## 2.1 Data

## Source details

**Canonical URL:** [2.1 Data](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025207-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025207-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025207-source-pdf.pdf.json)

---

### Sample and coverage
- Quarterly panel dataset covering the period from 1994 to 2022 for 190 U.S. metropolitan statistical areas (MSAs).
- Sample excludes MSAs with populations under 150,000 as of 1990 (16 least populated MSAs dropped); the 190 MSAs represented over 76.7% of total U.S. population in 2022 and spanned 46 States.
- MSAs defined based on the 2019 American Community Survey.
- Detailed dataset description provided in Online Appendix A.

### Housing price, rent, and valuation measures
- Monthly house price indices at the MSA level from Freddie Mac, aggregated to quarterly frequency; indices rebased to 2019 and multiplied by the median home value in each MSA based on the 2019 American Community Survey to express prices in levels.
- Rents constructed from annual county-level series in HUD’s Fair Market Rents dataset, aggregated to the MSA level using county population weights and linearly interpolated to quarterly frequency.
- Both house price and rent series deflated using the national CPI series for urban consumers (U.S. Bureau of Economic Analysis).
- Price-to-rent ratio (PRR) = real house prices ÷ annualized real rents (median monthly real rent for a two-bedroom apartment × 12).
- Data for rents and house prices start in 1983; used to construct backward-looking trends in PRR after 1994.
- Price-to-income ratio (PIR) = median house prices ÷ MSA-level annual personal income per capita (interpolated to quarterly).

### Mortgage origination and buyer composition
- MSA-level home-purchase mortgage originations derived from the Home Mortgage Disclosure Act (HMDA) dataset (sourced from Neil Bhutta’s website).
- Monthly county-level originations aggregated to MSA and quarter; seasonality adjusted with a four-quarter moving average.
- Owner-occupied (OO) share aggregated from county-level OO shares weighted by county originations; non-owner-occupied (NOO) share = 1 − OO share (interpolated to quarterly).
- Volumes of NOO and OO originations computed by multiplying total originations by respective shares.

### Additional controls and supply constraints
- MSA-level controls: total employment, population, share of young population (20–34 age group).
- Housing supply constraints proxied by two vintages of the Wharton Residential Land Use Regulation Index (WRLURI) from Gyourko et al. (2008) and Gyourko et al. (2019), and by housing supply elasticity estimates from Saiz (2010).
- WRLURI measures taken from Chodorow-Reich et al. (2024); Saiz supply elasticities from Aastveit and Anundsen (2022).

### Monetary policy shocks (MPS) used as instruments
- Three instruments for changes in the federal funds rate:
  - MPS BS: Bauer and Swanson (2023) high-frequency shocks (pre-purge unorthogonalized series).
  - MPS RR: narrative shocks from Romer and Romer (2004) extended by Miguel Acosta (residuals from regression on Fed Greenbook forecasts).
  - MPS DM: analysts’ policy rate forecast errors from Bloomberg as collected by De Stefani and Mano (2025) and IMF (2024).
- Orthogonalized Bauer and Swanson shocks were not used due to weak first-stage relevance (reported first-stage F-statistics around 2.6 in other work and lower in the authors’ richer specification).

### Data construction notes and sample statistics
- Observations for main five-year-ahead regressions: 22,300 in each reported column (Table 1 / Table A).
- Appendix A: geographical distribution summary lists regional MSAs and Pop. share (%) and Avg. HP ($) (e.g., All: States 46, MSAs 190, Pop. share (%) 100, Avg. HP ($) 315,991).

---

### 2.2 Empirical approach

#### Predictive regressions for five-year ahead growth
- Baseline specification for five-year ahead growth:
  - ∆t+20 yj,t = β PRRj,t + εj,t
  - ∆t+20 yj,t = log change in real rents or house prices in MSA j from quarter t to t+20.
  - Standard errors clustered at MSA and quarter levels.
  - Results shown without MSA fixed effects to avoid potential look-ahead bias; conclusions robust to including them.
- Decomposition of PRR into trend and deviation:
  - ∆t+20 yj,t = β1 PRR_T j,t + β2 PRR_D j,t + εj,t
  - Trend PRR_T j,t from backward-looking rolling regression over τ ∈ [t−79, t] (80 quarters, 20 years): PRR j,τ = MSAj + Trendτ × MSAj + εj,τ.
  - PRR_D j,t = PRR j,t − PRR_T j,t.
  - Rolling regressions start in 1993 so that backward series using data from 1983 ensure at least 10 years of data.

