the effect of ltv-based risk weights on house prices ... · ltv-based risk weights limit the ltv...
TRANSCRIPT
The Effect of LTV-Based Risk Weights on House Prices:
Evidence From an Israeli Macroprudential Policy
Nitzan Tzur-Ilan, Northwestern University and Bank of Israel
Steven Laufer, Federal Reserve Board
MAY 29, 2020
2020 AREUEA National Meeting
DISCLAIMER: The views expressed here are those of the authors and do not necessarily reflect the views of the Bank of Israel or the Board of
Govenrnors.
Motivation
• House prices and residential mortgages play central roles in the credit cycle that
sparked the Global Financial Crisis.
• As a result, many of the macroprudential policies imposed in the wake of the crisis
have specifically focused on banks’ provision of mortgage credit.
• Those policies serve two main purposes (Krznar and Morsink (2014); Lim et al.
(2013)):
1. discourage banks from originating riskier mortgages which reduce bank losses
during economic downturns.
2. Limiting the build up of financial imbalances by moderating the growth in
house prices.
Introduction 1
Motivation - Cont.
• A large literature has found that that an easing of mortgage credit leads to
stronger house price growth (e.g. Mian and Sufi (2009); Favara and Imbs
(2015); Di Maggio and Kermani (2017)).
• Therefore, one can expect that MPPs that limit mortgage credit would
affect the growth rate of house prices.
• However - open question in the literature regarding the effect of
macroprudential policy on house prices.
Introduction 2
LTV Limit - Affecting Housing Prices?
• This question received considerable attention in the literature,
but with mixed conclusions.
• Some studies do find that LTV limits reduce house price growth
(e.g. Igan and Kang (2011); Galati and Moessner (2013);
Akinci and Olmstead-Rumsey (2018)).
• Others fail to find any such effects.(e.g. Wong et al. (2011);
Kuttner and Shim (2016); Cerutti et al. (2017)).
Introduction 3
Literature - Identification Challenges
• Implementation of MPPs is highly endogenous to housing prices.
• These policies are typically used in combination with other
policies - challenge to attribute outcomes to specific tools.
• Challenges in controlling for country characteristics, quality of
MPP supervision, use and intensity of MPP and the phases of
the financial cycle.
• Availability of data.
Introduction 4
Goals of this paper
1. A cleaner identification of the effect of LTV caps on house
prices by studying policy that only affects part of the market.
2. The heterogeneity effect: which type of areas may be more
affected by these policies.
3. Generates an estimate of the semi-elasticity of housing prices
with respect to mortgage rates.
Introduction 5
MPP Measures Implemented
MPPs Date Type of MPP
MPP1 Oct-10 Increase capital provision for high-LTV-ratio loans
MPP2 May-11 Limit share of adjustable interest rate loans
MPP3 Nov-12 Limit LTV to 75% for FTHB, 50% for investors
MPP4 Feb-13 Raise risk weights for capital adequacy requirements
MPP5 Aug-13 PTI limited to 50% of HH income
Limit variable interest share of the loan to two-thirds
Limit loan period to 30 years
MPP6 Sep-14 Additional Tier One capital requirement
Introduction 6
The Housing Market and MPPs in Israel
The Rate of Change in Housing Prices in Israel, 01/2007-12/2015:
Introduction 7
Data and Identification
Data
• Property-level data from the Israel Tax Authority on the
universe of household purchases of residential properties.
• Detailed information on each property: date, location, price,
size and building year.
• Our analysis focuses on the period between Jan 2010 to May
2011 (90K obs.).
Data and Identification 8
LTV-based risk weights limit
• October 2010: risk-weight factor was raised from 35 to 100
percent for mortgages with:
1. An LTV of at least 60 percent.
2. A mortgage value higher than NIS 800,000 (USD 220,000).
Data and Identification 9
LTV-based risk weights limit
• The LTV limit required banks to set aside more capital against
risky loans.
• Regulation increased interest rates on high-LTV loans by
0.31-0.36 PP (Tzur-Ilan, 2017).
• LTV increases the yearly interest rate payments, on average, by
2,700-3,250 NIS (4% of average household gross yearly income).
