Search factory serm search engine reputatie management cc 03 handout
D Gimpelevich SERM Paper (JPIF)
-
Upload
dgimpelevich -
Category
Documents
-
view
47 -
download
2
Transcript of D Gimpelevich SERM Paper (JPIF)
Simulation-based excess returnmodel for real estate development
A practical Monte Carlo simulation-basedmethod for quantitative risk management andproject valuation for real estate developmentprojects illustrated with a high-rise office
development case study
David GimpelevichMid-Market Securities LLC, New York, New York, USA
Abstract
Purpose – An acute need exists for a practical quantitative risk management-based real estateinvestment underwriting methodology that clearly helps guide decision making and addresses theshortcomings of discounted cash flow (DCF) modeling by evaluating the full range of probableoutcomes. This paper seeks to address this issue.
Design/methodology/approach – The simulation-based excess return model (SERM) is an originalmethodology developed based on an application of Monte Carlo simulation to project risk assessmentcombined with the widely practiced DCF modeling. A case study is provided where results of themodeling are compared with traditional DCF risk models and with prior projects with knownoutcomes.
Findings – This paper lays out a practical method for stochastic quantitative risk managementmodeling for real estate development projects and illustrates that for identical projects risk-adjustedreturns derived with the use of SERM may differ significantly from returns provided by traditionaldiscounted cash flow analysis. SERM corrects serious shortcomings in the DCF methodology byincorporating stochastic tools for the measurement of the universe of outcomes. It further serves tocondense the results of Monte Carlo simulations into a simplified metric that can guide practitionersand which is easily communicational to decision makers for making project funding decisions.
Practical implications – SERM offers a simple, practical decision-making method for underwritingprojects that addresses the limitations of the prevailing methodologies via: stochastic assessment ofthe range of outcomes; interdependence of input variables; and objective risk premium metrics.
Originality/value – This paper presents an original methodology for making project-fundingdecisions for real estate development projects that is based on Monte Carlo simulation combined withDCF analysis. The methodology presented here will have value for real estate developers, investors,project underwriters, and lenders looking for a practical and objective method for project valuationand risk management than is offered by traditional DCF analysis. A review of literature did not revealanalogous methodologies for risk management.
Keywords Monte Carlo simulation, Underwriting, Real estate, Risk management, Modelling,Return on investment
Paper type Technical paper
I. IntroductionReal estate development is a risky endeavor. Some projects generate outsizedreturns while others languish or else entirely erase investment capital. It is also an
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1463-578X.htm
Simulation-basedexcess return
model
115
Received June 2010Accepted October 2010
Journal of Property Investment &Finance
Vol. 29 No. 2, 2011pp. 115-144
q Emerald Group Publishing Limited1463-578X
DOI 10.1108/14635781111112765
area of investment management that has historically seen slow adoption of financialrisk modeling by practitioners. Too often, developers operating with their ownfunds proceed with projects with little more than back-of-the-envelope calculationsbased on intuition and developers’ yield or return on cost metrics. Since institutionaland private equity investors have become more active in funding developmentprojects, many developers have adopted discounted cash flow (DCF) as theirprimary risk management methodology, often augmented by multi-point sensitivityanalysis. A few practitioners, recognizing the limitations of DCF, have exploredstochastic methods and real options in their search for more optimal quantitativerisk management methods. However, these methodologies are sufficiently complexin their applications to restrict their usefulness for a majority of practitioners.Furthermore, they often fail to provide a clear go/no-go decision signal whenemployed.
While a number of papers address stochastic methods in real estate asset valuationand non-real estate related project evaluation, research did not reveal any analogousquantitative risk management models or methodologies specifically for use in realestate development, nor did research reveal a useful method for applying the results ofMonte Carlo simulations to practical decision making.
In this paper we formulate a methodology for development of a stochastic modelthat represents a reasonable approximation of reality for development projects andcombining it with DCF in such a way as to generate a clear decision signal for thepractitioner. This paper will have value for real estate developers, investors, projectunderwriters, and lenders looking for a practical and objective method for projectvaluation and risk management than is offered by traditional DCF analysis. SERM isequally applicable to development of all product types; however, this paper onlyaddresses SERM as it applies to ground-up development of office projects.
Limitations of DCFThe limitations of the DCF method have been extensively documented. Baroni et al.(2006), as well as Atherton et al. (n.d.) and Young (2007), for instance, discuss thechallenges of proper identification of discount rates, future rent and expense levelsand terminal value estimates. Selection of these values can be in fact sufficientlyarbitrary as to make results of DCF analysis to be essentially meaningless. Manypractitioners, for instance, will estimate rents, expenses, and the exit capitalizationrate, with perhaps two or three sensitivity scenarios, and solve for the internal rateof return (IRR), looking for a set number as decision-making guideline, usually onthe order of 20 percent, derived from a wide-sample average. However, the selectedIRR hurdle often bears no relationship with the actual risks inherent in theparticular project under consideration. Risk depends on the variability of incomethat the developed property can be expected to generate and variability of themarket value at completion.
Too frequently the decision to proceed is based on nothing more than an intuitionthat the income necessary to clear the hurdle can be achieved at the time the project iscompleted, making it no more rigorous than a decision made on the basis of intuitionalone. Finally, even when input variables are rigorously selected, DCF can onlygenerate point estimates of returns, which, by ignoring the full range of probableoutcomes necessarily tend to distort the picture used for decision-making.
JPIF29,2
116
Prior applications of stochastic methodologies to real estate investmentsA number of papers have addressed application of Monte Carlo methods to real estateinvestments. Many have focused on the problem of asset valuation, including Baroniet al. (2006) and Young (2007). Atherton et al. (n.d.) examine, using a Monte Carlosimulation tool for development profitability analysis. In this paper we build on theirwork to formulate a decision-making methodology.
All of papers examined by the author use the mean valuation as the expected valueof an asset or a project and use probability measures to represent the risk. Knowing theprobability of achieving a certain outcome, however, is not sufficient to make adecision to proceed. What is needed is a way to weigh expected returns in the contextof specific project risk.
II. The simulation-based excess return modelExisting tools for evaluation of simulated results concentrate on the probability ofmissing expected returns. As such they provide valuable information, but the questionwhether this probability is excessive or acceptable remains with the practitioner’sjudgment, rendering an objective determination of project feasibility impossible. Whatis needed is an objective way to determine what return is sufficient to compensate forthe risk taken in undertaking a project. Risk is generally defined in terms of volatilityof returns. Finance professionals have long used the Sharpe ratio in order to measurerelative performance of portfolios with the returns of a market benchmark. The Sharperatio, however, suffers from two significant limitations, one of which is general and theother specific to the practice of real estate investment. The first limitation is the fact theSharpe ratio defines volatility in terms of the standard deviation of returns, whichincludes both the undesirable, downside volatility as well as the desirable, upsidevolatility. As such, it produces a distorted metric. The second limitation is that theSharpe ratio measures relative volatility with an index, or relative risk, whereas a realestate practitioner is generally concerned with absolute risk, both because direct realestate investments are essentially not diversifiable and because sponsor compensationmetrics are based on absolute measures.
The first limitation of the Sharpe ratio has been addressed by Satchell and Pedersen(2002) in formulating the Sortino ratio, which expresses risk as only the downside tailof the distribution of returns. The Sortino ratio also incidentally eliminates the secondlimitation of the Sharpe method by substituting a target rate of returns for the actualmarket returns. As with the Sharpe ratio, the higher values of the Sortino ratio signifygreater returns per unit of risk.
Having addressed the key limitations of the Sharpe ratio, however, the Sortino ratiostill only allows comparison of projects to one another and does not offer an objectivemetric of sufficiency of returns for a given project. Here we submit theSimulation-Based Excess Return Model (SERM), a methodology to address thisproblem.
Since the Sortino and Sharpe ratios have been developed to measure risk-adjustedperformance for investment managers in the equity markets, they take as theirfoundation volatility of actual portfolio and index returns. We apply an analogousmethodology to the universe of simulated outcomes.
First we express the downside risk DR in terms of net present value NPVdiscounted the project’s weighted average cost of capital WACC. We use NPV as a
Simulation-basedexcess return
model
117
measure of returns because during simulation a large proportion of iterations produceundefined values for internal rate of return. In other words, downside risk is heredefined as the semi-standard deviation of returns expressed as NPV divided by meanNPV:
DR ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiR 0
21NPV 2f NPVð ÞdNPV
q
NPVð1Þ
where NPV is the simulated net present value of returns as discounted by the weightedaverage cost of capital for the project.
Next, we determine the mean expected internal rate of return MEIRR. We cancalculate the MEIRR from the mean investment and gains from the simulation andknown effective project duration. Effective project duration is not equal to calendarduration because project cash flows are asymmetric, and is defined as the durationover which the known mean investment and gain amounts would produce a knownIRR. A good value for the known IRR is generally the risk-free rate, and meaninvestments and gains discounted by the risk-free rate are simulated for the purpose.The MEIRR can be calculated based on this effective project duration.
