## 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/cr/2023/english/1bgdea2023002.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/cr/2023/english/1bgdea2023002.pdf.md)
- [Structured JSON version](/-/media/files/publications/cr/2023/english/1bgdea2023002.pdf.json)

---

### Introduction and Mission Objectives
- Technical assistance mission: February 26 to March 2, 2023; funded by the Data for Decisions (D4D) Trust Fund.
- Pre-mission meeting: October 18, 2022.
- Objectives:
  - Capacity building on RPPI methods.
  - Advanced data analysis using microdata provided by Bangladesh Bank (BB).
  - Preparation for compiling an experimental RPPI.
- Staffing and follow-up:
  - Current RPPI team: three members; mission provided end-to-end training due to staff turnover.
  - On-site meetings attended by approximately 20 staff.
  - Mission to continue remote collaboration as recommendations are implemented.

### Data Sources for the RPPI — Findings
- Administrative property transfer data unsuited as source data because:
  - Lack of stakeholder acceptance about accuracy of declared prices for taxation and registration.
  - Administrative data not accessible to BB; registration system not digitalized in a central database.
  - No agreed mechanism for data sharing between BB and other public agencies.
- Current primary source: quarterly Delta Brac Housing Finance Corporation Ltd (DBH) home loan data.
  - DBH dataset fields: (i) date of disbursement of the loan, (ii) the valuation price, (iii) the size in square feet of the dwelling, (iv) the location of the dwelling.
  - Data available since 1998 with over 42,000 observations in total until Q2 2022.
  - BB has compiled experimental indices internally using simple means, medians, hedonic methods.
- Valuation prices used as proxy for transaction prices; challenges noted:
  - Valuation prices can lag transaction prices and may deviate from transaction prices over time.
- New reporting template pilot:
  - Initiated January 2022 with eight financial institutions and non-banks.
  - Collects 42 variables (outlined in Appendix A) and uses an MS Access database developed by the RPPI team.
  - Expanded collection expected to increase sample sizes (e.g., areas outside Dhaka) and provide additional location/property characteristics.
  - Response rate has been low; impediments include voluntary participation and limited RPPI staff resources.

### Data Sources — Recommended Actions
- Use the DBH dataset to compile an experimental RPPI for Dhaka using a single preferred method aligned with international best practice.
- Use valuation prices for compilation while recognizing potential downsides.
- Make the new reporting template mandatory.
- Provide additional human resources for data collection, respondent management and data cleaning.
- Use targeted respondent management based on market share of each financial institution and non-bank (market share defined as individual proportions of the overall value of properties transacted over a 12-month period, updated regularly).
- In medium to long term, explore listings data from property websites to expand coverage to land and commercial property.

### Methods for Compilation — Scope, Stratification, and Weighting
- Scope and sample period:
  - Compile experimental RPPI for the city of Dhaka using data from Q1 2007.
  - Rationale: low observations outside Dhaka and before Q1 2007 may not represent the market well.
- Stratification:
  - Stratify sample into four regional strata within Dhaka using the DBH location variable (proposed breakdown in Appendix B).
  - Calculate experimental index using these four strata initially; reassess if sub-indices or headline index are too volatile.
- Weighting:
  - Use expenditure weights (flow weights) for strata to compile overall RPPI for Dhaka; BB had used number of observations, which is incorrect.
  - Expenditure weights calculated by summing price information within each strata over a 12-month period.
- Proposed expenditure weights by strata for 2019 (Appendix B, TABLE 1):
  - Stratum 1 — Weights (Expenditure): 37.6%, No. of observations (% of Total): 26.5%
  - Stratum 2 — Weights (Expenditure): 17.4%, No. of observations (% of Total): 26.9%
  - Stratum 3 — Weights (Expenditure): 25.8%, No. of observations (% of Total): 23.1%
  - Stratum 4 — Weights (Expenditure): 19.2%, No. of observations (% of Total): 23.5%

### Outliers, Data Quality, and Diagnostics
- Implement outlier and error detection by strata and quarter, with particular reference to the floor area variable.
- Visual inspection: number of floor area outliers increased significantly from 2017.
- Recommended treatment:
  - Use interquartile range approaches for floor area outliers.
  - Explore thresholds and use Cooke’s Distance as part of hedonic regression diagnostics.
- Typical outlier share in experimental models: around 5 to 10 percent of observations.

