## 1. House Prices Growth in Malaysia, 2001–2012 ## Source details **Canonical URL:** [1. House Prices Growth in Malaysia, 2001–2012](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2014/_cr1499.pdf) ## Other formats - [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2014/_cr1499.pdf.md) - [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2014/_cr1499.pdf.json) --- ### Introduction and scope - Focus: price developments in the Malaysian housing market and its sustainability using quarterly house price index data beginning 2000Q1 (annual growth rates from 2001Q1). - Methodology: two-step econometric procedure constructing baseline “predictors” from a sample ending 2010Q2 and producing out-of-sample predictions starting 2010Q3. - Data limitation: limited timely micro-data (households’ income and assets); results are indicative and meant as warnings and guidance for further study. ### Key empirical facts on house prices and households - From 2001 through 2009 house price growth averaged 3 percent (annual). - Since 2010 the growth rate increased, reaching 12 percent at the beginning of 2011. - Average inflation (end of period) changed from 2.7 percent for 2007–2009 to 2.6 percent for 2010–2011. - Regional variation in house price growth (beginning 2012): 18 percent in Kuala Lumpur to -1.1 percent in Perlis. - Household debt-to-GDP ratio reached 74 percent at the end of 2011 (historical peak). - Total household financial and liquid assets are 2.3 times and 1.5 times household debt, respectively. - Visual and formal analysis indicate a structural break in house price dynamics beginning around 2010Q3; the second half of 2010 marks a clear change. ### Risks to financial stability - Sharp rise in house prices combined with relatively high household indebtedness could be a source of financial stability risk. - Mechanism: a sharp fall in prices would reduce collateral values, impair borrower repayment capacity in a recessionary environment, and could affect financial institutions’ balance sheets. - Aggregate household financial buffers could mitigate risks, but distributional micro-level variation may be important and is not captured by aggregate data. ### Policy measures enacted (timeline and specifics) - November 2010: BNM introduced maximum 70 percent LTV ratio on the third and subsequent home mortgages. - January 2011: banks’ capital risk weights raised to 100 percent for mortgages with LTVs exceeding 90 percent. - December 2011: maximum LTV ratio of 60 percent set for residential property loans taken by non-individual borrowers. - January 2010: Real Property Gains Tax (RPGT) set at 5 percent after exemption during 2007–2009. - 2011 revision: RPGT reintroduced as tiered structure ranging between 5 and 10 percent for disposal within five years of purchase. - 2013 budget (announced September 2012): RPGT revised to 10–15 percent for disposal within five years. - January 2010: minimum unit price for house purchases by foreigners raised from RM 250,000 to RM 500,000. ### Predictor of House Prices Growth — model construction and results - Baseline sample period: 2001Q2–2010Q2 (Included observations: 37 after adjustments). - Dependent variable: annual percentage growth rate of the house price index. - Regressors: CPIRP (CPI component: rents, power and utilities), RGDP (Real GDP), SENT (consumer sentiment index), BLR (base lending rate). - Data transformations: regressions on percentage annual changes for dependent and regressors (BLR entered as annual difference); appropriate lag structure employed. Model estimation results (Least Squares, sample 2001Q2–2010Q2): - Constant: -1.20 (Std. Error 1.00; t-Stat -1.20; Prob 0.24) - RGDP(-1): 0.58 (Std. Error 0.09; t-Stat 6.71; Prob 0.00) - CPIRP(-2): 1.30 (Std. Error 0.54; t-Stat 1.43; Prob 0.02) - BLR: -1.79 (Std. Error 0.52; t-Stat -3.41; Prob 0.00) - SENT: 0.05 (Std. Error 0.01; t-Stat 6.35; Prob 0.00) Model diagnostics: - R-squared: 0.71 - Adjusted R-squared: 0.68 - S.E. of regression: 0.92 - AIC-corrected: 4.73 - Schwarz criterion: 3.02 - Durbin-Watson stat.: 1.96 - Engel-Granger test for cointegration: strongly rejects the null hypothesis that house prices growth and the predictor are not cointegrated (p-values below 0.0 percent). Key findings from the predictor analysis: - Predictor explains 71 percent of in-sample variation in house prices growth. - All regressors’ coefficients have the expected signs. - Starting with the second half of 2010, actual house price growth permanently exceeds the predictor (out-of-sample). - Re-estimating the regression for