We use IRR as a component of SERM because it offers an objective measuresimulated returns given all cash flows and is the equivalent to the actual equityinvestment return measured by the Sharpe and Sortino ratios. In addition, it is widelyunderstood and accepted by the practitioners. Naturally, in practice the size of theprofit and loss projection needs to be taken into account even when the MEIRR appearsattractive.
Given the MEIRR, we can calculate the mean return per unit of risk by dividingMEIRR by DR. we can then discount the MEIRR by the resulting metric to obtain therisk adjusted return for the project. Finally, we adjust the resulting value for the timevalue of money by subtracting from it the risk-free rate to arrive at excess risk-adjustedreturn ERR, resulting in the following formulation:
SERM ¼E r½ �
E r½ �DR2 rf ð2Þ
where E[r] is the mean expected internal rate of return (IRR) for the project derivedfrom simulation; DR is the simulated downside risk for the project; and is the risk-freerate.
Equation (2) reduced to its normal form is the definition of the excess risk-adjustedreturn under the SERM:
SERM ¼E r½ �
E r½ �DR2 rf : ð3Þ
In other words, SERM adjusts the expected return by the ratio of the expected returnand downside risk that is derived from simulation and measures the excess of thisadjusted return over the risk-free rate.
Once the ERR is obtained, it becomes possible to derive the required rate of returnMAR for the given project, by computing the rate of return at ERR ¼ 0, as follows:
JPIF29,2
118
MAR ¼ffiffiffiffiffiffiffiffiffiffiffiffiffirf DRj j
p: ð4Þ
SERM MAR offers a useful metric to demonstrate the appropriate level of risk for thegiven project expressed in familiar terms. Care should be taken when applying theSERM MAR as a shortcut hurdle rate to other projects with potentially differentcharacteristics because they may face vastly different probable outcomes.
Additional metricsWhile the single-number ERR is a useful decision-making metric of the risk/returnprofile of a project, it does not address every aspect of the risk continuum for thepractitioner. An informed decision to undertake a project requires additionalinformation from the simulation, such as:
. Probability of loss, defined as the probably that the WACC-discounted NPV isless than zero.
. Median expected NPV, defined as the median NPV from simulation multiplied bythe probability of loss.
. Probability of clearing an IRR hurdle is an additional metric that is useful forpractitioners working with equity partners, enabling a practitioner tounderstand the probability to delivering promised return levels to investors.
. Mean expected loss in the event that the project is unsuccessful.
Risk-free rateThe most appropriate risk-free rate to use for SERM analysis is the interest rate paidby the Treasury obligation of a similar maturity to the development project, less anadjustment factor to back out the interest rate risk premium inherent in the Treasurysecurity’s interest rate:
rf ¼ rfT 2 rrp ð5Þ
where rP is the risk-free rate for the project; rfT is the US Treasury rate for an obligationof an equivalent term; and is the interest rate risk premium adjustment factor.
Discount rate: WACCBecause the purpose of the SERM analysis is to determine risk premium, it woulddistort results if a risk premium were also included in the discount rate. Therefore thediscount rate for SERM is determined as follows:
rP ¼ LTVð ÞrD þ 1 2 LTVð ÞrE ð6Þ
where rP is the weighted average cost of capital WACC for the project; rD is theeffective construction period interest rate of the debt portion of the total cost; rE is thebreak-even equity IRR applied to the equity portion of the total cost; and LTV is theloan to value ratio for the construction loan.
A normal profit for the developer needs to be included in the break-even IRR toensure that the developer’s overhead is at least covered. This is best implemented inthe form of a developer’s fee incorporated into the project cost model, or alternatively inthe form of a predetermined markup to the project break-even IRR. Issues with themarkup method include the fact that the normal profit fluctuates with the total return,
Simulation-basedexcess return
model
119
thus putting overhead coverage at risk, as well as the potentially arbitrary nature ofthe predetermined markup that can undermine the objectivity of the SERM analysis.
It is imperative to use realistic component rates of return when constructing theWACC. Assuming the standard development capital structure prevailing in the UnitesStates where land acquisition is financed by developer’s equity and construction isfinanced by a combination of investor equity and construction debt, for instance wouldproduce a structure as follows. The rate of return on debt can be approximated byusing the nominal interest rate on the construction loan. The rate of return on theinvestor’s equity can be approximated by the preferred rate demanded by the investorplus a variable component that depends on the actual return based on the payoutwaterfall for the project, making the WACC an output variable of the simulation. Forsimplicity’s sake it is possible to state the investor’s equity rate of return equals theinvestor’s stated required rate or return. Finally, the developer’s minimal required rateof return should be incorporated in the WACC. This rate may represent the holdingcost of the development parcel, opportunity cost of capital, or even zero in some cases.
Interdependence of input variablesThe following variables are commonly assessed in simulation of outcomes in the caseof office development. Variables for other product types may vary; however, analysisof other product types is outside the scope of this paper.
. General inflation. General inflation can be expressed as the average inflationfrom time T0 to the time of project stabilization TS.
. Market vacancy. The market vacancy of most interest is the vacancy at the timeof project stabilization TS. It may be possible to instead model vacancy as theendpoint of a Brownian-motion series from T0 to TS, but it is not clear that theadded complexity of modeling is justified by the potential enhancement inherentin this methodology as only the terminal value is of interest and the range ofterminal values is empirically derived.
. Starting occupancy and stabilization period. Starting occupancy at completion ofa project and the duration of the stabilization period (TD-TS) are functions ofmarket vacancy.
. Market rent. The market rent of interest is the rent at the of project stabilizationTS. While there is evidence that market rents follow a Brownian random walkover time, a simplification of modeling a range of static rents at the time ofstabilization is possible because interim rent flows between time zero andstabilization are not required.
. Rent escalation. Rent escalation from the time of project stabilization TS to thetime of disposition TD is best expressed as a spread above general inflation inorder to capture the close interrelationship between these variables.
. Hard (construction) cost inflation. Hard cost inflation from time T0 to thecommencement of construction (TC) is best expressed as a spread above generalinflation in order to capture the close interrelationship between these variables.
. Construction loan interest rate. Construction loan interest rate is best expressedas a spread above general inflation in order to capture the close interrelationshipbetween these variables.
JPIF29,2
120
. Direct capitalization rate. Direct capitalization rate of interest is the rate at TD is avariable that is exogenous to the model as it primarily depends onmacroeconomic factors and macro-level investment allocation decisions.
. Construction period duration (TS-TC).
. Actual rate of return on invested equity (rE). Because real estate developmentinvestment agreements frequently employ a waterfall payout schedule, the exactcost of equity depends on the profitability of the project, making the WACC adependent variable in computing the developer’s return. When analysis isperformed from the developer’s perspective rather than the investor’sperspective, the variability of the cost of capital should be modeled as adependent input variable, whereas in the converse case it may be treated as afixed quantity.
For practitioners it is intuitively obvious that market rent levels, rent escalations, newproject vacancy levels, and stabilization period durations, and possibly hard costescalation, are all dependent on the supply and demand for new space, for whichMarket Vacancy can stand as a proxy. The statistical analysis of this relationship ispresented in the case study in this paper.
There is a further relationship among general inflation, hard cost inflation,construction loan interest rates, and the direct capitalization rate.
To add complexity, hard cost inflation and the direct capitalization rate are alsosomewhat correlated with supply and demand for new space as proxied by marketvacancy, but are also dependent on outside factors such as global demand for steel,relative demand for real estate as an asset class, and risk premiums demanded by realestate investors. For a simplified practical implementation, these variables can betreated as independent.
Construction period duration is an independent variable as it relates toproject-specific considerations such as entitlement risk; however, it also exhibits adependency on market conditions, insofar as few practitioners would contemplate orhave the financing for launching new construction in the face of elevated vacancies[1].
The use of interdependent input variables in a simulation context is essentially anapplication of the familiar scenario-based sensitivity analysis extended to a continuumif a large number of intermediate scenarios that extends sensitivity analysis withstatistical and probabilistic analysis tools.
1. Derivation. The relationship among the interdependent variables can beapproximated by the following empirically-derived relationships by using curve fittingmethodologies.
With the exception of new project vacancy, all variables can be derived fromcommercially available sources. Expert estimates of the spread need to be used inpractice because there is little available hard data about vacancy for new projects vsgeneral market vacancy. In selecting data observations for curve fitting the historicalperiod needs to be sufficiently long to capture at least one and for preference multiplereal estate cycles.
It is not expected that the relationships among the variables will hold universallytrue for all MSA, submarket and product type permutations, and in the absence ofcommercially available data sources practitioners will need to perform their own curvefitting exercises to obtain valid relationships.
Simulation-basedexcess return
model
121
2. Additional variables. The following independent variables can be additionallysimulated in special circumstances:
. Project size, for cases where the size of the project developed depends on marketor regulatory variables.
. Environmental and municipal fees.
. Construction commencement date, including modeling of potential delaysresulting from subpar vacancy and rent for the particular simulation instance, aswell as for project delays due to construction or regulatory issues.
