View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.




First Quarter 2019
Volume 4, Issue 1

Why Are Recessions So Hard
to Predict? Random Shocks
and Business Cycles
Banking Trends: Estimating
Today's Commercial Real
Estate Risk

First Quarter 2019


Volume 4, Issue 1

Why Are Recessions So
Hard to Predict?
Random Shocks and
Business Cycles
Economists aren't soothsayers.
They can't pinpoint the start
of the next recession. But as
Thorsten Drautzburg explains,
their models can at least help
us understand why a recession
is happening, and what can be
done about it.

A publication of the Research
Department of the Federal
Reserve Bank of Philadelphia


Banking Trends:
Estimating Today's
Commercial Real
Estate Risk
Pablo D'Erasmo traces the link
between exposure to commercial
real estate loans and bank failure,
and estimates how much more
capital banks would need to withstand a plunge in prices like in the
financial crisis.

Research Update
Abstracts of the latest
working papers produced
by the Philadelphia Fed.

The views expressed by the authors are not
necessarily those of the Federal Reserve.
The Federal Reserve Bank of Philadelphia
helps formulate and implement monetary
policy, supervises banks and bank and
savings and loan holding companies, and
provides financial services to depository
institutions and the federal government. It
is one of 12 regional Reserve Banks that,
together with the U.S. Federal Reserve
Board of Governors, make up the Federal
Reserve System. The Philadelphia Fed
serves eastern and central Pennsylvania,
southern New Jersey, and Delaware.

Patrick T. Harker
President and
Chief Executive Officer
Michael Dotsey
Executive Vice President and
Director of Research
Adam Steinberg
Managing Editor, Research Publications
Brendan Barry
Data Visualization Manager

ISSN 0007–7011

About the Cover
Philadelphia's most famous citizen, Benjamin Franklin, has graced the $100 bill
since the newly created Federal Reserve began issuing “Federal Reserve Notes” in
1914. This particular image is taken from H.B. Hall's engraving of Joseph-Siffred
Duplessis's 1785 portrait of Franklin, which is currently on view at the National
Portrait Gallery in Washington, D.C. In the background are details from the 2009
redesign of the $100 bill, including a reproduction of the Declaration of Independence. Franklin served on the “Committee of Five” that drafted the Declaration
and presented it to the Second Continental Congress, then meeting at the Pennsylvania State House, on July 4, 1776. The State House still stands today, just
two blocks from the Federal Reserve Bank of Philadelphia, and is now known as
Independence Hall.
Photo by Rich Wood.

Connect with Us
We welcome your comments at:


E-mail notifications:


Previous articles:


Why Are Recessions So Hard to
Predict? Random Shocks and
Business Cycles
Economists are like doctors, not soothsayers. They
can't predict recessions, but they can help us
understand why one is happening. And that can
make all the difference for policymaking.

Thorsten Drautzburg is a senior
economist at the Federal
Reserve Bank of Philadelphia.
The views expressed in this
article are not necessarily those
of the Federal Reserve.


Economists can't tell you when the next downturn is
coming […]. Expansions don't die of old age: They're
murdered by bubbles, central-bank mistakes or some
unforeseen shock to the economy's supply (e.g., energy
price spike, credit disruption) and/or demand slide
(e.g., income/wealth losses).
—Jared Bernstein, Washington Post, 7/5/2018
Economists cannot predict the timing of the next recession
because forecasting business cycles is hard. For example, at the
onset of the 2001 recession, the median forecaster in the Survey of
Professional Forecasters (SPF) expected real U.S. gross domestic
product (GDP) growth of 2.5 percent over the next year, while in
reality output barely grew. Again, on the eve of the Great
Recession, forecasters were expecting GDP to grow 2.2 percent
over the next four quarters, and we all know how that worked
out.1 Why is it so hard to predict downturns—even while they
are happening?
Most economists view business cycle fluctuations—contractions
and expansions in economic output—as being driven by random
forces—unforeseen shocks or mistakes, as Bernstein writes.
As I will show, a model in which purely random events interact
with economic forces can resemble U.S. business cycles. This
randomness of economic ups and downs poses a challenge for
macroeconomic forecasters because random events, by their
very nature, are unpredictable.
One might be tempted to conclude that if the origins of business cycles are random forces, then analyzing business cycles
must be a pointless endeavor. However, not all random forces
are alike. For our purposes, economists distinguish between two
main types of random forces—demand shocks and supply shocks.2
As the term implies, shocks are surprise events that, when put
into a mathematical model of the economy, generate patterns in
economic variables that resemble those of business cycles.

Because the economy responds differently depending on which
type of random shock has occurred, knowing which type it was,
even after the fact, is important for getting economic models
right. And creating the right economic model is important for
choosing the right policy response if the economy is in the midst
of a recession.
If designing better models is the key, how is that research
progressing? What has prompted the recent thinking on the importance of shocks? I will summarize why early research focused
on productivity shocks (an important supply shock), and then
discuss why later models emphasized demand shocks. Perhaps
unsurprisingly after the Great Recession, more recent research
has focused on incorporating shocks to financial conditions.
I will also look beyond the mainstream research to two recent
critical contributions to traditional macroeconomic modeling.
First, though, let's consider more carefully what a business cycle
is, what the key characteristics of U.S. business cycles have been
over time, and just how random they have been.

What Is a Business Cycle?

Business cycles are recurrent expansions and contractions that
are common to large parts of the economy. The National Bureau
of Economic Research (NBER)—the private organization that is
the de facto arbiter of U.S. business cycle dating—defines a recession as “a significant decline in economic activity spread across
the economy, lasting more than a few months, normally visible in
real GDP, real income, employment, industrial production, and
wholesale-retail sales.”3
But even though business cycles recur, they are unpredictable
because the length of the expansions and contractions varies. In
the post-WWII era, expansions have lasted between one and 10
years. When the longest expansion ended after 10 years in 2001,
SPF forecasters were still surprised.

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


On a more practical level, we typically measure cycles as the
difference between the data as currently observed and the longerrun trend, defined as a movement that lasts eight or more years.4
Figure 1 illustrates this by plotting the level of real per capita
GDP and its estimated trend in the top panel. The difference between the level and the trend is the estimate of the cycle, shown
in the bottom panel. Qualitatively, economists typically focus on
how volatile such a detrended series is and how it comoves. We
typically measure volatility by the standard deviation, often
expressed relative to that of output. The correlation captures the
comovement, specifically that with the business cycle (as measured by GDP) and its own past realizations of a series (Figure 2).5
What characterizes U.S. business cycles? Three qualitative
properties of key economic indicators over the business cycle are
robust and form the key features that business cycle models try
to explain. First, investment and consumption are both
procyclical. They rise in expansions and fall in recessions. This
makes economic sense because output and income are higher in
expansions. Second, hours worked are strongly procyclical,
while unemployment shows the opposite pattern. In contrast,
labor productivity is only moderately procyclical, and real wages
are nearly acyclical. Third, investment is about three times more
volatile than GDP, whereas private consumption is one-third
less volatile, which makes sense if households prefer to smooth
their consumption—that is, to keep their rate of spending steady
through good times and bad.

Can Chance Drive Business Cycles?

Recall that even though business cycles are recurrent, they are
unpredictable because the length of expansions and contractions
varies. Economists have formalized this notion by building
models of business cycles that are driven by random events.
Mainstream economics views business cycles as comparable
to the “random summation of random causes,” to quote Eugen
Slutzky (1927, in English 1937). What does this mean, though?
Back in 1927, Slutzky observed that summing random numbers,
such as the last digits from the Russian state lottery, can generate
patterns that have properties similar to those we see in business
cycles. (See Figure 4 for his experiment.) Around the same
time, George Yule observed that other cyclical patterns, such as
those of actual sunspots, are well described by random shocks
that are fed into a simple linear model, again implying that we
can think of business cycles as random shocks that are averaged
over time. In 1933, Ragnar Frisch, the first Nobel laureate in
economics, took these insights about how random shocks can
combine to produce cyclical patterns to build a business cycle
model. Following Frisch, most economists now contend that good
models of the business cycle rely on combinations of current
and past shocks to accurately account for business cycle elements
such as those in Figure 2.
Broadly speaking, the models serve two purposes. First, they
provide a way to think about the economic origins of shocks. To
fix ideas, assume we observe data on prices and quantities.


Level and Trend (top) and Cycle (bottom) in U.S. Real GDP Per Capita Since 1870
GDP level

GDP trend

GDP level (100 × log)




















GDP (deviation from trend)
Source: Data retrieved from FRED, Federal Reserve Bank of St. Louis:; author's calculations.


Federal Reserve Bank of Philadelphia
Research Department

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

respond to economic shocks. For the
models discussed here, these individual
responses can be averaged to provide us
with a linear relationship between shocks
and macroeconomic data. This also allows
one to compute counterfactuals.


