## Section 3: Research Design and Identification (wpiea2024082-print-pdf)

## Source details

**Canonical URL:** [Section 3: Research Design and Identification (wpiea2024082-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024082-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024082-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024082-print-pdf.pdf.json)

---

### Data: sample, coverage, and key variables
- Estimation period: 1990 to 2019.
- Sample: yearly dataset covering 54 emerging and developing countries.
  - 32 upper middle-income countries.
  - 15 lower middle-income countries.
  - 10 low-income countries.
- Regions: the sample covers all the regions in the world.
- Key series included:
  - Output: country-specific real GDP from World Bank’s World Development Indicators (WDI).
  - Real exchange rates.
  - Domestic interest rates: Central Bank Policy Rate (IFS and Haver).
  - Capital flows: gross capital inflows and outflows (IMF IFS); inflows defined as net inflows from foreign residents into the domestic economy; outflows defined as net outflows from domestic residents to the rest of the world; net inflows = inflows − outflows.
  - Sovereign spreads: EMBI spreads (J.P. Morgan’s EMBI global diversified index) and BAA spreads (Federal Reserve Bank of St. Louis FRED).
  - Export and import prices and commodity prices (Px construction detailed below).
- Export price index (Px) construction:
  - Country-specific Px indices denominated in U.S. dollars using sectoral export shares, commodity prices, and disaggregated U.S. PPI data as a proxy for manufacturing prices.
  - Export shares: disaggregated product export values from the MIT Observatory of Economic Complexity.
  - Commodity prices: World Bank’s Commodity Price Data.
  - U.S. PPI for manufacturing categories and U.S. CPI: FRED.
  - Export and import price indices are deflated by the U.S. CPI (real dollar export prices).
  - Methodology follows the IMF Export and Import Prices Manual and Di Pace et al. (2020).
- Coverage and data-source details: presented in Appendix A of the source.

### Stylized facts motivating the analysis
- Px booms and busts strongly associate with EMDE business cycles and with surges and flights of capital flows (Panels (a) and (b) in Figure 1).
- Commodity prices are not the only drivers of emerging market capital flows; global financial conditions matter:
  - BAA spread used as an indicator of global financial conditions.
  - Worldwide financial stress typically triggers capital flights from EMDEs (Panel (c)).
  - Increases in commodity prices and improved global financial conditions are associated with lower EMBI spreads.
  - Most common variation in domestic financing conditions can be explained by a combination of commodity prices and global financial conditions (Panel (d)).
- Relationship between commodity prices and global financial conditions:
  - Commodity prices lack a distinct contemporaneous correlation with global financial conditions (Panel (e)).
  - Shifts in global financial conditions consistently precede changes in commodity prices (Panel (f)).
- Implication: empirical analysis must account for global shocks (commodity price fluctuations and global financial conditions) to avoid biased inference about commodity price effects.

### Identification: goals and challenges
- Research goal: investigate role of world shocks—commodity price fluctuations and changes in global financial conditions—in shaping business and capital flow cycles of EMDEs.
- Transmission channel for commodity price fluctuations: export prices (Px).
  - Median share of primary commodities in total exports across the sample: 0.68.
  - Commodity price volatility is significantly higher than goods prices; commodity fluctuations dominate variation in export prices.
  - Using export prices accounts for time-varying export specialization.
- Small open economy assumption: domestic conditions unlikely to affect global variables for sample countries; global variables may be endogenous to global economic activity.
- Identification challenge:
  - Fluctuations in export prices reflect both commodity-specific idiosyncratic shocks and endogenous responses of international prices to global economic activity.
  - Causal effects of changes in the BAA spread may vary depending on the underlying cause of the shift.
  - Solution: use external instrumental variables to identify causal impacts of Px and BAA spread shifts.

### Commodity-prices instrument: event-based quasi-natural experiment
- Approach: use commodity-market-specific events associated with large, plausibly exogenous price swings as instruments.
- Event selection and construction:
  - Historical documents and newspaper articles used to identify episodes unrelated to major macroeconomic developments with disproportionate impacts on specific commodity prices.
  - Total events identified: 24 (16 positive, 8 negative).
  - When an event directly affected a producing country, that country is excluded from being treated as instrumented for that event (see Appendix B.5 for exclusions).
- Surprise metric for each event:
  - Surprise defined as e_{c,t} = p_{c,t} – E_{t−1}[p_{c,t}], where p_{c,t} is the (log real) price of commodity c at time t and E is expectation.
  - Expectations estimated from a regression conditioning on the commodity’s own price history and the overall (log) level of real commodity price indices (lags) for commodity groups to which the commodity does not belong.
  - Commodity-group controls: agricultural, energy, and metals indices used as appropriate.
  - Surprise isolates unanticipated variation attributable to the event (approach follows Hamilton (2003)).
- Event-time indicator and instrument aggregation:
  - For each event j, q_{j,t} = e_{c,t} for the event year and q_{j,t} = 0 otherwise.
  - Export weight w_{i,c,t} denotes export weight of commodity c for country i at time t.
  - Instrument: z_{i,t} = Σ_j 1(w_{i,c,t−1} > w) w_{i,c,t−1} q_{j,t}, with lower bound w = 2%.
  - Interpretation: unanticipated commodity price variations during events, scaled by country-specific export exposure, generate exogenous fluctuations in country export prices; cross-sectional variation in exposures identifies causal effects (LATE interpretation).
- First-stage evidence:
  - Appendix C Table C1 reports first-stage regressions of ∆p_{x,it} on z_{it} (controls and country fixed effects).
  - Reported F-statistic: above 700 in one specification, indicating z_{it} is not a weak instrument.
  - Additional first-stage statistics in Appendix C: Commodity Events F-statistics shown as 60.157 (No Controls) and 10.68 (With Controls) in Table C1 variants; U.S. Monetary Policy and Global Risk Appetite instruments also report F >10.