These can be dependent on the outcome of entitlement processes such as zoningchanges, density limits, height limits, etc. Modeling regulatory outcomes may requireintroducing a binary variable (1 for approval, 0 for disapproval) with a probabilitydistribution based on prior experience for the market in question.
III. Comparison of SERM with common methodologiesSERM vs “pure” DCFWhile DCF, which has in recent years gained wide acceptance among the moresophisticated developers, is a significant improvement on the older underwritingpractices such as developer’s yield and return on cost, it lacks the ability to accuratelyassess the risk/return profile of a project, forcing the practitioner to rely on rules ofthumb for determination of risk premium.
The key difference between “pure” DCF and SERM is that SERM essentially adjuststhe universe of outcomes from reliance of discount rates to serve as proxy for risk byadding stochastic methods.
SERM vs “naked” simulationMonte Carlo simulation is widely acknowledged to overcome the static limitation ofDCF; however, prevailing implementations of simulation practices suffer from anumber of limitations:
. They treat all input variables as mutually independent despite the fact there areclose interrelationships among them, and thereby producing skewed results.
. While they output a range of return outcomes, they offer no guidance as towhether sufficient risk premium is being generated by a project to justify(continued) investment.
SERM vs real optionsReal options have been recently introduced into the real estate developmentunderwriters’ toolkit. While this approach hold considerable promise, it is not yetpractical for use because:
. Problem setup for real options is highly complex and in the absence of standard“cookbooks” can be highly error-prone.
. No public data are available regarding periodic volatility of returns.
. Real options still produce point estimates of returns and offer no information onthe statistical risk metrics of returns.
JPIF29,2
122
LimitationsThe SERM methodology as described here is intended for modeling individualdevelopment projects indented for disposition on completion rather than portfolios ofprojects or investment performance of existing assets. With additional research it maybe feasible to extend the methodology beyond the current application.
Quality of results produced necessarily depends on the quality and completeness ofinput assumptions. For instance, if the probability spread used for vacancy and rent isexceedingly optimistic, the model will fail to account for the probability of extrememovements and will produce skewed results. The financial crisis of 2008 hasdemonstrated that extreme ends of a probability spectrum must be included in anyinvestment model of significant duration lest the impact of extreme outcomes overstatethe return expectation.
IV. Practical implementationModeling environmentSERM relies on modeling using commercially available spreadsheet software with anda commercially available simulation engine. No purpose-designed commercial toolsexist at present to automate the creation of the model suitable for SERM analysis.
Interpretation of simulation results for decision makingA zero-value SERM ERR return for a project indicates that the forecast returns areexactly equal to the risk inherent in the project, and effectively a zero NPV. A greaterthan zero-value ERR indicates a positive NPV for the project, while a negative ERRimplies a negative NPV.
Further research and applicationsThe SERM methodology as described in this paper is limited to individualdevelopment project investment decisions. With additional theoretical work, the SERMconcept may be possible to expend to portfolios of development project investments,individual and portfolio investments in stabilized real assets, and potentially outside ofreal estate in the areas of investment where stochastic methods may improve thequality of discounted cash flow modeling such as private equity-financed mergers andacquisitions. The author welcomes opportunities to collaborate with practitioners inextending the research into such fields.
V. Case study: high-rise office developmentThe following case study for a high-rise office development in the financial district ofSan Francisco, California is based on an amalgam of characteristics of several realprojects and conditions, costs, and prices prevailing during the second quarter of 2010.
Project descriptionA development project sponsor owns a 0.41 acre parcel of land in the financial districtof San Francisco with a fully-leased structure much smaller than permitted by currentzoning regulations. The property is unencumbered by debt and generates $525,000annually in free cash flow. Current zoning allows structures with floor area ratio up to26:1 for a maximum net rentable area of 461,760 square feet. According to brokers’opinion, the parcel’s market value is approximately $31.00 per square foot of FAR, or
Simulation-basedexcess return
model
123
approximately $14,315,000. The sponsor has commissioned a preliminary design studyfor construction of an office building and obtained cost estimates with expectedescalations. The design study suggests a concrete-framed 32-story tower with a grossarea 572,831 square feet with retail space at street level. Construction costs areestimated to be $170.50 per gross foot in 2010 and are expected to escalate at 100 basispoints above prevailing inflation which is currently running at 2 percent annually. Theproject is expected to take approximately 6.5 years to complete, at which point theaverage full-service gross rent for the office portion of this building is expected reach$51.95 which is an annual average escalation of 6 percent from the current potentialaverage rent of $39.00. Annual step rent is expected to be 1.5 percent. Table I detailsthe building and rental structure of the project. Lease terms and market-ratecommissions are outlined in Table II. The sponsor believes that the project can befinanced at the start of the construction phase with an interest-only mini-permanentloan of 65 percent of the terminal project value at a 6.5 percent interest rate. Themini-perm instrument can remain in place through construction and stabilizationphases and be released upon final disposition of the project, which the sponsor intendsto effect upon stabilization. The sponsor expects based on recent averages and trendsthat the project will fetch a disposition value based on price direct capitalization rate of6.0 percent.
DCF analysisA project budget and schedule constructed based above parameters, as presented inTables III and IV and Figure 1. Table V summarizes forecast returns for the project.Because the sponsor uses an internal hurdle rate of 20.00 percent, this project is likelyto be approved.
Static sensitivity analysis of varying key inputs by 10 percent in unfavorabledirections reveals a generally acceptable level of risk, as illustrated in Table VI, withnone of the scenarios indicating a direct financial loss.
Based on the DCF analysis, this project is likely to be approved to proceed.
Simulation parametersKey drivers of financial performance of development projects of this type areunderstood to include the vacancy rate for the product type, construction cost, debtinterest rate, exit capitalization rate, market rent and project duration. Other costdrivers such as entitlement, legal and architecture and engineering are generallypredictable at the outset of projects and as such do not require simulation. Relativesensitivity of results to input variables within their respective ranges was analyzedwith the use of a Tornado chart presented in Figure 2, confirming that rent, schedule,construction loan interest rate and exit capitalization rate offer the largest impact toprofitability necessitating particularly careful attention to modeling their behavior forsimulation[2].
Market rent is generally assumed to be a function of vacancy. Additionally, itassumed that construction cost is also related to vacancy insofar as it is a function ofeconomic growth. These hypotheses were tested using publicly available sourceswherever possible. Where public data were not available, information was gatheredthrough private interviews with market practitioners. Sources not explicitly citedremain anonymous by request. A strong relationship between market vacancy and
JPIF29,2
124
2010
2015
Flo
orU
sab
lear
eaR
enta
ble
area
Ren
tP
SF
Tot
alre
nt
Ren
tP
SF
Tot
alre
nt
Offi
cere
nts
,fu
ll-s
erv
ice
gro
ss32
12,5
0015
,000
45.2
567
8,77
560
.28
904,
159
3112
,500
15,0
0045
.25
678,
775
60.2
890
4,15
930
12,5
0015
,000
45.2
567
8,77
560
.28
904,
159
2912
,500
15,0
0045
.25
678,
775
60.2
890
4,15
928
12,5
0015
,000
45.2
567
8,77
560
.28
904,
159
2712
,500
15,0
0045
.25
678,
775
60.2
890
4,15
926
12,5
0015
,000
39.2
558
8,77
552
.28
784,
275
2512
,500
15,0
0039
.25
588,
775
52.2
878
4,27
524
12,5
0015
,000
39.2
558
8,77
552
.28
784,
275
2312
,500
15,0
0039
.25
588,
775
52.2
878
4,27
522
12,5
0015
,000
39.2
558
8,77
552
.28
784,
275
2112
,500
15,0
0039
.25
588,
775
52.2
878
4,27
520
12,5
0015
,000
39.2
558
8,77
552
.28
784,
275
1912
,500
15,0
0039
.25
588,
775
52.2
878
4,27
518
12,5
0015
,000
37.2
555
8,77
549
.62
744,
313
1712
,500
15,0
0037
.25
558,
775
49.6
274
4,31
316
12,5
0015
,000
37.2
555
8,77
549
.62
744,
313
1512
,500
15,0
0037
.25
558,
775
49.6
274
4,31
314
12,5
0015
,000
37.2
555
8,77
549
.62
744,
313
1312
,500
15,0
0037
.25
558,
775
49.6
274
4,31
312
12,5
0015
,000
37.2
555
8,77
549
.62
744,
313
1112
,500
15,0
0037
.25
558,
775
49.6
274
4,31
310
12,5
0015
,000
37.2
555
8,77
549
.62
744,
313
912
,500
15,0
0037
.25
558,
775
49.6
274
4,31
38
12,5
0015
,000
35.2
552
8,77
546
.96
704,
352
712
,500
15,0
0035
.25
528,
775
46.9
670
4,35
26
12,5
0015
,000
35.2
552
8,77
546
.96
704,
352
512
,000
14,4
0035
.25
507,
624
46.9
667
6,17
84
12,0
0014
,400
35.2
550
7,62
446
.96
676,
178
312
,000
14,4
0035
.25
507,
624
46.9
667
6,17
8T
otal
offi
ce73
,500
448,
200
39.0
017
,479
,800
51.9
523
,283
,871
Ret
ail
ren
ts,
net
29,
500
12,0
0022
.00
264,
000
29.3
035
1,66
01
1,45
01,
450
40.0
058
,000
53.2
877
,259
Tot
alre
tail
10,9
5013
,450
23.9
432
2,00
031
.89
428,
918
Ov
eral
lre
nts
384,
450
461,
650
17,8
01,8
0023
,712
,789
Ad
dit
ion
alre
nt
An
ten
na
120,
000
159,
845
Sig
nag
e10
0,00
013
3,20
4P
ark
ing
312,
000
415,
598
Sto
rag
e8,
000
20.0
016
0,00
021
3,12
7T
otal
692,
000
921,
775
Bu
ild
ing
18,4
93,8
0024
,634
,564
Table I.Rent roll
Simulation-basedexcess return
model
125
rents was empirically demonstrated, as was an anecdotal relationship betweenvacancy and construction costs. Costs of funds, loan rates and capitalization rates werefound to be related to spot inflation, but not to historic inflation patterns and notrelated to vacancy. Tenant Improvement costs, which may be reasonably assumed tobe tied to construction costs, were found to have no clear relationship to either becauseonce prevailing rates are established for a given market they tend to follow apredictable appreciation pattern tied to inflation. In practice, these costs bear somerelationship to relative market power of landlords and tenants but fluctuations aroundthe mean appear to be self-canceling.