#### Local projections with instrumental variables (LP-IV) for policy transmission
- Baseline LP-IV dynamic specification:
  - yj,t+h − yj,t−1 = βh1 ∆Ratet × PRRj,t−1 V + βh2 PRRj,t−1 + Σ_{l=1}^4 βh3 Xj,t−l + MSA^h_j + τ^h_t + εj,t+h
  - ∆Ratet × PRRj,t−1 V: interaction between change in the monetary policy rate in quarter t and lagged PRR standardized by its sample standard deviation; instrumented using MPS BS, MPS RR, MPS DM.
  - Controls X: four lags of quarterly log changes in house prices, employment, population, share of young population, and four lags of quarterly changes in share of NOO home-purchase loans.
  - MSA and quarter fixed effects included; standard errors double-clustered at MSA and quarter levels.
  - Projection horizon h up to 16 quarters ahead.
  - Regressions weighted by population in 1990.
- Separate regressions estimated for trend and deviation components to preserve first-stage power; joint estimation used in robustness checks.

#### Composition channel: NOO share as dependent variable
- Same LP-IV specifications estimated with the share of non-owner-occupied home-purchase loan originations as the dependent variable to assess buyer composition responses to monetary policy.
- Rationale: NOO purchases proxy for investor demand; theoretical expectation that more sophisticated investors react faster to fundamentals.

---

### 3 Main results

### 3.1 Relationship between PRR and five-year ahead growth in rents and house prices
- Table 1 estimates (effect of PRR on five-year ahead growth):
  - Column (1): ∆t+20 Rent — PRR coefficient = 0.008 ∗∗∗ (standard error (0.001))
  - Column (2): ∆t+20 HP — PRR coefficient = −0.020 ∗∗∗ (standard error (0.004))
  - Column (3): ∆t+20 Rent — Trend PRR coefficient = 0.007 ∗∗∗ (0.001); Detrended PRR coefficient = 0.011 ∗∗∗ (0.001)
  - Column (4): ∆t+20 HP — Trend PRR coefficient = −0.017 ∗∗∗ (0.003); Detrended PRR coefficient = −0.029 ∗∗∗ (0.005)
  - Constants: Column (1) −0.086 ∗∗∗ (0.011); Column (2) 0.357 ∗∗∗ (0.044); Column (3) −0.068 ∗∗∗ (0.014); Column (4) 0.305 ∗∗∗ (0.036)
  - Observations: 22,300 in each column.
  - Significance notation: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
- Key findings:
  - Elevated PRR predicts stronger future rent growth (ˆβ = 0.008) and predicts negative future house price growth (ˆβ = −0.020).
  - Both trend and detrended components of PRR are positively associated with future rent growth (ˆβ_T = 0.007; ˆβ_D = 0.011, both significant at the 1 percent level).
  - Both trend and detrended components are negatively associated with future house price growth, with the detrended component having a larger negative coefficient (ˆβ_T = −0.017; ˆβ_D = −0.029).
  - Interpretation: larger quantitative effects of temporary deviations from trend point to a role for belief-driven, extrapolative expectations and temporary price overshooting followed by correction; a fully rational explanation would require highly volatile or persistent changes in risk aversion.

### 3.2 Impact of monetary policy on house prices (state-contingent transmission)
- Impulse response summary (per one-standard-deviation increase in PRR, to a 25bps policy rate rise):
  - Across MPS BS, MPS RR, MPS DM, a higher PRR by 1 standard deviation is associated with:
    - Decline in real house prices of about 0.3-0.5 percentage points after 8 quarters.
    - Decline of about 0.7-1 percentage points after 12-14 quarters.
  - Effect is delayed, strengthens over time, and remains statistically significant through the 16-quarter horizon.
  - Shape of impulse responses similar across instruments, with slight variation in magnitude.
- Implication:
  - Monetary policy transmission to house prices is strongly state-contingent: areas with higher PRR exhibit amplified house price sensitivity to monetary policy tightening.

### 3.3 Impact of monetary policy on the composition of buyers (NOO share)
- Key empirical findings:
  - Monetary tightening linked to a larger and more pronounced decline in the share of NOO home-purchase loans in MSAs with higher PRR.
  - Decline is more temporary than the house price effect and partially reverts by quarter 16.
  - Timing and magnitude:
    - Decline becomes statistically significant after around 4–8 quarters.
    - Peaks approximately after 6 quarters at about 0.1–0.15 percentage points (decline) before reversing as NOO share begins to recover.
- Interpretation:
  - Results consistent with models where more sophisticated investors (proxied by NOO buyers) time the market more effectively and exit earlier during monetary tightening, leaving naive buyers who may adjust later; subsequent re-entry by investors occurs when house prices are more depressed.
- Tabulated baseline NOO estimates (Table B1 / B4–B7, h = 6):
  - Baseline (∆Ratet × PRRj,t−1): MPS BS -0.149 ∗∗, MPS RR -0.100 ∗∗, MPS DM -0.117 ∗∗∗.
  - Example first-stage F-statistics reported across tables: 8.3, 39.8, 577.7 (illustrative examples in Appendix tables).

---

### 3.4 Trend versus deviations from trend: decomposing the PRR

#### Main finding
- Amplification of monetary policy effects on real house prices and buyer composition in overvalued housing markets is mainly driven by temporary departures of PRR from its trend rather than by the trend component itself.
- Interaction between a 25 basis point change in the policy rate and the deviation from PRR trend produces larger and more persistent impacts than the interaction with the PRR trend.