### Financial conditions instrument (BAA spread): separation of drivers
- Indicator: BAA spread used as indicator of global financial conditions.
- Distinction: decompose changes in the BAA spread into:
  - U.S. monetary policy spillovers.
  - Global risk perception shifts (global risk appetite).
- Instruments for U.S. monetary policy-driven BAA movements:
  - Construct a common factor from an unbalanced panel of (standardized) monetary policy shock proxies.
  - Individual proxies are cumulated over the calendar year; first principal component estimated per Stock and Watson (2002).
  - Rationale: many proxies minimize contamination by the “central bank informational effect.”
- Instruments for global risk appetite-driven BAA movements:
  - Proxy for uncertainty shocks computed from variations in the price of gold around uncertainty-related events (Piffer and Podstawski (2017)).
  - A measure of U.S. financial uncertainty from Ludvigson et al. (2021).
  - Both are orthogonal to U.S. monetary policy shocks.
- Distinctness and robustness:
  - Regressing the U.S. monetary policy shock on the two risk instruments yields insignificant coefficients under an outlier robust estimation method.
  - Results robust to alternative choices (e.g., wat1% or 0.5%).

### Empirical model: Local Projections with IV
- Estimation approach: Local projections (Jordà, 2005) estimated with instrumental variables (IV).
- Outcome variables:
  - log of GDP (detrended)
  - log of export prices (detrended)
  - log of the policy rate
  - log of the EMBI spread
  - log of the real exchange rate (detrended)
  - capital inflows and outflows in terms of trend GDP
  - foreign exchange reserves in terms of trend GDP
- Model specification (for h = 0,1,...,H):
  - y_{i,t+h} − y_{i,t−1} = μ^h_i + f_{i,t} β^h + (x_{i,t} − \bar{x}_i) γ^h_0 + f_{i,t} (x_{i,t} − \bar{x}_i) θ^h_x + ω_{i,t+h}
  - Dependent variable: cumulative change from t−1 to t+h.
  - f_{i,t}: intervention/treatment (change in export prices or in the BAA spread).
  - μ^h_i: country fixed effects.
  - x_{i,t}: covariates in deviation from mean; interaction terms f_{i,t} (x_{i,t} − \bar{x}_i) included to allow heterogeneity in causal effects per Cloyne et al. (2023).
- Controls used (x_{i,t}): two lags of real GDP growth, Px growth, the BAA spread, net capital inflows, and the lag of the dependent variable.
- Interpretation: β^h capture average dynamic causal effect; with treatment instrumented, IRFs are interpretable as LATE.

### Main empirical findings: commodity-specific shocks (one standard deviation increase in Px driven by commodity-specific shocks)
- Baseline average dynamic responses:
  - Domestic GDP: steady increase.
  - Real exchange rate: appreciation.
  - Policy rate: gradual decrease over time (suggesting a moderately pro-cyclical domestic monetary policy response).
  - EMBI spread: mild decrease.
  - Capital outflows: increase, mainly driven by other investment flows (mostly bank-related).
  - Foreign exchange reserves: accumulation.
  - Direct investment flows (both inflows and outflows): small positive response.
  - Capital inflows: more muted response.
  - Net capital inflows (inflows − outflows): decline.
- Additional notes:
  - These commodity-specific shocks are idiosyncratic commodity events.
  - IRFs exhibit mean reversion at longer horizons; confidence bands become considerably larger after four periods due to relatively short samples for many countries.
- Implication:
  - The negative association between the EMBI spread and Px is not solely attributable to commodity price shocks; other factors (e.g., GFC or shifts in global demand) also influence this relationship.
  - Commodity shocks are associated with significant net outflows rather than simultaneous increases in inflows and outflows (i.e., do not always match a “when it rains, it pours” pattern).

### Main empirical findings: decline in the BAA spread (common effects and differences by driver)
- Common characteristics of a reduction in the BAA spread (regardless of driver):
  - Increase in GDP and Px.
  - Real exchange rate: appreciation (counterpart of weakening U.S. dollar).
  - EMDE central banks: increase interest rates in reaction to easier global financial conditions.
  - EMBI spread: immediate decrease despite higher domestic policy rates.
  - Negative correlation between EMBI spread and Px consistent with prior studies.
- Differences by driver (U.S. monetary policy shock vs global risk appetite shock):
  1. Temporal profile:
     - Global risk appetite shock: more temporary; IRFs display spike-like patterns in GDP, Px, policy rate, and real exchange rate.
     - U.S. monetary policy shock: humped-shaped response in Px and output, indicating more persistent influence through global commodity prices.
  2. EMBI spread dynamics:
     - Global risk appetite shock: initial reduction in EMBI spread is very large but quickly reverses.
     - U.S. monetary policy shock: decline in EMBI spread is more persistent and builds over time.
  3. Capital flow responses:
     - Both shocks: increases in both inflows and outflows.
     - Global risk appetite shock: rise in inflows and outflows almost identical → net inflows relatively stable; large surge in portfolio inflows and outflows.
     - U.S. monetary policy shock: net inflows tend to decrease; muted portfolio flow response.
  - Global risk shock pattern aligns with “when it rains, it pours” dynamics (Kaminsky et al., 2004).
- Implications:
  - U.S. monetary policy plays a pivotal role in shaping global investor risk perceptions and impacts capital flows and credit spreads.
  - Identifying the origin of BAA spread fluctuations matters; treating a generic “BAA spread shock” can mask important differences.
  - Focus on gross capital movements (inflows and outflows separately) is important; both inflows and outflows increase following a reduction in global risk appetite.

