## wpiea2021132-print-pdf

## Source details

**Canonical URL:** [wpiea2021132-print-pdf](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021132-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021132-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021132-print-pdf.pdf.json)

---

### Introduction and objectives
- Focus: quantify exchange rate pass-through (ERPT) to domestic prices in Mozambique and assess its evolution over time, accounting for the inflationary environment, financial crisis, and monetary policy regime.
- Rationale: pronounced depreciations coinciding with high inflation episodes (e.g., 2015-2017 external debt crisis) provide an opportunity to estimate ERPT.
- Methodological approach: Autoregressive Distributed Lag (ARDL) applied to monthly data from January 2001 to December 2019 to address small-sample issues, permit mixed integration orders, and reduce reverse causality concerns via lagged interactions.

### Data, variables, and controls
- Sample period: monthly data from January 2001 to December 2019 (Observations: 228.00 for listed variables).
- Main dependent variable: Mozambique Consumer Price Index (CPI).
- Exchange rate measures:
  - NEER (nominal effective exchange rate); increase indicates depreciation.
  - USD/MZN (bilateral); units of domestic currency per unit of foreign currency (increase = depreciation).
  - ZAR/MZN (bilateral); units of domestic currency per unit of foreign currency (increase = depreciation).
  - 2019 trade-weight shares for 10 major trading partners: South Africa (47%), India (23%), China (18%), Singapore (10%), United Arab Emirates (10%), Italy (8%), Netherlands (7%), United Kingdom (6%), Japan (5%) and Belgium (5%).
- Controls:
  - Import Price Indices (MPIs) from the IMF.
  - Real money supply (M3) gap (Bandpass filter); robustness checks use M2 and M1 gaps and the Economic Activity Index.
  - Rainfall index from the National Institute of Statistics; supply gap computed as deviation from Bandpass-trend in some specifications.
- Seasonal adjustment: X-13 ARIMA-SEATS.
- Descriptive statistics highlights:
  - CPI: Mean 96.15; Median 95.10; Maximum 171.77; Minimum 34.58; Std. Dev. 39.96.
  - NEER: Mean 130.11; Std. Dev. 40.88.
  - USDMZN: Mean 34.30; Std. Dev. 15.42.
  - ZARMZN: Mean 3.60; Std. Dev. 0.78.
  - M3 (Million MT): Mean 162,706.70; Std. Dev. 138,265.10.
  - Rainfall (mm): Mean 60.65; Std. Dev. 68.61.
  - Agriculture accounts for around 25 percent of GDP.

### Econometric approach and identification
- Stationarity: ADF and PP tests indicate most series are nonstationary in level and stationary in first differences, but real money supply and rainfall gaps are exceptions.
- Method choice: ARDL (Pesaran et al., 2001) selected because it:
  - Does not require same integration order across variables (provided none are I(2) or higher).
  - Is suitable for small samples.
  - Allows asymmetric short-run and long-run effects and helps mitigate endogeneity via lag structure.
- Long-run specification (logs): lnCPI_t = φ0 + φ1 lnE_t + φ2 lnMPI_t + φ3 T + ε_t.
- Cointegration test: ARDL Bounds F-test (Wald F-statistic) with linear trend; optimal lags by AIC.

### Cointegration and long-run elasticities
- ARDL Bounds F-test F-statistics (Baseline, Asymmetric, Inflationary environment, Monetary policy regime, Financial Crisis models) all exceed upper bound critical value at 5% (I(1)=5.85) for NEER, USD/MZN, and ZAR/MZN specifications, leading to rejection of no cointegration.
  - Example F-statistics (Baseline Model): NEER 18.22; USDMZN 11.71; ZARMZN 10.73.
- Estimated long-run exchange-rate elasticities (coef., p-value):
  - NEER: 0.418 (0.000)
  - USDMZN: 0.531 (0.000)
  - ZARMZN: 0.296 (0.000)
- Estimated long-run import-price elasticities (coef., p-value):
  - NEER: 0.133 (0.000)
  - USDMZN: 0.331 (0.000)
  - ZARMZN: 0.069 (0.026)
- Trend coefficients (coef., p-value): NEER 0.004 (0.000); USDMZN 0.003 (0.002); ZARMZN 0.006 (0.000).
- Interpretation (long run):
  - A 10 percent depreciation of the Metical:
    - relative to NEER increases domestic prices by 4.2 percent, all else equal.
    - relative to USD increases domestic prices by 5.3 percent, all else equal.
    - relative to ZAR increases domestic prices by 3 percent, all else equal.

### Short-run dynamics, speed of adjustment, and decomposition
- Short-run monthly pass-through (one-month lag):
  - 10 percent depreciation of the Metical against NEER increases domestic prices by 0.46 percent in the following month.
  - 10 percent depreciation against USD increases domestic prices by 0.71 percent in the following month.
  - 10 percent depreciation against ZAR increases domestic prices by 0.42 percent in the following month.
- Error-correction term (ECT):
  - ECT coefficients negative and significant; 6.6 percent of the short run disequilibrium of the CPI are corrected toward the long run in one month for NEER, USD and ZAR models.
  - Convergence to steady state long run equilibrium: about 11 months (NEER), 20 months (USD), 17 months (ZAR).
- Impulse-response timing:
  - About 50 percent of ERPT from NEER and ZAR/MZN to CPI occurs in about 4 months.
  - About 50 percent of ERPT from USD/MZM to CPI occurs in about 5 months.
- Variance decomposition highlights:
  - NEER variations explain up to 80 percent of total variation in CPI in 20 months (Table 0-5).
  - USD/MZN variations explain up to 59 percent of total variation in CPI in 40 months.
  - ZAR/MZN variations explain up to 27 percent of total variation in CPI in 50 months.
  - Specific period shares: Period 10 — NEER model: NEER contributes 69 percent; USDMZN model: USDMZN contributes 38 percent; ZARMZN model: ZARMZN contributes 20 percent. Period 50 — NEER model: NEER contributes 77 percent; USDMZN model: USDMZN contributes 58 percent; ZARMZN model: ZARMZN contributes 10 percent.