### Hedonic Modelling and Quality Adjustment
- Model approach:
  - Use rolling window time-dummy approach for quality adjustment within strata due to low observation counts (pooling improves model quality and coefficient stability).
  - Models for dwellings should include property-mix adjustments based on (i) floor area and (ii) location.
  - Combine detailed location categories before including the location variable in regional-stratum models to ensure adequate observations per category (combine particularly for Strata 1 and 4; some combining for Strata 3; limited options for Strata 2).
  - Finalize model specifications (e.g., whether to take log of floor area) by Jul 31, 2023.
- Further testing required:
  - Staff should complete testing of alternative models and examine stability of coefficients over time as a key criterion.

### Sample Regression Output — Stratum 3, Dhaka, 2021 (Appendix D)
- Coefficients and diagnostics:
  - (Intercept) — Estimate: 8.360, Std Error: 0.214, Statistic: 39.099, P-value: 0.000 ***
  - log_area — Estimate: 1.036, Std Error: 0.029, Statistic: 35.838, P-value: 0.000 ***
  - Uttara (reference) — Estimate: 0.000, Std Error: 0.000, Statistic: 0.000, P-value: 0.000
  - Basundhara — Estimate: 0.077, Std Error: 0.027, Statistic: 2.867, P-value: 0.004 **
  - Bhatara — Estimate: -0.449, Std Error: 0.072, Statistic: -6.245, P-value: 0.000 ***
  - Cantonment — Estimate: -0.332, Std Error: 0.165, Statistic: -2.007, P-value: 0.045 *
  - Dakkhinkhan — Estimate: -0.386, Std Error: 0.032, Statistic: -12.143, P-value: 0.000 ***
  - Dhaka_cantonment — Estimate: -0.393, Std Error: 0.069, Statistic: -5.692, P-value: 0.000 ***
  - Hajipara — Estimate: -0.333, Std Error: 0.165, Statistic: -2.026, P-value: 0.043 *
  - Joar_shahara — Estimate: -0.394, Std Error: 0.072, Statistic: -5.469, P-value: 0.000 ***
  - Khilkhet — Estimate: -0.506, Std Error: 0.084, Statistic: -6.016, P-value: 0.000 ***
  - Nadda — Estimate: -0.368, Std Error: 0.232, Statistic: -1.583, P-value: 0.114
  - Shahjadpur — Estimate: -0.158, Std Error: 0.135, Statistic: -1.173, P-value: 0.241
  - Uttarkhan — Estimate: -0.283, Std Error: 0.055, Statistic: -5.175, P-value: 0.000 ***
  - 2021 Q1 (reference) — Estimate: 0.000, Std Error: 0.000, Statistic: 0.000, P-value: 0.000
  - 2021 Q2 — Estimate: 0.030, Std Error: 0.033, Statistic: 0.918, P-value: 0.359
  - 2021 Q3 — Estimate: 0.005, Std Error: 0.032, Statistic: 0.170, P-value: 0.865
  - 2021 Q4 — Estimate: 0.037, Std Error: 0.030, Statistic: 1.255, P-value: 0.210
- Model diagnostics:
  - Residual standard error: 0.2312 on 464 degrees of freedom
  - Multiple R-squared: 0.8203
  - Adjusted R-squared: 0.8145
  - F-statistic: 141.2 on 15 and 464 DF, p-value: < 2.2e-16

### Aggregation and Index Formula
- Aggregation recommendation:
  - Use an annually chained Laspeyres-type aggregation formula to calculate the headline RPPI.
  - Rationale: annually updated weights ensure weights remain representative of the structure of property transactions.
- Implementation guidance:
  - Assess volatility of sub-indices and headline index; if volatile, consider smoothing before publication.

