Explaining the Boom-Bust Cycle in the U.S. Housing Market: A Reverse-Engineering Approach

Transcription

1 Explaining the Boom-Bust Cycle in the U.S. Housing Market: A Reverse-Engineering Approach Paolo Gelain Norges Bank Kevin J. Lansing FRBSF Gisle J. Navik Norges Bank October 22, 2014 RBNZ Workshop The Interaction of Monetary and Macroprudential Policy Any opinions expressed here do not necessarily reflect the views of the managements of the Norges Bank, the Federal Reserve Bank of San Francisco, or the Board of Governors of the Federal Reserve System.

2 U.S. housing market boom and bust, 1995 to 2012 Correlated booms and busts in house prices, mortgage debt, and consumption. Housing rent-income ratio did not increase during the boom. Start of NBER recession = 2007.Q4.

3 What explains the U.S. housing market boom and bust? Rational explanations of the boom often appeal to exogenous housing demand shocks (increase in housing preference). Problem with this story: An increase in housing preference would increase the housing service flow, as measured by the imputed rent. But this did not happen in the data.

4 What explains the U.S. housing market boom and bust? Rational explanations of the boom often appeal to exogenous housing demand shocks (increase in housing preference). Problem with this story: An increase in housing preference would increase the housing service flow, as measured by the imputed rent. But this did not happen in the data. Empirical evidence: Changes in lending standards contributed to the boom-bust cycle.

5 What explains the U.S. housing market boom and bust? Rational explanations of the boom often appeal to exogenous housing demand shocks (increase in housing preference). Problem with this story: An increase in housing preference would increase the housing service flow, as measured by the imputed rent. But this did not happen in the data. Empirical evidence: Changes in lending standards contributed to the boom-bust cycle. This Paper: Investigate how a simple asset pricing model with 4 shocks can account for the boom-bust patterns in U.S. data over the period 1995 to 2012.

6 What explains the U.S. housing market boom and bust? Rational explanations of the boom often appeal to exogenous housing demand shocks (increase in housing preference). Problem with this story: An increase in housing preference would increase the housing service flow, as measured by the imputed rent. But this did not happen in the data. Empirical evidence: Changes in lending standards contributed to the boom-bust cycle. This Paper: Investigate how a simple asset pricing model with 4 shocks can account for the boom-bust patterns in U.S. data over the period 1995 to Model with moving-average forecast rules and long-term mortgages does best in matching the U.S. data. Matches data with small housing preference shocks and plausible lending standard shocks.

7 Competing explanations for the boom-bust episode Interest rates too low (empirical). Taylor (2007, Jackson Hole)

8 Competing explanations for the boom-bust episode Interest rates too low (empirical). Taylor (2007, Jackson Hole) Shifting lending standards (empirical). Mian and Sufi (2009, IMF Review) Demyanyk and Van Hemert (2009, Rev. Financial Studies) Duca, Muellbauer & Murphy (2011, Economic Journal) Dokko, et al. (2011, Economic Policy)

9 Competing explanations for the boom-bust episode Interest rates too low (empirical). Taylor (2007, Jackson Hole) Shifting lending standards (empirical). Mian and Sufi (2009, IMF Review) Demyanyk and Van Hemert (2009, Rev. Financial Studies) Duca, Muellbauer & Murphy (2011, Economic Journal) Dokko, et al. (2011, Economic Policy) Shifting lending standards plus rational learning (model). Boz and Mendoza (2012, JME)

10 Competing explanations for the boom-bust episode Interest rates too low (empirical). Taylor (2007, Jackson Hole) Shifting lending standards (empirical). Mian and Sufi (2009, IMF Review) Demyanyk and Van Hemert (2009, Rev. Financial Studies) Duca, Muellbauer & Murphy (2011, Economic Journal) Dokko, et al. (2011, Economic Policy) Shifting lending standards plus rational learning (model). Boz and Mendoza (2012, JME) Housing preference shocks, not LTV shocks (model). Justiniano, Primiceri & Tambalotti (2013, NBER WP 18941)