### Asymmetry, inflation environment, financial crisis, and monetary policy regime
- Asymmetry testing: Shin et al. (2014) extension splitting exchange-rate changes into partial sums of positive (depreciations) and negative (appreciations) changes to obtain separate long-run and short-run elasticities; Wald tests for symmetry in short and long run.
  - Reported pattern: depreciations transmit more strongly to prices than appreciations in magnitude; Wald tests suggest symmetry in the long run but asymmetry in the short run (short-run asymmetry significant).
- Inflation environment effects:
  - Two measures: MAINF (12-month moving average of inflation) and SDINF (standard deviation of inflation).
  - Findings:
    - Level of inflation (MAINF) does not have a significant effect on ERPT when interaction terms used.
    - Inflation volatility (interaction between standard deviation of inflation and exchange rate) has a positive and significant impact on ERPT for NEER and ZAR/MZN models: ERPT is higher in more volatile inflation periods.
- Financial crisis (global financial crisis 2007-2009; dummy from September 2009) effects:
  - ERPT increased during the crisis and reverted to long-run equilibrium later; crisis effect temporary.
  - Table 0-7 reports interaction coefficients (Exchange Rate*Financial Crisis2008) that are in some cases negative or not significant (e.g., NEER Exchange Rate*Financial Crisis2008 -0.092 (0.210)).
- Monetary policy regime shift (MPReforms17 dummy = 1 from April 2017):
  - Table 0-8 indicates the change to an interest rate-based regime did not significantly affect the magnitude of ERPT to date (interaction terms not significant).
  - Possible reasons: insufficient post-reform time span or regime may not influence long-run ERPT yet.

### Robustness checks and diagnostics
- Omitted regressor checks:
  - Demand proxied by money supply (M3; M2 and M1 checked) and Economic Activity Index.
  - Supply proxied by rainfall index (gap from trend).
  - Table 0-9: inclusion of Money Supply and Rainfall Index does not materially change ERPT estimates; money supply and rainfall not significant determinants of CPI in these specifications.
- Unit-root tests with structural breaks and alternative break dates: results broadly unchanged.
- Short-run ARDL-ECM diagnostics (Table 0-3):
  - ECM(-1) coefficients: NEER -0.095* (0.014); USDMZN -0.049* (0.009); ZARMZN -0.066* (0.013). (* significance at 1%).
  - Short-run D(ER(-1)) coefficients: NEER 0.046** (0.018); USDMZN 0.071** (0.031); ZARMZN 0.042** (0.016). (** significance at 5%).
  - Model diagnostics reported (R-squared, Durbin-Watson, LM Test, Ramsey, BPG, JB) in tables.

### Conclusions and policy implications
- The exchange rate is an important determinant of domestic prices in Mozambique; significant ERPT exists for USD/MZM, ZAR/MZM, and NEER.
- Exact long-run pass-through for a 10 percent depreciation:
  - USD/MZM: 5.3 percent increase in domestic prices.
  - ZAR/MZM: 2.96 percent increase in domestic prices.
  - NEER: 4.2 percent increase in domestic prices.
- Short-run: 6.6 percent of short-run CPI disequilibrium corrected toward long run in one month; substantial share of pass-through realized within 4–5 months.
- Policy guidance:
  - Targeting the exchange rate level or its variation to limit ERPT could be risky and unsustainable given reserve constraints and fiscal risks.
  - A floating exchange rate combined with a credible interest rate-based forward-looking monetary policy could help anchor inflation expectations and reduce ERPT over time.
  - Transition to the interest rate-based regime (April 2017) has potential to reduce ERPT but evidence to date does not show a decline; consolidation and time are needed.
  - Prudent macroeconomic policies and consistent pursuit of monetary objectives are likely necessary for declining ERPT as suggested by literature.

### Open questions and future research avenues
- What explains higher ERPT in periods of high macroeconomic volatility?
- Does short-run asymmetric ERPT amplify during periods of high volatility?
- Will the interest rate-based monetary policy regime ultimately reduce ERPT as expected?
- Further research needed on expectations of further depreciations, coping strategies in stressed periods, and the degree of dollarization.

*Source: wpiea2021132-print-pdf*

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

### wpiea2021132-print-pdf - References .............................................................................................................

### Introduction and objectives
- Focus: quantify exchange rate pass-through (ERPT) to domestic prices in Mozambique and assess its evolution over time, accounting for the inflationary environment, financial crisis, and monetary policy regime.
- Rationale: Mozambique experienced pronounced depreciations coinciding with high inflation episodes (e.g., 2015-2017 external debt crisis), providing an opportunity to estimate ERPT.
- Methodological approach highlighted: Autoregressive Distributed Lag Approach (ARDL) applied to monthly data from 2001 to 2019 to address small-sample issues, permit mixed integration orders, and reduce reverse causality concerns by exploiting lagged interactions.

### Key findings (empirical results summarized)
- Asymmetric ERPT: exchange rate variations have an asymmetric effect on domestic prices.
- Point estimates of long-run ERPT (response of CPI to a 10 percent depreciation):
  - 10 percent depreciation against the nominal effective exchange rate (NEER) increases domestic prices by 4.2 percent.
  - 10 percent depreciation against the United States Dollar (USD) increases domestic prices by 5.3 percent.
  - 10 percent depreciation against the South Africa Rand (ZAR) increases domestic prices by 3 percent.
- Time variation:
  - ERPT was broadly stable over the sample period but tended to be relatively high in periods of high volatility of the exchange rate and inflation.
  - Notably higher ERPT during the 2008 global financial crisis and the 2016 debt crisis in Mozambique.
- Asymmetry detail: depreciations transmit more strongly to prices than appreciations (consistent with cited literature and results).