Identifying statistical patterns in the independent drivers and their relationshipswith the dependent drivers comprised the bulk of the work in compiling this case studyand, while the resulting patterns may possible to generalize across US real estatemarkets, the specific curve shapes and interrelationships will be specific to eachmarket, and quite possibly submarket, so caution should be used in their applications.
Primary driversMarket rent. Figure 3 details the history of market vacancy and average Class A officerent for the financial district of San Francisco from 1980 through 2009[3]. In order totest the hypothesis that rent is a function of vacancy, rent was expressed in terms ofconstant dollars according to three different metrics: the gross domestic product (GDP)deflator, consumer price index, and the producer price index for real estate. Rents werefound to be relatively weakly tied with the CPI, with the correlation coefficient of20.47, while correlations with the GDP deflator and PPI were nearly identical at 20.54and 20.55 respectively. The sparseness of the PPI data set of only 23 samplessuggested the use of the GDP deflator data for developing a regression. Linearregression analysis results in slope m ¼ 2122:09 with intercept y ¼ 55:95. The dataand regression charts are presented in Table VII and Figure 4.
The regression alone is not sufficient to present a complete picture of therelationship of vacancy and rents because sponsors and are reluctant to initiate – andlenders reluctant to fund – new construction projects at times when vacancies areelevated. In order to reflect this condition, commencement of construction is delayed ifthe simulated vacancy is above average according to the assumption that for everypercentage point of difference between simulated vacancy and average vacancyconstruction will be delayed by 1.75 years. This assumption was based on thehistorical pattern of reversion to the mean observed in Figure 3. However, as vividlydemonstrated by the financial crisis of 2008, vacancy rates can also plunge afterconstruction has been substantially complete[4]. This eventuality is modeled with asecond independent variable for vacancy at completion. The complete picture of the
Office Retail
Term, years 5 5Average escalation (%) 2.50 2.50Commission basis 121,528,045 2,198,206Commission rate (%) 7.0 6.0Total commissions 8,506,963 131,892Commissions PSF ($) 18.98 9.81
Table II.Lease structures andcommissions
JPIF29,2
126
Pro
ject
star
t7/
1/20
10A
vg
infl
atio
n2.
0%P
roje
cten
d12
/1/2
016
Con
stru
ctio
nd
elay
(mon
ths)
–S
tart
ing
occu
pan
cyR
even
ues
2010
$/u
nit
Un
its
2010
$to
tal
Esc
alat
ion
Esc
alat
ed$
Sta
rtD
ura
tion
En
d(%
)In
-pla
cen
etre
nt
525,
000
0.0%
525,
000
7/1/
2010
258/
1/20
1210
0O
ccu
pan
cy6/
1/20
1518
12/1
/201
6O
ffice
ren
t39
.00
448,
200
17,4
79,8
006.
0%23
,283
,871
6/1/
2015
1812
/1/2
016
70R
etai
lre
nt
23.9
413
,450
322,
000
6.0%
428,
918
6/1/
2015
17/
1/20
1550
Ad
dit
ion
alre
nt
692,
000
6.0%
921,
775
6/1/
2015
612
/1/2
015
100
Ste
pre
nt
1.5%
627,
928
6/1/
2015
1812
/1/2
016
70L
ess:
offi
ceg
ener
alv
acan
cy5.
00%
(924
,690
)6.
0%(1
,231
,728
)6/
1/20
1518
12/1
/201
610
0L
ess:
pro
per
tyta
x1.
20%
(1,5
48,6
06)
6/1/
2015
1812
/1/2
016
100
Les
s:in
sura
nce
2.00
461,
650
(923
,300
)2.
0%(1
,017
,795
)6/
1/20
1518
12/1
/201
610
0L
ess:
oper
atin
gex
pen
se9.
2546
1,65
0(4
,270
,263
)2.
0%(4
,707
,303
)6/
1/20
1518
12/1
/201
610
0
Net
oper
atin
gin
com
e12
,375
,548
16,7
57,0
60
Tas
k20
10$/
un
itU
nit
s20
10$
tota
lE
scal
atio
nE
scal
ated
$S
tart
Du
rati
onE
nd
Acq
uis
itio
np
has
e7/
1/20
101
8/1/
2010
Lan
d31
.00
461,
760
14,3
14,5
600.
0%14
,314
,560
7/1/
2010
18/
1/20
10T
itle
insu
ran
ce25
0,00
00.
0%25
0,00
07/
1/20
101
8/1/
2010
Leg
alex
pen
ses
350,
000
0.0%
350,
000
7/1/
2010
18/
1/20
10
En
titl
emen
tp
has
e8/
1/20
1036
8/1/
2013
Tra
nsf
erab
led
evel
opm
ent
rig
hts
7,50
0,00
00.
0%7.
500,
000
8/1/
2010
62/
1/20
11A
rch
itec
ture
and
eng
inee
rin
g4,
500,
000
2.0%
4,50
7,57
58/
1/20
1024
8/1/
2012
En
titl
emen
tap
pro
val
250,
000
2.0%
250,
421
8/1/
2010
368/
1/20
13F
ees
and
per
mit
s47
5,00
00.
0%47
5,00
08/
13/2
010
248/
13/2
012
(continued
)
Table III.Budget and cash flow
forecast
Simulation-basedexcess return
model
127
Con
stru
ctio
np
has
e8/
1/20
1234
6/1/
2015
Imp
act
fees
12,5
00,0
002.
0%13
,027
,598
8/1/
2012
19/
1/20
12D
emol
itio
n52
5,00
02.
0%55
8,10
28/
1/20
132
10/1
/201
3C
onst
ruct
ion
155.
0057
2,83
188
,788
,728
3.0%
97,7
55,3
1110
/1/2
013
184/
1/20
15C
omm
issi
onin
g40
0,00
02.
0%43
9,48
14/
1/20
152
6/1/
2015
Pro
per
tyta
xes
1.20
1,29
1,23
92.
0%1,
345,
740
8/1/
2012
346/
1/20
15A
rtw
ork
887,
887
2.0%
946,
999
10/1
/201
318
4/1/
2015
Insu
ran
ce42
5,00
02.
0%44
2,93
88/
1/20
1234
6/1/
2015
Sec
uri
tysy
stem
s20
0,00
02.
0%21
9,74
14/
1/20
152
6/1/
2015
Sta
bil
izat
ion
ph
ase
8/1/
2013
2610
/1/2
015
Ten
ant
spac
ep
lan
nin
g10
0,00
02.
0%10
9,87
04/
1/20
156
10/1
/201
5L
egal
exp
ense
s65
0,00
02.
0%71
4,15
74/
1/20
156
10/1
/201
5M
ark
etin
gan
dP
R1,
500,
000
2.0%
1,59
4,57
88/
1/20
1324
8/1/
2015
Ten
ant
imp
rov
emen
ts–
offi
ce60
.00
373,
500
22,4
10,0
002.
0%24
,621
,932
4/1/
2015
610
/1/2
015
Lea
sin
gco
mm
issi
ons
–of
fice
17.1
037
3,50
06,
386,
396
8,50
6,96
312
/3/2
014
66/
3/20
15T
enan
tim
pro
vem
ents
–re
tail
40.0
010
,950
438,
000
2.0%
481,
232
4/1/
2015
610
/1/2
015
Lea
sin
gco
mm
issi
ons
–re
tail
9.04
10,9
5099
,015
131,
892
12/3
/201
43
3/3/
2015
Mis
cell
aneo
us
leas
ing
350,
000
2.0%
384,
546
4/1/
2015
610
/1/2
015
Ad
min
istr
atio
n7/
1/20
1063
10/1
/201
5G
ener
alan
dad
min
istr
ativ
e3,
500,
000
2.0%
3,50
0,00
07/
1/20
1063
10/1
/201
5D
evel
opm
ent
fee
2.50
3,64
1,90
74,
000,
352
7/1/
2010
6310
/1/2
015
Pro
ject
con
tin
gen
cy3.