Disentangling Changes in a
Price-Quantity Pair into Demand
and Supply Shocks










While accepting the paradigm set out by
Frisch, economists differ on which models
and shocks are most useful for understanding business cycles. Identifying
shocks that cause movements in economic
variables is not just of academic interest. It
is important for policymakers such as the
Federal Reserve and other central banks
to know whether inflation falls because of,
say, a shock that leads to unexpectedly
high productivity, or because of a shock
that leads households to unexpectedly
increase the rate at which they save.
So, what specific shocks, when put into
a model, might generate patterns that
look like business cycles? Most economists
think that economic cycles are the result
of multiple shocks, although a single
shock may dominate specific episodes
such as the Great Recession.7 The two
theories that currently dominate research
emphasize different types of shocks. Real
business cycle (RBC) theory focuses on
real (as opposed to monetary) factors and
supply-side shocks. New Keynesian (NK)



The Search for Shocks


Picture the famous “scissors” representing demand and supply, as in Figure 3. The
economy moves from origin to the new
equilibrium at point A, the intersection
of demand D0 and supply S0. Identifying
the origin of shocks corresponds to dissecting this change in prices and quantities.
Here, a supply shock moved the supply
curve from the line labeled S0 to the S1
line. By itself, it would have lowered prices
and increased quantities, moving the
economy from point A to point B. A demand shock, from D0 to D1, accounts for
the remaining movement from B to C.
We need models to give us the correct
slope of the curves because otherwise
we cannot decompose the price-quantity
change into demand and supply changes
even in this simple example.6 The business
cycle model analogous to this example
typically implies that negative supply
shocks cause rising inflation and falling
output. In contrast, falling inflation and
falling output may point to a negative
demand shock. Further details, for example on the composition of output changes
or on relative prices, allow models to be
even more specific.
The second benefit that models bring is
that they allow us to have a mapping from
current and past shocks to observed macroeconomic data: The models' assumptions
on preferences and technologies imply
how individual firms and households will

Source: Following Uhlig 2017.

theory also incorporates nominal factors
and stresses the role of demand-side
In addition to allowing us to think
about the origins of shocks, these theories
and their implied models allow us to map
these shocks to data counterparts, such
as output or wages. This is necessary to
allow us to compare them to the data and
validate them, albeit indirectly.

Real Business Cycles

The RBC paradigm8 proposes that random
changes in total factor productivity relative



Volatility, Cyclicality, and Persistence of GDP and
Other Key Macroeconomic Indicators

Index of UK Business Cycle, 1855 to 1877
vs. Moving Average of Lottery Numbers

Correlated with GDP:

Business cycle

Most procyclical

Private consumption
Real wage
Labor productivity
Hours worked


The unemployment
rate is highly
negatively correlated:
Fewer people tend to
be out of work when
gdp is above its trend.
Unemployment rate

Most countercyclical

Components of GDP
Measures of labor market

Moving average of
lottery numbers


Volatility: Deviation from Trend
Private consumption
Real wage
Labor productivity
Hours worked
Unemployment rate












Source: Slutzky (1927, in English 1937).

Source: Data retrieved from FRED, Federal Reserve Bank of St. Louis:; author's calculations.
Note: All variables except the unemployment rate are %-deviations from trend. The volatility of investment, consumption, the real
wage, and productivity are measured relative to GDP. All series are persistent, with autocorrelations around 0.9 or higher: Their cyclical
value today tends to be close to the cyclical value yesterday.

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


to its trend are the key shock. Total factor
productivity determines how much firms
and, ultimately, the economy can produce
given inputs such as capital and labor.
These random changes can reflect both
actual changes in technology, such as selfdriving cars, and, more broadly, changes
in the legal or regulatory environment.9 To
map these shocks to the data, the model
makes certain assumptions about how willing households are to forgo consumption
today in order to consume more tomorrow
and how willing they are to work more in
response to higher wages.10
This simple model—with only productivity driving business cycles and a few
linear equations—matches most of the
qualitative behavior of the U.S. economy
described in Figure 2, including the
procyclicality and relative volatility of
consumption. Because households prefer
smooth consumption, they respond to
economic conditions by adjusting their
investment more than their consumption.
This explains the relatively low volatility
of consumption. Procyclical hours worked
result from households' rational choice to
work more while the economy is more
productive, even though they like leisure.11
However, the basic RBC model has
difficulty explaining changes in wages and
employment. In this type of model, firms
pay their workers according to how
productive they are, implying a high
correlation between wages and productivity and output—in contrast to their
low correlation in the data (Figure 2).12

Michael Woodford (1999) argued that
nominal frictions are also important because they help us understand how prices
vary relative to the costs of production.
Formally, the NK paradigm adds two
elements to the RBC paradigm. First,
there is market power, which on the side
of firms allows them to set prices and on
the side of workers allows them to set
wages. Second, there are limits to firms'
ability to adjust prices and households'
ability to adjust the wages they demand.
These limits arise because adjusting prices
or wages may be too costly. Or, some
firms or households might not have an
opportunity to adjust prices or wages, for
example due to fixed contract terms.
As the example from Galí makes clear, the
extra ingredients of the NK model change
how shocks affect observables such as
output compared with the RBC model.
They also give scope to think about new
sources of shocks, such as monetary
policy shocks to nominal interest rates.
Estimated versions of these models
have shaped how central banks today
analyze business cycles.13 These models
are also called dynamic stochastic general
equilibrium (DSGE) models. They are

dynamic because how much people work
or consume in the model depends on
their assessment of past and current conditions and their expected future paths.
They are stochastic because they are
driven by random shocks. Absent shocks,
the models imply that business cycles
are predictable. And they are general
equilibrium models because there is full
feedback of the choices of individual
firms and households onto one another.
In a key breakthrough, Smets and
Wouters (2007) showed that such a DSGE
model could match state-of-the-art statistical models for forecasting. At the same
time, DSGE models allow us to interpret
the forces at play in the economy. Other
models, such as a no-change forecast or
a vector-autoregressive model, also often
produce good forecasts. But compared
with these purely statistical models, the
DSGE model allows us to open up the black
box of what had driven an economic forecast and where the forecast fell short. Even
in hindsight, this information is important
for policymaking and for improving
models. For example, as I will discuss, the
Great Recession prompted economists
to look at shocks to financial conditions.


Historical Decomposition of GDP Growth Implied by Smets and Wouters



New Keynesian Economics

The NK extension of the RBC model adds
nominal, or price-related, elements that
nevertheless have real, quantity-related
effects. Jordi Galí (1999) argued that
nominal factors are key to understanding
that people work less after a positive
productivity shock: Because firms initially
cannot lower prices when productivity
rises, their labor demand falls temporarily.
That is, firms use the higher productivity
to economize on labor rather than to lower
prices and increase sales and production.
This explains why productivity is not
more closely correlated with output and
employment and allows the NK model
to fit the data better than the RBC model
does. Similarly, Julio Rotemberg and


Federal Reserve Bank of Philadelphia
Research Department



GDP Decomposed

















Source: Author’s calculations based on
Smets and Wouters.

Note: The demand and supply contributions add up to total real GDP
growth per capita in the historical Smets-Wouters data.

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

New Keynesian DSGE models feature
many shocks and decompose business
cycles into the effects of these various
shocks (Figure 5). With these types of
models, it is useful to distinguish between
supply shocks that affect the quantity or
cost of what can be produced with given
inputs and demand shocks that determine
how much firms or households want to
purchase at a given point in time. These
models are therefore useful to monetary
policymakers because, to pursue their
mandates such as price stability and full
employment, central banks may want
to lower interest rates in the event of
unexpected increases in supply and may
have to raise interest rates if demand
unexpectedly rises.
Seen through the lens of the Smets and
Wouters (2007) model, demand shocks
have accounted for most of the variation in
GDP growth from 1965 to 2004, as seen
in Figure 5. The two largest contributors to
short-run fluctuations have been demand

Bond Spread Shocks Contributed
a Significant Amount to the GDP
Decline During the Great Recession
Shock contributions to GDP

shocks: A shock to government consumption and net exports and a shock to the
desire to save each accounted for about
25 percent of the fluctuation in GDP
growth.14 Together, four supply shocks
have accounted for slightly less than half
of the observed GDP growth. The two
most important supply shocks have been
shocks to the productivity of all firms, as
in the RBC model, and shocks specific to
firms producing investment goods.

