## sipea2025116

## Source details

**Canonical URL:** [sipea2025116](https://www.imf.org/-/media/files/publications/selected-issues-papers/2025/english/sipea2025116.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/selected-issues-papers/2025/english/sipea2025116.pdf.md)
- [Structured JSON version](/-/media/files/publications/selected-issues-papers/2025/english/sipea2025116.pdf.json)

---

### Inflation developments and context
- Paper objective: examines drivers of inflation in São Tomé and Príncipe and develops a forecasting model using a Phillips Curve approach within an Autoregressive Distributed Lag framework.
- Principal drivers identified:
  - Supply-side: inflation inertia, import food prices, global or regional crises.
  - Demand-side: weakened macroeconomic discipline in the lead-up to elections.
- Emphasis: strengthening inflation forecasting tools is crucial for informing policy responses to reduce and stabilize inflation.

### Recent inflation trajectory and exchange rate regimes
- Exchange rate regimes:
  - Flexible exchange rate until 2009.
  - Peg to the euro from 2010 onward.
- Inflation performance by regime:
  - 16.4 percent (average inflation, 1999–2009) under the flexible regime.
  - 8.4 percent (average inflation, 2010–2020) under the peg.
  - 6.5 percent (standard deviation, flexible regime).
  - 2.5 percent (standard deviation, pegged regime).
- Major recent movements:
  - 16.1 percent (inflation in 2009) → 3.9 percent (inflation in 2015).
  - 8 percent (average annual food-driven inflation, 2017–2021).
  - Over 25 percent (inflation by 2023) driven by global inflationary shocks affecting imported and then locally produced food.
  - Core inflation: 2 percent → 15.9 percent (core inflation rise from 2021 to end-2023).
  - December 2024: 11.6 percent (headline inflation, y/y) and 3.0 percent (core inflation), attributed to unwinding food-price increases in 2023–2024, fiscal consolidation, and monetary tightening.

### CPI composition and sensitivity
- CPI basket (2015 update): 12 subindexes, 79 categories.
- Food share: 70 percent (food share of CPI).
  - Nearly half of food CPI items are locally produced.
- Fuel and electricity share: 6 percent (fuel and electricity share of CPI).
- Core inflation share (excluding food, fuel, electricity): 24 percent (core inflation share).
- Implication: food-dominated CPI makes overall inflation highly sensitive to imported food-price shocks and domestic agricultural developments; fuel/electricity adjustments can have substantial second-round effects.

### VAT impact (Box 1 — empirical summaries)
- VAT structure:
  - Main rate: 15 percent.
  - Reduced rate: 7.5 percent for certain products and services.
  - Some items fully exempt.
- Mechanical estimate: complete and immediate coverage would raise CPI by 5 percentage points in June 2023.
- Difference-in-differences (DiD) with Cabo Verde:
  - DiD indicates a 2.7 percentage point inflation increase in June 2023.
  - Assuming broad VAT coverage by end-2023, DiD for June–December 2023 indicates an impact of 4.9 percentage points on inflation in 2023.
- Cross-country Panel Least Squares (11 Sub-Saharan African countries, excluding São Tomé and Príncipe):
  - Dummy VAT implementation: Coef. 3.17, P value 0.00
  - Lagged inflation: Coef. 0.52, P value 0.00
  - c Constant: Coef. 1.9, P value 0.00
  - R2: 0.40
  - Adjusted R2: 0.33
  - Observations: 362
- Caution: the 4.9 percentage point estimate for São Tomé and Príncipe should be interpreted with caution due to differing domestic policy responses not accounted for in the model.