#### Evidence and magnitudes (real house prices)
- Interaction coefficients on the deviation from PRR trend are larger and more persistent than those for the trend component; trend coefficients largely flat and insignificant.
- In markets where PRR is one standard deviation above trend, a 25-basis-point policy rate increase leads to an additional 2–3 percentage point decline in real house prices after 12 quarters.
- Baseline LP-IV estimates for horizon h = 12 are -0.88∗∗∗, -0.98∗∗∗, and -0.67∗∗∗ for MPS BS, MPS RR, and MPS DM, respectively.

#### NOO share (composition of buyers)
- NOO share reacts more strongly when monetary policy is interacted with the deviation of overvaluation from trend than when interacted with the trend component.
- Using volumes, following a 25 bps increase in the policy rate, NOO loan originations decline additionally by 2.5-3 percentage points after six quarters in MSAs with a one-standard-deviation higher PRR, whereas OO originations decline by roughly 1 percentage point.

#### Asymmetries in policy effects
- Decomposing ∆Rate into contractionary (∆Rate_t > 0) and expansionary (∆Rate_t < 0):
  - Using MPS RR and MPS DM as instruments, estimated effect of contractionary monetary policy is roughly twice as large as that of expansionary policy, though differences not statistically significant (p-values 0.39 and 0.46).
  - Using short-run deviations from PRR trend shows much larger asymmetry: contractionary policy has significant and large effects while expansionary policy has null effects (appendix evidence).
  - First-stage strength caveat: using MPS BS as an instrument yields a very low F-statistic, precluding testing asymmetries in that specification.

#### Robustness checks and extensions
- Results robust when controlling for housing supply elasticities (WRLURI and Saiz measures); including interaction of monetary policy with supply constraints does not change the negative and significant coefficient on monetary policy interacted with overvaluation.
- Results hold when:
  - Not weighted by population.
  - Including 8 or 12 lags of controls.
  - Sequentially dropping controls.
  - Using alternative PRR (real rent of three-bedroom instead of two-bedroom).
  - Using PIR and its deviation from trend as alternative valuation measures (PIR results: Column (1) PIR coefficient 0.011 ∗∗∗ on ∆t+20 Rent (0.001); Column (2) PIR coefficient -0.032 ∗∗∗ on ∆t+20 HP (0.009); decomposed PIR Detrended effects notably large in Table B8).
- Inference details:
  - Focal policy shock magnitude reported as 25 basis points.
  - Confidence intervals often at 90%; standard errors double-clustered at MSA and quarter levels.
  - Baseline controls: four lags of (log) changes in real house prices, employment, population, share of young population, and NOO share; MSA and quarter fixed effects.

#### Implications and interpretation
- Time-varying shifts in sentiment (short-run deviations from price-to-rent trends) are likely key drivers of the observed amplification effects of overvaluation on monetary policy transmission.
- Suggestive evidence that contractionary monetary policy is at least twice as effective in mitigating housing market overvaluation than expansionary policy is in exacerbating it, particularly when overvaluation defined by short-run deviations from long-term PRR trends.
- Findings consistent with models embedding extrapolative expectations into monetary frameworks and imply monetary policy can play a significant role in stabilizing excessive asset price fluctuations.

*Source: wpiea2025207-source-pdf.*

### 2.1  Data

### 2.1 Data

### Sample and coverage
- Quarterly panel dataset covering the period from 1994 to 2022 for 190 U.S. metropolitan statistical areas (MSAs).
- Sample excludes MSAs with populations under 150,000 as of 1990, dropping the 16 least populated MSAs; the 190 MSAs represented over 76.7% of total U.S. population in 2022 and spanned 46 States.
- Detailed dataset description provided in Online Appendix A.
- MSAs are defined based on the 2019 American Community Survey.

### Housing price, rent, and valuation measures
- Monthly house price indices at the MSA level from Freddie Mac, aggregated to quarterly frequency; indices rebased to 2019 and multiplied by the median home value in each MSA based on the 2019 American Community Survey to express prices in levels.
- Rents constructed from annual county-level series in the U.S. Department of Housing and Urban Development’s Fair Market Rents dataset, aggregated to the MSA level using county population weights and linearly interpolated to quarterly frequency.
- Both house price and rent series deflated using the national CPI series for urban consumers from the U.S. Bureau of Economic Analysis.
- Price-to-rent ratio (PRR) constructed by dividing real house prices by annualized real rents (median monthly real rent for a two-bedroom apartment multiplied by 12).
- Data for rents and house prices start in 1983; used to construct backward-looking trends in PRR after 1994.
- Price-to-income ratio (PIR) constructed as the ratio between median house prices and MSA-level annual personal income per capita, interpolated to quarterly frequency.

