## wpiea2024147-print-pdf

## Source details

**Canonical URL:** [wpiea2024147-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024147-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024147-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024147-print-pdf.pdf.json)

---

### Overview
- Purpose: Present an integrated suite of forecasting models for nowcasting and near-term forecasting (NTF) of Cambodia’s economy, specifically real GDP and inflation.
- Nowcasting defined as forecast of the present to assess the current state of the economy.
- Near-term forecasting defined as the economic outlook for the next couple of quarters (near term in this paper refers to the next two quarters).

### Four-step integrated approach (process summary)
- Step 1 — Interpolate historical quarterly (real) GDP:
  - Estimate historical quarterly GDP using high-frequency indicators closely related to economic activity (monthly exports, credit, foreign reserves, broad money).
  - Methodology referenced: regression-based Chow-Lin disaggregation method; robustness checked with Litterman method.
- Step 2 — Nowcast current-quarter GDP:
  - Use high-frequency data on credit, exports of garment, and tourist arrivals, and quarterly GDP of Cambodia’s largest trading partners (US, China, EU).
  - Nowcasting models: Bridge, MIDAS, U-MIDAS.
- Step 3 — Current-quarter headline CPI inflation forecast (bottom-up):
  - Forecast each key CPI component for the current quarter and aggregate to obtain headline CPI forecasts.
  - Models referenced for monthly inflation forecasting: ARMA, ARMAX, VAR, VARX.
- Step 4 — Joint near-term forecast of GDP and inflation:
  - VAR models (and VARX with exogenous foreign variables) to jointly forecast GDP and inflation for the next two quarters, capturing feedback effects.
  - Models include other macro variables: credit growth and exchange rate.
  - Latest available “historical” data in VARs include the GDP nowcasts and inflation forecasts from Steps 2 and 3.
  - Key foreign exogenous variables: global commodity prices and key trading partners’ GDP growth and inflation.

### Key features of Cambodia’s economy (selected indicators and facts)
- Garment exports account for about 60 percent of total exports and about 1/3 of growth on average during 2003-2020.
- Tourism and related services contribute about 15% of economic activities.
- Since the Covid-19 outbreak, strong growth in non-garment exports (travel goods, electronic components, bicycles).
- Top export destinations: European Union (EU), the United States (US), and China.
- China: key source of foreign direct investment, input materials for production, and tourism (until Covid-19).
- Important import partners for food and input materials: Vietnam and Thailand.
- Riels have floated within a range of 4000-4200 Riels per US dollar since 2000.
- Financial system highly dollarized: more than 80 percent of loans, deposits, and payments denominated in US dollar.
- Credit growth: average credit growth was about 20 percent on average between 2000-2022.
- Credit reached more than 150 percent of GDP in 2023.
- Inflation drivers: more associated with global commodity and energy prices, or inflation in neighboring trading partners (Vietnam and Thailand) rather than exchange-rate pass-through.
- Policy context: high dollarization limits scope and effectiveness of monetary policy; NBC influences credit growth and liquidity via reserve and capital requirements; macroeconomic management relies more on fiscal policy.

### Interpolation procedure and indicator selection (Appendix 1 and Annex I)
- Interpolation methods discussed: Chow-Lin (residuals follow an AR(1) process), Fernandez (residuals follow a random walk), Litterman (residuals corrected with white noise assumption).
- High-frequency variables converted to real terms using monthly CPI as deflator and seasonally adjusted using the X-13 procedure.
- Output gap estimated using Hodrick-Prescott (HP) filter with a lamda of 1600; forecasting period included to address end-point bias.
- Candidate high-frequency series (start years shown in source tables): Exports (Monthly, Start Year 2002), Imports (Monthly, Start Year 2002), Broad Money (Monthly, Start Year 2002), Credit to Private Sector (Monthly, Start Year 2002), Government Expenditure/Revenue (Monthly, Start Year 2002), China GDP (Quarterly, Start Year 2000), US GDP (Quarterly, Start Year 2000), EU GDP (Quarterly, Start Year 2000), Real Effective Exchange Rate (Monthly, Start Year 2000).
- Correlations (year-on-year percent change) highlight six potential candidate variables: broad money (M2), exports, imports, private credit, tourist arrival, foreign reserves, and China GDP.
  - Selected correlations (excerpt): RGDP — M2: 0.1; RGDP — EXP_GMENT: 0.5; RGDP — EXP_RICE: -0.2; RGDP — IMP_GMT: 0.4; RGDP — IMP_OIL: 0.0; RGDP — CREDIT: 0.5; RGDP — RESERVES: 0.1.
- Final baseline interpolation choice after consultation with authorities: quarterly RGDP1 — interpolation with foreign reserves, broad money, and exports.
  - Rationale: key historical drivers included and results consistent with authorities’ real sector insights.
- Non-overlapping interpolation option to avoid overlap with nowcasting predictors: RGDP2 using imports and broad money.
- Rebasing note: newly rebased annual GDP data for 2000-2022 published by NIS is overall about one-third higher than the old GDP data; in terms of growth rate the new GDP data is not much different from the old one.

### Nowcasting methods, specification, and diagnostics
- Nowcasting approaches considered: Bridge equation, MIDAS, U-MIDAS; mixed-frequency factor models, DFMs, and machine learning referenced as alternatives.
- Bridge model: linear regression linking aggregated high-frequency indicators to quarterly GDP growth; lag length chosen by Akaike information criterion; often lag length set to zero for static specification.
- MIDAS and U-MIDAS: MIDAS imposes distributed lag polynomials; U-MIDAS uses skip-sampled variables directly allowing OLS.
- Data preparation: transform to growth rate (quarter on quarter); Augmented Dicky-Fuller tests applied — all variables in growth rate are stationary; seasonal adjustment with X-13 where needed.
- Model selection criteria: RMSE, Akaike Information Criteria, R Squared.
- Nowcasting candidate predictors found consistently positive and significant: exports, tourist arrival, private credit, foreign reserves, and China real GDP.

