## Section I. Detailed Technical Assessment and Recommendations ## Source details **Canonical URL:** [Section I. Detailed Technical Assessment and Recommendations](https://www.imf.org/-/media/files/publications/tar/2024/english/tarea2024037.pdf) ## Other formats - [Markdown version](/-/media/files/publications/tar/2024/english/tarea2024037.pdf.md) - [Structured JSON version](/-/media/files/publications/tar/2024/english/tarea2024037.pdf.json) --- ### Summary of Mission Outcomes and Priority Recommendations - A technical assistance (TA) mission was conducted from January 15–19, 2024, to assist the National Bank of Cambodia (NBC) with further improvement of their Residential Property Price Index (RPPI). The mission was conducted under the Data for Decisions (D4D) Trust Fund. - Tasks completed: - review of methods used for the RPPI; - identification of areas for improvement consistent with international best practice; - practical training to NBC staff; - assessment of the suitability of the template used for data collection from commercial banks; - analysis of potential for accessing and using more comprehensive data sources. - Authorities are strongly committed to improving methodology and associated R scripts for the RPPI. - NBC published a new monthly RPPI data series in June 2022 using residential property loan data from Cambodian mortgage lending institutions. NBC receive the data 10 days after the reference month and results are disseminated at the start of the following month. - Key areas for improvement identified: - Filtering and cleaning of the loan data. - Stratification of the sample. - Outlier detection method. - Model specification for hedonic regressions. - Inclusion of additional loan data from micro-finance institutions. - Recommended methodological changes: - Continue using time dummy hedonic method with an 18-month window length. - Amend filters (e.g., lower minimum threshold for land area from 40 to 20 square meters). - Merge certain property characteristics into homogeneous groups (location categories within and outside Phnom Penh using provinces, districts/khans, communes/sangkats). - Apply more detailed stratification: four strata — (i) Central & East Phnom Penh, (ii) Inner Suburbs Phnom Penh, (iii) Outer Suburbs Phnom Penh, (iv) Cambodia excluding Phnom Penh. - Replace interquartile-range outlier detection (which excluded up to 18.3 percent in some months) with Cook’s Distance. - Reduce overfitting of hedonic model by updating model specification and applying robustness checks. - Impact of updated methods (houses): - Published national index: inflation of 16.0 percent between 2020M1 and 2021M6. - Updated index: inflation of 11.0 percent between 2020M1 and 2021M6. - NBC should implement updated methods into regular compilation and analyze results before deciding on retrospective revision. - Priority recommendations (assigned to NBC) and timing: - Jun-2024: Complete the calculation of the updated indices for the most recent period and compare with the published RPPI. - Dec-2024: Implement the improved methods into their regular compilation of the RPPI. - Dec-2024: Decide whether to revise the retrospective data. ### RPPI Workplan for Cambodia, 2024 (Selected actions and targets) - Data sources: - Continue to use the loan data to compile the RPPI in the medium term. Target Completion Date: Feb-2024. - Include additional loan data from micro-finance institutions in the sample as part of the overall update of the methods. Target Completion Date: Dec-2024. - Estimate existing coverage of the RPPI using loan data and administrative data sources. Target Completion Date: Dec-2024. - Devise a longer-term strategy to close coverage gap by researching potential new data sources (e.g., website listings). Target Completion Date: Dec-2024. - Compilation methods (status and targets): - Exclude transactions outside scope (e.g., loans issued for building a property). Status: Largely complete. - Apply more detailed stratification of the loan data. Status: Completed. - Use Cook’s Distance for outlier detection. Status: Completed. - Update hedonic model specification and apply robustness checks. Status: Largely complete. - Aggregate sub-indices using weights to higher-level indices including the national RPPI. Status: Completed. - Update R-code and organize external review of R scripts. Target Completion Date: Feb-2024. Priority: H. - Complete calculation of updated indices for recent period and compare with published RPPI. Target Completion Date: Jun-2024. Priority: H. - Monitor differences between updated and published indices. Target Completion Date: Aug-2024. - Re-run compilation with micro-finance institutions included and calculate updated RPPIs. Target Completion Date: Dec-2024. - Dissemination: - Implement improved methods into regular compilation of the RPPI. Target Completion Date: Dec-2024. Priority: H. - Decide whether to revise retrospective data. Target Completion Date: Dec-2024. Priority: H. - Disseminate more detailed RPPIs. Target Completion Date: Feb-2025. Priority: H. - Draft and publish a detailed methodological document on the authorities’ website. Target Completion Date: Feb-2025. Priority: H. ### Data Sources — Key findings and recommendations - Data source: monthly survey of lending institutions. Data available from 2019; RPPI currently compiled with data from January 2020 (2019 data judged lower quality). - Average monthly observations: - 2020: 830 - 2021: 1,030 - 2022: 1,260 - 2023: 880 - Loan reporting template sections: (i) administrative information, (ii) borrower information, (iii) loan information, (iv) property information. - Property information fields relevant to RPPI: total price of purchased property (final market price), price of purchased land, appraisal price, location variables (province/city, district/khan, commune/sangkat, village, street number), size variables (floor area, plot size, number of bedrooms, width), and other variables (property type, title of property, floor number/number of floors, date of construction). - Date used to assign transaction to reference month: drawdown date for the loan. - Coverage limitations: - sample limited to market financed by loans from commercial banks; - segments not covered include (i) cash purchases, (ii) transactions financed by developer loans, (iii) transactions financed by foreign banks. - Recommended actions: - Continue using loan data in the medium term. - Include micro-finance institution loan data (predominantly covers areas outside Phnom Penh) as part of the overall methods update. - Estimate coverage by comparing loan data with administrative sources (e.g., property tax data at General Department of Taxation, MEF). - Research new data sources (e.g., listings from real estate websites) to close coverage gaps. - Condominiums and apartments: - few observations in loan data; published RPPI covers houses only. - Legal context: foreigners cannot own land or houses built on the ground; foreigners can own apartments/condominiums above ground floor as co-owned properties provided foreign ownership does not exceed 70 percent of total floor area. ### Compilation Methods — Technical adjustments and outcomes - Current approach: time dummy hedonic method with an 18-month window length; recommended to continue but with improvements. - Data inspection: visualizations and descriptive statistics (histograms, bar charts, scatter plots, mean/median price per district) used to assess data quality and relationships. - Transactions to exclude: loans for building a property (funds used to purchase materials/pay builders) should be excluded as no residential property transaction occurs. - Revised filters: - Minimum land area threshold lowered from 40 to 20 square meters. - Exclude properties with number of bedrooms greater than 9. - Exclude properties with number of floors greater than 4. - Property characteristic grouping: - Create location categories within Phnom Penh using khan and sangkat; outside Phnom Penh, merge provinces. About 6–10 categories per strata recommended depending on observations. - Create four categories for number of floors and seven categories for number of bedrooms. - Stratification detail and sample counts (before outlier detection) — Number of observations per Year and Strata: - 2020 — Central & East Phnom Penh 2,495; Inner Suburb Phnom Penh 3,669; Outer Suburb Phnom Penh 2,313; Cambodia excluding Phnom Penh 1,483; Total 9,960 - 2021 — Central & East Phnom Penh 2,304; Inner Suburb Phnom Penh 3,837; Outer Suburb Phnom Penh 3,519; Cambodia excluding Phnom Penh 2,671; Total 12,331 - 2022 — Central & East Phnom Penh 2,362; Inner Suburb Phnom Penh 3,748; Outer Suburb Phnom Penh 5,525; Cambodia excluding Phnom Penh 3,492; Total 15,127 - 2023 — Central & East Phnom Penh 1,394; Inner Suburb Phnom Penh 2,635; Outer Suburb Phnom Penh 3,815; Cambodia excluding Phnom Penh 2,700; Total 10,544 - All years — Central & East Phnom Penh 8,555; Inner Suburb Phnom Penh 13,889; Outer Suburb Phnom Penh 15,172; Cambodia excluding Phnom Penh 10,346; Total 47,962 - Outlier detection: - Existing method: interquartile range by month and strata; excluded up to 18.3 percent in some months. - Recommendation: replace with Cook’s Distance (regression-based) to