Leverage, Re-leveraging, and Household Spending

Leverage, Re-leveraging, and Household Spending Thomas Crossley (Essex) Peter Levell (IFS) Hamish Low (Cambridge) NIESR March 2018 1 / 35

Introduction How does borrowing and spending of more leveraged households respond to changes in house prices? (Leverage: Loan-to-value ratio on homes) Distinguish different reasons to respond Distinguish types of response: investing in housing or extra-consumption 2 / 35

Households Re-leverage as House Prices Rise (ages 25-45) -5 0 5 10 15 20 Prop. taking out new loans/house price growth % 0 20 40 60 80 100 LTV ratio 1995 2000 2005 2010 2015 Year Prop. taking out new loans LTV ratio House price growth Source: Authors calculations from British Household Panel Survey/Understanding Society, ONS 3 / 35

Value-to-income Ratio Increased (ages 25-45) -5 0 5 10 15 20 Prop. taking out new loans/house price growth % 4 5 6 7 Value-to-Income ratio 1995 2000 2005 2010 2015 Year Prop. taking out new loans Value-to-income ratio House price growth Source: Authors calculations from British Household Panel Survey/Understanding Society, ONS 4 / 35

Current Interpretation Credit constraints and consumption Finding: Households with highest leverage see biggest spending/borrowing responses when prices rise i.e effect of house price rise highly nonlinear in leverage extra borrowing = consumption Mian and Sufi (2011), Disney et al. (2010), Cooper (2013) etc. Also: High leverage suppresses consumption growth (Dynan, 2012) 5 / 35

Contribution of This Paper Simple portfolio model where: 1. Households have a target leverage 2. Prices rise: desire to re-leverage Borrow more and invest in housing 3. Price fall: desire to de-leverage Save, repay or sell 4. Wealth effects highly nonlinear (even w/o credit constraints) Empirical evidence for the UK that when prices rise, more leveraged households: 1. Borrow more 2. Increase residential investment 6 / 35

Outline 1. Framework for: understanding household choice of leverage response to house price shocks 2. Evidence on re-leveraging 3. Evidence on spending responses 7 / 35

Framework: Portfolio Returns and Leverage Two assets: Housing is risky asset (h t ) with price p t Risk-free asset (bond, b t ) price 1, interest rate r Can short bond (mortgage) 8 / 35

Framework: Portfolio Returns and Leverage Two assets: Housing is risky asset (h t ) with price p t Risk-free asset (bond, b t ) price 1, interest rate r Can short bond (mortgage) Return on housing: r t = p t p t 1 1 No credit constraints 8 / 35

Framework: Portfolio Returns and Leverage Leverage (Loan-to-value ratio): L t = loans house value = b t p t h t 9 / 35

Framework: Portfolio Returns and Leverage Leverage (Loan-to-value ratio): L t = Portfolio share of housing: ω t = gross housing wealth net wealth loans house value = b t p t h t = p th t p t h t + b t = 1 (1 L t ) 0 < L t < 1 implies ω t > 1 for household with 95% mortgage, L = 0.95: ω t = 20 for outright owners: ω t = 1 9 / 35

Framework: Leverage Magnifies Risk and Return The effect of shocks to returns: x t = (1 + r + ω t 1 (r t r)) (x t 1 c t 1 ) + y t x t = ( ) 1 1 + r + (rt r) (x t 1 c t 1 ) + y t 1 L t 1 10 / 35

Framework: Leverage Magnifies Risk and Return The effect of shocks to returns: x t = (1 + r + ω t 1 (r t r)) (x t 1 c t 1 ) + y t x t = ( ) 1 1 + r + (rt r) (x t 1 c t 1 ) + y t 1 L t 1 Effects are highly nonlinear Leveraged households: greater wealth gain for a given house price shock But variance of returns also higher 10 / 35

Framework: Leverage Dynamics in a Simple Model Ignore labour income, y; no adjustment costs CRRA preferences; Housing does not yield utility 11 / 35

Framework: Leverage Dynamics in a Simple Model Ignore labour income, y; no adjustment costs CRRA preferences; Housing does not yield utility Linear consumption function Constant desired leverage L c t = α t x t ω t = ω Meets consumers desires for appropriate risk and return L evolves smoothly over time (Merton (1969), Cocco et al. (2005)) Empirical support figure 11 / 35

Framework: Leverage Dynamics in a Simple Model How do households respond to shocks to wealth (positive house price shock)? x t E [x t ] = ω (r t E [r ]) (x t 1 c t 1 ) x t E [x t ] = (p t E [p t ]) h t 1 12 / 35

