## 3.1 Data ## Source details **Canonical URL:** [3.1 Data](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020011-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020011-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020011-print-pdf.pdf.json) --- ### Importance of housing and motivation - Housing serves both as a long-term investment and as a consumption good that generates considerable utility for households. - Housing consumption and investment accounted for about one sixth of the GDP in the US and euro area economies in 2017. - Large drops in house prices reduce households’ net worth and can reduce consumption; mortgages and housing-related lending make up a large fraction of banks’ assets, making sudden house price declines a risk to banks’ portfolio quality, profitability, and stability. - During the COVID-19 pandemic crisis, the housing market has been exceptionally resilient, sustained by continued accommodative monetary policy and strong demand for new homes. - Sustained rapid growth in house prices can create expectations of continued rises, potentially leading to excessive risk-taking and rising vulnerabilities. - As central banks tighten monetary policy in response to post-pandemic inflation, identifying the size of future downside risks to house prices and implications for financial stability is crucial. ### House-Prices-at-Risk (HaR): definition and methodology - HaR definition and implementation: - HaR uses the 5th percentile of the conditional house price growth distribution to capture tail downside risk (τ = 5 percent). - Implemented using panel quantile regressions and local projections to explore risk evolution over forecast horizons h. - Out-of-sample performance is evaluated using only information available at or before the time of prediction; recursive out-of-sample estimates are reported as stable and effective for real-time prediction. - Estimation framework (two-step Canay (2011) approach): - First step: estimate unobserved fixed effects via within-estimators: - ∆hY_{i,t+h,τ} = α_{i,h,τ} + β_{h,τ} X_{i,t} + e_{i,t,h,τ} - Remove fixed effects: ∆hȲ_{i,t+h,τ} ≡ ∆hY_{i,t+h,τ} − ˆα_{i,h,τ} - Second step: quantile regression of ∆hȲ_{i,t+h,τ} on X_{i,t}: - ˆβ(τ) ≡ argmin E_{nT} ρ_τ(∆hȲ_{i,t+h} − X′_{i,t} β_{h,τ}) - Predicted conditional quantile: ˆQ_{i,t+h|x_i,t}(τ) = X_{i,t} ˆβ - Inference: pairs-bootstrap used to account for heteroskedasticity when computing standard errors. - Key regressors X: - Past log changes in real house prices, GDP growth, credit boom indicator, FCI (financial conditions index), and house price misalignment (overvaluation). - Credit boom indicator: cyclical deviation of credit-to-GDP above HP trend (smoothing parameter 1600); indicator = 1 when cyclical component > 0. - Overvaluation (baseline): deviation of price-to-GDP per capita ratios from an estimated linear trend. ### Baseline empirical findings on drivers of HaR - General patterns: - HaR broadly respond to past price dynamics and fundamental factors; lagged house prices capture momentum and persistence. - Fundamental factors typically have stronger effects on the left tail (5th percentile) than at the median. - Financial conditions (FCI): - A one-standard-deviation tightening of financial conditions is associated with 0.3 to 0.7 percentage point higher downside risk to house prices in the short term (stronger impact in emerging market economies). - Over longer horizons, impact diminishes to 0.1 percentage point in advanced economies and becomes insignificant for emerging market economies. - If overvaluation measures are excluded, medium-term association between financial conditions and HaR becomes positive. - Comparative shock sizes: global financial crisis entailed a 2.3 standard deviation shock to financial conditions in advanced economies and 1.4 standard deviations in emerging market economies. - Real GDP growth: - A one-standard-deviation higher real GDP growth does not significantly reduce downside risks one to three quarters ahead in advanced economies, but shows an opposite and significant relationship over longer horizons in advanced economies. - In emerging market economies, association is positive but not statistically significant. - Comparative shock sizes: GDP growth shock was 2.2 standard deviations in advanced economies and 1.7 standard deviations in emerging market economies. - Overvaluation (house price misalignment): - A one-standard-deviation higher price misalignment ratio is linked to 0.5 to 1.0 percentage point increase in downside risks to house prices in advanced economies and 0.7 to 1.1 percentage point increase in emerging market economies. - Overvaluation shock magnitude referenced as about 0.2 standard deviation across both groups. - Credit booms: - Credit booms worsen HaR by up to 0.5 percentage points at short horizons (three quarters ahead) in advanced economies and up to 1.1 percentage point at medium-term horizons (up to eight quarters ahead) in emerging market economies. - Left-tail vs median: - Shocks to house price-to-GDP ratio and credit booms show stronger relationships with downside HaR than with median house price outcomes. ### Quantified links between HaR, financial crises, and growth-at-risk (GaR) - HaR to financial crisis probabilities: - A negative 12 percent on the HaR gauge — corresponding to a 5 percent probability of at least a 12 percent drop in house prices — indicates: - a 31 percent probability of a financial crisis two years later in advanced economies, and - a 10 percent probability of a financial crisis two years later in emerging markets. - The highest impact of HaR on financial stability is four to eight quarters into the future. - HaR to GaR: - GaR defined as the conditional growth at the lower 5th percentile of the GDP growth distribution. - A 1 percentage point decline in the HaR measure precedes on average a 0.3 percentage point decline in growth at risk (GaR) (paper reports 0.4 percentage point for advanced economies and 0.5 percentage point for emerging market economies in other specifications). - Growth-at-Risk regression: ∆h yi,t+h,τ = αi,h,τ + θh,τ yi,t + βh,τ HaRt+h i,t,τ + λh,τ FCIi,t + ei,t,h,τ - Largest impact of HaR on GaR is four quarters ahead. - Crisis-prediction enhancement: - Adding HaR to standard crisis-prediction models (GDP growth, FCI, credit-to-GDP gap) improves accuracy across horizons of one, two, and three years for both advanced and emerging market economies. - Example: an annual HaR of −16 percent (an estimated 5 percent probability of a 12 percent decline in real house prices one year ahead) implies: - a 34 percent probability of a financial crisis one year ahead