Time-To-Plan Lags for Commercial Construction Projects

NBER WORKING PAPER SERIES

TIME-TO-PLAN LAGS FOR COMMERCIAL CONSTRUCTION PROJECTS

Jonathan N. Millar

Stephen D. Oliner

Daniel E. Sichel

Working Paper 19408

http://www.nber.org/papers/w19408

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue

Cambridge, MA 02138

September 2013

The views expressed in this paper are ours alone and are not necessarily those of the Board of Governors

of the Federal Reserve System, its staff, any other institutions with which we are affiliated, or the National

Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been peer-

reviewed or been subject to the review by the NBER Board of Directors that accompanies official

NBER publications.

sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided

that full credit, including © notice, is given to the source.

Time-To-Plan Lags for Commercial Construction Projects

Jonathan N. Millar, Stephen D. Oliner, and Daniel E. Sichel

NBER Working Paper No. 19408

September 2013

JEL No. E22,E32,L50,L74,R52

ABSTRACT

We use a large project-level dataset to estimate the length of the planning period for commercial

construction projects in the United States. We find that these time-to-plan lags are long, averaging

about 17 months when we aggregate the projects without regard to size and more than 28 months

when we weight the projects by their construction cost. The full distribution of time-to-plan lags

is very wide, and we relate this variation to the characteristics of the project and its location. In

addition,we show that time-to-plan lags lengthened by 3 to 4 months, on average, over our sample

period (1999 to 2010). Regulatory factors are associated with the variation in planning lags across

locations, and we present anecdotal evidence that links at least some of the lengthening over time to

heightened regulatory scrutiny.

Jonathan N. Millar

Federal Reserve Board

20th and C Streets, NW

Washington, DC 20551

[email protected]

Stephen D. Oliner

American Enterprise Institute

1150 17th S., NW

Washington, DC 20036

and UCLA Ziman Center for Real Estate

[email protected]

Daniel E. Sichel

Department of Economics

Wellesley College

106 Central Street

Wellesley, MA 02481

and NBER

[email protected]

- 1 -

I. INTRODUCTION

The U.S. economy emerged from recession in mid-2009, but commercial construction

activity did not hit bottom until nearly two years later and even now is up only slightly from its

low. Although the recent weakness partly reflects the tight credit conditions induced by the

global financial crisis, commercial construction also lagged behind the broader recovery in each

of the three previous business cycles. Averaging across those recoveries, which date back to

1982, commercial construction reached its low about three quarters after the official business

cycle trough determined by the National Bureau of Economic Research.

The cyclical dynamics

of commercial construction matter for the macroeconomy because activity in this sector is quite

volatile and contributes significantly to economic downturns.

A plausible explanation for why commercial construction lags the business cycle is that

these buildings require substantial planning before construction can begin, which delays the

upturn in this sector. This explanation resonates with the use of gestation lags in some

macroeconomic models, starting with Kydland and Prescott (1982). Two distinct types of

gestation lags can be defined. The time-to-plan lag of a project represents the time elapsed

between the initiation of planning and the start of construction, while its time-to-build lag

represents the period from the start of construction until completion.

Although time-to-build lags have garnered most of the attention in the literature, time-to-

plan lags are also important for understanding how economic fundamentals affect construction

This calculation is based on data in the National Income and Product Accounts for real investment in commercial

and health care structures.

From 1959:Q1 to 2012:Q4, the standard deviation of quarterly percent changes in real investment in commercial

and health care structures was about five times that of real GDP. And, around the time of recessions, this sector

typically makes a negative contribution to aggregate activity that far outstrips its small share of nominal GDP (1.2

percent on average from 1959:Q1 to 2012:Q4). For example, in the latest recession, which began at the end of 2007,

real GDP dropped 4.7 percent, and the associated decline in real investment in commercial and health care structures

reduced real GDP by about 0.5 percentage point. In each of the prior two recessions, which began in 1990 and

2001, the drop in this category amounted to an even larger share of the decline in real GDP.

- 2 -

and other types of investment activity. In particular, the decision to undertake a construction

project will reflect information available during the planning period, including current and

expected economic conditions as well as a host of regulatory and other factors specific to the

locale of the project. Once construction has started, it typically becomes very costly to defer or

abandon the project if circumstances change.

Thus, the planning period is the critical window

for the formation of expectations and the perceptions of risk that drive construction spending.

Existing empirical work, however, provides very limited information about time-to-plan

lags for investment projects. One strand of this literature incorporates gestation lags into the

structure of a model and estimates the lags in conjunction with other parameters.

These studies

generally point to time-to-build lags of one to two years for structures or broader aggregates of

capital spending but are largely silent about the length of time-to-plan lags.

The second strand

of the literature directly measures gestation lags. To our knowledge, only one study, Mayer

(1960), provides direct measures of the time-to-plan lag. Using survey data on new industrial

plants and additions to existing plants, he estimated a mean time-to-plan period of 7 months and

a mean time-to-build period of 15 months. Other studies (Krainer 1968; Taylor 1982;

Montgomery 1995; and Koeva 2000) measure either the time-to-build lag or a combination of

that lag and part of the time-to-plan lag; these lags are estimated to be long ― about 1½ years or

more. None of these studies, however, separately identifies the planning period.

For a discussion of investment under uncertainty and the role of irreversibility, see Dixit and Pyndyck (1994). For

applications to commercial real estate, see Holland, Ott, and Riddiough (2000) and Sivitanidou and Sivitanides

(2000).

For estimates of gestation lags from investment models, see Oliner, Rudebusch, and Sichel (1995), Zhou (2000),

Koeva (2001), Millar (2005a), and Del Boca et al. (2008); for estimates from dynamic factor demand models, see

Palm, Peeters, and Pfann (1993) and Peeters (1998); and for estimates from business cycle models, see Altug (1989)

and Christiano and Vigfusson (2003). Other business cycle models — including Kydland and Prescott (1982),

Christiano and Todd (1996), Casares (2006), and Edge (2007) — have incorporated gestation lags that were

calibrated using information outside the model.

One exception is Millar (2005b), who found that it takes at least one year for shocks to total factor productivity to

induce changes in spending on business fixed capital, a lag that was interpreted as the time-to-plan lag.

- 3 -

This paper begins to fill that void by providing comprehensive estimates of time-to-plan

lags for commercial construction projects and by exploring the sources of variation in these lags.

We estimate time-to-plan lags using a large project-level dataset from CBRE Economic

Advisors/Dodge Pipeline, a commercial vendor of real estate data, supplemented with

information about the project's locality from the Census Bureau and other governmental sources.

The dataset covers more than 80,000 commercial construction projects in the United States from

1999 to 2010. Our goal is to present new stylized facts about time-to-plan lags from a rich and

previously untapped dataset. We do not estimate structural models that could be used to identify

causal relationships. As valuable as that would be, it is beyond the scope of this paper.

Our analysis generates four main results. First, time-to-plan lags are quite lengthy for

commercial construction projects, averaging about 17 months across the projects in our dataset.

Large projects — which account for a disproportionate share of total construction spending —

tend to have even longer lags. Indeed, when we weight the projects by their construction cost,

we find that the average time-to-plan lag associated with a given dollar of commercial

construction spending is more than 28 months. Second, time-to-plan lags vary considerably

around these averages depending on the characteristics of the building and its location. For

example, as would be expected, time-to-plan lags are longer for larger, more complex projects;

we also find that the metropolitan statistical areas (MSAs) with the longest time-to-plan lags are

concentrated in California and the Northeast corridor. Third, time-to-plan lags lengthened

significantly from 1999 to 2010, rising by an average of 3 to 4 months. This increase was

widespread, occurring for all types of buildings, in MSAs across the population spectrum, and in

most regions of the country. Finally, we find that the variation in planning lags across locations

is associated with differences in land-use regulation, and we present anecdotal evidence that

- 4 -

links at least some of the lengthening in planning lags over time to heightened regulatory

scrutiny.

