Systemic Banks, Mortgage Supply and Housing Rents Pedro Gete and - PowerPoint PPT Presentation

Systemic Banks, Mortgage Supply and Housing Rents Pedro Gete and Michael Reher Georgetown & Harvard September 2016

Motivation

What drives recent housing rents and HOR dynamics? � Tight credit supply (among other factors) � A 1pp increase in mortgage denials leads to... � � 2.3% increase in housing rents � � 2.4pp reduction in a city’s homeownership � � 40% increase in multifamily building permits

What drives recent housing rents and HOR dynamics? � Stress-testing since 2011 discourages risk-taking � SIFIs: BofA, Citi, JPM-Chase, Wells Fargo � Department of Justice invoking the False Claims Act since 2011 � Big-4 banks (plus Ally) paid $25 billion in 2012 � In addition, each of the Big-4 also faced other settlements: from $82 million for Wells Fargo in 2015 to $16.65 billion for Bank of America in 2014

� “If you guys want to stick with this programme of ‘ putting back ’ any time, any way, whatever, that’s fine, we’re just not going to make those loans and there’s going to be a whole bunch of Americans that are underserved in the mortgage market.” Wells Fargo’s CEO (August 2014, Financial Times) � Similar remarks by JP Morgan’s CEO

Our theory � Tight credit supply of Big-4 banks � More households denied credit � Frictions to substitute across lenders � Higher demand for rental housing, supply sluggish � Higher rents, HOR down, rental vacancies down � Increase construction of rental housing (multifamily)

� Each point groups around 15 MSAs

Identification strategy 1. Estimate national propensity to deny mortgage application by Big4 and non-Big4 banks (Khwaja and Mian 2008) Pr ( denial i , l , m , t = 1 ) = X i , l , m , t β + L l , t + α m , t + α m , l � Control for borrower’s characteristics ( X ilmt ) , lender, time, and regional shocks ( α m , t , α m , l ) � Focus on L l , t , a lender-year fixed effect (propensity to deny loan)

Big4 deny relatively more mortgages, especially after 2011

More denials among FHA loans

More denials among Black and Hispanics loans

Create credit shock à la Bartik � Wedge between lenders’ national propensity to deny weighted by market share : V m , t = ( L t , Big4 − L t , ∼ Big4 ) · share 2008 m � We control for other factors driving rents (population, income, MSA’s age, lagged rents, unemployment, past foreclosures...)

Use Bartik shock as IV for denial rates Stage 1: ∆ Denial Rate m , t = V m , t − 1 δ + ∆ X m , t η + λ m + λ t + v mt , Stage 2: ∆ log ( Rent ) m , t = ∆ Denial Rate m , t β + ∆ X m , t γ + α m + α t + u mt

IV Estimation (Stage 2) Table: Denial Rates and Rent Growth based on IV Estimation (Stage 2). Outcome: ∆ log(Rent m , t ) ∆ log(Rent m , t ) 2.342 ∗∗∗ 2.329 ∗∗ ∆ Denial Rate m , t (0.845) (0.940) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes # Observations 1380 1380

Table: Denial Rates and Homeownership Rate based on IV Estimation Outcome: ∆ HR m , t ∆ HR m , t -2.014 ∗ -2.367 ∗∗ ∆ Denial Rate m , t (1.128) (0.933) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes # Observations 358 358

Table: Denial Rates and Rental Vacancies Based on IV Estimation Outcome: ∆ Vacancy Rate m , t ∆ Vacancy Rate m , t ∆ Denial Rate m , t -1.256 -2.501 (1.399) (2.051) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes # Observations 348 348

Table: Denial Rates and New Building Permits Based on IV Estimation Outcome: ∆ log(Multi Unit) m , t ∆ log(Multi Unit) m , t 41.671 ∗∗∗ 49.529 ∗∗∗ ∆ Denial Rate m , t (15.264) (9.546) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes # Observations 1223 1223

Frictions to substitute among lenders 1. Internet accessibility (use of online lenders): � # inhabitants over 50yrs old to inhabitants 25-49 � Forbes.com rank of internet accessibility 2. Competition among credit suppliers: � States with tighter requirements to license brokers � Herfindahl index among non Big-4 lenders

Table: Credit Shock and Homeownership Rate by Internet Access Outcome: ∆ HR m , t ∆ HR m , t ∆ HR m , t ∆ HR m , t -1.620 ∗∗∗ -1.336 ∗∗∗ V m , t − 1 -0.293 0.238 (0.220) (0.279) (0.359) (0.152) -0.510 ∗∗∗ -0.509 ∗∗∗ V m , t − 1 × Older m (0.168) (0.173) -0.941 ∗∗∗ -1.136 ∗∗∗ V m , t − 1 × LowInternet m (0.360) (0.307) -0.538 ∗ V m , t − 1 × WRLURI m -0.398 (0.309) (0.281) MSA-Year Controls Yes Yes Yes Yes MSA FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes R-Squared 0.084 0.085 0.086 0.087 # Observations 358 358 358 358