in advanced economies. - a 25 percent probability of a financial crisis one year ahead in emerging market economies. ### Out-of-sample and robustness evidence - Real-time forecasting setup: - Recursive estimation with earliest out-of-sample start dates in 2006; forecast horizons h = 1, 4, 8, 12 quarters. - Out-of-sample projections of the left tail track well in-sample predictions for the United States across h = 1,4,8,12 with generally very small differences. - Quantile R^2 and predictive accuracy (United States, 5th percentile): - Panel-Based Estimates Pseudo-R2: H=1 → 59.96; H=4 → 48.09; H=8 → 38.18; H=12 → 47.77. - Country-Level Estimates Pseudo-R2: H=1 → 80.07; H=4 → 72.66; H=8 → 86.41; H=12 → 83.78. - Robustness checks: - Alternative panel quantile estimators (Powell 2022; Machado and Silva 2019) produce broadly consistent signs and magnitudes. - Excluding 2008–2009 does not qualitatively change results. - Alternative definitions tested: credit boom definitions, FCI from time-varying VAR, HP filters for misalignment, price-to-rent misalignment — results are quantitatively similar. ### Data coverage and stylized facts - Data coverage: - House price data for 22 major advanced economies and 10 emerging market economies; quarterly data typically go back to 1990:Q1 in advanced economies; EM series generally start later. - Nominal house prices deflated using overall CPI when necessary. - Stylized facts: - Advanced economies: historical average (annualized) one-year and three-year real house price growth about 2 percent a year. - Emerging market economies: historical average (annualized) one-year and three-year real house price growth about 2.6 percent a year. - Negative real house price growth occurs in about half of the observations in advanced economies and in a third of the observations in emerging market economies over a one-year horizon. - House prices tend to co-move during crises; some countries experienced steep declines (up to 20 percent) in early 1990s and widely negative growth in 2008–2009. - Key data constructions: - Financial Conditions Index (FCI): principal component of 11 macrofinancial variables; FCI purged of house price growth variation by regressing on quarterly changes in real house prices with country-fixed effects and using residuals as "FCI purged". - Credit boom: cyclical deviation of credit-to-GDP above HP trend (smoothing parameter 1600). *Source: IMF Working Paper (content unit: 3.1 Data, sections 1–2 and related introductory material)* ### 3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 ### 3.1 Data ### Importance of housing and motivation - Housing serves both as a long-term investment and as a consumption good that generates considerable utility for households. - Housing consumption and investment accounted for about one sixth of the GDP in the US and euro area economies in 2017. - Large drops in house prices reduce households’ net worth and can reduce consumption; mortgages and housing-related lending make up a large fraction of banks’ assets, making sudden house price declines a risk to banks’ portfolio quality, profitability, and stability. - During the COVID-19 pandemic crisis, the housing market has been exceptionally resilient, sustained by continued accommodative monetary policy and strong demand for new homes. - Sustained rapid growth in house prices can create expectations of continued rises, potentially leading to excessive risk-taking and rising vulnerabilities. - As central banks tighten monetary policy in response to post-pandemic inflation, identifying the size of future downside risks to house prices and implications for financial stability is crucial. ### House-Prices-at-Risk (HaR): definition and methodology - The paper proposes a novel non-parametric approach to predict downside risks to house prices, termed house-prices-at-risk (HaR). - HaR uses the 5th percentile of the conditional house price growth distribution to capture tail downside risk. - The approach is implemented using panel quantile regressions and local projections to explore risk evolution over forecast horizons. - Out-of-sample performance is evaluated using only information available at or before the time of prediction; recursive out-of-sample estimates are reported as stable and effective for real-time prediction. ### Key empirical findings on drivers of HaR - The left tail of the house price growth distribution is mainly driven by fundamental factors: financial conditions, GDP growth, credit growth, and house price overvaluation. - A one-standard-deviation tightening of financial conditions is associated with an up to 0.7 percentage point higher downside risk to house prices in the short term. - Credit booms exacerbate the incidence of large negative house price corrections at short- and medium-term horizons by up to 0.5 percentage points. - The model and quantile regressions show that current house price overvaluation, excessive credit growth, and tighter financial conditions jointly forecast higher HaR up to three years ahead. - The HaR measure can predict future financial crises and economic downturns. ### Quantified links between HaR, financial crises, and growth-at-risk (GaR) - A negative 12 percent on the HaR gauge — corresponding to a 5 percent probability of at least a 12 percent drop in house prices — indicates: - a 31 percent probability of a financial crisis two years later in advanced economies, and - a 10 percent probability of a financial crisis two years later in emerging markets. - The highest impact of HaR on financial stability is four to eight quarters into the future. - A 1 percentage point decline in the HaR measure precedes on average a 0.3 percentage point decline in growth at risk (GaR). - GaR is defined as the conditional growth at the lower 5th percentile of the GDP growth distribution. - Forecasts illustrated include a one-year predictive horizon for a panel of 22 major advanced economies and a panel of 10 emerging market economies. ### Theoretical framework and policy channels - The macroeconomic model describes housing crises as a vicious cycle of real GDP and house price declines driven by binding collateral constraints when household debt is high. - Mechanism: income declines → collateral constraint binds → fire sales → house price declines → tighter collateral constraint → further drops in aggregate demand and incomes. - Macroprudential measures, such as a Pigouvian tax on household debt, can alleviate negative effects by preventing tightening of collateral constraints from house price declines. - Monetary policy does not have direct effects on house price growth in distress other than through general