### REER/NEER developments and passthrough (Box 2)
- 2010–2023: REER experienced substantial appreciation, indicating a decline in competitiveness tied to persistent high inflation under the peg.
- NEER rose by 20 percent during 2010–2023, largely driven by the Euro's strengthening against the US dollar.
- NEER passthrough (panel of seven micro-islands, fixed exchange rate regimes, 2001–2023):
  - Estimated passthrough: approximately -0.33.
    - Interpretation: a 10 percent appreciation/depreciation of the NEER is associated with a decline/increase in annual inflation by 3.3 percentage points.
  - Empirical highlights:
    - Lagged NEER coefficient negative but not statistically significant → passthrough occurs within the same year.
    - Inflation inertia and global crises effects significantly contribute to inflation.
  - Panel model results:
    - Inflation (t-1): Coef. 0.24, P value 0.00
    - NEER (t): Coef. -0.33, P value 0.00
    - NEER (t-1): Coef. -0.20, P value 0.10
    - Global crises: Coef. 2.10, P value 0.02
    - Constant: Coef. 1.63, P value 0.00
    - R2: 0.36
    - Adjusted R2: 0.31
    - Observations: 140
    - Number of countries: 7
    - Country fixed effects: yes

### Methodology and data
- Model framework: Phillips Curve approach within an Autoregressive Distributed Lag Model.
- Annual data: 1998 to 2023.
  - Variables: end-of-period inflation and real GDP growth in São Tomé and Príncipe; end-of-period food inflation in Portugal; changes in fuel prices in São Tomé and Príncipe.
- Dummy variables:
  - D1 = elections in São Tomé and Príncipe.
  - D2 = structural breaks associated with global and regional crises.
  - D3 = transition to a pegged exchange rate regime.
- Pre-estimation testing:
  - Stationarity: Augmented Dickey-Fuller and Phillips-Perron tests.
  - Lag selection: Akaike Information Criterion (AIC).
- Attempted additional variables (found unsatisfactory or statistically insignificant): government expenditure, nominal effective exchange rate, international reserves, BCSTP credit to the government.
- Estimated equation (notation preserved in source):
  - 휋휋푖 = 푐푐 + 훼훼휋휋푖−1 + 훽훽1 푖푖푖푖푖푖.푖푖푖푖푖푖푖푖 + 훽훽2 푖푖푖푖푖푖.푖푖푖푖푖푖 푖−1 + 훿훿 푐푐푐푐푐푐푐푐푐푐 푖−1 + 휃휃 푖푖 푓푓푐푐 푐푐 푖 + 휑휑 퐷퐷1 푖 + 훾훾 퐷퐷2 푖 + 휔휔 퐷퐷3 푖 + μ푖

### Estimation results — key coefficients and significance
- 휋휋푖−1 (Lagged end-of-period inflation): Coeff.: 0.485, P value: 0.0001
  - Interpretation: A one-percentage point increase in the lagged end-of-period inflation results in a 0.49-percentage point rise in end-of-period inflation.
- Portugal food inflation (end-of-period): Coeff.: 0.761, P value: 0.0005
  - Interpretation: A one-percentage point increase in Portugal's food inflation correlates with a 0.76-percentage point rise in São Tomé and Príncipe inflation.
- Lagged Portugal food inflation: Coeff.: 0.167, P value: 0.2433 (not statistically significant).
- Lagged Real GDP growth: Coeff.: 0.454, P value: 0.0490
  - Interpretation: A one-percentage point increase in lagged GDP growth results in a 0.45-percentage point rise in inflation.
- STP fuel price change (fuel_t): Coeff.: 0.063, P value: 1.1766 (not statistically significant; retained to account for supply-side shocks).
- Dummy – elections (D1): Coeff.: 2.365, P value: 0.0480
  - Interpretation: Elections lead to a 2.37-percentage point rise in end-of-period inflation.
- Dummy – global crises (D2): Coeff.: 3.172, P value: 0.0317
  - Interpretation: A global crisis would result in an increase of 3.17 percentage points in end-of-period inflation.
- Dummy – pegged regime (D3): Coeff.: -4.014, P value: 0.0176
  - Interpretation: Implementation of the pegged regime contributed to a 4.01-percentage point slowdown in end-of-period inflation.
- Constant (c): Coeff.: 3.766, P value: 0.1125
- Model fit and sample:
  - R2: 0.964835
  - Adjusted R2: 0.941392
  - Observations: 21