Table: Credit Shock and Homeownership Rate by Broker and Lender Competition Outcome: ∆ HR m , t ∆ HR m , t ∆ HR m , t ∆ HR m , t -0.791 ∗∗∗ -3.378 ∗∗∗ -3.057 ∗∗∗ -0.329 V m , t − 1 (0.248) (1.027) (0.527) (0.976) V m , t − 1 × License m -0.223 -0.381 (0.208) (0.318) -2.583 ∗∗ -2.769 ∗∗ V m , t − 1 × HHI m (1.135) (1.176) -0.690 ∗∗ V m , t − 1 × WRLURI m -0.438 (0.341) (0.339) MSA-Year Controls Yes Yes Yes Yes MSA FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes R-Squared 0.082 0.107 0.084 0.111 # Observations 358 358 358 358

Conclusions � SIFI banks contracted credit supply � Effects on rents, HOR, vacancies � Effects to weaken as frictions to switch to new lenders are overcome � Once new buildings are complete, rent growth should slow

Appendix

Table: Determinants of Big-4 Share in 2008. Outcome: Share m , 08 1.845 ∗∗∗ ∆ Unempl Rate m , 07-08 (0.510) 1.116 ∗∗∗ ∆ log ( Rent ) m , 00-08 (0.393) -2.283 ∗∗∗ ∆ log ( Income ) m , 00-08 (0.554) -0.122 ∗∗ ∆ log ( Population ) m , 00-08 (0.055) -3.200 ∗∗∗ ∆ log ( Age ) m , 00-08 (1.023) -14.404 ∗∗∗ ∆ Unempl Rate m , 00-08 (2.849) 0.118 ∗∗∗ Big-4 Headquarter m (0.020) R-squared 0.302 Number of Observations 299

Geography of Big-4 market share

Bartik type regression ∆ log ( Rent ) m , t = V m , t − 1 β + ∆ X m , t γ + α m + α t + u m , t � X m , t control for: MSA’s age, unemployment, income, population, past rents and lags

Table: Credit Shock and Housing Rents in Bartik-type Regressions Outcome: ∆ log(Rent m , t ) ∆ log(Rent m , t ) 1.373 ∗∗∗ 1.373 ∗∗∗ V m , t − 1 (0.471) (0.526) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes R-squared 0.019 0.108 # Observations 1380 1380

Table: Credit Shock and Homeownership Rate in Bartik-type Regressions Outcome: ∆ HR m , t ∆ HR m , t -0.983 ∗∗∗ -1.003 ∗∗∗ V m , t − 1 (0.277) (0.135) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes R-squared 0.015 0.082 # Observations 358 358

Table: Rental Vacancies and Big-4 Credit Shock in Bartik-type Regressions Outcome: ∆ Vacancy Rate m , t ∆ Vacancy Rate m , t -0.923 ∗ -0.593 V m , t − 1 (0.641) (0.523) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes R-squared 0.052 0.290 # Observations 348 348

Table: New Building Permits and Big-4 Credit Shock in Bartik-type Regressions Outcome: ∆ log(Multi Unit) m , t ∆ log(Multi Unit) m , t 24.534 ∗∗ 29.796 ∗∗∗ V m , t − 1 (12.273) (8.899) MSA-Year Controls No Yes MSA FE Yes Yes Year FE Yes Yes R-squared 0.331 0.430 # Observations 1223 1223

Fly to quality?

IV estimation � What are effects of higher denial rates on rents, HOR, vacancies, construction? � Mortgage denial rates are likely endogenous with respect to housing rents: � lower rents = ⇒ ⇒ lower-quality borrowers choose to rent � = ⇒ quality of the pool of borrowers improves � = ⇒ denial rates decrease � =

� Instrument for denial rate with Bartik shock: � Valid instrument? hard to justify that either the systematic tightening of the Big-4’s approval standards or the historical presence of the Big-4 in an MSA are endogenous with respect to MSA-level rents. � We perform robustness checks based on pre-trends and alternate credit shocks

Robustness #1: Idiosyncratic Big-4 Share � Obtain idiosyncratic part of share 2008 m s m = share 2008 m − ˆ β X m � X m = set of variables that affect market share and rent dynamics over 2008-2014 � Re-estimate core specifications using a different definition of the V m , t shock: W m , t = ( L t , Big4 − L t , NoBig4 ) · s m .

Table: Determinants of Big-4 Share in 2008. Outcome: Share m , 08 1.845 ∗∗∗ ∆ Unempl Rate m , 07-08 (0.510) 1.116 ∗∗∗ ∆ log ( Rent ) m , 00-08 (0.393) -2.283 ∗∗∗ ∆ log ( Income ) m , 00-08 (0.554) -0.122 ∗∗ ∆ log ( Population ) m , 00-08 (0.055) -3.200 ∗∗∗ ∆ log ( Age ) m , 00-08 (1.023) -14.404 ∗∗∗ ∆ Unempl Rate m , 00-08 (2.849) 0.118 ∗∗∗ Big-4 Headquarter m (0.020) R-squared 0.302 Number of Observations 299

Table: Robustness Check: Bartik Regression and Second Stage IV Estimation Outcome: ∆ log(Rent m , t ) ∆ log(Rent m , t ) 1.245 ∗∗∗ W m , t − 1 (0.397) 2.226 ∗∗ ∆ Denial Rate m , t (0.901) MSA-Year Controls Yes Yes MSA FE Yes Yes Year FE Yes Yes # Observations 1368 1368