financial conditions. ### Policy evaluation and empirical policy findings - The authors construct: - a proxy for the intensity of the macroprudential policy stance by combining information on tightening and loosening of interventions and purging variation due to credit-to-GDP, and - monetary policy shocks as residuals from a Taylor rule specification. - Empirical results: - Tightening macroprudential policies is associated with a reduction of downside risks to house prices. - Policies aimed at strengthening borrower resilience (limits on loan-to-value and debt-service-to-income ratios) are especially effective. - Monetary policy’s ability to mitigate downside risks beyond its effect on financial conditions is limited, and appears confined to the short term and to advanced economies. - Financial conditions, partly driven by monetary policy, have a clear relationship with downside risks to house prices. ### Contributions to literature and methodological positioning - Novelty: focus on the 5th percentile of the house price distribution as a real-time, tail-risk measure of house-prices-at-risk that does not rely on ex-post boom-and-bust classifications. - First study to quantify how fundamental factors predict house price tail risks and to quantify their implications for financial stability via GaR. - Methodological contributions build on Adrian et al. (2019) and Giglio et al. (2016) by utilizing panel quantile regression methods to examine primary drivers of downside risks in house prices and linking HaR to the lower tail of GDP growth. *Source: IMF Working Paper (content unit: 3.1 Data, sections 1–2 and related introductory material)* ### Section 5 concludes. ### Section 5 concludes. ### Theoretical framework and model setup - Model type: nonlinear dynamic stochastic general equilibrium (DSGE) with occasionally binding housing collateral constraints; follows Guerrieri and Iacoviello (2017) and focuses on household sector with two household types (borrowers and lenders). - Households choose consumption, leisure, and housing; housing is the only collateral for borrowing; house prices determined by a forward-looking asset pricing formula. - Borrower preferences and key parameters (as calibrated in the exercise): - Utility: E0 ∑t=0∞ βt [c1−σt/(1−σ) + χ log ht − ΨL1+ωt/(1+ω)] - Calibrated parameter values: β = 0.96; σ = 2.0; ω = 0.5; χ = 0.2; Ψ = 0.6. - Borrower budget constraint and collateral constraint: - Budget: ct + bt Rt + qt ht = wt Lt + bt−1 + qt ht−1. - Collateral constraint: bt Rt ≤ (1−γ) κ qt ht − γ bt−1. - Collateral tightness parameters: κ = 0.65; γ = 0.35. - Housing supply is fixed (ht = 1); equilibrium conditions include borrower Euler equation and house price asset pricing formula with Lagrange multiplier μt for collateral constraint and complementary slackness conditions. - Monetary policy: nominal interest rate Rt set by a Taylor rule: - Rt = R [πt/π*]φπ [yt/y*]φy with φπ = 1.5 and φy = 0.5. ### Model dynamics and crisis mechanism - Housing crises mechanism: - When household debt is high and income declines, the collateral constraint binds; households reduce borrowing and engage in "fire sales" (or foreclosures), causing house prices to fall, which tightens the collateral constraint further. - Binding collateral constraints force consumption reductions, lowering aggregate demand, output, wages, and household income, producing a vicious cycle of deteriorating conditions. - Model replication of crises: - The ergodic distribution of output gaps in the baseline simulation shows massive declines in output during housing crises (blue line in Figure 2a). - Positive association between initial household debt and incidence of housing crises: higher debt-to-GDP ratios significantly increase the probability of housing crises (Figure 2b). Median effects on house price growth are negligible or slightly positive, but tail risks rise. - Crisis definition used in policy evaluation: output growth less than −3 percent (declines in output of more than 3 percent). ### Policy experiments and quantitative findings - Three policy measures examined: 1. Macroprudential measures (MPMs): modeled as a Pigouvian tax on debt (state-contingent macroprudential tax ωt approximated as linear in household debt). 2. Monetary policy augmented by a response to household debt (central bank raises rates in run-up and lowers rates during deleveraging). 3. Monetary policy augmented by a response to credit spreads (central bank lowers rates in response to higher secured–unsecured interest rate gaps). - Social planner Euler equation comparison motivates the macroprudential tax design so borrower first-order conditions match the social planner's. - Key quantitative results from simulations (Figure 2 panels summarized): - MPMs (green line) significantly decrease variance of the output gap relative to baseline (blue line), especially by shrinking the left tail of the ergodic distribution—reducing the frequency and severity of crises. - During crisis periods (output decline > 3 percent): - MPMs mitigate declines in output growth and house price growth and lead to a significantly lower average level of household debt than baseline, both in run-up periods and in normal times. - Monetary policy responding to household debt: - Slightly mitigates adverse effects on output growth during crises but has no effect on house price growth in crisis periods. - Central bank lowers nominal interest rates more in crises (relative to baseline) because of deleveraging response; aggregate effect on house prices is negligible. - Overall policy effect is insignificant relative to MPMs; monetary policy is described as potentially too blunt for crisis management. - Monetary policy responding to credit spreads: - Slight mitigation of adverse effects on output growth in crises but no effect on house price growth in crises. - Requires pronounced declines in nominal interest rates during crises; feasibility may be limited in low-interest-rate environments. - Interpretation and caveats: - Augmented Taylor rule used is a specific linear response to household debt; other monetary policy rules may perform differently. - Monetary policy can be implemented more promptly than MPMs during crises, which may increase practical usefulness despite model results. - Surcharges on lending rates are mathematically equivalent to debt taxes in the model; differences in performance arise because monetary policy affects other interest rates (including deposit rates) as well. ### Data, empirical methodology, and stylized facts - Data coverage: - House price data for 22 major advanced economies and 10 emerging market economies; quarterly data typically