### Mortgage origination and buyer composition
- MSA-level home-purchase mortgage originations derived from the Home Mortgage Disclosure Act (HMDA) dataset (sourced from Neil Bhutta’s website).
- Monthly county-level originations aggregated to MSA and quarter; seasonality adjusted with a four-quarter moving average.
- Owner-occupied (OO) share aggregated from county-level OO shares weighted by county originations; non-owner-occupied (NOO) share = 1 − OO share (interpolated to quarterly).
- Volumes of NOO and OO originations computed by multiplying total originations by respective shares.

### Additional controls and supply constraints
- MSA-level controls: total employment, population, share of young population (20–34 age group).
- Housing supply constraints proxied by two vintages of the Wharton Residential Land Use Regulation Index (WRLURI) from Gyourko et al. (2008) and Gyourko et al. (2019), and by housing supply elasticity estimates from Saiz (2010).
- WRLURI measures taken from Chodorow-Reich et al. (2024); Saiz supply elasticities from Aastveit and Anundsen (2022).

### Monetary policy shocks (MPS) used as instruments
- Three measures used as instruments for changes in the federal funds rate:
  - MPS BS: shocks from Bauer and Swanson (2023), based on high-frequency identification of changes in money market futures within a 30-minute window around FOMC announcements (unorthogonalized measure used).
  - MPS RR: narrative shocks from Romer and Romer (2004) extended by Miguel Acosta (isolate unanticipated component via residuals from regression on Fed Greenbook forecasts).
  - MPS DM: analysts’ policy rate forecast errors from Bloomberg as collected by De Stefani and Mano (2025) and IMF (2024), following Checo et al. (2024).
- Orthogonalized Bauer and Swanson shocks were not used due to weak first-stage relevance (reported first-stage F-statistics around 2.6 for orthogonalized shocks in other work and even lower in the authors’ richer specification).

---

### 2.2 Empirical approach

### Predictive regressions for five-year ahead growth
- Baseline specification:
  - ∆t+20 yj,t = β PRRj,t + εj,t
  - ∆t+20 yj,t is the log change in real rents or house prices in MSA j from quarter t to t+20 (five-year ahead growth).
  - Standard errors clustered at MSA and quarter levels.
  - Results shown without MSA fixed effects to avoid potential look-ahead bias; conclusions robust to including them.
- Decomposition of PRR into trend and deviation:
  - ∆t+20 yj,t = β1 PRR_T j,t + β2 PRR_D j,t + εj,t
  - Trend PRR_T j,t obtained from predicted values of a backward-looking rolling regression:
    - PRR j,τ = MSAj + Trendτ × MSAj + εj,τ for τ ∈ [t−79, t] (80 quarters, 20 years).
    - For each quarter t, regression estimated using the previous 80 quarters; PRR_T j,t is the predicted value.
  - PRR_D j,t = PRR j,t − PRR_T j,t.
  - Rolling regressions start in 1993 to ensure at least 10 years of data given house price and rent series back to 1983.

### Local projections with instrumental variables (LP-IV) for policy transmission
- Baseline LP-IV (dynamic response of house prices to monetary policy):
  - yj,t+h − yj,t−1 = βh1 ∆Ratet × PRRj,t−1 V + βh2 PRRj,t−1 + Σ_{l=1}^4 βh3 Xj,t−l + MSA^h_j + τ^h_t + εj,t+h
  - yj,t+h − yj,t−1: cumulative log change in real house prices from t−1 to t+h.
  - ∆Ratet × PRRj,t−1 V: interaction between change in the monetary policy rate in quarter t and lagged PRR standardized by its sample standard deviation; this interaction is instrumented using the three MPS measures.
  - Controls X include four lags of quarterly log changes in house prices, employment, population, share of young population, and four lags of quarterly changes in share of NOO home-purchase loans.
  - MSA and quarter fixed effects included; standard errors double-clustered at MSA and quarter levels.
  - Projection horizon h extends up to 16 quarters ahead.
  - Regressions weighted by population in 1990.

- Separate regressions for trend and deviation components:
  - yj,t+h − yj,t−1 = βh,i1 ∆Ratet × PRR_i j,t−1 V + βh,i2 PRR_i j,t−1 + Σ_{l=1}^4 βh3 Xj,t−l + MSA^h_j + τ^h_t + εj,t+h for i ∈ {T, D}
  - Trend and deviation components estimated separately to preserve first-stage power; robustness checks include joint estimation.

### Composition channel: NOO share as dependent variable
- Same specifications (3) and (4) estimated with the share of non-owner-occupied home-purchase loan originations as the dependent variable to assess buyer composition responses to monetary policy.
- Rationale: NOO purchases proxy for investor demand, which may respond differently to policy than owner-occupier demand (theory suggests more sophisticated investors react faster to fundamentals).