### Exogenous versus endogenous commodity-price movements (OLS vs IV)
- OLS yields a mean response combining effects of both exogenous and endogenous shifts in export prices; IV isolates commodity-specific exogenous effects exploiting cross-country exposure.
- Comparative patterns:
  - OLS: exchange rate appreciation substantially larger and more persistent.
  - OLS: policy rate exhibits an immediate increase.
  - Reserves and capital flows: OLS suggests persistent surge in capital inflows; IV (LATE) depicts a more muted reaction.
- Caution: OLS may misattribute capital flow fluctuations to commodity price shocks because OLS conflates commodity shocks with other global shocks (e.g., U.S. monetary and risk appetite shocks).

### Channels of transmission and heterogeneity: methodology and findings
- Two channels examined:
  - Financial channel: operates through endogenous response of country spreads (EMBI spread).
  - Commodity channel: operates through changes in export prices (Px).
- Methodology:
  - Extend LP with interaction terms capturing country-specific sensitivity of EMBI spread or Px to the shock.
  - Use Kitagawa-Blinder-Oaxaca (KBO) decomposition to separate direct effect (β^h) and indirect effect (Θ^h_i θ^h_f) mediated by movements in EMBI spread or Px.
  - Construct mediating variable Θ^h_i by estimating country-specific responses e_{Θ^h_i} from a parsimonious LP; normalize against cross-country average to obtain Θ^h_i.
  - Counterfactuals: E[GDP_i,t+h − GDP_i,t−1 | f_t = 1, Θ^h = κ] − E[GDP_i,t+h − GDP_i,t−1 | f_t = 0] = β^h + θ^h_f × κ, with κ set to −1.5σ_{Θ,h} or 1.5σ_{Θ,h}.
- Key empirical findings:
  - GDP response to surge in Px (commodity-specific shocks):
    - Expansionary effects larger in countries with significant decline in EMBI spread; muted/shorter-lived where EMBI spread increases.
    - Differences are statistically different at a 10% level after the second year.
    - Economic magnitudes:
      - With a 4.5% decrease in the EMBI spread (approximately 20 basis points per period, consistent with the LATE reported in Figure 2), GDP increases by 0.18% for every percentage point rise in Px.
      - Counterfactual: if EMBI spread drops 1.5 standard deviations more than the average (about 47% or 210 basis points), an equivalent rise in export prices leads to GDP expansion of around 0.40% for every percentage point increase in Px.
      - Median value of average EMBI spreads for sample countries is around 450 basis points (used to convert percent decrease in EMBI spread into basis points).
    - Interpretation: financial channel (borrowing costs) operates alongside terms-of-trade channel; countries with institutions that insulate domestic financing from global shocks fare better.
  - GDP response to a fall in the BAA spread mediated by EMBI spread:
    - Expansion of domestic GDP greater when fall in EMBI spread is larger.
    - Effect statistically significant and economically meaningful only when BAA spread reduction is driven by global risk appetite.
  - GDP response to a fall in the BAA spread mediated by Px:
    - Endogenous response of Px significantly shapes and amplifies transmission of U.S. monetary policy to EMDEs; indirect effect statistically significant.
    - A modest increase in Px yields a restrained expansionary effect of a decline in the BAA spread linked to accommodative U.S. monetary policy.
    - Responses of GDP are less sensitive to Px movements when BAA spread reduction is driven by global risk appetite.
    - In countries with robust Px reaction, peak effect of similar-sized fall in BAA spread is nearly twice as large when global financial conditions are eased by U.S. monetary policy rather than by global risk appetite.
- Synthesis:
  - U.S. monetary policy shocks are amplified by the endogenous reaction of export prices.
  - Transmission of shifts in global risk appetite is strongly tied to endogenous domestic borrowing costs (EMBI spread).
  - Policy responses need to account for these different transmission mechanisms.

### Robustness checks and supplementary evidence
- Omitting events:
  - Excluding crude oil event of 1998 and all events in 2008: results remain robust.
  - Instrument constructed with subset of 15 events from Di Pace et al. (2020): results remain robust.
- Omitting countries:
  - Excluding one country at a time does not change results; no single country drives findings disproportionately.
- Subsample analysis (post-2000):
  - Effects qualitatively similar when sample starts in 2000.
  - Effects on Px and the real exchange rate are higher in the post-2000 subsample.
  - In response to a looser U.S. monetary policy, GDP impact is smaller but responses of the real exchange rate, policy rate, EMBI spread, reserves, capital inflows, and outflows are amplified in the 2000s subsample.
  - Post-2000, response of Px to a global risk appetite shock is notably larger; exchange rate shows larger appreciation; reserves, capital inflows and outflows exhibit larger responses.
- Energy-commodity-specific events:
  - Energy commodities have distinct attributes; effects of an increase in Px driven by energy commodity-specific shocks are not very different from the baseline.
  - Cross-sectional variation for non-energy commodities is limited, constraining identification for those commodities.
- Additional first-stage statistics (Appendix C):
  - Table C1 family reports first-stage F-statistics indicating instruments are not weak (e.g., Commodity Events and U.S. Monetary Policy and Global Risk Appetite instruments report F >10 in reported specifications).

### Policy implications and recommendations
- Three critical policy implications for EMDEs:
  - The financial channel is instrumental in transmitting both commodity price shocks and global risk appetite shocks; policymakers must monitor vulnerabilities of the real economy to large movements in borrowing costs resulting from global shocks.
  - Given commodity prices as a transmission channel of U.S. monetary policy, policies promoting a more diversified export portfolio are essential to mitigate external shocks, especially for countries reliant on primary commodity exports.
  - Disentangling channels through which fiscal policy interacts with commodity cycles is an important area for future research.
- Targeted policy measures discussed:
  - To mitigate impacts of U.S. monetary policy shocks: diversify export portfolios and develop domestic markets to reduce reliance on commodity exports.
  - For global risk appetite shifts: strengthen domestic financial institutions and frameworks to manage borrowing costs; implement proactive financial regulation or macro-prudential policies to better manage capital flows and domestic borrowing conditions.
  - Emphasize importance of robust institutions to insulate domestic financing conditions from global shocks.