go back to 1990:Q1 in advanced economies; EM series generally start later but efforts made to extend toward early 1990s. - Nominal house prices deflated using overall CPI when necessary. - Historical stylized facts: - Advanced economies: historical average (annualized) one-year and three-year real house price growth about 2 percent a year. - Emerging market economies: historical average (annualized) one-year and three-year real house price growth about 2.6 percent a year. - Negative real house price growth occurs in about half of the observations in advanced economies and in a third of the observations in emerging market economies over a one-year horizon. - House prices tend to co-move during crises; some countries are more cyclical than others (examples: early 1990s steep declines of up to 20 percent in some AEs; global financial crisis 2008–2009 widely negative). - Variables linked to downside risk in house prices: - Financial conditions index (FCI) constructed from a principal component analysis of 11 macrofinancial variables (real short-term rate, interbank spread, term spread, sovereign local debt spread, sovereign dollar debt spread, corporate local currency spread, corporate dollar debt spread, equity prices, equity price volatility, exchange rate and real house prices). An increase in the FCI represents a tightening of pricing of risk. - To mitigate endogeneity, FCI is purged of house price growth variation by regressing the index on quarterly changes in real house prices with country-fixed effects; the residuals are used as "FCI purged". - Empirical relationships across quantiles (Figure 4): - Tighter FCI is associated with lower one-year-ahead real house price growth, with strongest effects in the left tail (5th percentile). - Real GDP growth is generally associated with lower house price growth in the distributional analysis. - Credit-to-GDP ratio displays a negative relationship with house price growth when the ratio is above its long-term mean. - Price-to-GDP per capita ratio acts as a valuation metric; its association with future house price growth is markedly stronger for the left tail of the distribution relative to the median and 95th percentile. *Source: wpiea2020011-print-pdf - Section 5 concludes.* ### 3.2 Modeling House-Prices-at-Risk ### 3.2 Modeling House-Prices-at-Risk ### Definition and estimation approach - HaR definition: - HaR is the measure of downside risk for the growth of real house prices over a given horizon at a 5 percent probability, corresponding to the fifth percentile of the distribution. - Baseline focuses on the fifth percentile (τ = 5 percent) and varying horizons h to establish a term structure of house price risks. - Estimation framework: - Panel quantile regressions are used with separate panels for advanced economies and emerging market economies. - Two-step estimation following Canay (2011): - First step: estimate unobserved fixed effects via within-estimators: - ∆hY_{i,t+h,τ} = α_{i,h,τ} + β_{h,τ} X_{i,t} + e_{i,t,h,τ} - Remove fixed effects: ∆hȲ_{i,t+h,τ} ≡ ∆hY_{i,t+h,τ} − ˆα_{i,h,τ} - Second step: quantile regression of ∆hȲ_{i,t+h,τ} on X_{i,t}: - ˆβ(τ) ≡ argmin E_{nT} ρ_τ(∆hȲ_{i,t+h} − X′_{i,t} β_{h,τ}) - Predicted conditional quantile: ˆQ_{i,t+h|x_i,t}(τ) = X_{i,t} ˆβ - Under independence restrictions, ˆQ_{i,t+h|x_i,t} is a consistent linear estimator of the quantile function. - Key regressors (X): - Past log changes in real house prices, GDP growth, credit boom indicator, FCI (financial conditions index), and a house price misalignment (overvaluation) metric. - Credit boom indicator: cyclical deviation of credit-to-GDP above HP trend (smoothing parameter 1600); indicator = 1 when cyclical component > 0. - Overvaluation (baseline): deviation of price-to-GDP per capita ratios from an estimated linear trend. - Inference and standard errors: - Pairs-bootstrap adopted to account for remaining heteroskedasticity when computing standard errors of quantile regression estimates. - HaR formalized: - Pr(∆hˆY_{i,t+h,τ} ≤ HaR_{i,t,h}(τ|X_{i,t})) = τ - Varying h yields the term structure: sequence of β_τ coefficients across horizons h shows how a change in X affects the τ-th quantile of future house price growth. ### Baseline empirical findings (Section 4.1) - General: - HaR broadly respond to past price dynamics and fundamental factors; lagged house prices capture momentum and persistence. - Fundamental factors typically have stronger effects on the left tail (5th percentile) than at the median. - Financial conditions (FCI): - A one-standard-deviation tightening of financial conditions is associated with: - 0.3 to 0.7 percentage point higher downside risk to house prices in the short term (stronger impact in emerging market economies). - Over longer horizons, impact diminishes to 0.1 percentage point in advanced economies and becomes insignificant for emerging market economies. - If overvaluation measures are excluded, the medium-term association between financial conditions and HaR becomes positive (suggesting easy financial conditions raise future downside risks through current overvaluation). - Comparative shock sizes (reference context): global financial crisis entailed a 2.3 standard deviation shock to financial conditions in advanced economies (1.4 standard deviations in emerging market economies). - Real GDP growth: - A one-standard-deviation higher real GDP growth: - Does not significantly reduce downside risks one to three quarters ahead in advanced economies, but shows an opposite and significant relationship over longer horizons in advanced economies. - In emerging market economies, association is positive but not statistically significant. - Comparative shock sizes: GDP growth shock was 2.2 standard deviations in advanced economies and 1.7 standard deviations in emerging market economies. - Overvaluation (house price misalignment): - A one-standard-deviation higher price misalignment ratio is linked to: - 0.5 to 1.0 percentage point increase in downside risks to house prices in advanced economies. - 0.7 to 1.1 percentage point increase in emerging market economies. - Overvaluation shock magnitude referenced as about 0.2 standard deviation across both groups. - Credit booms: - Credit booms are related to a worsening of HaR by: - Up to 0.5 percentage points at short horizons (three quarters ahead) in advanced economies. - Up to 1.1 percentage point at medium-term horizons (up to eight quarters ahead) in emerging market economies. - Immediate effect likely due to the credit boom variable signaling overstretched household balance sheets instantaneously. - Left-tail vs median: - Shocks to house price-to-GDP ratio