## wpiea2024186-print-pdf ## Source details **Canonical URL:** [wpiea2024186-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024186-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024186-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024186-print-pdf.pdf.json) --- ### Methodology - Objective and approach: - Use Jorda (2005)’s local projection method to trace impulse response functions of effective exchange rates (nominal and real) and consumer price index (CPI) to natural disaster shocks. - Estimation horizon h runs from 0 (contemporaneous response) up to 24 months ahead to capture high-frequency dynamics. - Baseline regression (symbols preserved as in source): - s_{i,t+h} − s_{i,t−1} = c + Σ_{j=1}^{p} α_{j} Δs_{i,t−j} + β_{1}^{h} D_{i,t} + β_{2}^{h} perc_{i,t} + β_{3}^{h} D_{i,t} * perc_{i,t} + α_{i} + δ_{t,h} + ε_{i,t+h} - Notation and key modeling choices: - i, t, h denote country, month, and horizon respectively. - s_{i,t} is the log form of the dependent variables: NEER, REER, and CPI. - Δs_{i,t−j} denotes lags of changes of dependent variables up to j lags (baseline j = 2). - D_{i,t} is a monthly dummy for natural disasters (1 if a disaster occurs in month t; 0 otherwise). - perc_{i,t} denotes the exchange rate regime dummy (1 = pegged regime; 0 = non-pegged). - α_{i} = country fixed effects (time-invariant characteristics). - δ_{t,h} = month fixed effects (global/time-varying effects). - β_{3}^{h} measures the difference in exchange rate reaction to natural disasters by regime (pegged versus non-pegged). - Interpretation example: - At h = 0, depreciation rate (s_{i,t} − s_{i,t−1}) = β_{1}^{0} for non-pegged regimes and (β_{1}^{0} + β_{3}^{0}) for pegged regimes. - Statistically significant β_{3}^{h} implies differential responses across regimes. - Parsimony and identification rationale: - Model is intentionally parsimonious because: - Natural disasters are treated as largely exogenous and random, reducing concerns about omitted variable bias. - Avoid overcontrolling bias where potential controls (growth, interest differentials, reserves, financial variables) might be endogenous to disaster shocks. - Monthly data limitations for many macro variables would reduce sample coverage if included. - Global influences are captured by month fixed effects. - Robustness analyses expand the baseline to include lagged control variables. - Robustness specification with controls: - s_{i,t+h} − s_{i,t−1} = c + Σ_{j=1}^{p} α_{j} Δs_{i,t−j} + β_{1} D_{i,t} + β_{2} perc_{i,t} + β_{3} D_{i,t} * perc_{i,t} + Σ_{j=1}^{q} γ_{j} X_{i,t−j} + α_{i} + δ_{t,h} + ε_{i,t+h} - Controls X_{i,t} include lagged interest rate differentials, changes in capital account openness, and changes in international reserves. - Lags for controls used up to q lags with baseline q = 1 to reduce potential correlation with contemporaneous disasters. - Key methodological parameters and choices (bullet summary): - Estimation method: Local projections (Jorda, 2005). - Horizons analyzed: h = 0 to 24 months. - Baseline lags of dependent variable: j = 2. - Robustness lags for control variables: q = 1. - Disaster sample selection: - Full EM‑DAT 1970–2019: 4,408 disasters. - Threshold: total damage cost ≥ 0.1 percent GDP → 1,132 disasters retained. - Climatic vs geological split among retained disasters: 83.0 percent climatic. - Disasters lasting > 12 months removed: 16 disasters. - Country coverage in final dataset: 177 countries, monthly data 1970–2019. - Regime classification: Ilzetzki et al. (2019) coarse categories 1–6; pegged = 1–2; non-pegged = 3–6; category 5 removed. ### Data - Sample period and coverage: - Time period: 1970 to end of 2019 (data after 2019 dropped to avoid COVID-induced volatility). - Final dataset: 177 countries with monthly data from 1970 to 2019. - Darvas NEER/REER dataset provides monthly NEER and REER for 178 countries historically; baseline regressions restrict to months and countries with both NEER and REER available. - Natural disasters (EM-DAT / CRED): - Data source: AFDA/CRED International Emergency Disasters Database (EM‑DAT). - 1970–2019: 4,408 disasters worldwide reported in the dataset. - Duration statistics: - 82.9 percent of disasters lasted within a month. - 99.5 percent ended within a year. - Economic relevance threshold: - Retain disasters with total damage cost equal or larger than 0.1 percent GDP. - Total damage cost definition: physical damages to property, crops, and livestock (EM‑DAT); excludes human loss and reconstruction cost. - Resulting sample of disasters: - 1,132 disasters meet the 0.1 percent GDP threshold. - 83.0 percent of these 1,132 disasters are climatic events (drought, flood, storm, extreme temperature). - Remaining disasters classified as geological events (volcanic activity, earthquake, landslide, wildfire). - Remove disasters lasting more than 12 months: 16 disasters dropped (entire prolonged disaster periods removed to avoid different transmission mechanisms and contamination of control group). - Construction of monthly disaster variable: - Monthly dummy = 1 for each month of a disaster period (for disasters lasting up to 12 months); 0 otherwise. - Exchange rates: - NEER and REER data source: Darvas (2012, 2021) dataset. - Darvas dataset provides monthly NEER and REER for 178 countries, back to 1960 in some cases, and is highly correlated