## wpiea2020291-print-pdf

## Source details

**Canonical URL:** [wpiea2020291-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020291-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020291-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020291-print-pdf.pdf.json)

---

### Role of house prices in the macroeconomy
- Housing is a key constituent of the household balance sheet and cyclical house price movements affect non-human wealth, household spending and borrowing, aggregate consumption, investment, and financial stability.
- Demand shocks and inelastic/sluggish supply can amplify wealth effects and facilitate rapid asset-bubble build-up.
- Refinancing of mortgages and procyclical increases in household borrowing capacity can magnify wealth effects and induce consumption booms.

### Empirical patterns, urbanization, and data contributions
- Urbanization:
  - Low and middle-income countries: urban population share rose from about 33 percent in the 1980s to 48 percent in the 2010s.
  - Upper-middle-income countries: urban population share rose from 39 percent to 63 percent in the same period.
- Cross-country evidence:
  - Positive relationship observed between credit growth in the non-financial sector and house price growth (regression y = 0.06x + 5.25; R² = 0.68 for 15 major EMEs, 2000 Q1 to 2019:Q4).
  - Positive relationship between housing prices and growth in mortgages (Sample 101 countries; NUMBEO).
- Data and methodological contributions:
  - Uses two novel, granular city-level house-price databases for India: Reserve Bank of India (RBI) HPI (10 cities, 2010-11 base) and NHB Residex (33/50 city coverage, 2017-18 base series backward-linked).
  - City-wise indicators: median LTV ratio, loan-to-income (LTI) ratio, EMI-to-income (ETI) ratio, asset provisioning, and risk weights.
  - Empirical period: annual data for 2008-09 to 2018-19 (some monthly/quarterly series averaged to annual).
  - Estimators: dynamic panel data (Arellano–Bond and Blundell–Bond) and fixed-effects (FE); diagnostics include Sargan and Arellano–Bond tests.

### Analytical framework and elasticities
- Reduced-form and dynamic panel specification link house prices to demand-side drivers (income, population, credit), supply-side drivers (construction cost, land), opportunity costs (relative asset returns), and macroprudential covariates (ltv, risk weights, asset provisioning).
- Elasticity sign conventions posited:
  - ∂p/∂y > 0; ∂p/∂pop > 0; ∂p/∂crd > 0; ∂p/∂rm < 0; ∂p/∂r < 0; ∂p/∂rp < 0.
  - Supply shocks: ∂p/∂concos > 0; ∂p/∂tax < 0.
  - Macroprudential: ∂p/∂ltv > 0; ∂p/∂rw < 0; ∂p/∂asstprov < 0.

### Stylized facts for India
- Urbanization increased from 23 percent to 34 percent between 1980 and 2018.
- Buildings and housing ≈ 45 percent of urban households’ total asset holdings (NSSO, 2014).
- Household fixed investment in dwellings ≈ 30 percent of real gross fixed capital formation in recent years.
- Housing loan portfolio of commercial banks ≈ 15 percent of total loan portfolio.
- Post-global financial crisis: nominal house price growth largely above nominal income growth until 2015-16; thereafter house price growth remained below nominal income growth.
- Cross-city dispersion in house price growth has narrowed over time for both RBI and NHB indices.

### Key estimated coefficients — nominal house prices (average of specifications)
- Backward-looking behavior:
  - Log of nominal HPI(t-1): 0.29 (Arellano–Bond), 0.30 (Blundell–Bond).
- Income and demographics:
  - Log of city real per capita income(t-1): 0.33, 0.16.
  - Log of city population (t-1): 0.53, 0.16.
- Supply-side (construction cost):
  - Log of nominal construction cost Index: 1.10, 1.31.
  - Log of nominal construction cost Index (t-1): 0.83, 1.11.
- Rent/user cost:
  - Log of housing rentals: 0.46, 0.36.
- Opportunity cost / asset substitution:
  - Log of the ratio of BSE Sensex to nominal HPI: -0.63, -0.55.
- Credit variables and rates:
  - Log of per account nominal bank credit for residential housing: 0.14 (reported); (t-1): 0.23, 0.15.
  - Real weighted average interest rate on banks’ housing loans (t-1): -0.07, -0.07.
  - RBI’s policy rate: -0.02, -0.04.
- Macroprudential tools (nominal):
  - Log of sample median LTV ratio: 0.40, 0.69.
  - Log of regulatory LTV for large-sized loans: 1.03, 0.94.
  - Log of regulatory LTV for small-sized loans: 0.55, 0.70.
  - Log of regulatory risk weights: -0.14, -0.14; (t-1): -0.21, -0.21.
  - Log of median LTI ratio: 0.51, 0.81.
  - Log of EMI-to-income ratio: -0.40, -0.63.
  - Log of asset provisioning for banks’ standard housing assets: -0.20, -0.13.
- Magnitude examples:
  - A 10-percentage point loosening of the median LTV ratio leads to an average 4-7 percent increase in nominal house prices via direct credit supply effects.
  - Unit tightening of risk weights on housing loans causes about 14-21 basis point decline in house prices; tightening asset provisioning causes a similar 13-20 basis point effect.

### Key estimated coefficients — real house prices (average of specifications)
- Backward-looking behavior:
  - Log of real HPI(t-1): 0.24, 0.36.
- Income and demographics:
  - Log of city real per capita income (t-1): 0.25, 0.25.
  - Log of city population (t-1): 0.40, 0.12.
- Supply-side (real construction cost):
  - Log of real construction cost Index: 0.96 (Arellano–Bond reported).
  - Log of real construction cost Index (t-1): 0.29, 0.91.
- Rent/user cost:
  - Log of real housing rentals: 0.33, 0.39.
- Substitution effect:
  - Log of ratio of real BSE Sensex to real HPI: -0.57, -0.55.
- Credit and rates:
  - Log of real per account nominal bank credit for residential housing: 0.15, 0.14; (t-1): 0.18 (reported).
  - Real weighted average interest rate on banks’ housing loans: -0.01, -0.02.
  - RBI’s real policy rate: -0.03, -0.02; (t-1): -0.02.
- Macroprudential tools (real effects):
  - Log of sample median LTV ratio: 0.41, 0.44; (t-1): 0.42.
  - Regulatory LTV for large-sized loans: 1.10, 1.14.
  - Regulatory LTV for small-sized loans: 0.40, 0.59.
  - Regulatory risk weights: -0.15, -0.14; (t-1): -0.15.
  - Log of median LTI ratio: 0.45; (t-1): 0.56.
  - Log of housing EMI-to-income ratio: -0.70.
  - Log of asset provisioning: -0.16, -0.10.
- Interpretation:
  - Macroprudential measures (LTV, risk weights, asset provisioning) have strong moderating impacts on real house prices.
  - Differentiated LTV: effects are larger for large-sized loans (1.10–1.14) than for small-sized loans (0.40–0.59), consistent with larger investment/speculative motive in larger loans.
  - Higher loan leverage (LTI) and lower affordability (higher EMI-to-income) meaningfully affect real house prices.