---

### 3 Main results

### 3.1 Relationship between PRR and five-year ahead growth in rents and house prices
- Table 1 estimates (effect of PRR on five-year ahead growth):
  - Column (1): ∆t+20 Rent — PRR coefficient = 0.008 ∗∗∗ (standard error (0.001))
  - Column (2): ∆t+20 HP — PRR coefficient = −0.020 ∗∗∗ (standard error (0.004))
  - Column (3): ∆t+20 Rent — Trend PRR coefficient = 0.007 ∗∗∗ (0.001); Detrended PRR coefficient = 0.011 ∗∗∗ (0.001)
  - Column (4): ∆t+20 HP — Trend PRR coefficient = −0.017 ∗∗∗ (0.003); Detrended PRR coefficient = −0.029 ∗∗∗ (0.005)
  - Constants: Column (1) −0.086 ∗∗∗ (0.011); Column (2) 0.357 ∗∗∗ (0.044); Column (3) −0.068 ∗∗∗ (0.014); Column (4) 0.305 ∗∗∗ (0.036)
  - Observations reported as 22,300 in each column.
  - Significance: ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

- Key findings:
  - Elevated PRR predicts stronger future rent growth (ˆβ = 0.008) and predicts negative future house price growth (ˆβ = −0.020).
  - Both trend and detrended components of PRR are positively associated with future rent growth (ˆβ_T = 0.007; ˆβ_D = 0.011, both significant at the 1 percent level).
  - Both trend and detrended components are negatively associated with future house price growth, with the detrended component having a larger negative coefficient (ˆβ_T = −0.017; ˆβ_D = −0.029).
  - Interpretation: larger quantitative effects of temporary deviations from trend point to a role for belief-driven, extrapolative expectations and temporary price overshooting followed by correction; a fully rational explanation would require highly volatile or persistent changes in risk aversion.

### 3.2 Impact of monetary policy on house prices (state-contingent transmission)
- Impulse response summary (effect per one-standard-deviation increase in PRR, to a 25bps policy rate rise):
  - Across three MPS instruments (MPS BS, MPS RR, MPS DM), a higher PRR by 1 standard deviation is associated with:
    - Decline in real house prices of about 0.3-0.5 percentage points after 8 quarters.
    - Decline of about 0.7-1 percentage points after 12-14 quarters.
  - Effect is delayed, strengthens over time, and remains statistically significant through the 16-quarter horizon.
  - Shape of impulse responses is similar across the three instruments, with slight variation in magnitude.
- Implication:
  - Monetary policy transmission to house prices is strongly state-contingent: areas with higher PRR exhibit amplified house price sensitivity to monetary policy tightening.

### 3.3 Impact of monetary policy on the composition of buyers (NOO share)
- Findings using share of non-owner-occupied (NOO) originations as dependent variable:
  - Monetary tightening is linked to a larger and more pronounced decline in the share of NOO home-purchase loans in MSAs with higher PRR.
  - Decline is more temporary than the house price effect and partially reverts by quarter 16.
  - Timing and magnitude:
    - Decline becomes statistically significant after around 4–8 quarters.
    - Peaks approximately after 6 quarters at about 0.1–0.15 percentage points (decline) before reversing as NOO share begins to recover.
- Interpretation:
  - Results consistent with models where more sophisticated investors (proxied by NOO buyers) time the market more effectively and exit earlier during monetary tightening, leaving naive buyers who may adjust later; subsequent re-entry by investors occurs when house prices are more depressed.

*Source: wpiea2025207-source-pdf - 2.1 Data.*

### 3.4  Trend versus deviations from trend: decomposing the PRR

### 3.4  Trend versus deviations from trend: decomposing the PRR

### Main finding
- The amplification of monetary policy effects on real house prices and the composition of buyers in overvalued housing markets is mainly driven by temporary departures of PRR from its trend rather than by the trend component itself.
- The interaction between a 25 basis point change in the policy rate and the deviation from PRR trend produces larger and more persistent impacts than the interaction with the PRR trend.

### Evidence and magnitudes (real house prices)
- The estimated coefficients on the interaction between the monetary policy shock and the deviation from PRR trend are larger in magnitude and more persistent than those for the trend component; the trend coefficients are largely flat and insignificant.
- In markets where the PRR is one standard deviation above trend, a 25-basis-point policy rate increase leads to an additional 2–3 percentage point decline in real house prices after 12 quarters.
- Baseline LP-IV estimates for the horizon h = 12 are -0.88***, -0.98***, and -0.67*** for MPS BS, MPS RR, and MPS DM, respectively (asterisks denote statistical significance levels reported in the source).

### NOO share (composition of buyers)
- The NOO share of home-purchase loan originations reacts more strongly to monetary policy when interacted with the deviation of overvaluation from trend than when interacted with the trend component.
- Using volumes rather than shares, following a 25 bps increase in the policy rate, non-owner-occupied loan originations decline additionally by 2.5-3 percentage points after six quarters in MSAs with a one-standard-deviation higher PRR, whereas owner-occupied originations decline by roughly 1 percentage point.

