Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
1
Political Uncertainty and Housing Markets
Van Nguyen† Carles Vergara-Alert‡
Abstract
This paper examines the causal effects of political uncertainty on housing markets. We use the U.S.
gubernatorial elections from 1982 to 2015 as a source of exogenous variation in political uncertainty
and exploit the regional variations in the residential housing market. We use neighboring states without
elections as a control group. We find that higher political uncertainty causes: (1) a decrease in house
prices; (2) a decrease in the number of housing transactions; and (3) an increase in the number of
building permits. These effects are stronger during election years in which the election outcomes present
higher uncertainty. We further examine the impact of political uncertainty on the mortgage market and
we find that both mortgage demand and supply decrease in election years.
Key words: Political uncertainty; housing; mortgages; house prices.
JEL codes: D80, R30, G21, D72
1. Introduction
Political uncertainty has increased in recent times. The U.S. Economic Policy Uncertainty Index
–normalized to 100 as of year 2010– has been only three times above the value of 200 during the
period 1985-2007, but it has been nine times above 200 since the collapse of Lehman Brothers in
September 2008.1 Recent research has shown that there is a direct relationship between political
† IESE Business School. Av. Pearson 21, 08034 Barcelona, Spain. Email: [email protected]. ‡ IESE Business School. Av. Pearson 21, 08034 Barcelona, Spain. Email: [email protected]. Vergara-Alert acknowledges
financial support from the State Research Agency of the Spanish Ministry of Science, Innovation and Universities and from the
European Regional Development Fund (Ref. PGC2018-097335-A-I00, MCIU/AEI/FEDER, UE). 1 See the index of economic policy uncertainty developed in Baker, Bloom, and Davis (2016). During the period 1985-2007, this
index has been only above the value of 200 during Gulf War I, 9/11, and Gulf War II, and only once above 250 (i.e., during the
9/11 event). However, during the period 2008-2019, this index has been nine times above the value of 200 and four times above
2
uncertainty and economic outcomes. However, the impact of political uncertainty on real estate
markets is an understudied area, despite the fact that real estate is one of the largest asset classes
in the economy. Since the Brexit referendum in June 2016, in the U.K., house sales in the prime
area of central London plunged by 19% and house prices went down 14% (between June 2016 and
October 2019).2 Commercial real estate in the U.K. also suffered from the Brexit and the
uncertainty surrounding the U.K.-Europe trade deal. Commercial real estate investments
experienced a sharp drop of 33% in 2016.3 In the U.S., policy uncertainty negatively affects the
growth rates of house prices and housing transactions. Figure 1 shows these negatives
relationships. Nevertheless, house price growth tends to slow down in election years.4 During the
period from 1980 to 2016, house price growth in off-election years have been 0.22% higher than
in election years, on average. In Italy, where political crisis has been ongoing for several years
without a stable government, residential house prices have been falling consistently in real term
since Q1 2007 and stood 29% lower by Q3 2019.5 Investors have long attributed those moves in
housing markets to the uncertainty associated with political events. However, there is a lack of
causal evidence linking political uncertainty to changes in housing markets.
[Insert Figure 1 around here]
In this paper, we attempt to fill this void by empirically examining whether and to what
extent uncertainty associated with potential changes in government leadership affect housing
markets. The public authority has the power to regulate the economy through various tools
including enforcing laws and regulations, imposing taxes, and providing rights and protection.
250 (i.e., during the debt ceiling debate in 2011, the Trump Election in 2016, and twice during the rising tariffs and trade policy
tensions between the U.S. and China in 2018 and 2019). 2 Financial Times, October 24, 2019: “Brexit bargains: London’s luxury homes take big price cuts”
https://www.ft.com/content/b788a366-f4c1-11e9-b018-3ef8794b17c6. 3 CBRE report, February 14, 2020: “From user to guest: The user in the center of Real Estate” (unpublished). 4 Data source from Freddie Mac: http://www.freddiemac.com/research/indices/house-price-index.page. 5 Data source from Bank of International Settlements: https://www.bis.org/statistics/pp_selected.htm?m=6%7C288%7C596.
https://www.ft.com/content/b788a366-f4c1-11e9-b018-3ef8794b17c6http://www.freddiemac.com/research/indices/house-price-index.page
3
These tools play an important role in shaping housing markets. Therefore, any changes in the
structure of the government as well as its policies could result in instability and uncertainty that
influence decisions in the housing markets.
We focus specifically on the U.S. residential housing markets and we study the causal
impact of political uncertainty surrounding elections on both housing and mortgage markets. We
do so because the majority of residential housing purchases are financed with mortgages, which
suggests that there is a strong link between mortgage and housing markets. The key feature of
housing investments is that although housing can offer investors a greater control over the value
and use of their investment, direct investments in housing present sizeable transaction costs. As
argued by the real option theory, transaction costs in housing investments can increase the real
option value of delaying the investment under high-uncertainty circumstances. This might drive a
dampening effect in house prices and transaction activity as investors wait for more and clearer
signals before making investment decisions.
In addition, we argue that political uncertainty can also have an impact on the number of
building permits. In the U.S. market, a building permit is obtained at the city where the construction
will take place. The city is required by law to enforce various construction and development
policies at the local, state, and federal level. Since obtaining a building permit can be taken as
buying an option to build or reconstruct a housing unit, investors might want to avoid the
uncertainty generated by gubernatorial elections as well as the associated potential changes in
construction and development regulations to be implemented by newly-elected government.
Therefore, we argue that investors are more likely to apply for building permits before an election
takes place, which leads to a surge in the number of building permits in election years. As the
number of potential new constructions increases, there would be an expectation of a future increase
in the supply of housing units, which in turn might have further dampening impact on house prices.
4
We use state-level data for 50 states in the U.S from 1982 to 2015 to examine the impact
of political uncertainty on both housing and mortgage markets. Our empirical study can be
summarized in three sets of results. Firstly, we show that political uncertainty has a negative
relationship with house prices and the number of housing transactions, but they have a positive
relationship with the number of building permits that are issued. Our results show that house prices
and the number of housing transactions are reduced by 0.6% and 6.5%, respectively, while the
number of building permits are increased by 2.2% in years of gubernatorial elections. To address
potential endogeneity concerns, we employ the “neighboring state estimation methodology” in
which we compare house prices, number of housing transactions, and the number of building
permits between a state with a gubernatorial election and its neighboring state without an election.
This methodology is based on the assumption that neighboring states share similar unobservable
shocks, which will be explained in detail later in the paper.
Secondly, we demonstrate that the degree of uncertainty matters, that is, the impact of
political uncertainty on housing markets increases with the degree of political uncertainty. In
particular, we provide evidence that there is a greater decrease in house prices and number of
housing transactions, but greater increase in number of building permits when gubernatorial
election outcomes are more unclear.
Thirdly, we demonstrate that political uncertainty causes a decline in residential mortgage
activity on both the demand and the supply sides of the mortgage markets. We find evidence that
suggests that both mortgage supply and mortgage demand fall during gubernatorial election years.
At the bank level, we show that banks reduce the volume of mortgages originated in their
headquarter states when gubernatorial elections are scheduled, while the number of mortgage
applications also drops. However, we find that the magnitude and the significance level are greater
on the demand side at both state and bank level.
5
The use of US gubernatorial elections as a source of political uncertainty has several
advantages in the study. Firstly, the sample for investigation is larger for gubernatorial elections
than it is for presidential elections. Over the course of 34 years from 1982 to 2015, there were 458
gubernatorial election compared to only 8 presidential elections in the US, which cannot be a
reliable source for making meaningful statistical inferences. Secondly, it is important to note that
policy changes are not exogenous by nature as they might be triggered by various political and
economic shocks. Nevertheless, US gubernatorial elections can be taken as exogenous events as
they are prescheduled and not held due to any economic situations. Hence, the use of gubernatorial
elections in our empirical setting helps to mitigate the potential endogeneity problem arising
between political uncertainty and economic conditions. Lastly, the housing market in any state is
significantly affected by changes in the state government’s policies and regulations including but
not limited to changes in building code, tax code, housing subsidies, employee relocation expense,
environmental protection rules. Different governing parties may favor different sets of industries,
and with respect to residential housing, those industry-specific policies can influence the house
demand, supply and prices in the area where firms are under those industry-specific policies.
Our paper contributes to two main bodies of literature. First, we contribute to the growing
body of literature that studies the channel through which economic factors respond to political
uncertainty. Past literature has examined several components of political uncertainty including
political instability (Alesina et al., 1996; Jong-A-Pin, 2009; Aisen el al., 2011), corruption (Leff,
1964; Mauro, 1995; Mo, 2001; Lash, 2004), as well as changes in legal and regulatory
environments (Mauro, 1995; La Porta et al., 1997; Battalio et al., 2011). The majority of theoretical
papers, however, center on the impact of political uncertainty on corporate investment decisions.