Financial and Uncertainty

In the aftermath of the financial crisis of
2008 and the subsequent Great Recession,
shocks to the financial sector have been
proposed as a missing ingredient in business cycle models. At the time, this was
new. While economists had long analyzed
the effect of the financial sector on the
economy, often the question was whether
financial institutions strengthen the
effects of other shocks, such as demand or
supply shocks.15 After the Great Recession,
economists began to ask: Do shocks to
the financial sector have important macroeconomic effects?
Harald Uhlig and I estimated a DSGE

model that includes the spread between the yields on private bonds and
government-issued bonds. These spreads
are important because firms cannot
borrow at the same rate as the government. Since they also pay the spread,
both the rate of government bonds and
spreads matter for private decisions,
while only the former were traditionally
modeled in DSGE. Our approach sidesteps
modeling the specific drivers of bond
spreads, such as, for example, changes
in default risk or in how markets price
default risk. We found that shocks to
bond spreads alone accounted for the
drop in output growth at the onset of
the Great Recession, even though these
shocks usually contribute much less
to fluctuations (Figure 6). Incorporating
bond spreads can also significantly
improve the forecasting performance
of these DSGE models.16
Christiano et al. (2014) provide a model
of the drivers of bond spreads. In their
model, bond spreads reflect default risk.
They model financial shocks as affecting
how much the returns vary between
different investment opportunities (within
the same asset class). These shocks then
move bond spreads. They find that such




Micro Shocks Lead to Macro Fluctuations


The approaches discussed so far focus on how aggregate shocks can explain aggregate
fluctuations. But the idea also applies to shocks to individual industries or even individual
firms. Could these shocks have aggregate effects, too? Detailed data on firms and
industries are now readily available to investigate this question. Economists have refined
the RBC approach to interpret these microeconomic data.


Government bond



Private bond spread
Price markup

Other shocks




Source: Drautzburg
and Uhlig, 2015.



Note: Real GDP per capita
level relative to trend.
2007Q4 normalized to zero.

If an individual firm or industry accounts for a large share of total sales in the economy, it
is possible that a shock to only that firm or industry will matter in the aggregate.17 Using
a simple formula to quantify this idea, firm-level shocks may account for about one-third
of aggregate fluctuations.18 More detailed measurement, however, has called this number
into question and suggests that firm-level fluctuations are more likely to account for only
one-sixth of aggregate fluctuations.19
Industry-specific shocks—say, an unexpected advance in drilling techniques for the
oil industry—can have outsize weight, too, if the industry is an important supplier or customer
for other industries. By one estimate, industry-specific shocks accounted for only one-fifth
of fluctuations in postwar U.S. output, although their contribution was higher during the
Great Moderation.20 But if it is hard for industries to switch from one type of input, such as
a certain material, to another, shocks to the productivity of the input-producing industry
would have a greater impact across the economy. Research that argues that this is the case
estimates that industry-specific shocks account for half of aggregate fluctuations.21

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


shocks account for about half of U.S. business cycle fluctuations.
Shocks that increase the variance of returns across investors
translate into higher borrowing costs and spreads because they
make it more likely that borrowers with limited liability may walk
away from projects and require lenders to step in. Anticipating
this greater likelihood of default, lenders charge higher interest
rates to cover expected losses from defaults. Higher borrowing
costs discourage firms from investing in their businesses and
households from purchasing durable goods, thereby generating
drops in output.
Individual uncertainty can also create aggregate fluctuations
through another mechanism. Economic activity can contract
when uncertainty rises because investors prefer to “wait and
see” rather than invest. This behavior is not due to financial
frictions but because it is more costly to undo investments than
to postpone them.

Is the Search for Shocks the Right Approach?

This article surveys two broad ideas in economics. First, business
cycles are driven by random forces. Second, after the fact, we
can trace these random forces back to economically meaningful
shocks using DSGE models. Both ideas have their critics, however.
Using DSGE models to quantify shocks as the driving forces of
business cycles has its limitations. First, shocks can be a measure
of our ignorance.22 In the spirit of “less is more,” economists
favor models that generate larger effects from small shocks.
Second, the way DSGE models and other statistical models are
typically estimated implies that they always point to specific
shocks to explain the observed changes in economic indicators,
without the ability to test whether they have identified the
right shocks. My recent research questions whether the identified
shocks in DSGE models are correct if one believes established

narrative accounts of these shocks.23 Related research allows
us to quantify how important shocks are without taking a stance
on how many shocks there actually are.24
The idea that business cycle fluctuations are driven purely
by random shocks also has its critics. In other business cycle
paradigms—for example, in the theories of Karl Marx or Hyman
Minsky—each boom carries the seeds of the next downturn. Paul
Beaudry and his coauthors have argued that economists should
revisit this idea and incorporate it into modern models.
Beaudry and his coauthors motivate their critique by arguing
that business cycles are more predictable than typically thought.
Using data on all U.S. recessions since the 1850s, they argue
that the likelihood of a recession has depended on the time
elapsed since the previous recession.25 Most models today imply
that business cycles are driven by the accumulation of positive
and negative shocks and that economic indicators such as output
or unemployment return smoothly to their long-run trends or
averages after a shock. In contrast, business cycles in intrinsically
cyclical models—that is, ones that assume that each cycle carries
the seeds of the next—could, in the extreme, explain business
cycles in the absence of shocks. Of course, Beaudry et al. do not
imply that business cycles are perfectly predictable—just that
ups and downs are somewhat predictable and that shocks are
smaller than commonly believed.

1 In the first quarter of 2001, forecasters expected cumulative GDP growth
of 2.5 percent over the next four quarters, whereas actual growth
(according to the first releases) averaged 0.5 percent. In the fourth quarter
of 2007, forecasters expected cumulative GDP growth of 2.2 percent
over the next four quarters, whereas actual growth (according to the first
releases) averaged 0.6 percent.

activities, followed by similarly general recessions, contractions, and
revivals which merge into the expansion phase of the next cycle; this
sequence of changes is recurrent but not periodic; in duration business
cycles vary from more than one year to ten or twelve years; they are
not divisible into shorter cycles of similar character with amplitudes
approximating their own.”

2 Bernstein's “central-bank mistakes,” labeled monetary policy shocks
later in this article, withdraw demand from the economy and are thus
also demand shocks. “Bubbles” could affect the credit supply by easing
collateralized borrowing, and their emergence or bursting would then be
a supply shock in financial markets.

4 See Baxter and King (1999) for a technical exposition.

3 The modern-day NBER definition quoted above (taken from http:// is very similar to the original concept of
Mitchell (1927, p. 468), one of the founders of the NBER business cycle
research program. He defines a business cycle as a “cycle [that] consists
of expansions occurring at about the same time in many economic

6 See Uhlig (2017) for a discussion of this decomposition and of statistical
techniques to identify the slopes.


Federal Reserve Bank of Philadelphia
Research Department

5 There has recently been debate on the details of detrending procedures
(Hamilton 2018; Beaudry et al. 2016). The results here, however, are
robust to details of the detrending procedure.

7 As I will discuss, the Great Recession may have been dominated by
a shock to financial intermediation.

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

8 The RBC paradigm was initiated by Kydland and Prescott in their 1982
9 See the discussion in Stadler (1994).
10 See Hansen and Heckman (1996) for a discussion.
11 See Chatterjee (1999) for more details.
12 Perhaps ironically, labor productivity was more procyclical at the time
that Kydland and Prescott invented the RBC paradigm. Before 1982, the
correlation of real wages and real GDP was 0.60, as compared with 0.23
for the full post-WWII sample in Figure 2. Huang (2006) also argues that
the comovement of real wages with output has changed before and
after WWII, consistent with the changing importance of supply shocks.
However, he argues that the structure of the economy has changed, not
the nature of shocks.
13 See Christiano et al. (2014) and Smets and Wouters (2007) for the
original articles and Dotsey (2013) for an overview.
14 A third type of demand shock, a monetary policy shock, has contributed
only about 5 percent. However, this does not imply that systematic
monetary policy has been irrelevant to the cyclical volatility of economic
output, but rather that monetary policy surprises unrelated to the state
of the economy have not played a large role in the postwar U.S. economy.
15 See Bernanke et al. (1999).
16 See the handbook chapter by Del Negro and Schorfheide (2013).
17 GDP measures value added (i.e., sales net of intermediate inputs), not
sales. One might therefore guess that value added weights matter.
However, sales matter because a firm whose value-added is small can
still affect large swaths of the economy if it uses inputs from or provides
key inputs to many other firms.
18 See Gabaix (2011).
19 See Yeh (2017).
20 See Foerster et al. (2011).