and credit booms show stronger relationships with downside HaR than with median house price outcomes. ### Country examples: United States and China (one-year-ahead HaR decomposition) - United States (one-year HaR decomposition highlights): - HaR deteriorated gradually from early 2000s leading to the global financial crisis. - Early deterioration mainly related to house price overvaluation; later negative contributions from past house price movements and credit. - Loose financial conditions partially offset negative contributors until the global financial crisis when tightening FCI weighed negatively. - Since late 2016, deterioration driven by overvaluation concerns and high credit growth, partly offset by easy financial conditions and past house price momentum. - China (one-year HaR decomposition highlights): - HaR more volatile and partly follows overall house price growth volatility. - Easy financial conditions contained house price risks until 2010. - After 2010, high credit-to-GDP gaps and tightening financial conditions increased downside risks. - Since 2016, house price overvaluation contributed to HaR deterioration. ### Out-of-sample evidence (Section 4.1.1) - Real-time forecasting setup: - Recursive estimation: forecasts for the 5th percentile at time t+h constructed using only data up to time t. - Earliest out-of-sample start dates are 2006 to allow sufficient observations. - Forecast horizons reported in examples: h = 1, 4, 8, 12 (quarters). - In-sample vs out-of-sample comparison: - Out-of-sample projections of the left tail (real-time/red line) track well in-sample predictions using the full sample (blue line) for the United States across h = 1,4,8,12. - Differences between real-time and full-sample predictions are generally very small. - Quantitative out-of-sample accuracy: - Quantile R^2 measure based on quantile loss function ρ_τ: - quantile R^2_i = 1 − (1/T) Σ_t ρ_τ(Y_{i,t+h} − ˆα_t − ˆβ_t X_{i,t}) / (1/T) Σ_t ρ_τ(y_{i,t+h} − ˆq_{i,τ}) - Higher quantile R^2 indicates larger improvement over historical unconditional quantile forecast; can be negative if unconditional quantile is better. - Statistical significance of quantile R^2 assessed by comparing sequences of forecast losses following Diebold and Mariano (2002). - For the United States, Table A.3 reports quantile R^2 and t-statistics at h = 1,4,8,12 using panel quantile forecasts and a country-level model. - Overall finding: HaR have significant out-of-sample predictive power for the 5th percentile of future house price growth using both panel and country-level quantile regressions. - Note: country-level models can differ in level; panel-based estimates are relatively more precise within first four quarters, while country-level models may outperform in long-term projections. ### Robustness checks (Section 4.1.2) - Alternative estimators: - Tested panel quantile estimators based on Powell (2022) and Machado and Silva (2019): - Powell (2022): method consistent for small T permitting nonadditive fixed effects with nonseparable disturbance term (avoids estimating fixed effects directly). - Machado and Silva (2019): allow individual fixed effects to affect entire distribution; jackknife bias correction allows inference for moderate T. - Coefficients from these alternative estimators reported in Appendix A.4 (Tables A.15 and A.16). - Crisis-period exclusion: - Tested sensitivity by dropping 2008 to 2009 (peaks of financial crisis) to check for overfitting from simultaneous cross-country declines; results in Table A.17. - Panel vs country-level comparisons: - Compared baseline panel quantile estimator with country-level estimated model for the United States. - Panel-based predictions remain close to country-level estimates, especially in the medium term. - Panel-based HaR tend to anticipate large house price declines (e.g., during the GFC) at shorter prediction horizons. - Alternative definitions of key determinants: - Tested: - i) alternative credit boom definition based on Mendoza and Terrones (2014); - ii) an FCI from a time-varying parameter VAR per Koop and Korobilis (2014); - iii) HP filters for house price misalignment construction; - iv) house price misalignment based on price-to-rent ratios. - Results reported in Appendix A.4 (Tables A.18–A.21). - Overall: robustness tests are quantitatively similar and qualitatively unchanged from the baseline model. *Source: wpiea2020011-print-pdf - 3.2 Modeling House-Prices-at-Risk* ### 4.2 Contribution of House-Prices-at-Risk to Macro-Financial Sta- ### 4.2 Contribution of House-Prices-at-Risk to Macro-Financial Stability ### Empirical model - Growth-at-risk (GaR) regression: - Model: ∆h yi,t+h,τ = αi,h,τ + θh,τ yi,t + βh,τ HaRt+h i,t,τ + λh,τ FCIi,t + ei,t,h,τ - ∆h yi,t+h,τ: average GDP growth h quarters ahead. - HaRt+h i,t,τ: estimated house-prices-at-risk measure h quarters ahead at time t. - Financial crisis probability model: - Fixed effects logit: (PrYi,t+h = 1) = Λ(X′t β) = eXtβ / (1 + eXtβ) - Yi,t+h equals 1 if economy i is experiencing a systemic banking crisis h quarters ahead. - Controls include output growth, financial conditions index (FCI), and credit-to-GDP gap. ### Findings on growth-at-risk (GaR) - Direction and significance: - An increase in downside risks to house prices (a lower, more negative HaR) is associated with an increase in future downside risk to GDP growth (i.e., GaR). - The association with downside risks is stronger than with median growth. - Magnitude and timing: - The largest impact of HaR is four quarters ahead. - A 1 percent deterioration (i.e. more negative) in the house-prices-at-risk measure precedes, on average, a 0.4 percentage point decline in growth at risk for advanced economies. - A 1 percent deterioration precedes, on average, a 0.5 percentage point decline in growth at risk for emerging market economies. - Robustness: - The association is robust to adding various credit quantity measures to the GaR model. - The association is robust when adding other measures of house price imbalances, such as growth in house prices or overvaluation metrics. - Interpretation: - HaR serves as a leading indicator for financial stability risks as captured by the GaR model. ### Findings on crisis prediction - Predictive improvement: - Adding HaR to standard crisis-prediction models (GDP growth, FCI, credit-to-GDP gap) improves model accuracy across horizons of one, two, and three years and for both advanced and emerging market economies. - Example quantified scenario: - An annual HaR of −16 percent—that is, an estimated 5 percent probability of a 12 percent decline in real house prices one year ahead—implies: - a 34 percent