### Compilation Process, Tools, and Staff Capacity
- Technical transfer:
  - Detailed presentations given on model specification, interpretation of regression results, calculating price indices from time-dummy coefficients, and aggregating sub-indices using annually updated weights.
- R code:
  - BB already had R code for experimental indices; additional R code provided to implement recommended improvements.
  - Intention: continue to use R code for RPPI compilation.
- Staff capacity:
  - Varying experience levels among participants; detailed instruction provided to support implementation.

### Dissemination Guidance
- Publication recommendation:
  - Consider publishing the experimental RPPI on the BB website along with sub-indices for the four regional strata to increase user benefits and transparency.
  - Accompany publication with metadata and methodology documents.
  - BB should obtain agreement from DBH in advance of wider publication and provide assurance of data confidentiality.
- Smoothing and publication options:
  - If indices suffer from volatility, apply smoothing (simplest: a two-quarter moving average).
  - Publication should include a short note explaining why volatility is present if smoothing is applied.
  - Alternative: publish the headline index but refrain from publishing one or more sub-indices; unpublished sub-indices should still be included in the calculation of the headline index.
- Documentation:
  - Draft a detailed methodological document regardless of publication decision.

### Medium- to Long-Term Development
- Geographic expansion:
  - Expand coverage beyond Dhaka to areas adjacent to Dhaka and other parts of Bangladesh by adding additional strata when data permit.
  - Option to maintain long Dhaka time series while publishing a wider-coverage aggregate with a shorter time series.
- Use of new reporting-template variables:
  - The 42 variables in the new reporting template (Appendix A) can improve stratification and model specification through additional location and property attributes.
  - Apply comprehensive testing of new methods before implementation.
- Coverage extension:
  - Expand property price indicators to include land and commercial property; listings data noted as potential source.

### Consolidated Recommended Actions and Priorities (selected milestones)
- Immediate technical implementation:
  - Apr 30, 2023: Carry out detailed visual examination of the DBH dataset; limit scope to dwellings in Dhaka from Q1 2007; use four regional strata.
  - May 31, 2023: Use expenditure weights for strata; implement outlier and error detection by quarter and strata (emphasize floor area).
  - Jun 30, 2023: Use rolling window time-dummy approach for quality adjustment within strata.
  - Jul 31, 2023: Combine location categories for regional-stratum models; finalize model specifications (e.g., log of floor area).
  - Aug 31, 2023: Use an annually chained Laspeyres-type aggregation formula to calculate the headline RPPI.
  - Sep 30, 2023: Update R code to implement compilation-method improvements.
- Publication and staffing:
  - Sep 2023: Implement recommended improvements to compilation methods for the experimental RPPI using the DBH dataset.
  - Oct 31, 2023: Consider publishing the experimental RPPI on the BB website with sub-indices for the four regional strata.
  - Nov 30, 2023: Apply data smoothing to the indices as required; provide training for new staff.
  - Dec 31, 2023: Draft detailed methodological document; make the new reporting template mandatory; provide additional human resources for data collection, respondent management and data cleaning.

*Source: Section I. Detailed Technical Assessment and Recommendations, Bangladesh Residential Property Price Index (RPPI) technical report.*

### Section I. Detailed Technical Assessment and Recommendations ...................................................... 5

### Section I. Detailed Technical Assessment and Recommendations

### Introduction
- A technical assistance mission was conducted from February 26 to March 2, 2023, to assist the Bangladesh Bank (BB) with the ongoing development of their Residential Property Price Index (RPPI). The mission was funded by the Data for Decisions (D4D) Trust Fund.
- A pre-mission meeting was held on October 18, 2022, attended by officials from the BB, the mission team, and the local IMF Resident Representative.
- Objectives of the mission included: capacity building on RPPI methods, advanced data analysis using microdata provided by BB, and preparation for compiling an experimental RPPI.
- There are currently three members of the RPPI team; due to staff turnover, the mission provided training on all aspects of RPPI compilation. On-site meetings were attended by approximately 20 staff.
- The mission will continue remote collaboration with BB as recommendations are implemented.

