Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
BETTING THE HOUSE Òscar Jordà? Moritz Schularick† Alan M. Taylor § ? Federal Reserve Bank of San Francisco; University of California, Davis † University § University of Bonn; CEPR of California, Davis; NBER; CEPR Disclaimer: The views expressed herein are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Federal Reserve Bank of San Francisco or the Board of Governors of the Federal Reserve System. 1/26 House prices in the news Are you worried about side effects of loose monetary policy? 2/26 Broad context Should we worry about mortgage and house price booms? Is there a tension between monetary and macro prudential policies? Then and now... ECB: Spain/Ireland versus Core Fed: Bernanke (“Monetary Policy and the Housing Bubble”) Norges Bank, Bank of Israel Riksbank versus Lars Svensson ECB decision not to buy real estate loans in TLTRO Bank of England versus Help to Buy Australia, Canada (?) 3/26 Focused questions These are rare events, so we turn to large-sample historical evidence, rather than rely on theory or small-samples... 1 2 Do low interest rates cause booms in mortgage lending? and in house prices? Do these booms in turn increase the odds of a financial crash? 4/26 What we do Two novel historical datasets: panel annual data on disaggregated lending + house prices The trilemma creates a natural experiment: exchange rate peg + open capital markets → exogenous fluctuations in monetary conditions Local projections + IV: to measure the effect of exogenous perturbations in monetary conditions on mortgage lending and housing prices Machine learning tools: use methods of binary classification to see if mortgage lending and housing prices affect likelihood of financial crises 5/26 What we find Answers are yes and yes; but effects are amplified after WW2 1 2 3 Mortgage lending ↑ After WW2, mortgage lending and home ownership rise strongly. Mortgages became dominant share of bank lending: from 1⁄3 to 2⁄3 Interest rates → mortgages & house prices Exogenous loose monetary conditions result in: (1) lower interest rates; (2) more mortgage lending; and (3) higher house prices Mortgages & house prices → financial crises Increased mortgage lending and house price booms are associated with an increased likelihood of financial crises 6/26 New annual data, N = 17, 143 years Real estate credit: Jordà, Schularick and Taylor (2014) House prices: Knoll (2014) Country Australia Belgium Canada Switzerland Germany Denmark Spain Finland France U.K. Italy Japan Netherlands Norway Portugal Sweden U.S. Mortgage loans 1870–2011 1885–2011 1874–2010 1870–2011 1883–2011 1875–2010 1904–2012 1927–2011 1870–2010 1880–2011 1870–2012 1893–2011 1900–2011 1870–2010 1920–2012 1871–2011 1896–2011 House prices 1870–2012 1878–2012 1921–2012 1900–2012 1870–2012 1875–2012 1970–2012 1905–2012 1870–2012 1899–2012 1970–2012 1913–2012 1870–2012 1870–2012 —— 1870–2012 1890–2012 Type of house price index Median price; partly mix-adj. Median price; partly mix-adj. Avg. prices Avg. prices; partly mix-adj. Avg. prices; partly mix-adj. Avg. prices; SPAR OECD after 1970 only Av. sq. m. price; partly mix-adj. hed. Repeat sales; partly mix-adj. hed. Avg. prices; partly mix-adj. OECD after 1970 only Avg. prices; partly mix-adj. Repeat sales; partly SPAR Repeat sales; hedonic No data Repeat sales; hedonic Repeat sales; partly mix-adj. 7/26 House prices Real CPI deflated BEL CAN CHE DEU DNK ESP FIN FRA GBR ITA JPN NLD NOR PRT SWE USA -2 -4 2 0 -2 -4 2 0 -2 -4 2 0 -2 -4 1900 1950 2000 -2 0 2 1850 -4 log real house price index (1990=0) 0 2 -4 -2 0 2 AUS 1850 1900 1950 2000 1850 1900 1950 2000 8/26 