identify outliers with greater precision and preserve sample size. - Hedonic model specification: - Existing model is overfitted with too many independent variables; can lead to misleading R-squared, coefficients, and p-values. - Recommendation: update specification, reduce overfitting, and apply robustness checks. ### Outlier detection: issue, mission results, and recommendation - Existing method: interquartile range by month and strata to identify outliers. - Problem: method excludes a very high proportion of the sample. - Within Phnom Penh, excludes between 6.9 percent and 15.9 percent of the data. - Outside Phnom Penh, excludes between 10.3 percent and 18.3 percent of the data. - Recommendation: replace the interquartile range method with Cook’s Distance (a regression-based method). - Rationale: identifies outliers with greater precision and preserves the sample to the extent possible. - Mission application results (proportion of identified outliers across strata and years): ranges from 4.1 percent to 7.0 percent across all strata and years. - Identified Outliers Per Year and Strata, Percent of Observations: - 2020: Central & East Phnom Penh 6.6; Inner Suburbs Phnom Penh 5.8; Outer Suburbs Phnom Penh 7.0; Cambodia excluding Phnom Penh 5.9; Total 6.3 - 2021: Central & East Phnom Penh 5.9; Inner Suburbs Phnom Penh 5.2; Outer Suburbs Phnom Penh 4.1; Cambodia excluding Phnom Penh 6.6; Total 5.3 - 2022: Central & East Phnom Penh 8.8; Inner Suburbs Phnom Penh 7.2; Outer Suburbs Phnom Penh 5.0; Cambodia excluding Phnom Penh 7.0; Total 6.6 - 2023: Central & East Phnom Penh 6.6; Inner Suburbs Phnom Penh 5.8; Outer Suburbs Phnom Penh 6.0; Cambodia excluding Phnom Penh 6.0; Total 6.0 - All years: Central & East Phnom Penh 7.0; Inner Suburbs Phnom Penh 6.0; Outer Suburbs Phnom Penh 5.4; Cambodia excluding Phnom Penh 6.5; Total 6.1 ### Model specification: overfitting concerns, preferred variables, diagnostics - Problem: existing RPPI hedonic model is overfitted with too many independent variables (e.g., sangkat dummies for 105 sangkats), producing misleading R-squared values, coefficients, and p-values. - Location variable recommendations: - Remove sangkat variable. - Replace with aggregated location variable such as khan or categories created using a combination of khan and sangkat. - Outside Phnom Penh, create location categories using a combination of province and district. - Variable-level finding: - Coefficient for width of the property was not statistically significant and could be removed. - Preferred model includes the following property characteristics: - (i) location e.g., district / khan / province - (ii) floor area - (iii) plot size - (iv) number of floors - (v) number of bedrooms - (vi) title of the property - (vii) property type - Appendix A: sample regression output for Central & East Phnom Penh (2020M1 to 2021M6) — selected model diagnostics: - Residual standard error: 0.2995 on 3361 degrees of freedom - Multiple R-squared: 0.7931, Adjusted R-squared: 0.7906 - F-statistic: 322.1 on 40 and 3361 DF, p-value: < 0.00000000000000022 - Robustness checks and diagnostics recommended: - Ensure estimated coefficients (shadow prices) are plausible in sign and magnitude. - Stability checks on coefficients for each rolling window and each stratum. - Analyze regression residuals to ensure no particular trends in the data. ### Aggregation, weights, and index compilation - Sub-indices should be aggregated using weights to higher-level indices including the national RPPI. - NBC have implemented weighting approach in published index. - 2020 expenditure weights within newly proposed strata: - Central & East Phnom Penh: Weights 39.4 (% of Total Expenditure); No. of observations 25.1 (% of Total) - Inner Suburbs Phnom Penh: Weights 34.0; No. of observations 36.8 - Outer Suburbs Phnom Penh: Weights 16.3; No. of observations 23.2 - Cambodia excluding Phnom Penh: Weights 10.3; No. of observations 14.9 - Total: Weights 100.0; No. of observations 100.0 - Observations: - Central & East Phnom Penh has weight of 39.4 percent. - Inner Suburbs Phnom Penh weight is 34.0 percent. - Cambodia excluding Phnom Penh accounts for 10.3 percent of the overall residential property market. ### Updated indices and comparison with published RPPIs - Price indices for houses compiled during mission using updated methods. - Comparison for 18-month period 2020M1 to 2021M6: - Published national RPPI shows year-on-year inflation of 16.0 percent. - Updated national RPPI shows inflation of 11.0 percent. - Recommendation: NBC should complete calculation of updated indices for the most recent period and