004,
595,
288
5,02
5,42
27/
1/20
1063
10/1
/201
5
Inte
rest
-on
lyd
ebt
serv
ice
6.50
22,2
58,7
6722
,258
,767
Tot
alex
pen
ses
198,
586,
786
213,
713,
177
(continued
)
Table III.
JPIF29,2
128
Con
stru
ctio
nfi
nan
cin
gM
ini-
per
mlo
an65
.00
114,
613,
213
124,
445,
367
8/1/
2012
19/
1/20
12L
ess:
len
der
fees
0.50
(573
,066
.06)
(622
,227
)8/
1/20
121
9/1/
2012
Net
bor
row
ing
114,
040,
147
123,
823,
140
8/1/
2012
19/
1/20
12L
oan
run
nin
gb
alan
ceD
isp
osit
ion
12/1
/201
61
1/1/
2017
Rev
ersi
onv
alu
eC
apra
te6.
0027
0,53
4,32
912
/1/2
016
11/
1/20
17L
ess;
dis
pos
itio
nex
pen
ses
0.35
(946
,870
)12
/1/2
016
11/
1/20
17N
etre
ver
sion
269,
587,
459
12/1
/201
61
1/1/
2017
Deb
tre
tire
men
t(1
24,4
45,3
67)
12/1
/201
61
1/1/
2017
Yie
ldIR
RR
OE
Pro
fit
Cas
hfl
owb
efor
ele
ver
age
8.48
16.5
5%70
.62%
120,
792,
536
Cas
hfl
owb
efor
ein
com
eta
xes
7.60
22.5
1%12
3.06
%97
,911
,543
Table III.
Simulation-basedexcess return
model
129
Per
iod
01
23
45
67
Yea
ren
din
g7/
31/2
011
7/31
/201
27/
31/2
013
7/31
/201
47/
31/2
015
7/31
/201
67/
31/2
017
Con
stru
ctio
nd
elay
(mon
ths)
Rev
enu
esIn
-pla
cen
etre
nt
525,
000
525,
000
1,45
8–
––
–O
ccu
pan
cyO
ffice
ren
t–
––
–17
,074
,839
21,8
09,2
26R
etai
lre
nt
––
––
428,
918
428,
918
Ad
dit
ion
alre
nt
––
––
921,
775
921,
775
Ste
pre
nt
––
––
–27
6,38
3L
ess:
offi
ceg
ener
alv
acan
cy–
––
–(1
,231
,728
)(1
,231
,728
)(1
,231
,728
)L
ess:
pro
per
tyta
x–
––
–(1
,548
,606
)(1
,548
,606
)(1
,548
,606
)L
ess:
insu
ran
ce–
––
–(1
,017
,795
)(1
,017
,795
)(1
,017
,795
)L
ess:
oper
atin
gex
pen
se–
––
–(4
,707
,303
)(4
,707
,303
)(4
,707
,303
)
Net
oper
atin
gin
com
e52
5,00
052
5,00
01,
458
–9,
920,
100
14,9
30,8
7016
,757
,060
Tas
k
Acq
uis
itio
np
has
eL
and
14,3
14,5
60–
––
––
–T
itle
insu
ran
ce25
0,00
0–
––
––
–L
egal
exp
ense
s35
0,00
0–
––
––
–E
nti
tlem
ent
ph
ase
––
––
––
–T
ran
sfer
able
dev
elop
men
tri
gh
ts7,
500,
000
––
––
––
Arc
hit
ectu
rean
den
gin
eeri
ng
2,26
3,11
32,
238,
244
6,21
7–
––
–E
nti
tlem
ent
app
rov
al84
,012
83,0
8983
,089
231
––
–F
ees
and
per
mit
s23
0,62
123
5,86
28,
517
––
––
Con
stru
ctio
np
has
e–
––
––
––
Imp
act
fees
––
13,0
27,5
98–
––
–D
emol
itio
n–
––
558,
102
––
–C
onst
ruct
ion
––
–54
,149
,651
43,6
05,6
60–
–C
omm
issi
onin
g–
––
–43
9,48
1–
–P
rop
erty
tax
es–
–49
0,83
148
5,43
736
9,47
2–
–A
rtw
ork
––
–52
4,57
242
2,42
7–
–In
sura
nce
––
161,
553
159,
777
121,
608
––
(continued
)
Table IV.Budget and cash flowforecast
JPIF29,2
130
Per
iod
01
23
45
67
Yea
ren
din
g7/
31/2
011
7/31
/201
27/
31/2
013
7/31
/201
47/
31/2
015
7/31
/201
67/
31/2
017
Sec
uri
tysy
stem
s–
––
–21
9,74
1–
–S
tab
iliz
atio
np
has
e–
––
––
––
Ten
ant
spac
ep
lan
nin
g–
––
–72
,646
37,2
74–
Leg
alex
pen
ses
––
––
472,
202
241,
955
–M
ark
etin
gan
dP
R–
––
800,
588
791,
790
2,19
9–
Ten
nan
tim
pro
vem
ents
–of
fice
––
––
16,2
80,0
758,
341,
857
–L
easi
ng
com
mis
sion
s–
offi
ce–
––
–8,
506,
963
––
Ten
ant
imp
rov
emen
ts–
reta
il–
––
–31
8,19
216
3,04
0–
Lea
sin
gco
mm
issi
ons
–re
tail
––
––
131,
892
––
Mis
cell
aneo
us
leas
ing
––
––
254,
263
130,
283
–A
dm
inis
trat
ion
––
––
––
–G
ener
alan
dad
min
istr
ativ
e67
6,69
267
6,69
267
6,69
267
6,69
267
6,69
211
6,54
1–
Dev
elop
men
tfe
e81
,361
80,8
4736
1,36
21,
433,
876
1,81
7,07
822
5.82
7–
Pro
ject
con
tin
gen
cy97
1,61
897
1,61
897
1,61
897
1,61
897
1,61
816
7,33
4–
Inte
rest
-on
lyd
ebt
serv
ice
––
1,97
7,12
34,
502,
006
7,69
0,68
98,
088,
949
–T
otal
exp
ense
s14
,914
,560
11,8
07,4
174,
286,
352
17,7
64,6
0064
,262
,550
83,1
62,4
8917
,515
,210
–C
onst
ruct
ion
fin
anci
ng
Min
i-p
erm
loan
Les
s:le
nd
erfe
es–
–(6
22,2
27)
––
––
Net
bor
row
ing
––
30,4
17.2
7338
,844
,354
49,0
56,6
706,
127,
070
–L
oan
run
nin
gb
alan
ce–
–30
,417
,273
69,2
61,6
2711
8,31
8,29
712
4,44
5,36
7–
Dis
pos
itio
nR
ever
sion
val
ue
Les
s;d
isp
osit
ion
exp
ense
sN
etre
ver
sion
––
––
––
269,
587,
459
Deb
tre
tire
men
t–
––
––
–(1
24,4
45,3
67)
Cas
hfl
owb
efor
ele
ver
age
(14,
914,
560)
(11,
282,
417)
(3,7
61,3
52)
(15,
786,
018)
(59,
760,
544)
(65,
551,
700)
5,50
4,60
928
6,34
4,51
8C
ash
flow
bef
ore
inco
me
tax
es(1
4,91
4,56
0)(1
1,28
2,41
7)(3
,761
,352
)12
,031
,905
(25,
418.
196)
(24,
185,
719)
3,54
2,73
016
1,89
9,15
2
Table IV.
Simulation-basedexcess return
model
131
Yield IRR ROE(%) (%) (%) Profit
Cash flow before leverage 8.48 16.55 70.62 120,792,536Cash flow before income taxes 7.60 22.51 123.06 97,911,543
Table V.Summary of expectedreturns
Figure 1.Project schedule
Yield IRR ROE(%) (%) (%) Profit
Rent StatConstruction cost Stat 8.48 16.55 70.62 120,792,536 Before leverageCap rate Stat 7.60 22.51 123.06 97,911,543 Before income taxesRent 210%Construction cost Stat 7.26 11.23 43.90 75,449,929 Before leverageCap rate Stat 6.50 14.39 65.07 52,650,356 Before income taxesRent StatConstruction cost þ10% 7.98 14.62 59.39 107,913,698 Before leverageCap rate Stat 7.15 19.82 99.70 83,849,959 Before income taxesRent StatConstruction cost Stat 8.48 13.81 56.29 96,284,586 Before leverageCap rate þ10% 7.60 18.52 92.26 73,403,592 Before income taxesRent 210%Construction cost 210% 7.75 13.30 54.79 88,328,767 Before leverageCap rate þ10% 6.94 17.60 87.34 66,711,941 Before income taxes
Table VI.Sensitivity analysis
JPIF29,2
132
effect of its elevated condition includes more than impact to rent because both thestarting occupancy at project completion and the duration of the stabilization phase arealso negatively affected. The extension of the stabilization phase may be assumed tofollow the same pattern as the postponement of constriction. Modeling startingoccupancy proved more challenging because hard data were scarce and the slopeprecipitous. A reasonable approximation of reality was achieved with a formulationwhere starting occupancy:
So ¼ MIN Max So; 1 2 MIN LN 100 MIN Vacancy; Avg:Vacancy� ��� �
11=100; 000; 100%� �� �
ðC1Þ
where Max So is a constant limiting the largest starting occupancy to that deemedpossible by the sponsor. In this model, Max So ¼ 80 percent.