### Regression and nowcasting model findings (selected statistics preserved)
- Sample period for regressions: 2004Q1–2022Q2; Observations: 74 (all reported regressions).
- Selected regression excerpts (QoQ, Growth rate; standard errors in parentheses):
  - One Lag: 0.148* (0.083) in column (1); 0.241*** (0.069) in column (2)
  - GDP_China: 0.002*** (0.001) in column (1); 0.001** (0.001) in column (2)
  - GDP_US: 0.001 (0.001) in column (1); 0.002 (0.001) in column (2)
  - Private Credit: 0.0510 (0.036) in column (1)
  - Reserves: 0.096*** (0.023) in column (1)
  - Exports: 0.081*** (0.017) in column (1)
  - Tourist: 0.015*** (0.004) in column (1)
- Summary of nowcasting models (Table 5):
  - R Squared: Bridge 0.701; MIDAS 0.73; U-MIDAS 0.68
  - Akaike Information Criteria: Bridge (6.36); MIDAS (6.70); U-MIDAS (6.38)
  - Durbin-Watson Statistics: Bridge 2.19; MIDAS 2.03; U-MIDAS 2.14
  - Variable signs and significance across models summarized (e.g., DLOG(RGDP_China): (+)*** in Bridge and MIDAS; DLOG(Exports): (+)*** in Bridge and U-MIDAS; DLOG(Reserves): (+)*** across Bridge and U-MIDAS; DLOG(Tourist): (+)*** across all three models).

### Nowcasting performance and example nowcasts
- Pseudo out-of-sample forecast evaluation for 2022Q3–Q4: U-MIDAS had the best RMSE performance.
- In-sample checks: models capture the Covid-19 turning point well due to significant drops in tourist arrival and exports.
- Nowcast estimates for 2023 (re-estimated with full sample until 2022Q4):
  - 2023Q1 real GDP growth (year-on-year) estimated to slightly decelerate to 4.3 percent.
  - 2023Q2 real GDP growth (year-on-year) projected to rise slightly to 4.6 percent.
  - 2023Q2 growth driven mostly by rebound in tourist arrival and exports despite slowdown of China’s growth.
- Nowcasting outputs presented as QoQ and YoY forecasts with fan-chart percentiles: 30th-, 60th-, 90th-percentiles; median as 50th-percentile.

### Inflation forecasting (monthly, bottom-up approach)
- Headline CPI decomposed into three components: food CPI, oil-related CPI, and core CPI; each component modeled separately.
- Models applied: ARMA, ARMAX, VAR, VARX for component-level and joint dynamics.
- Key foreign exogenous variables for ARMAX/VARX: global food (or rice) prices, global energy (or oil) prices, global fertilizer prices, and headline inflation of trading partners (China, Thailand, the U.S., Vietnam).
- Correlations (percent change in prices, Table 7 excerpt): CPI_FOOD — CPI_CORE: 0.7; CPI_OIL correlations with P_RICE 0.6, P_OIL 0.6.
- Model diagnostics (Table 8 example Adjusted R-squared values): reported values include 0.49, 0.37, 0.59, 0.58, 0.44, 0.21 across specifications shown.
- Forecast combination: simple average of forecasts from ARMA, ARMAX, VAR, VARX used for each CPI component; aggregated by CPI weights into headline CPI forecast.
- Forecast inputs for foreign exogenous variables can be taken from IMF WEO and World Bank CMO (fertilizer).

### Joint near-term forecasting (quarterly VAR/VARX)
- Two VAR specifications estimated quarterly (2004Q4–2023Q2; two lags selected by Akaike):
  - VAR 1 endogenous: ∆log(GDP), ∆log(CPI)
  - VAR 2 endogenous: ∆log(GDP), ∆log(CPI), ∆log(Credit), ∆log(EXR)
- VARX augmentations include exogenous foreign variables: real GDP growth of U.S., China, EU; global commodity prices; headline inflation of trading partners.
- Key estimated effects (Table 9 summary):
  - DL_RGDP_CN: (+)*** in VARX 1 and VARX 2
  - DL_RGDP_US: (+)*** in VARX 1 and VARX 2
  - DL_P_RICE: (+)** in VARX 1; (+)*** in VARX 2
  - DL_P_ENERGY: (-)* in VARX 1 and VARX 2
  - DL_CPI_VN: (+)*** in VARX 1; (+)** in VARX 2
- Adjusted R-squared examples:
  - VARX 1 (GDP): 0.40; VARX 2 (GDP): 0.37
  - VARX 1 (Inflation): 0.58; VARX 2 (Inflation): 0.72
- Akaike info criterion examples reported: 3.21, 3.23, 3.41, 3.04 across VARX specifications.
- Pseudo out-of-sample evaluation for 2023Q1–Q2: VAR 2 model had the lowest RMSE; model averaging adopted as baseline (simple average of four models).
- Joint NTF baseline mean forecasts and VAR 2 fan charts (30th-, 50th-, 60th-, and 90th-percentiles) produced for 2023Q3–Q4.

### Key near-term forecast outcomes (baseline textual statements)
- Baseline near-term forecasts imply some recovery in both real GDP growth and inflation for H2 2023:
  - y/y growth in H2 2023 would have rebounded to about 6.5 percent, compared to 4.6 percent in the first half of 2023.
  - y/y inflation would have recovered somewhat to 2.9 percent in H2 2023, compared to 1.2 percent in the first half of 2023.
- For the whole year of 2023:
  - real GDP would have grown by about 5½ percent (compared to 5.2 percent in 2022).
  - inflation would have fallen significantly in 2023 to about 2¼ percent (compared to over 5 percent in 2022).
- Baseline forecasts broadly in line with IMF’s 2023 Article IV Staff Report for Cambodia (IMF, 2024) as noted in the source.