### Asymmetries in policy effects
- When decomposing ∆Rate into contractionary (∆Rate_t > 0) and expansionary (∆Rate_t < 0) components:
  - Using MPS RR and MPS DM as instruments, the estimated effect of contractionary monetary policy is roughly twice as large as that of expansionary policy, suggesting a stronger response of house prices to monetary tightening in overvalued markets.
  - The difference between contractionary and expansionary coefficients is not statistically significant in those cases, with p-values of 0.39 and 0.46.
  - When using short-run deviations from trends in the PRR, there is evidence of much larger asymmetry: contractionary policy has significant and large effects while expansionary policy has null effects (see referenced online appendices in the source).
- First-stage strength caveat: when using MPS BS as an instrument, the first-stage regression yields a very low F-statistic, precluding testing for asymmetric effects in that specification.

### Robustness checks and extensions
- Results remain when controlling for housing supply elasticities (WRLURI and Saiz measures). Including interaction of monetary policy with supply constraints does not change the negative and significant coefficient on monetary policy interacted with overvaluation.
- Results hold when:
  - Not weighted by population.
  - Including 8 or 12 lags of controls.
  - Sequentially dropping controls.
  - Using an alternative PRR measure (real rent of three-bedroom instead of two-bedroom apartments).
  - Using the price-to-income ratio (PIR) and its deviation from trend as alternative valuation measures (higher PIR and especially its deviations from trend associate with higher future rent growth but lower future house price growth; monetary policy has larger impacts where PIR deviates from trend).
- Baseline controls include four lags of (log) changes in real house prices, employment, population, share of young population, and share of non-owner-occupied home-purchase loans; MSA and quarter fixed effects; double-clustered standard errors at the MSA and quarter level.
- Online Appendix B.3: results robust when both trend and deviation coefficients are estimated simultaneously.

### Implications and interpretation
- Time-varying shifts in sentiment (short-run deviations from price-to-rent trends) are likely a key driver of the observed amplification effects of overvaluation on monetary policy transmission.
- There is suggestive evidence that contractionary monetary policy is at least twice as effective in mitigating housing market overvaluation than expansionary policy is in exacerbating it, particularly when overvaluation is defined by short-run deviations from long-term price-to-rent trends.
- Findings are consistent with models embedding extrapolative expectations into monetary frameworks and imply that monetary policy can play a significant role in stabilizing excessive asset price fluctuations.

*Source: wpiea2025207-source-pdf - 3.4  Trend versus deviations from trend: decomposing the PRR.*

### References

### wpiea2025207-source-pdf - References

### Bibliographic references (selected)
- Aastveit, Knut Are and André K. Anundsen (2022) “Asymmetric Effects of Monetary Policy in Regional Housing Markets”, American Economic Journal: Macroeconomics, 14 (4), pp. 499–529.
- Abreu, Dilip and Markus K Brunnermeier (2002) “Synchronization risk and delayed arbitrage”, Journal of Financial Economics, 66 (2), pp. 341–360.
- Adam, Klaus, Oliver Pfäuti, and Timo Reinelt (2025) “Subjective housing price expectations, falling natural rates, and the optimal inflation target”, Journal of Monetary Economics, 149, p. 103647.
- Bauer, Michael D. and Eric T. Swanson (2022) “A Reassessment of Monetary Policy Surprises and High-Frequency Identification”, NBER Working Papers, Number: 29939.
- Bauer, Michael D. and Eric T. Swanson (2023) “An Alternative Explanation for the "Fed Information Effect"”, American Economic Review, 113 (3), pp. 664–700.
- Bernanke, Ben S. and Mark Gertler (2001) “Should Central Banks Respond to Movements in Asset Prices?”, American Economic Review, 91 (2), pp. 253–257.
- De Stefani, Alessia and Rui Mano (2025) “Long-Term Debt and Short-Term Rates: Fixed-Rate Mortgages and Monetary Transmission”, IMF Working Papers, Number: 2025/024.
- IMF (2024) “Feeling the Pinch? Tracing the Effects of Monetary Policy Through Housing Markets”, World Economic Outlook, April 2024, Chapter 2.
- Romer, Christina D. and David H. Romer (2004) “A New Measure of Monetary Shocks: Derivation and Implications”, American Economic Review, 94 (4), pp. 1055–1084.
- Saiz, Albert (2010) “The Geographic Determinants of Housing Supply”, The Quarterly Journal of Economics, 125 (3), pp. 1253–1296.

