opposite to the initial assumption (B2 - 'Profit-driven supply’), it regulates the supply side, in terms
eking equilibrium in the Stuttgart real estate
of the quantity supplied (i.e., new construction) and prices, s
market.
Furthermore, the supply side aims at satisfying demand. Thereby, constructors are aware of the oversupply
resulting in increasing vacancy rates and newly completed buildings cannot be sold. Unsold buildings or new
apartments implicate no revenues, so that suppliers are left with uncovered costs and insolvency risk. For this
reason, they try to forecast excess demand in the real estate market as accurately as possible. The according
structure in the STREM-model is described in the following.
Excess Demand
Investors on the supply side react to demand as
for many business decisions" (Lyneis, 2000, p. 3). Yet, as they cannot know the precise actual excess demand,
umptions about future demand and performance are essential
they estimate it (Figure 11). We model the structure similar to the price and cost expectations, as described
above. Once again, it takes time to form expectations, so 'Expected excess demand' is adjusted by an average
delay time (‘Time to form expectation of excess demand’).
The aim of the supply side is to meet 'Expected excess demand". 'Desired new construction' emerges from
‘Excess demand in buildings' and initiates new construction to be planned, thus determining the inflow into the
aging chain, ‘Construction start rate’. If Excess demand! increases, so does 'Expected excess demand’ after a time
delay, boosting 'Desired new construction’. After construction has begun and finally ends up as new ‘Buildings
completed’, the number of 'Vacant apartments' grows likewise. These, in turn, reduce 'Excess demand’, which is
contrary to the starting point. Hence, this substructure entails the balancing feedback loop B3 - 'Supply line
control’, which keeps the supply line under control in terms of construction activity. Thus, new construction is
only started according to given excess demand in the market.
Time to Form
Expectation 0
IN Expectea excess Bi
Excess Demand
ied
excess
ao cman
— pe Expected Excess,
Apartments
Buildings spre
Average Average Average Beas ym
Planning Time wildings under Consitucion Time yy Bui sales Tene INButdings TH" oF vito
ree naafe seuior
uitsings under Buliings Buliings
+ Craton P| Consiucton oan, Me] Completes [AP PP occupied [bemoiion rage
Desired Newer Start Rate ‘Completion Rate
* ‘Sele Goneraton
Potentia
Bales,
' Demand
demand in eman ean
Buildings satisfaction Rate increase Rate
Figure 11: Substructure: Excess demand.
Ce ining the ab p of the STREM-model, Figure 12 illustrates the complete model
structure. As already explained with the CLD (Figure 6), two balancing feedback loops B1 and B2 primarily
determine the structure. On the one hand, demand responses to changes in price, on the other hand, the supply
side reacts with new construction if it is profitable. Next, the model is tested and validated. An extended
overview of the STREM-model is also provided in Appendix A, including all variable types specified.
onsite Ban Soe rete ae AI
pects Came rece gh
Bais ‘Seay g_ Sing rt
pry seri a 5 = ae
vs . SE ange
s .o > 5 ee
7 orgs BA +
Figure 12: Stuttgart real estate market model (STREM) ~ full model.
Model Validation and Testing
Sterman (2000) describes model validation as a "continuous process of testing and building confidence in the
model" (p.81). Nonetheless, models can never be validated in the sense of verified, as they are simplified
representations of the real world's systems (Sterman, 2000, p. 846). Still, model testing is a crucial step in the
system dynamics modeling process. For this purpose, there is a wide range of testing methods that can be applied
in order to increase confidence in the model (Sterman, 2000, pp. 859-861). They are categorized into behavior
and structural testing.
Structural Testing
Model Boundary Adequacy Test
The model boundary has been laid out above in the form of a model boundary chart (Table 1) and a causal loop
diagram. By reviewing adequate literature and from the insights gained from the expert interviews, we identified
variables and causalities relevant for the system's behavior. In doing so, we modeled important variables
endogenous to the system, while others needed to be left as exogenous input or even omitted.
Structure Assessment Test
Structure assessment testing refers to the consistency between the model and the real system in the context of the
model's purpose (Sterman, 2000, p. 863-864). Tools used in boundary assessment can be applied likewise.
Existing literature and expert's statements have helped to model the system's structures and formulate valid
equations compared to the real system. An important question to ask is whether the model adopts basic physical
laws (Sterman, 2000, p. 846). The Stuttgart real estate market is especially constrained by its topographical site,
yet the simulation model does not capture this aspect of natural limits to expansion. As the supply chain of the
model involves a source and sink, it implies infinite in- and outflow. However, adjustment times and other
variables control accumulation and depletion of stocks and prevent them from generating surrealistic behavior.
For instance, ‘Construction start rate' can only start to flow when 'Desired new construction’ is available.
Dimensional Consistency Test
We tested the model for di ional i th ‘hout the entire modeling process. All variables are
specified with units. We distinguish between the units buildings and apartments since supply constructs
whereas | demand ap . In order to link the variables, the units are transformed where
necessary with the parameter ‘Apartments per Building’. With an average of four apartments per buildings, the
parameter is set as an actual value of the Stuttgart's housing market (Heilweck-Backes & Straub, 2007, p. 104).
The variables 'Price' and ‘Construction costs' are specified as indexes, and thus are consistent. Yet, the base value
is set at two different years, creating a limitation of the model. However, as they are not directly linked in the
model, the limitation is mitigated. For instance, ‘Accepted minimum price' takes into account only 'Expected
construction costs' and computes 'Accepted profit margin’ only based on costs.
Parameter Assessment Test
Several parameter values of the model are estimations based on qualitative data retrieved from literature and
personal interviews. Where available, they have been adopted as numerical data from other references. In those
cases, their sources are cited either in the text when described or in the model equations. Statis
ical methods, as
proposed by Sterman (2000) to estimate parameters should be considered in future research and optimization of
the model. Despite some weakne: of the model assessed through structural testing, e.g., capacity constraints,
price indexes, p imati the followi i testing helped to further build confidence in the
model.
Behavioral Testing
Extreme Conditions
Behavioral testing involves assessing the system's behavior under extreme conditions. It reflects whether the
system's behavior and the model equations still make sense when the model is exposed to extreme values of
inputs (Sterman, 2000, p. 869-870). This way, one extreme condition te: sumption that
demand drops to zero. A physical reaction would be, for example, that supply does not initiate any new
construction projects, so that the inflow ‘Construction start rate! falls to zero as well. A simulation under this
condition shows that the model responds plausibly to unavailable demand (Figure 13).
the model under the a
500
375
lings/year)
250
125
Construction Start Rate (bui
0
1970 1980 1990 2000 2010 ©2020» 20302040
Figure 13: Extreme condition: zero demand.
We conducted further tests under likewise extreme conditions, such as shocks in supply or extremely high profit
margins that result in price is lhening the confidence in the model.
Partial Testing
Partial testing helps reducing complexity in a model's behavior by cutting feedback loops or 'switching off
certain variables (Morecroft, 1985; 1988). This way, we tested several substructures of the model when
developing the model on whether they behave in a reasonable manner. In order to be able to analyze the drivers
17
of the real estate cycles, we conducted several partial tests. This also increased our understanding of the behavior
of the full model. For instance, to analyze the origins of the cyclical 'Price' behavior, the substructures
encompassing interest rates and construction costs were cut off. The effects on prices are shown in Figure 14:
Regularly recurring oscillations are visible throughout the simulated time horizon (dashed and dotted-dashed).
The partial test thus confirms that oscillatory behavior arises from within the system, and is not caused by
irregular developments in costs and interest rates. Including exogenous inputs rather mitigates oscillations (base
run, solid), whereas partial testing without interest rates result in intensified cycles (dashed). In addition, feeding
the model with actual data on construction co:
s over the given time period, accounts for a likewise upward
development in housing prices. Thus, cutting off these costs results in an almost constant average price index (an
equilibrium), around which oscillations reoccur in approximately ten-year intervals (dotted-dashed).
300
No interest rate
No interest rate + constant
225 construction costs oo x
!
3 cN rt
Bi .
2 1so |- ese Vv,
7 toy
78
Base run
°
1970 1980-1990 2000-2010 «= 202020302040
Figure 14: Partial testing: Oscillating price behavior continuous.
Behavior Reproduction
The behavior testing method as
ses whether the model reproduces the real system's behavior as illustrated in
the reference modes (Figure 4 and 5). "The good fit between simulation run and real world data is an important
step in ensuring that the model structure correctly estimates short-term and long-term interdependencies between
variables and depicts realistically the development of (...) market data" (Kapmeier et al., 2011, p. 16).
Considering the mode of price in the Stuttgart real estate market, the base run of the
STREM-model reproduces the real price movement well (Figure 15, left graph), yet, overestimating actual price
since the beginning of the 1990s.
250
historical \
1875
simulated
Price (index)
historical
Buildings occupied (# buildings)
1970 1980 1981 2001 2011 1370 1979 1988 1997 2006 2015
Figure 15: Behavior reproduction for 'Price' (left) and ‘Buildings occupied’ (right).
Likewise, the construction activity of the supply side results in a simulated growth rate of housing stock similar
to the real data, yet not with a same good fit as the price behavior (Figure 15, right graph). The simulation is
underestimating actual data. An explanation might be that the aging chain is too simplified and omi
factors of the construction side, which might result in the observed inaccuracy.
several
Sensitivity Analysis
Since various parameters needed to be estimated when no real data are available, a sensitivity analysis helps to
determine to understand how sensitive the model reacts to certain parameters. Parameters included in the model
18
encompass adjustment times, such as ‘Average planning time' and 'Average construction time’ as well as
elasticity of demand and supply (‘Sensitivity of demand to price’ and 'Sensitivity of supply to price’). The
sensitivity analysis has been conducted by assessing the sensitivity of the several parameters on the key variable
Price’.
As Figure 16 illustrates, selected parameters cause different behavior of 'Price’. 'Price', for example, exhibits
highest sensitivity when the p sitivity of price to demand supply ratio' takes on values ranging from
zero to ten. Consequently, the parameter strongly influences the oscillating behavior of Price’, in particular the
amplitudes. Compared to that, the ining sensitivity itivity of demand to price’ and
‘Sensitivity of supply to price’, result in a lower sensitivity of 'Price'. In addition, the lower left chart of Figure 16
indicates that time delay in the aging chain, i.e., supply chain of the model, influences the cyclicality in price
movement. Testing the sensitivity of price on a ranging 'Average planning time! between zero and ten years, the
chart shows the sensitivity of price in terms of its oscillatory behavior.