with IMF IFS series. - IMF IFS offers monthly effective exchange rate data with more limited country coverage and only nominal EER for some periods. - Baseline regressions restrict to observations with both NEER and REER. - NEER comparison statistics (Darvas vs IMF IFS) — preserved values from Table 1: - Country group: AEs - lifs_neer: No. of observations = 14,821; Mean = 4.63; Standard Deviation = 0.51; Min = 2.69; Max = 12.14; Correlation = 0.93 - lneer: No. of observations = 20,241; Mean = 4.63; Standard Deviation = 0.84; Min = 3.11; Max = 13.74 - Country group: EMDEs - lifs_neer: No. of observations = 31,836; Mean = 5.33; Standard Deviation = 2.24; Min = -0.89; Max = 29.79; Correlation = 0.89 - lneer: No. of observations = 78,482; Mean = 5.84; Standard Deviation = 3.54; Min = -14.35; Max = 37.96 - Note: lifs_neer and lneer denote natural logarithms of NEER from IMF IFS dataset and Darvas dataset, respectively. - Exchange rate regime classification: - Regime source: Ilzetzki et al. (2019) dataset (refines Reinhart and Rogoff (2004) classification; covers multiple currency poles; 6 coarse categories and 15 fine categories). - Construction of regime dummy: - Pegged regime dummy = 1 for coarse classifications 1 and 2 (no separate legal tender up to de-facto crawling band within +/-2 percent). - Non-pegged regime dummy = 0 for coarse classifications 3 to 6. - Countries classified as coarse category 5 (freely falling) are removed following Ramcharan (2007) due to very high inflation (over 40 percent per annum). - Ilzetzki et al. (2019) dataset ends in 2019; period after 2019 excluded. - Other variables and sources: - CPI data: Ha et al., (2021) global inflation database. - Capital account openness: Chinn and Ito (2006). - Policy rates: Haver Analytics. - Interest rate differentials: Calculated as the difference between home country policy rate and base country policy rate as specified in Shambaugh (2004). - Monthly international reserves: IMF International Reserves and Foreign Currency Liquidity (IRFCL). - Appendix Table 1 contains descriptive statistics for all variables (not reproduced here). ### Empirical Results - 3.1 All disasters — Nominal exchange rates: - Baseline regression (1) cumulative effects on NEER reported in Figure 2. - Non-pegged regimes (dash line): nominal exchange rates depreciate by 4 percents over two years (all countries). - Pegged regimes (solid line): nominal exchange rate appreciation of 1 percent over two years (all countries). - Interaction coefficients statistically significant at 5 percent level for the last 9 periods, indicating a stark difference by regime. - By country group: - EMDEs: non-peggers witness a statistically significant nominal depreciation of 6 percents two years after a disaster shock; peggers nominal exchange rates appreciate 2 percents toward 24 months after a shock. Interaction terms strongly significant from the 11th month after the shock. - AEs: magnitudes considerably smaller; neither estimated coefficients nor interaction terms are statistically significant. - 3.1 All disasters — Real exchange rates: - REER responses to natural disasters are insignificant for both regimes for all countries (Figure 3, Panel A). - EMDEs (Figure 3, Panel B): REER significantly depreciates 5 percents in non-pegged regimes and mildly appreciates 1 percent in pegged regimes. - AEs (Figure 3, Panel C): mild (statistically insignificant) real appreciation of 2 percents for non-pegged regimes; muted REER impact for pegged regimes. Interaction terms insignificant — REER responses regime-independent in AEs. - Consistency with literature: developing countries with non-pegged regimes show weaker currencies after weather shocks; fixed regimes in EMDEs may be disadvantaged as both real and nominal exchange rates did not depreciate after shocks. - Concern: large depreciations in EMDEs could lead to higher domestic inflation amid reconstruction demand and/or limited supply due to infrastructure destruction. - CPI/inflation impulse responses (Figure 4): price stickiness present; inflation largely unchanged in most cases, especially for flexible regimes. Depreciation in flexible regimes does not trigger higher inflation in the short run. - 3.2 Disaster types (climatic vs geological): - Climatic events: - All countries: nominal rates depreciate 5 percents two years after shock for non-pegged regimes. - EMDEs: nominal depreciation of 7 percent in non-pegged regimes; pegged regimes show 2 percent nominal appreciation. - AEs: insignificant nominal change in non-pegged regimes; pegged regimes show 2 percent nominal depreciation. - Interaction terms significant in all country groups — meaningful regime differences for climatic disasters. - REER: non-pegged regimes show large real depreciation of 4 percent for all countries and 7 percent for EMDEs; AEs show mild real appreciation of 2 percent. - Summary: non-pegged EMDEs experience depreciation in both nominal and real terms after climatic shocks; pegged EMDEs do not observe highly significant impacts; impacts larger in EMDEs than AEs. - Geological events: - Impact insignificant in most cases. - None of the interaction terms significant — impact similar (and insignificant) across regimes and country groups. - 3.3 Country characteristics: - Groups examined: low-income countries (LICs), small-island EMDEs, larger