### Results from NHB Residex (33 cities) — summary
- Nominal house price dynamics (Table 3 averages; all coefficients significant at the 5% level):
  - Log of nominal HPI(t-1): 0.29 / 0.42.
  - Log of city real per capita income(t-1): 0.28 / 0.21.
  - Log of nominal construction cost Index: 0.59 / 0.67; (t-1): 0.49 / 0.59.
  - Log of sample median LTV ratio: 0.40 / 0.66.
  - Log of regulatory LTV for large-sized loans: 0.59 / 0.60.
  - Log of regulatory LTV for small-sized loans: 0.37 / 0.20.
  - Log of regulatory risk weights: -0.14 / -0.14; (t-1): -0.15 / -0.11.
  - Log of median LTI ratio: 0.57 / 0.58.
  - Log of housing EMI-to-income ratio: (blank) / -0.75; (t-1): (blank) / -0.31 / -0.25.
  - Log of asset provisioning: -0.17 / -0.11.
- Real house price dynamics (Table 4 averages; all coefficients significant at the 5% level):
  - Log of real house price(t-1): 0.31 / 0.42.
  - Log of city real per capita income(t-1): 0.28 / 0.21.
  - Log of real construction cost Index: 0.37 / 0.55; (t-1): 0.63 / 0.65.
  - Log of sample median LTV ratio: 0.56 / 0.73.
  - Log of regulatory LTV for large-sized loans: 0.76 / 0.80.
  - Log of regulatory LTV for small-sized loans: 0.50 / 0.41.
  - Log of regulatory risk weights: -0.11 / -0.16; (t-1): -0.13 / -0.17.
  - Log of median LTI ratio: 0.18 / 0.74.
  - Log of housing EMI-to-income ratio: -0.40 / (blank); (t-1): (blank) / -0.23.
  - Log of asset provisioning: -0.10 / -0.12.
- Credit-channel evidence:
  - Both LTV ratios (large and small) strongly affect real housing credit demand.
  - Increases in risk weights and asset provisioning have significant negative effects on real housing credit demand.

### Panel fixed-effect model diagnostics (Appendix Table A6 — RBI HPI)
- No. of groups: 10; T ranges from 9 to 11 across specifications [1]–[11].
- R2 (within) ranges: 0.87 to 0.96 across models.
- lnddppc (real per capita income) positive and significant across models (examples: 0.22**, 0.29***, 0.58**).
- lnrbse_rhpi10 (relative stock price ratio) negative and strongly significant across models (examples: -0.77***, -0.90**, -0.97***).
- lnltvm / lnltvl / lnltvs (various LTV measures) positive and frequently significant (examples: lnltvm 0.73***; lnltvl 1.63*; lnltvs 0.58***).
- lnrwm (regulatory risk measure) negative and highly significant (examples: -0.20***, -0.24***).
- Monetary policy rate (rrepo) generally negative and significant where included (examples: -0.06***, -0.05***).
- Credit variables (lnrcrdpa) positive and significant in some specifications (example: 0.11**).

### Policy implications and recommendations
- Macroprudential tools are effective countercyclical instruments in India:
  - Tightening LTV ratios, raising regulatory risk weights, and increasing asset provisioning dampen nominal and real house price growth.
  - Differentiated LTVs targeting large-sized loans can better restrain investment/speculative demand given larger estimated effects on house prices.
- Monetary policy affects housing and credit markets but transmission can be impeded by credit market imperfections (maturity mismatches, interest rate rigidities, competition from other saving instruments, loan portfolio deterioration).
- Supply constraints matter: increases in construction costs have large positive effects on house prices; small slowdowns in supply can lead to more than proportionate increases in nominal house prices.
- Financial stability: as mortgage markets deepen and housing comprises a larger share of household assets, procyclicality and systemic risk increase—active macroprudential management is warranted.
- Robustness: findings are robust across dynamic panel and fixed-effects specifications; instruments and diagnostics support validity.

_Italic source: wpiea2020291-print-pdf_

### REFERENCES

### REFERENCES

### Role of house prices in the macroeconomy
- Housing asset is a key constituent of the household balance sheet and cyclical movements in house prices have a profound impact on non-human wealth.
- Significant movements in house prices affect the stock of household wealth and thereby influence spending and borrowing decisions of households, aggregate consumption, and investment.
- House price behavior affects financial stability by impacting the profitability and balance sheets of banks and other financial intermediaries because mortgages constitute an important component of their asset portfolio.
- Demand shocks from macroeconomic factors can cause price surges; inelastic or sluggish supply can amplify wealth effects and facilitate rapid asset-bubble build-up.
- Refinancing of mortgages in developed markets can magnify wealth effects and induce consumption booms.
- Procyclical increases in household borrowing capacity may amplify collateral-based spending cycles (Almeida et al., 2006).

### Empirical patterns, urbanization, and cross-country evidence
- Rapid urbanization and expansion of housing markets in developing countries have shifted demand outward:
  - Low and middle-income countries: urban population share rose from about 33 percent in the 1980s to 48 percent in the 2010s.
  - Upper-middle-income countries: urban population share rose from 39 percent to 63 percent during the same period.
- Importance of asset prices in EMEs has risen with housing market development and greater household participation.
- Historical pattern: significant growth in real house prices in the pre-global financial crisis period was generally followed by sizeable deceleration or decline in the post-crisis period, indicating strong house price cycles in EMEs.
- Cross-country relationships highlighted in the source:
  - Figure 1 (15 major EMEs during the period 2000 Q1 to 2019:Q4) shows a positive relationship between credit growth in the non-financial sector and house price growth; regression: y = 0.06x + 5.25; R² = 0.68. Source: BIS database.
  - Figure 2 (Sample 101 countries) shows a positive relationship between housing prices and growth in mortgages. Source: NUMBEO.