### Appendix A — Data: sources and variable definitions
- Sample and coverage:
  - Quarterly panel covering 1994 to 2022 for 190 U.S. metropolitan statistical areas (MSAs), defined based on the 2019 American Community Survey.
  - MSAs kept represent over 76.7% of total U.S. population in 2022 and span 46 States.
  - Exclusion: MSAs with populations under 150,000 as of 1990.
- Table A1 (geographical distribution summary):
  - East North Central: States 5, MSAs 21, Pop. share (%) 32, Avg. HP ($) 13,784
  - East South Central: States 3, MSAs 12, Pop. share (%) 4, Avg. HP ($) 1,435
  - Middle Atlantic: States 3, MSAs 16, Pop. share (%) 42, Avg. HP ($) 70,204
  - Mountain: States 6, MSAs 13, Pop. share (%) 7, Avg. HP ($) 33,671
  - New England: States 6, MSAs 12, Pop. share (%) 5, Avg. HP ($) 28,702
  - Pacific: States 5, MSAs 11, Pop. share (%) 18, Avg. HP ($) 68,718
  - South Atlantic: States 7, MSAs 39, Pop. share (%) 21, Avg. HP ($) 30,872
  - West North Central: States 7, MSAs 16, Pop. share (%) 5, Avg. HP ($) 32,675
  - West South Central: States 4, MSAs 19, Pop. share (%) 13, Avg. HP ($) 16,867
  - All: States 46, MSAs 190, Pop. share (%) 100, Avg. HP ($) 315,991
- Main series and transformations:
  - Real House Prices: monthly MSA-level Freddie Mac House Price Index (Haver); aggregated to quarterly by monthly averages; scaled by median home value from 2019 ACS; deflated by national CPI for urban consumers (U.S. Bureau of Economic Analysis).
  - Real Rents: annual county-level Fair Market Rent series (U.S. Department of Housing and Urban Development); adjusted to approximate median rent per Kermani and Wong (2021); aggregated to MSA with population weights; deflated by national CPI; linearly interpolated to quarterly.
  - Price-to-Rent Ratio (PRR): ratio of median house price (USD) to annualized median monthly cost of renting a 2-bedroom apartment (12×2-BD Median Rent).
  - Price-to-Income Ratio (PIR): median house price (USD) divided by annual personal income per capita (BEA); annual series interpolated to quarterly.
  - HMDA-based series (from Neil Bhutta’s website):
    - Mortgage purchase origination: monthly county-level series summed to MSA and quarters; seasonality adjusted with a 4-quarter moving average.
    - Non-owner-occupied share (NOO share): computed as (1−OO Share) from owner-occupied share aggregated with origination-weighted averages and interpolated to quarterly.
    - NOO volume and OO volume: products of total mortgage purchase originations and NOO/OO shares.
  - Employment: monthly county-level employment (BLS); quarterly value from last month of each quarter; aggregated to MSA and seasonality-adjusted with 4-quarter moving average.
  - Population: annual MSA population (Census Bureau via Haver), supplemented by Chodorow-Reich et al. (2024); linearly interpolated to quarterly.
  - Share of young population: annual MSA population by age (Census via Haver); share aged 20-34 computed and linearly interpolated to quarterly.
  - Wharton Regulatory Index: two vintages from Gyourko et al. (2008) and Gyourko et al. (2019); MSA-level series obtained from Chodorow-Reich et al. (2024).
  - Saiz elasticity: housing supply elasticities at MSA level from Aastveit and Anundsen (2022).

### Appendix A.2 — Monetary policy measures
- Policy interest rate:
  - Monthly policy rates from FRED; quarterly change in the fed funds rate computed as the sum of the corresponding monthly changes.
- Monetary Policy Shocks (three measures):
  - “MPS BS” (Bauer and Swanson (2023)):
    - High-frequency event-study: market reactions within a 30-minute window around scheduled FOMC announcements.
    - Shocks constructed as first principal component of changes in interest rate futures for contracts covering the next four quarters.
    - Include only scheduled meetings; quarterly values by summing monthly shocks.
    - Two versions available (before and after purging information effects); the pre-purge series is used.
  - “MPS RR” (Romer and Romer (2004), extended by Miguel Acosta):
    - Narrative identification: residuals from regressing change in federal funds rate on Fed Greenbook forecasts of output and inflation.
    - Quarterly values by summing monthly shocks.
    - Sourced from https://github.com/miguel-acosta/RomerRomer2004.
  - “MPS DM” (De Stefani and Mano (2025)):
    - Difference between policy rate decisions and analysts’ policy rate forecasts (median Bloomberg forecasts up to day prior to announcement).
    - Quarterly values by summing monthly shocks.
    - Builds on Checo et al. (2024) and IMF (2024).