The common argument is that uncertainty associated with changes in financial regulation and
macroeconomic policy would hinder the firm’s investment. Early theoretical works (Cukierman,
6
1980; Bernanke, 1983) suggest that firms are more likely to delay investments when they are in
state of uncertainty in order to collect more information. Others (Pindyck, 1988; Ingersoll et al.,
1992; Bloom, 2009) provide theoretical frameworks to identify the effects of political uncertainty
shocks on corporate investment incentives. Several empirical evidences have supported these
theories. Julio and Yook (2012) document across different countries that firms reduce investment
expenses during national election years in which political uncertainty is heightened. Jens (2017)
suggests that firms’ investment at state level in the US is reduced in the period of escalating
political uncertainty due to gubernatorial elections after controlling for the general state economic
conditions. Colak, Durnev and Qian (2017) provide evidence that there are fewer IPOs in a state
when it is holding a gubernatorial election.
Second, our paper contributes to the growing body of literature on the impact of policy and
political uncertainty on asset pricing. Theoretical works by Pastor and Veronesi (2012; 2013)
provide a general equilibrium model showing that political uncertainty can dampen asset prices
and command a risk premium whose magnitude is greater in times of weak economic conditions.
They argue that investors demand higher return on their investment, raising the cost of capital for
firms, which creates more constraints on firms’ investment decisions. Their theoretical prediction
that political uncertainty has negative impacts on asset prices is supported by analysis of within-
country and cross-country evidence. In stock markets, Brogaard and Detzel (2015) employ the
economic policy uncertainty index (EPU) constructed by Baker, Bloom and Davis (as in Baker,
Bloom and Davis, 2016) to identify the negative relation between economic policy uncertainty and
US stock market return. Liu, Shu and Wei (2017) document that political uncertainty associated
with Bo scandal in China caused a significant drop in prices of stocks traded in Chinese market.
In bond markets, evidence found by Gao and Qi (2012) suggest that bond investors are less likely
7
to invest and demand higher risk premium to invest in municipal government bonds when political
uncertainty is high due to gubernatorial elections. Handler and Jankowitsch (2018) study the
impact of political uncertainty on prices and trading activity of Italian sovereign bonds and find a
significant price fall together with lower liquidity in the market before a political event. Our
findings in the residential housing market lend support to their theoretical framework. Recently,
Cao, Li and Liu (2019) showed that political uncertainty affects the volume and outcome of cross-
border acquisitions because upcoming national elections encourage firms to perform cross-border
acquisitions to target countries with free-trade agreements, military allies, or countries with better
governance. They also show that stock returns related to the announcement of cross-border deals
incorporate political uncertainty.
The rest of the paper is organized as follows: Section 2 uses existing option pricing models
in the literature in order to develop our testable hypotheses. Section 3 describes our data. Section
4 defines the identification strategies and reports the empirical results. Section 5 examines the
reverse effect, that is, how housing variables behave in post-election years. Section 6 provides
cross-sectional analyses with two subsamples of high- and low- uncertainty in gubernatorial
election outcomes. Section 7 analyses the potential impact of gubernatorial elections on mortgage
lending activities. Section 8 provides the conclusions of the paper.
2. Hypothesis development
We rely on the real option argument to construct our hypotheses about the impact of
political uncertainty on housing market. Myers (1977) was the one who coined the term “real
options” with the rationale of applying option pricing theory to the valuation of real investments,
8
while taking into account investors’ learning and adaptive behavior. Past theoretical and empirical
studies document that uncertainty negatively affects corporate investment decisions because the
value of the real option of waiting and learning more about the markets increases. (Dixit and
Pindyck, 1994; Leahy and Whited, 1996; Abel and Eberly, 1999; Julio and Yook, 2012; Jens
2016). The intuition behind the real option theory is that uncertainty, in general, can result in bad
economic outcomes; therefore, the option value of delaying investments until uncertainty is
resolved increases. This delaying effect is even stronger with investments that entail high reversal
costs, which have been termed “irreversible investments” by scholars.
Pastor and Veronesi (2012) present a theoretical model to prove that political uncertainty
dampens stock prices and commands a risk premium. In their theoretical prediction, they argue
that there are two effects generated from a political uncertainty on stock prices including an
increase in firms’ expected profitability and an increase in discount rates. Both of the foregoing
result in a decrease in stock prices. Their model also follows the logic of real option argument: the
greater the uncertainty triggered by political events, the greater the real option value of delaying
investing and financing decisions. Pastor and Veronesi (2013), Brogaard and Detzel (2015), Liu,
Shu and Wei (2017) document evidence that stock prices, on average, are dampened in period of
heightened uncertainty surrounding a political event as investors demand higher risk premium to
compensate for taking extra risk generated by political uncertainty.
The real option theory can also be applied to housing investments because these
investments are highly irreversible, which further increases the value of the real option to postpose
investments under circumstances of uncertainty. In particular, political uncertainty is triggered by
potential changes in regulatory environment, political leadership and government policies, which,
in turn, can highly affect investing decisions. In our paper, we employ gubernatorial elections as a
9
source of political uncertainty, which comes in three main forms: (1) uncertainty about the winning
parties; (2) uncertainty about possible changes in policies to be implemented by the newly elected
governors; and (3) uncertainty about the impact of possible policy changes on housing market.
State governments are responsible for implementing state housing policies and programs
using various tools. Housing transaction activity, therefore, can be affected either through direct
channels by the state government’s choices of policies regarding to taxes, housing subsidies,
building codes, landlord-tenant rights and responsibilities; and through indirect channels such as
changes in state citizens’ consumption behavior and living standards. It is from this argument that
we expect a lower house price and a lower number of housing transactions in election years.
Following the real option argument and the expected impact of political uncertainty on the
housing market, we propose the first hypothesis as follows.
Hypothesis 1: The return on house prices and the number of housing transaction are lower in
election years than in non-election years, while the number of building permits is greater in
election years than in non-election years.
The real option argument provides that under risky and uncertain environment, investors
can update their knowledge and adjust their behavior to increase the potential upside and reduce
the potential downside of their investment outcomes. The majority of past literature has come to
support the real option argument to delay investments in period of uncertainty, which means
uncertainty has a negative impact on investment activity. However, there are also other theories
suggesting that the real option framework might not hold when applied to real world situations
(Hartman, 1972; Abel, 1983; Caballero, 1991). The main issue with the real option framework is
that it is based on a strong assumption of replicating portfolio. In essence, it assumes that we can
construct a portfolio of traded investments to replicate the return of the real option free of arbitrage.
10
Moreover, the real option theory shows that the higher the uncertainty, the higher the value
of option to delay investments subject to transaction costs. This equilibrium result leads to our
second hypothesis with regard to the degree of political uncertainty surrounding gubernatorial
elections and its impact on housing market.
Hypothesis 2: The higher the degree of political uncertainty, the larger its impact on the housing
market. Specifically, the decrease in the house prices and number of housing transactions in
election years and the increase in the number of building permits are greater when the election
outcome is more uncertain.
Finally, we argue that political uncertainty matters for mortgage markets. As the majority
of home purchases are performed using leverage, the housing and the mortgage markets are highly
connected. This argument leads to our third hypothesis.
Hypothesis 3: The residential mortgage lending activity is lower in election years than in non-
election years, with reduction in both mortgage volume and number of mortgage applications.
3. Data
The full sample in this study covers state-level housing market data, election data and
economic data for the period from 1982 to 2015 for all 50 states in the U.S. Housing market data
include house price index; number of transactions and number of building permits. House price
indices provides a general picture of the health of the housing market, and is closely watched by a
market participants. In the United States, the two main house price indices frequently cited are
Federal Housing Finance Agency (FHFA) House price index and S&P/Case-Shiller index. Both
attempts to track the repeat sales of single-family housing units in order to measure the housing
value appreciation or depreciation in a particular market. The S&P/Case-Shiller index only cover
11
selected markets including specific metropolitan areas, top-20, top-10 metro areas and nation-
wide, whereas the FHFA House price index covers in-depth more geographical areas including
states, counties and zip codes. The FHFA House price index provides two types of indices:
“purchases only” index and “all transactions” index, which covers the longest history, and offers
the most granular level of housing market data in the U.S. Within the scope of this study, we opt
for state-level “all transactions” FHDA House price index as a proxy for house prices for the period
from 1982-2015 covering 50 states in the U.S. Then, we create a variable RH which measures the
annual return in the house price index. Transaction is the state-level number of housing
transactions per year provided by CoreLogic. Permit is the number of building permits issued by
a state within a year.
The gubernatorial election data is from Congressional Quarterly (CQ) Electronic library.
The data collected includes all election years, candidate names, the winning parties, and voting
margins. Election is a binary variable equal to one if a gubernatorial election occurred in a state.
Gubernatorial elections are held on the first Tuesday of November, with different sates holding
their elections in different years (see Appendix for more detailed description of gubernatorial
elections in the US). There was only one special non-prescheduled recall election that happened
in California in 2003; however, neither the exclusion nor inclusion of this observation affects the
empirical results. Pres_election is a binary variable equal to one if there is a presidential election
during a year in the sample. Incumbent_absence is a hand-collected binary variable equal to one
if the incumbent governor does not take part in the election either due to term limit regulation or
other personal reasons. Winning_margin is the difference between the percentage votes won by
the first and second place candidates. We sort gubernatorial elections based on winning margin
into terciles, then define Election_closeness as a binary variable equal to one if the election
winning margin is in the lowest tercile (5.1%).