Sensitivity: ‘Sensitivity of Price to Demand Supply Ratio’ on ‘Price’ Sensitivity: ‘Sensitivity of Demand to Price’ on ‘Price’
Runs: 200, Min: 0, Max: 10, Noise seed: 1234 Runs: 200, Min: 0, Max: 10, Noise seed: 1234
Sensitivity: ‘Average Planning Time’ on ‘Price’ Sensitivity: ‘Sensitivity of Supply to Price’ on ‘Price”
Runs: 200, Min: 0, Max: 10, Noise seed: 1234 Runs: 200, Min: 0, Max: 10, Noise seed: 1234
Figure 16: Sensitivity analysis: Sensitivity of price to demand supply ratio on Price, Sensitivity of demand to
price on Price, Sensitivity of supply to price on Price, Average planning time on Price (clock-wise from upper
left).
After conducting several tests according to recommendations given by Sterman (2000) and Morecroft (1985;
1988), including structural and behavior testing, the model can be assessed as valid. In the following section, its
behavior is analyzed in a simulation base run.
Model Base Run
The base run is conducted for the predefined time horizon of 1970 to 2045 and comprises available historical
data. A 'business as usual' (BAU) strategy is applied, which adopts a continuation in several parameters’ behavior
as usual. Accordingly, it is assumed that the household growth rate continues to rise with its average growth rate
of the last 20 years. Similarly, the exogenous construction costs take on the average value of the last 20 years.
The interest rate remains constant as of its last available data in 2016 (Osterreichische Nationalbank, 2017).
Future ink
costs and interest rates will play a significant role in scenarios,
so that their assumed values are comprised as separate variables in the model (highlighted in green in the
STREM-model; Figure 12): 'Future household growth rate assumption’, 'Future construction cost assumption’,
and 'Future interest rate assumption’. Settings of the base run, scenarios and policy are provided in Appendix C.
19
The following figures depict the behavior of selected variables in the STREM-model over the time horizon. As
shown in the left graph in Figure 17, 'Demand' oscillates despite an almost linear increase in 'Households of
Stuttgart’. Thus, the oscillation originates not from 'Households' but from another source: The balancing
feedback loop B1 is the driver that makes 'Demand' to respond to changes in 'Price' and vice versa, resulting in a
cyclical behavior. As the balancing loop B1 - 'Demand response' is strongly interacting with the second one on
the supply side, B2 - 'Profit-driven supply’, it correspondingly affects the demand cycles as well.
20000 ap
400008 ovo tes
300 inden
soo Dernandin
00008 ‘semble walle Ny
Households in
15,000 ap Stuttgart | ™s
3000 bides
80 index
300,000 hh
7500 a9 500 bidgs
250,000hh
75 index
Demand
Hovseholdsin Stutteart (households)
ents)
Demand (apa
oa | Buildings completed
200,000 hh“ bide! ings compl
1970 1980 199020002010 «2020 20302080 ‘70 1980 1990 + 2000«2010:«=«2020«=« 20302080
Figure 17: Base run: Behavior in 'Households' and 'Demand' (left) and Behavior in 'Price', Demand in buildings’,
and ‘Buildings completed’ (right).
The right graph in Figure 17 depicts that the supply side responds with a lag to the demand development, due to
the time delays in the stock and flow structure of the model, i.e., in the supply chain, as well as in the forecasting
process of expected excess demand. With increasing demand, prices also increase. Consequently, expected
profitability increases, which makes new real estate construction attractive. Due to the time delays in the supply
side's response to satisfy new demand, buildings are only completed years later, which is reflected in lagging
cycles (see Buildings completed’ in Figure 17). After reaching a peak in 2020, high prices make demand fall
again according to the macroeconomic law of demand. This reverse behavior translates into declining prices,
which in turn offset declining demand. Consequently, demand starts to increase again, starting a new cycle.
The delays in the aging chain are illustrated in Figure 18: When 'Desired new construction' is assessed (blue), the
behavior of the aging chain is initiated. However, the 'Average planning time' delays new construction, so that
the 'Buildings under construction’ are lagging behind (red). After construction projects are started, it takes the
‘Average construction time! until the buildings are completed and ‘flow’ into the stock 'Buildings completed’
(green). One can clearly observe that the peaks as well as the lowest turning points of the three variables are
lagging one after the other. Thus, completed buildings are placed on the market relatively late to satisfy the prior
increased demand. In contrary, when apartments are ready for sale, demand is in decline again. Consequently,
the balancing loop B2 inhibits ‘Desired new construction’ in response to lower demand, lower prices and lower
expected profitability and turns the cyclical behavior in the aging chain into a reverse movement. The variations
in amplitudes of the supply chain result from the varying demand cycles (Figure 17, red), as well as from
differences in feasible construction capacities.
20,000
15,000
2 Desired new
2 construction \
2 10,000 a
= Buildings under
e construction
3 Buildings
& 5000 completed
0
1970 1980 1990 2000 2010 2020 2030 © 2040
Figure 18: Base run: Behavior in Aging Chain.
20
The oscillating behavior of supplied buildings is also present in reality as indicated with the red line in Figure 19.
Since the STREM-model is only a simplified structure of the actual supply chain and is not constrained by
significant factors, such as construction capacity or available building land, the amplitudes of the simulated
cycles do not reproduce the behavior of real data.
2,000
= 1,500
2
=
= 1,000
2 simulated
e
8
2 500
2 . Peace
3 wo
historical
oO
1970 1980 1990 2000 + +-2010+~=«2020:~«=«080~=~=« 0D
Figure 19: Base run: Behavior in 'Buildings completed’.
Consequently, the observed real estate cycles arise due to significant delays and a diverging interplay between
overshooting and undershooting variables. These time delays include the supply side responding to changing
demand by forecasting excess demand. The time delay it takes for the completion of new buildings, including
both 'Average planning time' and 'Average construction time', causes prices to increase in that period, as demand
in the
cannot be satisfied and shortage in supply dominates the market, or, as the Real Estate Director states
interview, "there is just little supply" (personal communication, September 23, 2015). Another interviewee
confirms that "the supply of high quality real estate is really short" (Real Estate Advisor 1, personal
communication, September 23, 2015).
When construction is finished, the stock ‘Buildings completed’ accumulates. However, since demand in the
meantime has asa toi d prices, supply excesses demand. This way, the variables
of the STREM-model undertake corrective actions that result in a new cyclical period.
The base run based on a BAU setting results into a model behavior that reproduces the behavior of the Stuttgart
M ', by comparing it to the reference
real estate market as observed and described by the interviewees.
modes, i.e., real data, the confidence in the model is increased. Hence, it can be continued with experimental
runs that comprise different scenarios as well as policy design.
Scenario 1: Increasing Interest Rates
As our interviewees stress that interest rates play a significant role for the Stuttgart real estate market, we analyze
the impact of changing interest rates in more detail in the first scenario. Interest rates in Germany have reached
an all-time low in the observed time ° horizon (Osterreichische Nationalbank, 2017). The downturn in financial
markets has resulted in an i of fi jal assets (real estate)" (Deutsche Bundesbank,
2014a, p. 46). ‘Cement gold’ as Real Estate Advisor 1 (personal communication, September 23, 2015) states,
determines the current trend in capital investment. Notably, our interviewees all agree that interest rates do have
a large impact on the development of the real estate market. The Real Estate General Manager (personal
communication, September 23, 2015), for example, emphasizes that the demand side is strongly driven by the
interest situation. He continues that in situations of low interest rates, clients tell him that:
*Lam not selling.” And when many more people act likewise, supply will collapse. Supply is just very
short. And when you have hundreds [of people] waiting in the waiting line who say ‘I want to buy a real
estate; I don’t want to have my money laying around on my bank account. And anyway, the credit
conditions are so attractive! Now or never! Now it is just great!” And what happens then? Then, demand
will continue to increase. If supply is decreasing and demand increasing, what happens with prices?
21
They will increase. This is the mechanism that has occurred since 2010/2011. (Real Estate General
Manager, personal communication, September 23, 2015)
He continues explaining that clients call him saying "’Ah, I have money on my bank account. Don’t you have
some nice real estate for me? It could also be a house, or an apartment building, or two or three apartments or
whatever™ (Real Estate General Manager, personal communication, September 23, 2015). So, it seems as if
people just continue buying to whatever price, which makes the boom stronger. Real Estate Advisor 1 further
explains that:
People always live in some real estate. And if it is not nec
rily a castle which will collapse in the
near future, people just buy it immediately. For example, there is really high demand on investment
property for speculation as there are currently no investment alternatives. These days, you can offer
property for four or five percent [per year] only. (Real Estate Advisor 1, personal communication,
September 23, 2015)
Consequently, while the market gets short, demand is increasing also at low return rates. Real Estate Advisor 2
(personal ication, Sep 23, 2015) also i the demand side when saying that "the second
player in this market that have led to real estate scarcity and to this high demand are those who can, with these
low interest rates, afford to buy their own place instead of paying rent".
When asked for the real ite supply, the Private Real Estate Director (personal communication, September 23,
2015) characterizes it as "a market with low supply". Real Estate Advisor | (personal communication,
September 23, 2015) is convinced that "(...) it will always stay like this, supply is just too tight." It seems as if
the expert believes the situation is out of control, as he admits that "usually, the market is determined by supply
and demand". And this
September 23, 2015).
all due to the interest level (Real Estate General Manager, personal communication,
At the end of 2015 however, the Federal Reserve raised the interest rates in the United states for the first time in
nearly a decade (Wiebe, 2015) and just recently for the third time since the financial crisis (The Economist
Newspaper Limited, 2017). One could assume that the European Central Bank will follow eventually. Contrary
to the base run, where interest rates remain as low as of 2015, the first scenario assumes a gradually increase in
interest rates after the end of 2016. Hence, 'Future interest rate assumption’ ent:
s an increase up to seven
percent, leaving the current low-interest environment behind. Regarding the insights gained from the interviews,
it is expected that prices will decline along with increasing interest rates (Real Estate Advisor 2, 2015; Real
Estate General Manager, personal communication, September 23, 2015). Figure 20 illustrates real estate losing
attractiveness (dotted-dashed), which into declining demand to the base run (solid).
E
= Aaeeran Base run
£15] B a
A ~ .
3° 2 © 3000
i senor ~ \ | 2
$s Bu
§ E scenstio1
£
+ 0.
197019601990 2000 2010-2020 ~—-2080~—~—«2000 ism 19801980 2000 2010 2020 2030 3040
Figure 20: Scenario 1: Increasing interest rates resulting in declining attractiveness of real estate (left) and
demand (right).