### Appendix B — Additional tables, figures, and empirical findings (high-level)
- Scatterplots (PRR vs. future growth):
  - Figures B1 and B2 present binscatter relationships between PRR (and its trend/deviation components) and 5-year ahead growth in real rents and real house prices, as defined in section 2.2.
- LP-IV results: rents and house prices (Figures B3, B4, B8, B10, B12)
  - Figure B3: effect of a 25 basis point change in the policy rate on real rents, conditional on PRR; instruments used: MPS BS, MPS RR, MPS DM; 90% confidence intervals computed with double-clustered standard errors (MSA and quarter); controls include four lags of (log) changes in real house prices, real rents, employment, population, share of young population and NOO share; MSA and quarter fixed effects included.
  - Figure B4: joint estimation decomposing amplification by PRR trend and PRR deviation components (equation (7)); outcomes: real house prices and NOO share; instruments: MPS BS, MPS RR, MPS DM; red line = trend component (ˆβh,T1), blue line = deviation component (ˆβh,D1); 90% confidence intervals with double-clustered SE.
  - Figure B8: estimates of equation (6) — effect of a 25 basis point change in the policy rate on real house prices conditional on PRR and controlling for supply constraints (WRLURI 2006); instruments: MPS BS, MPS RR, MPS DM.
  - Figures B10 and B12: robustness using alternative PRR (3-BD rents) and PIR as measures of overvaluation; same LP-IV setup and instrument sets.
- Asymmetries and heterogeneity (Figures B5–B7)
  - Figure B5: binscatters of changes in policy rate vs. MPS series (BS, RR, DM) showing separate linear fits for expansionary (negative) and contractionary (positive) changes; BS shows slope flattening in contractionary case (consistent with a weaker first-stage).
  - Figures B6 and B7: asymmetric effects of monetary policy by decomposing ∆Ratet into contractionary and expansionary components (equation (5)); outcome: real house prices conditional on PRR and PRR deviation from trend; instruments: MPS BS, MPS RR, MPS DM.
- Transmission to mortgage origination volumes (Figure B9)
  - Figure B9: effect of a 25 basis point change in the policy rate on NOO and OO home-purchase mortgage originations, conditional on PRR; instruments: MPS BS, MPS RR, MPS DM; controls include lags of loan originations and other controls; MSA and quarter FE included.
- Key tabulated estimates and robustness (Tables B1–B8)
  - Table B1 (Dependent variable: log(NooSharej,t+6) − log(NooSharej,t−1)):
    - Baseline estimates (∆Ratet × PRRj,t−1): MPS BS -0.149 ∗∗, MPS RR -0.100 ∗∗, MPS DM -0.117 ∗∗∗ (standard errors reported in table).
    - Supply constraint interaction (∆Ratet × WRL2006j): coefficients reported (e.g., -0.036, -0.051 ∗∗, -0.039 ∗∗).
    - Asymmetry decomposition: entries for ∆Rateexpt and ∆Ratecontrt reported (e.g., ∆Ratecontrt × PRRj,t−1 = -0.295 ∗∗∗ for one column).
    - Observations and sample sizes: Observations 17,200; 17,014; 15,165 across some columns; MSAs 190 or 183 depending on column. First stage F-stat examples: 8.3, 39.8, 577.7.
  - Table B2 (Dependent variable: log(HPj,t+12) − log(HPj,t−1)):
    - Baseline (∆Ratet × PRRj,t−1) estimates: examples include -0.809 ∗∗∗, -0.897 ∗∗∗, -0.570 ∗∗∗ across instruments.
    - Interactions with WRL2006, WRL2018, Saiz elasticity reported (with coefficients and significance).
    - Observations example: 16,556; MSAs 183; First stage F-stat examples: 7.2, 17.0, 423.2.
  - Table B3 (Dependent variable: log(NooSharej,t+6) − log(NooSharej,t−1)) — supply elasticity robustness:
    - Baseline (∆Ratet × PRRj,t−1) examples: -0.135 ∗, -0.081 ∗, -0.101 ∗∗∗.
    - Interactions with WRL2006, WRL2018, Saiz reported.
    - Observations examples: 16,556; MSAs 183; First stage F-stat examples: 7.2, 17.0, 423.2.
  - Table B4–B7: robustness checks (weighting, lags, excluding controls) with reported baseline estimates for house prices and NOO share:
    - Baseline house price estimates (equation 3, h = 12): -0.88 ∗∗∗, -0.98 ∗∗∗, -0.67 ∗∗∗ (reported in notes).
    - Baseline NOO share estimates (equation 3, h = 6): -0.149 ∗∗, -0.100 ∗∗, -0.117 ∗∗∗ (reported in notes).
    - Robustness panels vary population weighting, number of lags (8, 12), exclusion of specific controls (NOO share, employment, population, fraction young).
    - First stage F-stat examples across tables: 8.0, 40.1, 568.2; 8.3, 39.8, 575.6; 8.3, 39.8, 577.0; 8.1, 41.8, 572.1.
  - Table B8 (PIR effects on Rent and House Price Growth):
    - Column (1): PIR coefficient 0.011 ∗∗∗ on ∆t+20 Rent (standard error 0.001).
    - Column (2): PIR coefficient -0.032 ∗∗∗ on ∆t+20 HP (standard error 0.009).
    - Decomposed: PIR Trend 0.009 ∗∗∗ (Rent) and -0.021 ∗∗∗ (HP); PIR Detrended 0.024 ∗∗∗ (Rent) and -0.117 ∗∗∗ (HP).
    - Observations 22,300 in each reported column.
- Notes on inference and estimation:
  - Many figures and tables report the effect of a 25 basis point change in the policy rate as the focal policy shock magnitude.
  - Confidence intervals: 90% intervals frequently displayed; standard errors are double-clustered at the MSA and quarter level.
  - Controls commonly include four lags of (log) changes in real house prices, employment, population, share of young population and the share of non-owner-occupied home-purchase loans; MSA and quarter fixed effects included throughout.

*Content unit derived from the PDF chapter "References" and associated Appendix material in the source.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025207-source-pdf.pdf_