12
Other variables included to control for the state-level economic conditions are retrieved
from the Bureau of Economic Analysis (bea.gov), the Bureau of Labor Statistics (bls.gov), the U.S
Census Bureau (census.gov) and the Federal Emergency Management Agency (fema.gov)
respectively: GDP as the annual growth rate in state gross domestic product; Unemployment as the
annual state unemployment rate; Population as the annual relative change in the state population;
Emergency as a binary variable equal to one if a state declares emergency in a given year.
Recession as a binary variable equal to one if the year is identified as being in a recession state by
the National Bureau of Economic Research (NBER). Other control variables include S&P500 as
annual return for the S&P500 index, and Bond_rate as the interest rates on the U.S. government
bonds from FRED database.
Table 1 presents descriptive statistics for the whole sample of 1700 state-year observations
as well as the sub-samples of observations with gubernatorial elections only. The return on house
price index (RH) in the two subsamples appears to be of smaller magnitude than in the whole
sample (3% in the subsamples versus 4% in the whole sample). The mean of the number of housing
transactions (Transaction) is also lower in the two subsamples than in the whole sample (46,508
in the subsample with gubernatorial elections only and 45,715 in the subsample with close
elections only versus 49,690 in the whole sample). The number of building permits (Permit), on
the contrary, exhibits higher means in the two subsamples (27,394 and 27,662 in the two
subsamples versus 26,386 in the whole sample). All these measures in the housing market display
stronger volatility in two subsamples with gubernatorial elections than in the whole sample.
[Insert Table 1 around here]
13
4. Identification strategies and empirical results
4.1. Baseline OLS multivariate analysis
We perform the following OLS regressions to examine the impact of political uncertainty
resulting from gubernatorial elections on housing variables including the return on HPI, the
number of housing transactions and the number of building permits:
RH i,t = a0 + a1Electioni,t + a2 Zi,t + a3Xt + μt + εi,t (1a)
Transactioni,t = b0 + b1Electioni.t + b2 Zi,t + a3Xt + μt + εi,t (1b)
Permiti,t = c0 + c1Electioni,t + c2 Zi.t + c3Xi,t + μt + εi,t (1c)
where RHi,t , Transactioni,t and Permiti,t are respectively the return on HPI ; the number of housing
transactions; and the number of building permits in state i and year t ; Our main variable of interest
is Electioni,t that equals 1 if there is a gubernatorial election in state i and year t, and 0 otherwise;
Zi.t is the set of control variables at state level; Xt is the set of control variable at country-wide
level; εi,t is the error term and μt is the state fixed effects. We include state fixed effects to address
the unobserved heterogeneity across states in terms of economic and political environment, and
use the 2-dimensional standard error clustering by state and year to address the concerns of
heteroskedasticity as well as correlations in the error terms.
Table 2 (Panel A, B, C) present the regression results of our multivariate analyses. The
results show that the coefficient of the election binary Election is always negative and highly
significant (with p_value ranging from 1% to 4%) for OLS regressions of equations (1a) and (1b),
implying that, after controlling for the overall health of the economy and heterogeneity across
states, political uncertainty caused by gubernatorial elections tends to decrease HPI (-0.6%) and
the number of housing transactions (approximately -3219 transactions). The regression on the
14
number of building permits, however, has a positive and significant at 5% level coefficient of the
election binary variable, suggesting an increase in permits granted in the year leading to
gubernatorial election (approximately +583 permits).
[Insert Table 2 around here]
4.2. Neighboring state analysis
One benefit of using data on gubernatorial elections is that these elections are considered
prescheduled events, so they can be taken as exogenous events. Hence, the problem of potential
endogeneity between political uncertainty and other economic movements is alleviated to some
extent. However, there could exist other unobservable shocks in socioeconomic environment that
are not adequately captured within the above model specifications. We address this concern by
employing the neighboring-state difference-in-difference methodology (Colak, Durnev and Qian,
2017) with the assumption that neighboring states share similar unobserved shocks. Then by taking
the differences in the dependent variables, the unobservable shocks denoted as αi,t can be canceled
out.
State i: RHi,t = a0 + a1Electioni,t + a2 Zi,t + a3Xt + a4αi,t + μt + εi,t (2)
State j: RHj,t = a0 + a1Electionj,t + a2 Zj,t + a3Xt + a4αj,t + μt + εj,t (3)
where RHi,t is the change in HPI in state i and year t ; Electioni,t is the main variable of interest
that equals to 1 if there is a gubernatorial election in state i and year t, and 0 otherwise; Zi.t is the
set of control variables at state level; Xt is the set of control variable at country-wide level; αi,t is
the unobserved state variables; εi,t is the error term and μt is the fixed effects.
Taking the difference of equations (2) and (3) to derive the difference-in-difference equation:
(RHi,t - RHj,t) = a0 + a1 (Electioni,t – Electionj,t) + a2 (Zi.t - Zj.t ) + a4(αi,t - αj,t) + (εi,t - εj,t ). (4)
15
The assumption made in this setting is that the housing market variables including the return on
HPI, the number of housing transactions, and the number of building permits are subject to similar
unobservable state shocks; therefore, the term (αi,t - αj,t) is canceled out, resulting in the following
regressions:
∆RHi,j,t = a0 +a1 ∆Electioni,j,t + a2 ∆Zi,j,t + ∆εi,j,t (5a)
∆Transactioni,j,t =b0 + b1 ∆Electioni,j,t + b2 ∆Zi,j,t + ∆εi,j,t (5b)
∆Permiti,j,t = c0 + c1 ∆Electioni,j,t + c2 ∆Zi,j,t + ∆εi,j,t. (5c)
We keep all observations with ∆Electioni,j,t =1 and 0 only; hence, the variable ∆Electioni,j,t becomes
a binary variable for all observations for use in our regressions. In total, there are 3575 state pair-
year observations out of which 903 observations with ∆Electioni,j,t =1 (a state with an election and
its neighbor without one), and 2672 observations with ∆Electioni,j,t = 0 (both states without
election).
Table 3 presents 50 states in the U.S and their neighboring states (denoted as NS). Two
states Alaska and Hawaii have no neighbors, and their data will be excluded from our tests using
this methodology. On average, each state has four neighboring states with Maine has only one
neighboring state, and Missouri and Tennessee share borders with eight other states.
[Insert Table 3 around here]
Table 4 (Panel A, B, C) report the OLS regression results for equations (5a), (5b), and (5c)
respectively, with different sets of control variables. We also perform weighted least squared
regression to account for the fact that our state pair-year observations have different probabilities
of being sampled, with weight being number of neighboring states. States with more neighboring
states have greater probability of being selected in the sample. The main variable of interest is
16
∆Election, and the dependent variables are ∆RH (the difference in the return on HPI); ∆Transaction
(the difference in the number of housing transaction) and ∆Permit (the difference in the number
of building permits) between a state and its neighboring state. Compare with the coefficients of
Election in regression equations (1a), (1b), (1c), the coefficients of ∆Election in regression
equations (5a), (5b), and (5c) are smaller in magnitude but much larger in statistical significance.
In Table 4.A, the coefficient of ∆Election is -0.001, implying the HPI is lower by 0.1% in a state
that holds a gubernatorial election compared to its neighboring state without one. The negative
impact of political uncertainty on housing transaction activities is illustrated in Table 4.B with the
coefficient of ∆Election equals -1029.329 which implies that there are approximately 1029 fewer
transactions. In Table 4.C, it is shown that there are approximately 183 more building permits in
a state with a gubernatorial election compared to its neighbor that does not hold one. We
acknowledge that different states have different number of neighboring states, and we address this
issue by including also the weighted least squared regressions as shown in specification (4) in
Table 4 (Panel A, B, C); however, the sign and the level of significance of the coefficient of the
election binary ∆Election remain unchanged.
[Insert Table 4 around here]
4.3. Falsification tests
In order to validate the neighboring-state estimation methodology employed in this study,
we perform two placebo tests (falsification testing) including:
(1) Keeping election years as in the sample but randomly mapping neighboring states (every state
with four other states (as the average number of neighboring states is four in the sample);
(2) Keeping neighboring states as in the sample, but randomly assigning election years within a
four-year election cycle.
17
We estimate regressions similar to equation (5a), (5b) and (5c). As expected, we do not
obtain any significant results for the coefficients (a1), (b1) and (c1).