Figure 21 presents the simulated behavior of model under the defined conditions of Scenario 1 exemplified by
the variable 'Price' (dotted-dashed).
22
300
Scenario 1
225
150
Base run
Price (index)
historical
0
1970 1980 1990 2000 ©2010 2020 2030 2040
Figure 21: Scenario 1: Increasing interest rates and Behavior of 'Price'.
Contrary to the initial expectation, ‘Price’ in this scenario (dotted-dashed) first behaves similar to the base run
(solid). The drop in 2025, however, is a little steeper, but it increases at a higher rate after 2025. Yet, in this
scenario, there is another sharp drop in the mid-2030s. After that, price recovers and increases on a higher level
than in the base run. Possible explanations for its behavior is the 'Minimum accepted price' that is determined by
the interest rates. As explained earlier, increasing interest rates cause suppliers
thus r:
to claim higher profit margins and
e the minimum accepted price level. Consequently, the market price increases likewis
drop in demand (between 2030 and 2040) is reflected by a decline in price as well. Despite the interviewee:
statement that interest rate affect demand in real estate, a study by the Deutsche Bundesbank (201 4a) indicates
that "interest rates seem to play a subordinate role in determining house prices (...), instead, the recent price
. However, the
increases are more likely the result of households’ productivity and income expectations" (p. 23). Hence,
modeling causal relations between interest rates and variables endogenous to the STREM-model needs to be
assessed carefully.
Scenario 2: Decline in Households
The second scenario is based on the previous one, assuming a light increase in interest rates. In addition, we
assume a drastic decline in h hold:
Stuttgart's attracti highly depends on its regional economic
strength, as laid out above, in particular on its cluster of automotive manufacturers and suppliers. In this scenario
we assess the effect of a declining economic importance of the automotive industry over a period of 30 years
(2015-2045). This might happen, for instance, consequently to possible future competitors like Tesla, Google, or
Apple, capturing higher market share with highly innovative, self-driving, and elect rs and making the
resident automotive manufacturers less important (Afhiippe et al., 2016). In this scenario, latter cut back jobs.
Although the STREM-model does not comprise any variables representing Stuttgart's regional economy directly,
this scenario can be modeled by a rapid decline in 'Future household fraction growth rate assumption Scenario 2'.
Naturally, as the city becomes less attractive and many inhabitants are not employed anymore, people are
moving away, resulting in a negative 'Household net growth rate' which gradually depletes the stock ‘Households
in Stuttgart’. A weak regional economy may also affect the construction industry, so that we assume a slight
decline in construction costs as well, modeled by ‘Future construction cost as
sumption Scenario 2'.
The following figures illustrate the resulting behavior of the STREM-Model of scenario 2 compared to the base
run. In Figure 22, after 2020, when the number of households decline (dashed), demand decreases faster
(dashed) than in the base run (solid) and continues to decline even further, while demand in the base run is
already turn upwards (around 2025). The rapid drop in demand leaves the supply side immensely overshooting
its equilibrium, which is reflected in Figure 23.
23
6,000 bldgs
400,000 hh
Demand in buildings Households
Base run >, ase run
= 4,500 bldgs
2 350,000 hh Demand in buildings
Scenario 2
3,000 bldgs
300,000 hh
Households
1,500 bldgs Scenario 2
250,000 hh .
oy
~ v3”
Households in Stuttgart (# households)
--
O bldgs
200,000 hte70 1980 1990 2000 2010 2020 2030 2040
Figure 22: Scenario 2: Declining households leading to a drop in ‘Demand’
i 20,000
e n 15,000
3 seenario2 S S
£ et g Desired new
a 2 construction - Sc 2
in) 2 >
2 10,000
z 14 = Buildings under
2 i s construction Sc 2
E rot S Buildings
21 et % 5000 completed~Sc2
3 H
1
° a g 0
197-1980 -~—«*1990~—=«2000-=«O10-=S=«0D-—«2030—~=S=—«OA 9701980 990 -—«2000=«ao10=«020~=—=—«2080 «2040
Figure 23: Scenario 2: Declining households leading to supply overshooting.
In this scenario, 'Supply demand ratio’ exhibits intensified amplitudes to the decline in households and its
corresponding demand (Figure 23, dashed in left chart). Moreover, the supply side responds with a rapid
decrease in 'Desired new construction' to almost zero after reaching its peak in 2020 (Figure 23, blue in right
chart) with lagging cycles in the aging chain (see 'Buildit
in green). Consequent to the high oversupply of housing, prices drop accordingly (Figure 24, dashed). Low
prices make real estate an attractive investment though, so that demand rises eventually and subsequently prices
under ion’ in red and 'B
as well. as Figure 24 i the decline in households in Scenario 2 leave price on a lower level
in average (dashed) compared to the base run (solid).
300
Base run
225
150
Price (index)
historical Scenario 2
0
1970 1980-1990 2000 ©2010 «= 2020.» -2030S 2040
Figure 24: Scenario 2: Declining households leading to a drop in 'Price'.
The behavior of the model though needs to be assessed with care since several variables, in particular
construction capacities take on surrealistic values ly to leave out ints and natural capacities of
the Stuttgart real estate market.
24
The second scenario simulates possible effects of a demand shock on the Stuttgart real estate market. In reality,
Real Estate Advisor 2 described a similar scenario, when referring to a weak economy affecting the Stuttgart real
estate market:
I do remember well that during a time when Daimler Benz was in financial struggle way back, the
entire region of Béblingen and Sindelfingen, where many Daimler Benz employees live, was
struggling. Prices did suddenly go down. As many former employees needed to leave the region, they
threw their houses on the market. (Real Estate Advisor 2, personal communication, September 23,
2015)
Interestingly, this statement came from the most experienced market expert. His younger colleagues have not
experienced a former bust market situation. So, this might lie outside of their range of thoughts.
Policy: Consideration of Underway Construction
The last experimental run implies a policy design, which is partially adopted from Barlas et al. (2007). Although
it is difficult to change the feedback structure of the STREM-model, this experimental setup aims at reducing the
1 s in certain variables that are ascribed to
strong cyclical ics. To the cycli
substantial time delays as described above, underway construction is considered and incorporated it into
planning when estimating 'Desired new ion’. This way, the supply side should be able to better respond
to changes in demand in order to avoid extreme over-/undershooting of its equilibrium that come along the time
delays within the aging chain. Therefore, the structure of the model needs to be adjusted by linking the two
stocks ‘Buildings under ion' and ‘Buildings ' directly with 'Desired new construction’ as
highlighted in Figure 25.
Expected Excess
Demand in Buildings B38
= ° Supply Li
~ pe ppl Line
Boral
a
Average _—
Planning time yy Buildings under Construction Time
Construction
Desired New Construction Genet Construction _
4 Construction Start Rei Completion Rate
Figure 25: Policy: Consideration of underway construction in the model.
Accordingly, the equation of the variable 'Desired new construction’ needs to be changed to:
DNC =(EEDB-BUC-BC)*EPDNC (14) where DNC Desired new construction
EEDB _ Expected excess demand in buildings
BUC Buildings under construction
BC Buildings completed
EPDNC Effect of Expected Profitability
on Desired New Construction
The resulting behavior is illustrated in Figure 26, where, indeed, the cycles are reduced. 'Desired new
construction’ (Figure 26, dashed in left chart) is significantly flatter than in the base run (solid), while the stock
‘Buildings completed’ respectively exhibits almost linear behavior (policy in dashed vs. base run in solid). The
policy prevents excess supply that arises when underway construction is ignored. Consequently, the 'Supply
demand ratio’ is fairly balanced, following an almost stable development until the end of the simulation. It is
indicated with a dotted line while the solid line shows the base run (Figure 26, right chart). Since cycles in the
aging chain are almost balanced out, the model does not exhibit major lagging booms and bust any longer. This
25
behavior is transferred to the demand side accordingly with the result of the almost linear 'Supply demand ratio!
offsetting cyclical price movements as presented in Figure 27.
20,000 4
Desired new
construction Base
Desired new
wa, Buildings under eatuction
construction - Base Paley
Scenario2 ~,
(dal)
5 ildings under fa Scenario 1
2 reco Builingsunde EB,
= serene fae 2
2 Buildings é
@ 3000 a1
0 o =i = 2 wai
‘sro 1980 1980 —~200~—«7010~=«7020~«=030~=~«OO ‘s7a 1980390 200020102020 208 2040
Figure 26: Policy: Consideration of underway construction reducing cycles in the construction stock-and-flow
(left) as well as in ‘Supply demand ratio’ (right).
The effect of offset cycles in key variables as described above lead to smoothened amplitudes in price (Figure
27, dotted). Since supply is not over- and undershooting demand time after time, price reacts with likewise
balanced developments over the simulated time horizon.
300
Scenario 1
Base run
225 ;
Policy
3
z 150
g
a
75
historical
0
1970 1980 1990 2000 2010 2020 2030 2040
Figure 27: Policy: Consideration of underway construction affecting price cycles.
Given the current design of the STREM-model, this policy accomplishes smoother behavior with almost offset
cycles, on both demand and supply side. Hence, this experimental run indicates that policy design in general can
be leveraged far more on the system's dynamics. Therefore, the dominant feedback loops of the model should be
changed by "redesigning the stock and flow structure, eliminating time delays, changing the flow and quality of
information available at key decision points, or fundamentally reinventing the decision processes of the actors in
the system" (Sterman, 2000, p. 104).
Discussion and Conclusions
Although often not i a dity, real estate belongs to those construction industries in
which long manufacturing times and asset lifetimes determine the market behavior. Similar to other markets,
e.g., copper, aluminum and coffee, and also aircraft and shipbuilding, real estate markets
in general exhibit cyclical dynamics (Sterman, 2000). However, the market’s complexity oftentimes prevents
decision-makers from capturing the underlying structure and thus understanding the origins of cyclical behavior.
such as raw material
The real estate market of the city of Stuttgart has seen an unprecedented increase in prices for the last decade.
Market experts expect prices to further increase. We have transferred findings from secondary data in the
26
literature and from conducting semi-structured interviews into a simulation model to gain further insights in how
market experts perceive the market and increase validity.
Yet, we acknowledge several limitati of our simul
model when ing the results of the paper. For
example, our model omits several structures that may be relevant, i.e., further aspects of the construction side
(e.g., capacities and available building land) or financial factors affecting demand, such as mortgages and loans
but also disposable income.
In addition, the model comprises weaknesses in numerical data and its equations. The utilized indices of price
and construction costs are valid for the Stuttgart real estate market, yet, they are not scaled to the same base
years. As a further limitation, we needed to estimate several model parameters, as secondary data was not
available for the specific case of Stuttgart.
