## Nowcasting Real GDP in Samoa

## Source details

**Canonical URL:** [Nowcasting Real GDP in Samoa](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025092-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025092-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025092-print-pdf.pdf.json)

---

### I. Purpose and context
- Motivations:
  - Samoa releases GDP figures roughly 90 days, or three months, after the end of a reference quarter, generating a need for timelier estimates.
  - Nowcasting provides timely estimates of real GDP using high-frequency indicators to support policy analysis and decision-making.
- Recent capacity building:
  - Following a request by the Central Bank of Samoa (CBS) in 2022, the IMF’s Institute for Capacity Development (ICD) initiated a Macroeconomic Framework technical assistance (TA) program to develop a nowcasting tool using the Forecasting and Policy Analysis System (FPAS).

### II. Stylized facts about Samoa’s economy (key statistics)
- Production side:
  - Services account for 80% of GDP.
  - Manufacturing comprises approximately 10% of GDP.
  - Agriculture & fisheries comprise approximately 10% of GDP.
- Expenditure side:
  - Household consumption accounts for about 80% of GDP.
  - Government consumption contributes around 20% of GDP.
  - Capital formation makes up approximately 40% of GDP.
  - Net exports of goods and services represent about -40% of GDP.
- Vulnerabilities and recent history:
  - Samoa faced negative GDP growth from 2020-2022 following a measles outbreak (November 2019) and the COVID-19 pandemic.
  - After reopening borders in August 2022, economic recovery was driven by tourism, remittances, and elevated public investment.
- Sectoral importance (2023 shares):
  - New Zealand and Australia accounted for approximately 44 percent and 31 percent, respectively, of Samoa's total tourism earnings.
  - New Zealand and Australia shares of remittances in 2023 are around 39 percent and 36 percent, respectively.
  - Remittances are approximately 45 percent of household consumption expenditure while tourism is around 30 percent of GDP.
- Trade:
  - Major import sources: Singapore (notably mineral fuels) and New Zealand (mainly food products).
  - Leading export destinations: American Samoa and New Zealand.

### III. Nowcasting framework: indicator selection and data handling
- Objective:
  - Estimate Samoa’s real GDP for the current and next quarter(s) using readily available high-frequency data.
- Indicator selection process:
  - Began with a broad set of high-frequency variables (monthly and quarterly) listed in Table 1 and reduced via regression analyses and model comparison based on RMSE, AIC, and R-squared.
  - Unit root testing: used the Augmented Dicky-Fuller (ADF) test; all variables in growth rate form were found stationary.
  - Monthly-to-quarterly conversion: aggregated "flow" variables by summation; "stock/index" variables by averaging or taking the last observation of the quarter.
  - Final model uses eight variables plus COVID-19 dummy and seasonality dummies (Table 2).
- High-frequency indicators in the current model (Table 2):
  - rgdp: Real GDP (2000:Q1–2022:Q3)
  - remitt: Total remittances (in millions WST) (1993:M01–2023:M03)
  - for_res_usd: Foreign reserves (in millions USD) (2000:M01–2023:M03)
  - bank_credit_m: Commercial banks’ total credit to private sector and public institutions (in millions WST) (1999:M01–2023:M03)
  - tour_rec: Tourist receipts (in millions WST) (2002:M08–2023:M03)
  - mark_svey_val: Fugalei market survey (index) (1993:M01–2023:M03)
  - imp: Total merchandise imports (in millions WST) (2000:M01–2023:M03)
  - total_tax_rev: Total tax revenue (in millions WST) (2008:M07–2023:M03)
  - covid: COVID-19 dummy (2008:Q1–2023:Q1)
  - @seas: Seasonality dummies (2008:Q1–2023:Q1)
  - nino_3: Southern Oscillation Index Niño 3 (1950:M01–2023:M03)
- Correlations with rgdp (Table 3; selected values):
  - remitt: 0.29
  - for_res_usd: 0.18
  - bank_credit_m: 0.60
  - tour_rec: 0.50
  - mark_svey_val: 0.73
  - imp: 0.48
  - total_tax_rev: 0.60
  - nino_3: 0.25