the entire sample: R-squared drops to 38 percent; introducing lagged house price growth raises R-squared to 81 percent and the lagged term is significant at the 1 percent level. - For pre-2010Q2 sample, lagged house price growth is highly insignificant when added to baseline regression—indicating a change in dynamics after 2010Q2. - Post-2010Q2, current house price growth is statistically well explained by past price growth or expected future growth rather than by historical predictor variables. Robustness and alternatives: - Adding population trend (POP) and more lags increased in-sample fit (R-squared 0.81; Adjusted R-squared 0.76) but raised overfitting concerns and worse AIC-corrected performance. - Substituting average lending rate on new loans for BLR and adult population growth for POP left results substantially unchanged but reduced explanatory power. - Replacing CPIRP with construction-materials prices underperformed relative to CPIRP. - Wage growth data unavailable before 2006 and could not substitute for RGDP. ### Exclusion of housing-supply regressors and endogeneity concerns - Supply of housing excluded due to "strong endogeneity bias". - Suggested methods to address endogeneity in future work: use vector auto regression (VAR) or simultaneous equation techniques. - Alternative supply measures tested and not significant: growth rate (current and lagged) in housing units; ratio between population and number of housing units. - Population treatment: quarterly population from spline interpolation of annual data; filtered with HP to capture long-term component; annual percentage change in trend population (POP) obtained. - Population trend coefficient: positive and significant but explains a small portion of variance. - Alternative model includes contemporaneous RGDP and CIR with positive and significant coefficients. ### Residential lending: objective, model, and results - Objective: construct predictor for residential loans up to Q2 2010 and compare predicted path with actual loans from Q3 2010 onward. - Data: BNM monthly series on approved loans by purpose; residential loans available as annual growth rates since January 2007. - Baseline regression sample: January 2007 to June 2010; out-of-sample: July 2010 to May 2012. - Regressors: RES(-1) (lagged residential loans), IP (Industrial Production Index), WCL (outstanding approved working-capital loans), OPR (overnight policy rate; annual difference used). - Transformations: regressions on percentage annual changes for dependent and regressors 1–3; OPR as annual difference. Predictor estimation results and diagnostics: - In-sample fit: predictor explains 74 percent of in-sample variation; Adjusted R-squared: 0.72. - Engel-Granger cointegration test: strongly rejects null (p-value below 0.0 percent). - Durbin-Watson statistic: 1.88. Reported coefficients: - Constant: Coeff. 0.08; Std. Error 0.05; t-Stat 1.73; Prob 0.09 - RES(-1): Coeff. 0.40; Std. Error 0.12; t-Stat 3.22; Prob 0.00 - IP: Coeff. 1.50; Std. Error 0.51; t-Stat 2.95; Prob 0.00 - WCL: Coeff. 0.26; Std. Error 0.10; t-Stat 2.69; Prob 0.01 - OPR(-1): Coeff. -19.60; Std. Error 4.87; t-Stat -4.02; Prob 0.00 Additional model statistics: - R-squared 0.74 - Adjusted R-squared 0.72 - S.E. of regression 0.17 - AIC-corrected 1.18 - Schwarz criterion -0.32 - Durbin-Watson stat 1.88 Out-of-sample behavior and robustness: - Growth rate of residential loans remained moderate after first half of 2010; hovered around 10 percent since second half of 2010. - No sign of a break-down in relation with estimated predictor. - Engel-Granger test extended to May 2012: strongly rejects null of no cointegration (p-value below 0.0 percent). - Chow test over entire sample: does not reject null hypothesis of no structural break in June 2010. - Alternative specification adding lagged WCL and "other loans" raises R-squared to 83 percent (Adjusted R-squared 80 percent) but small sample raises overfitting concerns; baseline preferred by corrected Akaike Information Criterion. - Out-of-sample behavior largely unaffected by model choice. Interpretation: - Residential loans evolution is econometrically stable, contrasting with house-price behavior which shows a structural shift since 2010. - No evidence that recent higher house-price growth is due to changes in credit policies specific to the mortgage market (a mortgage-specific easing would have produced a structural break with residential loans growth permanently above the predictor path starting Q3 2010). - Macro-prudential policies appear to have changed lending