### Novelty and contributions relative to prior Mozambican studies
- Methodological novelty: use of the ARDL approach combined with longer monthly time series (January 2001 to December 2019).
- Additional analyses not addressed in previous Mozambique studies:
  - Impact of the inflation environment on ERPT.
  - Impact of financial crisis episodes on ERPT.
  - Impact of the monetary policy regime shift (transition from monetary-based to interest rate-based framework since 2017) on ERPT.
  - Examination of asymmetric ERPT effects.
- Broader exchange rate coverage: considers NEER and USD in addition to the traditional focus on the ZAR.

### Data: sample, key variables, definitions, and controls
- Sample period: monthly data from January 2001 to December 2019.
- Main dependent variable: Mozambique Consumer Price Index (CPI) obtained from Mozambique’s National Institute of Statistics.
- Exchange rate measures:
  - Bilateral exchange rates: MZN vis-à-vis ZAR and USD obtained from the Bank of Mozambique; defined as units of domestic currency per unit of foreign currency (an increase represents depreciation).
  - Nominal Effective Exchange Rate (NEER) obtained from the International Monetary Fund; calculated as a trade-weighted average of bilateral exchange rates of the Mozambican Metical vis-à-vis Mozambique’s main trading partners (an increase indicates depreciation).
  - 2019 trade-weight shares for the 10 major trading partners: South Africa (47%), India (23%), China (18%), Singapore (10%), United Arab Emirates (10%), Italy (8%), Netherlands (7%), United Kingdom (6%), Japan (5%) and Belgium (5%).
- Seasonal adjustment: all data series seasonally adjusted through the X-13 ARIMA-SEATS seasonal adjustment method.
- Control variables included:
  1. Import Price Indices (MPIs) from the IMF — weighted average of price indices for elementary aggregates using relative trade values as weights.
  2. Real money supply (M3) gap from the Bank of Mozambique — real money supply computed by dividing broad money supply by CPI; gap defined as deviation from trend derived via the Bandpass filter. Robustness checks use M2 and M1 gaps and the economic activity index from the National Institute of Statistics.
  3. Rainfall index (used in some specifications; referenced in table listings).

### Econometric and identification considerations
- ARDL advantages emphasized:
  - Suited for small samples.
  - Does not require same-order integration among variables.
  - Reduces concerns over reverse causality by exploiting lag structure.
- Causality and omitted variables:
  - Acknowledges potential endogeneity between exchange rate and prices.
  - Notes omitted factors (external volatility, climate shocks) can drive correlations; the model includes controls (import prices, money supply, rainfall index) and explores lagged dynamics to mitigate these concerns.
  - References prior work showing potential causality from domestic prices to exchange rates (Hassan and Simione (2013)).

### Contextual background and theoretical framing
- Micro and macro determinants:
  - ERPT magnitude depends on market structure (monopoly vs imperfect competition), share of domestically produced goods in consumption, transport costs, trade barriers, distribution and retail costs.
- Role of inflation and monetary policy:
  - Lower and stable inflation environments associated with lower ERPT (Taylor (2000); Gagnon and Ihrig (2004); McCarthy (2007); Özyurt (2016)).
  - Credible and focused monetary policy (e.g., inflation targeting) can anchor expectations and reduce ERPT.
  - Low ERPT can reduce “fear of floating” and allow greater monetary policy independence, though it may weaken the exchange rate transmission of monetary policy.

### Comparative literature evidence (high-level)
- Developed countries: evidence of modest and declining ERPT in many studies (Taylor, 2000; Gagnon and Ihrig, 2004; McCarthy, 2007; Özyurt, 2016).
- Developing countries and Sub-Saharan Africa: ERPT often incomplete and asymmetric; determinants include inflation environment, exchange rate regime, monetary and fiscal policy credibility (Choudhri and Hakura, 2006; Akofio-Sowah, 2009; Razafimahefa, 2012; Lariau et al., 2016; Helmy et al., 2018; Kassi et al., 2019).
- Mozambique prior estimates (reported range): pass-through estimates between 10 and 74 percent across different studies, periods, methodologies, and price aggregations (examples reported: 20 percent for 1989-1996; 74 percent for 1993-2001; 50 to 70 percent for 2000-2005; 15 percent for 2001-2006).

*Source: wpiea2021132-print-pdf - References .............................................................................................................*

### 3. Rainfall index: the rainfall index was obtained from the National Institute of Statistics

### 3. Rainfall index: the rainfall index was obtained from the National Institute of Statistics

### Role of the rainfall index and data properties
- The rainfall index intends to capture supply conditions and food scarcity given the agriculture sector accounts for around a quarter of GDP and is heavily dependent on rainfall.
- Scarce rainfall tends to reduce the supply of domestically produced food with adverse impacts on inflation.
- A robustness check controls for a supply gap computed as the deviation of actual rainfall index from the trend derived through the Bandpass filter.
- Descriptive statistics (Table 1 in appendix A) indicate substantial variation across variables (relatively large standard deviations and large min–max gaps).
- Except for the USD/MZN, distributions of series in the baseline regression model are approximately symmetric (skewness values).
- Time-series plots (Figure 1 in appendix B) show:
  - CPI, exchange rates and import prices exhibit an overall positive trend.
  - Money supply and rainfall index gaps fluctuate around the mean.
  - ZAR/MZN, import price, and to a lower extent USD/MZN and NEER reveal apparent structural breaks coinciding with the global financial crisis in 2007-2009.