5. Reverse effect: Post-election analysis
As explained earlier with the real option value theory, the combined effect of a reduced
investment in housings and an increased number of building permits issued would further dampen
house prices in election years. In this section, we investigate the reverse effect of housing market
variables in the first two years following a gubernatorial election. Specifically, we test whether the
return on HPI and the number of housing transactions would increase, while the number of building
permits would decrease in the two years post-election. We estimate regressions similar to
equations (1a), (1b) and (1c); however, we replace the Election variable by the binary variable for
year T=1 and T=2 (with the election year set at T=0). The full set of control variables at the state-
and country-level is still applied. We estimate the following regressions:
RH i,t = a0 + a1Post_election_1i,t + a2 Zi,t + a3Xt + μt + εi,t (6a)
RH i,t = a0 + a1Post_election_1i,t + a2Post_election_2i,t + a3 Zi,t + a4Xt + μt + εi,t (6b)
Transactioni,t = b0 + b1Post_election_1i.t + b2 Zi,t + a3Xt + μt + εi,t (6c)
Transaction i,t = b0 + b1Post_election_1i,t + b2Post_election_2i,t + b3 Zi,t + b4Xt + μt + εi,t (6d)
Permiti,t = c0 + c1Post_election_1i,t + c2 Zi.t + c3Xi,t + μt + εi,t (6e)
Permit i,t = c0 + c1Post_election_1i,t + c2Post_election_2i,t + c3 Zi,t + c4Xt + μt + εi,t (6f)
Figure 2 illustrates the average return on HPI, the average number of housing transactions,
and the average number of building permits over the 4-year election cycle. Table 6 presents the
18
results of six OLS regressions. In columns 1, 3, and 5, the main variable of interest is the 1-year
post election binary (Post_election_1), and the coefficient of the binary variable is highly
significant and positive for RH and Transaction, but negative for Permit as expected. In columns
2, 4 and 6, the two main variables of interest are 1-year and 2-year post election binary
(Post_election_1 and Post_election_2), and the coefficients of the binary variables are
significantly positive for RH and Transaction, but negative for Permit. Economically, these results
imply that the return on HPI one-year post-election increases by 0.5%; the number of transactions
increases by approximately 915, the number of building permits issued decreases by approximately
418.
[Insert Figure 2 around here]
The results are consistent with our argument that investors delay making housing
transactions while applying for more building permits in the environment of greater political
uncertainty due to elections; however, once the uncertainty is resolved, there is a substantial
reverse effect observed in the housing market at state-level.
[Insert Table 5 around here]
6. Cross-sectional analysis: The uncertainty degree of election outcomes
There are 903 observations in the merged difference-in-difference neighboring state
sample in which gubernatorial election is held only in a state and not in its neighboring state. We
first classify elections into two subsamples: High_Uncertainty and Low_Uncertainty. These
subsamples are based on two measures: (1) Close as a binary variable equal to one if the election
winning margin is in the lowest tercile (5.1%), and zero otherwise; (2) Incumbent_absence is a
hand-collected binary variable equal to one if the incumbent governor did not take part in the
19
election either due to term limit regulation or other personal reasons. As most of the incumbent
governors are re-elected (82.8% in our sample spanning from 1982-2015) if they are not restricted
by term-limit regulation or voluntarily withdraw from re-election due to other personal reasons,
we argue that the absence of the incumbent might lead to higher uncertainty surrounding an
election.
To examine whether the impact of political uncertainty on housing markets differs
between two subsamples, we then re-estimate the equations (5a), (5b), and (5c) for the
High_Uncertainty and Low_Uncertainty election subsamples. We find that the estimated
coefficient of ∆Election is more negative in equation (5a) and equation (5b), but more positive in
equation (5c) for the High_Uncertainty subsample than for the Low_Uncertainty subsample. We
further perform Wald test of the difference in the coefficients between each pair of
High_Uncertainty and Low_Uncertainty regressions. As reported in Table 6 (Panel A, B, C), the
difference is significant for the classification of elections based on both Election_closeness and
Incumbent_absence.
Accordingly, these results show that higher- uncertainty gubernatorial elections lead to a
greater decline in house price index and the number of housing transactions, but a greater increase
in number of building permits compared to lower-uncertainty elections. In other words, the
negative impact on house prices and transaction activities is mainly due to elections with highly
uncertain outcomes.
[Insert Table 6 around here]
20
7. Gubernatorial elections and mortgage market
Most housing investors need leverage to finance their investments with mortgage loans. As
a result, housing and mortgage markets are closely connected. Therefore, effects of political
uncertainty on housing markets could come from the supply of housing financing. In this section,
we explore the mortgage lending channel. We investigate whether high political uncertainty is also
associated with reduced mortgage lending activities at both state level and bank level.
7.1. Data
We use the annual mortgage loan data between 1990 and 2015 from the Home Mortgage
Disclosure Act (HMDA). The loan-level HMDA data provides details on a substantial proportion
of mortgage market as it has been a mandatory reporting for most regulated depository institutions
including banks, credit unions, thrifts and other finance companies in the United States. The
HMDA data includes loan application, approval status, loan amount, as well as details on
borrowers and lenders. We only include home-purchase mortgage loans from bank lenders, and
we drop mortgages with missing loan characteristics such as loan size, loan type, lender’s approval
decision. We winsorize the mortgage loans based on their sizes at 1% level. We also exclude
mortgages subsidized by the Veterans Administration and the Federal Housing Authority. Then,
we aggregate the data and derive three main variables at both state level and bank level for our
analysis: log(Volume) as the natural logarithm of the volume of mortgages originated (with volume
in thousands of dollars), Application as the annual relative change in the number of mortgage
applications and Approval_rate as the ratio of the number of mortgage applications approved over
the total number of mortgage applications.
For the bank-level analysis, we use the data on banks’ financial statements collected from
Call Reports. The Call Reports data also provide us with the location of a bank’s headquarter,
21
which allows us to identify the home state of each bank in the sample. To examine the within-bank
variations in mortgage lending over time, we merge the annual HMDA data with the Call Reports
data in the period 1990-2015. Our sample results in 104,171 observations at the bank-, state-, and
year-level. We derive from the Call Reports data various variables representing banks’ financial
characteristics that may affect their lending decisions: Size as the logarithm of total Assets;
Return_on_equity as the ratio of net income to total equity; Liquid_assets as the ratio of liquid
assets to total assets; Home_mortgages as the ratio of residential mortgage loans to total loans;
Core_deposit as the ratio of core deposit to total assets; and Deposit_cost as the ratio of interest
expense on core deposit to core deposit. Detail descriptions of the variables used in this section
are in the Appendix. Table 7 presents the summary statistics of the variables that we use in this
analysis.
[Insert Table 7 around here]
7.2. Methodology
At the state level analysis, we still employ the baseline multivariate regression analysis as
well as the difference – in – difference neighboring state methodology as described in section 4.
We examine both the supply side and the demand side of the mortgage market in election years.
In the baseline analysis, we use two dependent variables to capture the mortgage supply over time.
First, we use log(Volume)i,t which measures the logarithm of total mortgage volume originated in
state i in year t . Second, we use Approval_ratei,t which measures the mortgage approval rate in
state i in year t. To capture the mortgage demand overtime, we use Applicationi,t which measures
the relative change in the total number of mortgage applications in state i in year t. We then do the
following regressions:
log(Volume)i,t = a0 + a1Electioni,t + a2 Zi,t + a3Xt + μt + εi,t (7a)
22
Approval_ratei,t = b0 + b1Electioni.t + b2 Zi,t + a3Xt + μt + εi,t (7b)
Applicationi,t = c0 + c1Electioni,t + c2 Zi.t + c3Xi,t + μt + εi,t (7c)
Our main variable of interest is Electioni,t that equals 1 if there is a gubernatorial election in state
i and year t, and 0 otherwise; Zi.t is the set of control variables at state level; Xt is the set of control
variable at country-wide level; εi,t is the error term and μt is the state fixed effects. Table 8 reports
estimation results for the baseline multivariate regressions (7a), (7b), and (7c).
[Insert Table 8 around here]
For the difference-in-difference neighboring methodology, we consider two dependent variables
to capture mortgage supply over time. First, we use ∆log(Volume)i,j,t which measures the difference
in the logarithm of total mortgage volume originated between state i and state j in year t . Second,
we use ∆Approval_ratei,j,t which measures the difference in the mortgage approval rate between
state i and state j in year t. For the demand side, we use ∆Applicationi,j,t which measures the
difference in the relative change in the total number of mortgage applications between state i and
state j in year t. We then do the following regressions:
∆log(Volume)i,j,t = a0 +a1 ∆Electioni,j,t + a2 ∆Zi,j,t + ∆εi,j,t (8a)
∆Approval_ratei,j,t =b0 + b1 ∆Electioni,j,t + b2 ∆Zi,j,t + ∆εi,j,t (8b)
∆Applicationi,j,t = c0 + c1 ∆Electioni,j,t + c2 ∆Zi,j,t + ∆εi,j,t. (8c)
We keep all observations with ∆Electioni,j,t =1 and 0 only; hence, the variable ∆Electioni,j,t becomes
a binary variable for all observations for use in our regressions. In total, there are 2635 state pair-
year observations out of which 589 observations with ∆Electioni,j,t=1 (a state with an election and
its neighboring state without an election), and 2037 observations with ∆Electioni,j,t = 0 (both states
23
without election) for the period from 1990 to 2015. Table 9 reports estimation results for diff-in-
diff neighboring state specifications (8a), (8b), and (8c).