composition: growth rate of individuals with multiple housing accounts slowed to 2.3 percent in June 2012, from 14.3 percent in December 2010. - No evident sign that macro-prudential policies have led to a slowdown in house-price growth to date. ### Conclusions, policy implications, and data needs - Stylized facts: - High house-price growth since 2010; since Q3 2011, annual growth rate at least 10 percent. - These rates are historical highs for Malaysia, though below some neighboring countries' recent rates. - Decoupling: - Since second half of 2010, house-price growth has decoupled from its structural relation with the historical predictor; decoupling remains unexplained. - Policy implications: - Structural break in 2010 suggests factors beyond historical predictors contributed to sharp rise; policy attention should consider mechanisms that amplify price momentum (e.g., expectations) and distributional vulnerabilities. - Macroprudential measures already adopted target LTVs and capital treatment; their effectiveness should be monitored against evolving price dynamics and lending behavior. - No evidence that mortgage-lending practices or macro-prudential measures explain the high house-price growth observed so far. - Recommended further research and data collection: - Collect timely micro-level data on household income, assets, and distributional characteristics to assess resilience and downside risks. - Collect granular micro-data including house ownership by income group, collateral values, and refinancing facilities taken by house buyers (especially for "equity withdrawal" purposes). - Investigate potential "missing variables": changes in tax treatment of real property gains, demand from foreigners (including Malaysians residing abroad), endogenous supply changes, regional dynamics, and microeconomic distributional shifts. *Prepared by Roberto Piazza (IMF) in the context of the 2013 Malaysia FSAP.* *Source: _cr1499 - 15. Of a somewhat different nature is the decision to exclude the supply of housing* ### 1. House Prices Growth in Malaysia, 2001–2012 ......................................................................3 ### 1. House Prices Growth in Malaysia, 2001–2012 ### Introduction and scope - Analysis focuses on price developments in the Malaysian housing market and its sustainability, using quarterly house price index data beginning 2000Q1 (annual growth rates from 2001Q1). - Methodology: two-step econometric procedure constructing baseline “predictors” from a sample ending 2010Q2 and producing out-of-sample predictions starting 2010Q3. - Cautionary note: limited timely micro-data (households’ income and assets) constrains the analysis; results are indicative and should be used as warnings and guidance for further study. ### Key empirical facts on house prices and households - From 2001 through 2009 house price growth averaged 3 percent (annual). - Since 2010 the growth rate increased, reaching 12 percent at the beginning of 2011. - Average inflation (end of period) changed from 2.7 percent for 2007–2009 to 2.6 percent for 2010–2011. - Regional variation in house price growth (beginning 2012): 18 percent in Kuala Lumpur to -1.1 percent in Perlis. - Household debt-to-GDP ratio reached 74 percent at the end of 2011 (historical peak). - Total household financial and liquid assets are 2.3 times and 1.5 times household debt, respectively. - Visual and formal analysis indicate a structural break in house price dynamics beginning around 2010Q3; the second half of 2010 marks a clear change. ### Risks to financial stability - Concern: sharp rise in house prices combined with relatively high household indebtedness could be a source of financial stability risk. - Mechanism described: a sharp fall in prices would reduce collateral values, impair borrower repayment capacity in a recessionary environment, and could affect financial institutions’ balance sheets. - Caveat: aggregate household financial buffers could mitigate risks, but distributional micro-level variation may be important and is not captured by aggregate data. ### Policy measures enacted (timeline and specifics) - November 2010: BNM introduced maximum 70 percent LTV ratio on the third and subsequent home mortgages. - January 2011: banks’ capital risk weights raised to 100 percent for mortgages with LTVs exceeding 90 percent. - December 2011: maximum LTV ratio of 60 percent set for residential property loans taken by non-individual borrowers. - January 2010: Real Property Gains Tax (RPGT) set at 5 percent after exemption during 2007–2009. - 2011 