compare with published RPPIs. ### Dissemination, methodological transparency, and operational notes - Timeliness: NBC receives loan data 10 days after reference month; validation checks normally completed by day 20; results disseminated at the start of the following month (example: results for November are published in early January). - Recommended dissemination actions and targets: - Implement improved methods into regular compilation of the RPPI by Dec-2024. - Decide on retrospective revision by Dec-2024 after analysis of updated indices. - Disseminate more detailed RPPIs by Feb-2025 (sub-indices for Phnom Penh strata). - Draft and publish a detailed methodological document on authorities’ website by Feb-2025. - Operational notes: - Bundle inclusion of micro-finance institution data with overall methods update to avoid multiple updates to published RPPI. - Monitor differences between updated and published indices (target Aug-2024 for initial monitoring). ### Recommended Actions (consolidated) - Transactions outside the scope of the RPPI should be excluded e.g., loans that were issued by the banks for the purpose of building a property. - Some property characteristics should be merged into homogenous groups. - Apply a more detailed stratification of the loan data. - Use Cook’s Distance for outlier detection. - Update the current model specification for the hedonic regressions. - Robustness checks should be applied to the results of the regressions. - Complete the calculation of the updated indices for the most recent period and compare with the published RPPI. - Implement the improved methods into their regular compilation of the RPPI. - Decide whether to revise the retrospective data. - Disseminate more detailed RPPIs. - Draft and publish a detailed methodological document on the authorities’ website. *Source: tarea2024037 - Section I. Detailed Technical Assessment and Recommendations* ### Section I. Detailed Technical Assessment and Recommendations ...................................................... 4 ### Section I. Detailed Technical Assessment and Recommendations ### Summary of Mission Outcomes and Priority Recommendations - A technical assistance (TA) mission was conducted from January 15–19, 2024, to assist the National Bank of Cambodia (NBC) with further improvement of their Residential Property Price Index (RPPI). The mission was conducted under the Data for Decisions (D4D) Trust Fund. - Tasks completed: (i) review of methods used for the RPPI; (ii) identification of areas for improvement consistent with international best practice; (iii) practical training to NBC staff; (iv) assessment of the suitability of the template used for data collection from commercial banks; (v) analysis of potential for accessing and using more comprehensive data sources. - Authorities are strongly committed to improving methodology and associated R scripts for the RPPI. - NBC published a new monthly RPPI data series in June 2022 using residential property loan data from Cambodian mortgage lending institutions. NBC receive the data 10 days after the reference month and results are disseminated at the start of the following month. - The mission identified key areas for improvement: - Filtering and cleaning of the loan data. - Stratification of the sample. - Outlier detection method. - Model specification for hedonic regressions. - Inclusion of additional loan data from micro-finance institutions. - Recommended methodological changes: - Continue using time dummy hedonic method with an 18-month window length. - Amend filters (e.g., lower minimum threshold for land area from 40 to 20 square meters). - Merge certain property characteristics into homogeneous groups (location categories within and outside Phnom Penh using provinces, districts/khans, communes/sangkats). - Apply more detailed stratification: four strata — (i) Central & East Phnom Penh, (ii) Inner Suburbs Phnom Penh, (iii) Outer Suburbs Phnom Penh, (iv) Cambodia excluding Phnom Penh. - Replace interquartile-range outlier detection (which excluded up to 18.3 percent in some months) with Cook’s Distance. - Reduce overfitting of hedonic model by updating model specification and applying robustness checks. - Impact of updated methods (houses): - Published national index: inflation of 16.0 percent between 2020M1 and 2021M6. - Updated index: inflation of 11.0 percent between 2020M1 and 2021M6. - NBC should implement updated methods into regular compilation and analyze results before deciding on retrospective revision. - Table 1 — Priority Recommendations (assigned to NBC): - Jun-2024: Complete the calculation of the updated