Commercial Real Estate in Crisis: Evidence from Transaction-Level Data

Abstract

During the past two decades, the commercial real estate (CRE) market has been impacted by major disruptions, including the global financial crisis and the Covid-19 pandemic. Using granular data from the U.S., we document how these crises have unfolded and elaborate on the role of heterogeneity and underlying shocks. Both a set of reduced-form approaches and a structural framework suggest a prominent role for demand-side local factors in the short run, along with significant shifts in preferences during crisis episodes. However, valuations become more closely linked to macro-financial factors over the long term. A one-standard deviation tightening in financial conditions is associated with a drop of about 3% in CRE prices in the following quarter, with a stronger impact on the retail sector and milder effects in states where household indebtedness is lower.

Full text

Commercial Real Estate in
Crisis: Evidence from
Transaction-Level Data
Adrian Alter, Cristian Badarinza, Elizabeth Mahoney
WP/23/15
IMF Working Papers describe research in
progress by the author(s) and are published to
elicit comments and to encourage debate.
The views expressed in IMF Working Papers are
those of the author(s) and do not necessarily
represent the views of the IMF, its Executive Board,
or IMF management.
2023
JAN

*Transaction-level data were provided by Zillow through the Zillow Transaction and Assessment Dataset (ZTRAX). More
information on accessing the data can be found at http://www.zillow.com/ztrax. The results and opinions are those of the authors
and do not reflect the position of Zillow Group, Inc., SafeGraph, or Real Capital Analytics. Business listings and footfall data
were provided by SafeGraph.
© 2023 International Monetary Fund WP/23/15
IMF Working Paper
Middle East and Central Asia Department
Commercial Real Estate in Crisis: Evidence from Transaction-Level Data
Prepared by Adrian Alter, Cristian Badarinza, Elizabeth Mahoney*
Authorized for distribution by Felix Fischer
January 2023
IMF Working Papers describe research in progress by the author(s) and are published to elicit
comments and to encourage debate. The views expressed in IMF Working Papers are those of the
author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.
ABSTRACT: During the past two decades, the commercial real estate (CRE) market has been
impacted by major disruptions, including the global financial crisis and the Covid-19 pandemic.
Using granular data from the U.S., we document how these crises have unfolded and elaborate
on the role of heterogeneity and underlying shocks. Both a set of reduced-form approaches and
a structural framework suggest a prominent role for demand-side local factors in the short run,
along with significant shifts in preferences during crisis episodes. However, valuations become
more closely linked to macro-financial factors over the long term. A one-standard deviation
tightening in financial conditions is associated with a drop of about 3\% in CRE prices in the
following quarter, with a stronger impact on the retail sector and milder effects in states where
household indebtedness is lower.
JEL Classification Numbers:
G12; E32; R33
Keywords:
Non-residential Real Estate; Financial Conditions; Liquidity; Big Data
Author’s E-Mail Address:
aalter@imf.org; cristian.badarinza@nus.edu.sg ;
elizabeth.mahoney@census.gov

IMF WORKING PAPERS
INTERNATIONAL MONETARY FUND
3
WORKING PAPERS
Commercial Real Estate in Crisis
Evidence from Transaction-Level Data
Prepared by Adrian Alter, Cristian Badarinza, Elizabeth Mahoney1
1 We are grateful to Prasad Ananthakrishnan, Mariarosaria Comunale, Guenter Franke, Chris Geiregat, Neesha Harnam, Prakash
Loungani, Camelia Minoiu, Nick Moschovakis, Tahsin Saadi Sedik, Alberto Sanchez, Radu Tunaru, Carlos van Hombeeck, and
IMF's seminar participants for useful comments and suggestions. The results and opinions are those of the authors and do not
reflect the position of Zillow Group, Inc., SafeGraph, Real Capital Analytics, the IMF, its Executive Board, or IMF management.

1 Introduction
The total value of the U.S. commercial real estate market (CRE) is roughly equal
to 100% of GDP, larger than the volume of Treasury certificates and corporate debt
outstanding (Ghent et al.,2019). A significant part of the portfolios of pensions funds,
life insurance companies, and other institutional investors are allocated to commercial
real estate, but 70% of the overall stock is owned by non-institutional private entities,
accounting for 30% of their overall firm assets.
Unsurprisingly, the CRE market has therefore been at the center of the two most re-
cent recessions. In 2008, the collapse of the system of mortgage securitization triggered
widespread turmoil, ultimately impacting real output and employment. In 2020, the
Covid-19 pandemic led to extensive periods of lockdown, effectively shifting the locus
of economic activity away from physical space into an emerging ecosystem of virtual
interactions, with a devastating impact on all industries for which the technological
transition was either sluggish or infeasible.
This paper seeks to uncover the drivers of price and liquidity dynamics in the CRE
market, exploiting newly available transaction-level data. To overcome limitations
inherent in aggregate indices, we obtain information about the universe of realized
commercial real estate transactions in the U.S. from Zillow’s ZTRAX database, which
covers the period between 1994 and 2020. The granularity of the data allows us to
address the differential exposure of CRE to crisis-induced cash flow disruptions by
looking into several dimensions of heterogeneity that would otherwise remain unob-
served. In particular, this concerns the functional use of the property, where we isolate
developments in the retail, office, industrial and hospitality sectors, and the geographic
location, relative to the urban core.
First, our results confirm the segmentation documented by previous studies, but
show remarkable consistency in the evolution of liquidity across these market segments:
While a common time component explains 48% of the price change variation, it also
explains 37% of the variation in volumes. This explains the resilience of the commercial
real estate market, especially in the period immediately following the pandemic. The
2

underlying mechanism here is tightly linked to the fundamental nature of investor
behaviour: since above 70% of the property stock is held by non-institutional investors,
with an average collateralization level of below 15% (Ghent et al.,2019;Ghent,2021),
in the absence of a supply shock that significantly affects asset valuations on the owners’
side, fire sales are rare. Instead, in response to a demand shock, the extensive margin
effect dominates—liquidity in the market decreases in response to increased uncertainty,
but it quickly rebounds when market conditions change.
The evolution of CMBS prices further supports the interpretation of recent devel-
opments as driven by factors that affect the demand side of the market. Exploiting
variation across tranches by issuance year and risk profile, we find that the magnitude
of the risk premium is well explained by the exposure of the underlying collateral to
the demand for space in its particular market segment.
Second, an important dimension of this micro-level evidence concerns spatial vari-
ation in prices and liquidity. Gupta et al. (2021) estimate the “spatial gradient” of
residential property, and find significant price and rent increases away from the center,
with larger effects in areas where working from home is more prevalent, housing mar-
kets are more regulated, and supply is less elastic. Ramani and Bloom (2021) validate
these findings with additional migration data, showing that real estate demand has
moved from dense central business districts (CBDs) towards lower density suburban
areas, and virtually no reallocation of population across cities (the “donut” effect). We
extend this analysis to the commercial real estate sector and document significant vari-
ation across property types, consistent with their exposure to such shifts of preferences
of end users, and their revealed residential choices.
Third, relying on two granular datasets that cover over a million CRE transactions
and over a hundred million data points on mobility trends during the pandemic, we
demonstrate the stability in the relationship between footfall and realized transaction
prices pre- and post-pandemic.1While this serves as validation for the role of demand
1Throughout the paper, "footfall" and "mobility" are used interchangeably. More specifically, we
refer to mobility trends as the realized footfall at a specific place (e.g., shopping mall, restaurant,
office) within a certain period of time (e.g., day, week, month).
3