middle-income countries (EMDEs excluding LICs and small islands). - Small islands (non-pegged): - NEER: immediate depreciation of about 10 percent 8 months after a natural disaster (largest among groups). - REER: more muted changes, consistent with high pass-through from nominal to domestic prices. - Larger middle-income countries (non-pegged): - NEER: smaller and more delayed depreciation; does not reach 10 percent until end of second year. - REER: shows real depreciation. - Low-income countries: - NEER: nominal exchange rate appreciates slightly. - Agriculture-intensive countries: - Defined using WDI agriculture, forestry, and fishing value added as percent of GDP: country labeled agriculture-intensive if share larger than the 75 percentiles in that year; time-invariant dummy equals 1 if country has at least 10 years labeled agriculture intensive during 1970-2019; advanced and upper middle-income countries excluded. - Finding: agriculture intensity seems not to influence exchange rate response. Insignificant response in flexible regimes and mild nominal appreciation in pegged regimes. - Tourism-dependent small islands: - Defined using UNWTO inbound tourism expenditure over GDP: country labeled tourism-dependent if among top 25 percent largest inbound expenditure (percent of GDP) in a year; time-invariant dummy equals 1 if country labeled tourism country for at least 5 years during 1995-2019; large countries excluded; final list: 19 small islands. - Findings: tourism-dependent small islands with flexible regimes see large nominal depreciation after disasters (similar to general small islands). - REER for tourism-dependent small islands appreciates and could go up to 7 percent two years after a disaster — much higher than small islands EMDEs. Possible explanations: higher exchange rate pass-through or limited ability to contain inflation due to concentrated tourism-dependent domestic industry. - 3.4 Larger versus smaller natural disasters: - Large disasters defined as those with economic cost in the top 25 percentiles or above; regressions include separate dummies for large and smaller disasters; baseline periods without disasters. - NEER (Figure 9): after large disasters, NEER of non-pegged regimes depreciates up to 15 percent over 24 months; NEER fluctuations much more muted after smaller disasters. - REER (Figure 10): after large disasters, REER of non-pegged regimes depreciates up to 10 percent over 24 months; REER fluctuations more muted after smaller disasters. - Conclusion: larger natural disasters have outsized effects on exchange rates; escalating disaster scale with climate change could be concerning. ### Robustness checks (summary) - Additional controls: - Pre-disaster lags of change in capital account openness, interest rate differential (home vs base country), and changes in international reserves. - Rationale: include pre-disaster lags to avoid multicollinearity with contemporaneous policy responses to disasters. - Data coverage: - 36,874 observations (country-month) with real interest rate differential data versus 86,121 observations with REER data (Appendix Table A1). - International reserves data available for about 20 percent of baseline observations. - Procedure: - Restrict baseline sample to match coverage of robustness checks for comparability; present baseline-with-same-coverage vs robustness model results (Figures 11 and 12). - Findings: - Results from robustness models (including additional controls) and baseline models with same sample coverage are similar to each other but differ from unrestricted-sample baseline — evidence of sample selection bias driven by data availability. - Estimation with interest rate data may suffer from sample selection bias because countries without interest rate data are likely low-income countries. - Results for AEs are similar to baseline; results for EMDEs differ considerably when sample restricted, corroborating sample selection concerns. - Overall, parsimonious baseline model is robust in terms of bias estimation when sample coverage is held constant: including other control variables does not materially alter results given the same sample. ### Appendix data and descriptive statistics (selected figures from Appendix Table 1) - NEER: No. of observation 98,723; Mean 5.59; Standard Deviation 3.22; Min -14.35; Max 37.97. Source: Darvas (2021). - REER: No. of observation 86,121; Mean 4.67; Standard Deviation 0.44; Min -1.57; Max 15.49. Source: Darvas (2021). - CPI: No. of observation 76,938; Mean 3.47; Standard Deviation 2.14; Min -17.33; Max 7.40. Source: Ha et. al (2021). - Nominal interest differential: No. of observation 43,409; Mean 4.93; Standard Deviation 7.95; Min -17.00; Max 59.91. Source: Haver Analytics. - Real interest differential: No. of observation 36,874; Mean 0.89; Standard Deviation 6.69; Min -99.03; Max 50.68. Source: Authors’ calculations. - Capital account openness: No. of observation 87,564; Mean 0.00; Standard Deviation 1.53; Min -1.93; Max 2.30. Source: Chinn and Ito (2006). - International reserves: No. of observation 15,742; Mean 10.00; Standard Deviation 1.75; Min 5.37; Max 15.19. Source: IMF. - Note: NEER, REER, CPI, and Reserves are in natural logarithm. Nominal interest differential is calculated as domestic policy rates minus policy rates of the base country. Real interest differential