The rest of the paper is organized as follows. Section II describes our data, and section

III presents the regression analysis. Section IV focuses on whether differences in land-use

regulation can account for the variation we find in planning lags across MSAs and over time.

Section V concludes.

II. DATA

We assembled our dataset using project-level data from the CBRE Economic

Advisors/Dodge Pipeline database, supplemented with information about localities from the

Census Bureau and other governmental sources. The Pipeline database includes more than

250,000 commercial construction projects planned in the United States from 1999 to 2010.

Pipeline aims to include all office, hotel, retail, and warehouse projects started since 1999 with

estimated construction costs — excluding land and design fees — that exceed $500,000 at the

time of the start.

Pipeline also includes projects meeting these criteria that have been planned

since 1999 but have not yet been started because they were abandoned, deferred, or are still in

the planning process. After accounting for missing data and other exclusions, our regression

sample consists of 82,303 projects — about 85 percent of which (69,723) were started before our

cutoff date of December 2010.

The key information in Pipeline for this study is that on project timelines. After field

representatives identify a potential new construction project, they track the dates at which the

We are not aware of other research on the effects of land-use regulation on the commercial real estate sector. On

the residential side, recent papers that have examined the effects of land-use regulations on housing supply and

home prices include Mayer and Somerville (2000), Ihlenfeldt (2007), Saks (2008), Glaeser and Ward (2009), Saiz

(2010), and Huang and Tang (2012). For reviews of this literature, see Quigley and Rosenthal (2005) and Gyourko

(2009).

This cost threshold is not a hard lower bound, as nearly 5 percent of the projects in our sample had nominal

construction costs below $500,000.

- 5 -

project transitions through its planning process. The planning timeline begins with the

preplanning phase, when the developer has announced an intention to build but has not yet hired

an architect, then transitions to the planning phase, when architects are hired to draw up

schematics for the building, and to the final planning phase, when specific plans have been (or

are about to be) finalized.

During these three phases, the developer also completes the many

legal steps (such as holding public hearings and obtaining zoning approvals from local

governments) needed to secure regulatory approval for the project. Once the planning phases are

completed, Pipeline representatives record when construction on the project started. If the

developer decides to defer the project or abandon it altogether at any time before completion,

these dates are recorded as well.

The Pipeline database shows the exact month and year during which construction on the

project began, but the other dates on the project timeline are recorded with less precision. These

other dates are less precise because the Pipeline field representatives are only required to check

on each project once every six months during the planning period. Thus, the dates recorded for

the beginning of each phase prior to the start represent the month and year during which the

representative discovered the change of status, rather than the actual date of the change. For this

reason, the reported dates for all changes in status before the start represent the end of a six-

month interval in which the actual change occurred.

We account explicitly for this interval

reporting in our empirical work.

In addition to these phases, Pipeline also documents when bids were solicited from general contractors, when a

contractor was hired, when construction permits were obtained, and when the project was completed.

Although a project could, in principle, be deferred or abandoned multiple times before completion, only the first

instance of each of these events is recorded in Pipeline.

In some cases, the interval containing the status change can be narrowed further by the constraint that planning

cannot occur after the measured start date. For instance, if a measured planning date is three months after the listed

start date, the actual planning date must have occurred within the three-month interval preceding the start, and so on.

- 6 -

For our analysis, we define the time-to-plan lag for started projects as the number of

months between the beginning of the planning phase, subject to the interval reporting issue just

noted, and the date of the construction start. The decision to exclude the earlier preplanning

phase was dictated largely by data availability. Dates for the preplanning phase were available

only for the small fraction of projects that could be identified at their inception through industry

contacts.

These projects tended to be much larger and more costly than the median project in

the dataset. All told, we were able to determine the beginning of the planning phase for nearly

90,000 projects that were started prior our cutoff in December 2010, compared to only about

12,000 projects based on the earlier preplanning date.

Many of the projects in Pipeline had not yet been started by our cutoff date of December

2010. Excluding such records would cause our sample to underrepresent projects with longer

planning durations, thereby exposing our results to truncation bias. To address this issue, we

employ an estimation method that regards the time to plan for unstarted projects as being right-

censored, with a true planning duration at least as long as the number of months between the

planning date and the cutoff date. Although this approach deals with the potential for truncation

bias, it can introduce a separate bias in the opposite direction. In particular, some unstarted

projects may be very unlikely to move forward, yet remain in the Pipeline database because the

developer has not formally declared the project abandoned. To deal with these potential

"zombie" projects, we exclude from our sample any unstarted projects whose status at the

December 2010 cutoff date had been listed as deferred for more than 60 consecutive months.

Field representatives attempt to identify projects as early as possible by canvassing neighborhoods, exploiting

information from industry contacts (such as developers, architects, and contractors), and — as a last resort —

through information from local permit offices.

We observe both the planning and preplanning dates for about 5500 projects that were started before December

2010. The beginning of the preplanning phase for this limited set of projects was about 8 months earlier, on

average, than the beginning of the planning phase, while the median difference between the two dates was 5 months.

- 7 -

We also excluded any projects that were listed as abandoned prior to the cutoff date, as

abandoned projects very rarely are resumed at a later date.

To account for the effects of project-specific factors on time-to-plan lags, we assembled

data on a number of project characteristics. From the Pipeline database, we obtained the location

of each project (including its MSA, county, and geocode), building type (office, hotel, retail, or

warehouse), construction type (new buildings or additions, alterations, or conversions of existing

structures), number of buildings, total number of floors, square footage, and dummy variables

indicating whether the project had been deferred prior to groundbreaking. For each project, we

calculated distance to the city center using geocodes for the project’s location and the

employment-weighted center (from the Department of Housing and Urban Development).

also obtained Pipeline data for each project’s construction costs in current dollars, which were

converted to real terms using the price deflator for nonresidential buildings from the National

Income and Product Accounts that prevailed at the planning date.

These project-specific variables were supplemented with MSA- and county-level controls

from the 2000 Decennial Census. Specifically, we use MSA-level estimates of population, and

county-level estimates of the average number of persons per household, the urban share of the

population, the homeownership rate for occupied housing units, the median house price, median

household income, and the share of the households who had high income (annual income above

$100,000). These characteristics could influence land-use regulations across localities and

thereby cause planning durations to differ across projects that are identical except for the

jurisdictions in which they are located.

We used information from outside sources (such as company reports) to verify whether some exceptionally large

deferred projects in Pipeline actually had been abandoned. This search eliminated one $5 billion project from our

sample that would not have been excluded by other criteria.

In a handful of instances for which geocodes for the MSA city center were unavailable from HUD, we used the

location of city hall from Google Maps as the city center.

- 8 -

Table 1 reports the distribution of projects along several dimensions. As shown in the top

part of the table, most of the projects in our sample had their primary property type listed as

office or retail, with the remainder split between warehouse and hotel properties.

About

95 percent of the projects were for new buildings, with the remainder involving some

combination of additions to existing buildings, alterations that do not affect square footage, or

conversions of property type (such as transforming retail space to office). Geographically, the

South Atlantic division, which stretches from Maryland to Florida, accounts for more than

20 percent of the projects, while New England contains about 5 percent, reflecting its much

smaller population. The shares for the other divisions are all between 10 and 20 percent. About

90 percent of the projects were located in MSAs in the top population quartile; the 25 most

populous MSAs alone contain nearly 40 percent of the observations in the dataset and account

for roughly half of total construction cost.

