## Annex Table 2.1.1. The empirical analysis is performed at the quarterly frequency. Data

## Source details

**Canonical URL:** [Annex Table 2.1.1. The empirical analysis is performed at the quarterly frequency. Data](https://www.imf.org/-/media/files/publications/weo/2024/april/english/ch2onlineannex.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/weo/2024/april/english/ch2onlineannex.pdf.md)
- [Structured JSON version](/-/media/files/publications/weo/2024/april/english/ch2onlineannex.pdf.json)

---

### Data construction and temporal transformations
- Monetary Policy Shocks (MPS) and Orthogonalized Monetary Policy Shocks:
  - Constructed at the monthly frequency and transformed to quarterly by summing the shocks within each quarter.
  - Sources: Bloomberg.
- Policy interest rate:
  - Retrieved from Bloomberg for each monetary policy announcement at the monthly frequency.
  - Monthly series are averaged to create quarterly data.
  - Note: Policy rates are averaged in months with more than one announcement (less than 2 percent of the sample).
- Temporal transformations:
  - Annual data converted into quarterly by assigning the annual value to each quarter in a given calendar year.
  - Monthly data converted into quarterly by assigning the value from the third month of each quarter.
  - For regional data: annual data series assigned to the fourth quarter of their respective year, then interpolated to obtain quarterly values.

### Housing and credit variable construction
- Housing variables:
  - Total homeownership rate and Average mortgage lending rate: combined from four different sources to maximize coverage.
  - Share of fixed rate mortgages: defined as the share of outstanding mortgages with rates that are fixed for at least 12 months (i.e., there is no rate-reset in the following 12 months) as a proportion of total outstanding mortgages.
- Credit variables:
  - Household Credit to GDP and Total Household Credit (NCU): Bank of International Settlements (BIS).
  - Share of Fixed Rate Mortgages in Stock: European Central Bank (ECB), national Central Banks.
  - Regulatory Loan-to-Value Limits (Average): Integrated Macroprudential Policy (iMaPP) Database.
  - Effective Rates on Outstanding Mortgage Loans: European Central Bank (ECB); Federal Reserve Board.

### Country-level panel dataset — key variables and sources
- Monetary Policy Shocks / Orthogonalized Monetary Policy Shocks: Bloomberg.
- Housing variables:
  - Residential House Price: Bank of International Settlements (BIS).
  - Commercial House Price: Morgan Stanley Capital International (MSCI).
  - House Sales, House Starts: Haver Analytics.
  - House Rents, Price-to-Rent Ratio, Price-to-Income Ratio: Organisation for Economic Co-operation and Development (OECD).
  - Asset Value Growth Index (Office, Retail): Morgan Stanley Capital International (MSCI).
- Other general economic indicators:
  - GDP (Constant and Current Prices), Headline CPI, GDP per Capita (Constant Prices), Private Consumption (Constant Prices), Gross Fixed Capital Formation (Constant Prices): World Economic Outlook database.
  - Policy Interest Rate: Bloomberg.

### Country groups composition
- Advanced Economies:
  - Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, Korea, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, United Kingdom, United States
- Emerging Markets:
  - Chile, Colombia, Hungary, Indonesia, Malaysia, Mexico, Poland, Russia, South Africa, Thailand

### Regional-level panel dataset: scope, variables, and transformations
- Coverage:
  - Regions in 9 countries: Belgium (11 regions), Denmark (5), Finland (4), France (26), Hungary (8), Mexico (32), Netherlands (12), Spain (19), United Kingdom (35) and United States (51).
  - Regions defined at the NUTS2 level (exceptions: United Kingdom Local Authority District; France NUTS3 aggregated to NUTS2).
- Monetary policy shocks:
  - National MPS are imputed to all regions in each country.
- House Price Index (regional handling and transformations):
  - Mexico, United Kingdom, United States: available regional House Price Index used directly.
  - United Kingdom: regional data at lower disaggregation than NUTS2 — computed as sales-weighted mean price by NUTS2 region-day, then aggregated as sales-weighted mean price by NUTS2 region-quarter.
  - Finland, Netherlands, Belgium: median house price for “all dwellings” or “all houses”; series smoothed using a 4-quarter moving average by region.
  - Spain: house price series smoothed using a 4-quarter moving average by region.
  - Hungary: mean price of all dwelling types within each region-quarter.
  - France: transaction-level data transformed to median sales price by region-quarter, then smoothed using a 4-quarter moving average.
  - Denmark: drop properties classified as "Holiday home"; mean house price by region-quarter across all dwelling types; smoothed using a 4-quarter moving average by region.
  - All house price series are normalized by setting the value in 2018 to 100.
- CPI (United States):
  - State-level price level obtained using nominal and real GDP from the Bureau of Economic Analysis (BEA) and computed as a GDP deflator: CPI = 100 * (nominal GDP / real GDP).
  - Rebasing: value in 2018 set to 100.
- GDP (regional dataset):
  - GDP series in national currencies converted into USD using PPP-adjusted exchange rates for each country.
  - GDP data for the US and Mexico is already in PPP-adjusted USD.
  - Data for the UK is in Pounds; data for all other countries is in PPP-adjusted Euros.
  - Exchange rates from local currency to PPP-adjusted for all countries retrieved from WEO database, except Denmark and Hungary where it equals the market exchange rate from euros to USD.
- Housing supply constraints:
  - United States: measured using the Wharton Residential Land Use Regulation Index (WRLURI) compiled by Gyourko and others, 2021.
  - For United States, WRLURI aggregated from county to state level as the population-weighted average of the county-level index.
- Regional data transformations:
  - All variables given at NUTS2 level for all countries except United Kingdom and France, where series are aggregated to NUTS2.