• Thus, although the policy is statutorily imposed on lenders, it
appears as if a large portion of the economic burden ends up
being born by borrowers in the form of higher interest rate
(DeFusco et al., 2020).Data and Identification 10
Change in LTV Distribution
Incentivize risky borrowers (LTV>60%) to reduce leverage:
Less credit for the purchase of a housing unit.
Data and Identification 11
Identification Strategy
• Because this MPP only applied to mortgages over a certain
size, we can measure its effect on house prices by comparing
price growth in different segments of the Israeli housing market.
• Only for housing units above a certain price would a mortgage
with a given LTV ratio be larger than the 800K threshold.
• Assume that buyers always use an 75% LTV. Then only for
units with transaction prices above NIS 1.06M would the
mortgage be larger than the 800K threshold.
Data and Identification 12
Identification Strategy - cont.
• Use a Diff-in-Diffs approach to compare units with prices above
and below this NIS 1.06M threshold, before and after the policy.
• Similar to Adelino et al. (2012), that study the effect on house
prices in the US caused by the ability of the GSE to purchase
mortgages below a certain size.
• In that context, the authors argue that one can safely assume
that the marginal buyer will use an 80 percent LTV loan.
Data and Identification 13
Identification Strategy - cont.
• This paper’s setting is more complicated, as the Israeli housing
market is not dominated by a single LTV ratio.
• Construct a more general treatment measure: uses the observed
distribution of LTV, capture the likelihood that a particular unit
would be purchased using a mortgage affected by the policy,
given the transaction price.
• Then perform the Diff-in-Diffs estimation.
Data and Identification 14
Identification Strategy: Different Effects at DifferentPrice Ranges
Distribution of LTV Ratios Before and After LTV Limit, by Sale
Price:
• As we consider transactions at higher prices, a wider range of LTV ratios would
place the purchase mortgage above the NIS 800,000 threshold.
Data and Identification 15
Construction of the Treatment Effect
• Treatment: probability that the unit would be purchased with a
mortgage above NIS 800K and an LTV 60%.
• For a transaction at price p:
Treat(p) =1
∑LTV=0.6
I (p ∗ LTV > NIS800, 000) ∗ f (LTV ), (1)
• p*LTV - Mortgage Size
• f(LTV) - fraction of units purchased in the previous year using a mortgage with
that LTV ratio.
Data and Identification 16
Construction of the Treatment Effect
• Using the observed LTV distribution before the policy:
Data and Identification 17
Graphical Illustration of the Treatment Measure
Data and Identification 18
Empirical Methodology
• Diff in Diff: Compare purchases before and after the policy,
between more and less treated apartments.
• We estimate the following hedonic equation.
• For a transaction at price p:
ln(PPSMit) = α+ β̂Xi +Areai +Γ ∗ θt + δ ∗Treat(p)+σ ∗Treat(p) ∗ θt + εit(2)
where ln(PPSMit) - log price per square meter for unit i sold at time t. X includes
number of rooms and log age of the building. θt - time dummy equal to zero before the
policy was implemented and one afterwards. εit - well-behaved error term clustered at the
locality statistical area level. Our primary interest is in the coefficient σ.
Data and Identification 19
Results
The Estimated Effect of LTV limit on Housing PricesPPSM PPSM PRICE
(1) (2) (3)
3.roomsgroup -0.183*** -0.101*** 0.233***
(0.00598) (0.00603) (0.00660)
4.rooms group -0.345*** -0.179*** 0.441***
(0.00775) (0.00815) (0.00930)
5.rooms group -0.490*** -0.241*** 0.570***
(0.00851) (0.00927) (0.0107)
lnage 0.00371*** -0.00346*** -0.0126***
(0.000810) (0.000887) (0.00111)
Treatment 0.156*** 0.744*** 1.010***
(0.00623) (0.0175) (0.0176)
After 0.0812*** 0.0998*** 0.0940***
(0.00360) (0.00362) (0.00380)
TreatmentAfter -0.0404*** -0.0309*** -0.0235***
(0.0108) (0.0119) (0.0092)
Geographic FE NO YES YES
Constant 2.113*** 2.319*** 5.517***
(0.00977) (0.0619) (0.183)
Observations 90,332 90,332 90,332
R-squared 0.891 0.902 0.919
Results 20
Alternative treatment variable
• Potential concern: price is both the outcome variable and the input used to compute the
treatment effect.