### Annex IV findings on newly rebased GDP and model implications
- The new GDP data would be about 1/3 bigger than the existing one.
- Industry and Service sector each contribute about half of the difference; Agriculture largely unchanged overall.
- In terms of growth rates:
  - The new growth rates are slightly higher between 2008 and 2019.
  - The new growth rates are exactly the same between 2020-2022.
- With rebasing, model coefficients and some forecast results change slightly (Table A1 shows Old vs New coefficients and R-squareds).
  - Example R-squared comparisons (Old vs New): 0.858 vs 0.789; 0.781 vs 0.724; 0.831 vs 0.786; 0.798 vs 0.738.
  - Selected coefficient examples (Old vs New, standard errors in parentheses): Reserves 0.109*** (0.017) vs 0.134*** (0.021); Exports 0.080*** (0.012) vs 0.047*** (0.016); Tourist 0.016*** (0.003) vs 0.021*** (0.003).

### Data sources and operational notes (Annex V summary)
- NBC database combines quarterly, monthly, and annual data from domestic and international sources.
- Monthly domestic variables (sources in parentheses):
  - Foreign Reserves — NBC
  - Tourist Arrival — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Broad Money (M2) — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - CPI_headline, CPI_food, CPI_core — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Exchange Rate — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Private Credit — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Exports, Imports — NBC, MEF — Monetary and Financial Statistics Data (nbc.gov.kh)
- Foreign/global monthly and quarterly variables (examples in source): CPI in Thailand — Bank of Thailand; CPI in Vietnam — Haver; Oil price — World Bank, IMF; GDP_US/GDP_China/GDP_EU — Haver, WEO.
- Quarterly/annual forecast inputs: Oil_forecast and Food_forecast from World Bank commodity prices; CPI_US_Forecast and GDP forecasts from IMF World Economic Outlook Database.
- Implementation note: EViews 12/13 built-in interpolation functions used to implement Chow-Lin, Fernandez, and Litterman methods, estimating β and ρ with maximum likelihood and the Kalman Filter.

### Conclusions, operational guidance, and recommended enhancements
- Recommends the four-step integrated framework: interpolate quarterly GDP, nowcast current activity, bottom-up CPI forecasting, and joint VAR/VARX near-term forecasting.
- Operational and policy recommendations:
  - Final selection of indicators and estimation methods should involve sectoral experts at NBC, MEF, and NIS.
  - Regular evaluation and fine-tuning of models as new data arrive; model staff should adjust models and incorporate judgment.
  - Enhance data availability and add potentially important information (PMI, business sentiment index, retail sales, nightlight index, emissions) to improve nowcasts and NTFs.
  - Extend analysis to more computationally demanding multivariate approaches (mixed-frequency VAR, DFMs, principal component analysis) as capacity and data improve.
- Practical note: avoid overlap of variables between interpolation and nowcasting when possible, or explicitly account for overlap via robustness checks (e.g., RGDP2 interpolation with imports and broad money).

*Source: IMF Working Paper — Nowcasting and Near-Term Forecasting Cambodia’s Economy (WP/2024/147), Executive Summary, Appendix 1, Annex I, Annex IV, Annex V, and related sections as provided in the source content.*

### Executive Summary ......................................................................................................

### Executive Summary

### Overview
- Purpose: Present an integrated suite of forecasting models for nowcasting and near-term forecasting (NTF) of Cambodia’s economy, specifically real GDP and inflation.
- Nowcasting: Forecast of the present to assess the current state of the economy.
- Near-term forecasting: Economic outlook for the next couple of quarters (near term in this paper refers to the next two quarters).

### Four-step integrated approach (process summary)
- Step 1 — Interpolate historical quarterly (real) GDP:
  - Estimate historical quarterly GDP using high-frequency indicators closely related to economic activity (monthly exports, credit, foreign reserves, broad money).
  - Methodology referenced: regression-based Chow-Lin disaggregation method; robustness checked with Litterman method.
- Step 2 — Nowcast current-quarter GDP:
  - Use high-frequency data on credit, exports of garment, and tourist arrivals, and quarterly GDP of Cambodia’s largest trading partners (US, China, EU).
  - Nowcasting models: Bridge, MIDAS, U-MIDAS.
- Step 3 — Current-quarter headline CPI inflation forecast (bottom-up):
  - Forecast each key CPI component for the current quarter and aggregate to obtain headline CPI forecasts.
  - Models referenced for monthly inflation forecasting: ARMA, ARMAX, VAR, VARX.
- Step 4 — Joint near-term forecast of GDP and inflation:
  - VAR models (and VARX with exogenous foreign variables) to jointly forecast GDP and inflation for the next two quarters, capturing feedback effects.
  - Models include other macro variables: credit growth and exchange rate.
  - Latest available “historical” data in VARs include the GDP nowcasts and inflation forecasts from Steps 2 and 3.
  - Key foreign exogenous variables: global commodity prices and key trading partners’ GDP growth and inflation.