### Objectives, data, and contributions of the paper
- Primary aims:
  - Explore the effectiveness of macroprudential policies in affecting house prices after controlling for fundamental drivers.
  - Examine the effectiveness of differential risk assigned on various mortgage sizes.
- Data and methodological contributions:
  - Utilizes novel, granular city-level house-price databases for India.
  - Employs city-wise indicators including median LTV ratio, loan-to-income ratio, and EMI ratio.
  - Uses granular data to examine the effectiveness of LTV ratio and risk weights prescribed for different sizes of mortgages on house prices.
- Policy relevance:
  - Focuses on macroprudential policies in a developing economy (India) that has implemented countercyclical macroprudential tools to dampen financial-system procyclicality and address systemic risks.
  - Emphasizes assessing macroprudential efficacy while giving due recognition to long-run fundamental drivers of housing prices.

*wpiea2020291-print-pdf - REFERENCES*

### Section IV. Section V analyses empirical results utilizing two separate novel databases on

### wpiea2020291-print-pdf - Section IV. Section V analyses empirical results utilizing two separate novel databases on housing prices in India

### II. THEORY AND EMPIRICAL LITERATURE
- Housing and financial assets influence lifetime wealth and consumption behavior; foundations in the permanent income hypothesis (Friedman, 1957) and life cycle hypothesis (Modigliani and Brumberg, 1954 and Ando and Modigliani, 1963).
- Households smooth consumption intertemporally; asset pricing models assume decisions compare marginal benefit of current consumption against expected asset returns.
- Main conduits of changes in household wealth: changes in household preference for holding an asset and price movements.
- Early literature focuses on house prices relative to present value of rents, construction cost, and fundamentals: incomes, demography, labour market, bank credit, mortgage cost.
- Co-movement of housing market and macroeconomy documented (Ito, 1993; Bowen, 1994; Green, 1997; Baffoe-Bonnie, 1998; Wen, 2001; Seko, 2003; Hwang and Quigley, 2004).
- Jud and Winkler (2002): real housing price appreciation influenced by population, real incomes, construction costs and interest rates.
- Case (2000) and Catte et al. (2004): transmission of macro shocks (credit supply, interest rates) to house prices with lag depending on propagation speed.
- Adams and Fuss (2010): short-run inelastic house supply → rise in economic activity raises housing demand → higher user costs and housing prices.
- Agnello and Schuknecht (2011): higher growth of personal income associated with higher probability of housing boom; lower growth with higher probability of bust.
- Other determinants: below-average long-run supply responsiveness, generous tax relief, stricter rent control (Geng, 2018).
- Demography as long-term determinant of house prices (Fitzpatrick and McQuinn, 2004; Terrones and Otrok, 2004; Ahearne et al., 2005; Egert and Mihaljek, 2007); Takáts (2012) finds robust positive relationship between population size and real housing prices.
- In developing countries, rising working-age population may:
  - improve per capita incomes and housing affordability;
  - reduce dependency ratio, generate savings, increase investible resources.
- Singh (2019): strong positive impact of increasing working-age population on real housing prices in EMDEs.
- Housing supply long-run elasticity view: price determined by cost factors; quantity by demand (Blackely, 1999). Somerville (1999): construction cost endogenous; material and labour cost shares ≈ 65 and 35 percent.
- Blackely (1999): long-run elasticity of residential construction ≈ 0.8 to input price changes.
- Higher returns on substitute assets (stocks) may shift demand away from housing; Égert and Mihaljek (2007) find negative correlation between house prices and equity prices in OECD.

### II.1 Credit, House Prices and Macroprudential Policies
- Financial factors (credit conditions, interest rate) influence mortgage access for housing investment.
- Adams and Fuss (2010): higher interest rates raise bond yields and opportunity cost of investing in real estate; higher mortgage rates reduce demand for housing credit and house prices.
- Monetary policy affects wholesale funding cost for banks → mortgage cost and housing credit demand.
- Taylor (2007): housing boom-bust linked to deviation of short-term interest rate path from appropriate level.
- Credit channel influences supply of new housing stock, housing prices, housing wealth.
- Financialization, mortgage credit, and leverage associated with housing boom-bust cycles; larger share of housing loans in financial institutions’ assets intensifies credit-housing nexus.
- Availability of bank credit identified as key driver of house prices (Collyns and Senhadji, 2002; Fitzpatrick and McQuinn, 2004; Tsatsaronis and Zhu, 2004).
- Reinhart (2012): boom in real house prices fuelled by ample domestic credit, large capital inflows, easy liquidity.
- Singh and Nadkarni (2020): in EMEs, both real stock prices and house prices rise with an expansionary credit shock; impact significant and persistent for house prices.
- Macroprudential policy rationale:
  - Monetary policy blunt in arresting asset price growth → macroprudential framework evolved to respond to credit and asset price boom-busts.
  - Tools protect bank balance sheets during booms and downturns.
- Macroprudential tools include: loan-to-value (LTV) ratio, risk weights on housing loans in capital-asset ratio, provisioning/dynamic provisioning, debt-to-income (DTI) ratio.
- LTV ratio: common, imposes hard-borrowing constraint on households to limit borrowing against collateral; countercyclical tool to mitigate credit-fed house price booms.
- DTI ratio: limits homebuyer leverage and moderates demand for credit-fueled residential properties.
- Effectiveness: tools can be unwound during recessions to enhance credit access; asymmetric effectiveness across boom/bust possible.
- Empirical literature finds positive relationship between LTV limits and house price changes (Crowe et al., 2011; IMF, 2011; Igan and Kang, 2011; Lim et al., 2011; Wong et al., 2011; Claessens et al., 2013; Cerutti et al., 2015; Morgan et al., 2015; Ozge and Olmstead-Rumsey, 2015).
- Lim et al. (2011), Wong et al. (2011), Crowe et al. (2011): LTV reduces growth in credit and asset prices.
- Vandenbussche et al. (2012): capital ratio requirements and non-standard liquidity measures slow house price inflation.
- Other supportive evidence: Igan and Kang (2011); Jiménez et al. (2012); Kuttner and Shim (2012); Arregui et al. (2013); Tressel and Zhang (2016); Richter et. al. (2018).
- Zhang and Zoli (2014): LTV, DTI, risk weights, loan loss provisions important in containing mortgage loan growth in Asia.
- Ozge and Olmstead-Rumsey (2018): only housing-related macroprudential policies constrain housing credit growth and house price appreciation.
- Alam et. al. (2019): strong and nonlinear effects of LTV changes on household credit; effects of LTV tightening diminish with adjustment size, possibly due to leakage.
- Literature trend: macroprudential tools generally effective in limiting excessive mortgage credit expansion and housing boom-bust risks, though some studies (Vandenbussche et al., 2012; Duffy, 2012) provide inconclusive evidence on LTV effectiveness.