Atalay, Enghin. “How Important Are Sectoral Shocks?” American
Economic Journal: Macroeconomics 9, no. 4 (October 2017): 254–280.
Baxter, Marianne and Robert G. King. “Measuring Business Cycles:
Approximate Band-Pass Filters for Economic Time Series.” The Review of
Economics and Statistics 81, no. 4 (November 1999): 575–593. https://
Beaudry, Paul. “What Should Business Cycle Theory Be Aiming to Explain?”
(keynote presentation, 31st Annual Meeting of the Canadian Macroeconomics Study Group / Groupe Canadien d'Études en Macroéconomie,
Carleton University, Ottawa, ON, Canada, November 10, 2017).
Beaudry, Paul, Dana Galizia, and Franck Portier. “Putting the Cycle Back
into Business Cycle Analysis.” NBER Working Paper No. 22825, Cambridge,
MA, November 2016.
Bernanke, Ben S., Mark Gertler, and Simon Gilchrist. “The Financial
Accelerator in a Quantitative Business Cycle Framework.” In Handbook
of Macroeconomics Vol. 1, Part C, edited by John B. Taylor and Michael
Woodford, 1341–1393. Amsterdam: Elsevier, 1999.
Chatterjee, Satyajit. “Real Business Cycles: A Legacy of Countercyclical
Policies?” Business Review 82, no. 1 (January/February 1999): 17–27.
Federal Reserve Bank of Philadelphia.
Christiano, Lawrence J., Roberto Motto, and Massimo Rostagno. “Risk
Shocks.” American Economic Review 104, no. 1 (January 2014): 27–65.
Cochrane, John H. “Shocks.” Carnegie-Rochester Conference Series on
Public Policy 41 (December 1994): 295–364.
Del Negro, Marco and Frank Schorfheide. “DSGE Model-Based Forecasting.” In Handbook of Economic Forecasting Volume 2 Part A, edited
by Graham Elliott and Allan Timmermann, 57–140. New York: Elsevier,

21 See Atalay (2017).
22 See Cochrane (1994).
23 See Drautzburg (2016).

Dotsey, Michael. “DSGE Models and Their Use in Monetary Policy.”
Business Review 96, no. 2 (Second Quarter 2013): 10–16. Federal Reserve
Bank of Philadelphia.

24 See Plagborg-Møller and Wolf (2017).
25 Beaudry and his coauthors also point out that current models miss
properties of the business cycle by throwing out too much information
in detrending procedures.

Drautzburg, Thorsten. “A Narrative Approach to a Fiscal DSGE Model.”
Federal Reserve Bank of Philadelphia Working Paper No. 16-11, Philadelphia,
PA, April 2016.

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


Drautzburg, Thorsten and Harald Uhlig. “Fiscal Stimulus and Distortionary
Taxation.” Review of Economic Dynamics 18, no. 4 (October 2015): 894–920.

Smets, Frank and Rafael Wouters. “Shocks and Frictions in US Business
Cycles: A Bayesian DSGE Approach.” American Economic Review 97, no.
3 (June 2007): 586–606.

Foerster, Andrew, Pierre-Daniel Sarte, and Mark Watson. “Sectoral
Versus Aggregate Shocks: A Structural Factor Analysis of Industrial
Production.” Journal of Political Economy 119, no. 1 (February 2011): 1–38.

Stadler, George W. “Real Business Cycles.” Journal of Economic Literature
32, no. 4 (December 1994): 1750–1783.

Frisch, Ragnar. “Propagation Problems and Impulse Problems in Dynamic
Economics.” In Economic Essays in Honor of Gustav Cassel. Oslo, Norway:
University of Oslo, 1933. Reprinted in Readings in Business Cycles, edited
by R. A. Gordon and L. R. Klein, 1–35. London: Allen and Unwin, 1966.
Gabaix, Xavier. “The Granular Origins of Aggregate Fluctuations.”
Econometrica 79, no. 3 (May 2011): 733–772.
Galí, Jordi. “Technology, Employment, and the Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?” American Economic Review 89, no. 1 (March 1999): 249–271.
Hamilton, James D. “Why You Should Never Use the Hodrick Prescott
Filter.” Review of Economics and Statistics 100, no. 5 (December 2018):

Uhlig, Harald. “Shocks, Signs, Restrictions, and Identification.” In Advances
in Economics and Econometrics Volume 2, edited by Bo Honoré, Ariel
Pakes, Monika Piazzesi, and Larry Samuelson, 95–127. New York:
Cambridge University Press, 2017.
Yeh, Chen. “Are Firm-Level Idiosyncratic Shocks Important for U.S.
Aggregate Volatility?” CES Working Paper 17-23, Center for Economic
Studies, U.S. Census Bureau, Washington, DC, 2017. https://
Yule, G. Udny. “On a Method of Investigating Periodicities in Disturbed
Series, with Special Reference to Wolfer's Sunspot Numbers.” Philosophical Transactions of the Royal Society A: Mathematical, Physical
and Engineering Sciences 226 (1927): 267–298.

Hansen, Lars Peter and James J. Heckman. “The Empirical Foundations
of Calibration.” Journal of Economic Perspectives 10, no. 1 (Winter 1996):
Huang, Kevin. “Ups and Downs: How Wages Change over the Business
Cycle.” Business Review 89, no. 2 (Second Quarter 2006): 1–8. Federal
Reserve Bank of Philadelphia.
Kydland, Finn E. and Edward C. Prescott. “Time to Build and Aggregate
Fluctuations.” Econometrica 50, no. 6 (November 1982): 1345–1370.
Econometric Society.
Mitchell, Wesley Clair. Business Cycles: The Problem and Its Setting. Boston: NBER, 1927.
Plagborg-Møller, Mikkel and Christian K. Wolf. “Instrumental Variable
Identification of Dynamic Variance Decompositions.” Princeton University
working paper. Princeton, NJ, September 17, 2017.
Rotemberg, Julio and Michael Woodford. “The Cyclical Behavior of Prices
and Costs.” In Handbook of Macroeconomics Volume 1 Part B, edited by
John B. Taylor and Michael Woodford, 1051–1135. Amsterdam, The Netherlands: Elsevier, 1999.
Slutzky, Eugen. “The Summation of Random Causes as the Source of
Cyclic Processes.” Econometrica 5, no. 2 (April 1937): 105–146. https://


Federal Reserve Bank of Philadelphia
Research Department

Why Are Recessions So Hard to Predict? Random Shocks and Business Cycles
2019 Q1

Banking Trends

Estimating Today's Commercial
Real Estate Risk
To survive a decline in commercial real estate prices
such as occurred during the financial crisis, how much
more capital do banks today need?

Pablo D'Erasmo is an economic
advisor and economist at the
Federal Reserve Bank of Philadelphia. The views expressed in
this article are not necessarily
those of the Federal Reserve.

PA B L O D ' E R A S M O


ince the mid-1990s, banks have increased their commercial real estate
(CRE) lending significantly, allowing
the CRE market to almost double as a share
of the nation's overall economic output.
This growing share of CRE mortgages on
bank portfolios presents a financial stability challenge, since CRE exposure has been
a key determinant of bank failures in the
past. As commercial property prices have
climbed back up since the financial crisis,
CRE capitalization rates—the expected
return to investors in commercial real
estate1—have fallen to historically low levels.
This fall suggests that commercial real
estate prices could be poised to tumble
again, potentially causing large numbers of
CRE borrowers to default, and leaving
banks with steeply devalued CRE mortgages on their books and too little capital to
match their liabilities.
This article presents evidence of the
link between exposure to commercial real
estate loans and bank failure, and then
estimates how much more capital banks
would need to withstand a decline in
commercial real estate values like that
observed during the financial crisis.
Preventing bank failures and keeping
capital levels in a position to absorb losses
protects taxpayers because it reduces
the expected cost to the federal deposit
insurance fund and the likelihood of
government intervention in the case that
the crisis becomes widespread. Moreover,
failures at small banks, which are generally

more directly exposed to commercial real
estate, tend to disproportionally affect
small savers and borrowers.

Small Banks Especially
Exposed to CRE

CRE loans finance the purchase or development of almost any type of incomeproducing property, from offices to retail
spaces to industrial locations to multifamily residential complexes.2 There are
three types of CRE loans, their use depending on the type of property involved
and the buyer's objective for it:3

Three Types of CRE Loans

Their most common loan maturities and
their average loan-to-value ratios.
Construction and land development
Loan maturity
50 yrs
0 yrs
Loan maturity

0 yrs

50 yrs



Nonfarm nonresidential loans
Loan maturity
0 yrs
Source: DiSalvo and Johnston, 2016.

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1

50 yrs

Construction and land development
loans cover the cost of acquiring the land
and constructing the buildings. Their
typical maturity is three years, and their
loan-to-value ratio is 75 to 85 percent. This
line of credit carries a balloon payment
due when construction is completed, and
is generally financed by a new loan.
Multifamily loans are used to purchase
residential buildings with five or more
units. Maturities range from 10 to 40 years,
with an average loan-to-value ratio of 75
Nonfarm nonresidential loans (also
referred to as commercial mortgages) are
used to buy retail, office, industrial, hotel,
and mixed-use properties. The most common length of these loans is 10 years, with
a loan-to-value ratio of 65 to 75 percent.
Commercial banks are key players in
the commercial real estate market, holding
over 50 percent of the outstanding stock
of CRE loans on their portfolios in 2016,
and are particularly important for the
nonfarm nonresidential and construction
and land development segments of the
market, in which they hold 60.8 percent
and 100.0 percent, respectively.4
However, within the banking sector,
the degree of exposure to commercial real
estate mortgages varies substantially by
bank size. The top 35 banks hold 75 percent
of all bank assets but just 43 percent of
the commercial real estate market. The
next-largest group of banks—those ranked
36th to 225th in terms of total assets—hold
Federal Reserve Bank of Philadelphia
Research Department



Degree of CRE Exposure Varies
Total assets and CRE exposure.
Market shares by bank size
Top 35


Top 225 (excl. Top 35)

Total Assets







Commercial Real Estate




Source: Federal Reserve Call Reports.