### Key features of Cambodia’s economy (data points preserved)
- Economic structure and drivers:
  - Garment exports account for about 60 percent of total exports and about 1/3 of growth on average during 2003-2020.
  - Tourism and related services contribute about 15% of economic activities.
  - Since the Covid-19 outbreak, strong growth in non-garment exports (travel goods, electronic components, bicycles).
- Trade partners:
  - Top export destinations: European Union (EU), the United States (US), and China.
  - China: key source of foreign direct investment, input materials for production, and tourism (until Covid-19).
  - Important import partners for food and input materials: Vietnam and Thailand.
- Exchange rate and reserves:
  - Riels have floated within a range of 4000-4200 Riels per US dollar since 2000.
  - Given large share of imports in CPI basket, exchange rate movements can affect inflation, but Riels’ stability has limited exchange-rate-driven inflation so far.
- Inflation drivers:
  - Inflation in Cambodia has been more associated with global commodity and energy prices, or inflation in neighboring trading partners (Vietnam and Thailand) rather than exchange-rate pass-through.
- Financial system and credit:
  - Financial system highly dollarized: more than 80 percent of loans, deposits, and payments denominated in US dollar.
  - Credit growth has been a key monetary indicator; average credit growth was about 20 percent on average between 2000-2022.
  - Credit reached more than 150 percent of GDP in 2023.
- Policy context:
  - High dollarization limits scope and effectiveness of monetary policy; NBC influences credit growth and liquidity via reserve and capital requirements, while macroeconomic management relies more on fiscal policy.
  - Timely nowcasts and near-term forecasts inform macroeconomic surveillance, the National Bank of Cambodia’s Forecasting and Policy Analysis System (FPAS), and budget preparation/negotiation cycles.

### Nowcasting and NTF modeling toolkit (methods and usage)
- Nowcasting models used to estimate current-quarter GDP:
  - Bridge models, Mixed-Data Sampling (MIDAS), Unrestricted MIDAS (U-MIDAS), Dynamic Factor Models (DFMs) mentioned as alternative methods.
- Interpolation of quarterly GDP from annual data:
  - Use correlation between annual GDP and high-frequency indicators, then apply quarterly high-frequency indicators to interpolate quarterly GDP ensuring quarterly sums equal official annual GDP (Chow-Lin approach).
- Joint near-term forecasting:
  - VAR and VARX models to capture feedbacks between GDP and inflation and incorporate credit growth, exchange rate, and exogenous foreign variables (global commodity prices, trading partners’ GDP growth and inflation).
- Monthly inflation forecasting:
  - Bottom-up approach forecasting key CPI components; ARMA, ARMAX, VAR, VARX models applied to component-level and to aggregate headline inflation.

### Contributions and implications
- Demonstrates how relatively high-frequency official indicators can be exploited for real-time economic assessment in countries with limited/lagged quarterly data.
- Shows an integrated, multi-model framework combining interpolation of historical quarterly GDP, nowcasting of current activity, component-level inflation forecasting, and joint near-term forecasting to inform policymaking.
- Emphasizes that, despite lack of innovative big-data sources (e.g., nightlight, emission), developing countries can still leverage monthly and quarterly official data for timely macroeconomic assessment.

_Executive Summary, Nowcasting and Near-Term Forecasting Cambodia’s Economy (IMF Working Paper) — Executive Summary and Introduction sections._

### Appendix 1 provides a summary of the methods.

### wpiea2024147-print-pdf - Appendix 1 provides a summary of the methods

### Interpolation procedure and indicator selection
- Interpolation requires a good selection of high-frequency indicator series and appropriate econometric models (Chow-Lin and Litterman approaches discussed).
- Key methodological notes:
  - Chow-Lin method assumes residuals follow an AR(1) process; Litterman method assumes residuals follow a random walk.
  - High-frequency variables are converted to real terms using monthly CPI as deflator and seasonally adjusted using the X-13 procedure.
  - Output gap estimated using Hodrick-Prescott (HP) filter with a lamda of 1600; to address end-point bias the forecasting period is included in the sample.
- Candidate high-frequency series (Table 1) include:
  - Exports (Monthly, Start Year 2002, Source: NBC/GDCE)
  - Imports (Monthly, Start Year 2002, Source: NBC/GDCE)
  - Broad Money (Monthly, Start Year 2002, Source: NBC)
  - Credit to Private Sector (Monthly, Start Year 2002, Source: NBC)
  - Government Expenditure/Revenue (Monthly, Start Year 2002, Source: MEF)
  - China GDP (Quarterly, Start Year 2000, Source: IMF)
  - US GDP (Quarterly, Start Year 2000, Source: IMF)
  - EU GDP (Quarterly, Start Year 2000, Source: IMF)
  - Real Effective Exchange Rate (Monthly, Start Year 2000, Source: NBC)

### Correlations and variable narrowing (Table 2 summary)
- Correlation matrix (real variables in percent change, yoy) highlights six potential candidate variables: broad money (M2), exports, imports, private credit, tourist arrival, foreign reserves, and China GDP.
- Selected correlation entries (year-on-year percent change):
  - RGDP — M2: 0.1
  - RGDP — EXP_GMENT: 0.5
  - RGDP — EXP_RICE: -0.2
  - RGDP — IMP_GMT: 0.4
  - RGDP — IMP_OIL: 0.0
  - RGDP — CREDIT: 0.5
  - RGDP — RESERVES: 0.1
- Note: headline CPI is used to derive real variables. Variables are in year-on-year percent change.

### Interpolation results and robustness checks
- Individual interpolations based on broad money, credit, and tourist arrival produced similar quarterly RGDP series; broad money could be used alone without materially changing historical estimates.
- Interpolations based on foreign reserves or imports produced similar patterns; exports-based interpolation showed higher volatility but similar pattern.
- Narrowed set of three variables that generate slightly different interpolated RGDP: foreign reserves, broad money, and exports.
  - Average of interpolations based on the three variables (DL4RGDPM1) is similar to interpolation based on foreign reserves alone (DL4RGDP1).
  - Rationale: foreign reserves summarize net effects of exports, imports, investment, and government expenditures.
- To avoid overlap with nowcasting predictors, a non-overlapping interpolation option used two variables: imports and broad money (RGDP2).
- Final baseline choice (after consultation with authorities): quarterly RGDP1 — interpolation with foreign reserves, broad money, and exports — for two reasons:
  - key historical drivers included in interpolation, and
  - results consistent with authorities’ real sector insights.
- Rebasing note: newly rebased annual GDP data for 2000-2022 published by NIS is overall about one-third higher than the old GDP data used in previous interpolations; in terms of growth rate the new GDP data is not much different from the old one. Effects on interpolation and nowcasts discussed in Annex V.