is calculated as real domestic policy rates (nominal rates minus annual inflation) minus real policy rates of the base country. ### Appendix Figure 2 — Figure description (selected) - Cumulated response of NEER to natural disaster shocks in countries with fixed (green line) and flexible (dash purple line) exchange rate regimes. - Green (purple) shaded area shows confident intervals at 90 percent level for fixed (flexible) regimes. - An increase (decrease) of NEER implies an exchange rate appreciation (depreciation). - Vertical axis label: percent with tick values: -15, -10, -5, 0, 5. - Horizontal axis label: Months with tick sequences including: 0 2 4 6 8 10 12 14 16 18 20 22 24. - Panels: (A) All Countries; (B) Advanced Countries; (C) Emerging Market and Developing Countries. - Disaster types shown: Climatic events; Geological events. ### Concluding Remarks - Key empirical findings: - The paper investigates the impact of natural disaster on exchange rate movements using a local projection model for a monthly dataset of 177 countries during the 1970-2019 period. - Main findings are three-fold: - Exchange rate movements are more sensitive to natural shocks in EMDEs than in AEs. Empirical tests largely point to an insignificant reaction of both nominal and real exchange rates to disasters in AEs, while exchange rates in EMDEs react significantly to disaster shocks. - Exchange rate reactions to natural disaster shocks depend on exchange rate regimes. In AEs the choice of regime does not play a role in how exchange rates react to natural shocks; in EMDEs regimes play a significant role: - The exchange rates could depreciate up to 6 percent two years after the shock in float regimes. - Exchange rates appreciate mildly in fixed regimes. - Exchange rates of small island economies are more sensitive to natural disasters, and larger disasters have stronger impacts on exchange rate fluctuations in general. - Policy implications and recommendations: - EMDEs should be prepared for possible exchange rate fluctuations after natural disasters hit. - A real depreciation is generally agreed to help countries recover faster after shocks via expenditure switching; therefore, a floating exchange rate regime could be more desirable (also see Elekdag and Tuuli 2023). - Price stickiness, as confirmed in the paper, would help lower the risk of higher imported inflation due to exchange rate depreciation in the short run. - EMDEs that float their exchange rates should be aware of feasible depreciation pressure, which could be particularly significant during large natural disasters. - Policymakers should prepare for possible policy challenges arising in terms of debt or financial stability, especially when financial vulnerabilities such as dollarized debt or currency mismatch are present (also see Reinhart and Rogoff, 2011; Asonuma, 2016). - Disaster-prone EMDEs intending to open their capital account and move to a more flexible exchange rate should be mindful about the additional exchange rate fluctuation coming from natural shocks. - The results could be helpful when building scenario-based macro-frameworks in disaster-prone countries, taking into account disaster-led exchange rate fluctuation. - Limitations and directions for future research: - Investigate possible sample bias when incorporating interest rates or other control variables. - Explore mechanisms behind exchange rate movements after disasters. - Study heterogeneity among prerequisite macroeconomic conditions and policy reactions post-disaster (such as inflation, interest rate, international reserves). *Source: IMF Working Paper — "Currencies in Turbulence: Exploring the Impact of Natural Disasters on Exchange Rates" (Methodology, Data, Empirical Results, and Concluding Remarks sections from wpiea2024186-print-pdf).* ### 2.1 Methodology _________________________________________________________________________________ 4 ### 2.1 Methodology ### Objective and approach - Use Jorda (2005)’s local projection method to trace impulse response functions of effective exchange rates (nominal and real) and consumer price index (CPI) to natural disaster shocks. - Estimation horizon h runs from 0 (contemporaneous response) up to 24 months ahead to capture high-frequency dynamics. ### Baseline regression (symbols preserved as in source) - Baseline specification: - s_{i,t+h} − s_{i,t−1} = c + Σ_{j=1}^{p} α_{j} Δs_{i,t−j} + β_{1}^{h} D_{i,t} + β_{2}^{h} perc_{i,t} + β_{3}^{h} D_{i,t} * perc_{i,t} + α_{i} + δ_{t,h} + ε_{i,t+h} - Notation and key modeling choices: - i, t, h denote country, month, and horizon respectively. - s_{i,t} is the log form of the dependent variables: NEER, REER, and CPI. - Δs_{i,t−j} denotes lags of changes of dependent variables up to j lags (baseline j = 2). - D_{i,t} is a monthly dummy for natural disasters (1 if a disaster occurs in month t; 0 otherwise). - perc_{i,t} denotes the exchange rate regime dummy (1 = pegged regime; 0 = non-pegged). - α_{i} = country fixed effects (time-invariant characteristics). - δ_{t,h} = month fixed effects (global/time-varying effects). - β_{3}^{h} measures the difference in exchange rate reaction to natural disasters by regime (pegged versus non-pegged). - Interpretation example: - At h = 0, depreciation rate (s_{i,t} − s_{i,t−1}) = β_{1}^{0} for non-pegged regimes and (β_{1}^{0} + β_{3}^{0}) for