### Time-series properties and choice of methodology
- Augmented Dickey-Fuller (ADF) and Phillips-Peron (PP) tests (Table 2 of Appendix A) indicate:
  - Except for real money supply and rainfall gaps, all variables are nonstationary in level but become stationary in first differences.
  - Unit-root tests with structural breaks produce broadly similar results.
- Because not all variables share the same order of integration, the Johansen cointegration approach is not applied.
- The paper adopts the Autoregressive Distributed Lag (ARDL) model (Pesaran et al., 2001) which:
  - Does not require series to be integrated to the same order (provided none are I(2) or more).
  - Is more reliable in small samples and can account for asymmetric short-run and long-run effects.
  - Mitigates endogeneity concerns due to its lagged structure.

### Empirical specification (summary)
- Long-run model in logs (Equation 1): lnCPI_t = φ0 + φ1 lnE_t + φ2 lnMPI_t + φ3 T + ε_t, where:
  - CPI = domestic consumer price index.
  - E = exchange rate (NEER, USD/MZN, ZAR/MZN alternatives).
  - MPI = import price index.
  - φ1 captures the exchange rate pass-through (ERPT) to CPI.
- ARDL error-correction and levels representations are used (Equations 2–6) to:
  - Test cointegration via bounds test (Wald F-statistic).
  - Estimate long-run elasticities from estimated ARDL coefficients (Equations 4–5).
  - Obtain short-run dynamics and error-correction speeds (Equation 6; ECT denotes error correction term).
- Asymmetric ERPT is tested using Shin et al. (2014) extension (Equations 7–8) splitting exchange-rate changes into partial sums of positive (depreciations) and negative (appreciations) changes to obtain separate long-run and short-run elasticities and Wald tests for symmetry.
- Effects of inflation environment, global financial crisis (structural break circa 2007-2009; dummy from September 2009), and monetary policy regime shift (dummy from April 2017) are modeled via interactions and dummy variables (Equations 9–10). MAINF = 12 months moving average of inflation; SDINF = standard deviation of inflation.

### Cointegration test results (ARDL Bounds F-test)
- Optimal lag lengths selected by AIC; linear trend added to cointegration test.
- Table IV-1 F-statistics (ARDL Bounds Test):
  - Baseline Model: NEER 18.22, USDMZN 11.71, ZARMZN 10.73
  - Assymetric Effects Model: NEER 21.65, USDMZN 10.55, ZARMZN 7.62
  - Inflationary Environmental Effects Model: NEER 12.91, USDMZN 9.68, ZARMZN 9.13
  - Monetary Policy Regime Model: NEER 11.48, USDMZN 7.60, ZARMZN 7.23
  - Financial Crisis Effect Model: NEER 11.86, USDMZN 9.48, ZARMZN 7.49
- Critical Values (5%): Lower bound I(0) = 4.87; Upper bound I(1) = 5.85.
- Interpretation:
  - For all three exchange-rate specifications (NEER, USD/MZN, ZAR/MZN) and across the five regressions, the computed F-statistics exceed the upper bound critical value of 5.85 at 5 percent, so the null hypothesis of no cointegration is rejected.
  - Conclusion: there is a stable long-run cointegration between CPI, the three exchange rates and other explanatory variables.

### Long-run elasticities and ARDL coefficient estimates (Table IV-2)
- ARDL model lag orders reported:
  - NEER: (2, 2, 2)
  - USDMZN: (2, 2, 1)
  - ZARMZN: (2, 2, 1)
- Estimated long-run exchange-rate elasticities (coef., p-value):
  - NEER: 0.418 (0.000)
  - USDMZN: 0.531 (0.000)
  - ZARMZN: 0.296 (0.000)
- Estimated long-run import-price elasticities (coef., p-value):
  - NEER: 0.133 (0.000)
  - USDMZN: 0.331 (0.000)
  - ZARMZN: 0.069 (0.026)
- Trend coefficients (coef., p-value):
  - NEER: 0.004 (0.000)
  - USDMZN: 0.003 (0.002)
  - ZARMZN: 0.006 (0.000)
- Constant terms (coef., p-value):
  - NEER: 5.507 (0.262)
  - USDMZN: 0.471 (0.439)
  - ZARMZN: 2.352 (0.000)
- Interpretation (reported in text):
  - A 10 percent depreciation of the domestic currency:
    - relative to NEER increases domestic prices by 4.2 percent, all else equal.
    - relative to USD increases domestic prices by 5.3 percent, all else equal.
    - relative to ZAR increases domestic prices by 3 percent, all else equal.
  - Import-price increases are also significantly associated with higher domestic prices.

### Short-run elasticities, dynamics, and variance/decomposition
- Short-run monthly pass-through (one-month lag) reported in text:
  - A 10 percent depreciation of the Metical against:
    - NEER increases domestic prices by 0.46 percent in the following month.
    - USD increases domestic prices by 0.71 percent in the following month.
    - ZAR increases domestic prices by 0.42 percent in the following month.
- Error correction term:
  - The ECT coefficient is negative and significant, indicating short-run disequilibrium of CPI adjusts towards the steady state.
- Impulse-response timing (Table 5 and Figure 2 in appendices):
  - About 50 percent of ERPT from NEER and ZAR/MZN to CPI occurs in about 4 months.
  - About 50 percent of ERPT from USD/MZM to CPI occurs in about 5 months.
- Variance decomposition (Table 5 of appendix A):
  - NEER variations explain up to 80 percent of total variation in CPI in 20 months.
  - USD/MZN variations explain up to 59 percent of total variation in CPI in 40 months.
  - ZAR/MZN variations explain up to 27 percent of total variation in CPI in 50 months.