revision: RPGT reintroduced as tiered structure ranging between 5 and 10 percent for disposal within five years of purchase. - 2013 budget (announced September 2012): RPGT revised to 10–15 percent for disposal within five years. - January 2010: minimum unit price for house purchases by foreigners raised from RM 250,000 to RM 500,000. ### Conclusions from overview - Despite policy measures, house price growth remains at historical highs. - Statistical evidence points to a structural break in 2010 between actual house price growth and the predictor that historically tracked price dynamics. - No structural break detected in the relation between residential loans growth and its corresponding predictor estimated from historical data. --- ### Predictor of House Prices Growth — model construction and results - Sample period used to estimate the baseline predictor: 2001Q2–2010Q2 (Included observations: 37 after adjustments). - Dependent variable: annual percentage growth rate of the house price index. - Regressors chosen to capture dividends from ownership, construction costs, income and expectations, and borrowing costs: - CPIRP (CPI component: rents, power and utilities) — source: Bank Negara Malaysia. - RGDP (Real GDP) — source: Bank Negara Malaysia. - SENT (consumer sentiment index) — source: Bank Negara Malaysia. - BLR (base lending rate) — source: Bank Negara Malaysia. - Data transformations: - Regressions performed on percentage annual changes for dependent variable and regressors (BLR entered as annual difference). - Appropriate lag structure employed (lags indicated in reported coefficients). Model estimation results (Least Squares, sample 2001Q2–2010Q2): - Constant: -1.20 (Std. Error 1.00; t-Stat -1.20; Prob 0.24) - RGDP(-1): 0.58 (Std. Error 0.09; t-Stat 6.71; Prob 0.00) - CPIRP(-2): 1.30 (Std. Error 0.54; t-Stat 1.43; Prob 0.02) - BLR: -1.79 (Std. Error 0.52; t-Stat -3.41; Prob 0.00) - SENT: 0.05 (Std. Error 0.01; t-Stat 6.35; Prob 0.00) Model diagnostics: - R-squared: 0.71 - Adjusted R-squared: 0.68 - S.E. of regression: 0.92 - AIC-corrected: 4.73 - Schwarz criterion: 3.02 - Durbin-Watson stat.: 1.96 - Engel-Granger test for cointegration: strongly rejects the null hypothesis that house prices growth and the predictor are not cointegrated (p-values below 0.0 percent). Key findings from the predictor analysis: - The predictor provides a good in-sample fit, explaining 71 percent of total in-sample variation in house prices growth. - All regressors’ coefficients have the expected signs. - Starting with the second half of 2010, actual house price growth permanently exceeds the predictor (out-of-sample). - If the regression is re-estimated for the entire sample, R-squared drops to 38 percent; introducing lagged house price growth raises R-squared to 81 percent and the lagged term is significant at the 1 percent level. - For the pre-2010Q2 sample, lagged house price growth is highly insignificant when added to the baseline regression, indicating a change in dynamics after 2010Q2. - Post-2010Q2, current house price growth is statistically well explained by past price growth or expected future growth rather than by the historical predictor variables. Alternative specifications and robustness checks: - Expanding regressors to include population trend (POP) and more complete lag structures increased in-sample fit (R-squared 0.81; Adjusted R-squared 0.76) but raised overfitting concerns and had worse AIC-corrected performance. - Out-of-sample behavior of predictors from alternative and baseline models is almost indistinguishable. - Substituting average lending rate on new loans for BLR and adult population growth for POP left results substantially unchanged but reduced explanatory power. - Adding price of steel and other construction materials in place of CPIRP underperformed relative to CPIRP. - Wage growth could not be used as substitute for RGDP because wage data are available only since 2006. --- ### Policy implications and areas for further study - The structural break in 2010 suggests that factors beyond historically reliable predictors contributed to the sharp rise in house price growth; policy attention should consider mechanisms that amplify price momentum (e.g., expectations) and distributional vulnerabilities. - Macroprudential measures already adopted target LTVs and capital treatment; their effectiveness should be monitored against evolving price dynamics and lending behavior. - Further data collection is recommended, particularly timely micro-level data on household income, assets, and distributional characteristics, to assess household resilience and downside risks more accurately. - Continued monitoring of the relation between residential loan growth and credit supply is warranted given the absence of a corresponding structural break in residential loans dynamics. *Prepared by Roberto Piazza (IMF) in the context of the 2013 Malaysia FSAP.