shocks, it also suggests a path for recovery, consistent with significant increases in
market valuations visible during the period in which lockdowns and restrictions have
been largely lifted.
Finally, we also look at long-term drivers of CRE prices. Combining macro-level
data at the national and state-level, we establish long-term relationships between the
CRE market and national financial conditions, as well as local factors such as rental
vacancies, business activity, and indebtedness level. Lessons from previous crises (par-
ticularly the global financial crisis (GFC)) and historical relationships point to a 0.5-
0.6% temporary drop in CRE prices for each 1% increase in vacancy rates, over the
following quarter. In addition, a one-standard deviation tightening in financial con-
ditions is associated with a 2.5%-3% decline in CRE prices. These effects are found
milder in states where households are less indebted relative to their incomes, and thus
less financially constrained.
To isolate the contributions of these factors at aggregate level, we complement
the reduced-form analysis described above by enriching a standard search-theoretic
framework with a risk shock that affects valuations idiosyncratically in each period.
In the model, the risk shock affects the market participants’ reservation values. In
the data, we proxy for these unobserved reservation values to financial market prices,
as captured by commercial mortgage-backed security (CMBS) spreads. This allows
us to contrast the very sharp adjustment of risk preferences during the financial crisis
with a more cash-flow-driven pattern of market activity during and after the pandemic.
When estimated against observed empirical patterns, our framework substantiates the
predominant contribution of demand shocks in both crises episodes.
Overall, the paper aims to contribute to the recent emerging literature on the drivers
of real estate market price cycles. Tuzel and Zhang (2017) and Duca et al. (2021) find
remarkable diversity in the international and regional behavior of house prices, and
document the need to improve the data tracking key local demand conditions. This is
the motivating factor for our exploration of local demand using detailed mobility data.
Our focus on the two major crisis episodes also complements recent work by Levitin
4

and Wachter (2013) and Duca and Ling (2020) on the 2008 period, and Chernozhukov
et al. (2020), Agarwal et al. (2020), and D’Lima et al. (2020) on the 2020 events around
the start of the Covid-19 pandemic.
The impact of adverse cash-flow developments on the commercial real estate market
has received significant attention, because of the direct link between price and liquidity
dynamics in this market —see Ghent et al. (2019) and Ghent (2021) for a review on
the nature of commercial real estate as an asset class. Ling et al. (2020) provide a
first look at the impact of Covid-19 on real estate prices, exploiting a novel measure
of exposure to local health shocks. We complement their evidence with additional
measures of local demand, and a longer time period that covers the staggered pace
of reopening, with the ensuing rebound of liquidity in the market. Bergeaud et al.
(2021) focus on the longer-term impact of the pandemic and find that increases in the
magnitude of remote work are associated with higher vacancy rates, less construction
and lower prices in the office sector. Analyzing the CRE market in Ireland, Kennedy
et al. (2021) find consistent evidence that vacancies rates are an important determinant
of downside risks to CRE prices, with a more pronounced impact on retail and office
segments after the Covid-19 shock.
In our analysis, we address the differential exposure of CRE to pandemic-induced
cash flow disruptions by looking at the composition of loan pools collaterialized with
properties with different functional use. Buchak et al. (2020) point to the role of liquid-
ity as a central determinant of the response of the real estate market to shocks, and the
ability of potential technology-enabled intermediaries (iBuyers) to perform arbitrage
functions. They find that iBuyers technology allows for additional supply liquidity,
but only in sub-markets where this is least valuable. During the early days of the
pandemic, this phenomenon became very transparent, with activity in the majority of
local markets moving to a grinding halt. Deghi et al. (2021) quantify the vulnerabili-
ties that arise as a consequence of the pandemic in the CRE sector and analyze how
macroprudential policy can mitigate financial stability risks posed by the CRE sector.
We contribute to this line of research by providing micro-level evidence to underpin
5

further estimations of aggregate impact in the post-pandemic period.
Finally, the pricing of CMBS instruments has received considerable attention in the
wake of the financial crisis—see Titman and Tsyplakov (2010) and An et al. (2011), but
not in the most recent period. We fill this gap, and exploit the dynamics of loan spreads
to pin down the market’s perceptions of the value of commercial real estate from the
perspective of its owners and operators. This allows for a more precise identification
of risk shocks.
The paper is structured as follows. Section 2 describes the data sources used to
compute price and liquidity dynamics in the US commercial real estate market. Section
3 introduces a simple search model and estimates the contribution of supply, demand
and risk shocks. Section 4 reports the results of several reduced-form estimation exer-
cises, which cover the role of local cash flow variation, the evolution of spatial gradients,
the long-run effects of crises, and the pricing of risk. Section 5 concludes.
2 Data
In tandem with other data sources, this paper relies on two comprehensive datasets to
measure the impact of Covid-19 pandemic and the GFC on the United States commer-
cial real estate sector.
2.1 Commercial Property Indices
Over the past couple of years, a number of alternative indices have emerged, to capture
the evolution of property values in the commercial real estate market.
The first is the Green Street Commercial Property Price Index, an appraisal-based
index covering the period since 1998. The index provides a limited view of the market,
focusing on properties owned by Real Estate Investment Trusts. It also does not allow
us to distinguish sectoral dynamics, weighting various types of commercial properties
such as retail (20%), office (17.5%), apartment (15%), health care (15%), industrial
(10%), lodging (7.5%), and other sectors (15%).
6