• An alternative model: compute a predicted price (p̂) for each unit based on its hedonic
characteristics:
ln(p̂i ) = α + β̂Xi +monthi + εi (3)
• Then, used this predicted price to compute a treatment effect:
treatment(p̂) =1
∑LTV=0.6
I (p̂ ∗ LTV > NIS800, 000) ∗ f (LTV ) (4)
• Then, use the Diff-in-Diff approach but instead of Treat(p) use Treat(p̂).
• When we generate our estimates we take the statistical variation embedded in our
approach into account through a bootstrap procedure.
Results 21
The Estimated Effect of LTV limit on Predicted Pricespredicted price
PPSM PRICE
(4) (5)
3.roomsgroup -0.0591*** 0.170***
(0.00699) (0.00950)
4.rooms group -0.0789*** 0.525***
(0.00913) (0.00967)
5.rooms group -0.0811*** 0.804***
(0.0157) (0.0120)
lnage -0.00741*** -0.0254***
(0.00469) (0.00672)
Treatment 0.744*** 1.012***
(0.0231) (0.0193)
After 0.0959*** 0.0846***
(0.00674) (0.00699)
TreatmentAfter -0.0212*** -0.0287***
(0.0231) (0.0090)
Geographic FE YES YES
Constant 2.324*** 6.288***
(0.0965) (0.0152)
Observations 90,332 90,332
R-squared 0.902 0.893Results 22
Pre-Trends Test: The Estimated Effect of LTV limit a Year Before the Policy
predicted price
PPSM PRICE PPSM PRICE
(1) (2) (3) (4)
3.roomsgroup -0.232*** 0.237*** -0.096*** 0.235***
(0.00538) (0.00729) (0.00654) (0.00767)
4.rooms group -0.414*** 0.448*** -0.171*** 0.449***
(0.00544) (0.01011) (0.00838) (0.01061)
5.rooms group -0.600*** 0.574*** -0.235*** 0.575***
(0.00613) (0.01177) (0.01010) (0.01257)
lnage -0.001*** -0.012*** -0.003*** -0.0118***
(0.00038) (0.00112) (0.00092) (0.00160)
Treatment 0.019 0.018 0.010 0.022
(0.01359) (0.02177) (0.02028) (0.03285)
After 0.021 0.011 0.028 0.012
(0.04068) (0.01157) (0.03506) (0.05201)
TreatmentAfter 0.015 0.006 -0.009 0.013
(0.02528) (0.02164) (0.01089) (0.01186)
Geographic FE YES YES YES YES
Geographic FEAfter YES YES YES YES
Constant 2.354*** 5.489*** 1.777*** 5.259***
(0.00554) (0.02453) (0.19597) (0.03679)
Observations 82,242 82,242 82,162 81,734
R-squared 0.811 0.897 0.855 0.898
Results 23
Placebo Tests: Treatments using other mortgage sizes
Results 24
Direct Observation of High-LTV Mortgages
• Instead of using the transaction price to estimate the probability that a buyer
would use a mortgage affected by the policy, use information about the
mortgage itself to conduct what might be considered the non-instrumented
OLS version of this exercise.
• Merge the housing transaction dataset to loan-level data from the Bank of
Israel.
• Directly identify buyers affected by the policy, i.e. those who purchase with
loan above NIS 800 thousand and with LTV above 60 percent.