[Insert Table 9 around here]
At the bank level analysis, we focus on bank lenders and drop all mortgages originated by
credit unions, thrifts, and other non-bank lenders. We also use the timing of gubernatorial elections
as a proxy for exogenous political shocks and employ difference-in-difference with fixed effects
methodology to study the dynamics of banks’ mortgage lending in election years. For the supply
side, we consider two dependent variables that capture banks’ mortgage lending activities:
log(Volume)b,i,t as the natural logarithm of mortgage volume originated by bank b in state i in year
t ; and Approval_rateb,i,t as the approval rate of bank b in state i in year t . For the demand side,
we use Applicationb,i,t which measures the annual relative change in the number of mortgage
applications received by bank b in state i in year t. We use difference-in-difference methodology
and estimate the following specifications:
log(Volume)b,i,t = a0 + a1Electioni,t + a2 Bb,i,t-1 + a3 Zi,t-1 + a4Xt-1 + μt + μb + εb,i,t (9a)
Approval_rateb,i,t = a0 + a1Electioni,t + a2 Bb,i,t-1 + a3 Zi,t-1 + a4Xt-1 + μt + μb + εb,i,t (9b)
Applicationb,i,t = a0 + a1Electioni,t + a2 Bb,i,t-1 + a3 Zi,t-1 + a4Xt-1 + μt + μb + εb,i,t (9c)
Our main variable of interest is Electioni,t , which is equal to 1 if there is a gubernatorial
election in state i and year t, and 0 otherwise. Let Bb,i,t-1 denote a set of control variables at the bank
level; Zi,t-1 is the set of control variables at the state level. Xt-1 denotes the set of control variables
at country-wide level. We also include time fixed effects and bank fixed effects in this estimation.
Table 10 exhibits the estimation results for the specifications (9a), (9b), and (9c).
[Insert Table 10 around here]
24
8. Concluding remarks
This paper shows large and significant dampening effects in the performance of the
residential housing market in the U.S. in periods of high political uncertainty. We employ
gubernatorial elections in the U.S. and an identification based on neighboring states to study the
causal effect of political uncertainty on the housing markets. We find that the return on house price
index and the number of transactions are both lower in the election year, suggesting a lower level
of house prices as investors demand higher premium for taking political risk. Our results also
suggest that there is an increase in the number building permits granted in the period leading to
the gubernatorial elections. This is because applying for a building permit is equivalent to buying
an option to construct in the future when the uncertainty is resolved. Therefore, permit holders can
hedge against potential regulatory changes from the newly elected government. Consequently, we
find that these results are stronger during election years in which the election outcomes are more
uncertain.
This study contributes to the growing body of literature on the impact of political and policy
uncertainty on the economy by demonstrating two points. First, uncertainty from the political
sources can significantly affect investors’ and households’ decision-making in the residential
housing market. Second, political uncertainty has a large effect on the lending activity in the
financial sector in both the supply and the demand sides of the mortgage markets.
25
References
Abel, A., 1983. Optimal Investment under Uncertainty. American Economic Review 73, 228-233
Aisen, A., José-Veiga, F., 2013. How does political instability affect economic growth?. European
Journal of Political Economy 29, 151-167.
Alesina, A., Ozler, S., Roubini, N., Swagel, P., 1996. Political Instability and Economic
Growth. Journal of Economic Growth 1, 189-211.
Baker, S., Bloom, N., Davis, S., 2016. Measuring Economic Policy Uncertainty. The Quarterly
Journal of Economics 131, 1593-1636
Battalio, R., Schultz, P., 2011. Regulatory Uncertainty and Market Liquidity: The 2008 Short Sale
Ban’s Impact on Equity Option Markets. Journal of Finance 66, 2013-2053
Bernanke, B., 1980. Irreversibility, Uncertainty, and Cyclical Investment. NBER Working
Papers 0502
Bloom, N., 2009. The Impact of Uncertainty Shocks. Econometrica 77, 623-685.
Brogaard J., Detzel, A., 2015. The Asset-Pricing Implications of Government Economic Policy
Uncertainty. Management Science 61, 3-18
Caballero, R., 1991. Earnings Uncertainty and Aggregate Wealth Accumulation. American
Economic Review 81, 859-871
Cao, C., Li, X. and Liu, G., 2019. Political uncertainty and cross-border acquisitions. Review of
Finance, 23, 439-470.
Chandrashekaran, V., 1999. Time-Series Properties and Diversification Benefits of REIT Returns.
Journal of Housing Research 17, 91-112.
https://econpapers.repec.org/article/aeaaecrev/v_3a73_3ay_3a1983_3ai_3a1_3ap_3a228-33.htmhttps://ideas.repec.org/a/eee/poleco/v29y2013icp151-167.htmlhttps://ideas.repec.org/s/eee/poleco.htmlhttps://ideas.repec.org/s/eee/poleco.htmlhttps://www.nber.org/people/alberto_alesinahttps://www.nber.org/people/sule_ozlerhttps://www.nber.org/people/nouriel_roubinihttps://www.nber.org/people/phillip_swagelhttps://ideas.repec.org/a/kap/jecgro/v1y1996i2p189-211.htmlhttps://ideas.repec.org/a/kap/jecgro/v1y1996i2p189-211.htmlhttps://ideas.repec.org/s/kap/jecgro.htmlhttps://econpapers.repec.org/article/oupqjecon/v_3a131_3ay_3a2016_3ai_3a4_3ap_3a1593-1636..htmhttps://econpapers.repec.org/article/blajfinan/v_3a66_3ay_3a2011_3ai_3a6_3ap_3a2013-2053.htmhttps://econpapers.repec.org/article/blajfinan/v_3a66_3ay_3a2011_3ai_3a6_3ap_3a2013-2053.htmhttps://ideas.repec.org/p/nbr/nberwo/0502.htmlhttps://ideas.repec.org/s/nbr/nberwo.htmlhttps://ideas.repec.org/s/nbr/nberwo.htmlhttps://ideas.repec.org/a/ecm/emetrp/v77y2009i3p623-685.htmlhttps://ideas.repec.org/s/ecm/emetrp.htmlhttps://econpapers.repec.org/article/inmormnsc/v_3a61_3ay_3a2015_3ai_3a1_3ap_3a3-18.htmhttps://econpapers.repec.org/article/inmormnsc/v_3a61_3ay_3a2015_3ai_3a1_3ap_3a3-18.htmhttps://econpapers.repec.org/article/aeaaecrev/v_3a81_3ay_3a1991_3ai_3a4_3ap_3a859-71.htm
26
Çolak, G., Durnev, A., Qian, Y., 2017. Political Uncertainty and IPO Activity: Evidence from U.S.
Gubernatorial Elections. Journal of Financial and Quantitative Analysis 52, 2523-2564
Cukierman, A., 1980. The Effects of Uncertainty on Investment under Risk Neutrality with
Endogenous Information. Journal of Political Economy 88, 462-75
Gao, P., Qi, Y., 2012. Political Uncertainty and Public Financing Costs: Evidence from U.S.
Municipal Bond Markets. Available at SSRN: https://ssrn.com/abstract=2024294
Handler, L., Jankowitsch, R., 2018. Political Uncertainty and Sovereign Bond Markets. Available
at SSRN: https://ssrn.com/abstract=3227790
Hartman, R., 1972. The effects of price and cost uncertainty on investment. Journal of Economic
Theory 5, 258-266
Hoesli, M., Lekander, L., Witkiewicz, W., 2004. International Evidence on Housing as a Portfolio
Diversifier. Journal of Housing Research 26, 161-206.
Ingersoll, J., Ross, S., 1992. Waiting to Invest: Investment and Uncertainty. The Journal of
Business 65, 1-29
Jens, C., 2017. Political uncertainty and investment: Causal evidence from U.S. gubernatorial
elections. Journal of Financial Economics, 124, 563-579.
Jong-A-Pin, R., 2009. On the measurement of political instability and its impact on economic
growth. European Journal of Political Economy 25, 15-29.
Julio B., Yook, Y., 2012. Political Uncertainty and Corporate Investment Cycles. Journal of
Finance 67, 45-84
https://econpapers.repec.org/article/cupjfinqa/v_3a52_3ay_3a2017_3ai_3a06_3ap_3a2523-2564_5f00.htmhttps://econpapers.repec.org/article/cupjfinqa/v_3a52_3ay_3a2017_3ai_3a06_3ap_3a2523-2564_5f00.htmhttps://econpapers.repec.org/article/ucpjpolec/v_3a88_3ay_3a1980_3ai_3a3_3ap_3a462-75.htmhttps://econpapers.repec.org/article/ucpjpolec/v_3a88_3ay_3a1980_3ai_3a3_3ap_3a462-75.htmhttps://ssrn.com/abstract=2024294https://ssrn.com/abstract=3227790https://econpapers.repec.org/article/ucpjnlbus/v_3a65_3ay_3a1992_3ai_3a1_3ap_3a1-29.htmhttps://econpapers.repec.org/article/eeejfinec/v_3a124_3ay_3a2017_3ai_3a3_3ap_3a563-579.htmhttps://econpapers.repec.org/article/eeejfinec/v_3a124_3ay_3a2017_3ai_3a3_3ap_3a563-579.htmhttps://ideas.repec.org/a/eee/poleco/v25y2009i1p15-29.htmlhttps://ideas.repec.org/a/eee/poleco/v25y2009i1p15-29.htmlhttps://ideas.repec.org/s/eee/poleco.htmlhttps://econpapers.repec.org/article/blajfinan/v_3a67_3ay_3a2012_3ai_3a1_3ap_3a45-84.htm
27
Kallberg, J., Liu, C., Greig, D., 1996. The role of housing in the portfolio allocation
process. Housing Economics 24, 359-377.