The second is the RCA US CRE property index, for the period since 2000, retrieved
through the MSCI CPPI US report. The index captures the universe of traded US
commercial property in the apartment, retail, industrial and office segments.
2.2 Commercial Property Transactions
To overcome the limitations posed by existing indices, we source actual CRE trans-
actions from Zillow’s ZTRAX database, which contains data from over 400 million
detailed public deed records across 2,750 U.S. counties. The ZTRAX database rep-
resents a rich source of transaction-level data, spanning over 20 years and containing
detailed information from deed transfers, mortgages, foreclosures, and property tax
delinquencies. The information collected from these records includes property char-
acteristics such as building square footage and land surface area. Additionally, the
dataset contains geographic and valuation information. The geographic information
includes data such as zip code and sale address, while the valuation variables refer to
mortgage amount, sale price, and loan amount. These data are available for approxi-
mately 150 million parcels across 3,100 counties nationwide, making the ZTRAX data
especially comprehensive and valuable.
Our final dataset includes 30-40 states that contain reliable, representative data and
account for more than 80 percent of the U.S. population between 1994-2020.2To ensure
we capture only CRE transactions, we consider only transactions greater than $250,000,
leaving us with around 1.3 million real estate sale transactions. Daily transaction-level
information is aggregated by quarter and zip-code. Through the lenses of land usage,
we are able to identify which commercial sector each transaction belongs to. To enable
identification of sector-specific trends, we aggregate transactions into six commercial
real estate types: retail, office, industrial, multi-family living units, lodging, and other.
Part of our analysis controls for population density, while testing the hypothesis
of the so-called "donut" effect. Amplified by the Covid-19 pandemic, working from
home has negatively affected office occupancy rates, leading to a decline in CRE prices
2For more details see also Alter and Dernaoui (2020).
7

particularly in crowded areas (Ramani and Bloom (2021)). To proxy for population
density, we rely on the 2013 urban-rural classification scheme provided by the National
Center for Health Statistics (NCHS). Aggregated at the county level, the NCHS’ survey
data distinguishes between six types of areas: 1) Large central metropolitan (metro);
2) Large fringe metro; 3) Medium metro; 4) Small metro; 5) Micropolitan; 6) Noncore.
The first two categories refer to counties that contain metropolitan statistical areas
(MSA) of 1 million or more population. Below this threshold, the third and fourth cat-
egories contain MSAs with a population in excess or below 250 thousand, respectively,
while the last two categories are non-metropolitan counties.
2.3 Local Economic Activity
In the literature, economic activity is typically captured in survey data sourced from
the U.S. Census Bureau, including monthly retail sales and food services. We vali-
date economic activity with granular visit data from SafeGraph which covers business
listings and footfall data for over 6 million points of interest (POIs) across the U.S.
and Canada. Some examples of POIs include major retail chains, shopping malls, con-
venience stores and airports. At the individual POI level, SafeGraph has daily data
covering a variety of visitor analytics, including foot-traffic counts and demographic
insights. This dataset can provide an insight into how frequently people visit these
POIs, where they come from, and where else they go.
The raw dataset contains 150 million observations. After cleaning, we have almost
95 million observations representing 3,098 U.S. counties with an average of 125 POIs
across each and a median of 21. Considering the coverage of POIs increased overtime,
we normalized the data to obtain accurate foot-traffic counts. Using business activity
codes, we are able to aggregate POIs into five different sectors: retail, auto, restaurants,
manufacturing, and wholesale trade.
To validate the SafeGraph data, we compare changes in aggregated monthly visits
in each individual sector (from SafeGraph) with changes in monthly sales for the entire
US economy (from Census data), as depicted in Figure 3. In particular, the correlation
8

for the restaurant sector is found in excess of 90% (3c), suggesting a nearly perfect
association between the trends found in SafeGraph and economic activity. Additionally,
monthly series corresponding to retail sales (3a) and manufacturing sector (3d) were
found strongly linked to SafeGraph data as well, with correlations around 70%. Perhaps
to a lesser extent but still correlated in excess of 60%, the auto sector (3b) and wholesale
trade (3e) confirm the relationship to aggregated Census data.
2.4 State- and National-Level Macroeconomic Activity
Macroeconomic factors can be important drivers of CRE prices. To test this rela-
tionship, we developed a model which includes a variety of state and national level
indicators which can influence CRE prices. In general, characteristics at the state-level
like GDP growth, population growth, inflation, imports and exports are important in-
dicators of economic activity, thus driving CRE demand. More specific to CRE prices,
state-level business elements like the cost of doing business, supply and demand factors,
and employment opportunities can each have an impact on the prices of CRE within
a state. To measure cost of entry, we use general corporation license and franchise
tax year-on-year growth across states. To proxy for supply and demand factors, we
use the year-on-year growth rates for business applications and rental vacancy rates.
In addition, the private sector net job creation is used as an indicator of employment
opportunities.
National-level variables also influence CRE prices across states- though the impact
can vary depending on state characteristics like debt levels. Financial conditions, such
as interest rates and credit availability, can have an impactful relationship on CRE
prices because as interest rates increase, the cost of a loan increases, thus driving down
the prices of real estate.3To test the impact of financial conditions at the state-level,
we use the National Financial Conditions Index (NFCI) from the Federal Reserve Bank
of Chicago and interact it with a dummy indicator of household indebtedness, with the
3For the residential sector, Alter and Mahoney (2021) find that financial conditions can be a good
leading indicator of downside risks to house prices.
9

expectation that states with higher debt levels will be more sensitive to changes in
financial conditions.
2.5 Commercial Mortgage-Backed Securities
To gauge the reaction in the financial market, we analyze the pricing of CMBX con-
tracts. These contracts are typically a credit default swap (CDS) on an underlying
portfolio of 25 CMBS deals.4Given that pricing is reliably available at the daily fre-
quency, the CMBX data allows us to investigate how the financial market perceives
valuations of commercial real estate assets during key events, similar to Driessen and
Van Hemert (2012). Zooming into different CMBX series and tranches (see Panel A of
Figure 7), we can identify the main factors driving prices of these contracts. For ex-
ample, the CMBX AAA tranche regularly references super-senior CMBS with a credit
enhancement (of about 30%). In contrast, CMBX AJ, AA, A, BBB, BBB- refer to
increasingly lower seniority tranches in the capital structure of the same portfolio of
CMBS. These data were provided by Markit and sourced through JP Morgan’s Data-
Query.
3 Prices and Volumes
3.1 Aggregate
To verify our data, we perform a variety of validation exercises. After identifying CRE
transactions from the ZTRAX data, we compare it to the two major CRE price indices
mentioned above. This allows us to verify that the aggregated ZTRAX CRE price data
exhibits similar behavior to these price indices.5Figure 1a depicts the two CRE price
indices tracked during the 2000-2021 period.
4Each CMBX series references a different portfolio of 25 CMBS deals. However, all tranches of the
same CMBX refer to the same portfolio of 25 underlying CMBS.
5The appendix contains more details about alternative validation exercises and robustness checks.
10