### Regional-level panel dataset — main sources (selected)
- Monetary Policy Shocks: Bloomberg.
- House Price Index: Belgium (STAT BEL), Denmark (Statistics Denmark), France (INSEE), Finland (StatFin), Hungary (Hungarian Central Statistical Office), Mexico (Sociedad Hipotecaria Federal), Netherlands (CBS Open Data), Spain (CEIC), United Kingdom (Local Authority District), United States (Federal Housing Finance Agency, FHFA).
- CPI: World Economic Outlook database, Bureau of Economic Analysis (United States).
- GDP and GDP per capita: Eurostat, OECD, Office of National Statistics (ONS), Bureau of Economic Analysis (BEA) as per country.
- Population and Population density: Eurostat, OECD, US Census Bureau as per country.
- Exchange rate: World Economic Outlook database.
- Office and Retail capital value: Morgan Stanley Capital International (MSCI).
- Land use regulation index: Wharton Residential Land Use Regulation Index (WRLURI) (only US).

### Housing finance characteristics and country coverage (selected points)
- Online Annex Table 2.2.1 fields include:
  - Share of Households (Owner with Mortgage) (data collected for 2020 unless otherwise specified)
  - Share of Households (Owner without Mortgage) (data collected for 2020 unless otherwise specified)
  - Share of Fixed rate Mortgages (data collected for 2023 unless otherwise specified)
  - Average of Regulatory LTV Limits (data collected for 2021 unless otherwise specified)
  - Typical Term to Maturity (data corresponds to 2015 from Cerutti, Dagher, and Dell'Ariccia (2017) unless otherwise specified)
  - Availability of Full Recourse (data corresponds to 2015 unless otherwise specified)
  - Housing Finance: Retail Funding (varies by country)
- Examples of exact reported data entries:
  - Argentina: Share of Households (Owner with Mortgage) = ––– ; Share of Households (Owner without Mortgage) = 10020 ; Share of Fixed rate Mortgages = No ; Housing Finance: Retail Funding = Retail Deposit
  - Australia: Share of Households (Owner with Mortgage) = 3231C10025 ; Availability of Full Recourse = Yes ; Housing Finance: Retail Funding = Other
  - United States: Share of Households (Owner with Mortgage) = 40269510030 ; Share of Fixed rate Mortgages = No ; Housing Finance: Retail Funding = Other
  - Note: The letter C refers to data points that are not published in this table for confidentiality reasons.
- Source aggregation for table: Bank for international settlements; country authorities; Haver; OECD; iMAPP; Cerutti, E., J. Dagher and G. Dell’Ariccia (2017); and IMF staff calculations.
- Data year notes:
  - 1 Data is collected for 2020 unless otherwise specified;
  - 2 Data is collected for 2023 unless otherwise specified;
  - 3 Data is collected for 2021 unless otherwise specified;
  - 4 Data corresponds to 2015 and is collected from Cerutti, Dagher, and Dell'Ariccia (2017);
  - 5 2017;
  - 6 2018;
  - 7 2019;
  - 8 2021;
  - 9 2022.
- Fixed rate mortgages exclude mortgages that adjust to inflation (like in Chile and Israel).

### Fixed rate mortgages: coverage and definitions
- Country examples (coverage windows and definitions preserved exactly):
  - Australia: 2019:Q3–2022:Q4, FRM if residual fixation > 12 months, Source: Reserve Bank of Australia
  - Austria: 2010:Q2–2022:Q4, FRM if residual fixation > 12 months, Source: ECB
  - Canada: 2016:Q3–2022:Q4, FRM if residual fixation > 12 months, Source: Bank of Canada
  - Denmark: 2013:Q4–2022:Q4, FRM if residual fixation > 12 months, Source: Danmarks Nationalbank
  - Japan: 2016:Q1–2022:Q1, FRM if residual fixation > 12 months, Source: Bank of Japan
  - Norway: 2013:Q4–2022:Q4, FRM if residual fixation > 3 months, Source: Statistics Norway
  - United States: 2013:Q1–2022:Q4, FRM if defined by duration of contract, Source: Federal Housing Finance Agency
- Classification notes:
  - Unless otherwise specified, loan classification is based on current fixed/floating status, rather than status at origination.
  - For countries where residual fixation is denoted by “duration of contract,” FRMs are loans not floating at any given quarter, irrespective of residual fixation.
  - Chile: all mortgages are inflation indexed and are thus classified as floating.
  - Israel: classification based on characteristics at origination; mortgages which are inflation–adjusted are classified as floating irrespective of fixation period.
  - Spain: comparisons across time use 2012Q1 at the request of authorities.
  - ECB: proportion of total outstanding loans to households (including, but not limited to, mortgages).
  - FRM = fix-rate mortgages.

### Additional stylized facts (summary of descriptive findings)
- Housing macro-criticality:
  - Activities related to housing account for about 15 percent of a country GDP on average, and about 7 percent of employment (Online Annex Figure 2.2.1).
- Affordability metrics:
  - Price-to-rent and price-to-income ratios experienced a boom and bust around the GFC; the pandemic period led to an increase in both ratios, which reached and surpassed pre-GFC levels in many countries.
- Housing activity:
  - Following the GFC, both housing starts and sales dropped by about 30 percent compared to pre-GFC levels and remained low for most of the 2010s.
  - The pandemic saw a surge in transactions; post-2022 rate hikes led to a sharp fall in new constructions and a drop in sales.
  - Recent drop-in housing activity is also illustrated by a drop in the number of new loans extended to households in available euro area data.
- Developer costs:
  - Inflation and rising capital costs pushed developers' costs up significantly as key material inputs increased in cost.
- Regulatory and credit changes:
  - Changes in regulatory LTV ratios and household credit-to-GDP ratios between 2022:Q4 (or latest available) and 2011:Q1 documented in Online Annex Figures 2.2.6 and 2.2.7.
- Demographic differentials:
  - Differences in the population growth differential between areas with high and low population density from 2019:Q4 to 2022:Q4 (or latest available) documented in Online Annex Figure 2.2.8.1.