### Residual diagnostics and multicollinearity
- Residual diagnostics (Table 2) — conclusions:
  - No serial correlation, homoskedastic errors, and residuals normally distributed.
  - Breusch–Godfrey Serial Correlation LM Test:
    - F-statistic Prob. F(2,10): 0.8262
    - Obs*R-squared Prob. Chi-Square(2): 0.6747
  - Heteroskedasticity Test: Breusch-Pagan-Godfrey:
    - F-statistic Prob. F(8,12): 0.4022
    - Obs*R-squared Prob. Chi-Square(8): 0.3351
    - Scaled explained SS Prob. Chi-Square(8): 0.9699
  - Normality Test: Jarque-Bera:
    - Jarque-Bera: 0.331581
    - Probability: 0.847224
- Multicollinearity (selected VIFs):
  - 휋휋푖−1: Uncentered VIF 11.88, Centered VIF 2.68
  - 푖푖푖푖푖푖.푖푖푖푖푖푖: Uncentered VIF 3.58, Centered VIF 2.58
  - 푖푖푖푖푖푖.푖푖푖푖푖푖 푖−1: Uncentered VIF 2.59, Centered VIF 1.87
  - 퐺퐺퐷퐷 퐺퐺 푖−1: Uncentered VIF 7.08, Centered VIF 1.50
  - ICRISIS: Uncentered VIF 2.29, Centered VIF 1.85
  - ELECTION: Uncentered VIF 1.94, Centered VIF 1.47
  - PEG: Uncentered VIF 10.00, Centered VIF 3.33
  - FUEL: Uncentered VIF 3.60, Centered VIF 2.20
  - C: Uncentered VIF 34.06, Centered VIF NA
- Interpretation: VIF tests indicate relatively low correlation among independent variables after centering.

### Forecasting performance and baseline projections
- Pseudo out-of-sample performance (Forecast sample: 1998-2029; Adjusted sample: 2003-2029; Included observations: 27):
  - Root Mean Squared Error: 1.538
  - Mean Absolute Error: 1.258
  - Mean Absolute Percentage Error: 14.810
  - Theil Inequality Coef.: 0.052
    - Bias Proportion: 0.000
    - Variance Proportion: 0.094
    - Covariance Proportion: 0.905
  - Theil U2 Coefficient: 0.405
  - Symmetric MAPE: 13.631
- Model forecast (assuming other factors remain constant):
  - 7.7 percent (end-of-period inflation expected to decline to 2025).
  - 9.1 percent (anticipated to rise in 2026 due to national elections).
  - 6.5 percent (gradually declining by 2029).
- Model evaluation: reasonably good pseudo out-of-sample performance over the past 15 years and low bias and variance per Theil measures.

### Main empirical findings (summary)
- Significant drivers:
  - Inflation inertia (past inflation) is a significant driver (Coeff. 0.485, P value 0.0001).
  - Imported food prices (Portugal food inflation) show strong pass-through (Coeff. 0.761, P value 0.0005).
  - Elections associated with a 2.37-percentage point increase in inflation (D1 Coeff. 2.365, P value 0.0480).
  - Global crises associated with a 3.17-percentage point increase in inflation (D2 Coeff. 3.172, P value 0.0317).
  - Pegged regime associated with a 4.01-percentage point reduction in inflation (D3 Coeff. -4.014, P value 0.0176).
  - Lagged GDP growth contributes to inflation (Coeff. 0.454, P value 0.0490).
  - Domestic fuel price changes exhibited expected sign but were not statistically significant (Coeff. 0.063, P value 1.1766).

### Policy recommendations
- Demand-side:
  - Maintain coherent macroeconomic policies, including ongoing fiscal consolidation and managing public debt levels.
  - Implement prudent monetary policy and liquidity management; consider raising interest rates to alleviate local demand pressures.
- Supply-side:
  - Increase local food production to reduce reliance on imported food:
    - Improve processing of primary agricultural products to add value and reduce waste.
    - Develop infrastructure such as greenhouses and agricultural facilities for year-round production and improved yields.
    - Promote sustainable agricultural practices to increase food production and enhance resilience against climate change.