3.2 Heterogeneity by property type
Figure 2shows the evolution over time of different types of CRE properties, both for
volumes and prices. Over the past two decades, CRE prices have generally doubled or
tripled, with a slightly stronger growth for industrial spaces (2d). Although prices have
substantially corrected after the GFC, particularly for retail (2a) and multi-family (2e),
they recovered and reached new highs right before the pandemic hit. Interestingly, the
cycles are even more visible for volumes, with the number of transactions substantially
dropping around the two major crises. It is worth noting that transaction volumes
peaked at the end of 2016 for most CRE types, while prices continued to rise. Im-
portantly, volumes drop significantly for all segments during the pandemic, but prices
show heterogeneous dynamics. While lodging, retail and multifamily prices were intu-
itively affected the most during the pandemic, given the nature of the crisis, industrial
and other segments remained relatively stable. This aspect is a peculiar feature of the
CRE market, which ensures its resilience. During crisis episodes, there are typically
not many transactions, liquidity dries up, absorbing the price shocks. Once liquidity
rebounds, prices recover as well (see e.g., Fig. 2b and 2f).
3.3 Heterogeneity across space
Beyond pure time-series effects, both the financial crisis of 2008 and the Covid-19 shock
have important implications across various CRE segments. Figure 6reports estimated
spatial gradients of CRE prices by property type. To calculate the gradient, we regress
the logarithm of the property price, controlling for hedonic characteristics, on a variable
that captures the distance from the closest urban core. Across all property types, a
clear trend is visible, toward higher prices close to the urban core, and lower prices
outside.
Interestingly, the effects around the financial crisis and the Covid-19 period are
rather similar, pointing towards a short-term inversion of that trend. For industrial
and office properties, the gradient was increasing in 2009, but decreasing in 2020. For
the office sector, our interpretation of the result is that the Covid-19 merely accelerated
11

a trend that was already visible before, in the sense that the sub-urban office had started
to be more attractive already for a number of years ahead of the pandemic’s impact.
For each property type, we estimate the following regression at the transaction
level:
Pi=α+β0Xi+β1UrbanCDC +β2U rbanCDC ∗Γt+ Γt+ Φc+εi.(1)
where the dependent variable Piis either (ln) CRE price per built surface, or (ln)
CRE price per land surface, or (ln) CRE price. When the price is not standardized, the
vector Xicontrols for property characteristics such as surface, building condition, and
year when the property was built, which could influence property valuations. Along
the lines described in Section 2.2,U rbanCDC is an ordered categorical variable taking
integer values from 1 to 6, thus ranging from (1) highest population density (i.e., in
large central metropolitan areas) to (6) the lowest density (i.e, in rural or noncore
areas). Γtand Φcrefer to year and county fixed effects, respectively.
Table 2provides transaction-level evidence on the quantification of spatial gradients
in the commercial real estate market over the past two decades, as a counterpart to
the results described above. The pronounced shift towards steeper spatial valuation
of closeness to the urban core is clearly visible across a wide range of specifications.
Consistent with the results of Gupta et al. (2021), we find a strong rebound of the
spatial gradient for multifamily housing after the pandemic, from a value of -0.2 to
roughly -0.16 within a single year. Such effects are only modestly visible in any other
market segment, where our estimation suggests a continuation of trends that were
building up over the previous years.
Taking the estimated coefficients in Table 2, the results therefore suggest that in
the early 2000s prices are around 30% higher in the most rural areas, compared to
the city core; to the contrary, in the early 2010s, they become 30% lower in the most
rural areas, relative to the city core. We see this as an economically very important
transition, consistent with an accumulating volume of past evidence, e.g., Dale-Johnson
12

et al. (2001) and Rosenthal et al. (2022).
3.4 Risk pricing
In Panel B of Figure 7we analyze the two periods of market disruption through the
lens of financial risk pricing. Using data on CMBX spreads tracked across vintages and
risk pools, we see remarkable consistency in the degree to which the financial market
responded to the two crises.
First, the left-hand plot reports changes in commercial spreads between June 2008
and December 2008, which is the period during which US financial markets have been
most affected by the collapse of the subprime mortgage market. During this period,
the average spread increased by roughly 12 percentage points, or 2.5 times higher than
the long-run average. This extreme magnitude is not surprising, given the wide spread
market panic that spilled over across markets during those months. Second, the right-
hand plot shows changes in commercial spreads between December 2019 and June
2020. This captures the direct impact at the onset of the Covid-19 pandemic, the set
of early lockdowns around the world, and the associated expectation of a major global
economic downturn.
The surprising feature of these results is the very different impact of the two shocks
across the risk rating spectrum. While the financial crisis affected low-risk tranches the
most, the events surrounding the start of the Covid-19 pandemic have mostly impacted
higher-risk tranches. This suggests the different nature of the two shocks as perceived
by the market. The former was expected to have a pervasive impact on the universe of
debt holders, while the latter was expected to materialize very heterogeneously, with
default risk only increasing in the more vulnerable sectors.
This risk adjustment pattern is consistent with the underlying structure of the
different loan cohorts. Panel A of Figure 7reports the allocation of different loan pools
across types of commercial real estate. The pools issued in earlier cohorts are more
heavily exposed to the retail sector, with the corresponding collateralized retail asset
share decreasing from around 40% to 25% in the latest cohort. This sector was most
13

intensely exposed to the effects of prolonged periods of lockdown and other mobility
restrictions: CMBX 6 has a delinquency rate in June 2020 that is double compared to
CMBX 12. This pattern is entirely consistent with the pricing of spreads during that
period, with the adjustment of the spread for CMBX 6 amounting to slightly more
than double the one for CMBX 12.
3.5 Long-run trends
Having explored the transaction-level drivers of value around two periods of significant
market disruption, we now turn to an analysis at the aggregate level using state-level
data. To formally gauge the long-run determinants, we estimate the following panel
regression specification:
Pst =α+βXst +δHHDebts∗N F CIt+ Γs+ Φi+εi.(2)
where the dependent variable (Pst) is the median price (per square foot) in each state
sat time t. In general, these regressions are estimated using state (Φs) and quarter
(Γt) fixed effects, over 2002-2020 period.6Xst is a vector of local characteristics such as
output growth, inflation, rental vacancies, business activity, net jobs creation, total ex-
ports, etc. The interaction between the state-specific dummy HHDebtsand nationwide
N F CItcaptures the heterogeneous effect of the financial conditions on the dependent
variable, subject to the level of indebtedness in each state. HHDebtstakes value one
in states where the level of indebtedness (proxied by household debt-to-income ratio
and averaged over the entire period) is below the cross-state median (i.e., 1.5), and zero
otherwise. Table 3presents the results for all types of CRE properties, while Table 4
focuses only on retail properties.
Table 3illustrates the fundamental drivers of CRE price dynamics, emphasizing the
strong effect of local economic conditions as captured by GDP growth (with a marginal
6Table A.1 presents the summary statistics for the dependent variable and its determinants. All
regressors are lagged by one quarter, with the exception of rental vacancy variable which is lagged
four quarters. The latter choice is based on a higher goodness-of-fit and significance.
14