* ### 15. Of a somewhat different nature is the decision to exclude the supply of housing ### _cr1499 - 15. Of a somewhat different nature is the decision to exclude the supply of housing ### Exclusion of housing-supply regressors and endogeneity concerns - Decision: the supply of housing was excluded from the list of regressors due to "strong endogeneity bias" that the variable could introduce. - Suggested methods to address endogeneity in future work: - Use vector auto regression (VAR). - Use simultaneous equation techniques. - Tested alternative measures of house supply (none significant): - Growth rate, current and lagged, in housing units. - Ratio between the population and the number of housing units. - Population handling (notes on alternative model): - Official publications report overall population only at an annual frequency. - Quarterly population levels obtained using spline interpolation of annual data. - Quarterly population series filtered using an HP procedure to capture long-term component and avoid artificial fluctuations from interpolation. - Annual percentage change in the trend population (POP) obtained. - Regression coefficient for population trend growth: positive and significant, but explains a small portion of the independent variable’s total variance. - Alternative model also includes contemporaneous values of RGDP and of CIR, both with positive and significant regression coefficients. ### Break-down in structural relation between house prices and macroeconomic variables - Observation: structural relation suggests an important shift in house-price dynamics since 2010; further investigation recommended. - Identification of "missing variables" is important to explain rapid house-price growth since 2010. - Examples and implications: - Sudden/unexplained increases in house prices are not necessarily harmful (e.g., improved liquidity). - If driven by heightened moral hazard in lending practices, systemic risk consequences would be substantial. - Suggested additional explanatory variables for alternative econometric models: - Changes in the tax treatment of real property gains. - Demand for domestic real-estate due to foreigners including Malaysians residing abroad. - The role of (endogenous) changes in the supply of houses. - Regional dynamics: substantial variation in house-price growth (as seen in Figure 1) may merit investigation. - Microeconomic phenomena not evident in aggregate variables: changes in distribution of income growth, of debt and of liabilities across population groups; requires more data collection. ### Residential lending: objective and approach - Objective: construct a predictor that explains the pattern of residential loans up to Q2 2010 and contrast its predicted path with actual residential loans from Q3 2010 onward to explore links with house-price dynamics. - Rationale: a break-down in the relation between loans growth and the predictor since Q3 2010 could indicate credit-policy changes specific to mortgages influencing house prices. - Data and sample period: - BNM monthly time series on approved loans by purpose; residential loans series available as annual growth rates since January 2007. - Baseline regression sample: January 2007 to June 2010. - Out-of-sample period: July 2010 to May 2012 (latest date with values for all regressors). - Regressors chosen: - RES(-1): lagged value of volume of loans approved for acquisition of residential property. [Source: BNM] - IP: Industrial Production Index (monthly proxy for aggregate income). [Source: BNM] - WCL: outstanding volumes of approved working-capital loans (about 10 percent of WCL to construction sector). [Source: BNM] - OPR: overnight policy rate; for OPR the annual difference is used. - Data transformation: - Regression performed on percentage annual changes for dependent variable, current and lagged, and for regressors 1–3. - For OPR, the annual difference is used. - Purpose of regressors: capture current and expected borrower income evolution and general credit-market conditions (volume and price of loans); include lagged dependent variable to improve fit. ### Predictor estimation results and diagnostics - In-sample fit: - Predictor explains 74 percent of total in-sample variation in growth rate of residential lending. - Adjusted R-squared: 0.72. - Engel-Granger cointegration test: strongly rejects null hypothesis (p-value below 0.0 percent) that growth rates in residential loans and predictor are not cointegrated. - Durbin-Watson statistic: 1.88 (close to 2), autocorrelation in residuals ruled out. - Reported estimated coefficients (as reported in Figure 3): - Constant: Coeff. 