Results 25
Direct Observation of High-LTV Mortgages
(1) (2) (3)
3.rooms group -0.677*** -0.840*** -0.605***
(0.178) (0.167) (0.00703)
4.rooms group -0.686*** -0.999*** -0.708***
(0.180) (0.170) (0.00279)
5.rooms group -0.652*** -1.144*** -0.519***
(0.182) (0.174) (0.00234)
ln age 0.00495*** 0.00515*** 0.00489***
(0.000484) (0.000474) (0.000490)
price 0.00128*** 0.00120***
(3.81e-05) (2.63e-05)
price sq’ -1.64e-07***
(7.22e-09)
Treatment 0.0764*** 0.00747 0.00831
(0.0149) (0.0157) (0.0106)
After 0.134*** 0.0533*** 0.0446***
(0.00822) (0.00930) (0.00310)
Treatment#After -0.0693*** -0.0749*** -0.0544***
(0.0259) (0.0238) (0.0208)
Constant 3.016*** 2.848*** 1.809***
(0.186) (0.172) (0.0173)
Observations 33,311 33,311 33,311
R-squared 0.514 0.550 0.586Results 26
Semi-Elasticity with respect to Interest Rates
• The mechanism underlying these results is that banks charge higher interest
rates on these high-LTV loans.
• Earlier research (Tzur-Ilan, 2017) has found that interest rates on mortgages
affected by this policy were higher by 0.31-0.36 percentage points.
• Combining with the baseline estimate of the effect on house prices, this
paper produces an estimate for the semi-elasticity in the range of 6-10,
consistent with the upper range of the results reported in the literature (e.g.
Adelino et al. (2012), Pinto et al. (2018), Kuttner (2014), Anenberg and
Kung (2017)).
Results 27
LTV limit - Welfare Implications
• To the extent that LTV limits are effective in reducing housing price growth,
they will also make housing more affordable.
• However, the reduction in price will come about via a reduction in demand,
as LTV limits make (some) mortgages more expensive.
• The effects may be especially strong on lower income households, which are
more likely to be liquidity constrained and to rely on riskier mortgages (e.g.,
high LTV mortgages).
Results 28
LTV limit - Welfare Implications
• The Israeli Central Bureau of Statistics publishes a socioeconomic index of
neighborhoods quality for each neighborhood in Israel.
• This index combines 16 different variables, including education, employment,
income, family size and standard of living into a single index.
• Neighborhoods are then classified into one of twenty clusters, 1 being the
lowest socioeconomic status and 20 being the highest.
• For our analysis, we divide neighborhoods into two groups: low-quality areas,
those neighborhoods that are graded from 1 to 10, and high-quality areas,
neighborhoods that are graded from 11 to 20.
• We repeat our main estimation separately on these two groups of
neighborhoods.Results 29
Differential Effects by Neighborhood Quality
Low-Graded Areas High-Graded Areas
(1) (2)
3.roomsgroup -0.220*** -0.270***
(0.00768) (0.00524)
4.roomsgroup -0.611*** -0.536***
(0.00345) (0.00517)
5.roomsgroup -0.549*** -0.691***
(0.0108) (0.00563)
lnage -0.0109*** -0.0126***
(0.000713) (0.00111)
Treatment -0.129*** 1.820***
(0.00397) (0.00865)
After 0.744*** 0.0334***
(0.0175) (0.00389)
TreatmentAfter -0.0571*** -0.0274***
(0.00380) (0.00362)
Geographic FE YES YES
Geographic FEAfter YES YES
Constant 2.254*** 2.566***
(0.00502) (0.00510)
Observations 38,585 51,747
R-squared 0.897 0.875Results 30
Larger Effects in Poorer Neighborhoods of More Expen-sive Areas
Low-Graded Areas High-Graded Areas
Jerusalem -0.0255** -0.02077***
(0.0113) (0.00679)
North -0.0102 0.00974
(0.00999) (0.00715)
Haifa -0.0137*** -0.00970
(0.00348) (0.0113)
Center -0.0761*** -0.0454***
(0.00742) (0.00703)
Tel-Aviv -0.0667*** -0.0324***
(0.00670) (0.0131)
South -0.00463 0.00239
(0.00487) (0.00348)
Results 31
Summary and Conclusions
• Provides new quantitative evidence for the impact of credit markets on house
prices and supports the interpretation that MPPs affect house prices through their
effects on mortgage interest rates.
• Finds a semi-elasticity of house prices with respect to interest rates in the range of
6-10, consistent with the upper range of estimates reported in the literature.
• Larger effects in poorer neighborhoods of more expensive areas. Suggests that
policies may disproportionately affect households that are already struggling to
afford their housing needs.
Summary 32
Thank you!
Summary 32