Despite the limitations, our analysis reveals some valuable insights for decision makers and academics, which
in the following. First, the model reproduces the behavior of the reference modes, indicating a decent
fit between simulation and real world data on price, for example. The model structure might not be as
sophisticated as it could be. Yet, it succeeds in simplifying a complex system, and thus enabling an
understanding and further analysis of accordingly complex structures.
we di
Second, our analysis shows that cyclical dynamics do exist in the Stuttgart real estate market, like in many other
real estate markets that have been analyzed before. The market experts we interviewed acknowledge the
comparison with the hog cycle — but when asked about possible future price development, they seem to neglect
possible market cycles. Instead, they point to the cyclical behavior of interest rates as the driver for an oscillating
real estate market: According to their mental model, the real estate market oscillates because interest rates go up
and down. Yet, our analysis reveals that the origins of oscillating behavior are created endogenously. Housing
prices react to changes in related variables, i.e., the balance between demand and supply. As both sides, demand
and supply, imply negative feedback loops characterized by significant time delays, the system is constantly
over- and undershooting its equilibrium. Thus, the system exhibits persistent cyclical instability. In particular,
the lagging supply of desired new construction leads to unsatisfied demand, hence increasing prices. When new
construction is completed, demand has decreased again as a response to the previously increased prices. The
subsequent supply surplus reduces prices and thus expected profits, so that construction activity is
lower in the
following period. It is the ongoing interplay of over- and undershooting variables results in the oscillating
behavior — which explains the hog cycle effect in the real estate market and not the interest rate’s ups and downs.
Third, the two scenarios provide additional insights into the system's behavior responding to exogenous changes
that are not influenced by the real estate market itself. The scenario in which we assume that interest rates will
increase again shows oscillating and increasing prices. In the second scenario we test, how the market will
behave when there is a period of economic downturn because of a change in the structure of the automotive
industry — which is discussed in the media but seems to be considered as an extreme scenario by many people.
Here, the market will go down — until finally demand and supply meet after a long period of excess demand.
Afterwards, prices will again increase. By designing a policy on how to manage the market differently, real
estate cycles could be almost eliminated. While cycles originating from the supply chain could be reduced
significantly, the policy design did not accomplish a similar effect on prices. Yet, the price level did not decrease
due to increasing construction costs accompanied by high profit margins — an assumption which could be
challenged in future research.
Fourth, Thornton (1992) as well as the personal interviews conducted show that real estate experts have
difficulties in ass
ing the impact of the endogenous feedback structure dominating real estate markets. This is
ascribed to both the complexity itself, but also to the physical time period between each cycle, so that many
have not yet experienced a plete cycle. Although all our interviewees are aware of cycles in the
real estate markets, they do not ascribe as much significance to endogenous causalities, rather, their focus
remains on macroeconomic factors, exogenous to the model.
27
Concluding, the paper has illustrated that endogenous structures play the critical role in dynamic system, and
thus should not be underestimated. With this analysis approach, difficulties in human information processing can
be overcome, resulting in more effective decision-making in any business environment.
Although the paper presents various relevant aspects of the real estate market, future research suggestions
involve a further extension of the model to a more sophisticated structure. Therefore, the model boundary could
be expanded by additional structures, i.c., capturing more accurately demand creation, capacity constraints and
interest rates for demand. Also, model equations could be redefined. In particular, the model's parameters as well
as the table functions could be tested and analyzed further.
The future outlook of the paper also reflects aspects of the problem statement that have been completely left out.
Considering the current situation with regard to concerning migration, the tremendous rise in the number of
refugee seeking asylum in the European Union boosts housing demand. Thus, the resulting increased housing
needs must be reassessed for the upcoming years. Appropriate policy design and effective decision-making today
will pay off in the near future.
List of References
Afhiippe, S., Fasse, M., & Murphy, M. (2016, October 31). "Wir betreten Neuland". Handelsblatt, pp. 1-4.
Atefi, Y., Minooei, F., & Darghi, R. (2010). Experimentation i in Learning Organizations; 1-21. Retrieved August
10, 2015, from http://www. 2010/p d/papers/P1115.pdf,
Barlas, Y., Ozbas, B., & Ozgiin, O. (2007). Modeling of real estate price oscillations in Istanbul. Presented at the
25th International Conference of the System Dynamics Society, Boston, MA. Retrieved August 10, 2015,
from http: 2007 /proceed/papers/BARLA342.pdf
Buchenau, M.-W. (2015, June 25). Stuttgart - Druck auf dem Talkessel. Handelsblatt, pp. 30-31.
Buchenau, M.-W. (2016, July 1). Ausweg Hochhaus. Handelsblatt, pp. 46-47.
Creswell, J. W. (2003). Research design: Qualitative, quantitative, and mixed methods approaches (2nd ed.).
Thousand Oaks, CA: Sage Publications.
Dalcomo, I. (2016, September 11). Das Angebot bleibt knapp. Stuttgarter Nachrichten.
Denzin, N. K. (2012). Triangulation 2.0. Journal of Mixed Methods Research, 6(2), 80-88.
http://doi.org/10.1177/15586898 12437186
Deutsche Bundesbank. (2013). Monthly Report October 2013 (Vol. 65, pp. 1-82). Retrieved December 19, 2015,
from https://www.bundesbank.de/Redaktion/EN/Downloads/Publications/
Monthly _Report/2013/2013_10_monthly_report.pdf?__blob=publicationFile
Deutsche Bundesbank. (201 4a). Mi ‘icht Februar 2014. Deutsche Bundesbank, 2, 1-166. Retrieved
December 19, 2015, from https://www. de/Redaktion/DE/D loads/Veroeffentli
Monatsberichte/2014/2014_02_monatsbericht.html
Deutsche Bundesbank. (2014b). Monthly Report February 2014. Deutsche Bundesbank (Vol. 66, pp. 1-82).
Retrieved December 19, 2015, from
http://www.bundesbank. de/S iteG lobals/Hotms/Arehiv/EN/Bublications/ Publications Formular. html?dateOf
31.12.2014. iff 11336. &inpu 1720&seare
hIssued.HASH=9a8a78268ba61 40ae 1 56&cl2Categories_ Typ.GROUP=1 &pageLocale=de&searchIssued=0
&dateOflssueA fter=01.01.2014&cl2Categories_Themen.GROUP=1 >p=29412_list%253D2&cl2Categori
es_Typ=Monatsbericht
DiPasquale, D., & Wheaton, W. C. (1992). The markets for real estate assets and space: a conceptual framework.
Journal of the American Real Estate and Urban Economics Association, 20(1), 181-197.
Eskinasi, M. (2014). Towards housing system dynamics. Eburon Academic Publishers, Delft, Netherlands.
Eskinasi, M., Rouwette, E., & Vennix, J. (2009). Simulating urban transformation in Haaglanden, the
Netherlands. System Dynamics Review, 25(3), 182-206. http://doi.org/10.1002/sdr.423
Forrester, J. W. (1961). Industrial Dynamics. Cambridge, MA: MIT Press.
Genta, P. J. (1989). Understanding the Boston Real Estate market: a System Dynamics approach. (Master
thesis). Mi I Institute of Technology. Boston, USA. Retrieved August 10, 2015, from
https://dspace.mit.edu/bitstream/handle/1721.1/14312/21719460-MIT.pdf?sequence=2
28
Gutachterausschuss fiir die Ermittlung von Grundstiickswerten in Stuttgart (Ed.). (1995). Der Grundstiicksmarkt
in Stuttgart - Jahresbericht 1995 (pp. 1-39). Stuttgart, Germany.
Gutachterausschuss fiir die Ermittlung von Grundstiickswerten in Stuttgart (Ed.). (2001). Jahresbericht 2001
zum Grundstiicksmarkt (pp. 1-54). Stuttgart, Germany.
Gutachterausschuss fiir die Ermittlung von Grundstiickswerten in Stuttgart (Ed.). (2011).
ticksmarktbericht 2011 (pp. 1-76). Stuttgart, Germany.
c| fiir die Ermittlung von Grundstiickswerten in Stuttgart (Ed.). (2015).
Grundstiicksmarktbericht 2015 (pp. 1-72). Stuttgart, Germany.
Gutachterausschuss fiir die Ermittlung von Grundstiickswerten in Stuttgart (Ed.). (2016).
Grundstiicksmarktbericht 2016 (pp. 1-48). Stuttgart, Germany.
Haar, M. (2017, April 26). Preisanstieg drangt Stuttgarter raus. Stuttgarter Zeitung. Retrieved May 25, 2017
from from http: Stutt; itung.de/inhalt.i kt-stuttgart-prei
stuttgarter-raus. 0202a192- Tale- 47ec-86f0-b78392c2a053.html
Hahn, S. (2016, May 17). Immobilien in Stuttgart: Der Ansturm auf Immobilien ist lirigebrochen Stuttgarter
Zeitung. iat March 23, 52017 from http://www.stuttgarter-zeitung.de/inhalt.i t in-stuttgart-
der-anstt hen.c16c6628-f28¢e-4d25-a8bf-dce020b94f6d3.html
Heilweck-Backes, I., & StrauB, M. @0r: Wohnungsmarkt Stuttgart - Ergebnisse der
Wohnungsmarktbefragung 2006. Statistik und Infe
Retrieved November 12, 2015, from http://service.stuttgart.de/lhs-
251_1 t_Stuttgart_2006.PDF
Hu, G G. & Lo, S. H. (1992). Understanding cyclical pattern of Taiwan's housing market: A system dynamics
approach. Presented at the 10th International Conference of the System Dynamics Society, Utrecht,
Netherlands. Retrieved August 12, 2015, from
http://www. 1992/p: d/pdfs/hu247.pdf
Kapmeier, F., Tilebein, M., Voigt, A., & Dillerup, R. 201 1). Applying system dynamics to overcome
unsuccessful su ctor research. Presented at the 29th International Conference of the System
Dynamics Society, Washington, DC. Retrieved August 21, , 2015, from
http://www. 201 1/proceed/papers/P1330.pdf,
Luna-Reyes, L. F., & Andersen, D. L. (2004). Collecting and analyzing qualitative data for system dynamics:
methods and models. System Dynamics Review, 19(4), 271-296. http://doi.org/10.1002/sdr.280
Lyneis, J. M. (2000). System dynamics for market forecasting and structural analysis, System Dynamics Review,
16(1), 3-25. doi: 10.1002/(SICI)1099-1727(200021)16:13.0.CO;2-5
Mankiw, N. G. (2010). Macroeconomics (7 ed.). New York, NY: Worth Publisher.