### Data Sources For The RPPI
Findings:
- Administrative data on property transfers are unsuited as source data for the RPPI:
  - Lack of acceptance among stakeholders about the accuracy of declared prices for taxation and registration.
  - Administrative data sources are not accessible to BB; the property registration system has not been digitalized in a central database.
  - No agreed mechanism for data sharing between BB and other public agencies.
- Current primary data source: quarterly data from Delta Brac Housing Finance Corporation Ltd (DBH) on home loans.
  - DBH dataset fields include: (i) date of disbursement of the loan, (ii) the valuation price, (iii) the size in square feet of the dwelling, and (iv) the location of the dwelling.
  - Data available since 1998 with over 42,000 observations in total until Q2 2022.
  - BB has compiled experimental indices internally using various methods (simple means, medians, hedonic methods).
- Valuation prices are used as a proxy for transaction prices, with noted challenges:
  - Valuation prices can lag transaction prices and may deviate from transaction prices over time.
- New reporting template pilot:
  - Initiated January 2022 with eight financial institutions and non-banks.
  - Collects 42 variables (outlined in Appendix A) and uses an MS Access database solution developed by the RPPI team.
  - Expanded data collection will increase sample sizes (e.g., areas outside Dhaka) and provide additional location/property characteristics.
- Response rate for the new reporting template has been low; impediments include voluntary participation and limited RPPI staff resources.

Recommended Actions:
- Use the DBH dataset to compile an experimental RPPI for Dhaka using a single preferred method that follows international best practice.
- Use valuation prices for compilation while being cognizant of their potential downsides.
- Make the new reporting template mandatory.
- Provide additional human resources for data collection, respondent management and data cleaning.
- Use targeted respondent management based on the market share of each financial institution and non-bank (market share defined as individual proportions of the overall value of properties transacted over a 12-month period, updated regularly).
- In the medium to long term, explore listings data from property websites to expand coverage to land and commercial property (websites examined included property advertising sites with residential, land and commercial listings).

### Methods For Compilation
Findings and Recommendations:
- A visual examination of the DBH dataset was carried out using frequency charts, histograms, and box & whisker plots to examine distributions, correlations and trends. A similar approach should be applied to data from the new reporting template.
- BB has developed 10 price indices using various methods (simple means, medians, hedonic regressions including imputation approach and time dummy method). The mission recommends refining methods and selecting a single headline approach aligned with international best practice.
- Scope and sample period:
  - Compile an experimental RPPI for the city of Dhaka using data from Q1 2007. Rationale: low number of observations for areas outside Dhaka and for periods before Q1 2007 may not represent the market well.
- Stratification:
  - Stratify the sample into four regional strata within Dhaka using the location variable in the DBH dataset (proposed regional breakdown in Appendix B).
  - Calculate an experimental index using these four strata initially and assess volatility of sub-indices and headline index; revisit stratification if indices are too volatile.
- Weighting:
  - Use expenditure weights (i.e., flow weights) for strata to compile overall RPPI for Dhaka. BB had used number of observations, which is incorrect.
  - Expenditure weights calculated by summing price information within each strata over a 12-month period.
  - Example for 2019: expenditure weight for stratum 1 (Gulshan and Dhanmondi) is 37.6 percent while stratum 1 contains 26.5 percent of total observations; stratum 2 contains 26.9 percent of observations while the expenditure weight is 17.4 percent.
- Outliers and errors:
  - Implement a system for outlier and error detection by strata and quarter, with particular reference to the floor area variable.
  - Visual inspection shows the number of floor area outliers increased significantly from 2017.
  - Recommended treatment of floor area based on interquartile range approaches; explore thresholds and use Cooke’s Distance as part of hedonic regression diagnostics.
- Hedonic modelling:
  - Use the rolling window time-dummy approach for quality adjustment within strata due to low numbers of observations (pooling improves model quality and coefficient stability).
  - Models for dwellings should include property-mix adjustments based on (i) floor area and (ii) location.
  - Combine detailed location categories before including the location variable in regional-stratum models to ensure adequate observations per category; combine particularly for Strata 1 and Strata 4, some combining for Strata 3, limited options for Strata 2 (most properties in Mirpur with no further detail).
  - Finalize model specifications (e.g., whether or not to take log of floor area) by Jul 31, 2023.