positive effect of 0.6% to 1% for each percentage point change in local output) and the
vacancy rate (with a marginal negative rate of 0.5% to 0.6% for a one percentage
point change in the rental vacancy rate). The local average inflation level has a very
modest and statistically insignificant parallel impact on realized CRE prices, most likely
because of the wider investor base present in the market.7In addition, the relationship
between CRE prices and corporate license state tax (a proxy for firm creation) is
statistically significant (with a marginal effect of 1%).
Consistent with the theory, the effects of national financial conditions (NFCI) are
found negative (columns 11-15) and significant.8A one-standard deviation tightening
in NFCI in the previous quarter leads to a drop of about 2.5% in CRE prices. Im-
portantly, the effects are found milder in states where households are less indebted
relative to their incomes (columns 14-15). These results can be interpreted as evidence
of weaker transmission of financial conditions in the presence of less financially con-
strained households through the consumption channel. Intuitively, households with
less debt relative to their incomes are able to maintain their consumption habits even
when monetary policy or financial conditions are tightening. In economic terms, the
impact of tighter financial conditions on CRE prices is about 1/3 of the average effect
(0.8%) in states with lower debt levels.
As expected, these effects are different when we move to the retail sector (Table
4). For instance, local inflation is a paramount driver of CRE valuation, reflecting the
concentrated exposure of the sector to cash flows generated locally. Similarly, the im-
pact of the rental vacancy rate becomes even more pronounced in the retail sub-sample
relative to the overall market, with a marginal effect of 0.9% price appreciation after
a 1 percentage point drop in vacancy. As far as financial conditions are concerned,
the effects are stronger on retail CRE prices than when all transactions are consid-
ered. A one-standard deviation tightening in NFCI in the previous quarter leads to
7Effects are found insignificant also for GDP deflator, population growth, business applications,
net jobs creation and total exports (columns 4-10).
8These regressions do not include time fixed effects (FE), given that NFCI is at the national level.
However, we introduce time FE in regression 15, and show that the interaction coefficient remains
robustly significant.
15

a drop of about 3% in retail CRE prices. Likewise, the effects are milder in states
where households are less indebted relative to their incomes, and thus less financially
constrained (columns 14-15), with a drop in retail CRE prices of 1.5% in states with
lower indebtedness levels.
3.6 Local cash-flow variation
The valuation of commercial real estate has a strong cash flow component. This is
either a direct income from the tenant to the landlord, e.g., in the case of retail and
hospitality, or an internal transfer price for the case of owner-occupied property, in the
industrial or healthcare sectors. While the former is directly observed and measured in
companies’ profit and loss statements, the latter is an imputed quantity, and depends
on the actual use of the property by its owner-occupant. The opportunity offered by
footfall data is that it accurately captures the degree to which real estate space is
actually being used (i.e., “consumed”), at any given point in time, and at any given
location.
The series of Covid-19-related lockdowns provides a source of clean exogenous vari-
ation to the level of consumption of commercial real estate, and the associated cash
flow variation across locations. Figure 4plots the evolution of footfall, measured for
various property types, across counties, and through time. The significant decrease
of economic activity in the period after March 2020 is clearly visible, with remarkable
consistency across counties. But more importantly, the plots show that for all property
types, the cross-sectional variation of footfall is not materially affected by the overall
level shift.
Focusing on the time dimension, Figure 4shows changes in mobility trends across
the full county distribution, expressed in year-over-year growth rates to avoid seasonal-
ity issues. Visits to retail locations dropped by 30% to 50% (Fig 4a) during the initial
phase of the Covid-19 shock. Although these trends were reversed in the following
months, the recovery was slightly below pre-Covid averages and the recovery started
to falter by 2021, coinciding with the emergence of a new variant. Similarly, restaurants
16

(4b) and hotels (4c), along with healthcare (4e) and other contact-intensive services
(4f), experienced substantial declines in visits. The recovery in mobility trends of the
contact-intensive sectors seems to have been slower and well below pre-pandemic lev-
els, in particular for healthcare and other services. Compared to pre-pandemic trends,
visits to industrial places (4d) were generally less affected, with a much faster recovery.
However, some counties experienced substantial declines in mobility even for industrial
places, as suggested by the lower band (p25).
There is no clear expectation for how the cross-sectional variation in mobility should
be affected. One possibility is that a national lockdown leads to a similar response
across all markets, reducing any amount of heterogeneity that would have been visible
before the impact of the shock. Alternatively, if some locations are more affected by the
health component of the pandemic, they should respond more strongly, which magnifies
the initial heterogeneity. Perhaps reflecting the combined effect of these two opposing
forces, we do not see any significant change in cross-county variation, throughout the
sample.
In Figure 5we run yearly regressions of average CRE prices and county-level visits,
focusing on the retail sector. For a change in footfall of 10%, we find a marginal
effect of a price adjustment equal to roughly 1.5%. While this is a significant economic
magnitude, more importantly, the size of the effect is remarkably stable across time,
and only very modestly higher in 2020.
The observed stability of the cross-sectional relationship between footfall and prices
provides an important validation opportunity for our hypothesis that the main driver
of Covid-19 market dynamics is a demand shock. At the same time, this result also
indicates a direct path for recovery, which was observed as a pervasive feature across
all locations, once economic activity has recovered, and once the associated footfall has
increased back to roughly pre-pandemic levels.
17

4 Structural Estimation
4.1 Theoretical framework
We start with a standard search framework (see Diaz and Jerez (2013)), in which
buyers of mass mscan the available set of property listings of mass sfor potential
opportunities. Upon a successful match, buyer and seller valuations determine the trade
surplus, and the allocation of the surplus is determine by a competitive equilibrium.
In each period t, we assume the model is driven by three exogenous shocks: (i) a
demand shock εB
t, (ii) a supply shock εS
t, and (iii) a risk shock εR
t, all of which are nor-
mally distributed with mean zero and standard deviations σB, σSand σR, respectively.
Before turning to the estimation procedure, we describe the dynamics of transaction
volumes, value functions, and equilibrium realized prices.
4.1.1 Transaction volumes
In a typical period t, the stock of sellers evolves according to the following flow equation,
where Sis the steady state mass of properties listed for sale:
st=S+εS
t.(3)
The corresponding stock of buyers comes from two sources: first, a fraction 1−πB
t−1
of buyers in period t−1have not been able to find a match; second, a fraction αof
buyers ntthat were not interested in a purchase in period t−1, but become interested
in period t:
bt= (1 −πB
t−1)bt−1+αnt−1+εB
t.(4)
We assume a constant total stock of properties, with no construction activity, which
implies that at any given point in time, any given individual will be either a buyer, a
seller, or a matched owner not listing a property for sale. The corresponding market
18