indices for the most recent period and compare with the published RPPI. - Dec-2024: Implement the improved methods into their regular compilation of the RPPI. - Dec-2024: Decide whether to revise the retrospective data. ### RPPI Workplan for Cambodia, 2024 (Selected actions and targets) - Topic: Data sources - Continue to use the loan data to compile the RPPI in the medium term. Target Completion Date: Feb-2024. - Include additional loan data from micro-finance institutions in the sample as part of the overall update of the methods. Target Completion Date: Dec-2024. - Estimate existing coverage of the RPPI using loan data and administrative data sources. Target Completion Date: Dec-2024. - Devise a longer-term strategy to close coverage gap by researching potential new data sources (e.g., website listings). Target Completion Date: Dec-2024. - Topic: Compilation methods - Exclude transactions outside scope (e.g., loans issued for building a property). Status: Largely complete. - Apply more detailed stratification of the loan data. Status: Completed. - Use Cook’s Distance for outlier detection. Status: Completed. - Update hedonic model specification and apply robustness checks. Status: Largely complete. - Aggregate sub-indices using weights to higher-level indices including the national RPPI. Status: Completed. - Update R-code and organize external review of R scripts. Target Completion Date: Feb-2024. Priority: H. - Complete calculation of updated indices for recent period and compare with published RPPI. Target Completion Date: Jun-2024. Priority: H. - Monitor differences between updated and published indices. Target Completion Date: Aug-2024. - Re-run compilation with micro-finance institutions included and calculate updated RPPIs. Target Completion Date: Dec-2024. - Topic: Dissemination - Implement improved methods into regular compilation of the RPPI. Target Completion Date: Dec-2024. Priority: H. - Decide whether to revise retrospective data. Target Completion Date: Dec-2024. Priority: H. - Disseminate more detailed RPPIs. Target Completion Date: Feb-2025. Priority: H. - Draft and publish a detailed methodological document on the authorities’ website. Target Completion Date: Feb-2025. Priority: H. ### Data Sources — Key findings and recommendations - Data source: monthly survey of lending institutions. Data available from 2019; RPPI currently compiled with data from January 2020 (2019 data judged lower quality). - Average monthly observations: - 2020: 830 - 2021: 1,030 - 2022: 1,260 - 2023: 880 - Loan reporting template sections: (i) administrative information, (ii) borrower information, (iii) loan information, (iv) property information. - Property information fields relevant to RPPI: total price of purchased property (final market price), price of purchased land, appraisal price, location variables (province/city, district/khan, commune/sangkat, village, street number), size variables (floor area, plot size, number of bedrooms, width), and other variables (property type, title of property, floor number/number of floors, date of construction). - Date used to assign transaction to reference month: drawdown date for the loan. - Coverage limitations: sample limited to market financed by loans from commercial banks; segments not covered include (i) cash purchases, (ii) transactions financed by developer loans, (iii) transactions financed by foreign banks. - Recommended actions: - Continue using loan data in the medium term. - Include micro-finance institution loan data (predominantly covers areas outside Phnom Penh) as part of the overall methods update. - Estimate coverage by comparing loan data with administrative sources (e.g., property tax data at General Department of Taxation, MEF). - Research new data sources (e.g., listings from real estate websites) to close coverage gaps. - Condominiums and apartments: few observations in loan data; published RPPI covers houses only. Legal context: foreigners cannot own land or houses built on the ground; foreigners can own apartments/condominiums above ground floor as co-owned properties provided foreign ownership does not exceed 70 percent of total floor area. ### Compilation Methods — Technical adjustments and outcomes - Current approach: time dummy hedonic method with an 18-month window length; recommended to continue but with improvements. - Data inspection: visualizations and descriptive statistics (histograms, bar charts, scatter plots, mean/median price per district) used to assess data quality and relationships. - Transactions to exclude: loans for building a property (funds used to purchase materials/pay