La Porta, R., Lopez- de-Silane, R., Shleiferand, A., Vishny, R., 1997. Legal Determinants of
External Finance. NBER Working Papers 5879
Lash, N. A., 2004. Corruption and economic development. Journal of Economic Asymmetries 1,
85-105
Leff, N. H., 1964. Economic Development through Bureaucratic Corruption. American Behavioral
Scientist 8 , 8–14.
Liu, L., Shu, H., Wei, J., 2017. The impacts of political uncertainty on asset prices: Evidence from
the Bo scandal in China. Journal of Financial Economics 125, 286-310.
Mauro, P., 1995. Corruption and Growth. The Quarterly Journal of Economics 110, 681-712.
Mo, P., 2001. Corruption and Economic Growth. Journal of Comparative Economics 29, 66-79.
Pastor, L., Veronesi, P., 2012. Uncertainty about Government Policy and Stock Prices. Journal of
Finance 67 , 1219-1264.
Pastor, L., Veronesi, P., 2013. Political Uncertainty and Risk Premia. Journal of Financial
Economics 110, 520-545.
Pindyck, R., 1988. Irreversible Investment, Capacity Choice, and the Value of the Firm. American
Economic Review 78, 969-85.
https://econpapers.repec.org/paper/nbrnberwo/5879.htmhttps://econpapers.repec.org/paper/nbrnberwo/5879.htmhttps://econpapers.repec.org/article/eeejfinec/v_3a125_3ay_3a2017_3ai_3a2_3ap_3a286-310.htmhttps://econpapers.repec.org/article/eeejfinec/v_3a125_3ay_3a2017_3ai_3a2_3ap_3a286-310.htmhttps://econpapers.repec.org/article/oupqjecon/v_3a110_3ay_3a1995_3ai_3a3_3ap_3a681-712..htmhttps://econpapers.repec.org/article/eeejcecon/v_3a29_3ay_3a2001_3ai_3a1_3ap_3a66-79.htmhttps://econpapers.repec.org/article/blajfinan/v_3a67_3ay_3a2012_3ai_3a4_3ap_3a1219-1264.htmhttps://econpapers.repec.org/article/aeaaecrev/v_3a78_3ay_3a1988_3ai_3a5_3ap_3a969-85.htm
28
Appendix. Variable definition
A. Housing market data
Housing data includes annual data on house price index, number of transactions and number
of building permits at state-level. The variables related to housing are the following:
- House price index is published by the Federal Housing Finance Agency (FHFA). The HPIi,t
is a weighted, repeat-sales index that measures average price changes in repeat sales or
refinancing on the same properties in state i in year t.
- RHi,t is the annual return or change in the HPI in state i in year t, and is calculated as follows
𝑅𝑖,𝑡𝐻 =
𝐻𝑃𝐼𝑖,𝑡 − 𝐻𝑃𝐼𝑖,𝑡−1𝐻𝑃𝐼𝑖,𝑡−1
- Transactioni,t is the number of housing transactions in state i in year t. The data is from
CoreLogic.
- Permiti,t is the number of building permits in state i in year t. The data is published by the
U.S. Census Bureau (census.gov)
B. Gubernatorial election data
The gubernatorial election data in our sample cover all gubernatorial elections across 50 states
in the U.S. in the period from 1982 to 2015. We collect most of the data from CQ Electronic
Library. There are 47 states that have 4-year election cycle; 3 states with 2-year election cycle
(New Hampshire, Vermont, Rhode Island adopted to 4-year election cycle in 1994). In general,
gubernatorial elections are held on the first Tuesday in November, with the earliest date being
the 2nd of November and the latest date being the 8th of November. The variables related to
gubernatorial elections are the following:
29
- Electioni,t is a yearly binary variable that is equal to 1 if there is a gubernatorial election in
state i in year t, and 0 otherwise.
- Pres_electiont is a yearly binary variable that takes value of 1 if there is a presidential
election in the U.S. in year t.
- Winning_margini,t is the difference between the percentage votes won by the first and
second place candidates in a gubernatorial election in state i in year t.
- Election_closenessi,t is a binary variable equal to one if the winning_margini,t is in the
lowest tercile (5.1%). In section 6, we divide our sample into two subsamples:
High_Uncertainty and Low_Uncertainty based on Election_closenessi,t :
Election_closenessi,t is equal to 1 for the High_Uncertainty subsample, and 0 for the
Low_Uncertainty subsample.
- Incumbent_absencei,t is a hand-collected binary variable equal to one if the incumbent
governor did not take part in a gubernatorial election in state i in year t either due to term
limit regulation or other personal reasons.
C. Mortgage data
Mortgage data are from Federal Financial Institutions Examination Council (FFIEC). Data are
made available publicly pursuant to the Home Mortgage Disclosure Act of 1975 (HMDA).
The HMDA requires most mortgage lenders to disclose information at the application level
about their mortgage lending activities every year. We aggregate mortgage data at the state
level and create three variables: log(Volume); Approval_rate; and Application
- log(Volume)i,t is the logarithm of the volume of mortgages originated in state i in year t,
with volume reported in thousands of dollars ; log(Volume)b,i,t is the logarithm of the
30
volume of mortgages originated by bank b headquartered in state i in year t, with volume
reported in thousands of dollars.
- Approval_ratei,t is the number of mortgage applications approved divided by the total
number of mortgage applications in state i in year t; Approval_rateb,i,t is the number of
mortgage applications approved divided by the total number of mortgage applications in
bank b headquarter in state i in year t.
Approval_ratei, t =𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑝𝑝𝑟𝑜𝑣𝑒𝑑 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒𝑠𝑖,𝑡
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑖,𝑡
Approval_rateb,i,t =𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑝𝑝𝑟𝑜𝑣𝑒𝑑 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒𝑠𝑏,𝑖,𝑡
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑏,𝑖,𝑡
- Applicationi,t is the relative change of the number of mortgage applications in state i in
year t; Applicationb,i,t is the relative change of the number of mortgage applications in bank
b headquartered in state i in year t.
Applicationi,t = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠 𝑖,𝑡− 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑖,𝑡−1
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑖,𝑡−1
Applicationb,i,t = 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠 𝑏,𝑖,𝑡− 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑏𝑖,𝑡−1
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑜𝑟𝑡𝑔𝑎𝑔𝑒 𝑎𝑝𝑝𝑙𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠𝑏,𝑖,𝑡−1
D. Call reports data
Commercial Banks’ financial statement data are extracted from Bank Regulatory data section
in Wharton Research Data Services. Commercial banks are required to report either
“Consolidated Reports of Condition and Income” FFIEC 031 for banks with Domestic and
Foreign Offices or “Consolidated Reports of Condition and Income” FFIEC 041 for banks with
Domestic Offices only. These reports cover detailed financial statements of commercial banks
in the U.S. as well as their headquarter states, which enables us perform analyses on the impact
31
of gubernatorial elections. All the variables described below are for bank b in state i in year t,
and we follow Acharya and Mora (2012) to derive these variables from Call reports.
- Sizeb,i,t is the logarithm of the total assets (item RCFD2170 in call reports), with total assets
reported in thousands of dollars.
- Core_depositb,i,t (implicit) is the sum of transaction deposits, savings and small time
deposits that form a stable source of funds divided by total assets. We take the quarterly
avergage of core deposits: RCON3485 + RCONB563 + RCON3469 (before 1997Q1)/
RCONA529 (after 1997Q1).
- Deposit_costb,i,t (implicit) is the interest expense on core deposits divided by the quarterly
average of core deposits. The interest expense on core deposits: RIAD4508 + RIAD0093
(RIAD4509 + RIAD4511 before 2001Q1) + RIADA518 (RIAD4512 before 1997Q1).
- Return_on_equityb,i,t is the net income divided by the total equity (RIAD4340 divided by
RCFD3210)
- Liquid_assetsb,i,t is the sum of liquid assets divided by total assets. Liquid assets are cash,
federal funds sold, reverse repos, and securities excluding mortgage-backed and asset-
backed securities. Cash: RCFD0010; federal funds sold: RCFD1350 (before 2002Q1) +
RCONB987 + RCFDB989 (from 2002Q1); securities excluding mortgage-backed and
asset-backed securities before 2009Q2: (RCFD1745 + RCFD1773) - (RCFD8500
+RCFD8504 + RCFDC026 + RCFD8503 + RCFD8507+ RCFDC027); after 2009Q2:
(RCFD1754+ RCFD1733) – (RCFDG300 + RCFDG303+ RCFDG304 + RCFDG307 +
RCFDG308 + RCFDG311 + RCFDG312 + RCFAG315+ RCFDG316 + RCFDG319+
RCFDG320 + RCFDG323+ RCFDG324 + RCFDG327 +RCFDG328 + RCFDG331+
RCFDG336 + RCFDG339 + RCFDG340 + RCFDG343 + RCFDG344 + RCFDG347 +
RCFDC026 + RCFDC027)
32
- Home_mortgageb,i,t is the volume of residential mortgage loans divided by total loans.
Residential mortgage loans: RCON1797 + RCON5367 + RCON5368. Total loans:
RCFD2122.