Priority Recommendations (from Summary):
- Sep 2023: The BB should implement the recommended improvements to the compilation methods for the experimental RPPI using the DBH dataset.
- Oct 2023: The BB should consider publishing the experimental RPPI along with the sub-indices for the four regional strata.
- Dec 2023: The BB should make the new reporting template mandatory and provide additional human resources for data collection, respondent management and data cleaning.

Action Plan — Selected Milestones and Target Completion Dates:
Topic: Improved compilation methods for the experimental RPPI
- Apr 30, 2023: Carry out a detailed visual examination of the DBH dataset.
- Apr 30, 2023: Limit the scope of the index to dwellings in the city of Dhaka for the period from Q1 2007.
- Apr 30, 2023: Use four regional strata within Dhaka.
- May 31,2023: Use expenditure weights (i.e., flow weights) for the strata.
- May 31, 2023: Implement a system for outlier and error detection by quarter and strata, with reference to the floor area variable in particular.
- Jun 30, 2023: Use the rolling window time-dummy approach for quality adjustment within strata.
- Jul 31,2023: Combine categories for location before including the variable in the models for the regional strata.
- Jul 31, 2023: Finalize the specifications of the models e.g., whether or not to take log of floor area.
- Aug 31, 2023: Use an annually chained Laspeyres-type aggregation formula to calculate the headline RPPI.
- Sep 30, 2023: Update the R code to implement the improvements to the compilation methods.

Topic: Publication of the experimental RPPI
- Oct 31, 2023: Consider publishing the experimental RPPI on the BB website along with the sub-indices for the four regional strata.
- Nov 30, 2023: Apply data smoothing to the indices as required.
- Dec 31, 2023: Draft a detailed methodological document.

Topic: Data collection using the new reporting template
- Sep 30, 2023: Provide additional human resources for data collection, respondent management and data cleaning.
- Nov 30, 2023: Provide training for new staff.
- Nov 30, 2023: Use targeted respondent management based on the market share of each financial institution and non-bank.
- Dec 31, 2023: Make the new reporting template mandatory.

### Dissemination
- The BB currently disseminates price indices internally only.
- Recommendation: Consider publishing the experimental RPPI on the BB website along with sub-indices for the four regional strata to increase benefits to users and transparency.
- Accompany publication with metadata and methodology documents.
- Apply data smoothing to indices as required before publication.

### Medium to long-Term Development
- Expand geographical coverage beyond Dhaka to include areas adjacent to Dhaka and other parts of Bangladesh by adding additional strata to the aggregation when data permit.
- The new reporting template’s 42 variables can be used to improve stratification and model specification:
  - Additional location and property attributes will allow better mix-adjustment by updating stratification and adding variables to regression models.
  - Comprehensive testing of new methods should be applied before implementation.
- Explore listings data from property websites (via webscraping or direct data provision) to expand coverage to land and commercial property after the RPPI is established.

### Officials Met During The Mission
- On-site meetings at BB were attended by approximately 20 staff to ensure wide dissemination of knowledge. (No further individual official names or titles are provided in the supplied content.)

### Appendices (referenced)
- Appendix A: New reporting template for pilot survey (42 variables).
- Appendix B: Proposed regional stratification in Dhaka and related expenditure weights (includes example weights for 2019: stratum 1 expenditure weight 37.6 percent vs. 26.5 percent of observations; stratum 2 observations 26.9 percent vs. expenditure weight 17.4 percent).
- Appendix C: Data analysis (includes graphs of observations by quarter and distribution of log floor area by quarter).
- Appendix D: Diagnostic results.