pegged regimes. - Statistically significant β_{3}^{h} implies differential responses across regimes. ### Parsimony and identification rationale - Model is intentionally parsimonious because: - Natural disasters are treated as largely exogenous and random, reducing concerns about omitted variable bias. - Avoid overcontrolling bias where potential controls (growth, interest differentials, reserves, financial variables) might be endogenous to disaster shocks. - Monthly data limitations for many macro variables would reduce sample coverage if included. - Global influences are captured by month fixed effects. - Robustness analyses expand the baseline to include lagged control variables. ### Robustness specification with controls - Extended specification: - s_{i,t+h} − s_{i,t−1} = c + Σ_{j=1}^{p} α_{j} Δs_{i,t−j} + β_{1} D_{i,t} + β_{2} perc_{i,t} + β_{3} D_{i,t} * perc_{i,t} + Σ_{j=1}^{q} γ_{j} X_{i,t−j} + α_{i} + δ_{t,h} + ε_{i,t+h} - Controls X_{i,t} include lagged interest rate differentials, changes in capital account openness, and changes in international reserves. - Lags for controls used up to q lags with baseline q = 1 to reduce potential correlation with contemporaneous disasters. --- ### 2.2 Data ### Sample period and coverage - Time period: 1970 to end of 2019 (data after 2019 dropped to avoid COVID-induced volatility). - Final dataset: 177 countries with monthly data from 1970 to 2019. - Darvas NEER/REER dataset provides monthly NEER and REER for 178 countries historically; baseline regressions restrict to months and countries with both NEER and REER available. ### Natural disasters (EM-DAT / CRED) - Data source: AFDA/CRED International Emergency Disasters Database (EM‑DAT). - 1970–2019: 4,408 disasters worldwide reported in the dataset. - Duration statistics: - 82.9 percent of disasters lasted within a month. - 99.5 percent ended within a year. - Economic relevance threshold: - Retain disasters with total damage cost equal or larger than 0.1 percent GDP. - Total damage cost definition: physical damages to property, crops, and livestock (EM‑DAT); excludes human loss and reconstruction cost. - Resulting sample of disasters: - 1,132 disasters meet the 0.1 percent GDP threshold. - 83.0 percent of these 1,132 disasters are climatic events (drought, flood, storm, extreme temperature). - Remaining disasters classified as geological events (volcanic activity, earthquake, landslide, wildfire). - Remove disasters lasting more than 12 months: 16 disasters dropped (entire prolonged disaster periods removed to avoid different transmission mechanisms and contamination of control group). - Construction of monthly disaster variable: - Monthly dummy = 1 for each month of a disaster period (for disasters lasting up to 12 months); 0 otherwise. ### Exchange rates - NEER and REER data source: Darvas (2012, 2021) dataset. - Darvas dataset provides monthly NEER and REER for 178 countries, back to 1960 in some cases, and is highly correlated with IMF IFS series. - IMF IFS offers monthly effective exchange rate data with more limited country coverage and only nominal EER for some periods. - Baseline regressions restrict to observations with both NEER and REER. ### NEER comparison statistics (Darvas vs IMF IFS) — preserved values from Table 1 - Country group: AEs - lifs_neer: No. of observations = 14,821; Mean = 4.63; Standard Deviation = 0.51; Min = 2.69; Max = 12.14; Correlation = 0.93 - lneer: No. of observations = 20,241; Mean = 4.63; Standard Deviation = 0.84; Min = 3.11; Max = 13.74 - Country group: EMDEs - lifs_neer: No. of observations = 31,836; Mean = 5.33; Standard Deviation = 2.24; Min = -0.89; Max = 29.79; Correlation = 0.89 - lneer: No. of observations = 78,482; Mean = 5.84; Standard Deviation = 3.54; Min = -14.35; Max = 37.96 - Note: lifs_neer and lneer denote natural logarithms of NEER from IMF IFS dataset and Darvas dataset, respectively. ### Exchange rate regime classification - Regime source: Ilzetzki et al. (2019) dataset (refines Reinhart and Rogoff (2004) classification; covers multiple currency poles; 6 coarse categories and 15 fine categories). - Construction of regime dummy: - Pegged regime dummy = 1 for coarse classifications 1 and 2 (no separate legal tender up to de-facto crawling band within +/-2 percent). - Non-pegged regime dummy = 0 for coarse classifications 3 to 6. - Countries classified as coarse category 5 (freely falling) are removed following Ramcharan (2007) due to very high inflation (over 40 percent per annum). - Ilzetzki et al. (2019) dataset ends in 2019; period after 2019 excluded. ### Other variables and sources - CPI data: Ha et al., (2021) global inflation database. - Capital account openness: Chinn and Ito (2006). - Policy rates: Haver Analytics. - Interest rate differentials: Calculated as the difference between home country policy rate and base country policy rate as specified in Shambaugh (2004). - Monthly international reserves: IMF International Reserves and Foreign Currency Liquidity (IRFCL). - Appendix Table 1 contains descriptive statistics for all variables (not reproduced here). --- ### Key methodological parameters and choices (bullet summary) - Estimation method: Local projections (Jorda, 2005). - Horizons analyzed: h = 0 to 24 months. - Baseline lags of dependent variable: j = 2. - Robustness lags for control variables: q = 1. - Disaster sample selection: - Full EM‑DAT 1970–2019: 4,408 disasters. - Threshold: total damage cost ≥ 0.1 percent GDP → 1,132 disasters retained. - Climatic vs geological split among retained disasters: 83.0 percent climatic. - Disasters lasting > 12 months removed: 16 disasters. - Country coverage in final dataset: 177 countries, monthly data 1970–2019. - Regime classification: Ilzetzki et al. (2019) coarse categories 1–6; pegged = 1–2; non-pegged = 3–6; category 5 removed. *Source: IMF Working Paper — "Currencies in Turbulence: Exploring the Impact of Natural Disasters on Exchange Rates" (Methodology and Data sections).