E. Other macroeconomics variables
Macroeconomic data and financial market data are retrieved from various sources including
the Bureau of Economic Analysis (bea.gov), the Bureau of Labor Statistics (bls.gov), the U.S
Census Bureau (census.gov), the Federal Emergency Management Agency (fema.gov), and
FRED (fred.stlouisfed.org). The list of control variables goes as follows:
- GDPi,t is the gross domestic product’s growth rate in state i in year t.
- Unemploymenti,t is the unemployment rate in state i in year t.
- Populationi,t is the annual relative change in the population in state i in year t.
- Emergencyi,t is a binary variable that takes value of 1 if state i in year t declares emergency
in a given year, and 0 other wise.
- Recessiont is a binary variable that takes value of 1 if year t is identified as being in a
recession state by the National Bureau of Economic Research (NBER).
- S&P500t is relative return of the S&P500 index in year t.
- Bond_ratet is the 10-year interest rate on the U.S. government bonds in year t.
33
Tables and Figures
Figure 1: Policy uncertainty and housing markets
Figure 1 exhibits the dynamics of the U.S. Economic Policy Uncertainty Index (left axis scale) and the annual growth in U.S. house
prices (right axis scale) in Panel A and the housing transactions per year (right axis scale) in Panel B from January 1991 to April
2020. The U.S. Economic Policy Uncertainty Index has been developed and constructed in Baker, Bloom and Davis (2016). The
annual growth in house prices is computed using the seasonally adjusted Federal Housing Finance Agency (FHFA) monthly house
price index for the U.S. The data on housing transactions comes from the U.S. National Association of Realtors.
Panel A. Policy uncertainty and growth in house prices
Panel B. Policy uncertainty and housing transactions
-10,0%
-5,0%
0,0%
5,0%
10,0%
0
50
100
150
200
250
300
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
20
09
20
10
20
11
20
12
20
13
20
14
20
15
20
16
20
17
20
18
20
19
20
20
Ann
ual
grow
th i
n ho
use
pri
ces
(%)
Polic
y un
cert
ainty
in
dex
(10
0 =
yea
r 20
10
)
Policy uncertainty index Growth in house prices
0
1.000
2.000
3.000
4.000
5.000
6.000
7.000
8.000
0
50
100
150
200
250
300
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
Hous
ing
tran
sact
ions
per
yea
r
Polic
y un
cert
ainty
in
dex
(100 =
yea
r 2010)
Policy uncertainty index Housing transactions
Stimulus
debate
2nd Gulf
War
9/11
Covid-19
Lehman and
TARP Eurozone
crisis
Debt ceiling
debate
Gov’t
shutdown Brexit
Trump
election
U.S.-China
trade war
Fiscal
cliff
Stimulus
debate
2nd Gulf
War
9/11
Covid-19
Lehman
and
TARP Eurozone
crisis
Debt ceiling
debate
Gov’t
shutdown
Brexit
Trump
election
U.S.-China
trade war Fiscal
cliff
34
Table 1: Summary statistics
Table 1 provide the summary statistics for the variables we use in our study. All variables are in annual term. RH is the return of
the FHFA house price index at state level; Transaction is the number of housing transactions at state level; Permit is the number
of building permits issued at state level; Election is a dummy variable that equals to 1 if a state holds a gubernatorial election and
0 otherwise; Pres_election is a dummy variable that equals to 1 if there is a presidential election and 0 otherwise;
Incumbent_absence is a dummy variable that equals to 1 if the incumbent is absent from the election due to term limit or other
personal reasons; Winning_margin is the difference between the percentage votes won by the first and second place candidates in
a gubernatorial election; Election_closeness is a dummy variable equal to 1 if the Winning_margin is in the lowest tercile (5.1%)
and 0 otherwise; GDP is the growth rate in gross domestic product at state level; Unemployment is the unemployment rate at state
level; Population is the growth rate in the number of residents at state level; Emergency is the dummy variable that equals 1 if a
state declares it is under an emergency situation and 0 otherwise; Recession is the dummy variable that equals 1 if a year is identified
as being in a recession; S&P500 is the relative return of the S&P500 index; and Bond_rate is the 10-year interest rate on the U.S.
government bonds. The Appendix shows a detailed description of the variables.
Whole sample
Housing variables Min Mean Max Std. deviation Num. of obs.
RH -0.30 0.04 0.53 0.06 1,700
Transaction 0 49,690 980,433 110,544 1,700
Permit 555 26,386 314,641 36,425 1,700
Political variables Min Mean Max Std. deviation Num. of obs.
Election 0.00 0.27 1.00 0.44 1,700
Pres_election 0.00 0.24 1.00 0.42 1,700
Incumbent_absence 0.00 0.40 1.00 0.49 458
Winning_margin 0.00 0.16 0.65 0.14 458
Election_closeness 1.00 0.34 0.00 0.47 458
Economic variables Min Mean Max Std. deviation Num. of obs.
GDP -0.26 0.05 0.25 0.037 1,700
Unemployment 0.02 0.06 0.18 0.02 1,700
Population -0.06 0.01 0.12 0.01 1,700
Emergency 0.00 0.30 1.00 0.46 1,700
Recession 0.00 0.06 1.00 0.24 1,700
S&P500 -0.37 0.13 0.37 0.16 1,700
Bond_rate -0.11 0.09 0.33 0.10 1,700
35
Table 1: Summary statistics (cont.)
Sub-sample with gubernatorial elections only
Housing variables Min Mean Max Std. deviation Num. of obs.
RH -0.15 0.03 0.53 0.07 458
Transaction 0 46,508 780,870 106,887 458
Permit 692 27,394 314,641 35,892 458
Sub-sample with close elections only (elections with winning margin is less than 5.1%)
Housing variables Min Mean Max Std. deviation Num. of obs.
RH -0.15 0.03 0.17 0.07 154
Transaction 0 45,715 780,870 113,913 154
Permit 789 27,662 203,238 33,849 154
36
Table 2: House price returns, transactions, and building permits in election years
Table 2 reports the results of the baseline OLS estimates of the specifications in equations (1a), (1b), and (1c) that examine the
impact of gubernatorial election on the return on house price index RH (Panel A), the number of housing transaction (Panel B), and
the number of building permit (Panel C). All the specifications include state and year fixed effects and the standard errors are
clustered at the state and year levels. *** significant at 1% level. ** significant at 5% level. * significant at 10% level. Sample
period: 1982-2015.
Panel A. Return on house price index (RH) and gubernatorial election: Baseline OLS
RH
(1) (2) (3) (4)
Election -0.005***
(-2.68)
-0.006***
(-3.40)
-0.006***
(-3.01)
-0.006***
(-2.80)
GDP (lagged) 0.625***
(10.10)
0.644***
(9.24)
0.324***
(3.67)
0.318***
(2.90)
SP500 (lagged) 0.369***
(5.70)
0.341***
(6.36)
0.337***
(4.86)
0.384***
(5.66)
Unemployment (lagged) -0.209**
(-2.21)
-0.216***
(3.23)
-0.327***
(3.48)
Bond_rate (lagged) -0.110***
(-7.67)
-0.085***
(-5.51)
-0.059***
(-3.73)
Emergency -0.001
(-0.36)
Recession -0.025***
(-4.30)
Population(lagged) 0.873***
(4.75)
Pres_election 0.002
(0.98)
R2 0.1487 0.1811 0.4539 0.4641
Num. of observations 1700 1700 1700 1700
37
Table 2: House price returns, transactions, and building permits in election years. OLS (cont.)
Panel B. Number of housing transactions (Transaction) and gubernatorial elections: Baseline OLS
Transaction
(1) (2) (3) (4)
Election -5547.575***
(-3.41)
-3726.202***
(-2.63)
-2909.319***
(-2.58)
-3218.847***
(-2.95)
GDP (lagged) 4371.155***
(-3.67)
3872.446***
(-3.40)
473.367***
(-2.90)
444.973***
(-2.69)
SP500 (lagged) 9844.930***
(-5.13)
10070.700***
(-5.21)
15872***
(-3.22)
6355.766*
(-1.81)
Unemployment (lagged) -1586.937*
(-1.69)
-1073.862***
(-3.68)
-1688.747***
(-4.18)
Bond_rate (lagged) -1056.700***
(-5.71)
-1591.724
(-0.52)
1997.908
(0.63)
Emergency 1162.844
(0.84)
Recession -14115.840***
(-4.81)
Population (lagged) -89939.100
(-0.27)
Pres_election 7536.735***
(4.71)
R2 0.040 0.047 0.931 0.933
Num. of observations 1700 1700 1700 1700
38
Table 2: House price returns, transactions, and building permits in election years. OLS (cont.)