Figure 3.San Francisco CBD class
office rent and vacancy1980-2011
Figure 2.Tornado analysis
Simulation-basedexcess return
model
133
Vac
ancy
rate
Yea
r(%
)
Av
erag
ere
nt,
nom
inal
GD
Pd
eflat
ora
GD
Pd
eflat
or,
2009
dol
lars
Av
erag
ere
nt,
2009
dol
lars
CP
IbC
PI
defl
ator
in19
80d
olla
rsC
PI
defl
ator
in20
09d
olla
rsR
ent
in20
09C
PI
dol
lars
PP
Ic
PP
Iin
2009
dol
lars
Ren
tin
2009
PP
Id
olla
rs
1980
1.00
2354
.043
181.
792
41.8
113
.510
012
6.1
2919
814.
0031
59.1
1917
3.20
653
.69
10.3
101.
103
125.
009
38.7
519
825.
7035
62.7
2616
7.45
558
.61
6.2
102.
165
123.
969
43.3
919
835.
9037
65.2
0716
3.65
60.5
53.
210
3.19
712
2.96
945
.519
8410
.10
3267
.655
160.
032
51.2
14.
310
4.24
121.
969
39.0
319
8513
.30
3169
.713
157.
0848
.69
3.6
105.
276
120.
985
37.5
119
8616
.60
2971
.25
154.
923
44.9
31.
910
6.29
512
0.02
634
.81
1987
13.9
030
73.1
9615
2.26
445
.68
3.6
107.
331
119.
061
35.7
210
1.4
156.
357
46.9
0719
8813
.20
2675
.694
148.
964
38.7
34.
110
8.37
211
8.1
30.7
110
6.2
151.
837
39.4
7819
8913
.60
2578
.557
145.
3236
.33
4.8
109.
4211
7.14
229
.29
110.
814
7.68
536
.921
1990
10.2
026
.89
81.5
914
1.60
238
.08
5.4
110.
474
116.
188
31.2
411
3.1
145.
652
39.1
6619
919.
9025
.29
84.4
4313
8.22
434
.96
4.2
111.
516
115.
254
29.1
511
4.2
144.
688
36.5
9219
929.
9022
.93
86.3
8613
5.97
431
.18
311
2.54
611
4.33
926
.22
115
143.
993
33.0
1819
9311
.10
22.6
288
.381
133.
717
30.2
53
113.
576
113.
432
25.6
611
7.8
141.
616
32.0
3319
948.
0023
.790
.259
131.
636
31.2
2.6
114.
602
112.
536
26.6
712
1.4
138.
6532
.86
1995
6.20
25.0
792
.106
129.
631
32.5
2.8
115.
6311
1.64
727
.99
126.
113
4.92
333
.825
1996
5.40
28.2
593
.852
127.
771
36.1
311
6.66
110.
765
31.2
912
8.2
133.
285
37.6
5319
972.
5038
.43
95.4
1412
6.13
448
.47
2.3
117.
683
109.
895
42.2
313
0.5
131.
523
50.5
4419
982.
4043
.88
96.4
7212
5.03
754
.87
1.6
118.
699
109.
039
47.8
513
1.2
130.
989
57.4
7819
993.
0051
.65
97.8
6812
3.61
163
.84
2.2
119.
721
108.
186
55.8
813
312
9.63
666
.957
2000
1.40
71.6
210
012
1.47
987
3.4
120.
755
107.
329
76.8
713
5.6
127.
718
91.4
7220
0114
.50
41.2
510
2.39
911
9.13
649
.14
2.8
121.
783
106.
485
43.9
313
612
7.42
452
.562
2002
18.5
030
.75
104.
187
117.
4236
.11
1.6
122.
799
105.
658
32.4
913
5.8
127.
571
39.2
2820
0322
.10
28.3
410
6.40
411
5.33
632
.69
2.3
123.
822
104.
832
29.7
113
7.6
126.
263
35.7
8320
0417
.90
30.0
610
9.46
311
2.54
233
.83
2.7
124.
849
104.
009
31.2
714
8.3
119.
048
35.7
8620
0514
.50
31.8
111
310
9.41
134
.83.
412
5.88
310
3.18
832
.82
159.
611
1.96
835
.614
2006
9.40
35.3
511
6.56
710
6.35
137
.63.
212
6.91
510
2.37
536
.19
169.
610
6.07
237
.499
2007
7.70
44.7
511
9.68
210
3.74
946
.42
2.8
127.
943
101.
571
45.4
517
5.2
102.
875
46.0
3320
088.
6044
.812
2.08
710
1.77
945
.59
3.8
128.
981
100.
766
45.1
418
9.1
95.5
2542
.791
2009
12.9
035
.45
124.
298
100
35.4
5-0
.412
9.97
710
035
.45
181
100
35.4
5
Sources:
aIn
tern
atio
nal
Mon
etar
yF
un
d,w
ww
.imf.
org
;bU
nit
edS
tate
sB
ure
auof
Lab
orS
tati
stic
s,w
ww
.bls
.gov
/cp
i/cp
id09
av.p
df;
c Un
ited
Sta
tes
Bu
reau
ofL
abor
Sta
tist
ics,
ftp
://ft
p.b
ls.g
ov/p
ub
/tim
e.se
ries
/pc/
pc.
dat
a.59
.Rea
lEst
ate
Table VII.Rent vs vacancy inconstant dollars
JPIF29,2
134
Interviews with experienced leasing agents established that asking and contact rentsfor new construction projects tend to be 5-15 percent greater than average. This caseassumes a constant 10 percent premium for new construction beyond the average.Because of low materiality, retail and additional rents were assumed to maintain aconstant ratio to with office rents and an unchanging starting occupancy andstabilization period.
Market vacancy. Statistical analysis of the population of vacancy values returnsresults and histogram are presented in Figure 5.
The observed population for this variable lends itself well for modeling with asimulated Beta distribution and reasonably well for a triangular set to producemeaningful results.
Average inflation. As further described above, rent and construction costparameters were modeled in constant dollars, making it necessary to predict theinflation from initiation to the commencement of construction at which point costs maybe locked in. Other costs and revenues are also affected by inflation. Because near-termexpenses and revenues are affected in a relatively minor way and interim values ofinflation are not material for this analysis we can model a single average inflation over
Figure 4.Rent vs vacancy
regression
Figure 5.Market vacancy
Simulation-basedexcess return
model
135
the life of the project. As a reasonable proxy to the interim period of the project weselected a five-year moving average inflation expressed as a rate of change of the GDPdeflator. Figure 6 demonstrates the data, statistical analysis and histogram for thefive-year moving average inflation. Since no mode value was found, the mean of 2.63percent was selected as the peak value for simulation using triangular distribution.
Construction cost. Unfortunately, no publicly available historical statistics forfluctuation in local construction costs were identified. Furthermore, nationalconstruction costs as measured by PPI Construction correlated very poorly withlocal metrics available, with the highest correlation coefficient obtained not exceeding0.104. In the absence of statistical data, interviews were conducted with generalcontractors with a significant history of operating in the area. The interviews yieldedinformation that construction costs depend broadly on cost of labor and materials, thelatter being impacted by global construction demand. This is particularly true in thecase of the cost of steel. Labor cost can be expected to be correlated with local buildingdemand, which in turn is driven by market vacancy. Because this particular project isexpected to be built on a concrete frame, the price of steel becomes less relevant andlocal variables assume primacy in determining a relationship with outside factors.Interviews with contractors yielded no information regarding absolute cost levels ofconstruction but did suggest that relative change in costs is correlated with marketvacancy in a relationship that can be represented a linear relationship with the slopem ¼ 20:57971 and intercept y ¼ 0:088 as presented in Figure 7.
Loan interest. Public information about construction loan terms has been as difficultto find as it was for construction costs. Lending agents interviewed indicated that
Figure 6.Inflation
JPIF29,2
136
construction loans are generally priced 300-400 basis points above prime. An averagespread of 350 bps was selected as a benchmark for the model. Prime rates from 1947 to2009 and statistical analysis of resulting interest rates are presented in Figure 8.