### III. ANALYTICAL FRAMEWORK
- Portfolio choice life cycle model: households receive labour income and allocate resources across consumption goods, financial assets, durable goods (houses); utility from housing services (w_H) and non-housing consumption (w_NH).
- Intertemporal utility in period τ expressed as: U [γ (w_Hτ, w_NHτ)]           (1)
- Housing market equilibrium when demand equals supply:
  - D_H ƒ(Zi) = S_H ƒ(Xi, Yi)           (2)
  - D_H = demand function; Zi = vector of demand-side covariates; S_H = supply function; Xi = macroeconomic covariates of supply; Yi = institutional factors affecting supply.
- Long-run demand drivers: household disposable income; demographic characteristics (size, density of urban population, workforce participation); tax exemptions; mortgage interest rates; debt financing flexibility; returns on competing assets.
- Supply-side drivers: availability and cost of land, construction cost, tax incentives, regulatory framework.
- Macroprudential policies (LTV, DTI, risk weights) affect cost and quantity of credit, thereby impacting housing demand.
- Reduced-form model posited:
  - ∑p_Hi(t)_{i=1}^n = φ ∑Z_i(t)_{i=1}^n + 1/λ ∑X_i(t)_{t=1}^T + 1/θ ∑Y_i(t)_{t=1}^T           (3)
- Elasticities of house prices to key macro variables:
  - ∂p/∂y > 0
  - ∂p/∂pop > 0
  - ∂p/∂crd > 0
  - ∂p/∂rm < 0
  - ∂p/∂r < 0
  - ∂p/∂rp < 0
  - where y = household income; pop = city population; rm = mortgage interest rate; r = monetary policy rate; rp = relative asset price return.
- Supply shock elasticities:
  - ∂p/∂concos > 0
  - ∂p/∂tax < 0
  - where concoix = construction cost; tax = tax concessions for affordable housing.
- Macroprudential covariates elasticities:
  - ∂p/∂ltv > 0
  - ∂p/∂rw < 0
  - ∂p/∂asstprov < 0
  - where ltv = LTV ratio ceilings; rw = risk weights; asstprov = asset provisioning.
- Dynamic panel data (DPD) estimator to address endogeneity and reverse feedback by including lags of dependent variable and panel-level effects:
  - P_it = ∑_{j=1}^k δ_k y_{i,t-j} + γ X_it + θ Z_it + φ Π_it + υ_i + ε_it, i = 1, ...k and t = 1, ..., n           (4)
  - P_it = house price in city i at time t; X_it = macro covariates; Z_it = supply-side variables; Π_it = macroprudential covariates; υ_i = panel effects; ε_it = i.i.d. low-order MA process with variance σ^2_ε.

### IV. DATA ISSUES
- Granular data scarcity on housing prices and stocks in developing countries; Indian city-level databases are relatively recent.
- Chosen indices for empirical exercise: Reserve Bank of India (RBI) House Price Index (HPI) and National Housing Bank (NHB) Residex (see Annex A1 for sample coverage).
- RBI HPI:
  - Weighted average price index using Laspeyres’ method with 2010-11 as base year.
  - Covers 10 major cities: Mumbai, Delhi, Chennai, Kolkata, Bengaluru, Lucknow, Ahmedabad, Jaipur, Kanpur, Kochi.
  - Computed from transaction data from housing registration authorities in ten major cities.
  - Construction steps:
    - Simple average of houses in each category (small, medium, large) by ward/zone per quarter based on floor space area (FSA).
    - Proportion of houses transacted in three FSA categories within ward/zone during base year used as weights.
    - Quarterly ward/zone weighted average price relatives computed.
    - Weighted relative prices averaged using proportion of houses transacted in each ward to total number of houses transacted in city during base period as weights.
  - Formula for city-wise price indices for each category: HPItj = Σ(RPijt*wj), where i = category (small/medium/large) and n = number of ward/zone in each city. City-wise category indices averaged using population proportion of ten cities to obtain all-India index.
- RBI Residential Asset Price Monitoring Survey (RAPMS) since July 2010 for housing loans disbursed by select banks/HFCs across 13 major cities: Mumbai, Chennai, Delhi, Bengaluru, Hyderabad, Kolkata, Pune, Jaipur, Chandigarh, Ahmedabad, Lucknow, Bhopal, Bhubaneswar.
  - Survey indicators: (i) loan to value (LTV) ratio; (ii) EMI-to-income (ETI) ratio; (iii) house price to income (HPTI) ratio; (iv) loan to income (LTI) ratio.
- NHB Residex:
  - Initially computed with financial year 2012-13 base; new series with base 2017-18 introduced; 2012-13 series backward-linked for continuity.
  - Geographical coverage: 50 cities including 18 State/UT capitals.
  - Three indices: registered prices (SROs), assessment prices (valuation data from primary lending institutions), market prices for under-construction properties (primary & secondary market survey).
- Variables used in study:
  - RBI's HPI for 10 major cities (2010-11 base)
  - NHB Residex for 33 major cities (2017-18 base)
  - real district/state domestic product (SDP) per capita (SDP used for particular city where domestic product not available)
  - real house price index (city house price index deflated by CPI index for respective city/state)
  - city population size in millions
  - construction cost index with base year 2011-12
  - state-wise CPI for housing (to capture housing rent)
  - average per account domestic commercial bank loans for residential housing
  - weighted average interest rate on housing loans of commercial banks
  - relative stock price index (BSE Sensex or Nifty ÷ HPI * 100) and relative stock price index relative to Residex (BSE Sensex or Nifty ÷ Residex * 100)
  - RBI's policy repo rate
  - median LTV ratio for housing loans in India
  - LTV ratio for large-size loans
  - LTV ratio for small-size loans
  - asset provisioning ratio on housing loans of commercial banks
  - median loan to income (LTI) ratio
  - median equal monthly loan installment (EMI) to income ratio
- Construction cost index variants:
  - Building cost index (Base October 2007=100) compiled by CEIC for select Indian cities.
  - For remaining cities, authors construct index (methodology details in source).
- LTV loan size definitions (2017 RBI guidelines):
  - small-sized loans correspond to lowest slab up to Rs. 3 million
  - large-sized loans relate to loans of Rs. [text truncated in source]