### IV. Nowcasting methodologies evaluated
- Methods compared:
  - Bridge equations: linear regressions that aggregate high-frequency variables into low-frequency counterparts and use OLS; forecast missing monthly observations (if any) via AR/ARMA before aggregation.
  - MIDAS (Mixed Data Sampling): allows different sampling frequencies in single regressions while controlling parameter proliferation by imposing functional forms (polynomial weighting); OLS not applicable due to nonlinearity of parameterization.
  - U-MIDAS (Unrestricted MIDAS): integrates high-frequency skip/split-sampled variables without predefining parameter shapes; allows OLS estimation and flexible lag structures.
- Variable inclusion note:
  - For tour_rec variable, first difference was applied given presence of 0 values during COVID-19 period.
  - Given nowcasting context, variables with significance level more than 15 percent were sometimes retained if signs were economically plausible.

### V. Model estimation results (summary from Table 4)
- Dependent variable: dlog(rgdp)
- Selected variable signs and significance (Bridge / MIDAS / U-MIDAS summary):
  - dlog(remitt): (−) in Bridge column (noted as included with negative sign)
  - dlog(for_res_usd(-1)): (+) in Bridge column
  - dlog(bank_credit_m(-1)): (+)* in Bridge column (* indicates 10 percent)
  - d(tour_rec): (+)* in Bridge column
  - dlog(mark_svey_val): (+)*** in Bridge column (*** indicates 1 percent)
  - dlog(imp(-2)): (+) in Bridge column
  - dlog(total_tax_rev): (+)*** in Bridge column
  - nino_3: (+) in Bridge column
  - covid: (−)* in Bridge column
  - @seas(1): (+) in Bridge column
  - @seas(2): (−)*** in Bridge column
  - @seas(3): (+)*** in Bridge column
- Model fit statistics (Table 4):
  - Bridge: R-squared 0.73; Akaike Information Criteria -3.82; Durbin-Watson Statistics 2.11
  - MIDAS: R-squared 0.80; Akaike Information Criteria -3.6; Durbin-Watson Statistics 2.07
  - U-MIDAS: R-squared 0.82; Akaike Information Criteria -3.43; Durbin-Watson Statistics 2.10

### VI. Forecast performance and evaluation (Table 5 and Figure 11 summary)
- Forecast evaluation metrics: RMSE and Theil U2 (lower is better)
- Full Sample (2008Q4-2023Q1):
  - RGDP_F_FC (Forecast Combination): RMSE 8.57; Theil U2 0.36
  - RGDP_F_BRIDGE: RMSE 12.46; Theil U2 0.52
  - RGDP_F_MIDAS: RMSE 10.26; Theil U2 0.44
  - RGDP_F_UMIDAS: RMSE 10.02; Theil U2 0.44
- Pre-COVID Sample (2008Q4-2019Q4):
  - RGDP_F_FC: RMSE 9.03; Theil U2 0.43
  - RGDP_F_BRIDGE: RMSE 12.29; Theil U2 0.57
  - RGDP_F_MIDAS: RMSE 11.04; Theil U2 0.53
  - RGDP_F_UMIDAS: RMSE 11.13; Theil U2 0.54
- COVID/Post-COVID Sample (2020Q1-2023Q1):
  - RGDP_F_FC: RMSE 5.96; Theil U2 0.19
  - RGDP_F_BRIDGE: RMSE 11.04; Theil U2 0.36
  - RGDP_F_MIDAS: RMSE 6.49; Theil U2 0.21
  - RGDP_F_UMIDAS: RMSE 4.41; Theil U2 0.13
- Key performance conclusions:
  - U-MIDAS performed best in RMSE terms for the full-sample and during/post-COVID sample.
  - Forecast combination (FC) of all three methods improved accuracy for the full sample and pre-COVID sample.
  - Nowcasts from the three models outperform a naïve forecast (Theil U2 < 1).

### VII. Practical implementation and use
- The developed nowcasting tool estimates Samoa’s real GDP up to three months before official release and has been integrated into CBS' policy analysis routine.
- Results are summarized in a one-page document showing the impact of indicators on projected real GDP growth, to support an up-to-date economic narrative for policymakers.
- Nowcasting process documentation and back-testing were conducted and repeated iteratively to improve model robustness.