As shown in table 2, the projects in our sample vary substantially by size, cost, proximity

to city center, and characteristics of the surrounding county. Although the typical project in our

sample involved the construction of a modest one-story building, the indicators of project size

(number of buildings, number of floors, and construction cost) have pronounced right tails that

push the mean values above the medians. In the location dimension, projects range from being

nearly at the city center to more than 65 miles away. As can be seen in the lower panel of the

table, the projects are located in counties with widely varying characteristics. The counties in the

dataset tend to be somewhat more urban and to have higher housing density than the overall

mean in the 2000 Census. However, the mean values for all the other county characteristics are

similar to the national averages in that year.

The shares of construction cost are roughly consistent with those implied by Census Bureau estimates of

aggregate nominal construction spending for these categories over the same period.

- 9 -

III. REGRESSION ANALYSIS

We use the project-level dataset described above to estimate how our control variables

affect time-to-plan lags and to characterize the distribution of these lags. To estimate the

regression, we employ a maximum likelihood procedure that accounts for both the interval

reporting of the initial planning date and the right-censoring of the measured planning duration

for unstarted projects, under the assumption that the true residuals are distributed normally.

A. Explanatory Variables

The explanatory variables for the regression consist of characteristics of the project itself,

characteristics of the county in which the project is located, and time and MSA fixed effects.

The project and county characteristics include all those shown in table 2; for each variable, the

regression contains both linear and quadratic terms to allow for nonlinear effects. We also

include dummy variables for the type of building (retail, warehouse, or hotel properties, with

office buildings as the omitted category); type of construction (additions, alterations, or

conversions, with new construction as the omitted category); and whether the project was ever

deferred. We interact the deferral dummy with both linear and quadratic terms for the project's

square footage to allow for the possibility that deferrals lengthen the planning period by differing

amounts for small and large projects.

We use the intreg routine in Stata. The Pipeline database includes many projects that we omitted from the

regression sample because of missing information for some variables, which raises the possibility of sample

selection bias. Although the intreg routine does not accommodate the usual Heckman two-stage procedure to assess

sample selection, we implemented the Heckman estimation in an OLS framework. We found both that the inverse

Mills ratio was insignificant and that the coefficient estimates were quite similar to the OLS estimates without the

Heckman correction. In light of these results, we did not pursue sample selection issues further. The results from

the Heckman estimation are available from the authors on request.

Variations in economic uncertainty — both over time and across MSAs — also could help to explain variation in

time-to-plan lags. In particular, greater uncertainty could lead to the deferral of a project start, thereby extending the

planning period. The role of uncertainty, however, extends beyond the scope of this paper; in research in progress,

we use hazard models to investigate the effect of uncertainty and other factors on the decision to break ground on a

commercial construction project.

- 10 -

The remaining variables in the regression are time and MSA fixed effects. We include a

year fixed effect for the year in which project planning began (with 2004 as the omitted year)

and a month fixed effect for the same event (with June as the omitted month). The regression

also includes 362 MSA fixed effects (with Atlanta as the omitted MSA). One of the 362 fixed

effects covers about 4900 projects located in MSAs with a population of less than 50,000 in the

2000 Census or for which the MSA could not be determined using location fields in Pipeline.

For ease of interpretation, we specify our regression so that the constant term represents

the planning lag for a baseline project. This baseline project is defined from the omitted

categories for the dummy variables, the omitted fixed effects, median characteristics of projects

in our sample, and county characteristics for Fulton County, Georgia (which contains the center

of Atlanta, the omitted MSA). The constant captures median project characteristics because we

normalize every non-dummy project variable (except distance to city center) prior to estimation

by subtracting its median across all projects in the sample; similarly, we normalize every county

characteristic by subtracting the value for Fulton County. With these conventions, the constant

represents the fitted planning lag for a new office project with median characteristics, located in

a city center with observed county characteristics akin to those of central Atlanta, for which

planning began in June 2004 and proceeded without deferral or abandonment.

B. Baseline Planning Lag and Effects of Characteristics

Table 3 presents the estimates, along with bootstrap standard errors, of the constant term

and the coefficients for all of the project and county characteristics. To begin, the constant term

indicates that the baseline project has a planning period of 16¼ months; this baseline planning

lag is estimated fairly precisely, with a 95 percent confidence band that runs from about

Note that we omitted dummy variables for the Census divisions and the MSA population groups shown in table 1.

Both of these sets of dummy variables are perfectly collinear with the set of individual MSA dummies and thus

cannot be identified separately from the MSA fixed effects.

- 11 -

15 months to 17½ months. This result confirms that the planning lags for a typical commercial

construction project is lengthy. The differences in the estimated planning lags for the various

types of buildings and types of construction are relatively small, though some of the differences

from the baseline are statistically significant. Among the significant results, the planning lags for

retail buildings and hotels are roughly one-half to a full month longer than for office buildings,

while the planning lag for additions to existing structures is a bit more than one month shorter

than for new construction.

Moving to the next block of the table, the dummy variable for project deferral has an

enormous effect on total planning time. Deferral adds about 25 months to the planning lag for a

project with median square footage.

The effect of deferral rises to about 29 months for the

largest projects in the dataset, those at the 99th percentile of the distribution of square footage.

Accordingly, projects that were ever deferred impart a long right-hand tail to the distribution of

planning lags for the full sample.

The variables measuring project size and complexity –– the number of buildings, number

of floors, square footage, and cost per square foot — all have positive and significant coefficients

on the linear terms, combined with negative and mostly insignificant coefficients on the

quadratic terms. One county characteristic, the median home price, has the same pattern, with

significant coefficients on both the linear and quadratic terms. However, the coefficients on all

the other project and county characteristics are insignificant.

To assess the quantitative implications of these results, table 4 uses the linear and

quadratic coefficients to calculate the change in the time-to-plan lag as each project or county

characteristic increases from the 1

percentile value of its distribution to the 99

percentile value.

This effect is shown by the first deferral coefficient in the table because the square footage variable is defined as

the deviation from the median.

- 12 -

The table also shows the 95 percent confidence interval for the effect of the change in each

variable.

Starting with the project characteristics, increasing the number of buildings in the project

from the 1

to the 99

percentile value lengthens the planning period by 6.5 months, all else

equal, with a 95 percent confidence band that runs from 5 months to 8 months. A parallel

change in the number of floors adds 7.2 months to the planning lag, with a slightly narrower

confidence band than for the number of buildings. Boosting the total square footage of a project

with a fixed number of floors and buildings also lengthens the planning period but by less than

those two variables. Thus, as would be expected, large projects take substantially longer to plan

than small projects. In contrast, a higher cost per square foot has only a small effect on time to

plan in the presence of our other regression controls, while the effect of distance from the city

center is negligible.

Among the county characteristics, median home price strongly affects planning times.

Moving from the 1

percentile to the 99

percentile of the home price distribution lengthens the

planning period by nearly 10 months, all else equal, with a 95 percent confidence band that runs

from about 6 months to 13½ months. This result is consistent with the notion that development

plans receive heightened scrutiny when county homeowners have a lot at stake through the value

of their homes.

The only other county characteristic with a substantial effect on the planning

period is median income. Surprisingly, counties with the highest median income have

considerably shorter planning lags than counties at the low end of the income distribution. The

effects of all the other county characteristics — housing density, the urban share of population,

the homeowner share, and the high-income share — are both small and statistically insignificant.

See Fischel (2001) and Saiz (2010).

- 13 -

We would caution, however, against interpreting the effects of these county

characteristics too strictly. First, even though our regression sample is large, the amount of

information used to estimate these effects is much more limited than for the project

characteristics. Our dataset contains 1593 counties, compared with roughly 82,000 projects.

Moreover, because our regression controls for MSA fixed effects and because the county

characteristics do not vary across time, the county effects are identified solely by cross-sectional

variations within the 362 MSAs in our sample. Second, the county characteristics could well be

endogenous. In particular, the factors influencing the development process in a county likely

affect both the length of the planning period and most, if not all, of the county characteristics in

our regression. To address this potential endogeneity, we estimated an IV version of our

regression with instruments for the county characteristics that included indicators of geographic

constraints on buildable land in the spirit of Saiz (2010) and demographic characteristics — such

as the age distribution of county residents — that should be much less affected by the regulatory

environment. Although the instruments pass standard specification tests, some of the estimated

county effects seemed implausible.

