House Prices, Credit Growth, and Excess Volatility:

House Prices, Credit Growth, and Excess Volatility: Implications for Monetary and Macroprudential Policy Paolo Gelain Kevin J. Lansing 2 Caterina Mendicino 3 4th Annual IJCB Fall Conference New Frameworks for Monetary Policy Analysis in an Era of Crises September 28, 202 Norges Bank 2 Federal Reserve Bank of San Francisco and Norges Bank 3 Banco de Portugal

What are the lessons of the nancial crisis for policy? Could policymakers have done more to prevent the buildup of nancial imbalances, particularly in the household sector?

What are the lessons of the nancial crisis for policy? Could policymakers have done more to prevent the buildup of nancial imbalances, particularly in the household sector? Should central banks take deliberate steps to prevent or de ate suspected bubbles? If so, what policy instruments should be used to do so? Standard macro-modeling approach: House price booms driven by preference shocks. Financial crises caused by capital quality shocks. All agents are fully-rational. This Paper: DSGE model of housing with excess volatility. Subset of agents employ moving-average forecast rules. Policy experiments: Interest-rate response to house price growth or credit growth. Tightening of lending standards (lower LTV). Weight on wage income in borrowing constraint. (best).

Related literature (partial list) Interest rate response to asset prices or credit in RE models Dupor (2005) Gilchrist and Saito (2008) Christiano, Ilut, Motto and Rostagno (200) Airaudo, Cardani, and Lansing (202) Macroprudential policy tools Galati and Moessner (20), BIS Working Paper 337 Macroprudential policy: A literature review. Bank of England (20), Discussion Paper Instruments of macroprudential policy. Countercyclical LTV rules in RE models Kannan, et al. (2009), Angelini, et al. (200), Christensen and Meh (20), Lambertini, et al. (20).

Household leverage, house prices, and consumption From Glick and Lansing (200), FRBSF Economic Letter 200-0.

% deviation from trend % deviation from trend % deviation from trend Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion U.S. Housing Boom of the mid-2000s New buyers with access to easy credit helped fuel an excessive run-up in house prices. 0.6 U.S. real house prices (in logs) U.S. real house prices 0.4 0 0.2 0 0 Linear trend 0 0.2 970 980 990 2000 200 970 980 990 2000 200 U.S. real household debt per capita (in logs) 2 U.S. real household debt per capita.5 0.5 Linear trend 20 0 0 20 0.5 970 980 990 2000 200 970 980 990 2000 200 0.8 U.S. real GDP per capita (in logs) 0 U.S. real GDP per capita 0.6 0.4 Linear trend 5 0 0.2 5 0 970 980 990 2000 200 0 970 980 990 2000 200

Housing Market Expectations Futures tend to overpredict prices when prices are falling (moving average forecast rule).

Survey Expectations about U.S. House Prices Survey expectations track past house price changes. Case and Shiller (2003): Surveys in 2002-3. 90% of survey respondents expect house prices to increase over the next several years. Over the next 0 years, respondents expect annual price appreciation in the range of 2 to 6% per year. Piazzesi and Schneider (2009): Starting in 2004, more and more households became optimistic after having watched house prices increase for several years. Shiller (2007): Surveys in 2006-7. Places with high recent house price growth exhibited high expectations of future price appreciation, while places with slowing price growth exhibited downward shifts in expected appreciation. Case, Shiller and Thompson (202): Survey in 2008. Respondents in prior boom areas now mostly expect declines in future house prices.

House Prices and Their Expectations in Four Cities From Case, Shiller, and Thompson (202), NBER Working Paper 8400.

Survey-Based In ation Expectations Survey forecasts exhibit -sided forecast errors, resemble moving-average of past in ation.