Postcrisis Exposure to CRE Still

30 percent of CRE assets. Small banks—all
those not in the top 225—hold 27 percent
of the market (Figure 2).5
Although small banks hold the smallest
slice of the CRE market, the historical
evidence hints that in terms of the share
of their loan portfolios, small banks tend
to specialize in commercial real estate
and are more exposed to this market than
large banks are (Figure 3).6
Small banks' CRE holdings account for
30 percent of their total assets, compared
with just above 5 percent for large banks.
And small banks' specialization in commercial real estate has increased over the
last few decades. Their specialization in
CRE has been driven mostly by construction and land development loans and
nonfarm nonresidential mortgages (Figure
3), which have higher rates of default than
other commercial real estate loans and, as
discussed here, are a main driver of the
link between commercial real estate and
bank failure.
At the peak of the last financial crisis,
commercial real estate loans accounted for
almost 50 percent of small banks' total
loans. Today, even after the decline of the
real estate market during the crisis, that
fraction remains above 40 percent, suggesting that concentration in the commercial
real estate loan market remains elevated.
The largest banks have increased their exposure to multifamily loans since the crisis,
but their share of CRE loans as a fraction of
their total loans has always been relatively
low, just above 15 percent in 2016.

CRE Exposure Determines Bank Failure
Historically, the commercial real estate
market has been cyclical, with relatively
pronounced oscillations between economic expansions and recessions. Its
cyclical properties make banks that concentrate their lending in this sector
particularly vulnerable and can amplify
business cycles via bank failure and
reduced lending.
Evidence shows that high exposure to
CRE lending, when coupled with depressed CRE markets, has contributed to
significant credit losses and bank failures
in the past.7 Two supervisory criteria—
described in a 2006 regulatory guidance
by the Board of Governors of the Federal
Reserve System, the Federal Deposit
Insurance Corporation (FDIC), and the
Office of the Comptroller of the Currency
(OCC)—provide good benchmarks for
evaluating whether a commercial bank is
overexposed to the CRE market:
If its holdings of construction and land
development (CLD) loans represent 100
percent or more of its total risk-based
capital, then the bank is High CLD.
If its holdings of CRE (including CLD)
loans represent 300 percent or more of its
total risk-based capital and have increased
by 50 percent or more during the previous
36 months, then the bank is High CRE.
At any point in time, a significant
fraction of banks is highly exposed to the
fluctuations in CRE prices (Figure 4).8
As Figure 4 also makes evident, CRE
loan exposure has a local peak in the

Loan portfolio specialization by bank size.
Loans-to-assets ratio for different loan types
Top 35

Top 225 (excl. Top 35)


Construction & land development




Nonfarm nonresidential







Source: Federal Reserve Call Reports.


Federal Reserve Bank of Philadelphia
Research Department

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1



Significant Share Highly Exposed
to CRE

Percentage of banks with high exposure to
the CRE market since 1984.

High CLD

High CRE


High CLD



Source: Federal Reserve Call Reports.
Note: High CLD & CRE refers to institutions that satisfy the criteria for High CLD and High CRE.

mid-1980s and another in the mid-2000s.
Both peaks were followed by surges in
bank failures that, among other factors,
the literature has identified with downturns in the CRE market.
To illustrate how relevant CRE exposure has been for bank failures, we can
trace the evolution of the number of
commercial banks that have failed since
1984 and compare the failure rates for all
banks and for banks conditional on their
degree of CRE concentration (Figure 5).
Number of banks failed
Peak years for
bank failures

multifamily starts rose from 390,000 in
1981 to 670,000 in 1985, with virtually all
of the increase in large buildings. What
triggered the decline? Further changes
in tax policies had also been identified as
the drivers of the decline. The Deficit
Reduction Act of 1984 and the Tax Reform
Act of 1986 reversed most of the changes
of the 1981 tax law. The net effect has
been a reduction in the tax incentives to
rental construction.11
Many of the banks that failed had
actively participated in the regional real
estate market booms, particularly in
commercial real estate. In 1991, the commercial real estate loan-to-asset ratio for
banks that failed was close to 30 percent,
while the same ratio for banks that
continued operating was just above 10
percent. Commercial real estate loan
exposure among banks that subsequently
failed was significantly higher than for
those that did not fail.

The Crisis of the Late 1980s and Early
During a boom in commercial real estate
lending in the early 1980s—primarily
in the Southwest, Alaska, Arizona, the
Northeast, and California—CRE loans
tripled, which was followed by a rapid
decline in the value of real estate in 1989
and 1990, leading to a large fraction of
nonperforming or foreclosed commercial
real estate loans in 1991.
What triggered the fantastic increase
in CRE lending? One of the factors that
the literature has identified (see James
Poterba's article) was the tax incentives
included in the 1981 tax reform, the
Economic Recovery Tax Act of 1981. Total

The Last Financial Crisis
In response to increased competition in
the consumer and residential real estate
loan markets during the early 2000s,
small banks—generally referred to as
community banks—turned increasingly to
commercial real estate lending (Figure 3).12
During the early 2000s and until the
issuance of the interagency guidance,
the fraction of banks with large CRE exposures grew steadily (Figure 4). In 2006,
just before the crisis, 40 percent of all
commercial banks in the U.S. had high CLD
concentrations, and close to 20 percent
had high CLD and CRE concentrations.
As the crisis deepened, deteriorating
conditions in the residential mortgage
market that had begun in 2007 spilled over


High Concentrations Correlate with Bank Failures
Number and rates of bank failures.

Note: High CLD & CRE refers to institutions that satisfy the criteria for High
CLD and High CRE, and Low CLD & CRE refers to institutions that do not
satisfy either of the criteria.


The banking crises in the late 1980s and
the 2008–2009 financial crisis resulted in
a large number of bank failures.9 In both
episodes, there were major differences in
failure rates for banks above and below
the concentration levels specified in the
interagency guidance. Failure rates for
banks that exceeded the criteria were
three to four times higher than those of
the rest of the banks. Most failures in the
late 1980s occurred among banks that
had high overall CRE exposure, and most
failures in the last crisis were among
banks with high CLD concentrations.10



Fraction of banks failed

High CRE

High CLD

High CRE & CLD














Source: Federal Reserve Call Reports.

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


to the CRE market in 2008.13 One important link between the two
markets was that many banks had made loans to developers
for the purpose of constructing multifamily residences, and demand for these residences fell sharply in the recession. The CRE
price declines—on average, more than 42 percent between
the peak in 2007 and 2010—had very negative consequences for
the financial sector.
The percentage of CRE loans that banks had to write off from
the end of 2007 through the end of 2010 was 10 times higher
than it had been between 2000 and 2007. As in the previous
crisis, banks that were more exposed to commercial real estate
suffered much more. Commercial real estate loan delinquencies
were not as high as delinquencies in the residential real estate
market but also increased dramatically. Yet, charge-off rates for
commercial real estate loans were higher than charge-off rates
for residential real estate loans at the peak of the crisis, with CRE
charge-offs driven primarily by land, development, and construction loans.
Are there other relevant differences between the banks that
failed and those that did not? To shed some light on the factors
influencing bank failure—and in particular whether there are
significant differences in commercial real estate exposure—we
can compare the balance sheet composition for large versus small
banks, and in the case of the small banks, for those that failed
versus those that did not fail during the financial crisis (Figure 6).
As Figure 6 shows, small banks held more safe assets (liquid
assets such as cash plus riskless securities such as U.S. Treasury
securities) and were more exposed to commercial real estate.
Their higher holdings of securities derives from differences in the
cost of borrowing between small and big banks, geographic
diversification, and the volatility of their deposit base, as small
banks are more exposed to local fluctuations. Moreover, those
that failed were more exposed to commercial real estate than
those that did not fail and had a negative net income, or return
on assets (ROA).

Small Banks That Failed Were More Exposed to CRE
Balance sheet composition by bank size and small bank failure.

Top 35 bank

Small bank, No-fail

Small bank, Fail

Ratio to Total Assets (%)
Liquid assets
Riskless securities
Residential RE Loans
Commercial RE Loans
C&I loans
Consumer loans
Other assets
Net income (ROA)


Source: Federal Reserve
Call Reports.






20% 25%

Note: We define large banks as those in the top 35 of
the asset distribution and small banks as all the rest.