Mashayekhi, A. N., Ghili, S., & Pourhabib, A. (2009). Real estate cycles: A theory based on stock-flow structure
of durable goods markets. Presented at the 27th International Conference of the System Dynamics Society,
Albuquerque, NM. Retrieved August 10, 2015, from
http://www. 2009/p {/papers/P1243.pdf
Morecroft, J. D. W. (1985). Rationality in the analysis of behavioral simulation models. Management Science,
31(7), 900-916.
Morecroft, J. D. W. (1988). System s and mi Ids for poli . European Journal of
Operational Research, 35(3), 301-320. http: //doi.org/10. 1016/0377- -2217(88)90221-4
Muth, R. F. (1988). Housing market dynamics. Regional Science and Urban Economies, 18(3), 345-356.
Osterreichische Nationalbank. (2017). Entwicklung des Kapi in Di in den Jahren von
1975 bis 2016. In Statista. Retrieved March 15, 201 q from http://de. satis com/statistik/daten/studie/
20141 gi ick] des
Ozbas, B., Ozgiin, O., & Barlas, Y. (2008). Sensitivity anal: of a real estate price oscillations model.
Presented at the 26th International Conference of the System Dynamics Society, Athens: Greece. Retrieved
August 10, 2015, from http://www. 2008/p d/papers/OZBA113.pdf
Patton, M. Q. (1999). Enhancing the quality and credibility of qualitative analysis. Health Services Research,
34(5 Part IT), 1189-1208.
Pyhrrn, S. A., Roulac, S. E., & Born, W. L. (1999). Real estate cycles and their strategic implications for
investors and portfolio managers in the global economy. Journal of Real Estate Research, 18(1), 7-68.
Rahmandad, H., & Sterman, J. D. (2012). Reporting guidelines for simulation-based research in social sciences,
System Dynamics Review, 28(4), 396-411. http://doi.org/10.1002/sdr.1481
Real Estate Advisor 1. (2015, September 23). Interview by Eleftheria Kapourani [Audio Recording]. Copy in
possession of author.
Real owaee Advisor 2. (2015, September 23). Interview by Eleftheria Kapourani [Audio Recording]. Copy in
ssion of author.
Real Estate Director. (2015, September 23). Interview by Eleftheria Kapourani [Audio Recording]. Copy in
possession of author.
Real Estate General Manager. (2015, September 23). Interview by Eleftheria Kapourani [Audio Recording].
Copy in possession of author.
5/2007, 101-152.
29
Reichel, R. (2014, June 25). Ende des Booms. Handelsblatt, pp. 30-31.
Reichel, R. (2017, January 6). Stadt, Land — alles im Fluss. Handelsblatt, pp. 48-51.
Schmitz-Veltin, A. (2009). Einwohnerprognose 2009 bis 2025 - Die Entwicklung der Zahl der Einwohner in
Stuttgart. Statistik Und h Me heft 11/2009, 324-344. Retrieved November 12,
2015, from http: rvice.stuttgart.de/lhs-
services/komunis/documents/8509_1_Einwohnerprognose_2009_bis_2025__Die_Entwicklung_der_Zahl_
der_Einwohner_in_Stuttgart.PDF
Statistisches Amt Stuttgart. (2015a). Gebaude- und Wohnungsbestand und Indikatoren zur
Wohnraumversorgung in Stuttgart seit 1950 [Komunistabelle: 193]. Retrieved November 15, 2015, from.
http://statistik | .stuttgart.de/statistiken/tabellen/193/jb193.php
Statistisches Amt Stuttgart. (2015b). Haushalte in Stuttgart seit 1992 nach der Zahl der Personen
[Komunistabelle: 4699]. Retrieved November 15, 2015, from
http://statistik1 .stuttgart.de: iken/tabellen/4699/jb4699.php
Statistisches Landesamt Baden-Wiirttemberg. (2014a). Bauferti gen im is Stuttgart.
Retrieved November 15, 2015, from http://www.statistik.baden-
wuerttemberg.de/SRDB/Tabelle.asp?H=ProdGew&U=01 &T=07015111&E=KR&R=KR111#doc
Statistisches Landesamt Baden-Wiirttemberg. (2014b). Baugenehmigungen im Wohnbau Stadtkreis Stuttgart.
Retrieved November 15, 2015, from http://www.-statistik.baden-
wuerttemberg.de/SRDB/Tabelle.asp?H=! ProdGew&U= 01&T=07015011&E=l KR&R= KRI i 1
Statistisches Landesamt Baden-Wiirttemberg. (2014c). Pri sowie
Stuttgart, L dt. Retrieved 15, 2015, from http://www.statistik.baden-
wuerttemberg.de/SRDB/Tabelle.asp?H=1 &U=07&T=99025080&E=GE&K=111&R=GE111000
Statistisches Landesamt Baden-Wiirttemberg. (2014d). Wohngebaude, Wohnungen nach Anzahl der Raume im
Stadtkreis Stuttgart. Retrieved November 15, 2015, from http://www.-statistik.baden-
wuerttemberg.de/SRDB/Tabelle.asp?H=ProdGew&U=05&T=07055011 &E=KR&R=KRI111
Statistisches Landesamt Baden-Wiirttemberg. (2016, March 16). Regierungsbezirk Stuttgart zahit zu den 15
fiihrenden EU-Regionen [Press release]. Retrieved May 12, 2017, from https://www.statistik-
bw.de/Presse/Pressemitteilungen/2016066
Statistisches Landesamt Baden-Wiirttemberg. (2017). Gebiet, Bevélkerung und Bevélkerungsdichte Stuttgart,
Landeshauptstadt. Retrieved February 5, 2017, from http://www.statistik-
bw.de/BevoelkGebiet/GebietFlaeche/01515020.tab?R=GS 111000
Sterman, J. D. (1987). Testing behavioral simulation models by direct experiment. Management Science, 33(12),
1572-1592.
Sterman, J. D. (2000). Business dynamics: Systems thinking and modeling for a complex world. Boston, MA:
McGraw-Hill.
The Economist Data Team. (2016, March a1), Location, location, location: global house ices. Retrieved March
20, 2017, from http://www. S ychart/2011/11/global-h p
The Economist Newspaper Limited (Ed.). (2017, March 18). As the Fed raises rates, Tenet Yellen’s ene is
pondered. The Economist. Retrieved March 20, 2017, trom, http: s-and-
finance/21718857-donald-trump-has-ch entral-bank-fed S
Thornton, L. (1992, Sep ), Real estate develo virms as learning organizations: Systems thinking as a
methodology for strategic planning. Massachusetts Institute of Technology, Cambrid;
Wheaton, W. C. (1999). Real estate "cycles": Some fundamentals. Real Estate Economics, 272), 209- 230.
Wiebe, F. (2015, December 17). Die groBe Wende in der Geldpolitik. Handelsblatt, p. 28.
Yin, R. K. (2003). Case study research: design and methods (3rd ed.). Thousand Oaks, CA: Sage.
30
Appendix
Appendix A. Full Model and Legend of Variable Types
=
"sg gee
spe UoUNSSY 105
> vopomsueg aiming <—
Baud paseaey Rud Pavey
oglu,
ferne|
a xepuy
1s09 vonomuisuog
\
‘51809
<ous| sis09 poratig
“> vonansuog are
4
‘s09 Uonon,feuo
evo oahu
weve
oney oaug uBnewen aod "
01 99H pEIDog ¥
uewog Uo oor i
20 paidoooy +
1p 9H 049101 20 <= inst
pid Pavodea
>< wwavemms ssp vOTEMTOS
24 0 puewna ‘oput 93h4 721000 Ni
AMIENS ‘viva $1809,
aia uononsuo
aoe + Lal oney Rear oe
pucuog isn a yes Pe
& ssolong aiasy ers ‘) smteinassog Aang jo;09)13
roa 0 Amory puetog ot ™F ccs uo ones
j ‘toa puewog ins +
so eIr370} 9198.
sang 0 hdang sto no
p> parsed vo tena
eens: payedsg Jo 198))3
_ nef
shedea 1en1e1ee— oe ot pu aang
A sss
«v0.9 HH + cottons
aby putlag
WS 01 BU joyasno} aa
oom 098 nerg Gy
+ sstitedy
we veins
poe 2 N01
10N spjoyosnoH uonannsudy
wud ey ywoD spun
yeuori2e14 proyasnoy uuu phonsueg
siping
onounesy. yg partnao0 fore
ee ae one d
ployasnoy eumng us i,
saupind
many ur pueweg y
_ se Sgpnsog ee
Figure Al: Full STREM-model.
Time to Realize Effect of ‘Smoothed Effect of ‘Smoothed Effect of
Interest Rate on Profit ——®= Interest Rate on Imarest rate On ag Time'te Realize Effedt of
Interest Rate on Demand
Profit Margin
+
+
Table for Effect of Effect of Interest Effect of Interest Table for Effect of
Interest Rate on Profit gs Rate on Profit Margin Rate on Demand “—— Interest Rate on
Margin 4 ae Demand
Interest Rate
—— a
<Time>
oF
Interest Rate DATA Future Interest
wee Aesuaneten ge
SWITCH Interest
Rate Scenario 1
Future Interest Rate Future Interest Rate
‘Assumption BAU Assumption Scenario 1
Figure A2: Substructure of STREM-model: Interest rate effects on supply and demand side.
Color Variable Type
| Endogenous variables, determined by the system's behavior as well as table
functions of effects
Exogenous input added (e.g., real data of times series and initial values)
Constant parameters implicating information delays (e.g., time adjustments)
Constant parameters implicating sensitivity of variables
Shadow variables: defined elsewhere and used to avoid clutter and overlapping of
structures
oO Variables i ing future ions and
it) Real data, which simulated variables are compared to
Table Al: Legend of STREM-model.
Appendix B. General Simulation Settings
The model was implemented in Vensim PLE for Macintosh, Version 6.3, and Vensim DSS for Windows
Version 6.3 Double Precision (x32). Model settings are described below (Table A2):
Initial Time 1970
Final Time 2045
Time Step 0.125
Units for Time Year
Integration Type Euler
Table A2: Model settings in Vensim.
The experimental runs were executed in Vensim PLE for Macintosh, Version 6.3 with a MacBook Air and in
Vensim 6.3 Double Precision with a ThinkPad. Exogenous data was imported through vdf-files and respective
variables are indicated in red and with the suffix 'DATA' in STREM-model (Figure A1).
Appendix C. Experimental Runs Settings
The settings described in Table A3 were used to execute all runs — base, scenarios and policy runs.