*Source: Section I. Detailed Technical Assessment and Recommendations, Bangladesh Residential Property Price Index (RPPI) technical report.*

### 27. Both semi-log and log-log models were examined in response to the positive distribution

### 27. Both semi-log and log-log models were examined in response to the positive distribution

### Model specification, diagnostics, and estimation
- Experimental models: semi-log and log-log regressions examined in two-stage runs:
  - Stage 1: identify outlier observations using Cook’s Distance.
  - Stage 2: rerun models without outliers.
- Typical outlier share: around 5 to 10 percent of observations.
- Regression performance:
  - Generally high explanatory power.
  - Explanatory variables were statistically significant and in line with a priori expectations (in respect of both sign and slope) over time.
  - Sample regression diagnostics presented for Stratum 3 in Appendix D.
- Further testing required:
  - Staff should complete testing of alternative models.
  - Decision on whether to include the log of floor area in the model should be examined further.
  - Key criterion: analyze the stability of the coefficients over time.

- Sample regression output for Strata 3 in Dhaka, 2021 (from Appendix D):
  - Coefficients and diagnostics:
    - (Intercept) — Estimate: 8.360, Std Error: 0.214, Statistic: 39.099, P-value: 0.000 ***
    - log_area — Estimate: 1.036, Std Error: 0.029, Statistic: 35.838, P-value: 0.000 ***
    - Uttara (reference) — Estimate: 0.000, Std Error: 0.000, Statistic: 0.000, P-value: 0.000
    - Basundhara — Estimate: 0.077, Std Error: 0.027, Statistic: 2.867, P-value: 0.004 **
    - Bhatara — Estimate: -0.449, Std Error: 0.072, Statistic: -6.245, P-value: 0.000 ***
    - Cantonment — Estimate: -0.332, Std Error: 0.165, Statistic: -2.007, P-value: 0.045 *
    - Dakkhinkhan — Estimate: -0.386, Std Error: 0.032, Statistic: -12.143, P-value: 0.000 ***
    - Dhaka_cantonment — Estimate: -0.393, Std Error: 0.069, Statistic: -5.692, P-value: 0.000 ***
    - Hajipara — Estimate: -0.333, Std Error: 0.165, Statistic: -2.026, P-value: 0.043 *
    - Joar_shahara — Estimate: -0.394, Std Error: 0.072, Statistic: -5.469, P-value: 0.000 ***
    - Khilkhet — Estimate: -0.506, Std Error: 0.084, Statistic: -6.016, P-value: 0.000 ***
    - Nadda — Estimate: -0.368, Std Error: 0.232, Statistic: -1.583, P-value: 0.114
    - Shahjadpur — Estimate: -0.158, Std Error: 0.135, Statistic: -1.173, P-value: 0.241
    - Uttarkhan — Estimate: -0.283, Std Error: 0.055, Statistic: -5.175, P-value: 0.000 ***
    - 2021 Q1 (reference) — Estimate: 0.000, Std Error: 0.000, Statistic: 0.000, P-value: 0.000
    - 2021 Q2 — Estimate: 0.030, Std Error: 0.033, Statistic: 0.918, P-value: 0.359
    - 2021 Q3 — Estimate: 0.005, Std Error: 0.032, Statistic: 0.170, P-value: 0.865
    - 2021 Q4 — Estimate: 0.037, Std Error: 0.030, Statistic: 1.255, P-value: 0.210
  - Residual standard error: 0.2312 on 464 degrees of freedom
  - Multiple R-squared: 0.8203
  - Adjusted R-squared: 0.8145
  - F-statistic: 141.2 on 15 and 464 DF, p-value: < 2.2e-16
  - Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1

### Aggregation, weights, and index formula
- Aggregation recommendation:
  - Use an annually chained Laspeyres-type aggregation formula to calculate the headline RPPI.
  - Rationale: annually updated weights ensure that the weights remain representative of the structure of property transactions.
- Strata and weighting guidance:
  - Limit the scope of the index to dwellings in the city of Dhaka for the period from Q1 2007.
  - Use four regional strata within Dhaka.
  - Use expenditure weights (i.e., flow weights) for the strata.
  - Assess volatility of resulting sub-indices and headline index.
- Proposed expenditure weights by strata for 2019 (Appendix B, TABLE 1):
  - Stratum 1 — Weights (Expenditure): 37.6%, No. of observations (% of Total): 26.5%
  - Stratum 2 — Weights (Expenditure): 17.4%, No. of observations (% of Total): 26.9%
  - Stratum 3 — Weights (Expenditure): 25.8%, No. of observations (% of Total): 23.1%
  - Stratum 4 — Weights (Expenditure): 19.2%, No. of observations (% of Total): 23.5%