*Italic source: wpiea2020291-print-pdf - Section IV. Section V analyses empirical results utilizing two separate novel databases on housing prices in India*

### 7.5 million or more.

### 7.5 million or more.

### Data and methodology
- Annual data for the period 2008-09 to 2018-19; some monthly or quarterly series averaged to annual.
- Data sources: Database on Indian Economy of the Reserve Bank of India; Basic Statistical Returns (BSR) of scheduled commercial banks in India, RBI; UN population database; NHB Residex database; CEIC; Central Statistical Office, Government of India.
- Construction cost index used as a weighted index of materials used in construction (metallic mineral products, and non-metallic mineral products); excludes cost of land.
- Nominal variables deflated using the CPI index.
- Estimation approach: dynamic panel data (DPD) estimators (Arellano–Bond and Blundell–Bond) and fixed-effects (FE) estimators; unrestricted lag structures to model sluggish adjustment; Sargan test supports validity of instruments; Arellano–Bond test does not reject no second-order serial correlation.
- Variable definitions and summary statistics provided in Annex A2 and Annex A3 (referenced in source).

### Stylized facts
- Urbanization in India rose from 23 percent to 34 percent between 1980 and 2018.
- Buildings and housing constitute about 45 percent of the total asset holdings of urban households (NSSO, 2014).
- Household fixed investment in dwellings, buildings and structures ≈ 30 percent of real gross fixed capital formation in recent years.
- Housing loan portfolio of commercial banks ≈ 15 percent of their total loan portfolio.
- Post-global financial crisis period: nominal house price growth largely above nominal income growth until 2015-16; thereafter house price growth remained below nominal income growth.
- Cross-city dispersion in house price growth has significantly narrowed over time for both RBI and NHB indices.

### Nominal house prices — fundamental drivers and macroprudential policies (summary of DPD results)
- Lagged house prices: evidence of backward-looking behaviour; Log of nominal HPI(t-1) coefficients: 0.29 (Arellano–Bond), 0.30 (Blundell–Bond).
- Income/liquidity: Log of the city real per capita income(t-1) coefficients: 0.33, 0.16.
- Demographics: Log of the city population (t-1) coefficients: 0.53, 0.16.
- Supply-side (construction cost): Log of nominal construction cost Index coefficients: 1.10, 1.31; Log of nominal construction cost Index (t-1): 0.83, 1.11.
- Rent/user cost: Log of housing rentals coefficients: 0.46, 0.36.
- Opportunity cost / asset substitution: Log of the ratio of BSE Sensex to nominal HPI coefficients: -0.63, -0.55.
- Credit variables:
  - Log of per account nominal bank credit for residential housing: 0.14 (Arellano–Bond reported), Log of per account nominal bank credit for residential housing(t-1): 0.23, 0.15.
  - Real weighted average interest rate on banks’ housing loans (t-1): -0.07, -0.07.
  - RBI’s policy rate: -0.02, -0.04.
- Macroprudential tools:
  - Log of sample median LTV ratio on bank’s housing loans: 0.40, 0.69.
  - Log of regulatory LTV ratio on banks’ large-sized housing loans: 1.03, 0.94.
  - Log of regulatory LTV ratio on banks’ small-sized housing loans: 0.55, 0.70.
  - Log of regulatory risk weights on the banks’ housing loans: -0.14, -0.14; Log of regulatory risk weights on the banks’ housing loans (t-1): -0.21, -0.21.
  - Log of median housing loan-to-income (LTI) ratio: 0.51, 0.81.
  - Log of housing EMI-to-income ratio: -0.40, -0.63.
  - Log of asset provisioning for banks’ standard housing assets: -0.20, -0.13.
- Magnitude examples and comparisons:
  - A 10-percentage point loosening of the median LTV ratio leads to an average 4-7 percent increase in nominal house prices via direct credit supply effects.
  - Duca et al. (2010) estimates (US): a 10-percentage point increase in LTV yields about an 8-11 percent increase in house prices.
  - Crowe et al. (2011) (cross-section of developed countries): a 10-percentage point increase in maximum LTV associated with a 13 percent increase in nominal house prices.
  - Unit tightening of risk weights on housing loans causes about 14-21 basis point decline in house prices; similar effect (13-20 basis points) from tightening asset provisioning on mortgages.
- Note: Coefficients in Table 1 represent averages from alternative specifications in Appendix Tables A1 and A2. All coefficients are significant at the 5% level.
- Robustness: panel fixed-effects specifications yield broadly similar conclusions.

### Real house price dynamics and macroprudential effectiveness (summary of FE and DPD results)
- Lagged real house prices: Log of real HPI (t-1) coefficients: 0.24, 0.36.
- Income and demographics:
  - Log of city real per capita income (t-1): 0.25, 0.25.
  - Log of the city population (t-1): 0.40, 0.12.
- Supply-side (real construction cost):
  - Log of real construction cost Index: 0.96 (Arellano–Bond reported).
  - Log of real construction cost Index (t-1): 0.29, 0.91.
- Rent/user cost: Log of real housing rentals: 0.33, 0.39.
- Substitution effect: Log of the ratio of real BSE Sensex to real HPI: -0.57, -0.55.
- Credit and rates:
  - Log of real per account nominal bank credit for residential housing: 0.15, 0.14; Log of real per account nominal bank credit for residential housing (t-1): 0.18 (reported).
  - Real weighted average interest rate on banks’ housing loans: -0.01, -0.02.
  - RBI’s real policy rate: -0.03, -0.02; RBI’s real policy rate (t-1): -0.02.
- Macroprudential tools (real effects):
  - Log of sample median LTV ratio on bank’s housing loans: 0.41, 0.44; (t-1): 0.42.
  - Regulatory LTV ratio on large-sized housing loans: 1.10, 1.14.
  - Regulatory LTV ratio on small-sized housing loans: 0.40, 0.59.
  - Regulatory risk weights on banks’ housing loans: -0.15, -0.14; (t-1): -0.15.
  - Log of median housing loan-to-income (LTI) ratio: 0.45; (t-1): 0.56.
  - Log of housing EMI-to-income ratio: -0.70.
  - Log of asset provisioning for banks’ standard housing assets: -0.16, -0.10.
- Interpretation:
  - Macroprudential measures (LTV, risk weights, asset provisioning) have a strong moderating impact on real house prices.
  - Differentiated LTV: change in LTV for small-sized loans has smaller impact (0.40-0.59) versus large-sized loans (1.10-1.14), consistent with larger speculative/investment motive in larger loans.
  - Higher loan leverage and lower affordability (EMI ratio) meaningfully affect real house prices.
- Note: Coefficients in Table 2 represent averages from alternative specifications in Appendix Tables A3 and A4. All coefficients are significant at the 5% level.