### Nowcasting methods, model specification, and diagnostics
- Nowcasting approaches considered: Bridge equation, MIDAS, U-MIDAS; also referenced mixed-frequency factor models, DFMs, and machine learning (but paper uses simpler univariate methods given NBC capacity).
- Bridge model:
  - Linear regression linking aggregated high-frequency indicators to quarterly GDP growth; lag length chosen based on Akaike information criterion; often lag length set to zero for static specification.
- MIDAS and U-MIDAS:
  - Use split/skip sampling instead of aggregation; MIDAS imposes distributed lag polynomials (reducing parameters) but requires nonlinear estimation; U-MIDAS adds skip-sampled variables directly allowing OLS but can lead to parameter proliferation.
  - U-MIDAS tends to perform better when frequency mismatch is relatively small (monthly to quarterly).
- Data preparation steps:
  - Transform to growth rate (quarter on quarter); Augmented Dicky-Fuller tests applied — all variables in growth rate are stationary.
  - Seasonal adjustment with X-13 where seasonal patterns are observed.
  - Monthly indicators converted to quarterly by summing (flow variables) or averaging (stock/index variables) for Bridge; skip/split sampling used for MIDAS/U-MIDAS.
- Model selection criteria include RMSE, Akaike Information Criteria, and R Squared.

### Regression and nowcasting model findings (Tables 4–5 summary)
- Bridge, MIDAS, and U-MIDAS regressions estimated over 2004Q1–2022Q2 show coefficients of exports, tourist arrival, private credit, foreign reserves, and China real GDP are positive and statistically significant across methods.
- Selected regression coefficients and statistics (excerpt from Table 4; QoQ, Growth rate; standard errors in parentheses):
  - One Lag: 0.148* (0.083) in column (1); 0.241*** (0.069) in column (2)
  - GDP_China: 0.002*** (0.001) in column (1); 0.001** (0.001) in column (2)
  - GDP_US: 0.001 (0.001) in column (1); 0.002 (0.001) in column (2)
  - Private Credit: 0.0510 (0.036) in column (1)
  - Reserves: 0.096*** (0.023) in column (1)
  - Exports: 0.081*** (0.017) in column (1)
  - Tourist: 0.015*** (0.004) in column (1)
  - Observations: 74 (all columns)
  - R-squared: 0.701 (column 1); 0.464 (column 2); 0.452 (column 3); 0.470 (column 4)
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1
- Summary of nowcasting models (Table 5):
  - R Squared: Bridge 0.701; MIDAS 0.73; U-MIDAS 0.68
  - Akaike Information Criteria: Bridge (6.36); MIDAS (6.70); U-MIDAS (6.38)
  - Durbin-Watson Statistics: Bridge 2.19; MIDAS 2.03; U-MIDAS 2.14
  - Variable signs and significance across models:
    - One Period Lag: (+)* in Bridge; (-)*** in MIDAS; (+) in U-MIDAS
    - DLOG(RGDP_China): (+)*** in Bridge; (+)*** in MIDAS; (+)* in U-MIDAS
    - DLOG(Exports): (+)*** in Bridge; (+)** in MIDAS; (+)*** in U-MIDAS
    - DLOG(Reserves): (+)*** in Bridge; (+)** in MIDAS; (+)*** in U-MIDAS
    - DLOG(Tourist): (+)*** across all three models

### Nowcasting performance and example nowcasts (validation and 2023Q1–Q2)
- Pseudo out-of-sample forecast evaluation conducted for 2022Q3–Q4; U-MIDAS had the best RMSE performance.
- In-sample checks show models capture the Covid-19 turning point well due to significant drops in tourist arrival and exports.
- Nowcast estimates for 2023 (re-estimated with full sample until 2022Q4):
  - 2023Q1 real GDP growth (year-on-year) estimated to slightly decelerate to 4.3 percent.
  - 2023Q2 real GDP growth (year-on-year) projected to rise slightly to 4.6 percent.
  - 2023Q2 growth driven mostly by rebound in tourist arrival and exports despite slowdown of China’s growth.
- Nowcasting outputs presented as QoQ and YoY forecasts with fan-chart percentiles (30th-, 60th-, 90th-percentiles; median as 50th-percentile).

### Inflation forecasting models (monthly, bottom-up approach)
- Headline CPI decomposed into three components: food CPI, oil-related CPI, and core CPI; each component modeled separately using ARMA and ARMAX, plus VAR and VARX models for joint dynamics.
- Key foreign exogenous variables for ARMAX/VARX: global food (or rice) prices, global energy (or oil) prices, global fertilizer prices, and headline inflation of trading partners (China, Thailand, the U.S., Vietnam).
- Correlations (percent change in prices, Table 7 excerpt):
  - CPI_FOOD — CPI_CORE: 0.7
  - CPI_OIL — CPI_VN (DL_CPI_VN): 0.4 (table shows CPI_OIL correlations with P_RICE 0.6, P_OIL 0.6)
  - CPI_FOOD highly correlated with Vietnam CPI inflation.
- Model diagnostics (Table 8 summary; Adjusted R-squared and information criteria across component models):
  - Examples: ARMAX and VARX models show DL_P_RICE and DL_P_OIL significant for food and oil components; DL_P_FERTILIZER significant in some specifications.
  - Adjusted R-squared values reported (e.g., 0.49, 0.37, 0.59, 0.58, 0.44, 0.21 across specifications shown).
- Forecast combination: simple average of forecasts from ARMA, ARMAX, VAR, VARX used for each CPI component; aggregated by CPI weights into headline CPI forecast.
- Forecast inputs for foreign exogenous variables can be taken from IMF WEO and World Bank CMO (fertilizer).