Bank mortgage and non-mortgage lending Ratio of bank lending to GDP .2 .4 .6 .8 Relative to GDP Nonmortgage lending 0 Mortgage lending 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 9/26 49 49 55 52 Great Depression vs. Great Recession 44 27 26 20 16 15 16 14 5 4 0 12 100 100 AUS 88 BEL CAN CHE DEU DNK ESP FRA ITA NLD PRT USA FIN 1928GBR JPN NOR SWE 80 73 51 51 48 85 45 70 70 51 58 41 88 56 49 49 96 84 95 86 74 71 84 56 61 48 71 93 54 64 84 19 39 1970 68 67 81 55 52 44 59 27 30 26 46 44 42 15 30 BEL CAN CHE DEU DNK ESP 12 33 4 FIN FRA 16 36 14 16 29 5 GBR ITA 7 JPN NLD NOR 16 PRT SWE USA 0 AUS 61 39 29 0 12 32 52 49 20 100 100 AUS CAN BEL 68 DEU CHE ESP DNK FRA ITA NLD PRT USA FIN 1970GBR JPN NOR SWE 70 70 51 58 41 88 56 57 49 13 49 40 42 57 67 48 71 93 54 64 84 19 39 37 52 54 45 32 56 81 42 32 46 68 55 36 2007 51 53 87 61 59 52 49 42 32 49 30 43 30 51 44 60 58 51 43 63 33 47 29 48 FRA ITA 12 46 68 58 44 16 7 0 AUS CAN DEU CHE ESP DNK FIN GBR NLD JPN PRT NOR USA SWE 0 BEL AUS 100 Realof estate and other lending, percent of total bank lending Real estate and other lending, percent total bank lending Share of mortgages in total lending 39 29 51 BEL 57 CAN 49 CHE 13 87 DEU 49 DNK 40 ESP 42 FRA ITA NLD PRT USA FIN 2007GBR JPN NOR SWE 57 53 37 Real estate 60 58 52 54 45 32 56 42 32 Other 68 63 55 68 58 10/26 Lending is now mostly about mortgages .1 Ratio of real estate lending to total lending .3 .2 .4 .5 .6 Share of real estate in total lending (17 country average) 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 11/26 Summary of main developments Post-WII total bank lending as a ratio to GDP doubled relative to pre-WWII Post-WWII tripling of real estate lending, specially mortgages to households Mortgage lending is about 2/3 of the loan book of banks 12/26 A European parable Germany versus Spain and Ireland: Interest rates, mortgages, house prices 0 5 10 15 (a) Taylor rule and actual policy interest rates (% per year) 1999q1 2000q1 2001q1 2002q1 Ireland 2003q1 2004q1 Spain 2005q1 2006q1 Germany 2007q1 2008q1 ECB policy rate 20 40 60 80 100 (b) Mortgage lending to GDP ratio (%) 1999 2000 2001 2002 2003 Ireland 2004 Spain 2005 2006 2007 2008 2007q1 2008q1 Germany 80 100 120 140 160 180 (c) House price to income ratio (1999q1=100) 1999q1 2000q1 2001q1 2002q1 Ireland 2003q1 2004q1 Spain 2005q1 2006q1 Germany 13/26 Methods: Local Projections + IV Usual LP: y = response vbl., r = impulse vbl. (s.t. interest rate) ∆h yit−1 = αhi + βh ∆rit + ∆Wit Γh + ∆Xit−1 Φh + uit+h Selection on observables: ∆Wit : all contemporaneous variables except y and r ∆Xit−1 : lags of all variables, including y and r Selection on unobservables: Now use IV for ∆rit using zit instrument But what instrument?... 14/26 The trilemma provides the instrument Trilemma provides a natural experiment to identify exogenous perturbations to home short-term interest rate ∆rit = a + b[PEGit × KOPENit × ∆rit∗ ] + uit Data sources: Obstfeld/Shambaugh/Taylor, Ilzetzki/Reinhart/Rogoff, Quinn, JST Base country UK (gold standard/BW base) UK/US/France composite (gold standard base) USA (BW/Post-BW base) Germany (EMS/ERM/ Eurozone base) Pre-WW1 All countries Interwar Bretton Woods Sterling bloc: AUS, CAN Post-BW All other countries Dollar bloc: AUS, CAN, CHE JPN, NOR All countries All other countries 15/26 Base and home short-rates are correlated Bivariate scatters ∆rit = a + b[PEGit × KOPENit × ∆rit∗ ] + uit Obstfeld/Shambaugh/Taylor simulations: b < 1 unless peg is ultra-hard (no band) (b) Post-WW2 Change in local short-term interest rate -5 0 5 -10 -4 Change in local short-term interest rate -2 0 2 4 10 (a) Pre-WW2 -2 -1 0 1 2 Change