11 Competing explanations for the boom-bust episode Interest rates too low (empirical). Taylor (2007, Jackson Hole) Shifting lending standards (empirical). Mian and Sufi (2009, IMF Review) Demyanyk and Van Hemert (2009, Rev. Financial Studies) Duca, Muellbauer & Murphy (2011, Economic Journal) Dokko, et al. (2011, Economic Policy) Shifting lending standards plus rational learning (model). Boz and Mendoza (2012, JME) Housing preference shocks, not LTV shocks (model). Justiniano, Primiceri & Tambalotti (2013, NBER WP 18941) Looser saving constraints > lower interest rates (model). Justiniano, Primiceri & Tambalotti (2014, WP)

12 Competing explanations for the boom-bust episode Interest rates too low (empirical). Taylor (2007, Jackson Hole) Shifting lending standards (empirical). Mian and Sufi (2009, IMF Review) Demyanyk and Van Hemert (2009, Rev. Financial Studies) Duca, Muellbauer & Murphy (2011, Economic Journal) Dokko, et al. (2011, Economic Policy) Shifting lending standards plus rational learning (model). Boz and Mendoza (2012, JME) Housing preference shocks, not LTV shocks (model). Justiniano, Primiceri & Tambalotti (2013, NBER WP 18941) Looser saving constraints > lower interest rates (model). Justiniano, Primiceri & Tambalotti (2014, WP) Bubble : Departure from rational expectations Adam, Kuang & Marcet (2012, NBER Macro Annual) Gelain, Lansing & Mendicino (2013, IJCB)

13 Relaxed lending standards and the run-up in house prices Source: B. Tal (2006), CIBC World Markets, Consumer Watch U.S. (October 18). House prices rose faster where lending standards were weaker.

14 What motivated the change in lending standards? [T]he financial services sector has been dramatically transformed by technology...where once more-marginal applicants would simply have been denied credit, lenders are now able to quite effi ciently judge the risk posed by individual applicants and to price that risk appropriately. These [technology] improvements have led to rapid growth in subprime mortgage lending.

15 What motivated the change in lending standards? [T]he financial services sector has been dramatically transformed by technology...where once more-marginal applicants would simply have been denied credit, lenders are now able to quite effi ciently judge the risk posed by individual applicants and to price that risk appropriately. These [technology] improvements have led to rapid growth in subprime mortgage lending. Mortgage Delinquencies by Loan Type

16 What motivated the change in lending standards? [T]he financial services sector has been dramatically transformed by technology...where once more-marginal applicants would simply have been denied credit, lenders are now able to quite effi ciently judge the risk posed by individual applicants and to price that risk appropriately. These [technology] improvements have led to rapid growth in subprime mortgage lending. Mortgage Delinquencies by Loan Type Fed Chairman Alan Greenspan, April 8, 2005.

17 House price changes and their expectations in four cities Source: Case, Shiller, and Thompson (2012), Brookings Papers on Economic Activity.

18 Expected house price changes vs. lagged price changes Source: Case, Shiller, and Thompson (2012), Brookings Papers on Economic Activity. 12-month expectations are fairly well described as attenuated versions of lagged actual 12-month price changes.

19 Inflation expectations: Moving-average of past inflation Data source: Survey of Professional Forecasters 1970.Q2 to 2012.Q4. ) Ê t π t+1 Ê t 1 π t = (π t Ê t 1 π t, (0.114) (0.025)

20 Simple asset pricing model with housing services Four shocks: housing preference, lending standard, income growth, interest rate. max Ê 0 c t, h t, b t+1, δ t+1 β t (c t ht θt ), t=0 c t + p t h t + (R t 1 + δ t ) b t = y t + p t h t 1 + l t, ] l t m t [Êt p t+1 h t b t+1, R t = 1 + r t, }{{} Next period home equity. b t+1 = (1 δ t ) b t + l t, l t = New loans δ t+1 = ( 1 l ) t δt α + l t (1 α) κ b t+1 b t+1 One-period mortgage: α = 0 δ t = 1 for all t. 30-yr. Mortgage α = κ 1

21 Payment schedule: Model vs. 30-year mortgage From Kydland, Rupert, and Sustek (2012), NBER Working Paper Solid line: Model. Dashed line: 30-year mortgage schedule.