Base Run: Scenario 1: ; ; Policy: Consideration of
i Scenario 2: Decline in :
Business as Increasing Underway Construction
Households
usual Interest Rates
~ Scenario 1, plus Policy
Interest rate Scenario 1, plus simulation of eae y
Design: Supply side
increases again a collapsing regional
considers underway-
assuming a automotive industry leading
Description construction, thus being able
step-by-step toa sharp decrease in
to estimate more accurately
desired new additions in
order to meet excess demand.
increase by the demand, i.e. growth rate of
households declines.
SWITCH "Interest Rate Scena
= Increase in interest rates as of 2017 i i
1 = Interest rates remain as low as of
2015 at 0.005
SWITCH ‘Scenario 2"
0 = Future Household Fractional Growth
Rate assumed constant at 0.004
1 = Sharp decline in households between
2017-2021, followed by constant
Fractional Growth Rate at -0.005
Table A3: Run settings.
Appendix D. Equations
Equations and Comments Units
(On) ‘Accepted Profit Margin= Normal Profit Margin*Smoothed Effect of Interest Rate on Profit Margin Dimensionless (Dmnl)
‘The accepted profit margin is determined by the annual interest rate level. Therefore, the effect variable adjusts the normal profit margin as follows: Supply side
accepts lower profit margins when interest rates are respectively low. They claim higher profit margins during periods of high interest rates. Eventually, the accepted
profit margin determines the minimum acceptable price by taking into account the construction costs.
(2) ‘Apartments per Building=4 apartments/building
In average each building is constructed in such way as to yield four apartments, This average number is based on real data of the real estate market Stuttgart. Source:
hup://service stuttgart.de/ths-s 7251_1\ Stuttgart_2006.PDF (retrieved October 29, 2015)
(03) ents pet Household=1 ‘apartments/household
Tris assumed that each household ‘occupies one apartment. This variable helps to equal units, i.e. 1 household = 1 apartment.
(4 “Attractivity of Real Estate Purchase=SMOOTH( Smoothed Effect of Interest rate on Demand* Effect of Price on Demand, Dial
Time to Smooth Attractivity trom Price Effect )
vane expresses the Attractivity of purchasing real estate by taking into account the effect of changing prices as well as interest rates on demand.
75
‘Average Construction Time=I Year
he average time to construct a building.
‘Average Life Time of Buildings=100 Year
The average life time of buildings until the building is fully demolished.
) ‘Average Planning Time=4 Year
‘The average time to plan the construction of a new building before the actual construction start
‘Average Sales Time=0.75 Year
‘The average time it takes to sell a building
09) Buildings Completed= INTEG (Construction Completion Rate-Sales Rate, IN Buildings Completed) buildings
The stock accumulates the fully constructed buildings.
(10) Buildings Completed DATA :INTERPOLATE: buildings
Jata was inserted into model for all buildings completed retrieved November 2015 from http://www.statistik.baden-
wuerttemberg.de/SRDB/Tabelle.asp?H=ProdGew&U=01&T=07015011&E-KR&R=KRI 11
i) Buildings in planning DATA -INTERPOLATE: buildings
Data was inserted into model for all buildings in planning retrieved November 2015 from hp! /Iwoww statistik baden-
wuerttemberg.de/SRDB/Tabelle.asp?H=ProdGew&U=01&T=07015011&E=KR&R=KRI1
(2 Buildings Occupied= INTEG (Sales Rate-Demolition Rate, IN Buildings Occup D buildings
‘The stock accumulates all buildings sold and thus occupied. The buildings being demolished are deducted through the outflow Demolition Rate
(3) Buildings Occupied DATA-INTERPOLATE: buildings
Data was inserted into model for all buildings in planning retrieved November 2015 from stuttgart /jb193.php
(5) Buildings under Construction= INTEG (Construction Start Rate-Construction Completion Rate,IN Buildings under buildings
rruction)
The stock of building that are constructed.
(15) Change in Expected Construction Costs=(Construction Costs-Expected Construction Costs)/Time to Adjust Expected Costs price index/Year
The flow adjusts the stock of Expected Construction Costs based on the input of the average construction costs per unit.
16) ‘Change in Expected Price=Price Variation/Time to Adjust Expected Price price index/Year
This inflow changes the expected price in response to the gap between the indicated price and the current expected price.
17) Construction Completion Rate=Buildings under Construction/A erage Construction Time bull ngs Year
This variable transforms the stock of buildings under construction into the stock of completed buildings. The higher the value the faster are buildings constructed.
(8) Constretion Cost Index DATA (197002016200) (1970 31.2)(1971342(1972.364, (197339. (1974, O9TS 412), price index
3), (1977,44.2), (1978.47.3),(1979,51.8),1980,57.5), (1981,60.2) 7)(1983,61.5), (1984,63.1),
{U98r55 1) 1088 66.9; 980.682 (1980-73 9) U99I-7E (105% a2 a ioH4es) (1904 ¥S 4) (998863) O86 HA 8 OSTEO,
(1998,84),(1999,84,2),(2000,85.2), 2001.85 9), (2002,86.1), (2003,85.6), (2004,86.7), (2005,87.3), (2006,89.4), (2007,95.7), (2008,98.5),
(2009,99.1), 2010, 100),201 1,103), (2012,105.5), 2013,107.4),(2014, 109.6), (2015,111.9),2016,114.5))
is table contains the index of construction costs for the federal state of Baden-Wiirttemberg for 1970-2016, retrieved from: hhttp://www.statistik.baden-
‘wWuerttemberg.de/GesamtwBranchen/K onjunktPreise/BPI_LR jsp (February 28, 2017) / Banana! BW --- After 2017: Data is calculated based on average growth
rate of previous 10 years (1996-2016) and is returned through the variable "Future Constru Assumptions’
33
0) Construction Cost
Assumption)
This is the exogenous input of Construction Costs per apartment indicated as an index. For 1970-2016 real data is returned via the variable "construction Cost Index
DATA", whereas, after 2017, Data is calculated based on average growth rate of previous 10 years (1996-2016) and is returned through the variable "Future
Construction Costs Assumptions"
IF THEN ELSE( Time<2017 , Construction Cost Index DATA(Time) , Future Construction Cost price index
(20) ee Costs DATA INTERPOLATE: price index
Dat ul ie ay del fo ne of corto i Ti Foil ado de W Ueno Wr 19702016 eye i
hhtip://wvww.statistik.bad e/BPL_LR jsp (February 28, 2017)
en Construction Start Rate=MAX< (0,Desired New Construction/Average Tae Time) buildings/Vear
‘The Construction Start Rate is the first inflow into the supply aging chain and takes into account the amount of desired new construction projects. The MAX function
usts the inflow not to become negative at any point of time.
i
(22) Delay Time in Demand Creation=1 ‘Year
This is the time people need to create actual demand in housing, meaning that people do not react immediately to changes in price, rather it takes time to decide to
move into a new apartment,
(23),
Demand= INTEG (Demand Increase Rate-Demand Satisfaction Rate, IN Demand) ‘apartments
This is the stock that accumulates demand in housing, i.e. how many buildings are demanded in total in a certain point of time.
(24) Demand in Buildings=IF THEN ELSE (D. per Building>0, ‘per Building, O) buildings
Demand of apartments transformed into number of buildings demanded
(25) increase Rate=SMOOTH(Attractivity of Real Estate Purchase*Potential Demand* Apartments per Household , apartments/Year
Delay Time in Demand Creation)
This is the net growth rate of demand. When positive, demand is created, otherwise, when negative demand is satisfied. The smooth function reflects the delay in
demand formation, since people do not react to price changes immediately.
(26) Demand Satisfaction Rate=Sales Rate* Apartments per Building apartments/Vear
‘The Demand Satisfaction Rate decreases the stock Demand along with the Sales Rate, i.e. when an apartment is sold/occupied.
en lished Space=Demolition Rate* Apartments per Building apartments/Year
‘This variable caleulates the total demolished space in units of apartments. Demolished space results in new demand for apartments.
(28) Demolition Rate=Buildings Occupied/Average Life Time of Buildings buildings/Vear
The outflow adjusts the stock Buildings Occupied since buildings need to be demolished after the average life time of a building.
29) Desired New Construction=Expected Excess Demand in Buildings*Effect of Expected Profitability on Desired New buildings
‘onstruction
Desired New Construction to satisfy Expected Excess Demand is supplied depending on Profitability High Profitability = high supply of new construction Low
profitability = decreases supply Equation for Policy 1: (Expected Excess Demand in Buildings-Buildings under Construction-Buildings Completed) *Effect of
Expected Profitability on Desired New Construction
G0) Effect of Expected Profitability on Desired New Construction=Table for Effect of Profit on Desired New Dial
mnstruction(Expected Profit)’Sensitivity of Supply to Price
Desired capacity is adjusted above or below current capacity in response to the expected profitability of new investment
GD Effect of Interest Rate on Demand=Table for Effect of Interest Rate on Demand( Interest Rate* 100) Dial
This is the effect that interest rates have on the demand. Low interest rates make Real Estate an attractive investment opportunity, thereby increasing the attractiveness
of purchasing Real Estate. Whereas, the higher interest rates increase the attractiveness of alternative investment options for potential real estate purchasers, such as
financial assets, savings on bank account.
fect of Interest Rate on Profit Margin=Table for Effect of Interest Rate on Profit Margin(Interest Rate* 100) Dmal
(2)
This is the effect that interest rates have on developer's Profit Margins. Low interest rates reduce profit margins, higher interest rates result in higher profit margins,
@3) ae ‘of Price on Demand=Table for Effect of Price on Demand(Expected Price to Price Ratio) Sensitivity of Demand to mn
Prict
Expected Price higher than current rice > Demand decreases
Expected Price lower than current Price --> Demand increases
G4) Effect of Supply Demand Ratio on Price=Table for Effect of Supply Demand Ratio on Price(Supply Demand Dial
Ratio)Sensitivity of Price to Supply Demand Ratio
The effect of the supply demand ratio on price is a power function of the demand/supply ratio. The Sensitivity of Price to the demand coverage controls the magnitude
of the response. The higher the sensitivity of price to the demand/supply ratio, the greater the change in price induced by any imbalance. Price rises when
demand/supply ratio is less than normal, and falls when it is greater.