Federal Reserve Bank of Philadelphia
Research Department

Current Vulnerability: Stress-Testing CRE

Although commercial real estate valuations have increased considerably since the end of the crisis and capitalization rates have
declined to historical lows, the recovery in CRE prices and sales
volumes is beginning to slow. There are indications that demand
for CRE loans has weakened and that lenders are tightening lending standards, according to recent Senior Loan Officer Opinion
Survey results.
Even though capital regulations have been strengthened and
bank risk-weighted capital ratios have increased in recent years,
the rise in real estate prices and declines in capitalizations raise
questions about the vulnerability of banks exposed to the
CRE market.14 In addition, declines in CRE market values could
reduce overall small business lending by community banks.
But how can we quantify the current level of risk in the system
posed by CRE lending? To estimate this risk, I perform an
experiment that computes capital losses across banks using CRE
delinquency rates and loss-given-default
rates observed during the last crisis.15 With F I G U R E 7
a measure of delinquencies and losses at
Stress Effects
hand, it is possible to estimate the losses
Predicted losses
that banks would stand to incur in their
in 4Q2016.
Capital Losses
CRE holdings under circumstances similar
to those of the last crisis and from this estimate derive the reduction in bank equity AVERAGE MEDIAN
that banks would sustain (Figure 7).16
Buffer over Minimum
For example, if a bank's CRE holdings
equal $100, and 10 percent of those loans
default, with an average recovery rate of
Source: Call Reports
70 percent, the bank's portfolio will be reFederal Reserve Bank.
duced by $3. If its ratio of CRE loans over
risk-weighted assets is 33 percent—its riskNote: Uses CRE
delinquency and
weighted assets equal $300—then its ratio
loss-given-default rates
of risk-weighted capital due to the losses
across banks during
suffered in the CRE portfolio is reduced by 2008–2009. Capital
0.01 (=$3/$300). Then, if the bank's capital
Losses is ratio of capital
buffer over and above the minimum
to risk-weighted assets
lost due to CRE losses.
required is less than 1 percent, its capital
Buffer over Minimum is
ratio will slip below the minimum.
amount of excess capital
This approach uses as a starting point
over minimum that
the 4Q2016 distribution of CRE loans and
banks hold after sustaincapital ratios, and provides a distribution
ing CRE losses.
of bank capital losses.
While similar in spirit, this experiment differs from the formal
stress test that the Federal Reserve conducts, since it does not
use loan-level data or an explicit model to calculate loan losses,
and it evaluates the losses suffered only during one period as
opposed to an extended period. In this respect, the results of the
exercise should be viewed as a lower bound on potential losses.17
While informative, this experiment is not designed to capture
the effects of a protracted crisis in the CRE market, in which case
banks are hit with repeated, consecutive losses, including those
deriving from the linkages across banks, commercial real estate
markets, and other asset markets.18
One question that arises when performing this type of experiment is whether CRE loans are particularly toxic. The results show

0.4% 0.1%

5.6% 5.3%

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1

that losses in this portfolio have the potential to affect a large
swath of small banks. On average, banks currently have enough
capital to remain adequately capitalized even after suffering losses as large as those observed during the last crisis (Figure 7).
The average bank has a capital buffer of more than 5 percent.
However, this statistic paints over wide differences in CRE exposure and capital ratios similar to those documented for previous
crises. A more in-depth analysis shows that when exposed to
this stress scenario, 117 banks—2.3 percent of the total number of
banks, holding 0.4 percent of the aggregate value of assets
and 1.3 percent of the value of CRE credit—would fall below the
7.25 percent Tier 1 capital ratio required.19
This number should be understood as a lower bound on the
potential effects of a stress scenario, not only because of the static
nature of the experiment but also because, as Figure 6 shows,
banks with capital ratios that were well above the minimum
required had failed. For example, the value of the bank for its
shareholders can become negative before capital reaches the
minimum required.
Moreover, banks that are vulnerable to CRE price declines do
not overlap exactly with those that have the largest CRE concentrations. Approximately 50 percent of those that go below the
7.25 percent capital threshold in the experiment have high
concentration ratios. Other banks with high concentrations have
capital ratios substantially above 7.25 percent and are able to
absorb the losses, but their reduction in capital ratios also has
the potential to reduce lending.
This stress experiment induces a clear shift in the distribution
of risk-weighted capital closer toward the minimum. If banks
are currently operating at or close to their optimal level of capital, this shift implies that losses in the CRE market could curtail
lending or other asset markets and impede the normal operation
of most banks in the industry.


This experiment shows that while the financial system appears
to be better prepared for a shock in the CRE market now than
it was leading up to the financial crisis, in the event of another
such crisis, most banks would be affected, and many might fail.
The CRE sector remains a potential source of instability for the
banking sector.

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


1 More specifically, capitalization rate refers to the ratio of a property's
annual net operating income to its price.

11 The Tax Reform Act of 1986 created the Real Estate Mortgage
Investment Conduit, facilitating the issuance of mortgage securitizations,
including commercial mortgage-backed securities (CMBS).

2 I use a conservative definition that excludes loans secured by farmland.
3 See James DiSalvo and Ryan Johnston's 2016 Banking Trends article
for a description of the commercial real estate market.
4 The other half of commercial real estate mortgages ends up in the
hands of other investors, such as insurance companies, government
agencies, and private investors, or in a pool of mortgages such as
commercial mortgage-backed securities (CMBS).
5 Banks in the top 35 have assets above $50 billion, banks ranked 36th
to 225th have assets between $3 billion and $50 billion, and all those not
in the top 225 have assets below $3 billion (measured in 2016 dollars).
6 Large banks originate a large fraction of CRE loans, but they tend to
securitize a much larger fraction of these loans than small banks do.
7 See the Federal Deposit Insurance Corporation, Office of the Comptroller
of the Currency, and the Board of Governors of the Federal Reserve
System's “Guidance on Concentrations in Commercial Real Estate Lending,
Sound Risk Management Practices” and the Federal Deposit Insurance
Corporation's 1997 “History of the Eighties—Lessons for the Future,” https://
8 See Keith Friend, Harry Glenos, and Joseph Nichols's article, "An
Analysis of the Impact of the Commercial Real Estate Concentration
Guidance,” for a detailed description of the guidance and its implications
for loan growth and bank failure.
9 The estimate of bank failure is very conservative. Mergers are separated
from clear failures, since the reasons banks fail can be different from
those that result in a bank merger. However, several bank mergers were
driven by the same fundamentals that drive bank failures—low returns
on assets, declines in charter value, and exposure to risky assets. Similarly,
a number of banks would have failed but for government bailouts. All the
banks that actually failed were outside the top 35.
10 The Eliana Balla, Laurel Mazur, Edward Prescott, and John Walter
article analyzed the factors driving bank failures during the crisis of the
late 1980s and the most recent financial crisis extensively. Consistent
with previous literature (for example, the articles by David Wheelock and
Paul Wilson, George Fenn and Rebel Cole, and Rebel Cole and Lawrence
White), they find that CRE, and in particular construction land and
development loans, is the main factor driving failure probabilities.


Federal Reserve Bank of Philadelphia
Research Department

12 In 2003, banks with assets of $100 million to $1 billion had commercial
real estate portfolios equal to 156 percent of their total risk-based capital,
and this ratio increased to 318 percent in 2006.
13 Adonis Antoniades' article describes the link between residential real
estate and commercial real estate.
14 Besides cyclical fluctuations in commercial real estate prices, other
risk factors include fluctuations in the CMBS market and softness in the
retail sector that could impact the value of collateral used in CRE loans.
15 For each commercial bank, the delinquency rate on CRE loans during
the crisis is computed as the maximum (yearly) delinquency rate on CRE
loans observed during years 2008, 2009, and 2010. The values reported
in Figure 5 refer to the average (or the median) across banks. The
loss-given-default is computed as the average during the crisis.
16 In addition to delinquency rates and the loss-given-default, estimating
capital losses requires a measure of the loan loss provision (the ratio of
the provision for loan losses over total loans), the ratio of CRE loans to
risk-weighted assets, and the current level of capital over risk-weighted
assets for each bank. At the height of the last crisis, average nonperforming CRE loans was 7.75 percent, and loss-given-default CRE loans
was 30.27 percent.
17 See the Jihad Dagher, Giovanni Dell'Ariccia, Luc Laeven, Lev Ratnovski,
and Hui Tong article for a similar approach used to estimate appropriate
levels of bank capital during a crisis.
18 These factors include the spillovers from one commercial real estate
market to another via securities prices or a reduction in lending by banks
affected by the initial shock as well as linkages across banks that disrupt
the normal flow of credit when one of the links in the network is in distress.
19 The minimum Tier 1 risk-weighted capital required is 6 percent plus
a 1.25 percent conservation buffer in 2017. The conservation buffer will
increase to 2.5 percent in 2019.

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1

Antoniades, Adonis. “Commercial Bank Failures During the Great
Recession: The Real (Estate) Story.” Working Paper 1779, European Central
Bank, Frankfurt am Main, Germany, April 2015. https://www.ecb.europa.
Balla, Eliana, Laurel Mazur, Edward S. Prescott, and John R. Walter.
“A Comparison of Small Bank Failures and FDIC Losses in the 1986–92
and 2007–13 Banking Crises.” Working Paper 1719, Federal Reserve
Bank of Cleveland, OH, December 4, 2017.