All in all, we would regard the county characteristics as

useful controls in the regression but would not attach a strong causal interpretation to the specific

results.

C. Variation in Planning Lags Across MSAs and Over Time

The "heat map" in figure 1 displays the estimated planning duration for the 362 MSAs in

our sample for a project with the baseline set of characteristics. For each MSA, the planning

For example, the IV results imply that moving from the 1st percentile to the 99th percentile of the home price

distribution would lengthen the time-to-plan period by roughly 40 months, four times the effect reported in table 4.

In addition, median household income no longer had a significant effect on the planning lag, and the effect of the

high-income share became negative and strongly significant. It is difficult to explain why counties with a large

share of high-income residents would have substantially shorter planning lags than an otherwise identical county

with relatively few such residents. Additional information about the IV regression is available from the authors on

request.

- 14 -

duration equals the sum of the regression constant plus the coefficient on the MSA dummy.

MSAs with longer planning times are represented by deeper shades of red. As shown, the MSAs

with the longest planning times are concentrated in California and the Northeast corridor, while

planning times generally are shorter in the interior of the country. We explore possible sources

of this variation in section IV below.

Table 5 presents results for the month and year effects in the regression. Looking first at

the monthly patterns, planning periods tend to be slightly longer — by about one-half to a full

month — when the planning starts late in the year rather than in June, the omitted month. The

reason for this minor seasonal pattern is not clear, though perhaps project planning gets a slow

start when it begins shortly before the year-end holidays.

The year effects are much more pronounced. Planning lags have increased over time,

with projects initiated in 2010 having planning periods that were roughly 3 months longer than

projects undertaken in 1999, the first year of our sample. If we measure the change instead from

2000, the year with the smallest estimated year effect, the increase in planning lags through 2010

was about 4 months. An F-test overwhelmingly rejects the null hypothesis that the year fixed

effects are equal in all years.

An important question is whether the upward trend is pervasive or is concentrated in a

particular set of observations, such as a single region of the country or just one type of

construction. We address this question by augmenting the previous regression with variables

formed by interacting the year effects with three other sets of factors: the building-type dummies

defined above, MSA population dummies, and geographic dummies for the nine Census Bureau

To examine whether the year effects reflect macroeconomic conditions, we added two explanatory variables to the

regression: the unemployment rate and the change in private payroll employment. Both variables are measured in

the project's MSA in the month when project planning was initiated. Neither variable was statistically significant,

and the year effects were little changed by adding these variables. In section IV below, we present some anecdotal

evidence that the upward trend in the year effects could be related to increased regulatory scrutiny of commercial

development projects.

- 15 -

divisions. To form the population dummies, we grouped the MSAs in our sample into the 25

most populous MSAs, the rest of the highest population quartile, the second quartile, and the

combination of the third and fourth quartiles. One variable from each set of these dummies must

be omitted to avoid perfect collinearity with the baseline year dummies. We omitted office

buildings, the largest 25 MSAs, and the South Atlantic Census division (which contains Atlanta,

the omitted MSA in the original regression). In this specification, the year dummies without any

interactions represent the year effects for office buildings in the largest MSAs in the South

Atlantic region. The interaction terms show the deviations from this baseline set of year effects

for projects with different characteristics.

We summarize the results in table 6 and report the underlying coefficient estimates for

the plan-year effects in the appendix.

The first column shows the change in the year effect

between 1999 and 2010, the endpoint years for our sample. This comparison, however, may not

adequately reflect the underlying trend in time-to-plan lags if the year effects for individual years

are volatile. To address this possibility, the second column presents the difference between the

average year effect in the first three years of the sample and that in the final three years.

As shown in the first column, planning lags rose for all four types of buildings over the

sample period, with increases of about 7 months for hotel projects, slightly more than 4 months

for office and warehouse projects, and 2½ months for retail stores. In addition, planning lags

increased 3 to 5 months for each of the MSA population groups. All of these increases are

statistically significant. Regionally, however, the results for the uptrend in planning lags are

more varied. Planning lags rose more than 8 months on the Pacific coast and 4 to 5 months in

the Mountain states and along the Middle and South Atlantic coast. These changes are all

significant at the one-percent level. In contrast, the planning lag rose much less in the four

The complete results for this augmented regression can be obtained from the authors on request.

- 16 -

central regions and New England, and none of these changes is significant. When we average

the three years at the beginning and end of the sample period, the increase in the year effects

becomes significant in New England and two of the central regions, but otherwise the results do

not change notably. All told, we find that the shift toward longer planning lags was concentrated

in the Far West and on the East coast but encompassed all types of buildings and MSAs of all

sizes.

D. Summing Up: The Full Distribution of Planning Times

Thus far, we have explored a variety of factors that affect planning times, generally from

the perspective of a standardized baseline project. Now we take a more comprehensive view and

ask: "What is the full distribution of planning times across all the projects in our sample?" The

construction of this distribution would be straightforward if we knew the true time to plan for

each project. However, we do not observe this information because of the six-month interval

reporting in the data and the right-censoring for projects still in the planning phase at the end of

the sample period. Given both of these measurement issues, we can only place bounds on the

true time to plan for a given project.

Despite this limitation at the project level, we can use results from our interval regression

to construct an overall distribution that reflects the variation within these known bounds.

Specifically, we form a notional planning lag for every project equal to the fitted value from the

regression plus a random draw of the error term from a truncated normal distribution consistent

with these measurement bounds, where the error variance is taken from the fitted regression. We

then assemble an overall distribution from the notional planning lags.

One question is whether these results are driven by the severe 2007-09 recession and the subsequent weak

recovery, which could have caused developers to delay the transition from the planning phase to the construction

phase of their projects. If this were the case, the null hypothesis of equal year effects would not be rejected for tests

that end in 2007. In fact, we found that the results from tests through 2007 were similar to those in table 6, showing

that the uptrend in the year effects existed before the recession.

- 17 -

As shown in the top panel of figure 2, this distribution of planning times is extremely

wide and sharply skewed to the right. The long right-hand tail reflects, in large part, the

influence of projects that have had deferrals –– recall from table 3 that deferral adds, on average,

more than two years to the planning period. This figure also illustrates the importance of

including censored projects in our sample. These projects, represented by the dark portion of the

bars, are concentrated in the right-hand part of the distribution and account for the vast majority

of projects with planning times of 60 months or more. Hence, excluding these cases would cause

projects in the upper portion of the planning lag distribution to be substantially under-

represented.

The distribution in the top panel counts the number of projects in each time-to-plan

bucket and does not distinguish between small and large projects. The bottom panel, in contrast,

weights each project by its share of the total cost of all projects in the sample. This cost-

weighted distribution shows the planning time for each dollar of construction spending and thus

is a more appropriate measure for assessing the macroeconomic consequences of planning lags.

As shown, the cost-weighted distribution is skewed even more sharply to the right than the

unweighted distribution. Given the long right tail, the mean cost-weighted time-to-plan lag is

more than 28 months. Even the typical (i.e. median) cost-weighted project has a time-to-plan lag

of 18 months. These statistics indicate that commercial buildings generally have a lengthy

planning period before any construction dollars are spent.