### VIII. Conclusions and recommended next steps
- Findings:
  - High-frequency indicators—remittances, foreign reserves, bank credit, tourist receipts, agriculture market surveys, imports, total tax revenue, and an El-Niño indicator—are useful to nowcast Samoa’s quarterly real GDP.
  - Readily available, relatively high-frequency data can meaningfully improve timeliness of macroeconomic assessment in LICs with delayed data.
- Recommendations and potential enhancements:
  - Continuously refine the models as new high-frequency indicators and data become available.
  - Consider incorporating new data types: business sentiment indices, retail sales, timely fiscal expenditure and revenue data.
  - Explore more sophisticated techniques: principal component analysis, mixed-frequency Vector Autoregression (VAR), Dynamic Factor Models (DFMs).
  - Maintain regular updates to nowcasts and combine model outputs with CBS sectoral expert judgment to produce the official growth forecast.
- Operational note:
  - Forecasting is not purely mechanical; nowcasts should be combined with CBS expert judgment for policy use.

*Source: IMF Working Paper — Nowcasting Real GDP in Samoa (References and chapter content as provided).*

### References .............................................................................................................

### Nowcasting Real GDP in Samoa

### I. Purpose and context
- Motivations:
  - Samoa releases GDP figures roughly 90 days, or three months, after the end of a reference quarter, generating a need for timelier estimates.
  - Nowcasting provides timely estimates of real GDP using high-frequency indicators to support policy analysis and decision-making.
- Recent capacity building:
  - Following a request by the Central Bank of Samoa (CBS) in 2022, the IMF’s Institute for Capacity Development (ICD) initiated a Macroeconomic Framework technical assistance (TA) program to develop a nowcasting tool using the Forecasting and Policy Analysis System (FPAS).

### II. Stylized facts about Samoa’s economy (key statistics)
- Production side:
  - Services account for 80% of GDP.
  - Manufacturing comprises approximately 10% of GDP.
  - Agriculture & fisheries comprise approximately 10% of GDP.
- Expenditure side:
  - Household consumption accounts for about 80% of GDP.
  - Government consumption contributes around 20% of GDP.
  - Capital formation makes up approximately 40% of GDP.
  - Net exports of goods and services represent about -40% of GDP.
- Vulnerabilities and recent history:
  - Samoa faced negative GDP growth from 2020-2022 following a measles outbreak (November 2019) and the COVID-19 pandemic.
  - After reopening borders in August 2022, economic recovery was driven by tourism, remittances, and elevated public investment.
- Sectoral importance (2023 shares):
  - New Zealand and Australia accounted for approximately 44 percent and 31 percent, respectively, of Samoa's total tourism earnings.
  - New Zealand and Australia shares of remittances in 2023 are around 39 percent and 36 percent, respectively.
  - Remittances are approximately 45 percent of household consumption expenditure while tourism is around 30 percent of GDP.
- Trade:
  - Major import sources: Singapore (notably mineral fuels) and New Zealand (mainly food products).
  - Leading export destinations: American Samoa and New Zealand.

### III. Nowcasting framework: indicator selection and data handling
- Objective:
  - Estimate Samoa’s real GDP for the current and next quarter(s) using readily available high-frequency data.
- Indicator selection process:
  - Began with a broad set of high-frequency variables (monthly and quarterly) listed in Table 1 and reduced via regression analyses and model comparison based on RMSE, AIC, and R-squared.
  - Unit root testing: used the Augmented Dicky-Fuller (ADF) test; all variables in growth rate form were found stationary.
  - Monthly-to-quarterly conversion: aggregated "flow" variables by summation; "stock/index" variables by averaging or taking the last observation of the quarter.
  - Final model uses eight variables plus COVID-19 dummy and seasonality dummies (Table 2).
- High-frequency indicators in the current model (Table 2):
  - rgdp: Real GDP (2000:Q1–2022:Q3)
  - remitt: Total remittances (in millions WST) (1993:M01–2023:M03)
  - for_res_usd: Foreign reserves (in millions USD) (2000:M01–2023:M03)
  - bank_credit_m: Commercial banks’ total credit to private sector and public institutions (in millions WST) (1999:M01–2023:M03)
  - tour_rec: Tourist receipts (in millions WST) (2002:M08–2023:M03)
  - mark_svey_val: Fugalei market survey (index) (1993:M01–2023:M03)
  - imp: Total merchandise imports (in millions WST) (2000:M01–2023:M03)
  - total_tax_rev: Total tax revenue (in millions WST) (2008:M07–2023:M03)
  - covid: COVID-19 dummy (2008:Q1–2023:Q1)
  - @seas: Seasonality dummies (2008:Q1–2023:Q1)
  - nino_3: Southern Oscillation Index Niño 3 (1950:M01–2023:M03)
- Correlations with rgdp (Table 3; selected values):
  - remitt: 0.29
  - for_res_usd: 0.18
  - bank_credit_m: 0.60
  - tour_rec: 0.50
  - mark_svey_val: 0.73
  - imp: 0.48
  - total_tax_rev: 0.60
  - nino_3: 0.25