### Policy implications and key takeaways
- Macroprudential tools are effective countercyclical instruments for moderating house price cycles in India:
  - Tightening LTV ratios, raising regulatory risk weights, and increasing asset provisioning dampen nominal and real house price growth.
  - Differentiated application of LTV across loan sizes can target investment/speculative demand (larger effect on large-sized loans).
- Monetary policy transmission to housing and credit markets is significant but may be impeded by credit market imperfections (maturity mismatches, interest rate rigidities, competition from other saving instruments, and loan portfolio deterioration).
- Supply constraints matter: construction cost increases have large positive coefficients, indicating that small slowdowns in supply can lead to more than proportionate increases in nominal house prices.
- Financial stability consideration: growth in mortgage market and higher share of housing in household assets imply stronger housing-credit dynamics and potential systemic risk; active macroprudential management is warranted.
- Robustness: findings hold across dynamic panel and fixed-effects specifications; instruments and diagnostics support validity of estimated relationships.

*Source: Extracted from the specified chapter/section of the provided IMF PDF content.*

### Appendix Table A6, yield broadly similar results.

### wpiea2020291-print-pdf - Appendix Table A6, yield broadly similar results.

### V.3 Results from Housing Price Models based on the NHB’s Residex
- Dataset: National Housing Bank (NHB) Residex for 33 cities for the period 2010-2019.
- Purpose: Validate earlier conclusions on the role of macroprudential policies using a broader house price dataset.

### V.3.1 Nominal House Price Dynamics and the Role of Macroprudential Policies
- Estimation approach:
  - Dynamic panel data estimators using one-step and two-step estimators (Arellano–Bond and Blundell–Bond).
  - Sargan test statistics used to test overidentifying restrictions and validate instruments.
  - Alternative specifications and diagnostic tests in Appendix Tables A7 and A8; fixed-effect robustness in Appendix Table A11.
- Main qualitative findings:
  - Significant lag effect of nominal house prices (backward-looking expectations).
  - Demand-side variables with significant effects: per capita income, city population, bank credit, mortgage rates.
  - Relative increase in stock price ratio vis-à-vis house prices has a significant negative impact on housing prices (opportunity cost channel).
  - Contemporaneous and lagged negative effects of tightening policy rates and rising mortgage rates.
  - Construction cost (supply-side) has significant contemporaneous and lagged effects.
  - Macroprudential tools (LTV ratio, risk weights, asset provisioning) show significant negative responses on housing prices with strong lagged effects.
  - Differentiated LTV policy matters: changes in regulatory LTV for large-sized loans have much higher effects on housing prices than for small-sized loans.
  - Rising EMI (worsening affordability) strongly negatively affects house prices.
  - Rising loan leverage in the housing market pushes up housing prices.

- Table 3 — Summary coefficients (average of alternative specifications; all coefficients significant at the 5% level):
  - Arellano–Bond / Blundell–Bond
  - Log of nominal HPI(t-1): 0.29 / 0.42
  - Log of city real per capita income(t-1): 0.28 / 0.21
  - Log of city population (t-1): 0.29 / (blank)
  - Log of nominal construction cost Index: 0.59 / 0.67
  - Log of nominal construction cost Index (t-1): 0.49 / 0.59
  - Log of housing rentals: 0.22 / 0.13
  - Log of the ratio of BSE Sensex to nominal HPI: -0.59 / -0.41
  - Real weighted average interest rate on banks’ housing loans: -0.05 / -0.01
  - Real weighted average interest rate on banks’ housing loans (t-1): -0.02 / (blank)
  - RBI’s policy rate: -0.02 / -0.04
  - Log of sample median LTV ratio on bank’s housing loans: 0.40 / 0.66
  - Log of regulatory LTV ratio on banks’ large-sized housing loans: 0.59 / 0.60
  - Log of regulatory LTV ratio on banks’ small-sized housing loans: 0.37 / 0.20
  - Log of regulatory risk weights on the banks’ housing loans: -0.14 / -0.14
  - Log of regulatory risk weights on the banks’ housing loans (t-1): -0.15 / -0.11
  - Log of median housing loan-to-income (LTI) ratio: 0.57 / 0.58
  - Log of housing EMI-to-income ratio: (blank) / -0.75
  - Log of housing EMI-to-income ratio(t-1): (blank) / -0.31 / -0.25  (table shows two entries)
  - Log of asset provisioning for banks’ standard housing assets: -0.17 / -0.11

### V.3.2 Real House Price Dynamics and the Role of Macroprudential Policies
- Estimation approach:
  - Wider cross-section of NHB Residex; alternative specifications in Appendix Tables A9 to A10; robustness in Appendix Table A13 (fixed effects).
- Main qualitative findings:
  - Strong lagged effect of real house prices (backward-looking expectations).
  - Lagged per capita income and population are key demand-side drivers.
  - Bank credit has a significant contemporaneous effect on real house prices (smaller relative magnitude).
  - Strong contemporaneous negative impact of a relative asset price shock (rise in stock price relative to house price) on house prices, indicating a strong asset substitution effect.
  - Construction cost (supply shock) has a strong positive impact on housing prices; more pronounced where supply is sluggish.
  - Macroprudential tools: LTV ratio is most dominant; unit increase in maximum LTV for large-sized loans has larger effect than for small-sized loans.
  - Risk weights on housing loans have significant contemporaneous and lagged negative effects.
  - Asset provisioning negatively influences house prices.
  - Rising leverage (loan-to-income ratio) pushes up real house prices.
  - Affordability (EMI-to-income ratio): rising EMI dampens real house prices.
  - Real monetary policy shock has a positive but relatively small effect.