clearing condition for a total housing stock with mass Nis given by:
nt+bt+st=N. (5)
This allows us to define market thinness θtas the relative mass of buyers and sellers,
consistent with the broader search literature:
θt=bt
st
(6)
Importantly, in this simple framework, market thinness θtis the single state variable
which determines the transition path of transaction volumes. In particular, it deter-
mines the probability that buyer and seller search will be successful, conditional on the
probability of a trade q, which we model as a constant structural parameter, and the
matched mass of buyers and sellers mt:
πS
t=q×mtand πB
t=q×mt
θ.(7)
Here, πS
tis the probability that any given seller with find a match, and πB
tis the
probability that any given buyer will find a match. This implies that transaction
volumes are then given by:
vt=πS
t×st,(8)
for a per-period matching function which takes the form of a standard non-linear
transformation of market thinness:
mt= 1 −e−θt.(9)
4.1.2 Valuations and prices
Given a particular structure of the market, prices arise in competitive equilibrium as a
function of individual buyer and seller valuations. The division of the surplus from the
transaction is captured by the variable ηt, which de facto indicates the seller’s relative
19

bargaining power in each period.
ηt=e−θt
1−e−θtθt.(10)
In a thin market, i.e., in a situation where bt<< st,ηtwill be low, and equilibrium
prices will reflect buyer reservation values. In a hot market with bt>> st, each listing
has a large number of potential buyers lined up, and equilibrium prices will reflect
seller reservation values.
The following system describes the value functions for the three types of agents:
WN
t=vB+βαEt[WB
t+1] + β(1 −α)Et[WN
t+1],(11)
WB
t=vS+βEt[WB
t+1] + βπB
tηtSt,(12)
WS
t=vS+βEt[WS
t+1] + βπS
t(1 −ηt)St+εR
t.(13)
Here, the magnitude of the total surplus Stis given by:
St=vB−vS+βEt[WN
t+1 −WB
t+1],(14)
and finally, equilibrium prices ptsolve the following non-linear equation:
vB−vS−pt+βEt[WS
t+1 −WB
t+1]
pt+βEt[WN
t+1 −WS
t+1]=ηt
1−ηt
.(15)
4.2 Time variation in the data
We match three sets of moments in the model and the data: transaction volumes vt,
prices pt, and seller valuations WS
t. While the former two are standard in the real estate
literature on search models (see, e.g., Genesove and Han (2012)), our contribution is
to include the average value of the CMBX spread as a novel financial variable in the
set of observed empirical moments. The CMBX spread allows us to identify the risk
shock separately. We calculate year-on-year quarterly differences for all variables, both
in the model and the data.
20

Panel A of Figure 8reports the three sets of moments that help pin down the values
of each of the exogenous shocks that drive equilibrium decisions and outcomes in the
model. Figure A.7 in the appendix illustrates the identification approach that allows
us to map these shocks onto the set of observable variables.
Supply shocks εS
tare identified by situations in which prices and volumes move
in opposite directions. Demand shocks correspond to situations in which prices and
volumes are positively correlated, with valuation spreads moving in the opposite direc-
tion. Risk shocks generate a similar positive correlation between prices and volumes,
with corporate spreads moving in the same direction as well. Before turning to the
results of the model, we briefly discuss the evolution of CMBX spreads, since they are
the novel element in our estimation.
4.3 Estimated shocks
We numerically linearize the model around its steady state and use a Bayesian Kalman
filter technique (see Herbst and Schorfheide (2015)) to match theoretical moments to
those observed in the data. The identification approach that we propose is equivalent
to a sign-restrictions method in a traditional VAR framework, in that the structure
of the model imposes restrictions on the direction of the impact of each shock. In
addition, we opt for a Bayesian estimation approach because it allows us to specify
prior distributions for the parameters, and is therefore analytically more tractable.9
Panel B of Figure 8reports the estimated time series of the three shocks. During
the financial crisis of 2008-2009, both demand and supply were at elevated levels,
suggesting an overall equilibrium in which high prices reflect high valuations and a
high propensity of trade. This is a period that is also characterized by an unusually
low level of the risk shock. This latter conclusion is particularly interesting, because
we did not feed the level of the CMBX rate to the model, and the spread itself is not
much lower during this period compared to the more recent years. The model therefore
9The model solution and estimation are implemented in Matlab, using Dynare version 5.2. (see
Adjemian et al. (2011)).
21

correctly attributes the pre-2008 developments to an extreme level of risk tolerance in
the market.
The situation before and after the Covid-19 pandemic is very different. First,
the pre-pandemic period sees a slow deterioration in supply entering the market, and
an increase in uncertainty. Especially after 2018, the decrease in demand is quite
pronounced as well, which suggests that even before the major disruption that ensued
in the early months of 2020, the commercial real estate market had entered a cooling
period. An additional reason for this development has to do with the monetary policy
regime, and the subsequent waves of proposed tightening.
Turning to the Covid-19 period itself, the strength of the risk shock is evident,
alongside the very dramatic collapse of demand. Indeed, this is the most significant
result from our structural estimation exercise—attributing the collapse of volumes and
prices in the 2020 and 2021 to a negative demand shock of a magnitude that is very
similar to the one observed during the financial crisis of 2008. In the next section, we
explore the nature and mechanics of this demand shock in more detail.
5 Conclusion
Historically, the CRE market has been highly intertwined with financial conditions and
the business cycle. During the recent downturns such as the GFC and the Covid-19
pandemic, the initial collapse in transaction volumes, a liquidity proxy, led to steep
declines in valuations. However, the transmission mechanisms of these two crises have
been different. While valuations swiftly rebounded during the pandemic, helped by
substantial policy support, the GFC had a more long-lasting impact, with prices re-
turning to pre-crisis levels after a some years.
To better understand these dynamics, we first test a few hypotheses in a reduced-
form setup, relying on two rich datasets covering over a million CRE transactions and
over a hundred million mobility patterns. This granular analysis allows us to establish
long-run relationships between the CRE market and local factors. Next, we investigate
22

the behaviour of CMBS spreads during the GFC and the Covid-19 pandemic. Finally,
we built a structural model which differentiates between risk shocks and fundamental
factors, allowing us to contrast the adjustment of risk preferences during the GFC with
a more cash-flow-sensitive market observed during the pandemic.
Our main contributions to the existing literature are threefold. First, our findings
suggest that patterns observed during the pandemic were broadly similar to those
experienced during the GFC. However, both demand and supply were found elevated
prior to the GFC and the overall market equilibrium reflected high valuations and
propensity to trade. In contrast, the CRE market was marked by a slow deterioration
in supply and increased uncertainty prior to the pandemic. Second, the stability of
cross-sectional relationships between retail traffic and prices suggests that the CRE
market during the pandemic was primarily driven by a demand shock. As far as
the office segment is concerned, our results point to an accelerated trend during the
pandemic that was already visible a few years before, with sub-urban office spaces
becoming relatively more attractive. Third, focusing on the long-run trends, we find
that a 1% increase in vacancy rates leads to a temporary drop of 0.5%-0.6% in CRE
prices over the following quarter. Importantly, a one-standard deviation tightening in
financial conditions is associated with a 2.5%-3% decline in CRE prices. These effects
are found stronger for the retail sector, which has typically been more sensitive to the
financial cycle, and milder in states with less financially constrained households.
Based on the long-run relationships, we conclude that the sensitivity to financial
conditions depends also on other macro-financial factors and local aspects such as
the level of leverage in the economy. Going forward, the transmission of monetary
policy tightening could be reflected in CRE prices through direct and indirect channels,
including higher vacancy rates, an increase in tenant bankruptcies, and tighter financial
conditions for investors.
23