### Asymmetric effects and environment/regime analysis (summary of approach and reported patterns)
- Asymmetry:
  - ERPT asymmetry is tested by decomposing exchange-rate changes into positive (depreciations) and negative (appreciations) cumulative sums and estimating separate long-run and short-run elasticities (Equations 7–8).
  - Long-run asymmetric elasticities are represented by −φ2+φ1 (depreciation) and −φ2−φ1 (appreciation).
  - Short-run asymmetric elasticities are sums of γi+ and γi− terms from the error-correction form.
  - Wald tests are used to test equality of depreciation vs. appreciation effects in short-run and long-run.
- Inflation environment, global financial crisis, and monetary policy regime:
  - Two measures captured via D: MAINF (12-month moving average of inflation) and SDINF (standard deviation of inflation).
  - Interaction terms are used to assess whether higher level or instability of inflation increases ERPT (Equation 9).
  - Dummy Dt−1 captures periods after global financial crisis (from September 2009) and the new monetary policy regime (from April 2017) and is interacted with exchange-rate terms to capture regime/crisis effects (Equation 10).
- Reported pattern:
  - Figure IV-1 (rolling coefficients) suggests long-run ERPT is volatile and tends to be higher during periods of higher exchange-rate and inflation volatility, notably during the 2007/9 global financial crisis and the 2015/17 debt crisis in Mozambique.
  - The ERPT is higher for the US Dollar (USD/MZN) relative to the South African Rand (ZAR/MZN), consistent with the US Dollar being the main currency used in foreign transactions despite South Africa being the largest single trading partner by import share.

*Source: wpiea2021132-print-pdf - 3. Rainfall index: the rainfall index was obtained from the National Institute of Statistics*

### 6.6 percent of the short run disequilibrium of the CPI are corrected toward the long run

### 6.6 percent of the short run disequilibrium of the CPI are corrected toward the long run

### Key findings on speed and magnitude of adjustment
- 6.6 percent of the short run disequilibrium of the CPI are corrected toward the long run equilibrium in one month for, respectively, the NEER, USD and ZAR models.
- It takes, respectively, about 11, 20 and 17 months to completely converge to the steady state long run equilibrium for the NEER, USD and ZAR models.
- In the long-run, each 10 percent depreciation of the Mozambican Metical vis-à-vis:
  - the US Dollar (USD/MZM) leads to a 5.3 percent increase in domestic prices, all else equal.
  - the Rand (ZAR/MZM) leads to a 2.96 percent increase in domestic prices, all else equal.
  - the NEER leads to a 4.2 percent increase in domestic prices, all else equal.
- About 50 percent of the pass-through for the ZAR/MZM and USD/MZM occur in 4 to 5 months, respectively.
- The exchange rate accounts for 27 percent and 59 percent of the variation in domestic prices in 50 months and 40 months, respectively.

### Asymmetry and relative contributions
- Wald Test results suggest:
  - exchange rate appreciations and depreciations have symmetric effects on domestic prices in the long run.
  - exchange rate appreciations and depreciations have asymmetric effects in the short run.
- Elasticities indicate that in the long run the magnitude of depreciations pass-through to inflation is higher than the appreciations pass-through, particularly for the USD/MZM exchange rate.
- Variations of the USD/MZN exchange rate have the highest contribution to inflation relative to the ZAR/MZN and NEER exchange rates.
- The short-run asymmetric effect possibly reflects retail market structure and competition issues (not analyzed further in the paper).

### Exchange rate pass-through (ERPT), inflation environment, monetary policy regime and financial crisis
- Inflation environment measurement approaches:
  - Level approach: interaction between exchange rate and the 12-month moving average of inflation rate (captures high vs low inflation).
  - Volatility approach: interaction between exchange rate and the 12-month moving average of the standard deviation of inflation rate (captures inflation volatility).
- Findings:
  - When defined by the 12-month moving average inflation (high vs low inflation), the inflation environment does not have a significant effect on the ERPT—no evidence ERPT differs in low and high inflation environments.
  - Using an alternative dummy (1 when inflation is higher than one digit levels for at least six months) likewise shows no significant effect on ERPT.
  - When measured by volatility of inflation (interaction coefficient between standard deviation of inflation and the exchange rate), there is a positive and significant impact on the ERPT for the NEER and ZAR/MZN models: ERPT is higher in more volatile inflation periods.
  - The ERPT was higher in periods of macroeconomic instability, including during the global financial crisis (2007-2009) and Mozambique’s debt crisis (2015-2017).
- Global financial crisis impact:
  - Table 7 in Appendix A suggests the global financial crisis changed the magnitude of the ERPT only temporarily: ERPT increased during the crisis and reverted to its long-run equilibrium in later years (as shown in Figure 4.1).
- Monetary policy regime shift:
  - Table 8 in Appendix A suggests the change from monetary to interest rate targeting (regime shift in April 2017) did not affect the magnitude of the ERPT to date.
  - Possible interpretation: the time since the regime shift is not long enough to change behavioral parameters governing ERPT, or the regime shift may not influence long-run ERPT (contrasting with literature that finds declining ERPT with more credible monetary frameworks).

### Robustness checks
- Two potential omitted regressors considered: demand and rainfall conditions.
- Demand proxied by money supply (M3; M2 and M1 also checked) and alternatively by Economic Activity Index.
- Supply proxied by rainfall index (agriculture accounts for around 25 percent of GDP; rainfall affects domestic food supply and inflation).
- Results:
  - Table 09 in Appendix A suggests money supply and rainfall do not have a significant effect on domestic prices.
  - ERPT estimates remain significant and broadly unchanged when adding money supply and rainfall index.
  - Results remain broadly unchanged when testing unit root hypothesis in presence of structural break with different break dates.
  - Replacing money supply with Economic Activity Index yields similar findings (demand not significant).