- Table 4 — Summary coefficients (average of alternative specifications; all coefficients significant at the 5% level):
  - Arellano–Bond / Blundell–Bond
  - Log of real house price(t-1): 0.31 / 0.42
  - Log of city real per capita income(t-1): 0.28 / 0.21
  - Log of the city population: 0.60 / (blank)
  - Log of real construction cost Index: 0.37 / 0.55
  - Log of real construction cost Index(t-1): 0.63 / 0.65
  - Log of real housing rentals: 0.25 / 0.15
  - Log of the ratio of real BSE Sensex to real house price: -0.51 / -0.51
  - Log of real per account nominal bank credit for residential housing: 0.12 / 0.12
  - Real weighted average interest rate on banks’ housing loans: -0.01 / -0.01
  - Real weighted average interest rate on banks’ housing loans(t-1): -0.01 / (blank)
  - RBI’s real policy rate: -0.02 / -0.01
  - RBI’s real policy rate(t-1): -0.02 / -0.02
  - Log of sample median LTV ratio on bank’s housing loans: 0.56 / 0.73
  - Log of regulatory LTV ratio on banks’ large-sized housing loans: 0.76 / 0.80
  - Log of regulatory LTV ratio on banks’ small-sized housing loans: 0.50 / 0.41
  - Log of regulatory risk weights on the banks’ housing loans: -0.11 / -0.16
  - Log of regulatory risk weights on the banks’ housing loans(t-1): -0.13 / -0.17
  - Log of median housing loan-to-income (LTI) ratio: 0.18 / 0.74
  - Log of housing EMI-to-income ratio: -0.40 / (blank)
  - Log of housing EMI-to-income ratio(t-1): (blank) / -0.23
  - Log of asset provisioning for banks’ standard housing assets: (blank) / -0.12
  - Log of asset provisioning for banks’ standard housing assets: -0.10 / (duplicate entry in table)

- Credit-channel evidence:
  - Real housing credit demand function estimates (Appendix Table A11) indicate:
    - Strong effect of both LTV ratios (large and small-size loans) on real housing credit demand.
    - Significant negative effect of increases in both risk weights and asset provisioning on real housing credit demand.

### VI Conclusions
- Objective: Examine effectiveness of macroprudential policies in influencing house price cycles in a developing country, controlling for fundamental factors.
- Robustness: Two novel city-level housing price databases broadly reinforce similar conclusions.
- Key macroeconomic determinants:
  - Per capita income significantly influences house prices.
  - Strong backward-looking house price expectations in both nominal and real estimators.
  - Demography (population) reinforces long-run drivers.
  - Opportunity cost: strong inverse effect of a positive shock to stock relative to house prices.
  - Mortgage credit is an important influence on housing prices.
  - Construction cost is a dominant supply-side factor due to sluggish supply response.
- Macroprudential policy findings:
  - LTV ratio is the most potent tool to respond to housing price fluctuations.
  - Change in maximum LTV ratio for large-sized loans has a much more dominant effect on house prices than changes in LTV for small-sized loans (large loans more investment-motivated; small loans more owner-occupier).
  - Strong asset substitution effect supports the role of speculative elements.
  - Increase in risk weights on housing assets causes significant downward pressure on nominal and real housing prices.
  - Standard asset provisioning on banks’ housing loans significantly affects house prices.
  - Regulatory measures have robust effects on credit demand, which in turn affect housing prices.
  - Build-up of credit leverage (loan-to-income ratio) substantially affects house prices by expanding access to borrowed funds.
  - Worsening affordability (higher loan EMI) has a significant negative effect on house prices.
- Policy implication:
  - As financial deepening and mortgage credit expansion occur in a developing economy, procyclicality in the housing market may increase, expanding the scope for macroprudential policies to contain systemic risks.
  - Evidence supports the efficacy of macroprudential policies in leaning against credit and housing boom-bust cycles.

*Source: wpiea2020291-print-pdf - Appendix Table A6, yield broadly similar results.*

### Annex A1. Sample of cities in house price database and their population

### Annex A1. Sample of cities in house price database and their population

### Sample coverage and HPI series
- Database includes cities with coverage by NHB House Price Index and/or RBI House Price Index as indicated.
- Population figures are reported in thousands (Population (in thousand)).

### Cities, HPI coverage, and population (exact entries)
- Ahmedabad — NHB: √ — RBI: √ — Population (in thousand): 7,681
- Bangalore — NHB: √ — RBI: √ — Population (in thousand): 11,440
- Bhopal — NHB: √ — RBI:  — Population (in thousand): 2,278
- Bhubaneshwar — NHB: √ — RBI:  — Population (in thousand): 1,100
- Chandigarh — NHB: √ — RBI:  — Population (in thousand): 1,110
- Chennai — NHB: √ — RBI: √ — Population (in thousand): 10,456
- Coimbatore — NHB: √ — RBI:  — Population (in thousand): 2,641
- Dehradun — NHB: √ — RBI:  — Population (in thousand): 871
- Delhi — NHB: √ — RBI: √ — Population (in thousand): 28,514
- Gandhinagar — NHB: √ — RBI:  — Population (in thousand): 356
- Guwahati — NHB: √ — RBI:  — Population (in thousand): 1,083
- Hyderabad — NHB: √ — RBI: √ — Population (in thousand): 9,482
- Indore — NHB: √ — RBI:  — Population (in thousand): 2,822
- Jaipur — NHB: √ — RBI: √ — Population (in thousand): 3,717
- Kanpur — NHB: √ — RBI: √ — Population (in thousand): 3,081
- Kochi — NHB: √ — RBI: √ — Population (in thousand): 2,858
- Kolkata — NHB: √ — RBI: √ — Population (in thousand): 14,681
- Lucknow — NHB: √ — RBI:  — Population (in thousand): 3,505
- Ludhiana — NHB: √ — RBI:  — Population (in thousand): 1,806
- Meerut — NHB: √ — RBI:  — Population (in thousand): 1,636
- Mumbai — NHB: √ — RBI: √ — Population (in thousand): 19,980
- Nagpur — NHB: √ — RBI:  — Population (in thousand): 2,808
- Nashik — NHB: √ — RBI:  — Population (in thousand): 1,952
- Patna — NHB: √ — RBI:  — Population (in thousand): 2,352
- Pune — NHB: √ — RBI:  — Population (in thousand): 6,276
- Raipur — NHB: √ — RBI:  — Population (in thousand): 1,521
- Rajkot — NHB: √ — RBI:  — Population (in thousand): 1,767
- Ranchi — NHB: √ — RBI:  — Population (in thousand): 1,370
- Surat — NHB: √ — RBI:  — Population (in thousand): 6,564
- Thiruvananthapuram — NHB: √ — RBI:  — Population (in thousand): 2,369
- Vadodara — NHB: √ — RBI:  — Population (in thousand): 2,110
- Vijayawada — NHB: √ — RBI:  — Population (in thousand): 1,911
- Visakhapatnam — NHB: √ — RBI:  — Population (in thousand): 2,076