### Monetary Policy Shocks: construction and coverage (summary)
- MPS measured as difference between actual monetary policy announcements and professional analyst forecasts submitted to Bloomberg up to the day prior to the announcement.
- Frequency and aggregation:
  - Constructed at the monthly frequency and transformed to quarterly by summing the shocks within each quarter.
- Dataset coverage:
  - Unbalanced panel of 30 countries and monetary unions.
  - Starting from as early as 1998.
  - Covering over 4,600 months of announcements.
- In rare cases with more than one announcement per month (less than 2 percent of the sample), policy rates and forecasts are averaged at the monthly level.
- Pegged countries included from the peg date if after 1998 using the same methodology.

### Orthogonalized Monetary Policy Shocks: identification adjustments
- Orthogonalized MPS constructed as residual from regressing each MPS on:
  - Two lags of GDP surprises.
  - Six lags of inflation surprises.
  - Change in national stock price index over the previous 6 months to the shock.
- Definitions and windows:
  - GDP surprises: actual release value for GDP minus mean of analyst forecasts from Bloomberg; 2 lags taken within a backward-looking window of 11 months up to the day prior to the monetary policy announcement.
  - Inflation surprises: actual release value for inflation minus mean of analyst forecasts from Bloomberg; 6 lags taken within a backward-looking window of 330 days up to the day prior to the monetary policy announcement.
  - Stock price changes: change in the domestic stock market price index on the day prior to each policy announcement relative to its value 180 days earlier; series refer to performance of the largest Exchange-Traded Funds (ETFs) listed on each country's stock exchange.
- Aggregation:
  - Orthogonalized MPS aggregated at the quarterly level by summing shocks within each quarter and expanded to include pegged countries using the same method.

### Empirical approach — transmission to house prices and real activity
- Dataset: unbalanced country-level panel covering 33 AEs and EMEs between 1998:Q4 and 2022:Q4.
- Method: instrumental variables local projections (LP-IV) following Stock and Watson (2018) and Jordà and others (2015).
- Key specification features:
  - Projection horizon h = 0,...,8.
  - Dependent variable: cumulative percentage change (log difference) in house prices, consumption, or other macro outcomes after h quarters.
  - Monetary policy measure: 2SLS estimate of the effect of a 100bpp change in policy rates over a given quarter; change in policy rates instrumented with surprises around monetary policy announcements among Bloomberg professional forecasters.
  - Controls X include 8 lags of: growth rate in the dependent variable; growth rate in real GDP; headline CPI inflation; nominal house prices; outstanding household credit in national currency.
  - Fixed effects: country and time fixed effects (μ_c^h and τ_t^h).
  - Standard errors: Driscoll and Kraay (1998) with three lags — robust to heteroskedasticity, autocorrelation, and cross-sectional dependence.
  - Charts report 90 percent confidence intervals based on these standard errors.

### Heterogeneity in transmission due to mortgage finance characteristics
- Focus variables: relative leverage ratio (proxied by outstanding household credit-to-GDP), maximum regulatory LTV limits, and share of fixed-rate mortgages (FRMs) in stock.
- Empirical augmentation:
  - LP-IV with interaction terms; interaction variable H_c,t−1 denotes lagged mortgage finance characteristic.
  - Models interact monetary policy shock with dummy denoting being above or below sample median for two out of three characteristics (household debt-to-GDP; share of FRMs in stock).
  - LTV dummy = 1 when the regulatory limit is below 100 percent, 0 otherwise.
  - Coefficient β2^h captures differential effects for different levels of H_c,t−1.
- Sign-sensitive specification:
  - Equation (3) expressed in reduced form to allow state variables to depend on the sign of the monetary policy impulse (tightening vs. loosening).
  - Absolute value of the monetary policy shock interacted with:
    - Dummy TIGHT^h = 1 if shock > 0, 0 otherwise.
    - Dummy LOOSE = 1 if shock < 0, 0 otherwise.
  - Coefficients β1^h and β2^h capture differential response to tightening and loosening shocks at different values of H_c,t−1.
- Additional empirical findings:
  - Using share of households with and without mortgages as H_c,t−1: House prices respond significantly more the more households have mortgages; consumption appears to respond more slowly (differences in consumption not significant).
  - Using simple share of homeowners as H_c,t−1: effects not statistically different between samples.

### Regional heterogeneity — methodology and definitions
- Regional dataset: unbalanced region-level panel covering 192 regions in 9 countries between 2005:Q1 and 2022:Q4.
- Estimation: Local projections with instrumental variables (LP-IV) augmented with interaction terms and country-time fixed effects.
- Main specification:
  - Models cumulative log change in house prices and GDP per capita after h quarters: h = 0,...,8.
  - 2SLS estimate of change in policy rates denoted D̂policy(c,t).
  - Regional dummy Hc,j,t−4 indicates past values of high population density or high house price overvaluation.
  - Controls Xc,j,t−l include 12 lags of changes in log house prices, GDP per capita, CPI inflation, and population.
  - Region fixed effects μc,jh and country-time fixed effects θc,th; standard errors clustered at the regional level.
  - Charts report 90 percent confidence intervals.
- Regional dummy definitions:
  - High population density: region’s density in the top 10th percentile within its country-year distribution.
  - High house price overvaluation: regional PIR deviations from long-term average in the top 25th percentile.
- Asymmetry test:
  - Equation (4) modifies (3) to capture asymmetric effects of tightening vs loosening by interacting instrumented policy changes with regional dummies and separate indicators for tightening vs loosening episodes.