Capitalization rate. Direct capitalization rates for San Francisco office propertieshave only been recorded in a publicly available way since 1999, offering a limiteddataset for ascertaining long-term patterns. However, the San Francisco cap ratesappear to closely track the national office cap rate trends as derived inChandrashekaran, 2000 for the period of 1985 to 1999, making it possible to extendthe series by 15 data points for a more statistically significant picture. The resultingrange of average cap rates is presented in Figure 9. One issue with the available sampleis the fact that the largest and the smallest values in the dataset, 3.93 percent and 10.20percent, represent single deals made during the top and bottom of the marketrespectively. These two points were therefore treated as outliers and eliminated themfrom the analysis.
Plotting cap rates against their spot market interest rate and spot GDP inflation rateresulted in poor correlation with the prime rate and, interestingly, a negativecorrelation with the spot inflation that becomes progressively smaller with the use of alonger moving average of rates. Since modeling spot inflation rates is outside the scopeof the model here, cap rates can be treated as an independently simulated variablefollowing a triangular distribution[5]. Figure 9 summarizes the statistics of the cap ratespread used.
Interviews with real estate agents in experienced in the San Francisco marketyielded information that new construction project when sold command pricesrepresented by cap rates in the lower ranges of the spread around the average value.The price premium was described as ranging between 100 and 150 basis points belowthe average value. A median value of 125 basis points below average was selected for
Figure 7.Construction cost
Simulation-basedexcess return
model
137
Figure 8.Loan interest
JPIF29,2
138
Figure 9.Cap rates
Term RateTreasury rate (years) (%)
Short term 5 2.08Intermediate term 10 3.25Long term 20 3.97
Table VIII.Treasury rates
Value Min Mode MaxIndependent drivers (%) Distribution (%) (%) (%)
Market vacancy 10.78 Triangular 1.00 9.90 22.10Average inflation 3.06 Triangular 1.60 2.63 5.25Prime rate 8.58 Triangular 1.75 4.50 21.50Exit cap rate 6.30 Triangular 4.40 6.20 8.40
Table IX.Independent simulation
drivers
Simulation-basedexcess return
model
139
Ris
k-f
ree
NP
VW
AC
CN
PV
Hu
rdle
NP
VL
ev.
gai
ns
Lev
.p
rofi
tL
ev.
RO
EU
nle
v.
RO
ER
isk
-fre
era
teW
AC
C
Min
.(2
06,8
49,8
20)
(92,
345,
744)
(79,
542,
871)
6,54
1,75
7(2
64,1
77,8
99)
297
.48%
210
0.00
%2.
30%
6.94
%M
ax.
262,
057,
754
166,
278,
328
63,9
47,6
6147
0,13
4,46
639
0,99
6,15
049
4.07
%26
6.07
%3.
88%
19.6
6%M
ean
29,7
40,0
871,
278,
197
(10,
863,
064)
156,
902,
902
45,5
89,8
9657
.85%
45.8
0%2.
85%
11.8
2%M
edia
n40
,191
,619
1,08
5,02
4(1
1,60
0,03
7)16
7,53
4,74
762
,322
,968
59.4
9%61
.69%
2.84
%11
.40%
Mod
eN
/AN
/AN
/AN
/AN
/AN
/A2
100.
00%
2.30
%N
/AA
vg
dev
.65
,411
,473
33,7
47,5
3717
,382
,149
70,5
66,4
0984
,142
,204
80.4
9%57
.98%
0.38
%2.
39%
SD
81,3
36,2
9141
,839
,430
21,6
25,6
5988
,522
,616
106,
196,
705
99.9
5%75
.29%
0.43
%2.
86%
Coe
f.v
ar.
2.73
4904
246
32.7
3316
525
21.
9907
5125
20.
5641
8724
2.32
9391
256
1.72
7706
712
1.64
3752
865
0.14
9831
912
0.24
2043
816
Sk
ewn
ess
20.
3542
4465
40.
1799
3585
40.
1326
4735
82
0.20
8035
519
20.
4613
8244
70.
1946
4725
12
0.67
3849
811
0.08
9627
981
0.49
7104
935
Ku
rtos
is2
0.43
9871
078
20.
3307
7139
62
0.21
0359
406
20.
5212
8249
82
0.27
8245
878
20.
3396
1685
20.
2963
8635
12
1.37
8681
998
20.
6330
8832
Sem
iS
D46
,340
,607
28,2
69,4
1521
,944
,920
259
,431
,972
39.6
5%39
.08%
0.00
%0.
00%
Sem
iC
oef.
var
.1.
5581
8663
822
.116
6353
12
2.02
0140
886
21.
3036
2158
68.5
4%85
.33%
0.00
%0.
00%
90p
erce
nt
inte
rval
low
erb
oun
d(1
16,2
62,3
32)
(66,
222,
812)
(44,
896,
340)
9,44
1,20
0(1
51,8
21,6
70)
(0.9
4)2
100.
00%
2.30
%7.
95%
90p
erce
nt
inte
rval
up
per
bou
nd
150,
619,
596
71,2
30,6
8725
,698
,444
293,
343,
785
198,
804,
041
2.25
146.
61%
3.49
%17
.12%
Eff
ecti
ve
du
rati
on3.
1798
1483
2P
rob
abil
ity
ofsh
ortf
all
35.7
3%48
.78%
69.2
3%3.
82%
33.3
9%E
xp
ecte
dN
PV
/re
ver
sion
25,8
30,5
2955
5,73
4(8
,030
,472
)16
1,14
1,78
641
,516
,217
Mea
nex
pec
ted
IRR
0.27
%R
isk
-ad
just
edex
pec
ted
IRR
0.14
%R
isk
-ad
just
edre
turn
onca
pit
al1,
085,
025
Sh
arp
era
tio
0.36
5643
514
0.03
0550
055
20.
5023
2292
91.
7724
6121
20.
4292
9670
90.
5788
0194
20.
6083
6395
9S
orti
no
rati
o0.
6417
7164
40.
0452
1483
42
0.49
5014
98#
DIV
/0!
0.76
7093
776
1.45
8966
981.
1719
4127
5B
ench
mar
kd
isco
un
tra
te2.
85%
11.4
0%20
.00%
Med
ian
pro
fit
40,1
91,6
191,
085,
024
(11,
600,
037)
167,
534,
747
Ex
pec
ted
med
ian
pro
fit
25,8
30,5
2955
5,73
4(8
,030
,472
)16
1,14
1,78
6M
edia
nIn
ves
tmen
t(1
27,3
43,1
28)
(166
,449
,723
)(1
79,1
34,7
84)
127,
343,
128
Tot
alp
roje
ctv
olat
ilit
y16
6.31
%48
1.02
%A
nn
ual
pro
ject
vol
atil
ity
57.8
8%16
7.39
%
Table X.Simulation results
JPIF29,2
140
use in the model. We have further elected to set a lower bound of 4.40 percent as thelowest observed cap rate excepting a single outlier.
Risk-free rate. Because the duration of the project is not fixed, and in fact was foundin simulation to vary between six and 19 years, the normal practice of selecting a singleTreasury rate as a proxy for the risk-free rate is inappropriate. The effective risk-freerate for the project is extrapolated from the available five-, ten- and 20-year Treasuryrates and is calculated based on the duration of the project in each iteration of thesimulation. The spot Treasury rates used in this case are presented in Table VIII.
Simulation analysisThe independent drivers thus selected for the simulation are market vacancy, averageinflation, prime rate and capitalization rate. Distributions, distribution modelingparameters and median simulated results are summarized in Table IX.
The simulation was run with 10,000 iterations, with the statistical results presentedin Table X[6]. As we can see, the median hurdle NPV result is negative, implying thatthe project as planned is unlikely to meet its profitability objectives. In fact, theprobability of doing so is only 33.39 percent. The probable shortfall in covering theweighted average cost of capital is 48.78 percent, but since the median return for thismetric is positive, the implication is that the project will at least not generate a cashloss for the sponsor. These may or may not be poor odds for the risk tolerance of aparticular sponsor, but they do not offer any information as to whether the size of theexpected profit justifies this level of risk. For the risk-adjusted return metric, we mustturn to SERM analysis.
SERM analysis and recommendationApplying SERM analysis to the results of the simulation, as presented in Table XI, weobtain the excess return metric ERR of -2.85 percent, which is a clear indication that therange of returns for this project does not justify the risk. In fact, once we solve for therequired rate of return MAR, we obtain a discount rate of 79.44 percent, indicating thatthe project’s viability is some distance from break-even.
Clearly, the returns from this project as presented do not justify the risks taken,barring significant mitigation of the primary risk factors such as the exit cap rate andrent and occupancy levels.
VI. ConclusionsThe SERM corrects serious shortcomings in the DCF methodology for evaluating riskin real estate development projects by incorporating stochastic tools for the
%
Downside risk (DR) 2,211.66Return per unit of risk 0.012080Risk-adjusted return 0.000032
Excess return above R-F rate (SERM ERR) 22.85365SERM required rate of return (SERM MAR) 79.44
Table XI.SERM analysis summary
Simulation-basedexcess return
model
141
measurement of the universe of outcomes. It further serves to condense the output ofMonte Carlo simulations into a simple metric that can generally understood andapplied by practitioners for making decisions concerning funding projects.Additionally, key profitability drivers are identified to be market vacancy,capitalization rate and construction debt interest, with rent, construction cost andproject schedule being functions of market vacancy.