_Italic: Source — Chapter content from sipea2025116 (PDF chapter/section)._

### 1. Inflation Developments _______________________________________________________________ 5

### 1. Inflation Developments

### Key findings and context
- The paper examines drivers of inflation in São Tomé and Príncipe and develops a forecasting model using a Phillips Curve approach within an Autoregressive Distributed Lag framework.
- Both supply and demand factors contribute to inflation:
  - Supply-side: inflation inertia, import food prices, global or regional crises.
  - Demand-side: weakened macroeconomic discipline in the lead-up to elections.
- Strengthening inflation forecasting tools is emphasized as crucial for informing policy responses to reduce and stabilize inflation.

### Recent inflation trajectory and exchange rate context
- Transition of exchange rate regimes:
  - Flexible exchange rate until 2009.
  - Peg to the euro from 2010 onward.
- Inflation performance by regime:
  - Average inflation 16.4 percent (1999–2009) under the flexible regime.
  - Average inflation 8.4 percent (2010–2020) under the peg.
  - Volatility (standard deviation) of inflation: 6.5 percent (flexible regime) and 2.5 percent (pegged regime).
- Major recent developments:
  - Inflation fell from 16.1 percent in 2009 to approximately 3.9 percent in 2015.
  - Inflation averaged about 8 percent annually from 2017 to 2021 (driven primarily by rising food prices).
  - By 2023, inflation surged to over 25 percent due to global inflationary shocks affecting imported and then locally produced food.
  - Core inflation rose from around 2 percent in 2021 to 15.9 percent by the end of 2023.
  - By December 2024, inflation fell to 11.6 percent year-over-year and core inflation decreased to 3.0 percent, attributed to unwinding food-price increases in 2023–2024, fiscal consolidation, and monetary tightening.

### CPI composition and sensitivity
- CPI basket (2015 update): 12 subindexes, 79 categories.
- Food represents approximately 70 percent of the CPI basket.
  - Nearly half of food CPI items are locally produced.
- Fuel and electricity represent around 6 percent of the CPI basket.
- Core inflation (excluding food, fuel, and electricity) represents about 24 percent of the CPI basket.
- Implication: food-dominated CPI makes overall inflation highly sensitive to both imported food-price shocks and domestic agricultural developments; fuel/electricity adjustments can have substantial second-round effects.

### Organization of the analysis
- Section layout of the paper: Section B (inflation structure and developments), Section C (methodology and data), Section D (model estimation results), Section E (model evaluation and forecasts), Section F (policy recommendations).

### Notable quantitative mentions (preserved exactly)
- 70 percent (food share of CPI).
- 6 percent (fuel and electricity share of CPI).
- 24 percent (core inflation share).
- 16.4 percent (average inflation, 1999–2009).
- 8.4 percent (average inflation, 2010–2020).
- 6.5 percent (standard deviation, flexible regime).
- 2.5 percent (standard deviation, pegged regime).
- 16.1 percent (inflation in 2009).
- 3.9 percent (inflation in 2015).
- 8 percent (average annual food-driven inflation, 2017–2021).
- Over 25 percent (inflation by 2023).
- 2 percent → 15.9 percent (core inflation rise from 2021 to end-2023).
- 11.6 percent (headline inflation, December 2024, y/y).
- 3.0 percent (core inflation, December 2024).
- 5 percentage points (estimated VAT contribution to price increases in 2023, noted in Section B).

---

### Box 1 — Estimating the VAT Impact on Inflation (summary)
- VAT introduction details:
  - Main rate: 15 percent.
  - Reduced rate: 7.5 percent for certain products and services.
  - Some items fully exempt.
- Mechanical estimate: if coverage were complete and impact immediate, CPI would have risen by 5 percentage points in June 2023.
- Difference-in-differences (DiD) approach using Cabo Verde as comparator:
  - The DiD estimate indicates the introduction of VAT in June 2023 was associated with an inflation increase of 2.7 percentage points for that month.
  - Assuming broad VAT coverage by end-2023, DiD for June–December 2023 indicates an impact of 4.9 percentage points on inflation in 2023.
- Cross-country Panel Least Squares (11 Sub-Saharan African countries, excluding São Tomé and Príncipe):
  - The model indicates a VAT implementation dummy is associated with a 3.17 percentage point increase in average inflation.
  - Panel model results (preserved exactly):
    - Dummy VAT implementation: Coef. 3.17, P value 0.00
    - Lagged inflation: Coef. 0.52, P value 0.00
    - c Constant: Coef. 1.9, P value 0.00
    - R2: 0.40
    - Adjusted R2: 0.33
    - Observations: 362
- Caution: the 4.9 percentage point estimate for São Tomé and Príncipe should be interpreted with caution due to differing domestic policy responses not accounted for in the model.