### Regional heterogeneity — key empirical results and validation
- Differential responses:
  - Online Annex Figure 2.6.1 shows cumulative responses of house prices and real GDP-per-capita to a 100 bpp tightening (blue) and loosening (red) in areas with high population density (supply constrained) and high house price overvaluations.
  - Shaded areas indicate 90th percentile confidence intervals.
- Distributional evidence:
  - Figure 2.6.2 reports distributions of population density and house price overvaluations standardized at country level (mean 0, standard deviation 1).
- Proxy validation:
  - Correlation between population density in 2019 and WRLURI (Wharton residential land use regulatory index) from a 2018 US survey approximately 0.6, supporting population density as a proxy for housing supply constraints.

### Model-based analysis (TANK model extensions and calibration)
- Model framework:
  - Two-agent New Keynesian (TANK) model extending Iacoviello and Neri (2010) with illiquid housing, long-term debt, collateral constraints, and macroprudential tools such as LTV ratios.
  - Households heterogeneous in discount rates (patient savers, impatient borrowers); housing and consumption goods produced with different technologies.
- Calibration adjustments:
  - Equity withdrawal and inflation target fixed at 1.5 percent and 2 percent, respectively.
  - Household preferences for housing services set to 0.19 for all households to raise steady state debt as percent of GDP.
  - Monetary policy via a Taylor Rule with smoothness parameter set at 0.95 to mimic peak consumption response.
  - Model exhibits standard-sized impulse responses for GDP, consumption, and inflation.
- Simulation scenarios (joint impact of LTV limits and FRM share):
  - Baseline scenario 1 — Restricted vs not restricted LTV: LTV changes from 0.75 to 0.9 as percentage of housing investments.
  - Baseline scenario 2 — High vs low FRM: Share of outstanding loans with a fixed-rate mortgage change from 95% to 70%.
  - Qualitative simulation findings:
    - Less restrictive LTV and low FRM represent highest degree of transmission (strongest consumption response to tightening).
    - High FRM and high LTV indicate a lower degree of transmission (weaker consumption response).

### Key empirical stylized findings (concise)
- Housing sector size:
  - Housing-related activities account for about 15 percent of GDP on average and about 7 percent of employment.
- Affordability and activity dynamics:
  - Price-to-rent and price-to-income ratios: boom and bust around the GFC; pandemic increased both ratios to at-or-above pre-GFC levels in many countries.
  - Housing starts and sales: dropped by about 30 percent after the GFC compared to pre-GFC levels; surged during the pandemic; post-2022 rate hikes produced a sharp fall in new constructions and sales.
  - New loans to households: recent drop evidenced in available euro area data.
- Developer costs and regulation:
  - Construction/developer costs rose substantially due to inflation and higher capital costs.
  - Regulatory LTV ratios and household credit-to-GDP ratios documented changes between 2022:Q4 (or latest available) and 2011:Q1.
- Mortgage market features:
  - FRM definitions and country coverage vary; FRMs typically defined by residual fixation > 12 months except where noted (e.g., Norway > 3 months; US defined by duration of contract).
  - Fixed-rate mortgage prevalence materially affects transmission of policy to effective mortgage rates, house prices, and consumption in reduced-form estimates.

*Source: Online Annex Tables and Figures accompanying Chapter 2, World Economic Outlook, International Monetary Fund | April 2024.*

### Annex Table 2.1.1. The empirical analysis is performed at the quarterly frequency. Data

### Annex Table 2.1.1. The empirical analysis is performed at the quarterly frequency. Data

### Data construction and temporal transformations
- Monetary Policy Shocks (MPS) and Orthogonalized Monetary Policy Shocks:
  - Constructed at the monthly frequency and transformed to quarterly by summing the shocks within each quarter.
  - Sources: Bloomberg.
- Policy interest rate:
  - Retrieved from Bloomberg for each monetary policy announcement at the monthly frequency.
  - Monthly series are averaged to create quarterly data.
  - Note: Policy rates are averaged in months with more than one announcement (less than 2 percent of the sample).
- Temporal transformations:
  - Annual data is converted into quarterly by assigning the annual value to each quarter in a given calendar year.
  - Monthly data is converted into quarterly by assigning the value from the third month of each quarter.
  - For regional data: annual data series have been assigned to the fourth quarter of their respective year, then interpolated to obtain quarterly values.

### Housing and credit variable construction
- Housing variables:
  - Total homeownership rate and Average mortgage lending rate: combined from four different sources to maximize coverage (see Online Annex Table 2.2.1).
  - Share of fixed rate mortgages: defined as the share of outstanding mortgages with rates that are fixed for at least 12 months (i.e., there is no rate-reset in the following 12 months) as a proportion of total outstanding mortgages. Details on coverage and definitions provided in Annex Table 2.2.2.
- Credit variables:
  - Household Credit to GDP and Total Household Credit (NCU): Bank of International Settlements (BIS).
  - Share of Fixed Rate Mortgages in Stock: European Central Bank (ECB), national Central Banks.
  - Regulatory Loan-to-Value Limits (Average): Integrated Macroprudential Policy (iMaPP) Database.
  - Effective Rates on Outstanding Mortgage Loans: European Central Bank (ECB); Federal Reserve Board.