### Joint near-term forecasting (NTF) of GDP and inflation (quarterly VAR/VARX)
- Two VAR specifications estimated quarterly (2004Q4–2023Q2; two lags selected by Akaike):
  - VAR 1 endogenous: ∆log(GDP), ∆log(CPI)
  - VAR 2 endogenous: ∆log(GDP), ∆log(CPI), ∆log(Credit), ∆log(EXR)
- VARX versions augment VARs with foreign exogenous variables: real GDP growth of U.S., China, EU; global commodity prices; headline inflation of trading partners.
- Key estimated effects (Table 9 summary):
  - DL_RGDP_CN: (+)*** in VARX 1 and VARX 2
  - DL_RGDP_US: (+)*** in VARX 1 and VARX 2
  - DL_P_RICE: (+)** in VARX 1; (+)*** in VARX 2
  - DL_P_ENERGY: (-)* (negative effect on GDP) in VARX 1 and VARX 2
  - DL_CPI_VN: (+)*** in VARX 1; (+)** in VARX 2
  - Adjusted R-squared examples:
    - VARX 1 (GDP): 0.40; VARX 2 (GDP): 0.37
    - VARX 1 (Inflation): 0.58; VARX 2 (Inflation): 0.72
  - Akaike info criterion examples: 3.21, 3.23, 3.41, 3.04 across VARX specifications.
- Pseudo out-of-sample evaluation for 2023Q1–Q2 indicates VAR 2 model had the lowest RMSE; model averaging adopted as baseline (simple average of four models).
- Joint NTF baseline mean forecasts and VAR 2 fan charts (30th-, 50th-, 60th-, and 90th-percentiles) produced for 2023Q3–Q4.

### Key near-term forecast outcomes (textual baseline statements)
- Baseline near-term forecasts imply some recovery in both real GDP growth and inflation for H2 2023:
  - y/y growth in H2 2023 would have rebounded to about 6.5 percent, compared to 4.6 percent in the first half of 2023.
  - y/y inflation would have recovered somewhat to 2.9 percent in H2 2023, compared to 1.2 percent in the first half of 2023.
  - For the whole year of 2023:
    - real GDP would have grown by about 5½ percent (compared to 5.2 percent in 2022).
    - inflation would have fallen significantly in 2023 to about 2¼ percent (compared to over 5 percent in 2022).
- Baseline forecasts broadly in line with latest forecasts in the IMF’s 2023 Article IV Staff Report for Cambodia (IMF, 2024).

### Conclusions, operational guidance, and recommended enhancements
- Four-step integrated approach proposed:
  1. Interpolate historical quarterly real GDP using high-frequency indicators (exports, credit, foreign reserves, broad money).
  2. Nowcast current economic activity using high-frequency data (credit, export of garment, tourist arrivals) and trading-partner quarterly GDP.
  3. Forecast headline CPI for current quarter using bottom-up component approach (food, oil-related, core).
  4. Jointly forecast GDP and inflation over next couple of quarters using VAR/VARX, incorporating credit growth, exchange rate, nowcasts, and exogenous foreign variables.
- Operational and policy recommendations:
  - Final selection of indicators and estimation methods should involve sectoral experts at NBC, MEF, and NIS to ensure consistency with historical events and sectoral knowledge.
  - Regular evaluation and fine-tuning of models as new data arrive; model staff should adjust models and incorporate judgment.
  - Enhance data availability and add potentially important information (PMI, business sentiment index, retail sales, nightlight index, emissions) to improve nowcasts and NTFs.
  - Extend analysis to more computationally demanding multivariate approaches (mixed-frequency VAR, DFMs, principal component analysis) as capacity and data improve.
- Practical note: interpolation and nowcasting choices should avoid overlap of variables between interpolation and nowcasting when possible, or explicitly account for such overlap via robustness checks (e.g., RGDP2 interpolation with imports and broad money).

*Source: IMF Working Paper — Nowcasting and Near-Term Forecasting Cambodia’s Economy (Appendix 1 and related sections, authors’ estimates).*

### Annex I. Correlation and Interpolation Method

### Annex I. Correlation and Interpolation Method

### Robustness check: annual regression of GDP growth on explanatory variables
- Purpose: Run a regression of GDP growth in the annual series on the proposed explanatory variables in annual series as a robustness check.
- Finding: "We observe that all key variables have expected signs and are statistically significant (except the foreign reserves)."
- Dependent Variable: DLOG(RGDP)
- Method: Least Squares
- Sample (adjusted): 2004 2022
- Included observations: 19 after adjustments
- Estimated coefficients and statistics:
  - DLOG(CREDIT): Coefficient 0.105, Std. Error 0.061, t-Statistic 1.720, Prob. 0.106
  - DLOG(EXPORTS): Coefficient 0.166, Std. Error 0.073, t-Statistic 2.280, Prob. 0.038
  - DLOG(TOURIST): Coefficient 0.016, Std. Error 0.007, t-Statistic 2.179, Prob. 0.046
  - DLOG(RESERVES): Coefficient 0.077, Std. Error 0.063, t-Statistic 1.236, Prob. 0.236
- Model fit and diagnostics:
  - R-squared 0.542581
  - Adjusted R-squared 0.451097
  - Mean dependent var 0.064038
  - S.D. dependent var 0.03503
  - S.E. of regression 0.025953
  - Akaike info criterion -4.280407
  - Sum squared resid 0.010103
  - Schwarz criterion -4.081578
  - Log likelihood 44.66386
  - Hannan-Quinn criter. -4.246757