Loan-to-Value (LTV) versus Loan-to-Income (LTI) Ratios LTI provided a much earlier warning signal of rising household leverage. 00 90 Leverage ratios U.S. data Household mortgage debt/household real estate assets Average LTV of mortgaged homeowners Household mortgage debt/personal disposable income 80 70 60 50 40 30 20 970 975 980 985 990 995 2000 2005 200

Understanding Household Debt Obligations Remarks at Credit Union National Association Governmental A airs Conference (2004) Overall, the household sector seems to be in good shape, and much of the apparent increase in the household sector s debt ratios over the past decade re ects factors that do not suggest increasing household nancial stress.

Households: Patient-lenders and Impatient-borrowers Basic setup is similar to Iacoviello (2005, AER). max be,t t=0 β t L +ϕ L,t log (c,t bc,t ) + ν,h log (h,t ) ν,l +ϕ, L c,t + I t + q t (h,t h,t ) + b,t R t π t = b,t + w t L,t + rt k k t + φ t. ψ 2 k t = ( δ)k t + [ It 2 I t ] It, max be 2,t t=0 β t 2 L +ϕ L 2,t log (c 2,t bc 2,t ) + ν 2,h log (h 2,t ) ν 2,L +ϕ, L c 2,t + q t (h 2,t h 2,t ) + b 2,t R t π t = b 2,t + w t L 2,t, b 2,t γ R t h b E,t q t+ π t+ i h 2,t, β 2 < β (Incentive to borrow)

Household Expectations Subset employ moving-average forecast rules. Remainder employ rational forecast rules. F t X t+ {z } Current forecast = F t X t {z } = λ Previous forecast + λ (X t F t X t ), 0 < λ, {z } Previous forecast error i h X t + ( λ) X t + ( λ) 2 X t 2 +..., where λ = weight on recent data in moving average. X t+ = object to be forecasted. = U c,t+ hq k t+( δ) + r k t+ i (example).

% deviation from steady state % deviation from steady state Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion Hybrid Expectations Model Exhibits Excess Volatility Moving-average forecast rule embeds a unit root which magni es volatility. House price 0 Baseline comparison Household debt Price of capital 30 6 Consumption 6 5 20 4 4 0 0 0 2 0 2 0 5 0 2 2 0 0 00 200 20 0 00 200 4 0 00 200 4 0 00 200 0 Output Labor hours 5 4 Inflation Policy interest rate 4 5 0 0 2 0 2 0 5 5 2 2 0 0 00 200 0 00 200 Rational expectations 4 0 00 200 Hybrid expectations 4 0 00 200

Monetary Policy and Macroprudential Policy What policy actions are e ective in dampening excess volatility in credit, output, etc.? Interest-rate response to house price growth or credit growth: πt.5 0.25 αq αb yt qt b2,t R t = ( + r) ς y t, q t 4 α q or α b 2 [0, 0.4], (baseline = 0) b 2,t 4 Lower LTV or move towards LTI constraint: b 2,t γ R t h b E,t q t+ π t+ i h 2,t γ 2 [0.2,.0], (baseline = 0.7) b 2,t bγ h i o nm w t L 2,t + ( m) E b,t q t+ π t+ h 2,t R t m 2 [0, ] (baseline = 0)

Volatility ratio Relative loss Volatility ratio Volatility ratio Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion Interest Rate Response to House Price Growth Reduces volatility of household debt but magni es volatility of output and n ation. Sensitivity to interest rate response to house price growth Hybrid expectations House price Household debt Price of capital.2.3.2. 0.95. 0.9 0.85 0 0.2 0.4 0 0.2 0.4 0 0.2 0.4 α q α q α q Consumption Output Labor hours.04.02 0.98 0 0.2 0.4 α q Inflation 2.5 0 0.2 0.4 α q..05.6.4.2 0 0.2 0.4 α q Policy interest rate 0 0.2 0.4 α q.4.2.4.2 0 0.2 0.4 α q Loss function Loss func. Loss func. 2 0.8 0 0.2 0.4 α q