probability of a financial crisis one year ahead in advanced economies. - a 25 percent probability of a financial crisis one year ahead in emerging market economies. ### Mechanisms, historical context, and auxiliary findings - Consistency with previous literature: - Results align with Claessens et al. (2012): recessions are deeper and last longer when house prices fall more and more quickly; more than two-thirds of the nearly 50 systemic banking crises in recent decades were preceded by boom-bust patterns in house prices. - Housing-market characteristics that elevate crisis risk: - Higher loan-to-value ratios. - Greater reliance on wholesale funding markets. - Theoretical mechanism: - Interactions between house prices and credit volumes can create self-reinforcing feedback loops: - Rising house prices → expansion in credit via collateral effects → further upward pressure on house prices. - Reversal: large house price declines → collapse in credit and GDP growth → higher probability of financial crisis. ### Implications for financial stability surveillance - HaR is a useful forward-looking indicator: - It complements standard macro-financial indicators (GDP growth, FCI, credit-to-GDP gap) in forecasting GaR and systemic banking crises. - Its predictive value holds across advanced and emerging market economies and multiple forecast horizons. - Policy use: - Monitoring HaR can improve early-warning systems for downside risks to growth and banking-sector stress, particularly when combined with indicators of credit conditions and housing-market-specific vulnerabilities. *Source: IMF working paper section 4.2, "Contribution of House-Prices-at-Risk to Macro-Financial Stability."* ### 4.3 Policies to Mitigate Downside Risks to House Prices ### 4.3 Policies to Mitigate Downside Risks to House Prices ### Model and empirical strategy - Expanded forecasting equation (equation (18)): - Δh_y_{i,t+h,τ} = α_{i,h,τ} + β_{h,τ} X_{i,t} + λ_{h,τ} M_{i,t} + e_{i,t,h,τ} - M_{i,t} is the proxy for policy measures; X_{i,t} denotes all other variables. - β_{h,τ} captures the marginal effects of policy tightening itself. - λ_{h,τ} captures the policy effects conditional on other variables X_{i,t} (how the policy measure mitigates the marginal effects of X_{i,t} on HaR). - Interaction of M_{i,t} with the financial condition index (FCI) is included among controls to mitigate confounding. - Confidence intervals reported in figures are 90 (68) percent. ### Macroprudential policy measures: construction and identification - Data source: IMF’s Integrated Macroprudential Policy (iMaPP) database. - Two borrower-based macroprudential interventions targeted at housing: - Caps on loan-to-value (LTV) ratios for mortgage loans. - Debt-to-income (DTI) or debt-service-to-income (DSTI) limits. - Quarterly aggregation: - If no action in a quarter → PM = 0. - Multiple actions in same quarter → sum changes over the quarter. - Tightening = +1, loosening = -1. - PM range in a given quarter can be {−2,−1,0,1,2,3}. - Intensity measure (MPM) to capture cumulative binding effects over time: - MPM_{i,t} = Sum_{j=1}^{16} PM_{i,t−j+1} - Rolling sum over a four-year (16-quarter) window to capture delayed effects. - Addressing endogeneity in macroprudential adoption: - Include interaction term between FCI and macroprudential measure among controls. - Construct orthogonalized macroprudential shocks (LMPM_{i,t}) by estimating an ordered Probit: - MPM_{i,t} = λ_{0,i} + β_{1} Creditgap_{i,t−1} + β_{2} Housepricesgap_{i,t−1} + Sum_{j=1}^{4} MPM_{i,t−M−j} + ε_{i,t} - Credit gap and house price gap use Hodrick and Prescott detrending. - LMPM_{i,t} = MPM_{i,t} − Ê_{t−1}[MPM_{i,t}] = MPM_{i,t} − Sum_{k=−4}^{8} p̂_{k}(x_{i,t−1})^{k} - Alternative orthogonalization uses instantaneous policy changes (PM_{i,t}) in the ordered Probit to produce LPM_{i,t}. ### Monetary policy: construction and alternative shock measures - Baseline monetary policy shock: - PolicyRate_{i,t} = λ_{0,i} + λ_{1} Z_{i,t} + u_{i,t} - Z_{i,t} includes contemporaneous and lagged inflation, log GDP, corporate spreads, log foreign GDP, lagged short-term rate, and quadratic time trend. - Residual u_{i,t} used as identified unexpected short-term rate deviation from an augmented Taylor rule. - Two alternative high-frequency identification approaches to address residual endogeneity: - Nakamura and Steinsson (2018): use bond price movements in a narrow window around scheduled FOMC meetings (applied for United States). - Cieslak and Pang (2021): use joint dynamics of government bond yields and equity returns around central bank releases for four leading central banks (Federal Reserve, ECB, Bank of England, Bank of Japan) through end-2017. - Foreign GDP index constructed by cumulating the average quarter-on-quarter GDP growth for countries in the sample. ### Empirical findings: macroprudential policies - Direction and robustness: - Tightening borrower-based macroprudential measures (LTV and DSTI) reduces House-prices-at-Risk (HaR) and shifts the entire term structure of HaR downward. - Results robust to orthogonalized intensity shocks (LMPM_{i,t}) and to orthogonalized instantaneous shocks (LPM_{i,t}); magnitudes and significance generally consistent. - Magnitudes and timing: - In advanced economies (AEs): - One standard deviation increase in the macroprudential policy intensity measure (corresponding to 1.2 units) has a maximum impact 7 quarters ahead equal to 0.3 percentage points. - In emerging market economies (EMs): - Impact is highest in the longer term and remains steady after 12 quarters. - A one-unit tightening of macroprudential measures could lower the three-years-ahead annualized average HaR by 0.2 (about 50 percent of the median HaR value). ### Empirical findings: monetary policy - Baseline result: - A one standard deviation increase in monetary policy shocks contributes to a deterioration of HaR over a short horizon in advanced economies. - Negative, short-lived relationship with HaR primarily in advanced economies; monetary policy shocks weaken the short-term relationship between financial conditions and HaR. - Overall, monetary policy influences downside risks to house prices mainly through its impact on financial conditions and has a more limited direct effect on HaR relative to targeted macroprudential policies. - Alternative shock definitions: - Cieslak and Pang (2021) shocks: similar short-term effects with significance spanning up to 5 quarters ahead in a reduced sub-panel of advanced economies. - Nakamura and Steinsson (2018) shocks for the United States: relationship peaks 8 quarters ahead but generally shows lower statistical significance. ### Interpretation