Poterba, James M. “Tax Reform and the Housing Market in the Late
1980s: Who Knew What, and When Did They Know It?” In Real Estate
and the Credit Crunch, Proceedings of a Conference Held in September
1992, edited by Lynn E. Browne and Eric S. Rosengren, 230–261. Boston:
Federal Reserve Bank of Boston, 1992.
Wheelock, David C. and Paul W. Wilson. “Why Do Banks Disappear?
The Determinants of U.S. Bank Failures and Acquisitions.” Review of
Economics and Statistics 82, no. 1 (February 2000): 127–138. https://

Cole, Rebel A. and Lawrence J. White. “Déjà vu All Over Again: The
Causes of U.S. Commercial Bank Failures This Time Around.” Journal of
Financial Services Research 42, no. 1-2 (October 2012): 5–29. https://
Dagher, Jihad, Giovanni Dell'Ariccia, Luc Laeven, Lev Ratnovski, and Hui
Tong. “Benefits and Costs of Bank Capital.” IMF Staff Discussion Note
16/04, International Monetary Fund, Washington, DC, March 3, 2016.
DiSalvo, James and Ryan Johnston. “Banking Trends: The Growing Role
of CRE Lending.” Economic Insights 1, no. 3 (Third Quarter 2016): 15–21.
Federal Deposit Insurance Corporation. “Guidance on Concentrations in
Commercial Real Estate Lending, Sound Risk Management Practices.”
Financial Institution Letter FIL-104-2006, Arlington, VA, December 12,
Federal Deposit Insurance Corporation. History of the Eighties—Lessons
for the Future.” Arlington, VA, 1997.
Cole, Rebel A. and George W. Fenn. “The Role of Commercial Real Estate
Investments in the Banking Crisis of 1985–92.” MPRA Paper 24692,
University Library of Munich, Germany, August 2006. https://mpra.
Friend, Keith, Harry Glenos, and Joseph B. Nichols. An Analysis of the
Impact of the Commercial Real Estate Concentration Guidance.
Washington, DC: Federal Reserve Board and Office of the Comptroller
of the Currency, April 2013.

Banking Trends: Estimating Today's Commercial Real Estate Risk
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


Research Update
These papers by Philadelphia Fed economists,
analysts, and visiting scholars represent
preliminary research that is being circulated
for discussion purposes.

Firm Wages in a Frictional Labor Market

The views expressed in these papers are
solely those of the authors and should not
be interpreted as reflecting the views of
the Federal Reserve Bank of Philadelphia
or Federal Reserve System.

How Big Is the Wealth Effect? Decomposing the
Response of Consumption to House Prices

This paper studies a labor market with directed search, where multiworker firms follow a firm wage policy: They pay equally productive
workers the same. The policy reduces wages, due to the influence
of firms' existing workers on their wage-setting problem, increasing
the profitability of hiring. It also introduces a time-inconsistency into
the dynamic firm problem, because firms face a less elastic labor
supply in the short run. To consider outcomes when firms reoptimize
each period, I study Markov perfect equilibria, proposing a tractable
solution approach based on standard Euler equations. In two applications, I first show that firm wages dampen wage variation over the
business cycle, amplifying that in unemployment, with quantitatively
significant effects. Second, I show that firm-wage firms may find
it profitable to fix wages for a period of time, and that an equilibrium
with fixed wages can be good for worker welfare, despite added
volatility in the labor market.
Working Paper 19-05. Leena Rudanko, Federal Reserve Bank of

We investigate the effect of declining house prices on household
consumption behavior during 2006–2009. We use an individuallevel dataset that has detailed information on borrower characteristics,
mortgages, and credit risk. Proxying consumption by individual-level
auto loan originations, we decompose the effect of declining house
prices on consumption into three main channels: wealth effect,
household financial constraints, and bank health. We find a negligible
wealth effect. Tightening household-level financial constraints can
explain 40–45 percent of the response of consumption to declining
house prices. Deteriorating bank health leads to reduced credit supply
both to households and firms. Our dataset allows us to estimate the
effect of this on households as 20–25 percent of the consumption
response. The remaining 35 percent is a general equilibrium effect that
works via a decline in employment as a result of either lower credit
supply to firms or the feedback from lower consumer demand. Our
estimate of a negligible wealth effect is robust to accounting for the
endogeneity of house prices and unemployment. The contribution
of tightening household financial constraints goes down to 35 percent,
whereas declining bank credit supply to households captures about
half of the overall consumption response, once we account for
Working Paper 19-06. S. Borağan Aruoba, University of Maryland and
Federal Reserve Bank of Philadelphia Research Department Visiting
Scholar; Ronel Elul, Federal Reserve Bank of Philadelphia; Şebnem
Kalemli-Özcan, University of Maryland.


Federal Reserve Bank of Philadelphia
Research Department

Research Update
2019 Q1

Incumbency Disadvantage of Political Parties:
The Role of Policy Inertia and Prospective Voting

The Roles of Alternative Data and Machine Learning
in Fintech Lending: Evidence from the LendingClub
Consumer Platform

We document that postwar U.S. elections show a strong pattern of
“incumbency disadvantage”: If a party has held the presidency of the
country or the governorship of a state for some time, that party tends
to lose popularity in the subsequent election. To explain this fact,
we employ Alesina and Tabellini's (1990) model of partisan politics,
extended to have elections with prospective voting. We show that
inertia in policies, combined with sufficient uncertainty in election
outcomes, implies incumbency disadvantage. We find that inertia
can cause parties to target policies that are more extreme than the
policies they would support in the absence of inertia and that such
extremism can be welfare reducing.
Supersedes Working Paper 17-43.
Working Paper 19-07. Satyajit Chatterjee, Federal Reserve Bank of
Philadelphia; Burcu Eyigungor, Federal Reserve Bank of Philadelphia.

Fintech has been playing an increasing role in shaping financial and
banking landscapes. There have been concerns about the use of
alternative data sources by fintech lenders and the impact on financial
inclusion. We compare loans made by a large fintech lender and
similar loans that were originated through traditional banking channels.
Specifically, we use account-level data from LendingClub and Y-14M
data reported by bank holding companies with total assets of $50
billion or more. We find a high correlation with interest rate spreads,
LendingClub rating grades, and loan performance. Interestingly, the
correlations between the rating grades and FICO scores have declined
from about 80 percent (for loans that were originated in 2007) to
only about 35 percent for recent vintages (originated in 2014–2015),
indicating that nontraditional alternative data have been increasingly
used by fintech lenders. Furthermore, we find that the rating grades
(assigned based on alternative data) perform well in predicting loan
performance over the two years after origination. The use of alternative data has allowed some borrowers who would have been
classified as subprime by traditional criteria to be slotted into “better”
loan grades, which allowed them to get lower-priced credit. In addition,
for the same risk of default, consumers pay smaller spreads on loans
from LendingClub than from credit card borrowing.
Supersedes Working Paper 17-17.
Working Paper 18-15 Revised. Julapa Jagtiani, Federal Reserve Bank
of Philadelphia; Catharine Lemieux, Federal Reserve Bank of Chicago.

Research Update
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


From Incurred Loss to Current Expected Credit
Loss (CECL): A Forensic Analysis of the Allowance
for Loan Losses in Unconditionally Cancelable
Credit Card Portfolios
The Current Expected Credit Loss (CECL) framework represents
a new approach for calculating the allowance for credit losses. Credit
cards are the most common form of revolving consumer credit and
are likely to present conceptual and modeling challenges during CECL
implementation. We look back at nine years of account-level credit
card data, starting with 2008, over a time period encompassing the
bulk of the Great Recession as well as several years of economic
recovery. We analyze the performance of the CECL framework under
plausible assumptions about allocations of future payments to
existing credit card loans, a key implementation element. Our analysis
focuses on three major themes: defaults, balances, and credit loss.
Our analysis indicates that allowances are significantly impacted by
specific payment allocation assumptions as well as downturn
economic conditions. We also compare projected allowances with
realized credit losses and observe a significant divergence resulting
from the revolving nature of credit card portfolios. We extend our
analysis across segments of the portfolio with different risk profiles.
Interestingly, fewer risky segments of the portfolio are proportionally
more impacted by specific payment assumptions and downturn
economic conditions. Our findings suggest that the effect of the
new allowance framework on a specific credit card portfolio will
depend critically on its risk profile. Thus, our findings should be
interpreted qualitatively, rather than quantitatively. Finally, the goal
is to gain a better understanding of the sensitivity of allowances to
plausible variations in assumptions about the allocation of future
payments to present credit card loans. Thus, we do not offer specific
best practice guidance.
Working Paper 19-08. José J. Canals-Cerdá, Federal Reserve Bank of

Investigating Nonneutrality in a State-Dependent
Pricing Model with Firm-Level Productivity Shocks
In recent years, there has been an abundance of empirical work
examining price-setting behavior at the micro level. First-generation
models with price-setting rigidities were generally at odds with much
of the microprice data. A second generation of models, with fixed
costs of price adjustment and idiosyncratic shocks, have attempted
to rectify this shortcoming. Using a model that matches a large set of
microeconomic facts, we find significant nonneutrality. We decompose
the nonneutrality and find that state dependence plays an important
part in the responses of output and inflation to a monetary shock. We
also examine how aggregating firm behavior can generate flat hazards.
Last, we find that the steady state statistic developed by Alvarez,
Le Bihan, and Lippi (2016) is an imperfect guide to characterizing
nonneutrality in our model.
Working Paper 19-09. Michael Dotsey, Federal Reserve Bank of
Philadelphia; Alexander L. Wolman, Federal Reserve Bank of Richmond.