### Country-level panel dataset — key variables and sources
- Monetary Policy Shocks / Orthogonalized Monetary Policy Shocks: Bloomberg.
- Housing variables:
  - Residential House Price: Bank of International Settlements (BIS).
  - Commercial House Price: Morgan Stanley Capital International (MSCI).
  - House Sales, House Starts: Haver Analytics.
  - House Rents, Price-to-Rent Ratio, Price-to-Income Ratio: Organisation for Economic Co-operation and Development (OECD).
  - Asset Value Growth Index (Office, Retail): Morgan Stanley Capital International (MSCI).
- Other general economic indicators:
  - GDP (Constant and Current Prices), Headline CPI, GDP per Capita (Constant Prices), Private Consumption (Constant Prices), Gross Fixed Capital Formation (Constant Prices): World Economic Outlook database.
  - Policy Interest Rate: Bloomberg.

### Country groups composition
- Advanced Economies:
  - Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, Korea, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, United Kingdom, United States
- Emerging Markets:
  - Chile, Colombia, Hungary, Indonesia, Malaysia, Mexico, Poland, Russia, South Africa, Thailand

### Regional-level panel dataset: scope, variables, and transformations
- Coverage:
  - Regions in 9 countries: Belgium (11 regions), Denmark (5), Finland (4), France (26), Hungary (8), Mexico (32), Netherlands (12), Spain (19), United Kingdom (35) and United States (51).
  - Regions defined at the NUTS2 level (exceptions: United Kingdom Local Authority District; France NUTS3 aggregated to NUTS2).
- Monetary policy shocks:
  - National MPS are imputed to all regions in each country.
- House Price Index (regional handling and transformations):
  - Mexico, United Kingdom, United States: available regional House Price Index used directly.
  - United Kingdom: regional data at lower disaggregation than NUTS2 — computed as sales-weighted mean price by NUTS2 region-day, then aggregated as sales-weighted mean price by NUTS2 region-quarter.
  - Finland, Netherlands, Belgium: median house price for “all dwellings” or “all houses”; series smoothed using a 4-quarter moving average by region.
  - Spain: house price series smoothed using a 4-quarter moving average by region.
  - Hungary: mean price of all dwelling types within each region-quarter.
  - France: transaction-level data transformed to median sales price by region-quarter, then smoothed using a 4-quarter moving average.
  - Denmark: drop properties classified as "Holiday home"; mean house price by region-quarter across all dwelling types; smoothed using a 4-quarter moving average by region.
  - All house price series are normalized by setting the value in 2018 to 100.
- CPI (United States):
  - State-level price level obtained using nominal and real GDP from the Bureau of Economic Analysis (BEA) and computed as a GDP deflator: CPI = 100 * (nominal GDP / real GDP).
  - Rebasing: value in 2018 set to 100.
- GDP (regional dataset):
  - GDP series in national currencies converted into USD using PPP-adjusted exchange rates for each country.
  - GDP data for the US and Mexico is already in PPP-adjusted USD.
  - Data for the UK is in Pounds; data for all other countries is in PPP-adjusted Euros.
  - Exchange rates from local currency to PPP-adjusted for all countries retrieved from WEO database, except Denmark and Hungary where it equals the market exchange rate from euros to USD.
- Housing supply constraints:
  - United States: measured using the Wharton Residential Land Use Regulation Index (WRLURI) compiled by Gyourko and others, 2021.
  - For United States, WRLURI aggregated from county to state level as the population-weighted average of the county-level index.
- Regional data transformations:
  - All variables given at NUTS2 level for all countries except United Kingdom and France, where series are aggregated to NUTS2.

### Regional-level panel dataset — main sources (selected)
- Monetary Policy Shocks: Bloomberg.
- House Price Index: Belgium (STAT BEL), Denmark (Statistics Denmark), France (INSEE), Finland (StatFin), Hungary (Hungarian Central Statistical Office), Mexico (Sociedad Hipotecaria Federal), Netherlands (CBS Open Data), Spain (CEIC), United Kingdom (Local Authority District), United States (Federal Housing Finance Agency, FHFA).
- CPI: World Economic Outlook database, Bureau of Economic Analysis (United States).
- GDP and GDP per capita: Eurostat, OECD, Office of National Statistics (ONS), Bureau of Economic Analysis (BEA) as per country.
- Population and Population density: Eurostat, OECD, US Census Bureau as per country.
- Exchange rate: World Economic Outlook database.
- Office and Retail capital value: Morgan Stanley Capital International (MSCI).
- Land use regulation index: Wharton Residential Land Use Regulation Index (WRLURI) (only US).