---

### Box 2 — REER/NEER and NEER passthrough (summary)
- REER and NEER developments:
  - From 2010 to 2023, the REER experienced substantial appreciation, indicating a decline in competitiveness tied to persistent high inflation under the peg.
  - NEER rose by 20 percent during 2010–2023, largely driven by the Euro's strengthening against the US dollar.
- NEER passthrough analysis (panel of seven micro-islands, fixed exchange rate regimes, 2001–2023):
  - Estimated passthrough: approximately -0.33.
    - Interpretation preserved exactly: a 10 percent appreciation/depreciation of the NEER is associated with a decline/increase in annual inflation by 3.3 percentage points.
  - Empirical highlights:
    - Lagged NEER coefficient is negative but not statistically significant → passthrough occurs within the same year.
    - Inflation inertia and global crises effects significantly contribute to inflation in these micro-islands.
  - Panel model results (preserved exactly):
    - Inflation (t-1): Coef. 0.24, P value 0.00
    - NEER (t): Coef. -0.33, P value 0.00
    - NEER (t-1): Coef. -0.20, P value 0.10
    - Global crises: Coef. 2.10, P value 0.02
    - Constant: Coef. 1.63, P value 0.00
    - R2: 0.36
    - Adjusted R2: 0.31
    - Observations: 140
    - Number of countries: 7
    - Country fixed effects: yes

---

### C. Methodology and Data (model specification highlights)
- Model framework: Phillips Curve approach within an Autoregressive Distributed Lag Model, building on Lanau, Robles, and Toscani (2018), with enhancements.
- Key drivers included in the model:
  - Inflation inertia/expectations: past inflation as proxy for backward-looking adaptive expectations.
  - Demand-side shocks: lagged real GDP growth (economic cycle) and an elections dummy capturing looser macroeconomic discipline during national elections.
  - Supply-side shocks: Portugal’s food inflation as proxy for imported inflation shocks (reflecting heavy reliance on imported food, the euro peg, and Portugal as main trading partner).
  - Domestic fuel price adjustments.
  - Dummy variables for significant crises: global economic and financial crisis (2007–2008), COVID-19 crisis (2020), and crisis from Russia’s invasion of Ukraine (2022).
  - Dummy variable to capture effects of the pegged exchange rate regime.

*Source: Selected Issues Paper — "Strengthening Inflation Forecasting in São Tomé and Príncipe" (Prepared by Alsis Cruz and Dilson Tiny, led by Bahrom Shukurov), July 16, 2025.*

### 12.      We tested several other variables in the model, but none yielded satisfactory results.

### 12.      We tested several other variables in the model, but none yielded satisfactory results.

### Data and Methodology
- Annual data from 1998 to 2023 were used, including:
  - end-of-period inflation and real GDP growth in São Tomé and Príncipe;
  - end-of-period food inflation in Portugal;
  - changes in fuel prices in São Tomé and Príncipe.
- Dummy variables included:
  - elections in São Tomé and Príncipe (D1);
  - structural breaks associated with global and regional crises (D2);
  - transition to a pegged exchange rate regime (D3).
- Prior to estimation:
  - stationarity tested with the Augmented Dickey-Fuller and Phillips-Perron tests;
  - number of lags selected by the Akaike Information Criterion (AIC).
- Additional model tests attempted:
  - government expenditure (negative coefficient, statistically insignificant);
  - nominal effective exchange rate (positive coefficients contrary to theory, statistically insignificant);
  - international reserves and BCSTP credit to the government (positive coefficients, statistically insignificant).
- Including these additional variables would fail residual diagnostic tests (serial correlation, heteroskedasticity, normality).