0.08; Std. Error 0.05; t-Stat 1.73; Prob 0.09 - RES(-1): Coeff. 0.40; Std. Error 0.12; t-Stat 3.22; Prob 0.00 - IP: Coeff. 1.50; Std. Error 0.51; t-Stat 2.95; Prob 0.00 - WCL: Coeff. 0.26; Std. Error 0.10; t-Stat 2.69; Prob 0.01 - OPR(-1): Coeff. -19.60; Std. Error 4.87; t-Stat -4.02; Prob 0.00 - Additional model statistics: - R-squared 0.74 - Adjusted R-squared 0.72 - S.E. of regression 0.17 - AIC-corrected 1.18 - Schwarz criterion -0.32 - Durbin-Watson stat 1.88 ### Out-of-sample behavior, robustness checks, and alternative specification - Post-2010 behavior: - Growth rate of residential loans remained moderate after first half of 2010; hovered around 10 percent since second half of 2010. - No sign of a break-down in relation with estimated predictor. - Engel-Granger test (when sample extended to May 2012): strongly rejects null of no cointegration (p-value below 0.0 percent). - Chow test over entire sample: does not reject null hypothesis of no structural break in June 2010. - Alternative model specification: - Added regressors: lagged WCL and "other loans" (residual credit variable excluding WCL, residential property, and construction financing). - Both additions highly significant with expected positive sign. - Fit improves: R-squared increases to 83 percent; Adjusted R-squared to 80 percent. - Concern: small sample size raises overfitting concerns. - Model selection: baseline model preferred according to corrected Akaike Information Criterion. - Out-of-sample behavior: largely unaffected regardless of model used. ### Interpretation and policy-relevant findings on residential lending and macro-prudential measures - Key contrasts and inferences: - Residential loans evolution is econometrically stable, contrasting with house-price behavior which shows a structural shift since 2010. - No evidence that recent higher house-price growth is due to changes in credit policies specific to the mortgage market: a mortgage-specific easing would have produced a structural break with residential loans growth permanently above the predictor path starting Q3 2010. - Macro-prudential policies: - No direct evidence that macro-prudential policies changed dynamics of house prices or aggregate residential loans. - Authorities’ objective: change composition of lending (reduce "speculative" lending, favor "non-speculative" investors), not necessarily reduce lending volume. - Evidence of policy effect on composition: growth rate of individuals with multiple housing accounts slowed to 2.3 percent in June 2012, from 14.3 percent in December 2010. - No evident sign that macro-prudential policies have led to a slowdown in house-price growth to date. - Interpretation: measures may have encouraged a more cautious approach to lending, "shutting the stable door before the horse bolted." ### Conclusions and data needs - Stylized facts: - High house-price growth recorded since 2010; since Q3 2011, annual growth rate at least 10 percent. - These rates are historical highs for Malaysia, though below some neighboring countries' recent rates. - Decoupling: - Since second half of 2010, house-price growth has decoupled from its structural relation with the predictor constructed from historical data; this decoupling remains unexplained and requires further investigation. - Implications: - Identifying the source(s) of the shift in house-price dynamics is crucial to determine whether growth is driven by harmful market distortions and whether targeted corrective actions are needed. - No evidence that mortgage-lending practices or macro-prudential measures explain the high house-price growth observed so far. - Recommended enhancements in data collection: - Collect granular micro-data including: - House ownership by income group. - Collateral values. - Refinancing facilities taken by house buyers, especially for "equity withdrawal" purposes. - Such data would aid analysis of systemic risk and design/implementation of macro-prudential policies. *Source: _cr1499 - 15. Of a somewhat different nature is the decision to exclude the supply of housing* --- _Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2014/_cr1499.pdf_