and policy implications - Aggregate interpretation: - Findings are consistent with model predictions (Section 1) and prior literature that macroprudential policy is preferable to leaning-against-the-wind monetary policy for reducing downside risks to house prices. - Tighter borrower-based macroprudential measures reduce volatility by strengthening collateral constraints and dampening financial accelerator effects. - Active use of monetary policy to reduce house-price downside risks can be counterproductive and increase house price volatility, potentially causing welfare losses. - In rational-bubble settings, monetary tightening may depress fundamentals but relax the growth constraint on bubble components; net effect depends on which component dominates and assumptions can overturn results. - Policy recommendations: - Add borrower-based macroprudential measures (caps on loan-to-value and debt-service-to-income ratios) to countries’ macroprudential toolkits and monitor them over time. - Prefer targeted macroprudential interventions over monetary policy for mitigating HaR and reducing financial stability risks. *Source: 4.3 Policies to Mitigate Downside Risks to House Prices (wpiea2020011-print-pdf).* ### References ### wpiea2020011-print-pdf - References (Appendices and Tables) ### Key empirical findings on determinants of House-Prices-at-Risk (HaR) - Advanced Economies (Table A.1 — 5th percentile, 1-quarter-ahead to 16-quarters-ahead): - House price growth (t): 1.289***, 1.151***, 0.908***, 0.820***, 0.775***, 0.562***, 0.414***, 0.344***, 0.361***, 0.381***, 0.391***, 0.392***, 0.372***, 0.357***, 0.324***, 0.300*** (standard errors reported). - House price misalignment (t): -0.473**, -0.779***, -0.788***, -0.874***, -0.931***, -0.976***, -1.033***, -1.015***, -0.972***, -0.956***, -0.920***, -0.928***, -0.945***, -0.985***, -0.982***, -0.987***. - Financial condition index (t): -0.339*, -0.283**, -0.114, -0.205*, -0.220**, -0.218***, -0.233**, -0.238***, -0.196***, -0.150***, -0.164***, -0.161***, -0.119***, -0.102**, -0.135***, -0.121***. - Credit boom (t): -0.275, -0.483*, -0.543**, -0.481***, -0.321**, -0.385**, -0.409**, -0.358**, -0.344**, -0.304*, -0.241*, -0.293**, -0.260*, -0.207**, -0.216**, -0.227**. - Observations: 2,389; 2,367; 2,345; 2,323; 2,301; 2,279; 2,257; 2,235; 2,213; 2,191; 2,169; 2,147; 2,125; 2,103; 2,081; 2,059. - Emerging Market Economies (Table A.1 — 5th percentile): - House price growth (t): 0.494, 1.025***, 0.665***, 0.529**, 0.478**, 0.528**, 0.521***, 0.469**, 0.474**, 0.371**, 0.428**, 0.261*, 0.200*, 0.211**, 0.203***, 0.183**. - House price misalignment (t): -0.748***, -0.812***, -0.837***, -0.846***, -0.962***, -0.965***, -0.976***, -1.015***, -1.085***, -1.076***, -1.015***, -1.051***, -1.047***, -1.037***, -0.978***, -1.023***. - Financial condition index (t): -0.619***, -0.525**, -0.660***, -0.674***, -0.539***, -0.453***, -0.225, -0.157, -0.094, -0.029, -0.020, -0.042, -0.086, -0.088, -0.048, -0.105. - Credit boom (t): -0.526, -0.685*, -0.934***, -1.039***, -0.959***, -1.106***, -1.065***, -0.784***, -0.561*, -0.282*, -0.125, -0.134, -0.084, -0.013, 0.017, 0.164. - Observations: 948; 938; 928; 918; 908; 898; 888; 878; 868; 858; 848; 838; 828; 818; 808; 798. - Robustness: alternative panel quantile estimators (Powell 2022; Machado and Silva 2019) and alternative sample exclusions (excluding 2008–2009) produce broadly consistent signs and magnitudes for key determinants (house price growth positive, misalignment negative, financial conditions negative, credit booms negative). ### HaR link to GDP downside risks and median GDP growth - Growth-at-Risk (GaR) augmented with HaR (Table A.4 — 5th percentile of GDP growth): - Advanced Economies: - HaR 1-year ahead (t): 0.342***, 0.445***, 0.364***, 0.377***, 0.350***, 0.344***, 0.298***, 0.285***, 0.245***, 0.236***, 0.227***, 0.218***, 0.207***, 0.204***, 0.204***, 0.195***. - Financial condition index (t): -0.279***, -0.113**, -0.113***, -0.049, -0.043, 0.009, 0.044*, 0.079***, 0.100***, 0.116***, 0.123***, 0.111***, 0.113***, 0.113***, 0.113***, 0.104***. - Observations: 2,394; 2,372; 2,350; 2,328; 2,306; 2,284; 2,262; 2,240; 2,218; 2,196; 2,174; 2,152; 2,130; 2,108; 2,086; 2,064. - Emerging Market Economies: - HaR 1-year ahead (t): 0.354, 0.301, 0.462**, 0.521***, 0.517***, 0.490***, 0.405***, 0.325***, 0.336***, 0.309***, 0.292***, 0.294***, 0.309***, 0.291***, 0.271***, 0.270***. - Financial condition index (t): 0.083, 0.344**, 0.442***, 0.607***, 0.582***, 0.493***, 0.461***, 0.405***, 0.398***, 0.348***, 0.287***, 0.259***, 0.247***, 0.244***, 0.237***, 0.233***. - Observations: 953; 944; 935; 926; 917; 907; 897; 887; 877; 867; 857; 847; 837; 827; 817; 807. - Effect on future median GDP growth (50th percentile, Table A.5): - Advanced Economies: HaR 1-year ahead (t) coefficients: 0.198***, 0.186***, 0.201***, 0.191***, 0.194***, 0.190***, 0.175***, 0.178***, 0.169***, 0.176***, 0.167***, 0.163***, 0.156***, 0.149***, 0.138***, 0.129***. - Emerging Market Economies: HaR 1-year ahead (t) coefficients: 0.090*, 0.139***, 0.155***, 0.128**, 0.145***, 0.156***, 0.169***, 0.191***, 0.175***, 0.171***, 0.155***, 0.126***, 0.109***, 0.079***, 0.079***, 0.062***. ### Probability of systemic banking crisis conditional on HaR (Table A.6) - Advanced Economies (HaR H=4 marginal predictive margins at specified HaR values): - HaR -160.34 → Predictive Margin 0.08, Std. Err. 0.04, z 0.06, P>z 0.00, [95% conf. interval] 0.18 0.51. - HaR -140.26 → Predictive Margin 0.05, Std. Err. 0.05, z 0.11, P>z 0.00, [95% conf. interval] 0.16 0.36. - HaR -120.19 → Predictive Margin 0.02, Std. Err. 0.03, z 0.08, P>z 0.00, [95% conf. interval] 0.15 0.24. - HaR -100.14 → Predictive Margin 0.00, Std. Err. 0.00, z 0.04, P>z 0.71, [95% conf. interval] 0.13 0.14. - HaR -80.09 → Predictive Margin 0.01, Std. Err. 0.01, z 0.02, P>z 0.19, [95% conf. interval] 0.07 0.11. - HaR -60.06 → Predictive Margin 0.02, Std. Err. 0.02, z 0.04, P>z 0.16, [95% conf. interval] 0.03 0.09. - HaR -40.04 → Predictive Margin 0.04, Std. Err. 0.02, z 0.02, P>z 0.01, [95% conf. interval] 0.01 0.07. - HaR -20.03 → Predictive Margin 0.03, Std. Err. 0.01, z 0.01, P>z 0.06, [95% conf. interval] 0.00 0.06. - HaR 0 → Predictive Margin 0.02, Std. Err. 0.01, z 0.01, P>z 0.14, [95% conf. interval] -0.01 0.04. - Emerging Market Economies (HaR H=4 predictive margins): - HaR -160.25 → Predictive Margin 0.03, Std. Err. 0.03, z 0.79, P>z 0.00, [95% conf. interval] 0.19 0.31. - HaR -140.17 → Predictive Margin 0.02, Std. Err. 0.02, z 0.63, P>z 0.00, [95% conf. interval] 0.14 0.20. - HaR -120.11 → Predictive Margin 0.01, Std. Err. 0.01, z 0.08, P>z 