Panel C. Number of building permits (Permit) and gubernatorial elections: Baseline OLS
Permit
(1) (2) (3) (4)
Election 1058.299**
(2.91)
1201.012***
(3.21)
813.855***
(2.80)
583.425**
(2.15)
GDP (lagged) 1235.050**
(3.47)
958.596***
(2.94)
388.769***
(3.61)
328.513***
(3.24)
SP500 (lagged) -2191.324*
(-1.73)
2464.412*
(1.78)
2632.739***
(4.00)
3306.844***
(4.35)
Unemployment (lagged) -2054.970***
(-5.57)
-706.088***
(3.80)
-714.569***
(3.94)
Bond_rate (lagged) -4683.753
(-1.52)
-8699.603***
(-3.60)
-4939.527**
(-2.54)
Emergency 694.743**
(2.17)
Recession -4335.058***
(-4.01)
Population (lagged) 99508.320***
(3.58)
Pres_election 71.058
(0.4)
R2 0.081 0.132 0.780 0.787
Num. of observations 1700 1700 1700 1700
39
Table 3: U.S. states and their neighboring state
State
Num. NS1 NS2 NS3 NS4 NS5 NS6 NS7 NS8
Alaska AK 0
Alabama AL 4 TN GA FL MS
Arkansas AR 6 MO TN MS LA TX OK
Arizona AZ 5 UT CO NM CA NV
California CA 3 OR NV AZ
Colorado CO 7 WY NE KS OK NM AZ UT
Connecticut CT 3 MA RI NY
Delaware DE 3 PA NJ MD
Florida FL 2 GA AL
Georgia GA 5 NC SC FL AL TN
Hawaii HI 0
Iowa IA 6 MN WI IL MO NE SD
Idaho ID 6 MT WY UT NV OR WA
Illinois IL 5 WI IN KY MO IA
Indiana IN 4 MI OH KY IL
Kansas KS 4 NE MO OK CO
Kentucky KY 7 OH WV VA TN MO IL IN
Louisiana LA 3 AR MS TX
Massachusetts MA 5 NH RI CT NY VT
Maryland MD 4 PA DE VA WV
Maine ME 1 NH
Michigan MI 3 OH IN WI
Minnesota MN 4 WI IA SD ND
Missouri MO 8 IA IL KY TN AR OK KS NE
Mississippi MS 4 TN AL LA AR
Montana MT 4 ND SD WY ID
North Carolina NC 4 VA SC GA TN
North Dakota ND 3 MN SD MT
40
Table 3: U.S. states and their neighboring state (cont.)
State Num. NS1 NS2 NS3 NS4 NS5 NS6 NS7 NS8
Nebraska NE 6 SD IA MO KS CO WY
New Hampshire NH 3 ME MA VT
New Jersey NJ 4 NY CT DE PA
New Mexico NM 5 CO OK TX AZ UT
Nevada NV 5 ID UT AZ CA OR
New York NY 5 VT MA CT NJ PA
Ohio OH 5 PA WV KY IN MI
Oklahoma OK 6 KS MO AR TX NM CO
Oregon OR 4 WA ID NV CA
Pennsylvania PA 6 NY NJ DE MD WV OH
Rhode Island RI 2 MA CT
South Carolina SC 2 NC GA
South Dakota SD 6 ND MN IA NE WY MT
Tennessee TN 8 KY VA NC GA AL MS AR MO
Texas TX 4 OK AR LA NM
Utah UT 6 ID WY CO NM AZ NV
Virginia VA 5 MD NC TN KY WV
Vermont VT 3 NH MA NY
Washington WA 2 ID OR
Wisconsin WI 4 MI IL IA MN
West Virginia WV 5 PA MD VA KY OH
Wyoming WY 6 MT SD NE CO UT ID
41
Table 4: House price returns, transactions, and building permits in election years
Difference-in-Differences neighboring state methodology
Table 3 reports the results from the difference-in-difference neighboring state methodology with the specifications in equations
(5a), (5b), and (5c) that examine the impact of gubernatorial election on the differences in return on house price index ∆RH (Panel
A), the number of housing transaction ∆Transaction (Panel B), and the number of building permit ∆Permit (Panel C) between a
state and its neighboring state. The standard errors are clustered by pairs of state-neighboring state and year in all the specifications.
*** significant at 1% level. ** significant at 5% level. * significant at 10% level. Sample period: 1982-2015.
Panel A. Return on house price index (RH) and gubernatorial election: Difference-in-differences
RH
(1) (2) (3) (4)
∆Election -0.002***
(-32.62)
-0.002***
(-19.58)
-0.001***
(-18.06)
-0.001***
(-17.08)
∆GDP (lagged) 0.124***
(2.72)
0.157
(1.19)
0.183
(1.65)
∆Unemployment (lagged) -0.166
(0.87)
-0.103
(0.69)
∆Populatiton (lagged) 1.160***
(4.7)
1.179***
(4.79)
∆Emergency 0.001
(0.53)
0.001
(0.70)
R2 0.324 0.340 0.337 0.335
Num. of observations 3575 3575 3575 3575
Note: Column (4) reports the weighted least square regression results, with weight
being the number of neighboring states.
42
Table 4: House price returns, transactions, and building permits in election years
Difference-in-Differences neighboring state methodology (cont.)
Panel B. Number of housing transactions (Transaction) and gubernatorial election: Difference-in-differences
Transaction
(1) (2) (3) (4)
∆Election -995.528***
(-53.47)
-1010.03***
(54.18)
-1029.329***
(-54.77)
-486.988***
(-14.97)
∆GDP (lagged) 391.475**
(-1.99)
435.226*
(-1.79)
422.988*
(-1.97)
∆Unemployment (lagged) -721.264
(1.22)
-690.718
(1.15)
∆Populatiton (lagged) 1583.364
(1.45)
1794.482*
(1.72)
∆Emergency 967.277
(0.94)
1704.148*
(1.85)
R2 0.881 0.882 0.883 0.884
Num. of observations 3575 3575 3575 3575
Note: Column (4) reports the weighted least square regression results, with weight being the
number of neighboring states.
43
Table 4: House price returns, transactions, and building permits in election years
Difference-in-Differences neighboring state method (cont.)
Panel C. Number of building permits (Permit) and gubernatorial election: Difference-in-differences
Permit
(1) (2) (3) (4)
∆Election 190.954***
(3.26)
193.056***
(3.35)
183.044***
(3.10)
232.871***
(6.53)
∆GDP (lagged) 160.544**
(2.58)
72.521
(1.20)
98.485*
(1.68)
∆Unemployment (lagged) -399.387
(1.56)
-402.084*
(1.69)
∆Populatiton (lagged) 1644.365***
(4.56)
1862.433***
(4.33)
∆Emergency 473.017
(1.2)
464.346
(1.35)
R2 0.749 0.751 0.758 0.758
Num. of observations 3575 3575 3575 3575
Note: Column (4) reports the weighted least square regression results, with weight being the
number of neighboring states.
44
Figure 2: Housing markets over the gubernatorial election cycle
Figure 2 plots the averaged-across state return on house price index RH (Panel A), averaged-across state number of housing
transactions Transaction (Panel B), and averaged-across state number of building permit Permit (Panel C) over the gubernatorial
election cycle during the sample period from 1982-2015. There are 458 gubernatorial elections in 50 states in which 47 states with
4-year election cycle and 3 states (NH, RI, VT) with of 2-year election cycle. In these graphs, we only include states with 4-year
election cycle. Sample period: 1982-2015.
Panel A. Average RH and Gubernatorial election cycle (Election year T=0)
Panel B. Average number of transactions and Gubernatorial election cycle (Election year T=0)
Panel C. Average number of building permits and Gubernatorial election cycle (Election year T=0)
3.35%
3.40%
3.45%
3.50%
3.55%
3.60%
3.65%
3.70%
3.75%
T=-1 T=0 T=1 T=2
Av
erag
e R
H
Gubernatorial election cycle
0
10,000
20,000
30,000
40,000
50,000
60,000
T=-1 T=0 T=1 T=2
Av
erag
e T
ran
sact
ion
s
Gubernatorial election cycle
0
5,000
10,000
15,000
20,000
25,000
30,000
35,000
T=-1 T=0 T=1 T=2
Av
erag
e P
erm
its
Gubernatorial election cycle
45
Table 5: Post-election analysis of house price returns, transactions, and building permits
Table 5 presents the results of six OLS regressions (6a), (6b), (6c), (6d), (6e), and (6f) to examine the post-election reversal
effect on the return on house price index RH (columns 1 and 2); the number of housing transactions Transaction (columns 3 and
4); and the number of building permits Permit (columns 5 and 6). All the scpeficications include state and year fixed effects
and the standard errors are clustered at the state and year level. *** significant at 1% level. ** significant at 5% level. *
significant at 10% level. Sample period: 1982-2015.
Variables RH Transaction Permit
(1) (2) (3) (4) (5) (6)
Post_election_1 0.004***
(-2.97)
0.005***
(-3.46)
734.263***
(-2.98)
914.872***
(-3.32)
-334.182***
(3.03)
-417.631**
(2.18)
Post_election_2 0.003*
(1.80)
1031.283**
(2.06)
-287.319**
(1.99)
GDP (Lagged) 0.302***
(3.90)
0.314***
(4.62)
423.617***
(-3.01)
444.973***
(-2.69)
334.471***
(3.73)
421.491***
(4.21)
SP500 (Lagged) 0.043***
(6.47)
0.055***
(5.89)
-5541.029*
(-2.18)
-6355.766*
(-1.81)
3218.438***
(3.32)
3834.591***
(4.99)
Unemployment (Lagged) -0.228***
(3.96)
-0.234***
(4.03)
-1594.387***
(-4.24)
-1688.747***
(-4.57)
-721.041***
(3.12)
-681.381***
(4.03)
Bond_rate (Lagged) -0.067***