Table 7 provides additional detail about mean time-to-plan lags along two dimensions:

the influence of deferred projects and differences across types of buildings. First, the table

shows that deferrals are an important factor behind the lengthy mean planning lags. Excluding

projects with any period of deferral shortens the unweighted mean lag from about 17 months to

about 14 months for the aggregation across all building types. The effect of excluding deferred

- 18 -

projects on the cost-weighted mean lag is even greater. Second, among the four types of

buildings, hotels have the longest mean planning periods — roughly 8 to 9 months longer than

office, retail, and warehouse projects on an unweighted basis and as much as 14 months longer

with cost weighting. Much of this difference owes to the differential effect of deferrals across

the four types of buildings. As can be seen, excluding deferred projects cuts the differential

about in half on an unweighted basis and eliminates most of the gap with cost weighting. The

differences that remain mostly reflect the relatively large scale of the typical hotel project.

IV. INTERPRETATION AND ADDITIONAL RESULTS

This section takes a closer look at two key results from section III: the wide variation in

planning times across MSAs for a baseline project and the upward trend in planning times over

our sample period. In particular, we explore whether differences in land-use regulations — as

measured by the Wharton survey on residential land-use regulation reported in Gyourko, Saiz,

and Summers (2008) — are associated with the variation in planning lags. We also summarize

what we learned from our own consultations with industry experts on changes in planning

periods over time.

A. Wharton Survey on Residential Land-Use Regulation

The Wharton survey asked officials in roughly 6900 municipalities across the country to

provide information about their process for regulating residential land use and about the

outcomes of that process.

The survey was mailed in June 2004, and responses were received

from about 2650 municipalities, which represent a finer level of geography — Census places —

than the MSA level. Gyourko, Saiz, and Summers (2008) used the results to create an overall

The Wharton survey did not collect information on land-use regulations for commercial property. In the absence

of such information, we assume that the regulations for residential land use are a good proxy for the unobserved

regulations on the commercial side. This assumption seems reasonable in that the rules governing each type of

property likely would reflect general community preferences toward development.

- 19 -

index of the stringency of land-use regulation in each Census place, along with eleven separate

component indexes.

Table 8 describes the eleven components of the aggregate index. The first four

components pertain to the length or intensity of the project approval process. The next four

components reflect the local rules that define permissible development activity. The final three

components measure the extent of local political influence on development, state-level political

involvement in local project decisions, and the tendency of the courts to uphold local land-use

regulation. Both the overall index and all the components are defined so that higher values

correspond to a tighter regulatory environment.

B. Variation in Planning Periods across Localities

To assess these regulatory effects, we re-ran the initial regression from section III using

fixed effects for Census places rather than the MSA fixed effects. After accounting for missing

or miscoded location information in the Pipeline database, the dataset for this regression included

74,409 projects and 5,984 places. However, most of these places had only a handful of projects,

so we ran the regression with separate fixed effects for the 1,712 places that had at least ten

projects and with a single catch-all fixed effect for all the other places. Recall that the fixed

effect represents the estimated time-to-plan lag in a given locality for our baseline project. After

setting aside the places that had no Wharton survey data and those in the catch-all group, we

regressed the fixed effects in the remaining 666 places on either the overall Wharton index or the

individual components of the index.

In each case, we estimated the second-stage regression in

We omit the local assembly index because that variable has no variation across the places with projects in our

sample.

- 20 -

two ways –– first without weights on the individual places and then by weighting the data for

each place by its share of the total project cost in the sample.

The results are reported in table 9. Regardless of whether we weight the data, the

estimated coefficient on the aggregate Wharton index is insignificant and the regression R

essentially zero. However, when we unpack the components of the aggregate index, the

explanatory power of the regression increases, especially for the cost-weighted regression. In

that regression, two components have significant positive coefficients and four have significant

negative coefficients. Among these six components, we focus on the coefficient on the approval

delay index (ADI) because we believe it is the only one with a straightforward interpretation.

The ADI measures the length of the review process in a locality and thus can be compared to the

fixed effects to see if both tend to be longer in the same places. The positive coefficient in the

regression provides evidence of a link between the regulatory review process and planning

timelines. Given the coefficient estimate, moving from the 1

percentile to the 99

percentile of

the ADI distribution raises the planning time for a given place by 7 months, with a tight

confidence interval. This result suggests that the length of the review process is associated with

a substantial part of the cross-sectional variation in planning times.

Unlike the ADI, the other significant index components are not measures of the time

required for planning review but instead characterize aspects of the regulatory environment that

could influence planning times. Because of the reduced-form nature of the regression, their

channels of influence are unclear. For these other index components, we can conclude only that

the regulatory environment is correlated ― in possibly complex ways ― with planning

timelines.

As shown in Saiz (2010), the stringency of land-use regulation, as measured by the Wharton index, reflects a

variety of characteristics of each locality and thus is endogenously determined. Accordingly, the regressions

estimated here have no structural interpretation and are intended only to assess whether project planning lags are

correlated with regulatory variables.

- 21 -

C. Planning Periods over Time

Although most of the information collected from the Wharton survey is cross sectional,

the survey includes a few questions about changes over time. In particular, one question asks:

"Over the last 10 years, how did the length of time required to complete the review and approval

of residential projects in your community change?" The possible responses are "no change,"

"somewhat longer," and "considerably longer," with separate responses for single-family and

multifamily projects. The question did not allow respondents to indicate that review periods had

become shorter. Because of this asymmetry, we cannot use the responses to examine our finding

that planning times became longer, on average, for the nation as whole. However, by comparing

the responses across Census divisions, we can assess whether the Wharton survey provides

independent confirmation of our finding that the shift toward longer planning periods was

concentrated on the East Coast and in the West.

To do this, we created an index of the Wharton responses at the Census division level.

For each of the 666 places in the Wharton regression sample, we converted the responses of "no

change," "somewhat longer," or "considerably longer" to values of zero, one, or two,

respectively, and then averaged the values for single-family and multifamily projects. Next, we

aggregated the results to the Census division level in two ways –– first as an unweighted average

of the included places and then by weighting each place by its share of the number of Pipeline

projects in that division. The unweighted average captures the general perception among the

survey respondents in each Census division, but it makes no allowance for differences in the

importance of each Census place to overall activity. Weighting each place by its number of

projects helps on the latter score, but a potential downside is that it emphasizes the views of

some survey respondents more than others, even though this weighting does not reflect their

- 22 -

degree of expertise about the regulatory review and approval process. On a priori grounds, we

could not see a clear case for one method over the other, so we tried both.

We regressed the 1999 to 2010 change in the plan-year effect by Census division on each

of the two Wharton proxies for the change in review and approval times. The results differed

across the two Wharton measures. The unweighted measure has a significant positive

relationship across Census divisions with the change in the estimated plan-year effects. In

contrast, the project-weighted measure shows a much looser and insignificant positive

relationship. Overall, these results provide limited confirmation that the regional pattern of

changes in planning lags is consistent with the regional pattern implied by the responses to the

Wharton question about changes in the length of review and approval times.

To gain further perspective on the upward trend in planning durations, we consulted with

firms that are directly involved in real estate development and with informed industry observers.

Although some of the comments reflected conditions across the country, a substantial fraction of

the real estate firms we contacted work primarily in the Washington, D.C. area, so the responses

are skewed somewhat toward that market.

An overwhelming majority of these contacts indicated that planning timelines had

lengthened over our sample period, consistent with our econometric results. In addition, several

contacts noted that the trend toward longer planning periods had started a decade or more before

the beginning of our sample period. Among the factors cited for this trend, by far the most

common was that the regulatory process for the review and approval of construction projects had

become more time-consuming. The specific features of the regulatory process seen as

contributing to this trend included greater citizen involvement in project review, tougher

environmental standards, and an increase in the number of government agencies whose approval

is required. Of course, it is important to emphasize that, even if changes in land-use regulations

- 23 -

do account for the increase in planning lags over time, we are not able to assess whether these

changes reflect a shift to over-regulation, a catch-up from too light regulation, or a change over

time in the optimal amount of regulation.