### Housing finance characteristics and country coverage (selected points)
- Online Annex Table 2.2.1 fields include:
  - Share of Households (Owner with Mortgage) (data collected for 2020 unless otherwise specified)
  - Share of Households (Owner without Mortgage) (data collected for 2020 unless otherwise specified)
  - Share of Fixed rate Mortgages (data collected for 2023 unless otherwise specified)
  - Average of Regulatory LTV Limits (data collected for 2021 unless otherwise specified)
  - Typical Term to Maturity (data corresponds to 2015 from Cerutti, Dagher, and Dell'Ariccia (2017) unless otherwise specified)
  - Availability of Full Recourse (data corresponds to 2015 unless otherwise specified)
  - Housing Finance: Retail Funding (varies by country)
- Examples of exact reported data entries (preserved as in source):
  - Argentina: Share of Households (Owner with Mortgage) = ––– ; Share of Households (Owner without Mortgage) = 10020 ; Share of Fixed rate Mortgages = No ; Housing Finance: Retail Funding = Retail Deposit
  - Australia: Share of Households (Owner with Mortgage) = 3231C10025 ; Availability of Full Recourse = Yes ; Housing Finance: Retail Funding = Other
  - United States: Share of Households (Owner with Mortgage) = 40269510030 ; Share of Fixed rate Mortgages = No ; Housing Finance: Retail Funding = Other
  - Note: The letter C refers to data points that are not published in this table for confidentiality reasons.
- Source aggregation for table: Bank for international settlements; country authorities; Haver; OECD; iMAPP; Cerutti, E., J. Dagher and G. Dell’Ariccia (2017); and IMF staff calculations.
- Data year notes:
  - 1 Data is collected for 2020 unless otherwise specified;
  - 2 Data is collected for 2023 unless otherwise specified;
  - 3 Data is collected for 2021 unless otherwise specified;
  - 4 Data corresponds to 2015 and is collected from Cerutti, Dagher, and Dell'Ariccia (2017);
  - 5 2017;
  - 6 2018;
  - 7 2019;
  - 8 2021;
  - 9 2022.
- Fixed rate mortgages exclude mortgages that adjust to inflation (like in Chile and Israel).

### Fixed rate mortgages: coverage and definitions
- Online Annex Table 2.2.2 provides country-level coverage windows and definitions for FRM in outstanding stock; examples preserved exactly:
  - Australia: 2019:Q3–2022:Q4, FRM if residual fixation > 12 months, Source: Reserve Bank of Australia
  - Austria: 2010:Q2–2022:Q4, FRM if residual fixation > 12 months, Source: ECB
  - Canada: 2016:Q3–2022:Q4, FRM if residual fixation > 12 months, Source: Bank of Canada
  - Denmark: 2013:Q4–2022:Q4, FRM if residual fixation > 12 months, Source: Danmarks Nationalbank
  - Japan: 2016:Q1–2022:Q1, FRM if residual fixation > 12 months, Source: Bank of Japan
  - Norway: 2013:Q4–2022:Q4, FRM if residual fixation > 3 months, Source: Statistics Norway
  - United States: 2013:Q1–2022:Q4, FRM if defined by duration of contract, Source: Federal Housing Finance Agency
- Classification notes:
  - Unless otherwise specified, loan classification is based on current fixed/floating status, rather than status at origination.
  - For countries where residual fixation is denoted by “duration of contract,” FRMs are loans not floating at any given quarter, irrespective of residual fixation.
  - Chile: all mortgages are inflation indexed and are thus classified as floating.
  - Israel: classification based on characteristics at origination; mortgages which are inflation–adjusted are classified as floating irrespective of fixation period.
  - Spain: comparisons across time use 2012Q1 at the request of authorities.
  - ECB: proportion of total outstanding loans to households (including, but not limited to, mortgages).
  - FRM = fix-rate mortgages.

### Additional stylized facts (summary of descriptive findings)
- Housing macro-criticality:
  - Activities related to housing account for about 15 percent of a country GDP on average, and about 7 percent of employment (Online Annex Figure 2.2.1).
- Affordability metrics:
  - Price-to-rent and price-to-income ratios experienced a boom and bust around the GFC; the pandemic period led to an increase in both ratios, which reached and surpassed pre-GFC levels in many countries (Online Annex Figure 2.2.2).
- Housing activity:
  - Following the GFC, both housing starts and sales dropped by about 30 percent compared to pre-GFC levels and remained low for most of the 2010s.
  - The pandemic saw a surge in transactions; post-2022 rate hikes led to a sharp fall in new constructions and a drop in sales.
  - Recent drop-in housing activity is also illustrated by a drop in the number of new loans extended to households in available euro area data (Online Annex Figure 2.2.4).
- Developer costs:
  - Inflation and rising capital costs pushed developers' costs up significantly as key material inputs increased in cost (Online Annex Figure 2.2.5).
- Regulatory and credit changes:
  - Online Annex Figures 2.2.6 and 2.2.7 show changes in regulatory LTV ratios and household credit-to-GDP ratios between 2022:Q4 (or latest available) and 2011:Q1.
- Demographic differentials:
  - Online Annex Figure 2.2.8.1 shows differences in the population growth differential between areas with high and low population density from 2019:Q4 to 2022:Q4 (or latest available).

*Source: Online Annex Tables and Figures accompanying Chapter 2, World Economic Outlook, International Monetary Fund | April 2024.*

### 1. Share of Activity and Employment

### ch2onlineannex - 1. Share of Activity and Employment

### Housing market indicators and cross-country medians
- Price-to-rent and price-to-income ratios: median indices (2005 = 100) show series for multiple advanced and emerging economies. The dotted vertical line corresponds to 2020:Q1.
- Median indices for housing starts and housing sales use 2005 = 100 as reference.
- Construction input costs: median index (2015 = 100) tracked from 2015:Q1 through 2023:Q3.
- Volume of new loans to households: median index (2005 = 100) for selected European countries, tracked through 2023.
- Online Annex Figure 2.2.8.2 reports the median price-to-income ratio (PIR) growth differential between overvalued and non-overvalued areas, from 2019:Q4 to 2022:Q4 (or latest available).
- Online Annex Figure 2.2.3 shows housing starts and sales (Median index, 2005 = 100) with the vertical lines at 2020:Q1 and 2022:Q1.
- Construction input costs chart spans values up to 240 on the median index (2015 = 100).