0.00, [95% conf. interval] 0.08 0.14. - HaR -100.07 → Predictive Margin 0.01, Std. Err. 0.02, z 0.24, P>z 0.00, [95% conf. interval] 0.04 0.10. - HaR -80.04 → Predictive Margin 0.04, Std. Err. 0.01, z 0.29, P>z 0.00, [95% conf. interval] 0.01 0.07. - HaR -60.02 → Predictive Margin 0.02, Std. Err. 0.01, z 0.05, P>z 0.04, [95% conf. interval] 0.00 0.05. - HaR -40.01 → Predictive Margin 0.01, Std. Err. 0.01, z 0.15, P>z 0.12, [95% conf. interval] 0.00 0.03. - HaR -20.01 → Predictive Margin 0.01, Std. Err. 0.01, z 0.12, P>z 0.21, [95% conf. interval] 0.00 0.02. - HaR 0 → Predictive Margin 0.00, Std. Err. 0.00, z 0.01, P>z 0.30, [95% conf. interval] 0.00 0.01. ### Macroprudential policy effects on HaR - Macroprudential Policy Intensity Measure (MPM) added to baseline (Table A.7): - Advanced Economies: - MPM (t): 0.230**, 0.197*, 0.237***, 0.235***, 0.282***, 0.284***, 0.288***, 0.204***, 0.165***, 0.153***, 0.120**, 0.065, 0.042, 0.059, 0.046, 0.043. - MPM x Financial Condition Index (t): 0.355***, 0.332**, 0.290***, 0.070, 0.071, 0.099, 0.097, 0.005, -0.029, -0.044, -0.048, -0.127, -0.139*, -0.121*, -0.131**, -0.122**. - Emerging Market Economies: - MPM (t): 0.089, 0.133, 0.150*, 0.151, 0.063, 0.061, 0.084, 0.186, 0.224**, 0.139, 0.137, 0.214***, 0.145***, 0.209***, 0.171***, 0.146***. - MPM x Financial Condition Index (t): -0.157, 0.036, 0.122*, 0.110, 0.101, 0.002, -0.074, -0.040, -0.052, -0.150, -0.130, -0.103**, -0.120***, -0.053, -0.045, -0.047. - Orthogonalized macroprudential shocks (LMPM / LPM) (Table A.8 and Table A.9): - Advanced Economies LMPM (t): 0.211***, 0.311***, 0.283***, 0.367***, 0.348***, 0.336***, 0.285***, 0.279***, 0.201***, 0.223***, 0.160*, 0.088, 0.068, 0.075, 0.055, 0.053. - Advanced Economies LPM (t) (instantaneous shocks): 0.310***, 0.313***, 0.257***, 0.251***, 0.296***, 0.273***, 0.272***, 0.240***, 0.210***, 0.194***, 0.157***, 0.127*, 0.094, 0.063, 0.065*, 0.067. - Emerging markets: LMPM/LPM coefficients show mixed significance across horizons; some later horizons show large positive and statistically significant coefficients (e.g., LMPM (t) 0.222***, 0.267***, 0.209*** at certain horizons). ### Monetary policy shocks and HaR (Table A.10 and alternatives) - Advanced Economies (Table A.10): - Monetary policy shock (t): -0.406***, -0.223***, -0.039, -0.016, 0.088, -0.046, -0.042, -0.064, -0.030, -0.019, 0.013, 0.017, -0.020, -0.007, 0.018, 0.002. - Interpretation: monetary policy shocks exert a statistically significant negative effect on HaR at the shortest horizons (notably -0.406*** at h=1). - Emerging Markets (Table A.10): - Monetary policy shock (t): -0.284, -0.044, -0.056, -0.161, -0.006, 0.179*, 0.051, 0.090, 0.113, 0.148, 0.113, 0.152, 0.154*, 0.206***, 0.109, 0.132*. - Alternative monetary shock measures (Table A.11): - AEs subsample shocks (Cieslak and Pang 2021): Alternative monetary policy shocks (t) coefficients (selected): -0.036, -0.158, -0.171***, -0.166***, -0.164***, -0.137, -0.141, -0.022, -0.125, -0.010, 0.003, -0.001, -0.019, -0.020, -0.020, -0.036. - US policy news shocks (Nakamura and Steinsson 2018): Policy news shock (t) coefficients (selected): -0.058, 0.019, -0.074, 0.062, 0.004, -0.104, -0.166, -0.248**, -0.192, -0.056, -0.096, -0.079, -0.052, 0.006, -0.026, -0.016. ### Model evaluation and out-of-sample performance - Table A.3 (Quantile R2 Accuracy Measure) — United States (5th percentile): - Panel-Based Estimates Pseudo-R2: H=1 → 59.96; H=4 → 48.09; H=8 → 38.18; H=12 → 47.77. T-Statistics: 0.01, 0.01, 0.01, 0.00. - Country-Level Estimates Pseudo-R2: H=1 → 80.07; H=4 → 72.66; H=8 → 86.41; H=12 → 83.78. T-Statistics: 0.00, 0.00, 0.00, 0.00. - Probability Integral Transform (PIT) analysis (A.5 and Figure A.14): - PIT results for one-year-ahead predictive distributions show empirical cumulative distributions that lie well within confidence bands of lower quantiles, supporting robustness of the quantile regression predictive distributions for future house price growth and their ability to capture downside vulnerabilities. ### Data, coverage, and summary statistics - Country coverage (Table A.12): panel includes Advanced Economies and Emerging Market Economies with sample start and end dates by country (e.g., United States 1990q2–2017q4). Note: data coverage limited by joint availability of all variables. - Summary statistics (Table A.14): - Advanced Economies (sample size = 2,384 quarterly observations): - Real House Prices (YoY): Mean 2.23, St.dev. 7.42, p25 -1.95, p50 1.95, p75 6.13, Min -40.55, Max 46.53. - Real House Prices (QoQ): Mean 0.48, St.dev. 2.36, p25 -0.66, p50 0.51, p75 1.64, Min -18.32, Max 16.5. - Real GDP Growth (YoY): Mean 2.53, St.dev. 1.09, p25 2.47, p50 3.79, Min -9.55, Max 29.07. - Total Credit to GDP: Mean 160.24, St.dev. 47.51, p25 125.6, p50 154.2, p75 187.56, Min 23, Max 98.5 (reported as 2398.5 in table — preserved as shown). - FCI: Mean 0.15, St.dev. 0.89, p25 -0.44, p50 0.01, p75 0.53, Min -3.33, Max 4.15. - Emerging Market Economies (sample size = 960 quarterly observations): - Real House Prices (YoY): Mean 2.78, St.dev. 8.6, p25 -1.38, p50 2.32, p75 6.39, Min -25.87, Max 67.48. - Real House Prices (QoQ): Mean 0.63, St.dev. 3.02, p25 -0.86, p50 0.54, p75 2.11, Min -26.04, Max 20.6. - Real GDP Growth (YoY): Mean 4.57, St.dev. 4.22, p25 2.28, p50 4.73, p75 7.37, Min -12.53, Max 16.76. - Total Credit to GDP: Mean 69.54, St.dev. 40.23, p25 39.8, p50 58.5, p75 95.9, Min 14.1, Max 213.4. - FCI: Mean -0.14, St.dev. 0.79, p25 -0.65, p50 -0.16, p75 0.32, Min -5.14, Max 3.13. ### Data sources and variable definitions (Table A.13 highlights) - Main data sources and variable descriptions include: - Real house price indices: Bank for International Settlements; CEIC Data Co. Ltd; Haver Analytics; IMF Research Department house price dataset; OECD; Thomson Reuters Datastream; IMF staff calculations. - Financial Conditions Index (FCI): IMF staff estimates; methodology and variables referenced to Appendix 3.2 of October 2017 GFSR. - Macroprudential policies: IMF Integrated Macroprudential Policy Database. - Credit measures: Bank for International Settlements; Haver Analytics; Jordà and Taylor (2016) for credit-to-GDP boom definition; Mendoza and Terrones (2014) used as alternative definition. - Systemic banking crisis dummy: Laeven and Valencia (2018). - Monetary policy shocks: IMF staff calculations (residuals from Taylor-rule regressions); alternatives from Cieslak and Pang (2021) and Nakamura and Steinsson (2018) used in robustness checks. *Italic: Source — wpiea2020011-print-pdf (References and Appendices as provided).* --- _Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020011-print-pdf.pdf_