V. CONCLUSION

Gestation lags have long been understood to be an important feature of the investment

process. However, previous research has focused on the time-to-build part of the gestation

period and has provided very little information on the earlier time-to-plan lag. In addition, most

of what we know about gestation lags has come from indirect inference in structural models

estimated with aggregate data. Only a handful of studies have estimated gestation lags using

project-level information.

This paper addresses both of these limitations. We estimate time-to-plan lags for

commercial construction projects using a rich project-level dataset that allows direct observation

of these lags. Our analysis demonstrates that time-to-plan lags for commercial construction

projects are long, averaging about 17 months when we aggregate all the projects in the dataset

without regard to size. When we weight the projects by their construction cost — which is

needed to measure the average planning time for a given dollar of commercial construction

outlays — the average lag rises to more than 28 months.

Our results also show that the distribution of time-to-plan lags spans a wide range, with

an especially long right-hand tail. The characteristics of the building to be constructed and its

location account for part of this variation, while project deferrals also contribute importantly to

the long tail. Another key result is that time-to-plan lags increased by several months, on

average, over the 1999-2010 period that we study. This lengthening occurred for all types of

buildings, in MSAs of all sizes, and in most regions of the country. Finally, we find that

- 24 -

differences in the regulatory environment across jurisdictions are associated with the cross-

sectional variation in time-to-plan lags, and we present some anecdotal evidence that the upward

trend in planning lags may be related as well to the regulatory review process. As noted, our

results do not say whether the increase in time-to-plan lags reflects a move toward or away from

the optimal amount of regulation.

These results contribute to the literature in both macroeconomics and urban/real estate

economics. Macroeconomists can use the results to calibrate business cycle models and to help

specify the lag structure in models of investment spending. For urban and real estate economists,

our findings provide new information on the geographic variation in planning periods for

commercial real estate projects and on the influence of the regulatory review process at the local

level.

The Pipeline database is a valuable source of information about the planning process for

commercial construction projects. We know of no similar data for other types of investment

spending. Accordingly, the Pipeline data have the potential to provide new insights into the

factors affecting firms’ decisions to continue, defer, or abandon investment projects. In research

in progress, we are using the Pipeline data to study the effect of uncertainty on these decisions in

the commercial real estate sector.

REFERENCES

Altug, S. “Time-to-Build and Aggregate Fluctuations: Some New Evidence.” International

Economic Review, 30(4), 1989, 889-920.

Casaras, M. “Time-to-Build, Monetary Shocks, and Aggregate Fluctuations.” Journal of

Monetary Economics, 53(6), 2006, 1161-76.

Christiano, L. J., and R. M. Todd. “Time to Plan and Aggregate Fluctuations.” Federal Reserve

Bank of Minneapolis Quarterly Review, 20(1), 1996, 14-27.

- 25 -

Christiano, L. J., and R. J. Vigfusson. “Maximum Likelihood in the Frequency Domain: The

Importance of Time-to-Plan.” Journal of Monetary Economics, 50(4), 2003, 789-815.

Del Boca, A., et al. “Investment and Time to Plan and Build: A Comparison of Structures vs.

Equipment in a Panel of Italian Firms.” Journal of the European Economic Association, 6(4),

2008, 864-89.

Dixit, A. K., and R. S. Pindyck. Investment Under Uncertainty. Princeton, NJ: Princeton

University Press. 1994

Edge, R. M. “Time-to-Build, Time-to-Plan, Habit-Persistence, and the Liquidity Effect.”

Journal of Monetary Economics, 54(6), 2007, 1644-69.

Fischel, William A. 2001. The Homevoter Hypothesis: How Home Values Influence Local

Government. Cambridge, MA: Harvard University Press.

Glaeser, E. L., and B. A. Ward. "The Causes and Consequences of Land Use Regulations:

Evidence from Greater Boston." Journal of Urban Economics, 65(3), 2009, 265-78.

Gyourko, J. "Housing Supply." Annual Review of Economics 1, 2009, 295-318.

Gyourko, J., A. Saiz, and A. Summers. “A New Measure of the Local Regulatory Environment

for Housing Markets: The Wharton Residential Land Use Regulatory Index.” Urban Studies,

45(3), 2008, 693-729.

Holland, A. S., S. H. Ott, and T. J. Riddiough. “The Role of Uncertainty in Investment: An

Examination of Competing Investment Models Using Commercial Real Estate Data.” Real

Estate Economics, 28(1), 2000, 33-64.

Huang, H., and Y. Tang. "Residential Land Use Regulation and the US Housing Price Cycle

Between 2000 and 2009." Journal of Urban Economics, 71(1), 2012, 93-9.

Ihlanfeldt, K. R. "The Effect of Land Use Regulation on Housing and Land Prices." Journal of

Urban Economics, 61(3), 2007, 420-35.

Koeva, P. “The Facts About Time-to-Build.” IMF Working Paper WP/00/138, 2000.

Koeva, P. “Time-to-Build and Convex Adjustment Costs.” IMF Working Paper WP/01/9, 2001.

Krainer, R. E. “The Time Profile of Capital Accumulation in the Automobile Industry.” Journal

of Political Economy, 76(5), 1968, 1049-57.

Kydland, F. E., and E. C. Prescott. “Time to Build and Aggregate Fluctuations.” Econometrica,

50(6), 1982, 1345-70.

Mayer, C. J., and C. T. Somerville. "Land Use Regulation and New Construction." Regional

Science and Urban Economics, 30(6), 2000, 639-62.

- 26 -

Mayer, T. “Plant and Equipment Lead Times.” Journal of Business, 33(2), 1960, 127-32.

Millar, J. N. “Gestation Lags and the Relationship Between Investment and Q in Regressions.”

Federal Reserve Board, Finance and Economics Discussion Series Paper 2005-28, 2005a.

Millar, J. N. “Gestation Lags for Capital, Cash Flows, and Tobin’s Q.” Federal Reserve Board,

Finance and Economics Discussion Series Paper 2005-24, 2005b.

Montgomery, M. R. “’Time-to-Build’ Completion Patterns for Nonresidential Structures, 1961-

1991.” Economics Letters, 48(2), 1995, 155-63.

Oliner, S., G. Rudebusch, and D. Sichel. “New and Old Models of Business Investment: A

Comparison of Forecasting Performance.” Journal of Money, Credit, and Banking, 27(3), 1995,

806-26.

Palm, F. C., H. M. M. Peeters, and G. A. Pfann. “Adjustment Costs and Time-to-Build in Factor

Demand in the U.S. Manufacturing Industry.” Empirical Economics, 18(2), 1993, 639-71.

Peeters, M.. “Persistence, Asymmetries, and Interrelation in Factor Demand.” Scandinavian

Journal of Economics, 100(4), 1998, 747-64.

Quigley, J. M., and L. A. Rosenthal. "The Effects of Land Use Regulation on the Price of

Housing: What Do We Know? What Can We Learn?" Cityscape, 8(1), 2005, 69-137.

Saiz, A. "The Geographic Determinants of Housing Supply." Quarterly Journal of Economics,

125(3), 2010, 1253-96.

Saks, R. "Job Creation and Housing Construction: Constraints on Metropolitan Area

Employment Growth." Journal of Urban Economics, 64(1), 2008, 178-95.

Sivitanidou, R., and P. Sivitanides. “Does the Theory of Irreversible Investments Help Explain

Movements in Office-Commercial Construction?” Real Estate Economics, 28(4), 2000, 623-61.

Taylor, J. B. “The Swedish Investment Funds System as a Stabilization Policy Rule.” Brookings

Papers on Economic Activity, no. 1, 1982, 57-99

Zhou, C. “Time-to-Build and Investment.” Review of Economics and Statistics, 82(2), 2000,

273-82.