builders) should be excluded as no residential property transaction occurs. - Revised filters: - Minimum land area threshold lowered from 40 to 20 square meters. - Exclude properties with number of bedrooms greater than 9. - Exclude properties with number of floors greater than 4. - Property characteristic grouping: - Create location categories within Phnom Penh using khan and sangkat; outside Phnom Penh, merge provinces. About 6–10 categories per strata recommended depending on observations. - Create four categories for number of floors and seven categories for number of bedrooms. - Stratification detail and sample counts (before outlier detection) — Number of observations per Year and Strata: - Strata / Year — Central & East Phnom Penh / Inner Suburb Phnom Penh / Outer Suburb Phnom Penh / Cambodia excluding Phnom Penh / Total - 2020 — 2,495 / 3,669 / 2,313 / 1,483 / 9,960 - 2021 — 2,304 / 3,837 / 3,519 / 2,671 / 12,331 - 2022 — 2,362 / 3,748 / 5,525 / 3,492 / 15,127 - 2023 — 1,394 / 2,635 / 3,815 / 2,700 / 10,544 - All years — 8,555 / 13,889 / 15,172 / 10,346 / 47,962 - Outlier detection: - Existing method: interquartile range by month and strata; excluded up to 18.3 percent in some months. - Recommendation: replace with Cook’s Distance (regression-based) to identify outliers with greater precision and preserve sample size. - Hedonic model specification: - Existing model is overfitted with too many independent variables; can lead to misleading R-squared, coefficients, and p-values. - Recommendation: update specification, reduce overfitting, and apply robustness checks. ### Dissemination and implementation - Timeliness: NBC receives loan data 10 days after reference month; validation checks normally completed by day 20; results disseminated at the start of the following month (example: results for November are published in early January). - Recommended dissemination actions and targets: - Implement improved methods into regular RPPI compilation by Dec-2024. - Decide on retrospective revision by Dec-2024 after analysis of updated indices. - Disseminate more detailed RPPIs by Feb-2025 (sub-indices for Phnom Penh strata). - Draft and publish a detailed methodological document on authorities’ website by Feb-2025. - Operational notes: - Bundle inclusion of micro-finance institution data with overall methods update to avoid multiple updates to published RPPI. - Monitor differences between updated and published indices (target Aug-2024 for initial monitoring). *Source: tarea2024037 - Section I. Detailed Technical Assessment and Recommendations* ### 26. The authorities should use Cook’s Distance for outlier detection. The existing methodology uses ### The authorities should use Cook’s Distance for outlier detection. The existing methodology uses ### Outlier detection: issue and recommendation - Existing method: interquartile range by month and strata to identify outliers. - Problem: method excludes a very high proportion of the sample. - Within Phnom Penh, excludes between 6.9 percent and 15.9 percent of the data. - Outside Phnom Penh, excludes between 10.3 percent and 18.3 percent of the data. - Recommendation: replace the interquartile range method with Cook’s Distance (a regression-based method). - Rationale: identifies outliers with greater precision and preserves the sample to the extent possible. - Mission application results (proportion of identified outliers across strata and years): - Ranges from 4.1 percent to 7.0 percent across all strata and years. - Identified Outliers Per Year and Strata, Percent of Observations (Table 4): - 2020: Central & East Phnom Penh 6.6; Inner Suburbs Phnom Penh 5.8; Outer Suburbs Phnom Penh 7.0; Cambodia excluding Phnom Penh 5.9; Total 6.3 - 2021: Central & East Phnom Penh 5.9; Inner Suburbs Phnom Penh 5.2; Outer Suburbs Phnom Penh 4.1; Cambodia excluding Phnom Penh 6.6; Total 5.3 - 2022: Central & East Phnom Penh 8.8; Inner Suburbs Phnom Penh 7.2; Outer Suburbs Phnom Penh 5.0; Cambodia excluding Phnom Penh 7.0; Total 6.6 - 2023: Central & East Phnom Penh 6.6; Inner Suburbs Phnom Penh 5.8; Outer Suburbs Phnom Penh 6.0; Cambodia excluding Phnom Penh 6.0; Total 6.0 - All years: Central & East Phnom Penh 7.0; Inner Suburbs Phnom Penh 6.0; Outer Suburbs Phnom Penh 5.4; Cambodia excluding Phnom Penh 6.5; Total 6.1 ### Model specification: overfitting concerns and preferred variables - Problem: existing RPPI hedonic model is overfitted with too many independent variables (e.g., sangkat dummies for 105 sangkats), producing misleading R-squared values, coefficients, and p-values. - Location variable recommendations: - Remove sangkat