Figure 1: US Commercial Real Estate Indices
(a) CRE Price Index Comparison
(b) CRE Price Index
(Transaction-level estimates)
(c) CRE all transactions
Source: Ztrax, Authors’ calculations. Note: Panel A depicts two aggregated CRE indices. Panel B compares estimates of median Sales
Price per built property surface with median Sales Price per property land surface, obtained from the ZTRAX transaction-level data.
Panel C depicts total transaction volume (LHS) and median Sales Price per property land surface (RHS).
24

Figure 2: Ztrax Data
(a) CRE Retail (b) CRE Other
(c) CRE Office (d) CRE Industrial
(e) CRE Multi-family (f) CRE lodging
Source: SafeGraph, Authors’ calculations. Note: YoY percent change
25

Figure 3: SafeGraph Data Validation
(a) Change in monthly visits and sales (Retail Sec-
tor)
(b) Change in monthly visits and sales (Auto Sec-
tor)
(c) Change in monthly visits and sales (Restaurant
Sector)
(d) Change in monthly visits and sales (Manufac-
turing Sector)
(e) Change in monthly visits and sales (Wholesale
Trade Sector)
(f) Correlations
Source: SafeGraph, Authors’ calculations, Census. Note: Log percent change
26

Figure 4: Distribution of SafeGraph Monthly Visits Across Counties
(a) Retail Visits (b) Restaurant Visits
(c) Hotel Visits (d) Industrial Visits
(e) Healthcare Visits (f) Other Services Visits
Source: SafeGraph, Authors’ calculations. Note: YoY percent change
27

Figure 5: Transaction-level relationship between CRE prices and county-level retail
visits in 2018, 2019, and 2020
Source: Authors’ calculations.
Note: This scatter plot depicts yearly relationships b etween CRE prices and number of retail visits. The retail visits are aggregated by
zipcode in a specific year.
28

Figure 6: Spatial Gradients
Source: Authors’ calculations.
Note: This figure plots the interaction coefficients between the location (or distance to city center proxied by the URB ANCDC
variable) and the yearly dummy, as presented in Table 2columns (13)-(18). The coefficient of URBANC DC is added to the interaction
coefficients.
29

Figure 7: CMBX Spreads and Composition
Panel A
Panel B
AAA AJ AA A BBB- BB
0%
5%
10%
15%
20%
25%
30%
Absolute change in spreads
Great Recession
(30 June 2008 to 31 Dec 2008)
CMBX 5
CMBX 4
CMBX 3
CMBX 2
CMBX 1
AAA AS A A A BBB- BB
0%
5%
10%
15%
20%
25%
30%
Absolute change in spreads
Covid-19 crisis
(31 Dec 2019 to 30 June 2020)
CMBX 11
CMBX 10
CMBX 9
CMBX 8
CMBX 7
CMBX 6
Source: JP Morgan DataQuery, Authors’ calculations.
Note: YoY percent change; CMBX 1 to 12 represent different vintages of CMBS pools, covering the years 2007–2019.
30

Figure 8: Data and shocks
Panel A
Overview of the data
2008 2010 2012 2014 2016 2018 2020
-20%
-10%
0%
10%
20%
Year-on-year changes
Prices
2008 2010 2012 2014 2016 2018 2020
-60%
-40%
-20%
0%
20%
40%
Year-on-year changes
Volumes
2008 2010 2012 2014 2016 2018 2020
-100%
0%
100%
200%
Year-on-year changes
CMBX Spreads
Panel B
Estimated structural shocks
2008 2010 2012 2014 2016 2018 2020
-15%
-10%
-5%
0%
5%
10%
Year-on-year changes
Supply shock
2008 2010 2012 2014 2016 2018 2020
-2%
-1%
0%
1%
Year-on-year changes
Demand shock
2008 2010 2012 2014 2016 2018 2020
-5%
0%
5%
Year-on-year changes
Risk shock
31

Table 1: Summary Statistics (ZTRAX, transaction-level)
Source: Ztrax, Authors’ calculations.
Note: This table presents summary statistics of all CRE transactions, sourced from ZTRAX data. The minimum sale price was set to
USD 250,000. To avoid outliers, other variables such as LTV were truncated at 1.5. p25 and p75 are the 25th and 75th percentile of the
distribution, respectively.
32

Table 2: Regression results: Spatial gradient (transaction-level)
Source: Ztrax, Authors’ calculations.
Note: Robust standard errors in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1. URB AN CDC takes integer values from 1 to 6, with 1 being the most urban area (e.g., closest to the city center).
33

Table 3: Regression results: Long-run - All CRE segments (state-level)
Source: Authors’ calculations.
Note: Robust standard errors in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1. All regressors are lagged.
34

Table 4: Regression results: Long-run - Retail Space (state-level)
Source: Authors’ calculations.
Note: Robust standard errors in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1. All regressors are lagged.
35

References
Adjemian, S.,Bastani, H.,Juillard, M.,Mihoubi, F.,Perendia, G.,Ratto,
M. and Villemot, S. (2011). Dynare: Reference manual, version 4.
Agarwal, S.,Ambrose, B. W.,Lopez, L. A. and Xiao, X. (2020). Pandemic risk
factors and the role of government intervention: evidence from covid-19 and cmbs
mortgage performance.
Alter, A. and Dernaoui, Z. (2020). Non-primary home buyers, shadow banking,
and the us housing market. IMF Working Papers,2020 (174).
—and Mahoney, E. M. (2021). Local house-price vulnerability: Evidence from the
us and canada. Journal of Housing Economics,54, 101791.
An, X.,Deng, Y. and Gabriel, S. A. (2011). Asymmetric information, adverse
selection, and the pricing of cmbs. Journal of Financial Economics,100 (2), 304–
325.
Bergeaud, A.,Eyméoud, J.-B.,Garcia, T. and Henricot, D. (2021). Working
from home and corporate real estate. Available at SSRN.
Buchak, G.,Matvos, G.,Piskorski, T. and Seru, A. (2020). iBuyers: liquidity
in real estate markets. Available at SSRN.
Chernozhukov, V.,Kasahara, H. and Schrimpf, P. (2020). Causal impact of
masks, policies, behavior on early covid-19 pandemic in the us. Journal of Econo-
metrics,220 (1), 23–62.
Dale-Johnson, D.,Brzeski, W. J. et al. (2001). Spatial regression analysis of com-
mercial land price gradients. In Asian Real Estate Society Sixth Annual Conference
at Keio University, Tokyo, Japan, Citeseer.
Deghi, A.,Mok, J. and Tsuruga, T. (2021). Commercial real estate and macrofi-
nancial stability during covid-19. IMF Working Papers,2021 (264).
36