*Source: wpiea2024082-print-pdf (Section 3: Research Design and Identification; Sections 3.2 and 4.3 excerpts).*

### Section 3 details the research design and the identification strategy. Our empirical methodol-

### Section 3: Research Design and Identification (wpiea2024082-print-pdf)

### Data: sample and variables
- Estimation period: 1990 to 2019.
- Sample: yearly dataset covering 54 emerging and developing countries.
  - 32 upper middle-income countries.
  - 15 lower middle-income countries.
  - 10 low-income countries.
- Regions: the sample covers all the regions in the world.
- Key series included:
  - Output (country-specific real GDP from World Bank’s World Development Indicators (WDI)).
  - Real exchange rates.
  - Domestic interest rates (Central Bank Policy Rate sourced from IFS and Haver).
  - Capital flows (gross capital inflows and outflows from IMF IFS; inflows defined as net inflows from foreign residents into the domestic economy; outflows defined as net outflows from domestic residents to the rest of the world; net inflows = inflows − outflows).
  - Sovereign spreads: EMBI spreads (spreads over U.S. Treasury bills using J.P. Morgan’s EMBI global diversified index obtained from Datastream, Bloomberg, and J.P. Morgan) and BAA spreads (from the Federal Reserve Bank of St. Louis FRED).
  - Export and import prices and commodity prices (see Px construction below).
- Export price index (Px) construction:
  - Country-specific Px indices denominated in U.S. dollars using sectoral export shares, commodity prices, and disaggregated U.S. PPI data as a proxy for manufacturing prices.
  - Export shares calculated from disaggregated product export values sourced from the MIT Observatory of Economic Complexity.
  - Commodity prices from the World Bank’s Commodity Price Data.
  - U.S. PPI for manufacturing categories and the U.S. CPI from Federal Reserve Bank of St. Louis FRED.
  - Export and import price indices are deflated by the U.S. CPI (real dollar export prices).
  - The methodology follows the IMF Export and Import Prices Manual and is explained in Di Pace et al. (2020).
- Coverage and data-source details: presented in Appendix A of the source.

### Stylized facts motivating the analysis
- Px (export prices) booms and busts play a key role in driving EMDE business cycles and are strongly associated with surges and flights of capital flows (Panels (a) and (b) in Figure 1).
- Commodity prices are not the only drivers of emerging market capital flows; global financial conditions matter:
  - BAA spread used as an indicator of global financial conditions (following Akinci (2013)).
  - Panel (c): worldwide financial stress typically triggers capital flights from EMDEs.
  - Increases in commodity prices and improved global financial conditions are associated with lower EMBI spreads.
  - Panel (d): most common variation in domestic financing conditions can be explained by a combination of commodity prices and global financial conditions.
- Relationship between commodity prices and global financial conditions:
  - Commodity prices lack a distinct contemporaneous correlation with global financial conditions (Panel (e)).
  - There is a robust dynamic connection: shifts in global financial conditions consistently precede changes in commodity prices (Panel (f)).
- Implication: global financial stress and fluctuations in commodity prices emerge as primary drivers behind broad comovement in asset prices, capital flows, and global crises; therefore empirical analysis must account for these global shocks to avoid biased inference about commodity price effects.

### Identification: goals and challenges
- Research goal: investigate the role of world shocks—commodity price fluctuations and changes in global financial conditions—in shaping business and capital flow cycles of EMDEs.
- Main transmission channel for commodity price fluctuations: price of exports (Px).
  - Median share of primary commodities in total exports across the sample: 0.68.
  - Commodity price volatility is significantly higher than goods prices; commodity fluctuations dominate variation in export prices.
  - Using export prices accounts for time-varying export specialization.
- Small open economy assumption: domestic conditions unlikely to affect global variables for sample countries, but global variables may be endogenous to global economic activity.
- Identification challenge:
  - Fluctuations in export prices reflect both commodity-specific idiosyncratic shocks and endogenous responses of international prices to global economic activity.
  - Causal effects of changes in the BAA spread may vary depending on the underlying cause of the shift.
  - Solution: use external instrumental variables to identify causal impacts of Px and BAA spread shifts.