### Conclusions and policy implications
- The exchange rate is an important determinant of domestic prices in Mozambique; significant pass-through effects exist for USD/MZM, ZAR/MZM, and NEER.
- High magnitude of ERPT and speed of adjustment pose a significant policy challenge that is difficult to tackle in the short term.
- Policy guidance:
  - Targeting the exchange rate level or its variation to limit ERPT could be risky and unsustainable for a given stock of international reserves, especially under weak fiscal discipline.
  - A floating exchange rate regime consistent with a credible interest rate-based forward-looking monetary policy regime could help anchor inflation expectations.
  - The transition to an interest rate-based monetary policy regime (introduced in April 2017) holds potential to reduce ERPT over time, although results do not yet show a decline; consolidation of the regime is ongoing.
  - Literature suggests ERPT should decline if monetary policy objectives are pursued consistently, accompanied by low and stable inflation outcomes and prudent macroeconomic policies.

### Open questions and suggested future research
- What explains the higher ERPT in periods of high macroeconomic volatility?
- Does the asymmetric ERPT observed in the short run also hold in periods of high macroeconomic volatility?
- Will the new interest rate-based monetary policy regime lead to lower ERPT as expected?
- The role of expectations of further depreciations, coping strategies during stressed periods, and the degree of dollarization in the economy warrant separate follow-up research.

### Key statistics and exact numeric findings
- 6.6 percent (portion of short run disequilibrium of the CPI corrected in one month for NEER, USD and ZAR models)
- Convergence to steady state long run equilibrium: about 11 months (NEER), 20 months (USD), 17 months (ZAR)
- Long-run pass-through for a 10 percent depreciation:
  - USD/MZM: 5.3 percent increase in domestic prices
  - ZAR/MZM: 2.96 percent increase in domestic prices
  - NEER: 4.2 percent increase in domestic prices
- About 50 percent of pass-through for ZAR/MZM and USD/MZM occur in 4 to 5 months
- Exchange rate accounts for 27 percent and 59 percent of variation in domestic prices in 50 months and 40 months, respectively
- Global financial crisis period referenced: 2007-2009
- Mozambique debt crisis period referenced: 2015-2017
- Monetary policy regime shift: April 2017
- Agriculture share of GDP: around 25 percent

*Source: wpiea2021132-print-pdf*

### APPENDIX A: TABLES

### APPENDIX A: TABLES

### Descriptive statistics (Table 0-1)
- Variables reported: CPI, NEER, USDMZN, ZARMZN, Foreign Prices, Rainfall (mm), M3 (Million MT).
- Observations: 228.00 for all variables.
- Selected summary statistics:
  - CPI: Mean 96.15; Median 95.10; Maximum 171.77; Minimum 34.58; Std. Dev. 39.96; Skewness 0.42; Kurtosis 2.16.
  - NEER: Mean 130.11; Median 128.55; Maximum 245.05; Minimum 59.10; Std. Dev. 40.88; Skewness 0.46; Kurtosis 2.99.
  - USDMZN: Mean 34.30; Median 27.50; Maximum 77.58; Minimum 17.12; Std. Dev. 15.42; Skewness 1.28; Kurtosis 3.18.
  - ZARMZN: Mean 3.60; Median 3.49; Maximum 5.60; Minimum 2.17; Std. Dev. 0.78; Skewness 0.46; Kurtosis 2.62.
  - Foreign Prices: Mean 70.55; Median 67.34; Maximum 121.86; Minimum 36.04; Std. Dev. 20.69; Skewness 0.27; Kurtosis 2.18.
  - Rainfall (mm): Mean 60.65; Median 37.15; Maximum 329.50; Minimum -; Std. Dev. 68.61; Skewness 1.73; Kurtosis 5.93.
  - M3 (Million MT): Mean 162,706.70; Median 120,701.30; Maximum 474,714.00; Minimum 17,318.53; Std. Dev. 138,265.10; Skewness 0.72; Kurtosis 2.09.
- Source: Researchers' calculation based on Eviews11.

### Unit root tests (Table 0-2: ADF and Phillips-Perron)
- Test details (variables shown with test statistics, critical values, lag orders, and integration order):
  - lnCPI c,t: ADF statistic 1-3.191 with critical -3.430; PP statistic 7-2.776 with critical -3.430 -> I(0).
  - D(lnCPI) c: ADF statistic 0-8.686 with critical -2.874; PP 4-8.743 with critical -2.874 -> I(1).
  - lnP* c,t: ADF 1-1.924 (-3.430); PP 6-1.851 (-3.430) -> I(1).
  - D(lnP*) c: ADF 0-10.493 (-2.874); PP 4-10.567 (-2.874) -> I(1).
  - lnUSDMZN c,t: ADF 1-1.749 (-3.430); PP 9-1.981 (-3.430) -> I(1).
  - D(lnUSDMZN) c: ADF 0-10.080 (-2.874); PP 7-10.624 (-2.874) -> I(1).
  - lnZARMZN c,t: ADF 1-2.547 (-3.430); PP 7-2.535 (-3.430) -> I(1).
  - D(lnZARMZN) c: ADF 0-10.685 (-2.874); PP 3-10.736 (-2.874) -> I(1).
  - lnNEER c,t: ADF 2-2.878 (-3.430); PP 8-2.608 (-3.430) -> I(1).
  - D(lnNEER) c: ADF 1-7.022 (-2.874); PP 5-10.631 (-2.874) -> I(0).
  - lnM3 c,t: ADF 0-3.538 (-3.430); PP 1-3.340 (-3.430) -> I(0).
  - D(lnM3) c: ADF 0-13.760 (-2.874); PP 5-13.741 (-2.874).
  - lnRainfall c,t: ADF 6-9.840 (-3.430); PP 8-10.072 (-3.430).
  - D(lnRainfall) c: ADF 12-10.750 (-2.87519); PP -40.506 (-2.874).
- Notes: Critical values obtained from MacKinnon (1996). Null hypothesis for ADF and PP is unit root. Source: Researchers' computation based on Eviews 11.