* ### 3. Empirical Results ### 3. Empirical Results ### 3.1 All disasters — Nominal exchange rates - Baseline regression (1) cumulative effects on NEER reported in Figure 2. - Non-pegged regimes (dash line): nominal exchange rates depreciate by 4 percents over two years (all countries). - Pegged regimes (solid line): nominal exchange rate appreciation of 1 percent over two years (all countries). - Interaction coefficients statistically significant at 5 percent level for the last 9 periods, indicating a stark difference by regime. - By country group: - EMDEs: non-peggers witness a statistically significant nominal depreciation of 6 percents two years after a disaster shock; peggers nominal exchange rates appreciate 2 percents toward 24 months after a shock. Interaction terms strongly significant from the 11th month after the shock. - AEs: magnitudes considerably smaller; neither estimated coefficients nor interaction terms are statistically significant. ### 3.1 All disasters — Real exchange rates - REER responses to natural disasters are insignificant for both regimes for all countries (Figure 3, Panel A). - EMDEs (Figure 3, Panel B): REER significantly depreciates 5 percents in non-pegged regimes and mildly appreciates 1 percent in pegged regimes. - AEs (Figure 3, Panel C): mild (statistically insignificant) real appreciation of 2 percents for non-pegged regimes; muted REER impact for pegged regimes. Interaction terms insignificant — REER responses regime-independent in AEs. - Consistency with literature: developing countries with non-pegged regimes show weaker currencies after weather shocks; fixed regimes in EMDEs may be disadvantaged as both real and nominal exchange rates did not depreciate after shocks. - Concern: large depreciations in EMDEs could lead to higher domestic inflation amid reconstruction demand and/or limited supply due to infrastructure destruction. - CPI/inflation impulse responses (Figure 4): price stickiness present; inflation largely unchanged in most cases, especially for flexible regimes. Depreciation in flexible regimes does not trigger higher inflation in the short run. ### 3.2 Disaster types (climatic vs geological) - Disasters classified as climatic (drought, flood, storm, extreme temperature) and geological (volcanic activity, earthquake, landslide, wildfire). - Climatic events: - All countries: nominal rates depreciate 5 percents two years after shock for non-pegged regimes. - EMDEs: nominal depreciation of 7 percent in non-pegged regimes; pegged regimes show 2 percent nominal appreciation. - AEs: insignificant nominal change in non-pegged regimes; pegged regimes show 2 percent nominal depreciation. - Interaction terms significant in all country groups — meaningful regime differences for climatic disasters. - REER: non-pegged regimes show large real depreciation of 4 percent for all countries and 7 percent for EMDEs; AEs show mild real appreciation of 2 percent. - Summary: non-pegged EMDEs experience depreciation in both nominal and real terms after climatic shocks; pegged EMDEs do not observe highly significant impacts; impacts larger in EMDEs than AEs. - Geological events: - Impact insignificant in most cases. - None of the interaction terms significant — impact similar (and insignificant) across regimes and country groups. ### 3.3 Country characteristics - Groups examined: low-income countries (LICs), small-island EMDEs, larger middle-income countries (EMDEs excluding LICs and small islands). - Small islands (non-pegged): - NEER: immediate depreciation of about 10 percent 8 months after a natural disaster (largest among groups). - REER: more muted changes, consistent with high pass-through from nominal to domestic prices. - Larger middle-income countries (non-pegged): - NEER: smaller and more delayed depreciation; does not reach 10 percent until end of second year. - REER: shows real depreciation. - Low-income countries: - NEER: nominal exchange rate appreciates slightly. - Agriculture-intensive countries: - Defined using WDI agriculture, forestry, and fishing value added as percent of GDP: country labeled agriculture-intensive if share larger than the 75 percentiles in that year; time-invariant dummy equals 1 if country has at least 10 years labeled agriculture intensive during 1970-2019; advanced and upper middle-income countries excluded. - Finding: agriculture intensity seems not to influence exchange rate response. Insignificant response in flexible regimes and mild nominal appreciation in pegged regimes. - Tourism-dependent small islands: - Defined using UNWTO inbound tourism expenditure over GDP: country labeled tourism-dependent if among top 25 percent largest inbound expenditure (percent of GDP) in a