- 27 -

Table 1

Summary Statistics: Type of Building, Type of Construction, and Location

Percent of

observations

Percent of

construction cost

Type of building

Office

33.7

39.8

Retail

42.5

25.8

Warehouse

17.9

17.1

Hotel

5.9

17.3

Type of construction

New structure

95.1

93.1

Addition

7.1

7.3

Alteration

5.7

4.1

Conversion

1.0

0.5

Census Division

New England

5.4

5.1

Middle Atlantic

10.2

15.0

South Atlantic

23.8

21.6

North Central

17.9

16.7

South Central

18.5

13.2

Mountain

11.5

11.6

Pacific

12.7

16.8

MSA population

Top 25 MSAs

39.0

52.0

Rest of first quartile

49.6

40.1

Second quartile

7.0

4.6

Third and fourth quartiles

4.4

3.2

Note: Based on 82,303 projects with planning dates recorded between 1999 and 2010. See the

text for details about the construction of the sample.

1. A given project may encompass multiple types of construction.

2. Combination of East North Central and West North Central.

3. Combination of East South Central and West South Central.

- 28 -

Table 2

Summary Statistics: Project and County Characteristics

Variable Mean

Percentile of distribution

Project characteristics

Number of buildings

1.5

Number of floors

1.8

Construction cost (millions of 2005 dollars)

5.6

1.5

64.4

Square footage (thousands)

600

Cost per square foot (2005 dollars)

110

466

Distance from city center (miles)

County characteristics (in year 2000)

Housing density (units per square mile)

626

218

11,675

Urban share of population (percent)

100

Homeowner share of occupied units (percent)

Median household income (thou. of dollars)

Median home price (thou. of dollars)

129

115

361

High-income share (percent of households)

Note: Based on 82,303 projects with planning dates recorded between 1999 and 2010. See the text for details about

the construction of the sample.

1. Share of households with income above $100,000.

- 29 -

Table 3

Regression Results: Baseline Planning Lag and Effects of Characteristics

Variable Linear effect Quadratic effect

Constant (planning lag for baseline project)

16.21*

(.68)

Type of building (omitted: office)

Retail

.58*

(.14)

Warehouse

-.18

(.20)

Hotel

.85*

(.31)

Type of construction (omitted: new structure)

Addition

-1.20*

(.20)

Alteration

-.09

(.30)

Conversion

-.42

(.79)

Other project characteristics

Ever deferred

24.92*

(.37)

Ever deferred * square footage

.016*

(.0041)

Ever deferred * (square footage)

-1.58e-6

(1.96e-6)

Number of buildings

.74*

(.096)

-.0015

(.0012)

Number of floors

.89*

(.084)

-.017*

(.0024)

Square footage

.0071*

(.0010)

-3.36e-8

(4.87e-7)

Cost per square foot

3.49*

(.84)

-.120

(.133)

Distance from city center

-.0084

(.0096)

6.6e-5

(7.5e-5)

County characteristics

Housing density

-1.03e-5

(2.22e-4)

-4.64e-9

(5.14e-9)

Urban share of population

1.62

(2.32)

-1.20

(1.83)

- 30 -

Table 3 (continued)

Regression Results: Baseline Planning Lag and Effects of Characteristics

Variable Linear effect Quadratic effect

Homeowner share of occupied units

-11.47

(11.52)

12.04

(8.28)

Median household income

-.095

(.12)

-4.26e-4

(1.35e-3)

Median home price

.063*

(.014)

-7.35e-5*

(2.61e-5)

High-income share

15.9

(10.4)

-25.4

(31.4)

Note: Estimated by a maximum likelihood procedure that accounts for right censoring and interval reporting of

the time-to-plan data. Bootstrap standard errors are shown in parentheses. A total of 500 bootstrap replications

were run; 495 drew samples in which all parameters could be identified. Asterisks indicate significance at the 5

percent level. The results for the month and year dummies are reported in table 5. The results for the MSA fixed

effects are available from the authors upon request.

- 31 -

Table 4

Effects of Change in Characteristics on Planning Duration

(in months)

Variable Total effect

95 percent

confidence interval

Bottom

Top

Project characteristics

Number of buildings

6.5

5.0

8.0

Number of floors

7.2

6.0

8.5

Square footage

4.2

3.2

5.3

Cost per square foot

1.5

0.8

2.2

Distance from city center

-0.3

-1.0

0.5

County characteristics

Housing density

-0.7

-4.3

2.9

Urban share of population

0.0

-1.0

1.0

Homeowner share of occupied units

1.2

-2.1

4.4

Median household income

-6.4

-9.8

-2.9

Median home price

9.8

6.1

13.5

High-income share

1.8

-2.2

5.8

Note: The total effect represents the change in planning duration, using the results reported in table 3, when one

explanatory variable at a time is adjusted from the 1

percentile of its distribution to the 99

percentile. The 95

percent confidence interval for each variable is calculated from the bootstrapped variance-covariance matrix of

the coefficient estimates.

- 32 -

Table 5

Regression Results for Time Effects

Month

(omitted: June)

Coefficient

Year

(omitted: 2004)

Coefficient

January

.22

(.26)

1999

-.67

(.35)

February

.39

(.23)

2000

-1.77*

(.31)

March

.32

(.23)

2001

-.32

(.32)

April

-.02

(.23)

2002

-.81*

(.27)

May

.07

(.23)

2003

-.82*

(.26)

July

.27

(.23)

2005

.74*

(.24)

August

.62*

(.23)

2006

1.12*

(.27)

September

.37

(.23)

2007

.74*

(.24)

October

.50*

(.25)

2008

2.02*

(.24)

November

.93*

(.24)

2009

1.52*

(.27)

December

.50*

(.24)

2010

2.58*

(.29)

Note: See table 3 for a description of the regression. The results for the constant term in the

regression and for the project and county characteristics are reported in table 3. The results for the

MSA fixed effects are available from the authors upon request.

- 33 -

Table 6

Change in Year-Effect Coefficients

Change, in months

1999 to 2010

1999-2001 to

2008-2010

Type of building

Office

4.2**

3.3**

Retail

2.5*

2.5**

Warehouse

4.2**

Hotel

6.9**

5.8**

MSA population group

Top 25

4.2**

3.3**

Rest of first quartile

4.3**

2.7**

Second quartile

3.1*

2.1**

Third and fourth quartiles

4.7**

2.8**

Census divisions

Pacific

8.2**

7.0**

Mountain

4.8**

4.7**

West North Central

1.5

2.2*

West South Central

East North Central

1.0

East South Central

1.7

2.9**

New England

1.7

2.6*

Middle Atlantic

4.6**

3.9**

South Atlantic

4.2**

3.3**

Note: ** and * indicate a significant difference at the one-percent and five-percent level,

respectively. All results are obtained from the augmented regression described in the text.

The results for office buildings, the top 25 population group, and the South Atlantic Census

division are identical because these are the omitted categories in the year-effect interactions,

so the year effects in each case are captured by the uninteracted year dummies.

- 34 -

Table 7

Mean Time to Plan

(in months)

Unweighted Cost-weighted

All projects Excl. deferred All projects Excl. deferred

All building types 17.3 13.9 28.6 20.5

Office

17.2

13.6

29.3

21.7

Retail

16.5

13.8

24.6

20.1

Warehouse

16.9

13.5

23.4

17.0

Hotel 25.2 18.1 37.9 22.4

Note: Figures shown are means of the distribution of planning lags for projects within each building type

with corrections for right-censoring and interval collection.

- 35 -

Table 8

Components of the Wharton Residential Land-Use Regulatory Index

Index Component Description

Approval delay index

An index of the time required to complete the review of

(i) residential construction projects, (ii) rezoning

applications, and (iii) subdivision applications.

Local zoning approval index

The number of local entities required to approve zoning

changes.

Local project approval index

The number of local entities required to approve new

projects.

Local assembly index

Dummy variable. Equals one if town meetings are

required to approve zoning changes.

Supply restrictions index

An index of limits on annual permit issuance and

allowable construction activity.