### Interpolation Method: deriving quarterly GDP from annual GDP using high-frequency indicators
- Objective: Derive Cambodia’s Quarterly GDP data from annual GDP Data using quarterly data of export, credit, and foreign reserves, each having strong correlation with GDP.
- Approach: Use statistical relations between low frequency data (annual GDP) and high frequency indicator variables (quarterly indicators) through regressions. Methods referenced: Chow-Lin (1971), Fernandez (1981), and Litterman (1983). EViews 12/13 built-in interpolation functions implement these methods.
- General Chow-Lin framework:
  - High-frequency representation: y_hf = X β + u
    - y_hf is the high frequency data (e.g., quarterly GDP) and X are selected indicators (such as quarterly export data).
  - Relation to low-frequency (annual) data via aggregation matrix C:
    - Y = C y = C (X β + u) = C X β + C u
    - E(u u′) = V
  - C is an n X 4n matrix that converts 4n quarterly observations into n annual observations.
  - Residual assumption: u_t = ρ u_{t−1} + ε_t (first-order auto-regression).
  - Estimation: β̂ and ŷ̂ can be estimated with generalized least squares (GLS). EViews built-in function estimates β and ρ with maximum likelihood and the Kalman Filter.
- Fernandez (1981) and Litterman (1983) distinctions:
  - Fernandez (1981): similar to Chow-Lin but assumes residuals follow a random walk: u_t = u_{t−1} + ε_t (non-stationary).
  - Litterman (1983): further corrects for serial correlation by assuming ε_t is a white noise process.

### Implementation notes and context
- Indicators used in interpolation and nowcasting models (as shown in subsequent annexes/tables within the source) include: CREDIT, EXPORTS, TOURIST, RESERVES (in logs or seasonally adjusted transformations indicated by variable names such as DLOG(...) and SA suffixes).
- EViews 12/13 provides built-in interpolation functions that implement Chow-Lin, Fernandez, and Litterman methods, estimating parameters via maximum likelihood and the Kalman Filter as described.

*Source: Annex I. Correlation and Interpolation Method — wpiea2024147-print-pdf*

### Annex IV: Newly Rebased GDP and Nowcasting

### Annex IV: Newly Rebased GDP and Nowcasting Models

### Key findings on the new rebased GDP
- The New GDP data would be about 1/3 bigger than the existing one.
- Industry and Service sector each contributes about half of the difference.
- Agriculture remains largely unchanged overall although there are changes in the sub-sectors.
- The rebased data for industry is driven by food/beverage/tobacco, textile, and manufacture of metallic products.
- The service sector data is driven by wholesale and trade, financial activities, and public administration.
- In terms of overall growth rate, the numbers are not much different:
  - The new growth rates are slightly higher between 2008 and 2019.
  - The new growth rates are exactly the same between 2020-2022.
- With the rebasing, the coefficients of the nowcasting/interpolation models are expected to change.
- The Annex reports model results with the new GDP data (Table A1) and shows that forecast results also slightly change (Figures A3 and A4).
- Note on forecast presentation:
  - The different green shades represent the 30th-, 60th-, and 90th-percentiles.
  - The blue line denotes the median (50th-percentile) forecast.

### Model comparison (Table A1): selected coefficients and statistics (Old vs New)
- Dependent variables presented: RGDP1, RGDP_Mean_All, RGDP_Mean_Lm, RGDP_Mean_CL, RGDP2
- Observations: 63 (for all specifications)
- R-squared values:
  - 0.858 (Old)
  - 0.789 (New)
  - 0.781 (Old)
  - 0.724 (New)
  - 0.831 (Old)
  - 0.786 (New)
  - 0.751 (Old)
  - 0.675 (New)
  - 0.798 (Old)
  - 0.738 (New)