Framework: Leverage Dynamics in a Simple Model How do households respond to shocks to wealth (positive house price shock)? x t E [x t ] = ω (r t E [r ]) (x t 1 c t 1 ) x t E [x t ] = (p t E [p t ]) h t 1 Partly consumed: c t E [c t ] = α t (x t E [x t ]) Partly saved (provides extra-downpayment): s t E [s t ] = (1 α t ) (x t E [x t ]) 12 / 35

Framework: Leverage Dynamics in a Simple Model This saving is leveraged by ω : p t h t E [p t h t ] = ω (1 α t ) (x t E [x t ]) (1) 13 / 35

Framework: Leverage Dynamics in a Simple Model This saving is leveraged by ω : p t h t E [p t h t ] = ω (1 α t ) (x t E [x t ]) (1) Using: x t E [x t ] = (p t E [p t ]) h t 1 (2) 13 / 35

Framework: Leverage Dynamics in a Simple Model This saving is leveraged by ω : p t h t E [p t h t ] = ω (1 α t ) (x t E [x t ]) (1) Using: x t E [x t ] = (p t E [p t ]) h t 1 (2) Extra active investment in housing: (p t h t E [p t h t ]) (p t h t 1 E [p t ] h t 1 ) }{{}}{{} housing wealth mechanical increase = (ω (1 α t ) 1) (p t E [p t ]) h t 1 Unexpected house price increase: invest more if leveraged Unexpected house price fall: sell if leveraged 13 / 35

Example: α = 0.05 and ω = 3 ( 600K house, 33% downpayment) House value goes up 5% ( 30K) Increases new investment in housing by 55K 14 / 35

Example: α = 0.05 and ω = 3 ( 600K house, 33% downpayment) House value goes up 5% ( 30K) Increases new investment in housing by 55K Model to quantify this effect Empirical work to estimate importance 14 / 35

Unleveraged Household ω t < 1: House Price Increase μ μ H ω H =1 ω H * σ H σ 15 / 35

Leveraged Household ω t > 1: House Price Increase μ μ H ω H =1 ω H* =1/(1-L * ) σ H σ 16 / 35

Re-leveraging μ - Borrow more and invest in housing μ H ω H =1 ω H* =1/(1-L * ) σ H σ 17 / 35

Framework: Leverage Dynamics L t = b t p t h t House price increase: if leveraged (ω > 1), leverage and portfolio share fall 18 / 35

Framework: Leverage Dynamics L t = b t p t h t House price increase: if leveraged (ω > 1), leverage and portfolio share fall = ω t < ω : Borrow, buy more housing 18 / 35

Framework: Leverage Dynamics L t = b t p t h t House price increase: if leveraged (ω > 1), leverage and portfolio share fall = ω t < ω : Borrow, buy more housing House price decrease: if leveraged (ω > 1), leverage and portfolio share rise = ω t > ω : sell housing (if you can), or save to reduce debt 18 / 35

Empirical Implications 1. When house prices rise consumers re-leverage to attain ω Effects greater for more leveraged households 19 / 35

Empirical Implications 1. When house prices rise consumers re-leverage to attain ω Effects greater for more leveraged households 2. Wealth effects nonlinear in leverage w/o credit constraints 3. Leveraged households want to invest in housing after price rise If transaction costs make moving costly, can invest in own home (extensions, renovations, repairs etc.) Housing investments are irreversible = asymmetric responses 19 / 35

Data Leverage from British Household Panel Survey/Understanding society Spending data from the Living Costs and Food Survey 1993-2013 Sample of homeowners, aged 25-45 20 / 35

Descriptives Mean % own home 68.7% LTV ratio 0.34 ω t (housing share) 3.04 Total spend ( ann.) 30,416 Nondurable 23,234 Durable 4,217 Residential inv. 2,965 % Res inv. > 0 79.6% 21 / 35

Evidence for Re-leveraging We know that Take logs L t = Debt t House value t log L t = log Debt t log HValue t log L t = β log HValue t + ε t H 0 : households passive when house prices rise (β = 1) 22 / 35

log L t = β log HValue t + ε t log LTV OLS IV IV β -0.881-0.562-0.606 95% Confidence Interval [-0.91,-0.85] [-0.62,-0.50] [-0.66,-0.55] Control for age N N Y R 2 0.240 0.216 0.213 N 26,829 26,694 26,694 Clusters 7,411 7,403 7,403 23 / 35

Evidence on Spending Now consider impact of house price rises on spending of more leveraged households 24 / 35

Evidence on Spending Now consider impact of house price rises on spending of more leveraged households We look at total spending, nondurable consumption and residential investment spending Issues 1. Leverage is endogenous 24 / 35

Empirical approach Our approach: 1. Two sample IV using cross-sectional budget survey and panel data with wealth 2SIV 2. Instrument for leverage: national average house price-to-income ratio at time households first moved into their current homes Previous studies mainly use lagged leverage 25 / 35