Exactions index

Dummy variable. Equals one if developers are required to

fund infrastructure improvements in order to build.

Density restrictions index

Dummy variable. Equals one if any area in the locality

has a minimum lot-size requirement of at least one acre.

Open space index

Dummy variable. Equals one if developers are required to

supply open space in order to build.

Local political pressure index

An index that measures (i) the degree to which various

local entities are involved in development decisions and

(ii) the importance of governmental and citizen opposition

to growth.

State political involvement

index

An index of the state-level involvement in local land-use

regulations and the enactment of state-level land-use

restrictions.

State court involvement index

An index for the tendency of appellate courts to uphold

local land-use regulation.

Note: See Gyourko, Saiz, and Summers (2008) for details.

- 36 -

Table 9

Regression of Estimated Place Fixed Effects on the Wharton Residential Land Use

Regulatory Index and Components

Variable

Unweighted

Cost-weighted

Constant

15.99*

(.25)

16.07*

(.34)

17.08*

(.16)

16.52*

(.24)

Aggregate index

.35

(.25)

-.07

(.18)

Approval delay index

.17

(.18)

1.69*

(.13)

Local zoning approval index

.41*

(.21)

.85*

(.19)

Local project approval index

.18

(.19)

-.06

(.16)

Supply restrictions index

-.04

(.23)

-.28

(.24)

Exactions index

-.13

(.20)

-.54*

(.15)

Density restrictions index

-.25

(.15)

-.35*

(.15)

Open space index

.06

(.18)

-.28

(.16)

Local political pressure index

-.15

(.15)

-.51*

(.09)

State political involvement index

.65*

(.29)

-.13

(.14)

State court involvement index

-.51*

(.26)

-.52*

(.16)

.006 .050 .0002 .356

Note: Each regression is estimated by OLS using the 666 places that have at least 10 projects. In the cost-weighted

regressions, each place is weighted by its share of total construction cost in the 666 places. The local assembly

index is excluded from the regressions because it takes the same value in every place. Robust standard errors are

shown in parenthesis. An asterisk indicates significance at the five-percent level.

- 37 -

- 38 -

APPENDIX: YEAR EFFECTS IN THE AUGMENTED REGRESSION

As described in section III, we estimated an augmented regression to assess whether the

upward trend in the year fixed effects was widespread across the sample or was limited to

specific locations or types of projects. The augmented regression includes all of the variables in

the baseline regression plus interaction terms between the year dummy variables and three sets of

other dummy variables: dummies for the type of building to be constructed by the project,

dummies for MSA population size groups, and dummies for the nine Census Bureau geographic

divisions. We estimated this augmented regression using the same maximum likelihood

procedure with bootstrapped standard errors as for the baseline regression.

Table A1 presents all the estimated coefficients that involve the year dummy variables.

Figures A1, A2, and A3 use these coefficients to plot the time series of year effects by building

type, MSA population group, and Census division.

- 39 -

Table A1

Coefficients on Year Dummies in Augmented Regression

Variable 1999 2000 2001 2002 2003 2005 2006 2007 2008 2009 2010

Uninteracted

year dummies

-2.02

(.97)

-2.13

(.86)

.30

(.85)

-.41

(.68)

.40

(.73)

2.14

(.74)

1.11

(.69)

.92

(.68)

2.02

(.72)

1.77

(.92)

2.18

(.84)

Interaction with:

Building type

(omitted: office)

Retail

1.75

(.73)

.58

(.64)

-.72

(.67)

-1.24

(.58)

-.74

(.55)

-.02

(.50)

-.75

(.55)

-.30

(.53)

.40

(.56)

-1.21

(.63)

.08

(.59)

Warehouse

-2.08

(.99)

-1.77

(.89)

-3.11

(.90)

-2.50

(.87)

-.63

(.84)

-1.10

(.77)

-.90

(.88)

-1.83

(.75)

-1.29

(.83)

-.81

(.97)

-2.12

(.86)

Hotel

-.84

(1.46)

-.16

(1.62)

-1.33

(1.54)

-1.82

(1.55)

.62

(1.55)

1.85

(1.49)

3.40

(1.35)

.10

(1.21)

.37

(1.19)

2.95

(1.30)

1.84

(1.50)

MSA population group

(omitted: top 25 MSAs)

First quartile excluding

top 25

.10

(.78)

.32

(.71)

-.19

(.76)

.53

(.65)

.09

(.62)

-.22

(.59)

-.90

(.60)

-.53

(.56)

-.94

(.58)

-.58

(.72)

.16

(.68)

Second quartile

1.01

(1.13)

1.05

(1.09)

-.17

(.94)

2.90

(.94)

.22

(.80)

-.25

(.81)

.41

(1.01)

-.74

(.81)

-1.38

(.84)

-.21

(.91)

-.09

(.84)

Third and fourth quartiles

-1.97

(1.14)

.45

(1.13)

-.72

(1.25)

-1.04

(1.09)

.95

(1.01)

-.32

(1.17)

-.07

(.96)

-.27

(.89)

-1.65

(1.11)

-.45

(1.21)

-1.44

(1.05)

Census Division

(omitted: South Atlantic)

Pacific

-.73

(1.00)

-.88

(.96)

-.31

(.99)

.44

(.96)

.47

(1.00)

-.40

(.93)

3.51

(1.00)

2.39

(.88)

2.65

(1.02)

3.43

(1.28)

3.24

(1.16)

Mountain

1.03

(1.13)

.65

(.93)

.09

(1.04)

.05

(.92)

-1.85

(.85)

-2.05

(.76)

.78

(.78)

1.05

(.79)

1.80

(.84)

2.63

(.96)

1.68

(1.02)

West North Central

3.26

(1.40)

1.25

(.94)

1.30

(1.28)

-.13

(.90)

-1.21

(.72)

-2.37

(.73)

.71

(1.09)

1.29

(.83)

2.03

(.97)

-.00

(1.29)

.58

(1.03)

West South Central

3.85

(1.18)

2.29

(1.11)

2.27

(.98)

1.22

(.87)

-1.19

(.84)

-2.71

(.78)

-.46

(.76)

-1.31

(.71)

-1.03

(.78)

.03

(.98)

-.11

(.79)

East North Central

2.48

(1.04)

.55

(.84)

.45

(.85)

-.55

(.69)

-2.43

(.64)

-2.38

(.69)

-1.07

(.61)

-.05

(.63)

-.09

(.69)

-1.62

(.83)

-1.55

(.77)

East South Central

2.28

(1.11)

-.75

(.96)

-1.10

(.83)

-.23

(.82)

-1.88

(.73)

-1.06

(.88)

.63

(.64)

1.02

(.70)

.35

(.74)

-.98

(.93)

-.20

(.89)

New England

4.38

(2.30)

1.11

(1.29)

1.57

(1.25)

-.16

(1.00)

-1.24

(.93)

-.98

(.98)

1.23

(.93)

.89

(.84)

1.01

(.92)

2.23

(1.20)

1.88

(1.09)

Middle Atlantic

2.16

(1.09)

1.49

(1.08)

1.26

(1.11)

1.49

(1.08)

-.08

(.83)

1.10

(.89)

3.20

(.87)

2.27

(.77)

2.44

(.79)

1.70

(.87)

2.58

(1.01)

Note: Estimated by a maximum likelihood procedure that accounts for right censoring and interval reporting of the time-to-plan

data. Bootstrap standard errors are shown in parentheses. A total of 500 bootstrap replications were run; 492 drew samples in

which all parameters could be identified. Shaded fields signify that the coefficient is significant at the 5 percent level, with blue

for negative coefficients and yellow for positive coefficients. The explanatory variables omitted from this table include a

constant, month-of-year dummy variables, MSA dummy variables, and the full set of project and county characteristics. Year

dummies for 2004 are omitted from the regression to avoid perfect collinearity.

- 40 -

- 41 -