### Monetary Policy Shocks: construction and coverage
- Monetary Policy Shocks (MPS) are measured as the difference between actual monetary policy announcements and professional analyst forecasts submitted to Bloomberg up to the day prior to the announcement.
- Frequency and aggregation:
  - MPS are constructed at the monthly frequency.
  - The monthly MPS are transformed to quarterly by summing the shocks within each quarter.
- Dataset coverage:
  - Unbalanced panel of 30 countries and monetary unions.
  - Starting from as early as 1998.
  - Covering over 4,600 months of announcements.
- In rare cases where there was more than one announcement per month (less than 2 percent of the sample), policy rates and forecasts are averaged at the monthly level.
- Pegged countries:
  - Countries pegged to currencies for which MPS information is available are added following the same methodology and included from the peg date if after 1998.
  - Online Table 2.3.1 lists countries pegged to the European Currency Unit/Euro, Dollar, and Euro–Dollar baskets.

### Orthogonal Monetary Policy Shocks: addressing central bank reaction to information
- Orthogonalized MPS are constructed as the residual from regressing each MPS on:
  - Two lags of GDP surprises.
  - Six lags of inflation surprises.
  - The change in the national stock price index over the previous 6 months to the shock.
- Definitions and windows:
  - GDP surprises = actual release value for GDP minus mean of analyst forecasts from Bloomberg. For each announcement, 2 lags of GDP surprises are taken within a backward-looking window of 11 months up to the day prior to the monetary policy announcement.
  - Inflation surprises = actual release value for inflation minus mean of analyst forecasts from Bloomberg. For each announcement, 6 lags of inflation surprises are taken within a backward-looking window of 330 days up to the day prior to the monetary policy announcement.
  - Stock price changes = change in the domestic stock market price index on the day prior to each policy announcement relative to its value 180 days earlier; series refer to performance of the largest Exchange-Traded Funds (ETFs) listed on each country's stock exchange.
- The orthogonalized MPS dataset is aggregated at the quarterly level by summing shocks within each quarter and expanded to include pegged countries using the same method.

### Transmission of monetary policy to house prices and real activity: empirical approach
- Dataset: unbalanced country-level panel covering 33 AEs and EMEs between 1998:Q4 and 2022:Q4.
- Empirical method: instrumental variables local projections (LP-IV) following Stock and Watson (2018) and Jordà and others (2015).
- Key specification features:
  - Projection horizon h = 0,...,8 (up to 8 quarters ahead).
  - Dependent variable: cumulative percentage change (log difference) in house prices, consumption, or other macro outcomes after h quarters.
  - Monetary policy measure: 2SLS estimate of the effect of a 100bpp change in policy rates over a given quarter; change in policy rates is instrumented with surprises around monetary policy announcements among Bloomberg professional forecasters.
  - Controls X include 8 lags of: growth rate in the dependent variable; growth rate in real GDP; headline CPI inflation; nominal house prices; outstanding household credit in national currency.
  - Fixed effects: country and time fixed effects (μ_c^h and τ_t^h).
  - Standard errors: Driscoll and Kraay (1998) with three lags — robust to heteroskedasticity, autocorrelation, and cross-sectional dependence.
  - Charts report 90 percent confidence intervals based on these standard errors.
- Results presentation:
  - Online Annex Figure 2.4.1 presents estimates of β1^h for real private consumption and nominal house prices.
  - Blue lines: cumulative percentage point response to a 100bpp change in policy rates instrumented with monetary policy shocks.
  - Horizontal axes represent quarters after the monetary policy action; vertical axes represent percentage points.
  - Shaded areas represent 90 percent confidence intervals.

### Heterogeneity in transmission due to mortgage finance characteristics
- Focus variables for heterogeneity: relative leverage ratio (proxied by outstanding household credit-to-GDP), maximum regulatory LTV limits, and share of fixed-rate mortgages (FRMs) in stock.
- Empirical approach: LP-IV augmented with interaction terms to capture differential effects at different levels of mortgage market characteristics.
- Specification highlights:
  - Interaction variable H_c,t−1 denotes the (lagged) value of each mortgage finance characteristic per country/quarter.
  - Models interact the monetary policy shock with a dummy denoting being above or below the sample median for two out of three characteristics (household debt-to-GDP; share of FRMs in stock).
  - LTV dummy = 1 when the regulatory limit is below 100 percent, 0 otherwise.
  - Coefficient β2^h captures differential effects of a given change in policy rates for different levels of H_c,t−1.
- Sign-sensitive specification:
  - Equation (3) modifies equation (2) to allow the role of state variables to depend on the sign of the monetary policy impulse (tightening vs. loosening).
  - Due to power loss from multiple interaction terms, equation (3) is expressed in reduced form (outcomes regressed directly on monetary policy shocks).
  - Absolute value of the monetary policy shock is interacted with:
    - Dummy TIGHT^h which takes value 1 if the shock has positive value, and 0 otherwise.
    - Dummy LOOSE which takes value 1 if the shock has negative value, and 0 otherwise.
  - Coefficients β1^h and β2^h capture differential response to tightening and loosening shocks at different values of H_c,t−1.
- Additional results:
  - Online Annex Figure 2.5.1 uses the share of households with and without mortgages as H_c,t−1; results are comparable to using household-debt-to-GDP ratios.
    - Finding: House prices respond significantly more the more households have mortgages.
    - Consumption appears to respond more slowly to a rate change (differences in consumption are not significant).
  - Online Annex Figure 2.5.2 examines differential responses depending on the simple share of homeowners; effects are not statistically different between samples.