- Coefficients (Old vs New) with standard errors in parentheses; statistical significance indicators preserved (*** p<0.01, ** p<0.05, * p<0.1):
  - One Lag:
    - 0.065 (Old) (0.079)
    - 0.051 (New) (0.085)
    - -0.039 (Old) (0.096)
    - -0.043 (New) (0.095)
    - 0.035 (Old) (0.085)
    - 0.020 (New) (0.087)
    - 0.271** (Old) (0.109)
    - 0.196* (New) (0.109)
    - 0.134 (Old) (0.097)
    - 0.093 (New) (0.098)
  - GDP_China:
    - 0.002*** (Old) (0.000)
    - 0.003*** (New) (0.001)
    - 0.002*** (Old) (0.001)
    - 0.003*** (New) (0.001)
    - 0.002*** (Old) (0.000)
    - 0.002*** (New) (0.000)
    - 0.001*** (Old) (0.000)
    - 0.001*** (New) (0.001)
    - 0.001*** (Old) (0.000)
    - 0.002*** (New) (0.000)
  - GDP_US:
    - 0.001 (Old) (0.001)
    - 0.000 (New) (0.001)
    - 0.001 (Old) (0.001)
    - -0.000 (New) (0.001)
    - -0.000 (Old) (0.001)
    - -0.001 (New) (0.001)
    - 0.000 (Old) (0.001)
    - -0.001 (New) (0.001)
    - 0.000 (Old) (0.001)
    - -0.001 (New) (0.001)
  - GDP_EU (-1):
    - 0.001 (Old) (0.000)
    - 0.000 (New) (0.001)
    - 0.001 (Old) (0.001)
    - 0.001 (New) (0.001)
    - 0.001 (Old) (0.000)
    - 0.000 (New) (0.001)
    - 0.000 (Old) (0.000)
    - 0.000 (New) (0.001)
    - 0.000 (Old) (0.000)
    - 0.000 (New) (0.001)
  - Private Credit:
    - 0.085*** (Old) (0.028)
    - 0.063* (New) (0.035)
    - 0.086** (Old) (0.037)
    - 0.065 (New) (0.042)
    - 0.085*** (Old) (0.030)
    - 0.066* (New) (0.034)
    - 0.074** (Old) (0.030)
    - 0.064* (New) (0.033)
    - 0.080*** (Old) (0.029)
    - 0.065* (New) (0.033)
  - Reserves:
    - 0.109*** (Old) (0.017)
    - 0.134*** (New) (0.021)
    - 0.119*** (Old) (0.021)
    - 0.143*** (New) (0.025)
    - 0.108*** (Old) (0.018)
    - 0.135*** (New) (0.020)
    - 0.041** (Old) (0.017)
    - 0.053** (New) (0.020)
    - 0.075*** (Old) (0.017)
    - 0.095*** (New) (0.020)
  - Exports:
    - 0.080*** (Old) (0.012)
    - 0.047*** (New) (0.016)
    - 0.078*** (Old) (0.016)
    - 0.045** (New) (0.018)
    - 0.111*** (Old) (0.013)
    - 0.093*** (New) (0.015)
    - 0.062*** (Old) (0.013)
    - 0.057*** (New) (0.015)
    - 0.087*** (Old) (0.013)
    - 0.075*** (New) (0.015)
  - Tourist:
    - 0.016*** (Old) (0.003)
    - 0.021*** (New) (0.003)
    - 0.017*** (Old) (0.003)
    - 0.022*** (New) (0.004)
    - 0.013*** (Old) (0.003)
    - 0.018*** (New) (0.003)
    - 0.010*** (Old) (0.003)
    - 0.014*** (New) (0.003)
    - 0.012*** (Old) (0.003)
    - 0.016*** (New) (0.003)
  - Crisis:
    - -0.005 (Old) (0.003)
    - -0.004 (New) (0.004)
    - -0.005 (Old) (0.004)
    - -0.004 (New) (0.005)
    - -0.003 (Old) (0.003)
    - -0.002 (New) (0.004)
    - -0.002 (Old) (0.003)
    - -0.002 (New) (0.004)
    - -0.003 (Old) (0.003)
    - -0.002 (New) (0.004)
  - Constant:
    - -0.000 (Old) (0.002)
    - 0.002 (New) (0.003)
    - 0.001 (Old) (0.003)
    - 0.003 (New) (0.003)
    - 0.000 (Old) (0.002)
    - 0.002 (New) (0.003)
    - 0.001 (Old) (0.002)
    - 0.004 (New) (0.003)
    - 0.001 (Old) (0.002)
    - 0.003 (New) (0.003)

- Standard errors are reported in parentheses.
- Significance legend: *** p<0.01, ** p<0.05, * p<0.1

### Data used for nowcasting and near-term forecasting (Annex V)
- Overview:
  - The data include Cambodia’s domestic high-frequency indicators and foreign and global variables such as real GDP and inflation of key trading partners and global commodity prices.
  - These data are combined into an NBC database which consists of quarterly, monthly, and annual data.
  - Domestic high-frequency indicators include credit, broad money, foreign reserves, tourist arrival, exports, exports of garment, consumer price index.
  - Domestic variables are largely collected and updated from the NBC, MEF, and NIS.
  - Foreign and global variables include Fed rates, US GDP, China GDP, EU GDP, Oil price, food price, Thai CPI, Vietnamese CPI, and US CPI.
  - Data on commodities and global environment (commodity prices, international interest rates) come from the World Bank (Pink sheet), IMF, and Fed.

- Monthly Data (variable — source):
  - Foreign Reserves — NBC
  - Tourist Arrival — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Broad Money (M2) — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - CPI_headline — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - CPI_food — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - CPI_core — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Exchange Rate — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - CPI in Thailand — Haver, Bank of Thailand — Bank of Thailand (bot.or.th)
  - CPI in Vietnam — Haver
  - CPI in China — Haver, PBOC
  - Oil price — World Bank, IMF — Commodity Markets (worldbank.org)
  - Food Price — World Bank, IMF — Commodity Markets (worldbank.org)
  - Private Credit — NBC Monetary and Financial Statistics Data (nbc.gov.kh)
  - Exports — NBC, MEF — Monetary and Financial Statistics Data (nbc.gov.kh)
  - Imports — NBC, MEF — Monetary and Financial Statistics Data (nbc.gov.kh)

- Quarterly Data (variable — source):
  - GDP_US — Haver, WEO — IMF World Economic Outlook Database, latest version
  - GDP_China — Haver, WEO — World Economic Outlook Databases (imf.org)
  - GDP_EU — Haver, WEO — World Economic Outlook Databases (imf.org)
  - Oil_forecast — Projected oil price in the future — https://www.worldbank.org/en/research/commodity-markets . Click on “commodity prices” link to obtain the data.
  - CPI_US_Forecast — Projected CPI in US — IMF World Economic Outlook Database, latest version

- Annual Data (variable — source):
  - GDP_US — WEO — World Economic Outlook Databases (imf.org)
  - GDP_KH — MEF, NBC, NIS.  KHM Macroframework or IMF World Economic Outlook Database, latest version
  - Oil_forecast — Projected oil price in the future — https://www.worldbank.org/en/research/commodity-markets . Click on “commodity prices” link to obtain the data.
  - Food_forecast — Projected Food price in the future — https://www.worldbank.org/en/research/commodity-markets . Click on “commodity prices” link to obtain the data.
  - CPI_US_Forecast — Projected CPI in US — IMF World Economic Outlook Database, latest version

*Source: Annex IV and Annex V, "Nowcasting and Near-Term Forecasting Cambodia’s Economy," Working Paper No. WP/2024/147*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024147-print-pdf.pdf_