variable. - Replace with aggregated location variable such as khan or categories created using a combination of khan and sangkat. - Outside Phnom Penh, create location categories using a combination of province and district. - Variable-level finding: - Coefficient for width of the property was not statistically significant and could be removed. - Preferred model includes the following property characteristics: - (i) location e.g., district / khan / province - (ii) floor area - (iii) plot size - (iv) number of floors - (v) number of bedrooms - (vi) title of the property - (vii) property type - Appendix A: sample regression output for Central & East Phnom Penh (2020M1 to 2021M6) — selected model diagnostics: - Residual standard error: 0.2995 on 3361 degrees of freedom - Multiple R-squared: 0.7931, Adjusted R-squared: 0.7906 - F-statistic: 322.1 on 40 and 3361 DF, p-value: < 0.00000000000000022 ### Robustness checks and diagnostics - Recommended analyses to apply to regression results: - Ensure estimated coefficients (shadow prices) are plausible in sign and magnitude. - Stability checks on coefficients for each rolling window and each stratum. - Analyze regression residuals to ensure no particular trends in the data. ### Aggregation, weights, and index compilation - Sub-indices should be aggregated using weights to higher-level indices including the national RPPI. - NBC have implemented weighting approach in published index. - 2020 expenditure weights within newly proposed strata (Table 5): - Central & East Phnom Penh: Weights 39.4 (% of Total Expenditure); No. of observations 25.1 (% of Total) - Inner Suburbs Phnom Penh: Weights 34.0; No. of observations 36.8 - Outer Suburbs Phnom Penh: Weights 16.3; No. of observations 23.2 - Cambodia excluding Phnom Penh: Weights 10.3; No. of observations 14.9 - Total: Weights 100.0; No. of observations 100.0 - Observations: - Central & East Phnom Penh has weight of 39.4 percent. - Inner Suburbs Phnom Penh weight is 34.0 percent. - Cambodia excluding Phnom Penh accounts for 10.3 percent of the overall residential property market. ### Updated indices and comparison with published RPPIs - Price indices for houses compiled during mission using updated methods. - Comparison for 18-month period 2020M1 to 2021M6: - Published national RPPI shows year-on-year inflation of 16.0 percent. - Updated national RPPI shows inflation of 11.0 percent. - Recommendation: NBC should complete calculation of updated indices for the most recent period and compare with published RPPIs. ### Dissemination and methodological transparency - Implement improved methods into regular compilation of the RPPI. - NBC should analyze updated indices before deciding whether to revise retrospective data. - Option: apply updated methods from the next update of annual weights only — note this would introduce a structural change in the index from the month the new methods were introduced. - Alternative: apply new methods from the start of the time series for consistent compilation over time. - Dissemination improvements: - Introduce a more granular stratification of the data to publish more detailed sub-indices within Phnom Penh. - Benefit: users obtain information on contribution of different areas of the capital region to overall residential property inflation. - Methodological documentation: - Draft and publish a detailed methodological document including: data source, coverage, data quality and cleaning, outlier detection, stratification, weighting information, and methods for index compilation and aggregation. - NBC already have a draft paper that can be used as the basis. ### Recommended Actions (as listed) - Transactions outside the scope of the RPPI should be excluded e.g., loans that were issued by the banks for the purpose of building a property. - Some property characteristics should be merged into homogenous groups. - Apply a more detailed stratification of the loan data. - Use Cook’s Distance for outlier detection. - Update the current model specification for the hedonic regressions. - Robustness checks should be applied to the results of the regressions. - Complete the calculation of the updated indices for the most recent period and compare with the published RPPI. - Implement the improved methods into their regular compilation of the RPPI. - Decide whether to revise the retrospective data. - Disseminate more detailed RPPIs. - Draft and publish a detailed methodological document on the authorities’ website. * IMF | Technical Assistance Report – Cambodia RPPI (excerpt) * --- _Source: https://www.imf.org/-/media/files/publications/tar/2024/english/tarea2024037.pdf_