*Source: ch2onlineannex - 1. Share of Activity and Employment (PDF chapter).*

### 1. House Price Response

### ch2onlineannex - 1. House Price Response

### House Price and Consumption Responses to Monetary Policy Shocks
- Charts plot the differential response of house prices and real private consumption to a 100 basis points change in policy rates between samples split by mortgage/homeownership measures.
- Shaded areas represent 90 percent confidence intervals; diamonds represent statistical significance, p value < 0.1.
- Differential samples shown:
  - Mortgage ownership: above vs below median (red vs blue).
  - Homeownership share: above vs below median (red vs blue).
- Findings highlighted in text:
  - Homeownership rates in isolation do not determine the degree of response to monetary policy shocks; the two groups show no clear differential response on house prices or consumption.

### Effective Mortgage Rates and Fixed-Rate Mortgages (FRMs)
- Online Annex Figure 2.5.3 and accompanying text:
  - Plots responses of effective rates (rates on all outstanding mortgages) to a monetary policy shock depending on prevalence of FRMs.
  - Red/blue lines: response to a 100 bpp monetary policy shock where FRMs are above/below the median.
  - Estimation is in reduced form here due to reduced sample size and loss of first-stage power in IV estimation.
- FRM definition and sample notes:
  - FRMs are defined as all mortgages for which rates do not reset in the following 12 months.
  - FRM is a dummy equal to 1 if the outstanding share of FRM mortgages in the quarter is above sample median; 0 otherwise.
  - Sample includes selected Eurozone economies and the US.
  - Diamonds denote pvalue<0.1.

### Heterogeneity Due to Housing Market Characteristics — Methodology
- Dataset:
  - Unbalanced region-level panel covering 192 regions in 9 countries between 2005:Q1 and 2022:Q4.
- Estimation:
  - Local projections with instrumental variables (LP-IV), augmented with interaction terms and country-time fixed effects, following Aastveit and Anundsen, 2022.
  - Main specification (equation (3)) models cumulative log change in house prices and GDP per capita after h quarters: h = 0, ... ,8.
  - The 2SLS estimate of change in policy rates is denoted D̂policy(c,t).
  - Regional dummy Hc,j,t−4 indicates past values of high population density or high house price overvaluation.
  - Controls Xc,j,t−l include 12 lags of changes in log house prices, GDP per capita, CPI inflation, and population.
  - Region fixed effects μc,jh and country-time fixed effects θc,th included; standard errors clustered at the regional level.
  - Charts report 90 percent confidence intervals.
- Definitions of regional dummies:
  - High population density: region’s density in the top 10th percentile within its country-year distribution.
  - High house price overvaluation: regional PIR deviations from long-term average in the top 25th percentile.
- Asymmetry test:
  - Equation (4) modifies (3) to capture asymmetric effects of monetary policy (tightening vs loosening), interacting instrumented policy changes with regional dummies and separate indicators for tightening vs loosening episodes.

### Heterogeneity — Key Empirical Results (Additional Results)
- Online Annex Figure 2.6.1:
  - Shows differential responses of house prices and real GDP-per-capita to a 100 bpp tightening (blue) and loosening (red) in areas with:
    - High population density (housing supply constrained).
    - High house price overvaluations.
  - Lines represent cumulative response to a 100 bpp tightening or loosening in the policy rate at any quarter; shaded areas indicate the 90th percentile confidence intervals.
- Distributional evidence:
  - Figure 2.6.2 shows distributions of population density and house price overvaluations standardized at country level (mean 0, standard deviation 1).
- Proxy validation:
  - Online Annex Figure 2.6.3 documents correlation between population density in 2019 and the Wharton residential land use regulatory index (WRLURI) from a 2018 US survey.
  - Correlation coefficient is approximately 0.6, supporting population density as a proxy for housing supply constraints.

### Model-based Analysis (TANK Model Extensions and Calibration)
- Model framework:
  - Two-agent New Keynesian (TANK) model extending Iacoviello and Neri (2010) with illiquid housing, long-term debt, collateral constraints, and macroprudential tools such as LTV ratios.
  - Households heterogeneous in discount rates (patient savers, impatient borrowers); housing and consumption goods produced with different technologies.
- Calibration adjustments (based on Chen and others (2023) with modifications):
  - Equity withdrawal and inflation target fixed at 1.5 percent and 2 percent, respectively.
  - Household preferences for housing services set to 0.19 for all households to raise steady state debt as percent of GDP.
  - Monetary policy via a Taylor Rule with smoothness parameter set at 0.95 to mimic peak consumption response.
  - Model exhibits standard-sized impulse responses for GDP, consumption, and inflation.
- Simulations to assess joint impact of LTV limits and FRM share:
  - Baseline scenarios:
    1. Restricted vs not restricted LTV: LTV changes from 0.75 to 0.9 as percentage of housing investments.
    2. High vs low FRM: Share of outstanding loans with a fixed-rate mortgage change from 95% to 70%.
  - Simulation findings (described qualitatively):
    - Less restrictive LTV and low FRM represent highest degree of transmission (strongest consumption response to tightening).
    - High FRM and high LTV indicate a lower degree of transmission (weaker consumption response).

*Source: ch2onlineannex - 1. House Price Response, Online annex material from the IMF World Economic Outlook chapter (excerpts provided).*

---


_Source: https://www.imf.org/-/media/files/publications/weo/2024/april/english/ch2onlineannex.pdf_