Diaz, A. and Jerez, B. (2013). House prices, sales, and time on the market: A
search-theoretic framework. International Economic Review,54 (3), 837–872.
D’Lima, W.,Lopez, L. A. and Pradhan, A. (2020). Covid-19 and housing market
effects: Evidence from us shutdown orders. Real Estate Economics.
Driessen, J. and Van Hemert, O. (2012). Pricing of commercial real estate securi-
ties during the 2007–2009 financial crisis. Journal of Financial Economics,105 (1),
37–61.
Duca, J. V. and Ling, D. C. (2020). The other (commercial) real estate boom and
bust: the effects of risk premia and regulatory capital arbitrage. Journal of Banking
& Finance,112, 105317.
—,Muellbauer, J. and Murphy, A. (2021). What drives house price cycles? in-
ternational experience and policy issues. Journal of Economic Literature,59 (3),
773–864.
Genesove, D. and Han, L. (2012). Search and matching in the housing market.
Journal of Urban Economics,72 (1), 31–45.
Ghent, A. C. (2021). Whats wrong with pittsburgh? delegated investors and liquidity
concentration. Journal of Financial Economics,139 (2), 337–358.
—,Torous, W. N. and Valkanov, R. I. (2019). Commercial real estate as an asset
class. Annual Review of Financial Economics,11, 153–171.
Gupta, A.,Mittal, V.,Peeters, J. and Van Nieuwerburgh, S. (2021). Flat-
tening the curve: Pandemic-induced revaluation of urban real estate. Journal of
Financial Economics.
Herbst, E. P. and Schorfheide, F. (2015). Bayesian estimation of dsge models.
In Bayesian Estimation of DSGE Models, Princeton University Press.
Kennedy, G.,Killeen, N.,Skouralis, A.,Wosser, M. and Velasco, S. (2021).
Covid-19 and the commercial real estate market in ireland. Financial Stability Notes.
37

Levitin, A. J. and Wachter, S. M. (2013). The commercial real estate bubble.
Harv. Bus. L. Rev.,3, 83.
Ling, D. C.,Wang, C. and Zhou, T. (2020). A first look at the impact of covid-19
on commercial real estate prices: Asset-level evidence. The Review of Asset Pricing
Studies,10 (4), 669–704.
Ramani, A. and Bloom, N. (2021). The donut effect of covid-19 on cities. National
Bureau of Economic Research Working Paper Series.
Rosenthal, S. S.,Strange, W. C. and Urrego, J. A. (2022). Jue insight: Are
city centers losing their appeal? commercial real estate, urban spatial structure, and
covid-19. Journal of Urban Economics,127, 103381.
Titman, S. and Tsyplakov, S. (2010). Originator performance, cmbs structures,
and the risk of commercial mortgages. The Review of Financial Studies,23 (9),
3558–3594.
Tuzel, S. and Zhang, M. B. (2017). Local risk, local factors, and asset prices. The
Journal of Finance,72 (1), 325–370.
38

6 Appendix
Figure A.1: Ztrax Data Validation
(a) Green St. Vs ZTRAX (b) FRB vs. Ztrax
Source: Ztrax, Green St., Federal Reserve Board, Authors’ calculations.
Figure A.2: Ztrax Data County-level Coverage
(a) (2000-2020)
Source: Authors’ calculations.
Note: Numb er of transactions by county
39

Figure A.3: SafeGraph Data County-level Coverage
(a) Number of normalized retail visits by county (January 2020)
Source: SafeGraph; Authors’ calculations.
Note: This figure depicts number of visits aggregated by county in January 2020.
Figure A.4: Stylized Facts (SafeGraph)
(a) Number of Visits by Month (All CRE types, in
millions)
(b) Number of Visits by Sector (Normalized, in mil-
lions)
Source: SafeGraph, Authors’ calculations.
40

Figure A.5: County-level relationship between visits and CRE prices in 2019 and 2020
(Dropping outliers and states lacking observations)
(a) 2019 (b) 2020
Source: SafeGraph, Authors’ calculations.
Figure A.6: State-level relationship between visits and CRE prices in 2019 and 2020
(Dropping outliers and states lacking observations)
(a) 2019 (b) 2020
Source: SafeGraph, Authors’ calculations.
41

Table A.1: Summary Statistics (state-level)
Source: US Census, ZTRAX, Chicago Fed, Authors’ calculations. Note: The low debt dummy variable is created based on state-level
debt-to-income data from the Federal Reserve; A state with an average debt-to-income ratio over the time period below 1.5 is considered
a state with low debt. The median CRE prices were winsorized by 1 percent. The CPI is regional, rather than state-level.
42

Figure A.7: Illustration of identification
Panel A
Supply shocks
0 2 4 6 8 10 12
Quarters after impact
-0.5%
0%
0.5%
Year-on-year changes
Prices
0 2 4 6 8 10 12
Quarters after impact
-3%
-2%
-1%
0%
1%
2%
Year-on-year changes
Volumes
0 2 4 6 8 10 12
Quarters after impact
-4%
-2%
0%
2%
4%
Year-on-year changes
CMBX Spreads
Panel B
Demand shocks
0 2 4 6 8 10 12
Quarters after impact
-0.5%
0%
0.5%
1%
1.5%
Year-on-year changes
Prices
0 2 4 6 8 10 12
Quarters after impact
-2%
0%
2%
4%
6%
Year-on-year changes
Volumes
0 2 4 6 8 10 12
Quarters after impact
-10%
-5%
0%
5%
Year-on-year changes
CMBX Spreads
Panel C
Risk shocks
0 2 4 6 8 10 12
Quarters after impact
-4%
-2%
0%
2%
Year-on-year changes
10-15 Prices
0 2 4 6 8 10 12
Quarters after impact
-15%
-10%
-5%
0%
5%
Year-on-year changes
10-15 Volumes
0 2 4 6 8 10 12
Quarters after impact
-15%
-10%
-5%
0%
5%
Year-on-year changes
10-15 CMBX Spreads
43

Commercial Real Estate in Crisis: Evidence from Transaction-Level Data
Working Paper No. WP/23/15