in base short-term interest rate * PEG * KOPEN -4 -2 0 2 4 Change in base short-term interest rate * PEG * KOPEN 16/26 The instrument is relevant Add vector of macro controls X ∆rit = a + b[PEGit × KOPENit × ∆rit∗ ] + ΘXit + uit Dependent variable: short-term interest rate (1) All Years No Controls (2) (3) PrePostWW2 WW2 (4) All Years Controls (5) PreWW2 (6) PostWW2 b̂ 0.68∗∗∗ (0.06) 0.36∗∗∗ (0.11) 0.81∗∗∗ (0.06) 0.43∗∗∗ (0.04) 0.29∗∗∗ (0.06) 0.46∗∗∗ (0.06) F-statistic Observations 150.17 1875 11.59 876 169.51 999 37.16 1220 9.26 375 29.84 845 Notes: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Country-based cluster-robust standard errors in parentheses. The dependent variable is the short-term interest rate regressed on the instrument, fixed effects and when appropriate, controls. The set of controls includes: (i) the change in short-term interest rate; (ii) the change in long-term interest rate; (iii) the change in mortgages to GDP ratio; (iv) the change in real house prices as a ratio to per capita income; (v) real per capita GDP growth; (vi) the change in the investment to GDP ratio; (vii) the change in the ratio of non-mortgage loans to GDP ratio; (viii) CPI inflation; and (ix) the current account to GDP ratio. We include contemporaneous terms and two lags. The full sample starts in 1870 and ends in 2010. The pre-WW2 sample ends in 1938. The post-WW2 sample begins in 1946. World Wars omitted from all samples. See text. 17/26 LP-IV estimated responses Baseline results: full control set, sample is all years Responses ∆h Short-term interest rate Kleibergen-Paap ∆h Long-term interest rate Kleibergen-Paap ∆h Mortgage Loans/GDP Kleibergen-Paap ∆h log (House Price/Income) Kleibergen-Paap Year h =0 1.00 Year h =1 1.31∗∗∗ (0.16) Year h =2 1.02∗∗∗ (0.19) Year h =3 0.80∗∗∗ (0.19) Year h =4 0.39∗∗∗ (0.14) 0.42∗∗∗ (0.05) 26.64 0.55∗∗∗ (0.09) 26.59 0.67∗∗∗ (0.13) 26.43 0.60∗∗∗ (0.15) 27.10 0.39∗∗∗ (0.13) 35.58 -0.45∗∗∗ (0.15) 35.24 -1.19∗∗∗ (0.38) 35.29 -1.87∗∗∗ (0.61) 34.66 -2.35∗∗∗ (0.76) 35.21 -2.82∗∗∗ (0.86) 28.44 -0.18 (0.79) 28.08 -1.76 (1.67) 27.90 -3.72∗ (2.05) 27.97 -5.02∗∗ (2.27) 28.49 -4.37∗∗ (1.88) 27.65 27.23 27.01 27.01 27.53 Notes: ∆h denotes change from year t − 1 to t + h. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Country-based cluster-robust standard errors in parentheses. Coefficient estimates of fixed effects and controls not reported. The set of controls includes: (i) the change in short-term interest rate; (ii) the change in long-term interest rate; (iii) the change in mortgages to GDP ratio; (iv) the change in real house prices as a ratio to per capita income; (v) real per capita GDP growth; (vi) the change in the investment to GDP ratio; (vii) the change in the ratio of non-mortgage loans to GDP ratio; (viii) CPI inflation; and (ix) the current account to GDP ratio. We include contemporaneous terms and two lags. The full sample starts in 1870 and ends in 2010. Kleibergen and Paap (2006) statistic for weak instruments reported. World Wars omitted. See text. 18/26 LP-IV estimated responses “Boom experiment”: exogenous perturbation in short-term rate of −100 bps -2 -1.5 Percentage points -1 -.5 Percentage points -1.5 -1 -.5 0 (b) Long-term interest rate 0 (a) Short-term interest rate 0 1 2 Year 3 4 0 1 2 Year 3 4 (d) log (House price/income ratio) 0 -5 1 0 5 100 x change in log Percentage points 2 3 4 5 10 (c) Mortgage loans/GDP ratio 0 1 2 Year 3 4 0 1 2 Year 3 4 19/26 LP-IV estimated responses Robustness check 1: OLS versus IV (attenuation bias) Responses ∆h Short-term interest rate ∆h Long-term interest rate ∆h Mortgage loans/GDP OLS Year h =0 1.00 Year h =1 