### IV. Nowcasting methodologies evaluated
- Methods compared:
  - Bridge equations: linear regressions that aggregate high-frequency variables into low-frequency counterparts and use OLS; forecast missing monthly observations (if any) via AR/ARMA before aggregation.
  - MIDAS (Mixed Data Sampling): allows different sampling frequencies in single regressions while controlling parameter proliferation by imposing functional forms (polynomial weighting); OLS not applicable due to nonlinearity of parameterization.
  - U-MIDAS (Unrestricted MIDAS): integrates high-frequency skip/split-sampled variables without predefining parameter shapes; allows OLS estimation and flexible lag structures.
- Variable inclusion note:
  - For tour_rec variable, first difference was applied given presence of 0 values during COVID-19 period.
  - Given nowcasting context, variables with significance level more than 15 percent were sometimes retained if signs were economically plausible.

### V. Model estimation results (summary from Table 4)
- Dependent variable: dlog(rgdp)
- Selected variable signs and significance (Bridge / MIDAS / U-MIDAS summary):
  - dlog(remitt): (−) in Bridge column (noted as included with negative sign)
  - dlog(for_res_usd(-1)): (+) in Bridge column
  - dlog(bank_credit_m(-1)): (+)* in Bridge column (* indicates 10 percent)
  - d(tour_rec): (+)* in Bridge column
  - dlog(mark_svey_val): (+)*** in Bridge column (*** indicates 1 percent)
  - dlog(imp(-2)): (+) in Bridge column
  - dlog(total_tax_rev): (+)*** in Bridge column
  - nino_3: (+) in Bridge column
  - covid: (−)* in Bridge column
  - @seas(1): (+) in Bridge column
  - @seas(2): (−)*** in Bridge column
  - @seas(3): (+)*** in Bridge column
- Model fit statistics (Table 4):
  - Bridge: R-squared 0.73; Akaike Information Criteria -3.82; Durbin-Watson Statistics 2.11
  - MIDAS: R-squared 0.80; Akaike Information Criteria -3.6; Durbin-Watson Statistics 2.07
  - U-MIDAS: R-squared 0.82; Akaike Information Criteria -3.43; Durbin-Watson Statistics 2.10

### VI. Forecast performance and evaluation (Table 5 and Figure 11 summary)
- Forecast evaluation metrics: RMSE and Theil U2 (lower is better)
- Full Sample (2008Q4-2023Q1):
  - RGDP_F_FC (Forecast Combination): RMSE 8.57; Theil U2 0.36
  - RGDP_F_BRIDGE: RMSE 12.46; Theil U2 0.52
  - RGDP_F_MIDAS: RMSE 10.26; Theil U2 0.44
  - RGDP_F_UMIDAS: RMSE 10.02; Theil U2 0.44
- Pre-COVID Sample (2008Q4-2019Q4):
  - RGDP_F_FC: RMSE 9.03; Theil U2 0.43
  - RGDP_F_BRIDGE: RMSE 12.29; Theil U2 0.57
  - RGDP_F_MIDAS: RMSE 11.04; Theil U2 0.53
  - RGDP_F_UMIDAS: RMSE 11.13; Theil U2 0.54
- COVID/Post-COVID Sample (2020Q1-2023Q1):
  - RGDP_F_FC: RMSE 5.96; Theil U2 0.19
  - RGDP_F_BRIDGE: RMSE 11.04; Theil U2 0.36
  - RGDP_F_MIDAS: RMSE 6.49; Theil U2 0.21
  - RGDP_F_UMIDAS: RMSE 4.41; Theil U2 0.13
- Key performance conclusions:
  - U-MIDAS performed best in RMSE terms for the full-sample and during/post-COVID sample.
  - Forecast combination (FC) of all three methods improved accuracy for the full sample and pre-COVID sample.
  - Nowcasts from the three models outperform a naïve forecast (Theil U2 < 1).