### Commodity prices instrument: event-based quasi-natural experiment
- Approach: use commodity-market-specific events associated with large, plausibly exogenous price swings as instruments.
- Event selection:
  - Historical documents and newspaper articles examined to identify episodes unrelated to major macroeconomic developments (e.g., natural disasters, weather shocks, significant local geopolitical events) that had disproportionate impacts on specific commodity prices.
  - Total events identified: 24 (summarized in Table 1).
  - Example: 2003 cotton price positive shock due to severe weather damage to cotton crops in China; treated as an exogenous shift for cotton exporters in that year.
  - When an event directly affected a producing country, that country is excluded from being treated as instrumented for that event (example: 2002 cocoa attempted coup in Cˆote d’Ivoire excluded for Cˆote d’Ivoire).
  - Appendix B provides detailed narratives and evidence; Appendix B.5 reports country exclusions.
- Construction of surprise metric for each event:
  - Surprise defined as e_{c,t} = p_{c,t} – E_{t−1}[p_{c,t}], where p_{c,t} is the (log real) price of commodity c at time t and E is expectation.
  - Expectations estimated from a regression that conditions on the commodity’s own price history and the overall (log) level of real commodity price indices (including lags) for commodity groups to which the commodity does not belong.
  - For commodity-group controls, the three main commodity indices considered are agricultural, energy, and metals; when evaluating an agricultural commodity surprise, lagged energy and metal indices proxy the global component.
  - The surprise isolates unanticipated variation attributable to the event, following the approach in Hamilton (2003).
- Event-time indicator and instrument aggregation:
  - For each event j, define q_{j,t} = e_{c,t} for t corresponding to the year of the event, and q_{j,t} = 0 for all other periods.
  - Export weight w_{i,c,t} denotes the export weight of commodity c for country i at time t.
  - Instrument constructed as z_{i,t} = Σ_j 1(w_{i,c,t−1} > w) w_{i,c,t−1} q_{j,t}, where 1(x) is an indicator equal to 1 when condition x is satisfied.
  - Lower bound chosen: w = 2% to limit noise from countries with limited exposure.
- Identification logic:
  - Unanticipated commodity price variations during events, scaled by country-specific export exposure, generate exogenous fluctuations in country export prices.
  - Cross-sectional variation in exposures (w_{i,c,t−1} ≠ w_{j,c,t−1} for i ≠ j) is exploited to identify causal effects (LATE interpretation per Imbens and Angrist (1994)).
- First-stage evidence:
  - Table C1 in Appendix C reports first-stage regressions of ∆p_{x,it} (change in detrended log of export prices for country i) on z_{it} (including controls and country fixed effects).
  - Reported F-statistic: above 700, indicating z_{it} is not a weak instrument.

### Empirical implications and estimation setup (summary)
- The instrument isolates exogenous commodity price shocks from movements driven by global economic conditions.
- The design leverages event-driven surprises and cross-country differences in export composition to estimate local impulse responses (LATE) of export prices on macro and financial outcomes in EMDEs.
- The identification strategy accounts for:
  - Time-varying export composition.
  - Exclusion of directly affected producing countries for specific events.
  - Conditioning on global commodity price indices to remove common global components.

*Source: wpiea2024082-print-pdf (Section 3: research design and identification) — canonical PDF content provided.*

### 3.2    Financial Conditions Instrument

### 3.2    Financial Conditions Instrument

### Instrument construction and identification
- Indicator: BAA spread used as an indicator of global financial conditions.
- Distinction: Separate changes in the BAA spread into two drivers:
  - U.S. monetary policy spillovers.
  - Global risk perception shifts (global risk appetite).
- Novelty: Use of instrumental variables and explicit separation of the two drivers distinguishes this approach from prior literature (notably Akinci (2013)), which identifies generic “BAA spread shocks” and thereby merges the two effects.
- Instrument for U.S. monetary policy-driven BAA movements:
  - Constructed as a common factor from an unbalanced panel of (standardized) monetary policy shock proxies.
  - Individual proxies are cumulated over the calendar year to get annual proxies; the first principal component is then estimated following Stock and Watson (2002).
  - Rationale: many proxies minimize contamination by the “central bank informational effect” (Nakamura and Steinsson, 2018) because the central bank’s information component is unlikely to be systematically associated with many proxies.
- Instruments for global risk appetite-driven BAA movements:
  - A proxy for uncertainty shocks computed from variations in the price of gold around uncertainty-related events (Piffer and Podstawski (2017)).
  - A measure of U.S. financial uncertainty from Ludvigson et al. (2021).
  - Properties: both are orthogonal to U.S. monetary policy shocks and identify events associated with significant shifts in “risk-off” behavior in financial markets.
- Distinctness: Regressing the U.S. monetary policy shock on the two risk instruments using an outlier robust estimation method yields insignificant coefficients, ensuring the monetary policy shock and risk appetite instruments capture distinct causal effects.
- Robustness: Results reported are robust to alternative choices (e.g., wat1% or 0.5%).

### Empirical model (LP with IV) and estimation details
- Estimation approach: Local projections (Jordà, 2005) estimated with instrumental variables (IV).
- Outcome variables: log of GDP (detrended), log of export prices (detrended), log of the policy rate, log of the EMBI spread, log of the real exchange rate (detrended), capital inflows and outflows in terms of trend GDP, foreign exchange reserves in terms of trend GDP.
- Model specification (for h = 0,1,...,H):
  y_{i,t+h} − y_{i,t−1} = μ^h_i + f_{i,t} β^h + (x_{i,t} − \bar{x}_i) γ^h_0 + f_{i,t} (x_{i,t} − \bar{x}_i) θ^h_x + ω_{i,t+h}
- Interpretation:
  - Dependent variable: cumulative change in country i outcome y from year t−1 to t+h.
  - f_{i,t}: intervention/treatment (change in export prices or in the BAA spread).
  - μ^h_i: country fixed effects.
  - x_{i,t}: additional covariates in deviation from their mean.
  - Interaction terms f_{i,t} (x_{i,t} − \bar{x}_i) included (following Cloyne et al., 2023) to allow heterogeneity in causal effects; θ^h_x need not equal 0.
- Controls used in empirical application (x_{i,t}): two lags of real GDP growth, Px growth, the BAA spread, net capital inflows, and the lag of the dependent variable.
- Interpretation of estimates: β^h capture average dynamic causal effect; because the treatment is instrumented, reported impulse response functions (IRFs) can be interpreted as the LATE (local average treatment effect).