*Source: RBI; NHB; UN Population Statistics database.*

### Appendix Table A6. Panel fixed-effect models for real house prices based on the RBI’s HPI

### Appendix Table A6. Panel fixed-effect models for real house prices based on the RBI’s HPI

### Model specifications
- Eleven panel fixed-effect model specifications [1] through [11].
- No. of groups: 10 for all models.
- T (time periods): 9, 11, 10, 10, 9, 11, 11, 11, 11, 11, 11 (for models [1]–[11], respectively).
- Figures in brackets are t-statistics or z-statistics. *, **, *** represent significance levels of 10%, 5% and 1%, respectively.
- Please see Annex A2 for the definition of the variables.

### Key coefficient estimates (coefficient followed by (t-statistic) as in table)
- constant:
  - [1] -0.77 (-0.80)
  - [2] -3.60 (-2.53)
  - [3] 5.53 (4.86)
  - [4] 2.79 (2.71)
  - [5] 2.72 (2.49)
  - [6] -8.62 (-4.58)
  - [7] -4.06 (-3.35)
  - [8] -2.68 (-1.23)
  - [9] 0.79 (0.66)
  - [10] 3.62 (4.01)
  - [11] 0.14 (5.23)

- lnddppc:
  - [1] 0.22** (2.42)
  - [2] 0.29*** (3.50)
  - [3] 0.14** (2.58)
  - [4] 0.19*** (3.07)
  - [5] 0.18*** (2.87)
  - [6] 0.58** (6.62)
  - [7] 0.58*** (6.62)
  - [8] 0.41*** (6.00)
  - [9] 0.39*** (6.00)
  - [10] 0.32*** (3.21)
  - [11] 0.23*** (5.72)

- lnpop_city:
  - [2] 0.43** (2.44)

- lnrconcoix:
  - [3] 0.46* (1.85)
  - [4] 0.46* (1.85)

- lnrbse_rhpi10:
  - [1] -0.77*** (-14.98)
  - [2] -0.75*** (-16.24)
  - [3] -0.91*** (-26.98)
  - [4] -0.90** (-13.23)
  - [5] -0.86*** (-20.38)
  - [6] -0.56*** (-6.22)
  - [7] -0.56*** (-6.22)
  - [8] -0.75*** (-10.86)
  - [9] -0.75*** (-10.86)
  - [10] -0.42*** (-5.65)
  - [11] -0.97*** (-40.43)

- lnrent_inx:
  - [1] 0.62*** (7.11)
  - [3] 0.26* (1.81)

- lnrcrdpa:
  - [3] 0.11** (2.19)

- rwarhg:
  - [2] -0.01** (-2.53)

- rrepo:
  - [1] -0.06*** -(6.33)
  - [2] -0.05*** (-12.29)
  - [4] -0.01** (-2.15)
  - [5] -0.01** (-2.15)
  - [6] -0.06*** (-8.68)
  - [7] -0.06*** (-8.68)
  - [9] -0.05*** (-14.37)

- lnltvm:
  - [1] 0.73*** (3.74)
  - [2] 0.89** (2.18)
  - [3] 0.26** (2.58)
  - [4] 0.86*** (3.16)
  - [5] 0.63*** (4.00)
  - [10] 0.34*** (3.01)

- lnltvl:
  - [4] 1.63* (-3.39)
  - [8] 1.25** (2.86)

- lnltvs:
  - [7] 0.58*** (-3.39)
  - [9] 0.44** (2.86)

- lnrwm:
  - [1] -0.20*** (-12.90)
  - [2] -0.24*** (-10.22)
  - [3] -0.22*** (-7.43)
  - [5] -0.19*** (-17.45)

- lnlti:
  - [4] 0.61*** (3.41)

- lnemiy:
  - [4] -0.68*** (-2.97)

- lnasstprov:
  - [4] -0.10*** (-9.43)

### Model fit and diagnostics
- R2 (within):
  - [1] 0.92
  - [2] 0.92
  - [3] 0.96
  - [4] 0.96
  - [5] 0.95
  - [6] 0.87
  - [7] 0.87
  - [8] 0.91
  - [9] 0.91
  - [10] 0.87
  - [11] 0.87

- F-stats. (figures in brackets are p-values):
  - [1] 232.14(0.0)
  - [2] 206.07(0.0)
  - [3] 299.30(0.0)
  - [4] 262.96(0.0)
  - [5] 254.32(0.0)
  - [6] 44.79(0.0)
  - [7] 44.79(0.0)
  - [8] 145.87(0.0)
  - [9] 145.87(0.0)
  - [10] 45.78(0.0)
  - [11] 65.39(0.0)

### Observations on substantive findings
- Real house prices (dependent variable) are:
  - Positively and significantly associated with lnddppc across all models: coefficients range from 0.14** to 0.89** with consistent statistical significance (see lnddppc rows).
  - Negatively and strongly associated with lnrbse_rhpi10 across all models: coefficients range from -0.42*** to -0.97*** with large magnitude t-statistics.
  - Positively associated with loan-to-value measures (lnltvm, lnltvl, lnltvs) where included, often statistically significant.
  - Negatively associated with measures of risk or housing market weakness (lnrwm) where included, with coefficients between -0.19*** and -0.24***.
  - Monetary policy rate (rrepo) generally shows a negative and statistically significant relationship where included.
  - Credit-related variables (lnrcrdpa) show positive and sometimes significant associations in specifications where included.

*Appendix Table A6. Panel fixed-effect models for real house prices based on the RBI’s HPI (source PDF content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020291-print-pdf.pdf_