35) Excess Demand=MAX(0 , Demand-Vacant Apartments ) ‘apartments
This is the gap in housing, i. the discrepancy between the supplied space (the stock ‘Buildings Completed’) and the demanded space (the stock ‘Demand!). It is a goal
seeking function. If positive, it is the excess of space demanded over space supplied. If negative, it is the excess in supply over demani
) Expected Construction Costs= INTEG (Change in Expected Construction Costs, IN Expected Construction Costs) price index
Expected Construction Costs represent beliefs among market participants about the unit costs of production (variable and fixed, including normal profit margins)
Expected Costs therefore represent beliefs about what a 'fair' price would be, or the long-run equilibrium price. Expected costs adjust to the actual costs with a delay
representing the time required for gain information and adjust beliefs about costs. Exogenous for partial model test
on) Expected Excess Demand= INTEG (Expected Excess Demand Net Growth Rate, IN Expected Excess Demand) ‘apartments
It isa first order information delay stock that represents the supply side’s expectation of real desired sp:
G8) ted Excess Demand in Buildings=Expected Excess Demand/Apartments per Buildin, buildings
The value sim the value of the stock Expected Excess Demand in such a way as to yield the unit of buildings
9) ted Excess Demand Net Growth Rate=(Excess Demand-Expected Excess Demand)/Time to Form Expectation of apartments/Year
Enoes Demund
It takes into account both the vacant space (=the gap between supplied and demanded space) and the demolished space. Since it cannot be accurately known, it is,
estimated by the supply side through an information delay structure,
(40) Expected Price= INTEG (Change in Expected Price, IN Expected Price) price index
‘The price market makers and traders believe would clear the market if demand and supply were in balance, and no other pressures to change price existed.
ai Expected Price to Price Ratio=ZIDZ(Expected Price, Price) Dial
When expected price increases the effect on demand shall decrease. When expected price decreases the effect on demand shall increase, having a positive effect on
demand,
(42) Expected Profit=(Expected Price-Expected Construction Costs)/Expected Price Dmnl
This is the expected profit of the supply side based on the expected price and expected costs,
(@) FINAL TIME = 2045 Year
‘The final time for the simulation.
(44) Future Construction Cost Assumption=IF THEN ELSE (SWITCH Scenario 2=0, Future Construction Cost Assumption price index
Scenario 1 (Time), Future Construction Cost Assumption Scenario 2 (Time))
The variable delivers the construction cots price index under each specified scenario.
Fur Construstion Cost Assumption Senaro (2017102045200 201711609 201817 5), 219,182) (200 20.70, price index
(ama 12231), 2022125 94, (03,125 '27.25),2025,128.95)(2026,130.66), (2027,132.4),(2028,134.16), 229,135.94),
37.73), 031.139.5208, 1413), (2085145 31) (2084 148.22) 2085 14918), 2086, 491
137,151.09 53.1), (2039, 155.13), (2040, 157.19) 2041, 159.28), (2042,161.), (2043,163.55), (2044, 165.72),(2045,167.92))
This i indicates the assungiion for future construction costs for the period from 2017 until the end of the simulation run (204s) for the Base Run, Scenario 1
oli uming "Business as usual" - i.e. Construction costs increase with a rate of the last 10
34
75) Tass Construction Cost Assumption Scenario 2 (((2017,100)(2045 200)),2017,116.02),2018,117.56), (209,119.13), (2020,120.7), price index
(2021 22.31, 2022,120.7, 2023 120.7,(2025, 120.7 (2025.119-7 (2026119. (20271192), 20281183), 2029118), 2031175),
(2035,115), (2086, 114.5), (2037,114), (2038,113.5), (2039,113), (2040,112.5), (2041,112),
(2031117), (2032,116.5) 2034,
(2042 1152083 111 fo088 T10.9)2088 110)
This table indicates the assumption for future construction costs for the period from 2017 until the end of the simulation run (2045) in Scenario 2: assuming a drop of
construction costs along with decline in HH, due to weak economy in Stuttgart area,
47) Future Household Fractional Growth Rate Assumption=IF THEN ELSE (SWITCH Scenario 2=0, Future Household Year
Fractional Growth Rate Assumption Scenario 1(Time), Future Household Fractional Growth Rate Assumption Scenario 2
(Time)
This variable takes on values of variables Future Household Fractional Growth Rate Assumption’, that change in Scenario 1 and 2
(48) Future Houschold Eracional Growth Rate Assumption Scenario I (217 )2045.)1(2017.0.00), (2048 000,219.00), Year
(2020,0.004) 0.004), (2022,0.004), (2023,0.004),(2024,0.004), (2025,0,004), 2026, 0.004), (2027,0.004), (2028,0.004) (2029.0.004),
(2030.0.008, 2031.0 008) (20324008), 2033 0.008) (20340.04),2038,0.008) (2036.0 04), (20370008), (2038,0.008),
(2039,0.004),(2040,0,004), (2041,0.004),(2042,0.0037), (2083,0.004), (2044,0.004), (2045,0.004))
This table indicates the assumption for future household growth for the period from 2017 until the end of the simulation run (2045) in the Base Run, Scenario 1 and
Policy 1: assuming a Steady growth in the Households of Stuttgart.
(49)
Futre HoursboldFraetional Growth Rats Asstipion Scenario 2 (017-417 O045 05% GO17 AOE 2H18 0000), Year
0.001), (2019.97,-0.004), (202 20 3), (2024, 0.005), 2025,-.005),(2026,-0.005),(202
(0.005),2029,-0.005), 2030, $2 05).2038<0 008), (2084-0008), “ans, “us 99)20 0,005), (2038,
1039-0,005)(2040 -0.005), (2041 -0.005),(2042,-0.005),2042.-0,005) (20430, 005) (2044 -0.005)2048 0.005)
‘This table indiates the assumption for future household provrth forthe peviod ftom 2017 unt the end ofthe simulation run (2088) under SCENARIO 2(a Drop in
Households Growth Rate due to decline in market's economy).
(50) Future Interest Rate Assumption=IF THEN ELSE (SWITCH Interest Rate Scenario 1=0, Future Interest Rate Assumption Dial
BAU(Time), Future Interest Rate Assumption Scenario I(Time))
‘The variable delivers the interest rate assumption under each specified scenario
a) Future Interest Rate Assumption BAU(@017, 2017,0.005), (2018,0.005} 2020,0,005),2021,000% Dial
(2022,0.005), (2023,0.005), (2024,0.003),(2025,0.00: 2029,0.005), (
(2032,0.005), (2033,0.005), (2034,0.008), cn {2036.0.003(2039.0.00) (2038,0.05) 2089.00
(2042,0.005), (2043,0.005)(2044,0.005), (2045,0.005)
This table indicates the assumption for future interest rate cal for the period from 2017 until the end of the simulation run (2045) for the Base Run assuming
0.005),
(2040,0.008), (2041,0.005),
“Business as usual" - i.e, interest rates remain as low as of 2015 (as defined in the variable "Interest Rate DATA").
() ote Interest Rate Assvnpion Scenario 1 (017 145,0.08)],(2017,0.006), (201 7020,0.021 Dinnl
(2021,0.032), (2022,0.037), (2023,0.039),(2 0.055), (2028,0.057),
(2030.17,0.068019), (2032,0.071),(2035,0.075), (2037,0.0755),(2040,0.078),(2042,0.08),(2045,0.08))
This table indicates the assumption for future interest rate development for the period from 2017 until the end of the simulation run (2045)
In Scenario 1, 2 and Policy 1: an increase in interest rates as of 2017.
(53)
Household Fractional Growth Rate DATA ({((1970.0.03)-(2016,0.03)],(1970,0.0041), (1971,0.0041), (1972,0.0041), (1973,0.0041), Year
(48740004), 995,041, 19%60.004 (1977, 0.0041 1978 0.041)(1979.0.004), (19809041), (1981 0.081), (1982.8.0040,
(1983,0.0041), (1984,0.0041), (1985,0.0041), (1986,0.0041),(1987,0.0041), (1988,0.0049), (1989,0.0049),(1990,0.0049), (1991,0.0049),
{1992.0.0088) 1988 0.0049 (19940 0049){0998 6.0089) (1996.0.008), 19970 0008) (1998 0.0008) {1999 0.0008) (20000 005)
(2001,0.0043), (2002,0.0043),(2003,0.0043), (2004,0.0043), (2005, 0.0043),(2006,0.0064) (2007,0.0012), (2008,0.0069), (2009,-0.0008), (2010,-
(0.0217),(2011,0.0125),(2012,0.0189), (2013,0.0122),2014,0.0095),(2015,0.0158),(2016,0.0116))
This is the lookup of the fractional household growth rate in Stuttgart from 1970 until 2030 calculated on basis of the households data of Stuttgart. For missing data,
the annual compound method is used to calculate fractional growth rates between two given values. Source Households 1970-1995: http://www.statistik.baden-
‘wuerttemberg.de/SRDB/Tabelle.asp?H=1&U=07&T=99025080&E=GE&K=I 11&R= GE111000 (Retrieved October 3, 2015)
Source Households since 1995: http://statistik stuttgart. 199/jb4699 php (Retrieved February 27, 2016).
(4) Households in Stuttgart= INTE tous Net Growth Rate,IN Households) households
This stock accumulates all households in Stut
65) Households in Stuttgart DATA “NTERTOTATE households
Source Households 1970-1995: http://www.stati
(Retrieved October 3, 2015)
de/SRDB/Tabelle.asp?H=1 &U=07&T-990250808
SE&K=111&R= GE111000
(56) Hourchols Net Growth Rate=IF THEN ELSE Time=2015 , Household Fractional Growth Rate DATA(Time)*Households households/Year
Stuttgart , Future Household Fractional Growth Rate Assumption*Households in Stuttgart )
This flow shige the stock of households in Stuttgart.
G7)
Buildings Completed= INITIAL(S66) buildings
The stil value of Buildings completed is estimated based on the number of buildings existing in Stuttgart in 1970 and 1971. Data is retrieved from
http://www statistik baden-wuerttemberg.de/SRDB/Tabelle.asp?H=ProdGew&U=05&T=07055011&E=KR&R=KR 1 (November 14, 2015).
(58) IN Buildings Occupied= INITIAL(59036) buildings
‘The initial value, 59036 buildings in 1970, is retrieved from Stuttgart data. From stuttgart, /jb193.php (November 16, 2015).
(59) IN Buildings under Construction= INITIAL(570) buildings
The initial value of Buildings under Construction is estimated based on the number of buildings completed after two years (in 1972) considering the fact that the
constructions turn into completed buildings after 2 years on average. The data for buildings completed is retrieved from http://www. statistik.badet
wuerttemberg.de/SRDB/Tabelle asp? H=ProdGew8&U=05&T=07055011&E=KR&R=KR 11 (November 14, 2015)
(60) TN Demand= INITIAL(1000) ‘apartments
The initial value of demand is assumed at 1000.