### Model Specification
- Estimated equation (notation preserved):
  - 휋휋푖 = 푐푐 + 훼훼휋휋푖−1 + 훽훽1 푖푖푖푖푖푖.푖푖푖푖푖푖푖푖 + 훽훽2 푖푖푖푖푖푖.푖푖푖푖푖푖 푖−1 + 훿훿 푐푐푐푐푐푐푐푐푐푐 푖−1 + 휃휃 푖푖 푓푓푐푐 푐푐 푖 + 휑휑 퐷퐷1 푖 + 훾훾 퐷퐷2 푖 + 휔휔 퐷퐷3 푖 + μ푖
- Variable definitions (as in source):
  - 휋휋 = end-of-period inflation rate in São Tomé and Príncipe;
  - c = constant;
  - 푖푖푖푖푖푖.푖푖푖푖푖푖 = end-of-period food inflation rate in Portugal;
  - 푐푐푐푐푐푐푐푐푐푐 = real GDP growth in São Tomé and Príncipe;
  - fuel = percentage change in domestic fuel prices;
  - D1 = dummy for elections in São Tomé and Príncipe;
  - D2 = dummy for structural breaks associated with the global crises;
  - D3 = dummy for structural break related to the introduction of the pegged exchange rate regime.

### Estimation Results — Key Coefficients and Significance
- Lagged end-of-period inflation (휋휋푖−1):
  - Coeff.: 0.485
  - P value: 0.0001
  - Interpretation: A one-percentage point increase in the lagged end-of-period inflation results in a 0.49-percentage point rise in end-of-period inflation (statistically significant).
- End-of-period Portugal food inflation (푖푖푖푖푖푖.푖푖푖푖푖푖푖):
  - Coeff.: 0.761
  - P value: 0.0005
  - Interpretation: A one-percentage point increase in Portugal's food inflation correlates with a 0.76-percentage point rise in São Tomé and Príncipe inflation (statistically significant).
- Lagged end-of-period Portugal food inflation (푖푖푖푖푖푖.푖푖푖푖푖푖 푖−1):
  - Coeff.: 0.167
  - P value: 0.2433
  - Interpretation: One-lagged coefficient of 0.17 percentage point is not statistically significant.
- Lagged Real GDP growth (푐푐푐푐푐푐푐푐푐푐 푖−1):
  - Coeff.: 0.454
  - P value: 0.0490
  - Interpretation: A one-percentage point increase in lagged GDP growth results in a 0.45-percentage point rise in inflation (statistically significant).
- STP fuel price change (fuel_t):
  - Coeff.: 0.063
  - P value: 1.1766
  - Interpretation: A one-percentage point increase in domestic fuel price results in a 0.06-percentage point rise in inflation (coefficient not statistically significant; retained to account for supply-side shocks).
- Dummy – elections (D1):
  - Coeff.: 2.365
  - P value: 0.0480
  - Interpretation: Elections lead to a 2.37-percentage point rise in end-of-period inflation (statistically significant).
- Dummy – global crises (D2):
  - Coeff.: 3.172
  - P value: 0.0317
  - Interpretation: A global crisis would result in an increase of 3.17 percentage points in end-of-period inflation (statistically significant).
- Dummy – pegged regime (D3):
  - Coeff.: -4.014
  - P value: 0.0176
  - Interpretation: Implementation of the pegged regime contributed to a 4.01-percentage point slowdown in end-of-period inflation (statistically significant).
- Constant (c):
  - Coeff.: 3.766
  - P value: 0.1125
- Model fit and sample:
  - R2: 0.964835
  - Adjusted R2: 0.941392
  - Observations: 21