### Short-run ARDL-ECM results (Table 0-3)
- Models: USDMZN (2,2,1); NEER (2,2,2); ZARMZN (2,2,1). Dependent variable: D(LnConsumer Price Index). All variables in logs; constant and trend included.
- Key coefficients (short-run and error-correction):
  - ECM(-1): NEER model -0.095* (Std. Error 0.014); USDMZN model -0.049* (0.009); ZARMZN model -0.066* (0.013). (* significance at 1%).
  - D(CPI(-1)): 0.257* (0.063), 0.266* (0.077), 0.307* (0.070) across models.
  - D(ER(-1)): 0.046** (0.018) NEER; 0.071** (0.031) USDMZN; 0.042** (0.016) ZARMZN. (** significance at 5%).
  - Other short-run coefficients for D(ER), D(P), lags provided with standard errors; several non-significant.
- Model diagnostics presented: R-squared, S.E. of regression, Prob(F-statistic), Durbin-Watson Stat, LM Test, Ramsey Test, BPG Test, JB Test (values listed in table but not all textually reproduced here).
- Source: Researchers' computation based on Eviews 11.

### CPI response to a one percent shock of exchange rate and import prices (Table 0-4)
- Cumulative impulse responses by period (with standard errors in parentheses). Three ARDL specifications (NEER, USDMZN, ZARMZN) with import prices:
  - Period 1: all zeros.
  - Period 2: NEER 0.077 (0.017) vs Import Prices -0.007 (0.011); USDMZN 0.107 (0.017) vs Import Prices 0.010 (0.010); ZARMZN 0.062 (0.012) vs Import Prices -0.008 (0.011).
  - Period 5: NEER 0.297 (0.038) vs Import Prices 0.033 (0.022); USDMZN 0.262 (0.039) vs Import Prices 0.072 (0.026); ZARMZN 0.182 (0.026) vs Import Prices 0.016 (0.023).
  - Period 10: NEER 0.429 (0.068) vs Import Prices 0.041 (0.038); USDMZN 0.291 (0.052) vs Import Prices 0.093 (0.040); ZARMZN 0.223 (0.037) vs Import Prices 0.010 (0.032).
  - Longer horizons show responses that rise and then partly reverse (periods 20, 30, 40, 50 with values and standard errors reported).
- Note: Standard errors reported in parentheses. Source: Researchers' computation based on Eviews 11.

### Variance decomposition of CPI (Table 0-5)
- Three model variants: NEER Model, USDMZN Model, ZARMZN Model. Period-by-period percent contributions to CPI forecast error variance (S.E. and shares):
  - Period 1: CPI 100; NEER/USDMZN/ZARMZN 0; Import Prices 0.
  - Period 2: NEER model CPI 96, NEER 4; USDMZN model CPI 94, USDMZN 6; ZARMZN model CPI 96, ZARMZN 4.
  - Period 10: NEER model CPI 29, NEER 69, Import Prices 2; USDMZN model CPI 53, USDMZN 38, Import Prices 9; ZARMZN model CPI 78, ZARMZN 20, Import Prices 2.
  - Period 50: NEER model CPI 16, NEER 77, Import Prices 7; USDMZN model CPI 32, USDMZN 58, Import Prices 10; ZARMZN model CPI 63, ZARMZN 10, Import Prices 27.
- Source: Researchers' computation based on Eviews 11.

### Short-run exchange rate asymmetric effects (Table 0-6)
- ARDL specifications: NEER (2,3,3,2); USDMZN (2,3,3,1); ZARMZN (2,3,1,1). Dependent variable: D(LnConsumer Price Index). All variables in logs; constant and trend included.
- Key short-run asymmetric coefficients (with Std. Error and significance markers):
  - ECM(-1): -0.105* (0.016), -0.069* (0.010), -0.083* (0.014). (* 1%).
  - D(CPI(-1)): 0.243* (0.064), 0.179* (0.067), 0.272* (0.066).
  - D(ER-Depreciation): NEER 0.024 (0.025); USDMZN 0.060* (0.021); ZARMZN 0.000 (0.021). (* 1%).
  - D(ER-Depreciation(-1)): NEER -0.034 (0.027); USDMZN 0.085* (0.022); ZARMZN 0.058* (0.021).
  - D(ER-Depreciation(-2)): NEER -0.082* (0.027); USDMZN 0.062* (0.023); ZARMZN 0.050** (0.021). (* 1%, ** 5%).
  - D(ER-Appreciation): NEER 0.026 (0.033); USDMZN -0.033* (0.042); ZARMZN -0.042*** (0.022). (*** 10%).
  - D(PF) and lags: mixed signs and some significance (e.g., D(PF(-1)) -0.019*** (0.011) in NEER model).
- Asymmetry test: F-stat(p-value) = 1.834 (0.177); Ho = No asymmetry (not rejected).
- Model diagnostics: R-squared NEER 0.529; USDMZN 0.529; ZARMZN 0.477. Durbin-Watson: 1.989, 1.998, 1.976 respectively.
- Source: Researchers' computation based on Eviews 11.