### Impact of commodity shocks (one standard deviation increase in Px driven by commodity-specific shocks)
- Main dynamic responses (baseline average effect):
  - Domestic GDP: steady increase.
  - Real exchange rate: appreciation.
  - Policy rate: gradual decrease over time (suggesting a moderately pro-cyclical domestic monetary policy response).
  - EMBI spread: mild decrease (lower borrowing costs).
  - Capital outflows: increase, mainly driven by other investment flows (mostly bank-related).
  - Foreign exchange reserves: accumulation.
  - Direct investment flows (both inflows and outflows): small positive response.
  - Capital inflows: more muted response.
  - Net capital inflows (inflows − outflows): decline.
- Additional notes:
  - These commodity-specific shocks are idiosyncratic commodity events.
  - If horizon extended, all IRFs exhibit mean reversion, but confidence bands become considerably larger after four periods due to relatively short samples for many countries.
- Implication:
  - The negative association between the EMBI spread and Px is not solely attributable to commodity price shocks; other factors (e.g., GFC or shifts in global demand) influence this relationship.
  - Observed comovement of commodity prices, capital flows, and economic fluctuations does not align with pure commodity price shock effects (commodity shocks are associated with significant net outflows rather than simultaneous increases in inflows and outflows).
  - Policy implication: contrary to conventional wisdom (Kaminsky et al., 2004), “it does not always pour when it rains.”

### Decline in the BAA spread: common effects and differences by driver
- Common characteristics of a reduction in the BAA spread (regardless of driver):
  - Increase in GDP and Px.
  - Real exchange rate: appreciation (counterpart of weakening U.S. dollar).
  - EMDE central banks: increase interest rates in reaction to easier global financial conditions.
  - EMBI spread: immediate decrease despite higher domestic policy rates (result of eased global financial conditions).
  - Negative correlation between EMBI spread and Px consistent with Drechsel and Tenreyro (2018) and Reinhart et al. (2016).
- Two drivers analyzed:
  - U.S. monetary policy shock (depicted in blue in Figures 4 and 5).
  - Shift in global risk appetite (depicted in magenta).
- Differences between global risk appetite shock and U.S. monetary policy shock:
  1. Temporal profile:
     - Global risk appetite shock: more temporary; IRFs display spike-like patterns in GDP, Px, policy rate, and real exchange rate.
     - U.S. monetary policy shock: humped-shaped response in Px and output, indicating a more persistent influence through global commodity prices.
  2. EMBI spread dynamics:
     - Global risk appetite shock: initial reduction in EMBI spread is very large but quickly reverses (typical risk-on/risk-off event).
     - U.S. monetary policy shock: decline in EMBI spread is more persistent and builds up over time, even though the BAA spread reverts faster after a U.S. monetary policy shock compared to a risk appetite shock.
  3. Capital flow responses:
     - Both shocks are associated with overall increases in both inflows and outflows, but:
       - Global risk appetite shock: rise in inflows and outflows is almost identical → net inflows relatively stable.
       - U.S. monetary policy shock: net inflows tend to decrease.
     - Composition differences:
       - Global risk appetite shock: large surge in portfolio inflows and outflows.
       - U.S. monetary policy shock: muted portfolio flow response.
     - The global risk shock pattern aligns with “when it rains, it pours” dynamics (Kaminsky et al., 2004).
- Implications for literature and policy:
  - U.S. monetary policy plays a pivotal role in shaping global investor risk perceptions and thereby impacts capital flows to and from EMDEs, causing direct fluctuations in credit spreads (consistent with Kalemli-Özcan (2019)).
  - The impact of U.S. monetary policy shocks on EMDEs can be considerable when the BAA spread is affected, contrasting with Akinci (2013) who finds minimal effects.
  - Importance underscored of identifying the origin of BAA spread fluctuations rather than treating a generic “BAA spread shock.”
  - Findings support focusing on gross capital movements (inflows and outflows separately) rather than net flows (Forbes and Warnock, 2012; Milesi-Ferretti and Tille, 2014).
  - Evidence: both capital inflows and outflows increase following a reduction in global risk appetite, consistent with Forbes and Warnock (2012) characterizations of episodes (stops, retrenchments, surges, flights).

### Figures and confidence intervals (as reported)
- Impulse responses:
  - Commodity-shock driven Px IRFs: gray areas denote 68% and 90% confidence intervals.
  - BAA spread decline IRFs (U.S. monetary policy vs global risk appetite): shaded areas denote 68% and 90% confidence intervals.
- Capital flow measures in figures: presented as a ratio with respect to the (quadratic) trend of GDP denominated in U.S. dollars.

*Source: IMF staff analysis in "3.2    Financial Conditions Instrument" (wpiea2024082-print-pdf).*

### 4.3    Exogenous vs. Endogenous Response of Commodity Prices

### 4.3    Exogenous vs. Endogenous Response of Commodity Prices

### Distinction between exogenous and endogenous commodity price movements
- Existing empirical literature treats global commodity prices as exogenous in a small open economy setting; under that premise, the impact of a shift in commodity prices on the domestic economy can be readily identified.
- OLS yields a mean response that combines effects of both exogenous and endogenous shifts in export prices; IV isolates effects of export price increases driven by commodity-specific shocks exploiting cross-country exposure to major commodity events.
- Comparative patterns (OLS versus IV):
  - With OLS, the appreciation of the exchange rate is substantially larger and more persistent.
  - With OLS, the policy rate exhibits an immediate increase.
  - Responses for reserves and capital flows deviate considerably between OLS and LATE (IV): OLS suggests a persistent surge in capital inflows, whereas the IV response depicts a more muted reaction.
- Caution: Inferring that fluctuations in capital flows are a consequence of commodity price shocks (based on OLS) can be misleading because OLS responses combine elements associated with other global shocks; the global factor in capital flows correlated with commodity prices appears to reflect U.S. monetary and risk appetite shocks rather than exogenous commodity price shocks.