### Model Residual Diagnostics and Multicollinearity
- Residual evaluation (Table 2):
  - Breusch–Godfrey Serial Correlation LM Test (No serial correlation up to 2 lags):
    - F-statistic Prob. F(2,10): 0.8262
    - Obs*R-squared Prob. Chi-Square(2): 0.6747
  - Heteroskedasticity Test: Breusch-Pagan-Godfrey (Null: Homoskedasticity):
    - F-statistic Prob. F(8,12): 0.4022
    - Obs*R-squared Prob. Chi-Square(8): 0.3351
    - Scaled explained SS Prob. Chi-Square(8): 0.9699
  - Normality Test: Jarque-Bera:
    - Jarque-Bera: 0.331581
    - Probability: 0.847224
  - Conclusion: No serial correlation, homoskedastic errors, and residuals normally distributed.
- Multicollinearity (Table 3) — selected VIFs preserved:
  - 휋휋푖−1: Uncentered VIF 11.88, Centered VIF 2.68
  - 푖푖푖푖푖푖.푖푖푖푖푖푖: Uncentered VIF 3.58, Centered VIF 2.58
  - 푖푖푖푖푖푖.푖푖푖푖푖푖 푖−1: Uncentered VIF 2.59, Centered VIF 1.87
  - 퐺퐺퐷퐷 퐺퐺 푖−1: Uncentered VIF 7.08, Centered VIF 1.50
  - ICRISIS: Uncentered VIF 2.29, Centered VIF 1.85
  - ELECTION: Uncentered VIF 1.94, Centered VIF 1.47
  - PEG: Uncentered VIF 10.00, Centered VIF 3.33
  - FUEL: Uncentered VIF 3.60, Centered VIF 2.20
  - C: Uncentered VIF 34.06, Centered VIF NA
- Interpretation: VIF tests indicate relatively low correlation among independent variables after centering.

### Forecasting Performance and Projections
- Pseudo out-of-sample forecast performance (Forecast sample: 1998-2029; Adjusted sample: 2003-2029; Included observations: 27):
  - Root Mean Squared Error: 1.538
  - Mean Absolute Error: 1.258
  - Mean Absolute Percentage Error: 14.810
  - Theil Inequality Coef.: 0.052
    - Bias Proportion: 0.000
    - Variance Proportion: 0.094
    - Covariance Proportion: 0.905
  - Theil U2 Coefficient: 0.405
  - Symmetric MAPE: 13.631
- Model forecast (assuming other factors remain constant):
  - End-of-period inflation expected to decline to 7.7 percent in 2025.
  - Anticipated to rise to 9.1 percent in 2026 due to national elections.
  - Gradually declining to 6.5 percent by 2029.
- Model evaluation:
  - The model shows reasonably good pseudo out-of-sample performance over the past 15 years and low bias and variance per Theil measures.

### Main Findings
- Inflation drivers in São Tomé and Príncipe:
  - Inflation inertia (past inflation) is a significant driver.
  - Imported food prices proxied by Portugal's food inflation have a strong, statistically significant pass-through.
  - National elections are associated with a statistically significant increase in inflation (2.37 percentage points during election years).
  - Global crises produce a statistically significant inflation increase of 3.17 percentage points.
  - Pegged exchange rate regime implementation associated with a 4.01-percentage point reduction in inflation.
  - Lagged GDP growth contributes to inflation (0.45 percentage point per one-percentage point increase in lagged GDP growth).
  - Domestic fuel price changes exhibited the expected sign but were not statistically significant in the sample.

### Policy Recommendations
- Demand-side measures:
  - Maintain coherent macroeconomic policies, including ongoing fiscal consolidation and managing public debt levels.
  - Implement prudent monetary policy and liquidity management; consider raising interest rates to alleviate local demand pressures.
- Supply-side measures:
  - Increase local food production to reduce reliance on imported food:
    - Improve processing of primary agricultural products to add value and reduce waste.
    - Develop infrastructure such as greenhouses and agricultural facilities for year-round production and improved yields.
    - Promote sustainable agricultural practices to increase food production and enhance resilience against climate change.

_Italic: Source — Chapter content from sipea2025116 (PDF chapter/section). _

---


_Source: https://www.imf.org/-/media/files/publications/selected-issues-papers/2025/english/sipea2025116.pdf_