22 Model (continued) Four stochastic AR(1) shocks: θ t = θ exp (u t ), u t = ρ u u t 1 + ε u,t ε u,t N ( 0, σ 2 ) u m t = m exp (v t ) v t = ρ v v t 1 + ε v,t ε v,t N ( 0, σ 2 ) v log y t x t = x + ρ y x (x t 1 x) + ε x,t ε x,t N ( 0, σ 2 ) x t 1 R t = R exp (τ t ), τ t = ρ τ τ t 1 + ε τ,t ε τ,t N ( 0, σ 2 ) τ

23 Model (continued) Four stochastic AR(1) shocks: θ t = θ exp (u t ), u t = ρ u u t 1 + ε u,t ε u,t N ( 0, σ 2 ) u m t = m exp (v t ) v t = ρ v v t 1 + ε v,t ε v,t N ( 0, σ 2 ) v log y t x t = x + ρ y x (x t 1 x) + ε x,t ε x,t N ( 0, σ 2 ) x t 1 R t = R exp (τ t ), τ t = ρ τ τ t 1 + ε τ,t ε τ,t N ( 0, σ 2 ) τ First-order condition (h t = 1): p t = θ t c t + m t µ 1 + m t Ê t p t+1 }{{ t } Dividend = Imputed rent + βê t p t+1

24 Moving-average forecast rules Ease of computation: Agent does not need to know the stochastic processes for shocks. Forecast = exponentially-weighted moving average of past values. Ê t z t+1 = λ z t + (1 λ) Ê t 1 z t, λ [0, 1], ] = Ê t 1 z t + λ [z t Ê t 1 z t, [ ] = λ z t + (1 λ) z t 1 + (1 λ) 2 z t 2... λ = weight on most-recent data in moving average. z t+1 = any stationary object to be forecasted. = p t+1 y t+1 exp (x t+1 ) (example). This setup is consistent with a wide variety of survey evidence.

25 Model parameter values Target ratios based on U.S. data in 1995.Q1. One-period Long-term Parameter Mortgage Mortgage Target α Approximate 30-year mortgage. κ Approximate 30-year mortgage. β House price/rent 83. θ Housing value/income 6.3. m Mortgage debt/income 2.4. x Quarterly income growth = 0.45%. R Quarterly real mortgage rate. λ Est. from inflation survey data.

26 Calibrating the stochastic shocks Four shocks: housing preference, lending standard, income growth, interest rate Parameter RE Model MA Model Target ρ u AR(1) house value/income. σ u Std. dev. house value/income. ρ v AR(1) debt/income. σ v Std. dev. debt/income. ρ x AR(1) income growth. σ x Std. dev. income growth. ρ τ AR(1) mortgage rate. σ τ Std. dev. mortgage rate. RE model requires a more volatile housing preference shock. Volatile housing preference shock implies volatile rent-income ratio.

27 Model simulations: Rational expectations Volatile housing preference shock implies volatile rent-income ratio.

28 Model simulations: Moving-average forecast rules Much less volatile housing preference shock and rent-income ratio.

29 Impulse response: Housing preference shock Moving average (MA) model: Higher volatility and persistence. Long-term mortgage: Peak in debt-income ratio comes later.

30 Impulse response: Lending standard shock Moving average (MA) model: Higher volatility and persistence. Long-term mortgage: Peak in debt-income ratio comes later.