(61) IN Expected Construction Costs= INITIALGI.2) price index
‘The initial value of the Construction Cost index for 1970 retrieved from http://www statistik.bad tIndex.asp
IN Eepeited tcisg Demin INETTAT3000) ‘apartments
The initial value of expected ex: mand is assumed at 3000.
(63) IN Expected Prices INIHIALGT) price index
The initial value of expected price is assumed at 37, close to actual price in beginning of time horizon.
(64) TN Households= INITIAL (264312) households
The initial value of households is retrieved from http://www.statistik.baden-wuerttemberg.de/SRDB/Tabelle.asp?H=1 &U=07 &T=99025080&E=GE&K=111&R=
GE111000 (Retrieved October 3, 2015)
INITIAL TIME = 1970 Year
‘The initial time for the simulation.
(66) Interest Rate=IF THEN ELSE( Time=2016 , Interest Rate DATA(Time) , Future Interest Rate Assumption) Dinnl
The function returns the values of real interest rate data for 1970 until 2015. After 2016 the Table for Interest Rate Assumption is returned.
(a) Tnterest Rate DATA ((1970,0)-(2015,0.2)), (1970,0.082), (1971,0.083),(1972,0.082)(1973,0.095){1974 0.106), (1975,0.0868), (1976,0.0804), Dini
(197700653), (1978,0.0613),(1979,0.0758),(1980,0.0843), (1981,0.1013),(1982,0.0894),(1983,0.0808),(1984,0.0798), (1985,0.0704),
{488600617 (1987.0.624, (198.0958) (19880070), (190 0.88, 1910.05), (982.0079), (995.0681, 1994.0. 068,
(4995,0.0685), (1996,0.0622),(1997,0.0564),(1998,0.0457), (1999,0.0449), (2000,0.0526).(2001,0.048), (2002,0.0478), (2003,0.0407),
{20030 080s) 200s 0.0838), 200600876), (2007400822) (20080398), 200800322) 2010100298), Oot l/ob2et, O12 001%,
2013,0.0157)(2014.0.0116), (2015 0.005), (2016,0.0009))
Exogenous input: interest rates for Germany from | od from http://de.statista di
Hipialinaiteacestzey ch euacitend/ “Dai fo we 1974 retrieved from hap: //www.digitalis.uni-koeln.de/Geldwesen/geldwesen279-284.pdf and
http://w 083
68) Minimum Accepted Price=Expected Costa Costs*(1-+Accepted Profit Margin) price index
This is the minimum price that the supply side accepts, based on expected costs plus an accepted profit margin.
(69) Normal Profit Margin=0.25 Dial
This is the normal profit margin for the construction side. It determines the accepted profit margin on basis of the annual interest rate level. See variable Accepted
Profit Margin.
35
(0) Potential Demand=SMOOTH( Households Net Growth Rate , Time to Smooth Demand from HH Growth)*(Demofished households/Year
Space/Apartments per Household)
Variable delivers potential demand calculated by the net growth in Houscholds as well as Demolished space
a) Potential Sales-MAX(0, MIN(Demand in Buildings, Buildings Completed )) idings
‘The MIN-Function returns the smaller value of either Demand or Buildings Completed. The function prevents Potential Sales, i.e. the number of buildings sold con
exceeding the existing demand in any point of time as naturally buildings can only be sold until total demand is satisfied.
(72) Price=Minimum Accepted Price*Effect of Supply Demand Ratio on Price price index
Trader's set prices by adjusting their belief about the underlying equilibrium price in response to market pressures such as the supply/demand balance, here represented
by inventory coverage relative to the normal level, and unit costs.
(3) Price Index Stuttgart DATA -INTERPOLATE: price index
Price index with 2010=100 retrieved from 2011, Stuttgart
(74) Price Variation=Price-Expected Price price index
The difference between Price and Expected Price adjusts the change in expected price.
(75) Sales Rate=Potential Sales/Average Sales Time buildings/Vear
The Sales Rate is the outflow that reduces the stock of Buildings completed. However, it is calculated on basis of Potential Sales
completed that can be sold must not exceed the given demand.
since the number of buildings
SAVEPER = TIME STEP Year
The frequency with which output is stored,
(a) Sensitivity of Demand to Price=0.5 Dial
This is the demand elasticity, which adjusts the effect ice on demand. Demand in real estate market is found to be rather inelastic (see Sterman, 2000; Muth,
1988). The lower the value, the less price sensitive is dem
(78) Sensitivity of Price to Supply Demand Ratio=0.75 Dinnl
Controls the response of price to the supply/demand coverage. Must be positive for high demand to lead to higher prices. Higher absolute values lead to greater price
changes for any given demand coverage
(79) Sensitivity of Supply to eal Dini
This is the supply elasticity, which adjusts the effect of price on supply. Supply in real estate market is found to be pretty elastic (DiPasquale,1999; Muth, 1988). The
higher the value, the more price sensitive is supply. The effect of elasticity is determined by this variable together with the lookup "Table for Effect of Profit on
Smoothed Effect of Interest rate on Demand=SMOOTH( Effect of Interest Rate on Demand , Time to realize effect of interest, Dmal
rate on demand)
This is the smoothed effect that interest rates have on the demand side. See also 'Table for Effect of Interest Rate on Demand’
(cH) Smoothed Effect of Interest Rate on Profit Margin=-SMOOTH( Effect of Interest Rate on Profit Margin , Time to realize Dial
effect of interest rate on profit margin)
This is the smoothed effect that interest rates have on developer's Profit Margins. Low interest rates reduce profit margins, higher interest rates result in higher profit
margins.
(82) Supply Demand Ratio=XIDZi( Buildings Completed, Demand in Buildings, 30) Dial
‘The supply demand ratio is the balance between demand and supply, expressed as a ratio. Supply is equal to the stock of buildings completed
SWITCH Interest Rate Scenario 1=0 Dini
This is a switch 0 = BAU 1 = Scenario 1, Scenario 2, Policy 1
SWITCH Scenario Dial
This is a switch 0= Base Run, Scenario 1, Policy 1 1 = Scenario 2
(85) ‘Table for Effet of Interest Rate on Demand (0.0}(1225}(03.18), (06354381495), (1297351 46667), (1.85936,45714) Dmal
(280424 41908) (37271-13743), (4.057031 276194 7169118099) 35234108571, (6) (668766089!
{Gast 40790475 48 2608301404579 (8969430.7087), (8.65993, 0 68971, 10587, 068714, 1 8ST S3HI8Sy(12.049,0615088)
Table determines the effect of interest rates on the attractiveness of purchasing real estate (i.e. demand) based on Germany's interest rates. Low interest rates generate
higher demand since investing into Real Estate seems as a better option, compared to high interest rates that moderate demand, since alternative investment options
become more attractive with higher interest rates (e.g. financial assets).
(86) Table for Effect of Interest Rate on Profit Margin (((0,0)-(13, (0.0571429)(1.29735, 7 Dinnl
(2.19756,0,0809524), (3.09776,0.12),(4.26273,0.209524),5,13646,0.31 9048) (5.85132,0.504762), (6.83096,0,795238), (7.96945,0.933333),
(©.29328,0.992857),11,1))
Table determines the acceptable profit margin based on Germany's interest rates. Low interest rat
requests a higher profit margi
decrease the profit margin acceptable, whereas higher interest rate
7) Table for Effent of Price on Demand ((0-(2),(0.03 0:25) (013482 024761903 027/05 0370.7 055400177 075280 Dinnl
(1,1,G.1,1.21), (.2,1.45), (1.33608, 1.70476),(1 491184762). (1.6,1.92),(1.8,1.98),
The S-shaped Table function adjusts demand as follows: When Expected Price increases -> effect decreases demand When expected price to price ratio decreases >
effect increases demand
(88) Table for Effect of Profit on Desired New Construction ([C05,0)(1,8)), -0.5,0:1),-0.2,0.15),0.13,0.26),-0.0997963,0.419048)(- Dini
(0.0448065,0.647619), (0.00101835,0.876191), (005,1.25), (0.098778, 1.67619), (0.15.21), (0.202648,2.81905),(0.25,3.3),
(0.294297,3.73333)40.35,3.9)40.404277.4),(0.45,4),0.54)1.4))
Table for Effect of Profit on Desired New Buildings. Depending on the expected profits, suppliers adjust the desired new constructions, with the assumptions that high
profits stimulate an increase in desired new while lower profits decrease the desired new
(89) Table for Effect of Supply Demand Ratio on Price ({(0.0)-6.2)|(0,1.49), (0.385321,1 42105), (0.7.1.28)(0.844037,1 14035)(1,1), Dini
(1.24771,0.815789), (1.72477 0.614035), (2.47706,0.535088), (3.37615,0.517544),(5,0.5),6,0.5))
When Supply > Demand = Effect on Price less 1 = decreases Price due to oversupply (low demand) When Supply < Demand = Effect on Price greater | = increases
rice due to supply shortagehigh demand, based on Barlas, 2007.
(0) TIME STEP Year
The time step for te si aoa
Time to Adjust Expected Costs=0.5 Year
The time to form expectations about the construction costs.
(2) Time to Adjust Expected Price=1.5 Year
The expected price adjusts to actual prices over this time period.
Time to Form Expectation of Excess Demand=2 Year
timated time that is needed in order to form the expectation of how much additional space is desired. It generates an information delay.
4
94) Time to realize effect of interest rate on deman Year
This is the time it takes until the effect of interest rates reach demand side.
(05) Time to realize effect of interest rate on profit margin=2 Year
This is the time it takes until the effect of interest rates reach supply side
96) Time to Smooth Attractivity from Price Effect=0.75 Year
This is the time that smooths Attractivity: Attractivity of houses resulting from price changes (Effect of price on demand) does not translate into demand immediately
but with the given time delay.
) ‘Smooth Demand from HH Growth=2.5 Year
‘his dly/ldne smooths tha amplitudes in Potential Demand
(98) Vacant Apartments=Buildings Completed* Apartments per Building ‘apartments
Variable calculates the number of vacant apartments based on buildings assumed being vacant once completed.
Table A4: Equations and comments of STREM-model.
36
Appendix E. Supporting Material
A Vensim file of the STREM-model is attached, including all experimental runs conducted and described in the
present paper. In addition, a vdf-file entails data time series that were used as real data input for validation
purposes.
Attachments:
e Vensim file of the "STREM-model", including
© Experimental runs:
Base Run: Business as Usual
Scenario 1: Increasing Interest Rates
Scenario 2: Decline in Households
Policy: Consideration of Underway Construction
37