### VII. Practical implementation and use
- The developed nowcasting tool estimates Samoa’s real GDP up to three months before official release and has been integrated into CBS' policy analysis routine.
- Results are summarized in a one-page document showing the impact of indicators on projected real GDP growth, to support an up-to-date economic narrative for policymakers.
- Nowcasting process documentation and back-testing were conducted and repeated iteratively to improve model robustness.

### VIII. Conclusions and recommended next steps
- Findings:
  - High-frequency indicators—remittances, foreign reserves, bank credit, tourist receipts, agriculture market surveys, imports, total tax revenue, and an El-Niño indicator—are useful to nowcast Samoa’s quarterly real GDP.
  - Readily available, relatively high-frequency data can meaningfully improve timeliness of macroeconomic assessment in LICs with delayed data.
- Recommendations and potential enhancements:
  - Continuously refine the models as new high-frequency indicators and data become available.
  - Consider incorporating new data types: business sentiment indices, retail sales, timely fiscal expenditure and revenue data.
  - Explore more sophisticated techniques: principal component analysis, mixed-frequency Vector Autoregression (VAR), Dynamic Factor Models (DFMs).
  - Maintain regular updates to nowcasts and combine model outputs with CBS sectoral expert judgment to produce the official growth forecast.
- Operational note:
  - Forecasting is not purely mechanical; nowcasts should be combined with CBS expert judgment for policy use.

*Source: IMF Working Paper — Nowcasting Real GDP in Samoa (References and chapter content as provided).*

### References

### wpiea2025092-print-pdf - References

### Cited academic and methodological literature
- Baffigi, A., Golinelli, R., & Parigi, G. (2004). Bridge models to forecast the euro area GDP. International Journal of Forecasting, 20(3): 447-460.
- Ghysels, E., Santa-Clara, P., & Valkanov, R. (2004). Predicting Volatility: Getting the Most Out of Return Data Sampled at Different Frequencies. Journal of Econometrics, 131(1-2), 59-95.
- Ghysels, E., & Wright, J. H. (2009). Forecasting Professional Forecasters. Journal of Business & Economic Statistics, 27(4), 504-516.
- Cashin, P., Raissi, M., and Mohaddes, K. (2017). Fair Weather or Foul? The Macroeconomic Effects of El Niño. Journal of International Economics, Vol. 106, pp. 37-54, 2017.
- Xie, J. (2023). Identifying Optimal Indicators and Lag Terms for Nowcasting Models. IMF Working Paper WP/23/45.

### IMF working papers, national reports, and forthcoming items
- Cashin, P., Han, F., Sabuga, I., Xie, J. & Zhang, F. (2025). Parameter Proliferation in Nowcasting: Issue and Approaches – An Application to Nowcasting China’s Real GDP. IMF Working Paper (Forthcoming).
- Ouliaris, S., Rochon, C. & Taumoepeau, D. (2025). A Quarterly Projection Model for Tonga. IMF Working Paper (Forthcoming).
- Heng, D., Han, F., Chey, S., Uch, R., Kuchsa, D., & Phork, P. (2024). Nowcasting and Near-Term Forecasting Cambodia’s Economy. IMF Working Paper WP/24/147.
- International Monetary Fund (2023). Samoa Article IV Staff Report.
- ______________________  (2024). Samoa Article IV Staff Report.
- Central Bank of Samoa (2023). Annual Report.

*Nowcasting   Real   GDP   in   Samoa  
Working Paper No. WP/2025/092*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025092-print-pdf.pdf_