### Additional results and robustness
- Omitting events:
  - Exclude crude oil event of 1998 and all events in 2008 to avoid contamination from weak global growth/Global Financial Crisis; results remain robust.
  - Construct instrument with subset of 15 events from Di Pace et al. (2020); results remain robust.
- Omitting countries:
  - Excluding one country at a time does not change results; no single country drives findings disproportionately.
- Subsample analysis (post-2000):
  - Effects are qualitatively similar when sample starts in 2000.
  - Effects on Px and the real exchange rate are higher in the post-2000 subsample, possibly indicating a more prominent role of the exchange rate as a shock absorber.
  - In response to a looser U.S. monetary policy, GDP impact is smaller but responses of the real exchange rate, the policy rate, the EMBI spread, reserves, capital inflows, and outflows are amplified in the 2000s subsample.
  - Post-2000, the response of Px to a global risk appetite shock is notably larger; exchange rate shows larger appreciation; reserves, capital inflows and outflows exhibit larger responses.
  - Broad conclusions are preserved when focusing on last 20 years of data.
- Events associated with energy commodities:
  - Energy commodities have distinct attributes (indispensable inputs, no viable substitutes; energy shocks historically associated with contractions in economic activity).
  - Effects of an increase in Px driven by energy commodity-specific shocks are not very different from the baseline.
  - Note: cross-sectional variation for non-energy commodities is limited, constraining identification for those commodities.

### Framework to explore channels of transmission
- Two channels examined:
  - Financial channel: operates through the endogenous response of country spreads (EMBI spread).
  - Commodity channel: operates through changes in export prices (Px).
- Methodological approach:
  - Extend standard local projection (LP) by introducing interaction terms capturing country-specific sensitivity of EMBI spread or Px to the shock.
  - Use Kitagawa-Blinder-Oaxaca (KBO) decomposition to separate direct effect (βh) and indirect effect (Θhi θhf) mediated by movements in EMBI spread or Px.
  - Construct mediating variable Θhi by estimating country-specific responses eΘhi from a parsimonious LP (two lags of ζi,t among controls); normalize eΘhi against the cross-country average to obtain Θhi.
  - Counterfactuals computed as: E[GDPi,t+h − GDPi,t−1 | ft = 1, Θh = κ] − E[GDPi,t+h − GDPi,t−1 | ft = 0] = βh + θhf × κ, with κ set to −1.5σΘ,h or 1.5σΘ,h.

### Empirical findings on channels and heterogeneity
- GDP response to surge in Px driven by commodity-specific shocks (Figure 6):
  - Expansionary effects are more pronounced in countries with a significant decline in the EMBI spread (green line) and muted/shorter-lived where EMBI spread increases (red line).
  - Differences are statistically different from each other at a 10% level after the second year.
  - Economic magnitude:
    - With a 4.5% decrease in the EMBI spread (approximately 20 basis points per period, consistent with the LATE reported in Figure 2), GDP increases by 0.18% for every percentage point rise in Px.
    - In a counterfactual where the EMBI spread drops 1.5 standard deviations more than the average (i.e., by about 47% or 210 basis points), an equivalent rise in export prices leads to a GDP expansion of around 0.40% for every percentage point increase in Px.
    - Median value of average EMBI spreads for sample countries is around 450 basis points (used to convert percent decrease in EMBI spread into basis points).
  - Interpretation: Financial channel (borrowing costs) operates alongside terms-of-trade channel; robust institutions that insulate domestic financing conditions from global shocks matter.
- GDP response to a fall in the BAA spread mediated by EMBI spread (Figure 7):
  - Expansion of domestic GDP is greater when the fall in the EMBI spread is larger (green line).
  - Effect is statistically significant and economically meaningful only when the BAA spread reduction is driven by global risk appetite.
  - Implies domestic financial channels that affect country spreads amplify effects of global financial shocks tied to risk appetite.
- GDP response to a fall in the BAA spread mediated by Px (Figure 8):
  - Endogenous response of Px significantly shapes and amplifies transmission of U.S. monetary policy to EMDEs; indirect effect is statistically significant.
  - A modest increase in Px yields a restrained expansionary effect of a decline in the BAA spread linked to a more accommodative U.S. monetary policy.
  - Responses of GDP are less sensitive to Px movements when BAA spread reduction is driven by global risk appetite.
  - In countries with robust Px reaction, the peak effect of a similar-sized fall in the BAA spread is nearly twice as large when global financial conditions are eased by U.S. monetary policy rather than by global risk appetite.
- Synthesis of channel differences:
  - U.S. monetary policy shocks are amplified by the endogenous reaction of export prices.
  - Transmission of shifts in global risk appetite is strongly tied to endogenous domestic borrowing costs (EMBI spread).
  - Policy responses need to account for these different transmission mechanisms.

### Policy implications and recommendations
- Three critical policy implications for EMDEs:
  - The financial channel is instrumental in transmitting both commodity price shocks and global risk appetite shocks; policymakers must be wary of vulnerabilities of the real economy to large movements in borrowing costs resulting from global shocks.
  - Given commodity prices as a transmission channel of U.S. monetary policy, policies promoting a more diversified export portfolio are essential to mitigate such external shocks, especially for countries reliant on primary commodity exports.
  - Disentangling channels through which fiscal policy interacts with commodity cycles is an important area for future research.
- Targeted policy measures discussed:
  - To mitigate impacts of U.S. monetary policy shocks: diversify export portfolios and develop domestic markets to reduce reliance on commodity exports.
  - For global risk appetite shifts: strengthen domestic financial institutions and frameworks to manage borrowing costs; implement proactive financial regulation or macro-prudential policies to better manage capital flows and domestic borrowing conditions.
  - Emphasize the importance of robust institutions to insulate domestic financing conditions from global shocks.