Volatility ratio Relative loss Volatility ratio Volatility ratio Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion Interest Rate Response to Credit Growth Tends to magnify volatility of household debt and other macro variables...05.05 Sensitivity to interest rate response to credit growth Hybrid expectations House price Household debt Price of capital 0 0.2 0.4 α b Inflation.5..05 0 0.2 0.4 α b Consumption..04 3 2 0 0.2 0.4 α b 0.95 0 0.2 0.4 α b.02 Output 0 0.2 0.4 α b Policy interest rate 3 2 0 0.2 0.4 α b..05 0.95 0 0.2 0.4 α b.5..05 Labor hours 0 0.2 0.4 α b Loss function 2.5 Loss func. Loss func. 2 0.5 0 0.2 0.4 α b

Monetary policy results depend on expectations Previous results obtained from rational expectations models may not be robust. Interest rate response to credit growth (α b = 0.2) House price Standard deviations HH debt Output In ation Rational Expectations Not responding 2.08 3.7 2.3 0.8 Responding 2.4 2.00 2.34 0.84 Volatility Ratio.03 0.63.0.04 Hybrid Expectations Not responding 3.62 6.55 3.4 0.90 Responding 3.72 6.68 3.8.65 Volatility Ratio.03.02.0.83 Standard deviations expressed as percent deviations from steady state.

Relative loss Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion Tighten Lending Standards: Lower LTV Reduces volatility of household debt but magni es volatility of other macro variables...05 House price 0.95 0.2 0.4 0.6 0.8 γ Consumption.03.02.0 0.99 0.2 0.4 0.6 0.8 γ Inflation.04.02 0.98 0.2 0.4 0.6 0.8 γ Sensitivity to LTV ratio Hybrid expectations Household debt.6.4.2 0.8 0.6 0.2 0.4 0.6 0.8 γ Output.05 0.95 0.2 0.4 0.6 0.8 γ Policy interest rate.05 0.95 0.2 0.4 0.6 0.8 γ..05 Price of capital 0.95 0.2 0.4 0.6 0.8 γ Labor hours.04.02 0.98 0.2 0.4 0.6 0.8 γ Loss function 2 Loss func. 2.5 Loss func. 0.5 0.2 0.4 0.6 0.8 γ

% deviation from steady state % deviation from steady state Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion Volatility Comparison: Wage Income versus Housing Value Wage income is less subject to bubble-induced distortions. 20 Volatility comparison: borrower's wage income versus housing value Borrower's wage income Borrower's housing value 20 5 5 0 0 5 5 0 0 5 5 0 0 5 0 50 00 50 200 Rational expectations 5 0 50 00 50 200 Hybrid expectations

Volatility ratio Volatility ratio Volatility ratio Relative loss Move Towards Loan-to-Income Constraint (Best) Reduces volatility of household debt as well other economic variables. Sensitivity to weight on wage income in borrowing constraint Hybrid expectations House price Household debt Price of capital.06.04.04.02 0.8.02 0.6 0.98 0 0.5 m Consumption.005 0.995 0 0.5 m Inflation.004.002 0.998 0 0.5 m 0.4 0 0.5 m Output.005 0.995 0.99 0 0.5 m Policy interest rate.03.02.0 0.99 0 0.5 m 0.98 0 0.5 m Labor hours 0.95 0.9 0 0.5 m Loss function 0.9 0.8 0.7 Loss func. 2 Loss func. 0 0.5 m

% deviation from steady state Overview Cross-Country Data U.S. Data Model Policy Experiments Conclusion Endogenous LTV acts like an automatic stabilizer Weight on wage income in borrowing constraint induces countercyclical LTV ratio. 3 Endogenous movements in LTV with generalized borrowing constraint Rational expectations Hybrid expectations 3 2 2 0 0 2 2 3 0 50 00 LTV 3 0 50 00 House value

Conclusion No policy was perfect but some did better than others. Interest rate response to either house price growth or credit growth had the serious drawback of substantially magnifying the volatility of in ation. A lower LTV ratio mildly raised the volatilities of output, in ation, and consumption, but reduced the volatility of household debt a nancial stability bene t. Best-performing policy: Require lenders to put substantial weight on wage income in the borrowing constraint. Promotes both economic and nancial stability (automatic stabilizer).