Notes
1. Modeling variable durations of the phases of a project using spreadsheet software can bechallenging. An acceptable approximation is the use of the Microsoft Excel XNPV functionthat computes NPVs for cash flows that do not follow a regular periodic pattern. However,such a model at best an approximation because it only captures the changes due to the timevalue of money and does nothing to account for additional debt service expenses that occurbecause of extended construction and stabilization periods. An alternative method, detailedin the case study to be published, utilizes a complex algorithm for adjusting calendar cashflows based on variable project duration.
2. Tornado analysis was performed with the use of the Dartmouth Sensitivity Toolkit,available at http://mba.tuck.dartmouth.edu/toolkit/index.html
3. Figure 3 data and chart courtesy of the San Francisco office of Cornish & Carey Commercial.
4. This is not the only extraordinary risk that may impact outcomes. For instance, the currenttenant may terminate early, or adverse market conditions may manifest themselves duringearly stages of construction. Such eventualities would be analogous to the delay ofconstruction commencement but without the attendant extension of cash flow. Themateriality of the interim cash flows would dictate whether such a case would requireinclusion in simulation. Additionally, the entitlement period here assumed to be fixed may infact vary, requiring an additional delay factor.
5. See above for information on the use of triangular vs beta distribution.
6. Simulation for the case study was performed with the use of ExcelSim2003 (Mayes, 2003), aMicrosoft Excel add-in simulation engine intended for academic use. The specificcharacteristic of this engine necessitated the use of triangular distributions in simulation ofhighly skewed sample populations. Analysis of the data suggests that in practicalapplication a more accurate picture may be obtained with the use of Beta distributions inpreference; nevertheless, triangular distributions offered sufficient accuracy for the purposesof this paper.
References
Atherton, E., French, N. and Gabrielli, L. (n.d.), “Decision theory and real estate development:a note on uncertainty”, working paper, unpublished.
Baroni, M., Barthelemy, F. and Mokrane, M. (2006), “Monte Carlo simulations versus DCF in realestate portfolio valuation”, ESSEC Working Papers DR 06002, ESSEC Research Center,ESSEC Business School, Cergy-Pontoise.
Chandrashekaran, V. (2000), “The predictability of real estate capitalization rates”, available at:http://homepage.mac.com/mikero1/MSY/articlesPDF/CapRates.pdf
Mayes, T.R. (2003), ExcelSim 2003 Simulation Engine, Department of Finance, MetropolitanState College of Denver, Denver, CO, available at: http://clem.mscd.edu/,mayest/../ExcelSim%202003%20Documentation.pdf/
JPIF29,2
142
Satchell, S.E. and Pedersen, C.S. (2002), “On the foundation of performance measures underasymmetric returns”, available at: www.sortino.com/htm/satchell.pdf
Young, M.S. (2007), “Real-time valuation: breathing new life into moribund DCF modeling”,Journal of Real Estate Practice and Education, Vol. 10 No. 1, pp. 25-40.
Further reading
Akalu, M.M. and Turner, R. (2002), “A Monte Carlo comparison between the free cash flow anddiscounted cash flow approaches”, Tinbergen Institute Discussion Paper, TI 2002-083/1,Tinbergen Institute, Rotterdam.
Barman, B. and Nash, K.E. (2007), “A streamlined real options model for real estatedevelopment”, MSRED thesis, Massachusetts Institute of Technology, Cambridge, MA.
Breidenbach, M., Mueller, G.R. and Schulte, K.-W. (2006), “Determining real estate betas formarkets and property types to set better investment hurdle rates”, Journal of Real EstatePortfolio Management, Vol. 12 No. 1, pp. 73-80.
Brotman, B.A. (2010), “The impact of market conditions using appraisal models”, Journal ofProperty Investment & Finance, Vol. 28 No. 3, pp. 237-42.
Brown, R.J. (2006), “Sins of the IRR”, Journal of Real Estate Portfolio Management, Vol. 12 No. 1,pp. 195-200.
Chaplin, R. (2000), “Predicting real estate rents: walking backwards into the future”, Journal ofProperty Investment & Finance, Vol. 18 No. 3, pp. 352-70.
Comeau, J. (2009), “Incorporating uncertainty into discounted cash flow equity”, available at:www.cob.calpoly.edu/media/files/econ/Top-Senior-Project-F2009.pdf
Damodaran, A. (2008), “The promise and peril of real options”, working paper, Stern School ofBusiness, Greenwich Village, NY.
de Neufville, R., Scholtes, S. and Wang, T. (2006), “Valuing options by spreadsheet: parkinggarage case example”, ASCE Journal of Infrastructure Systems, Vol. 12 No. 2, pp. 107-11.
Fekete, I. and Katona, T. (2003), “Modelling the fulfilment of the real estate utilisation plan for2002 by Monte-Carlo simulation”, working paper, Matav Rt (Hungarian Telecom),Budapest.
French, N. (2006), “Value and worth: probability analysis”, Journal of Property Investment& Finance, Vol. 24 No. 4, pp. 374-80.
French, N. and Gabrielli, L. (2005), “Discounted cash flow: accounting for uncertainty”, Journal ofProperty Investment & Finance, Vol. 23 No. 1, pp. 75-89.
Geltner, D.M., Miller, N.G., Clayton, J. and Eichholtz, P. (2007), Commercial Real Estate Analysis& Investments, 2nd ed., Thompson South-Western, Mason, OH.
Greig, W.D. and Young, M.S. (1991), “New measures of future property performance and risk”,Real Estate Review, Vol. 21 No. 1, Spring.
Hengels, A. (2005), “Creating a large practical model using real options to evaluate large-scalereal estate development projects”, MSRED thesis, MIT Center for Real Estate, Cambridge,MA.
Hoesli, M., Jani, E. and Bender, A. (2006), “Monte Carlo simulations for real estate valuation”,Journal of Property Investment & Finance, Vol. 24 No. 2, pp. 102-22.
Holtan, M. (2002), “Using simulation to calculate the NPV of a project”, 2002, available at: www.investmentscience.com/Content/howtoArticles/simulation.pdf
Hudson-Wilson, S. (Ed.) (2000), Modern Real Estate Portfolio Management, Frank J. FabozziAssociates, New Hope, PA.
Simulation-basedexcess return
model
143
Hughes, W.T. (1995), “Risk analysis and asset valuation: a Monte Carlo simulation usingstochastic rents”, Journal of Real Estate Finance and Economics, Vol. 11 No. 2, pp. 177-87.
Kariya, T., Kato, Y., Uchiyama, T. and Suwabe, T. (2005), “Tenant management and leasevaluation for retail properties: a real options approach”, International Real Estate Review,Vol. 8 No. 1, pp. 44-82.
Lucius, D.I. (2001), “Real options in real estate development”, Journal of Property Investment& Finance, Vol. 19 No. 1, pp. 73-8.
Magni, C.A. (2008), “CAPM-based capital budgeting and nonadditivity”, Journal of PropertyInvestment & Finance, Vol. 26 No. 5, pp. 388-98.
Sabour, S.A.A. and Poulin, R. (2006), “Valuing real capital investments using the least-squaresMonte Carlo method”, Engineering Economist, Vol. 51 Nos 2, Summer, pp. 141-60.
Schurman, G. (2009), “Volatility and discount rate”, available at: www.appliedbusinesseconomics.com/files%5Cgvsolv01.pdf
Sing, T.F. (2001), “Optimal timing of a real estate development under uncertainty”, Journal ofProperty Investment & Finance, Vol. 19 No. 1, pp. 35-52.
Souza, L.A. (n.d.), “Modern real estate portfolio management (MREPM)”, doctoral dissertation,Corporate Finance, Golden Gate University, available at: http://the-commercial-group.com/uploads/3_Real%20Estate%20Portfolio%20Theory-Management.pdf/
Tsolacos, S. (2006), “An assessment of property performance forecasts: consensus versuseconometric”, Journal of Property Investment & Finance, Vol. 24 No. 5, pp. 386-99.
Waddell, P. and Ulfarsson, G.F. (2003), “Dynamic simulation of real estate development and landprices within an integrated land use and transportation model system”, paper presented atthe TRB 2003 Annual Meeting.
Wessa, P. (2009), Maximum-likelihood Beta Distribution Fitting (v1.0.4) in Free StatisticsSoftware (v1.1.23-r6), Office for Research Development and Education, available at: www.wessa.net/rwasp_fitdistrbeta.wasp/
Yang, M. and Blyth, W. (2007), “Modeling investment risks and uncertainties with real optionsapproach. Climate policy uncertainty and investment risk”, Working Paper NumberLTO/2007/WP01, IEA, Paris, February.
Corresponding authorDavid Gimpelevich can be contacted at: [email protected]
JPIF29,2
144
To purchase reprints of this article please e-mail: [email protected] visit our web site for further details: www.emeraldinsight.com/reprints