Frictional Intermediation in Over-the-Counter
We extend Duffie, Gârleanu, and Pedersen's (2005) search-theoretic
model of over-the-counter (OTC) asset markets, allowing for a decentralized interdealer market with arbitrary heterogeneity in dealers'
valuations or inventory costs. We develop a solution technique that
makes the model fully tractable and allows us to derive, in closed
form, theoretical formulas for key statistics analyzed in empirical
studies of the intermediation process in OTC markets. A calibration to
the market for municipal securities reveals that the model can generate
trading patterns and prices that are quantitatively consistent with
the data. We use the calibrated model to compare the gains from trade
that are realized in this frictional market with those from a hypothetical,
frictionless environment, and to distinguish between the quantitative
implications of various types of heterogeneity across dealers.
Supersedes Working Paper 15-22.
Working Paper 19-10. Julien Hugonnier, EPFL and Swiss Finance
Institute; Benjamin Lester, Federal Reserve Bank of Philadelphia;
Pierre-Olivier Weill, UCLA and Visiting Scholar, Federal Reserve Bank
of Philadelphia Research Department.


Federal Reserve Bank of Philadelphia
Research Department

Research Update
2019 Q1

The Paper Trail of Knowledge Spillovers:
Evidence from Patent Interferences

A Dynamic Model of Intermediated Consumer
Credit and Liquidity

We show evidence of localized knowledge spillovers using a new database of U.S. patent interferences terminated between 1998 and
2014. Interferences resulted when two or more independent parties
submitted identical claims of invention nearly simultaneously. Following the idea that inventors of identical inventions share common
knowledge inputs, interferences provide a new method for measuring
knowledge spillovers. Interfering inventors are 1.4 to 4 times more
likely to live in the same local area than matched control pairs of
inventors. They are also more geographically concentrated than
citation-linked inventors. Our results emphasize geographic distance
as a barrier to tacit knowledge flows.

We construct a model of consumer credit with payment frictions, such
as spatial separation and unsynchronized trading patterns, to study
optimal monetary policy across different interbank market structures.
In our framework, intermediaries play an essential role in the functioning of the payment system, and monetary policy influences the
equilibrium allocation through the interest rate on reserves. If interbank
credit markets are incomplete, then monetary policy plays a crucial
role in the smooth operation of the payment system. Specifically, an
equilibrium in which privately issued debt claims are not discounted is
shown to exist provided the initial wealth in the intermediary sector
is sufficiently large relative to the size of the retail sector. Such an equilibrium with an efficient payment system requires setting the interest
rate on reserves sufficiently close to the rate of time preference.

Working Paper 17-44 Revised. Ina Ganguli, University of
Massachusetts–Amherst; Jeffrey Lin, Federal Reserve Bank of
Philadelphia; Nicholas Reynolds, Brown University.

Working Paper 19-12. Pedro Gomis-Porqueras, Deakin University;
Daniel Sanches, Federal Reserve Bank of Philadelphia.

Toward a Framework for Time Use, Welfare, and
Household-Centric Economic Measurement
What is meant by economic progress, and how should it be measured?
The conventional answer is growth in real GDP over time or compared
across countries, a monetary measure adjusted for the general rate of
increase in prices. However, there is increasing interest in developing
an alternative understanding of economic progress, particularly in
the context of digitalization of the economy and the consequent
significant changes Internet use is bringing about in production and
household activity. This paper discusses one alternative approach,
combining an extended utility framework considering time allocation
over paid work, household work, leisure, and consumption with
measures of objective or subjective well-being while engaging in
different activities. Developing this wider economic welfare measure
would require the collection of time use statistics as well as wellbeing data and direct survey evidence, such as the willingness to pay
for leisure time. We advocate an experimental set of time and
well-being accounts, with a particular focus on the digitally driven
shifts in behavior.
Working Paper 19-11. Diane Coyle, University of Cambridge; Leonard
Nakamura, Federal Reserve Bank of Philadelphia.

Research Update
2019 Q1

Federal Reserve Bank of Philadelphia
Research Department


A Shortage of Short Sales: Explaining the
Underutilization of a Foreclosure Alternative
The Great Recession led to widespread mortgage defaults, with borrowers resorting to both foreclosures and short sales to resolve their
defaults. I first quantify the economic impact of foreclosures relative
to short sales by comparing the home price implications of both.
After accounting for omitted variable bias, I find that homes selling as
short sales transact at 9.2% to 10.5% higher prices on average than
those that sell after foreclosure. Short sales also exert smaller negative externalities than foreclosures, with one short sale decreasing
nearby property values by 1 percentage point less than a foreclosure.
So why weren't short sales more prevalent? These home price
benefits did not increase the prevalence of short sales because free
rents during foreclosures caused more borrowers to select foreclosures,
even though higher advances led servicers to prefer more short sales.
In states with longer foreclosure timelines, the benefits from foreclosures increased for borrowers, so short sales were less utilized.
I find that one standard deviation increase in the average length of the
foreclosure process decreased the short sale share by 0.35 to 0.45
standard deviation. My results suggest that policies that increase the
relative attractiveness of short sales could help stabilize distressed
housing markets.
Working Paper 19-13. Calvin Zhang, Federal Reserve Bank of

Banking Regulation with Risk of Sovereign Default
Banking regulation routinely designates some assets as safe and thus
does not require banks to hold any additional capital to protect
against losses from these assets. A typical such safe asset is domestic
government debt. There are numerous examples of banking regulation
treating domestic government bonds as “safe,” even when there is
clear risk of default on these bonds. We show, in a parsimonious
model, that this failure to recognize the riskiness of government debt
allows (and induces) domestic banks to “gamble” with depositors'
funds by purchasing risky government bonds (and assets closely
correlated with them). A sovereign default in this environment then
results in a banking crisis. Critically, we show that permitting banks
to gamble this way lowers the cost of borrowing for the government.
Thus, if the borrower and the regulator are the same entity (the
government), that entity has an incentive to ignore the riskiness of
the sovereign bonds. We present empirical evidence in support of the
key mechanism we are highlighting, drawing on the experience of
Russia in the run-up to its 1998 default and on the recent Eurozone
debt crisis.

We Are All Behavioral, More or Less: Measuring
and Using Consumer-Level Behavioral Sufficient
Can a behavioral sufficient statistic empirically capture crossconsumer variation in behavioral tendencies and help identify whether
behavioral biases, taken together, are linked to material consumer
welfare losses? Our answer is yes. We construct simple consumer-level
behavioral sufficient statistics—“B-counts”—by eliciting 17 potential
sources of behavioral biases per person, in a nationally representative
panel, in two separate rounds nearly three years apart. B-counts
aggregate information on behavioral biases within-person. Nearly all
consumers exhibit multiple biases, in patterns assumed by behavioral
sufficient statistic models (a la Chetty), and with substantial variation
across people. B-counts are stable within-consumer over time, and
that stability helps to address measurement error when using B-counts
to model the relationship between biases, decision utility, and experienced utility. Conditional on classical inputs—risk aversion and
patience, life-cycle factors and other demographics, cognitive and noncognitive skills, and financial resources—B-counts strongly negatively
correlate with both objective and subjective aspects of experienced
utility. The results hold in much lower-dimensional models employing
“Sparsity B-counts” based on bias subsets (a la Gabaix) and/or fewer
covariates, illuminating lower-cost ways to use behavioral sufficient
statistics to help capture the combined influence of multiple behavioral biases for a wide range of research questions and applications.
Working Paper 19-14. Victor Stango, University of California, Davis
and Federal Reserve Bank of Philadelphia Visiting Scholar; Jonathan
Zinman, Dartmouth College and Federal Reserve Bank of Philadelphia
Visiting Scholar.

Working Paper 19-15. Pablo D'Erasmo, Federal Reserve Bank of
Philadelphia; Igor Livshits, Federal Reserve Bank of Philadelphia; and
Koen Schoors.


Federal Reserve Bank of Philadelphia
Research Department

Research Update
2019 Q1


Exploring the
Economic Effects of
the Opioid Epidemic
Monetary Policy
in a Changing Federal
Funds Market
Regional Spotlight:

Growing a Healthier
Regional Economy
Connect with Us
Read previous articles
Economic Insights

Business Review

Sign up for e-mail notifications


Federal Reserve Bank of Philadelphia


Ten Independence Mall
Philadelphia, Pennsylvania 19106-1574

You can find Economic
Insights via the
Research Publications
part of our website.