Estimating Equation { ( )} prt ln C i,t = β 0 + γ crt + β 1 ω i,t 1 + β 2 ω t 1 1 + X i,t β 3 + ε i,t p rt 1 γ crt are cohort-region-time effects p rt are regional house prices X is a vector of controls: education, years at address, house type etc. LHS: log total spending, log consumption and residential investment Approach derived from a first order condition with i.i.d housing shocks (random walk in prices) 26 / 35

Identification Instrument for leverage: measure of borrowing needs at the point moved into current home 27 / 35

Identification Instrument for leverage: measure of borrowing needs at the point moved into current home Average price-to-income ratio on new purchases (P/Y ) T and its interaction with house price growth ( ) p Instruments for ω i,t 1 and ω r,t i,t 1 p r,t 1 1 27 / 35

Price-to-income Ratios 2004 5.0 Price to income ratio (new purchases) 4.5 4.0 3.5 3.0 1989 1996 2.5 1970 1980 1990 2000 2010 28 / 35

Loan-to-value Ratios by Year Moved in (1960s cohort) 0.8 Loan to value ratio 0.6 2004 0.4 1996 0.2 1989 1995 2000 2005 2010 29 / 35

Identification Instrument for leverage: measure of borrowing needs at the point moved into home Average price-to-income ratio on new purchases P/Y T and its interaction with house price growth ( ) p Instruments for ω t 1 and ω r,t t 1 p r,t 1 1 Good first stage First Stage Uncorrelated with possible confounders Validity 30 / 35

Results (Total Spending) log Total ω it 1 ( pr,t p r,t 1 1) 0.050 (0.066) ω it 1-0.018 (0.011) R 2 0.355 N 60,357 31 / 35

Results Nondurables Durables Res inv. (IHS) (IHS) ω it 1 ( pr,t p r,t 1 1) -0.020-0.168 0.687*** (0.060) (0.243) (0.257) ω it 1-0.015-0.036-0.045 (0.010) (0.041) (0.043) R 2 0.377 0.113 0.082 N 60,357 60,357 60,357 32 / 35

Conclusions Simple portfolio model explains re-leveraging and de-leveraging as house prices rise/fall Even w/o credit constraints and Even w/ iid return expectations 33 / 35

Conclusions Simple portfolio model explains re-leveraging and de-leveraging as house prices rise/fall Even w/o credit constraints and Even w/ iid return expectations If house prices rise, then: to smooth consumption and leverage, must increase the size of the balance sheet 33 / 35

Conclusions Empirical analysis with attention to endogeneity of (lagged) leverage composition of spending 35 / 35

Conclusions Empirical analysis with attention to endogeneity of (lagged) leverage composition of spending Evidence for the UK that more leveraged households Re-leveraging as prices rise Extra borrowing spent on residential investment Suggests significant feedback mechanisms price rises feed housing demand Leveraged households take on more debt as prices rise = more sensitive to shocks 35 / 35

First Stage Results ω it 1 ω it 1 ( prt p rt 1 1) P/Y T 0.437-0.087 (0.049) (0.003) P/Y T ( prt p rt 1 1) 0.651 0.736 (0.407) (0.057) Shea partial R 2 0.008 0.028 F-stat (p-value) 39.51 111.29 (<0.001) (<0.001) Kleibergen-Paap (p-value) 72.40 (<0.001) N 28,823 Clusters 7,843 Back 35 / 35

Instrument Validity Panel a Credit conditions Dependent var. log(hvalue) Invest inc. > 1000 Invest inc.= 0 Mover t+1 P/Y T 0.012-0.002-0.006-0.001 (0.012) (0.0089) (0.014) (0.004) P/Y T ( prt p rt 1 1) -0.051-0.061 0.089 0.014 (0.105) (0.077) (0.123) (0.037) F-test (p-values) 0.467 0.731 0.625 0.876 Panel b Lagged leverage LTV t 1-0.197*** -0.166*** 0.334*** 0.0001 (0.0207) (0.0138) (0.0217) (0.00829) LTV t 1 ( prt p rt 1 1) -0.595*** -0.142 0.192 0.051 (0.197) (0.136) (0.216) (0.0850) F-test (p-values) < 0.001 < 0.001 < 0.001 0.821 Back 35 / 35

Leverage by Age and Cohort Figure: LTV ratios by age and cohort, 1994-2013 0.8 0.6 Loan to value ratio 0.4 0.2 0.0 20 30 40 50 60 70 Age 1940 1949 1950 1959 1960 1969 1970 1979 Back 35 / 35

Two Sample IV Back Angrist and Krueger (1992) β 2SIV = (Z 1X 1 ) 1 (Z 2Y 2 ) Two-stage least squares version more efficient in finite samples (Inoue and Solon, 2010) 35 / 35