0.69∗∗∗ (0.06) Year h =2 0.45∗∗∗ (0.04) Year h =3 0.38∗∗∗ (0.05) Year h =4 0.35∗∗∗ (0.07) IV 1.00 1.02∗∗∗ (0.19) 0.32∗∗∗ (0.04) 0.80∗∗∗ (0.19) 0.26∗∗∗ (0.03) 0.39∗∗∗ (0.14) 0.26∗∗∗ (0.03) OLS 0.34∗∗∗ (0.03) 1.31∗∗∗ (0.16) 0.33∗∗∗ (0.04) IV 0.42∗∗∗ (0.05) -0.11∗∗∗ (0.04) 0.55∗∗∗ (0.09) -0.15∗∗ (0.06) 0.67∗∗∗ (0.13) -0.25∗∗∗ (0.08) 0.60∗∗∗ (0.15) -0.29∗∗ (0.11) 0.39∗∗∗ (0.13) -0.45∗∗∗ (0.15) -0.45∗∗∗ (0.15) 0.35 (0.33) -1.19∗∗∗ (0.38) 0.15 (0.40) -1.87∗∗∗ (0.61) -0.33 (0.48) -2.35∗∗∗ (0.76) -0.67 (0.51) -2.82∗∗∗ (0.86) -0.90 (0.56) -0.18 (0.79) -1.76 (1.67) -3.72∗ (2.05) -5.02∗∗ (2.27) -4.37∗∗ (1.88) OLS IV ∆h log House prices/income OLS IV 20/26 LP-IV estimated responses Robustness Check 2: Excluding controls (attenuation also) Responses ∆h Short-term interest rate Kleibergen-Paap ∆h Long-term interest rate Kleibergen-Paap ∆h Mortgage loans/GDP Kleibergen-Paap ∆h log House prices/income Kleibergen-Paap Year h =0 1.00 Year h =1 1.34∗∗∗ (0.08) Year h =2 1.08∗∗∗ (0.09) Year h =3 0.91∗∗∗ (0.11) Year h =4 0.76∗∗∗ (0.13) 0.40∗∗∗ (0.05) 65.14 0.55∗∗∗ (0.07) 65.01 0.64∗∗∗ (0.10) 64.50 0.61∗∗∗ (0.12) 64.43 0.49∗∗∗ (0.11) 70.42 -0.20∗∗∗ (0.07) 70.00 -0.54∗∗∗ (0.18) 69.45 -0.85∗∗∗ (0.31) 69.74 -1.11∗∗∗ (0.39) 69.67 -1.41∗∗∗ (0.49) 64.50 -0.06 (0.52) 64.44 -0.81 (1.02) 64.09 -2.00 (1.26) 64.06 -2.87∗∗ (1.40) 63.64 -2.96∗∗ (1.22) 72.01 71.98 71.98 72.03 71.69 21/26 LP-IV estimated responses Robustness check 3: Subsample sensitivity, including control set, year-4 Responses in year 4 ∆4 Short-term interest rate All Years 0.39∗∗∗ (0.14) PreWW2 0.36∗∗ (0.18) PostWW2 0.31∗ (0.18) Set z = 0 1946–72 0.36∗∗ (0.16) Exclude 1946–72 0.39∗∗ (0.16) Exclude 1914–72 0.30 (0.23) Kleibergen-Paap ∆4 Long-term interest rate 27.10 0.39∗∗∗ (0.13) 9.67 0.40∗∗∗ (0.15) 26.59 0.24 (0.15) 27.84 0.40∗∗∗ (0.14) 25.07 0.41∗∗∗ (0.14) 21.16 0.36∗ (0.18) Kleibergen-Paap ∆4 Mortgage loans/GDP 35.21 -2.82∗∗∗ (0.86) 8.89 -1.94 (1.32) 33.23 -2.67∗∗∗ (0.91) 34.07 -2.95∗∗∗ (0.95) 32.21 -3.10∗∗∗ (0.93) 25.27 -3.47∗∗∗ (1.06) Kleibergen-Paap ∆4 log House prices/income 28.49 -4.37∗∗ (1.88) 9.49 -1.34 (4.82) 28.29 -5.37∗∗ (2.12) 29.48 -4.66∗∗ (2.14) 26.52 -4.38∗∗ (2.04) 22.84 -5.88∗∗∗ (2.25) Kleibergen-Paap 27.53 7.99 24.99 27.92 25.25 20.01 22/26 Financial crisis prediction Binary classification based on mortgage lending and house prices (a) Logit models, country fixed effects Mortgage loans/GDP, lagged 5-year change (1) All years 0.17∗∗ (0.08) (2) Pre-WW2 0.11 (0.14) (3) Post-WW2 0.26∗∗ (0.10) log (House prices/income), lagged 5-year change 0.07∗∗ (0.03) 0.08 (0.05) 0.07 (0.05) 1275 415 860 0.66 (0.04) 0.63 (0.06) 0.71 (0.06) 0.53 (0.03) 0.53 (0.03) 0.54 (0.06) 0.02 0.17 0.02 Observations (b) Correct classification frontier statistics Model AUC Benchmark AUC, country fixed effects only H0 : AUC model = AUC benchmark (p-value) 23/26 Financial crisis prediction 0.00 True positive rate 0.25 0.50 0.75 1.00 Binary classification based on mortgage lending and house prices 0.00 0.25 0.50 True negative rate Benchmark AUC = 0.53 (0.03) Reference 0.75 1.00 Model AUC = 0.66 (0.04) 24/26 Conclusions Patterns that emerge strongly after WW2... 1 2 3 Changing role of banking: Mortgage lending share from 1⁄3 to 2⁄3 Fluctuations over the business cycle: short rates → mortgages + house prices Predicting financial crises: mortgages + house prices → financial crises 25/26 Background papers available in working paper form Betting the House. Just discussed. The Great Mortgaging: Discussion of new historical dataset of credit broken down by mortgages and non-mortgages. Long-run trends and business cycle fluctuations. Financial crises and type of credit buildups. 26/26