### Data handling, outlier detection, and quality adjustment
- Implement an outlier and error detection system by quarter and strata, with particular reference to the floor area variable.
- Use the rolling window time-dummy approach for quality adjustment within strata.
- Combine categories for location before including the variable in the models for the regional strata.
- Finalize model specifications (e.g., whether or not to take log of floor area).

### Compilation process, tools, and staff capacity
- Compilation process notes:
  - Technical aspects of index compilation and aggregation were presented during the mission.
  - Detailed information was presented on specifying models, interpreting regression results, calculating price indices from time-dummy coefficients, and aggregating sub-indices using annually updated weights.
- R code:
  - BB already had R code for experimental indices.
  - Additional R code was provided to assist required improvements to compilation methods.
  - Intention of BB: continue to use R code for RPPI compilation.
- Staff capacity:
  - Varying levels of experience among participants; detailed instruction provided to support understanding and implementation.

### Dissemination guidance
- Publish experimental RPPI on BB website along with the sub-indices for the four regional strata is recommended.
  - Currently, price indices compiled by BB are disseminated internally.
  - Transparent publication with accompanying metadata and methodology documents would mitigate representativeness concerns and improve user benefits.
  - BB should obtain agreement from DBH in advance of wider publication and provide assurance of data confidentiality.
- Smoothing and publication options:
  - If indices suffer from volatility, a smoothing method can be applied; simplest method: a two-quarter moving average.
  - Publication should include a short note explaining why volatility is present if smoothing is applied.
  - Alternative: publish the headline index but not publish one or more sub-indices; unpublished sub-indices should still be included in calculation of the headline index.
- Documentation:
  - A detailed methodological document should be drafted regardless of publication decision to finalize compilation steps, ensure transparency, and solidify staff knowledge.

### Medium to long-term development
- Geographic coverage:
  - Expand the geographical coverage of the RPPI to include areas adjacent to Dhaka and other parts of Bangladesh.
  - Option when expanding: maintain a long time series for Dhaka while publishing a new aggregate index with wider coverage but a shorter time series.
- Use of additional variables:
  - The new reporting template contains 42 variables (see Appendix A).
  - Additional location and property attributes can improve mix-adjustment via updated stratification or additional regression variables.
  - Comprehensive testing of new methods before implementation is required.
- Coverage extension:
  - Expand property price indicators to include land and commercial property.
  - Listings data is noted as a potential source for price information on land and commercial property.

### Recommended actions (consolidated)
- Carry out a detailed visual examination of the DBH dataset.
- Limit the scope to dwellings in the city of Dhaka from Q1 2007.
- Use four regional strata within Dhaka; assess volatility of sub-indices and headline index.
- Use expenditure weights (flow weights) for the strata.
- Implement quarter- and strata-level outlier and error detection, with emphasis on the floor area variable.
- Use rolling window time-dummy approach for quality adjustment within strata.
- Combine location categories before including location in regional models.
- Finalize model specifications (e.g., whether to take log of floor area).
- Use an annually chained Laspeyres-type aggregation formula for the headline RPPI.
- Update R code to implement compilation-method improvements.
- Consider publishing experimental RPPI and sub-indices on BB website with metadata and methodology.
- Consider smoothing (e.g., two-quarter moving average) if indices are volatile; accompany publication with explanatory note.
- Draft a detailed methodological document.
- Expand geographic coverage, use additional reporting-template variables to improve stratification and model specification, and expand coverage to include land and commercial property.

*IMF | Technical Report – Bangladesh Residential Property Price Index (RPPI) — content unit 27–35 and appendices (excerpts).*

---


_Source: https://www.imf.org/-/media/files/publications/cr/2023/english/1bgdea2023002.pdf_