### Estimated long-run elasticities accounting for Financial Crisis effects (Table 0-7)
- ARDL exchange rate models with Financial Crisis2008 shift dummy (value one from September 2009, zero before):
  - NEER (2,2,0,0,2): Exchange Rate coeficient 0.488 (p-value 0.000); Exchange Rate*Financial Crisis2008 -0.092 (0.210); Financial Crisis2008 -0.444 (0.212); Import Prices 0.108 (0.000); Trend 0.004 (0.000); Constant 5.960 (0.000).
  - USDMZN (2,2,0,0,1): Exchange Rate 0.661 (0.004); Exchange Rate*Financial Crisis2008 -0.110 (0.618); Financial Crisis2008 0.510 (0.590); Import Prices 0.319 (0.000); Trend 0.002 (0.002); Constant -0.004 (0.937).
  - ZARMZN (2,2,0,0,1): Exchange Rate 0.226 (0.001); Exchange Rate*Financial Crisis2008 0.092 (0.268); Financial Crisis2008 -0.404 (0.248); Import Prices 0.088 (0.005); Trend 0.006 (0.000); Constant 2.556 (0.000).
- Notes: All variables in logs; constant and trend included. Source: Researchers' computation based on Eviews 11.

### Estimated long-run elasticities accounting for Monetary Policy Regime (Table 0-8)
- ARDL exchange rate models with MPReforms17 dummy (value one from April 2017, zero before):
  - NEER model (2,2,2): Exchange Rate 0.439 (0.000); Exchange Rate*MPReforms17 0.151 (0.657); MPReforms17 0.619 (0.673); Import Prices 0.116 (0.000); Trend 0.004 (0.000); Constant 5.685 (0.000).
  - USDMZN model (2,2,1): Exchange Rate 0.588 (0.000); Exchange Rate*MPReforms17 -0.380 (0.706); MPReforms17 1.865 (0.718); Import Prices 0.316 (0.000); Trend 0.003 (0.002); Constant 0.297 (0.649).
  - ZARMZN model (2,2,1): Exchange Rate 0.324 (0.000); Exchange Rate*MPReforms17 0.279 (0.287); MPReforms17 -1.307 (0.265); Import Prices 0.037 (0.344); Trend 0.006 (0.000); Constant 2.357 (0.000).
- Notes: All variables in logs; constant and trend included. Source: Researchers' computation based on Eviews 11.

### Estimated long-run elasticities including Money Supply and Rainfall Index (Table 0-9)
- Multiple ARDL specifications reported (NEER, USDMZN, ZARMZN variants) with inclusion of Money Supply and Rainfall Index.
- Selected coefficients (coeficient and p-value) for Exchange Rate, Import Prices, Rainfall Index, Trend, Constant, Money Supply where applicable:
  - NEER variants: Exchange Rate coefficients 0.392 (0.000), 0.385 (0.000), 0.325 (0.000); Import Prices coefficients 0.141 (0.000), 0.128 (0.000), 0.128 (0.000); Rainfall Index coefficients 0.015, 0.014, 0.053 (p-values vary); Trend coefficients 0.004/0.004/0.005; Constants 5.323, 5.344, 5.023 (p-values 0.000).
  - USDMZN variants: Exchange Rate 0.279 (0.000), 0.413 (0.000), 0.316 (0.000); Import Prices 0.231, 0.272, 0.233 (p-values 0.000); Rainfall Index negative in some specifications (-0.163, -0.107, -0.023) with mixed p-values.
  - ZARMZN variants: Exchange Rate 0.194 (0.000), 0.271 (0.000), 0.281 (0.000); Import Prices 0.308, 0.407, 0.063; Rainfall Index and Money Supply coefficients reported with p-values 0.000 in many cases.
- Notes: All variables in logs; constant and trend included. Dependent variable: Consumer Price Index. Source: Researchers' computation based on Eviews 11.

### Figures (Appendix B highlights)
- Figure 0.1: Research variables plots (in logs) for Consumer Price Index, Nominal Effective Exchange Rate, Exchange Rate USDMZN, Exchange Rate ZARMZN, Import Price Index, Rainfall Index Gap, Real Money Supply Gap (time axis 0–18 shown).
- Figure 0.2: Research variables plots in first log-difference for same variables (visual inspection series over time).
- Figure 0.3: CPI impulse response plots to NEER, USDMZN, ZARMZN and Import Price shocks (horizon up to 50 periods; ±2 S.E. bands shown).
- Figure 0.4: CUSUM and CUSUM Squares plots for NEER ARDL Model, USD ARDL Model, ZAR ARDL Model (5% significance lines shown).
- Source for figures: Researchers' computation based on Eviews 11.

### Appendix C: Previous studies on exchange rate pass-through in Mozambique (summary table)
- Ubide (1997): Error-correction model; sample 1989M1-1996M12; exchange rate variable ZAR/MZM; price variable CPI. Finding: One percent depreciation associated with 0.2 percent increase in inflation; effect lasts about 10 months and dies after 20 months.
- Omar (2003): Error-correction model; sample 1993M1-2001M12; ZAR/MZM; CPI. Finding: One percent depreciation associated with 0.74 percent increase in inflation.
- Cirera and Nhate (2006): OLS controlling for product-specific heterogeneity; sample 2000M1-2005M12 (provinces panel); ZAR/MZM; selected import products. Finding: Degree of ERPT is high (50 to 75 percent depending on specification). ERPT tends to be symmetric.
- Vicente (2007): Co-integrated VAR; sample 2001M1-2006M12; ZAR/MZM; CPI. Finding: One percent exchange rate depreciation associated with 0.15 percent increase in the price level.

*Source: wpiea2021132-print-pdf - APPENDIX A: TABLES (Researchers' computation based on Eviews 11).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021132-print-pdf.pdf_