year; time-invariant dummy equals 1 if country labeled tourism country for at least 5 years during 1995-2019; large countries excluded; final list: 19 small islands. - Findings: tourism-dependent small islands with flexible regimes see large nominal depreciation after disasters (similar to general small islands). - REER for tourism-dependent small islands appreciates and could go up to 7 percent two years after a disaster — much higher than small islands EMDEs. Possible explanations: higher exchange rate pass-through or limited ability to contain inflation due to concentrated tourism-dependent domestic industry. ### 3.4 Larger versus smaller natural disasters - Large disasters defined as those with economic cost in the top 25 percentiles or above; regressions include separate dummies for large and smaller disasters; baseline periods without disasters. - NEER (Figure 9): after large disasters, NEER of non-pegged regimes depreciates up to 15 percent over 24 months; NEER fluctuations much more muted after smaller disasters. - REER (Figure 10): after large disasters, REER of non-pegged regimes depreciates up to 10 percent over 24 months; REER fluctuations more muted after smaller disasters. - Conclusion: larger natural disasters have outsized effects on exchange rates; escalating disaster scale with climate change could be concerning. ### 4. Robustness Check (summary) - Additional controls: pre-disaster lags of change in capital account openness, interest rate differential (home vs base country), and changes in international reserves. - Rationale: include pre-disaster lags to avoid multicollinearity with contemporaneous policy responses to disasters. - Data coverage: - 36,874 observations (country-month) with real interest rate differential data versus 86,121 observations with REER data (Appendix Table A1). - International reserves data available for about 20 percent of baseline observations. - Procedure: restrict baseline sample to match coverage of robustness checks for comparability; present baseline-with-same-coverage vs robustness model results (Figures 11 and 12). - Findings: - Results from robustness models (including additional controls) and baseline models with same sample coverage are similar to each other but differ from unrestricted-sample baseline — evidence of sample selection bias driven by data availability. - Estimation with interest rate data may suffer from sample selection bias because countries without interest rate data are likely low-income countries. - Results for AEs are similar to baseline; results for EMDEs differ considerably when sample restricted, corroborating sample selection concerns. - Overall, parsimonious baseline model is robust in terms of bias estimation when sample coverage is held constant: including other control variables does not materially alter results given the same sample. *Source: wpiea2024186-print-pdf — 3. Empirical Results* ### 5. Concluding Remarks ### 5. Concluding Remarks ### Key empirical findings - The paper investigates the impact of natural disaster on exchange rate movements using a local projection model for a monthly dataset of 177 countries during the 1970-2019 period. - Main findings are three-fold: - Exchange rate movements are more sensitive to natural shocks in EMDEs than in AEs. Empirical tests largely point to an insignificant reaction of both nominal and real exchange rates to disasters in AEs, while exchange rates in EMDEs react significantly to disaster shocks. - Exchange rate reactions to natural disaster shocks depend on exchange rate regimes. In AEs the choice of regime does not play a role in how exchange rates react to natural shocks; in EMDEs regimes play a significant role: - The exchange rates could depreciate up to 6 percent two years after the shock in float regimes. - Exchange rates appreciate mildly in fixed regimes. - Exchange rates of small island economies are more sensitive to natural disasters, and larger disasters have stronger impacts on exchange rate fluctuations in general. ### Policy implications and recommendations - EMDEs should be prepared for possible exchange rate fluctuations after natural disasters hit. - A real depreciation is generally agreed to help countries recover faster after shocks via expenditure switching; therefore, a floating exchange rate regime could be more desirable (also see Elekdag and Tuuli 2023). - Price stickiness, as confirmed in the paper, would help lower the risk of higher imported inflation due to exchange rate depreciation in the short run. - EMDEs that float their exchange rates should be aware of feasible depreciation pressure, which could be particularly significant during large natural disasters. - Policymakers should prepare for possible policy challenges arising in terms of debt or financial stability, especially when financial vulnerabilities such as dollarized debt or currency mismatch are present (also see Reinhart and Rogoff, 2011; Asonuma, 2016). - Disaster-prone EMDEs intending to open their capital account and move to a more flexible exchange rate should be mindful about the additional exchange rate fluctuation coming from natural shocks. - The results could be helpful when building scenario-based macro-frameworks in disaster-prone countries, taking into account disaster-led exchange rate fluctuation. ### Robustness and heterogeneity insights (from figures and robustness checks) - Cumulative impulse responses of NEER and REER to natural disasters are shown separately for fixed and flexible regimes, with 90 percent confidence intervals for each regime. - Results are presented across subsamples: All Countries; Advanced Countries; Emerging Market and Developing Countries. - Disaster-type heterogeneity: responses are shown for climatic events and geological events. - Country-group heterogeneity: responses are shown for Low Income Countries, Small Islands EMDEs, Large Middle Countries. - Sectoral heterogeneity: responses are shown for Agriculture-Intensive Countries and Tourism-Dependent Small Islands. - Disaster-size heterogeneity: responses are shown for Large Disasters and Small Disasters (Large disasters defined as disasters with economic cost of top 25 percentiles). - Robustness checks include additional control variables: - Robustness Check 1 adds lags of interest rate differential and change in capital account openness. - Robustness Check 2 adds lags of interest rate differential, change in capital account openness, and change in international reserves. - Robustness figures compare the robustness model with baseline regressions on restricted samples excluding observations without interest rate or reserves data. ### Data and descriptive statistics (Appendix Table 1) - NEER: No. of observation 98,723; Mean 5.59; Standard Deviation 3.22; Min -14.35; Max 37.97. Source: Darvas (2021). - REER: No. of observation 86,121; Mean 4.67; Standard Deviation 0.44; Min -1.57; Max 15.49. Source: Darvas (2021). - CPI: No. of observation 76,938; Mean 3.47; Standard Deviation 2.14; Min -17.33; Max 7.40. Source: Ha et. al (2021). - Nominal interest differential: No. of observation 43,409; Mean 4.93; Standard Deviation 7.95; Min -17.00; Max 59.91. Source: Haver Analytics. - Real interest differential: No. of observation 36,874; Mean 0.89; Standard Deviation 6.69; Min -99.03; Max 50.68. Source: Authors’ calculations. - Capital account openness: No. of observation 87,564; Mean 0.00; Standard Deviation 1.53; Min -1.93; Max 2.30. Source: Chinn and Ito (2006). - International reserves: No. of observation 15,742; Mean 10.00; Standard Deviation 1.75; Min 5.37; Max 15.19. Source: IMF. - Note: NEER, REER, CPI, and Reserves are in natural logarithm. Nominal interest differential is calculated as domestic policy rates minus policy rates of the base country. Real interest differential is calculated as real domestic policy rates (nominal rates minus annual inflation) minus real policy rates of the base country. ### Exchange rate regime classification (Appendix Table 2) - Coarse classification and corresponding regime codes are provided, with Peg regimes including codes 1 and 2 (e.g., No separate legal tender; Pre-announced peg or currency board arrangement; Pre-announced crawling peg), and Non-peg regimes including codes 3, 4, 6 (e.g., Managed floating; Freely floating). Source: Ilzetzki et. al (2019). ### Limitations and directions for future research - Data and technical limitations remain in this paper. Future research should investigate: - Possible sample bias when incorporating interest rates or other control variables. - Mechanisms behind exchange rate movements after disasters. - Possible heterogeneity among prerequisite macroeconomic conditions and policy reactions post-disaster (such as inflation, interest rate, international reserves) that could influence the way exchange rates respond to natural shocks. *Source: IMF Working Paper — chapter “5. Concluding Remarks” from the provided PDF content.* ### Appendix Figure 2 : Robustness Check - Cumulative Impulse Responses of NEER to ### Appendix Figure 2 : Robustness Check - Cumulative Impulse Responses of NEER to Disaster Shock by Disaster Type (Alternative Classification)*, 1970-2019 ### Figure description - The figure shows the cumulated response of NEER to natural disaster shocks in countries with fixed (green line) and flexible (dash purple line) exchange rate regimes. - The green (purple) shaded area shows confident intervals at 90 percent level for fixed (flexible) regimes. - An increase (decrease) of NEER implies an exchange rate appreciation (depreciation). - Source: Authors’ calculations. - Note: *See footnote 5 for more details. ### Panel labels and axes (as shown) - Vertical axis label: percent - Tick values shown: -15, -10, -5, 0, 5 - Horizontal axis label: Months - Tick sequences shown in various panels: - 0 2 4 6 8 10 12 14 16 18 20 22 24 - 0 2 4 6 8 10 12 14 16 18 20 22 24 (repeated across panels) - Variations displayed in figure images include sequences printed as: 0246810 12 1416 1820 22 24; 024681012141618202224; 02 4 68101214 16 18202224; 0 24 68 10 121416 18 2022 24 ### Panels reported - (A) All Countries - Axis ticks: percent: -15, -10, -5, 0, 5 - Months ticks: 0 2 4 6 8 10 12 14 16 18 20 22 24 (as displayed) - (B) Advanced Countries - Axis ticks: percent: -15, -10, -5, 0, 5 - Months ticks: 0 2 4 6 8 10 12 14 16 18 20 22 24 (as displayed) - (C) Emerging Market and Developing Countries - Axis ticks: percent: -15, -10, -5, 0, 5 - Months ticks: 0 2 4 6 8 10 12 14 16 18 20 22 24 (as displayed) ### Disaster types shown - Climatic events - Geological events *Source: Authors’ calculations.* --- _Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024186-print-pdf.pdf_