31 What sequence of shocks explains the U.S. data? Use log-linear decision rules from model to reverse-engineer the exogenous shock sequences needed to exactly match the observed patterns in U.S. data from 1995.Q1 to 2012.Q4. x t = U.S. per capita real disposable income growth (smoothed). R t 1 = U.S. quarterly real mortgage interest rate (smoothed). (p t h t /y t ) Take observed time series from U.S. data. (b t /y t ) θ t = θ exp (u t ) m t = m exp (v t ) Given decision rules & data, solve for shocks.

32 U.S. housing market boom and bust, 1995 to 2012

33 Inputs to the reverse-engineering excercise. Shocks identified using smoothed versions of U.S. income growth and mortgage rate

34 Reverse engineering the shocks to match the data RE model: Requires much larger housing preference shock. One-period mortgage: Post-2007 loosening of lending standards.

35 Two indicators of U.S. bank lending standards Source: Federal Reserve Senior Loan Offi cer Opinion Survey (SLOOS). Banks started to tighten standards before the Great Recession.

36 Properties of reverse-engineered shocks Moving-average model: Smaller housing preference shocks. Long-term mortgage: Larger lending standard shocks. RE Model MA Model One-period Long-term One-period Long-term Shock Housing Preference Statistic Mean Std. Dev. Mortgage Mortgage Mortgage Mortgage Lending Standard Mean Std. Dev

37 Model-implied rent-income ratio RE Model: Large preference shocks make rent-income ratio too volatile. Recall: Rent t = θ t c t + µ t m t 1+m t Ê t p t+1

38 Model-implied consumption-income ratio By construction, all models imply similar paths for the consumption-income ratio. c t /y t = 1 + b t+1 /y t (R t 1 + δ t )b t /y t

39 Counterfactual 1: No housing preference shock Moving average model can still generate a significant boom-bust episode.

40 Counterfactual 2: No lending standard shock One-period mortgage model implies a rapid deleveraging when housing value falls.

41 Counterfactual 3: No income growth shock MA model with long-term mortgage debt shows a smaller boom-bust cycle.

42 Counterfactual 4: No mortgage interest rate shock All models continue to exhibit a substantial boom-bust episode.

43 Conclusion RE model: Requires a large housing preference shock to match boom-bust in U.S. housing value. This implies a boom-bust in rent-income ratio, which is counterfactual.

44 Conclusion RE model: Requires a large housing preference shock to match boom-bust in U.S. housing value. This implies a boom-bust in rent-income ratio, which is counterfactual. One-period mortgage debt: Implies rapid deleveraging after 2007 as housing values fell. Hence, the model requires an implausible, post-2007 relaxation of lending standards ( m t ) to explain the slow deleveraging observed in the data.

45 Conclusion RE model: Requires a large housing preference shock to match boom-bust in U.S. housing value. This implies a boom-bust in rent-income ratio, which is counterfactual. One-period mortgage debt: Implies rapid deleveraging after 2007 as housing values fell. Hence, the model requires an implausible, post-2007 relaxation of lending standards ( m t ) to explain the slow deleveraging observed in the data. All models: Mortgage interest rate decline was not an important driver of the U.S. boom-bust episode.

46 Conclusion RE model: Requires a large housing preference shock to match boom-bust in U.S. housing value. This implies a boom-bust in rent-income ratio, which is counterfactual. One-period mortgage debt: Implies rapid deleveraging after 2007 as housing values fell. Hence, the model requires an implausible, post-2007 relaxation of lending standards ( m t ) to explain the slow deleveraging observed in the data. All models: Mortgage interest rate decline was not an important driver of the U.S. boom-bust episode. Credit-fueled bubble: Model with (1) moving-average forecast rules, (2) long-term mortgages, and (3) shifting lending standards does best in plausibly matching the U.S. data.

47 Impulse response: Housing preference shock All model versions hit with the same shock.

48 Impulse response: Lending standard shock All model versions hit with the same shock.

49 Impulse response: Income growth shock All model versions hit with the same shock.

50 Impulse response: Interest rate shock All model versions hit with the same shock.