*Source: wpiea2024082-print-pdf - 4.3    Exogenous vs. Endogenous Response of Commodity Prices*

### References

### References and Appendix A — Data

### References (literature cited)
- Bibliography includes empirical and theoretical studies on:
  - Global Financial Conditions, Country Spreads and Macroeconomic Fluctuations (AKINCI, O., 2013).
  - Commodity-Price Comovement and Global Economic Activity (ALQUIST, R., S. BHATTARAI, AND O. COIBION, 2020).
  - Identifying Monetary Policy Shocks: A Natural Language Approach (ARUOBA, B. AND T. DRECHSEL, 2022).
  - Wage Discrimination: Reduced Form and Structural Estimates (BLINDER, A. S., 1973).
  - Cross-Border Banking and Global Liquidity (BRUNO, V. AND H. S. SHIN, 2014).
  - The Macroeconomic Impact of Financial and Uncertainty Shocks (CALDARA, D., C. FUENTES-ALBERO, S. GILCHRIST, AND E. ZAKRAJˇSEK, 2016).
  - Capital Inflows and Real Exchange Rate Appreciation in Latin America (CALVO, G. A., L. LEIDERMAN, AND C. M. REINHART, 1993).
  - Inflows of Capital to Developing Countries in the 1990s (CALVO, G. A., L. LEIDERMAN, AND C. M. REINHART, 1996).
  - Capital Flows in Risky Times: Risk-On / Risk-Off and Emerging Market Tail Risk (CHARI, A., K. D. STEDMAN, AND C. T. LUNDBLAD, 2020).
  - State-Dependent Local Projections: Understanding Impulse Response Heterogeneity (CLOYNE, J., O. JORD`A, AND A. M. TAYLOR, 2023).
  - Global Drivers of Gross and Net Capital Flows (DAVIS, J. S., G. VALENTE, AND E. VAN WINCOOP, 2021).
  - If the Fed sneezes, who catches a cold? (DEDOLA, L., G. RIVOLTA, AND L. STRACCA, 2017).
  - The Global Transmission of U.S. Monetary Policy (DEGASPERI, R., S. S. HONG, AND G. RICCO, 2023).
  - Terms-of-Trade Shocks are Not all Alike (DIPACE, F., L. JUVENAL, AND I. PETRELLA, 2020).
  - Monetary Policy for Commodity Booms and Busts (DRECHSEL, T., M. MCLEAY, AND S. TENREYRO, 2019).
  - Commodity Booms and Busts in Emerging Economies (DRECHSEL, T. AND S. TENREYRO, 2018).
  - Sharing a Ride on the Commodities Roller Coaster (FERNÁNDEZ, A., A. GONZÁLEZ, AND D. RODRÍGUEZ, 2018).
  - World Shocks, World Prices, and Business Cycles: An Empirical Investigation (FERNÁNDEZ, A., S. SCHMITT-GROHÉ, AND M. URIBE, 2017).
  - Capital Flow Waves: Surges, Stops, Flight, and Retrenchment (FORBES, K. J. AND F. E. WARNOCK, 2012).
  - The Effect of Monetary Policy on Real Commodity Prices (FRANKEL, J. A., 2008).
  - Monetary Policy in Emerging Markets (FRANKEL, J. A., 2010).
  - Determinants of Global Spillovers from US Monetary Policy (GEORGIADIS, G., 2016).
  - Monetary Policy Surprises, Credit Costs, and Economic Activity (GERTLER, M. AND P. KARADI, 2015).
  - Natural Resources and Sovereign Risk in Emerging Economies (HAMANN, F., J. C. MENDEZ-VIZCAINO, E. G. MENDOZA, AND P. RESTREPO-ECHAVARRIA, 2023).
  - Oil and the Macroeconomy since World War II; What is an Oil Shock? (HAMILTON, J. D., 1983, 2003).
  - The Prebisch-Singer hypothesis: four centuries of evidence (HARVEY, D. I., N. M. KELLARD, J. B. MADSEN, AND M. E. WOHAR, 2010).
  - Identification and Estimation of Local Average Treatment Effects (IMBENS, G. W. AND J. D. ANGRIST, 1994).
  - Estimation and Inference of Impulse Responses by Local Projections (JORDÀ, O., 2005).
  - The Effects of Quasi-Random Monetary Experiments (JORDÀ, O., M. SCHULARICK, AND A. M. TAYLOR, 2020).
  - Speculation in the Oil Market (JUVENAL, L. AND I. PETRELLA, 2015).
  - U.S. Monetary Policy and International Risk Spillovers (KALEMLI-ÖZCAN, S., 2019).
  - Terms of Trade Shocks and Fiscal Cycles (KAMINSKY, G. L., 2010).
  - When It Rains, It Pours: Procyclical Capital Flows and Macroeconomic Policies (KAMINSKY, G. L., C. M. REINHART, AND C. A. VÉGH, 2004).
  - On the Prevalence of Trends in Primary Commodity Prices (KELLARD, N. AND M. E. WOHAR, 2006).
  - Not All Oil Price Shocks Are Alike (KILIAN, L., 2009).
  - Components of a Difference Between Two Rates (KITAGAWA, E. M., 1955).
  - Uncertainty and Business Cycles: Exogenous Impulse or Endogenous Response? (LUDVIGSON, S. C., S. MA, AND S. NG, 2021).
  - The Great Retrenchment: International Capital Flows During the Global Financial Crisis (MILESI-FERRETTI, G.-M. AND C. TILLE, 2014).
  - U.S. Monetary Policy and the Global Financial Cycle; The Global Financial Cycle (MIRANDA-AGRIPPINO, S. AND H. REY, 2020, 2021).
  - The Transmission of Monetary Policy Shocks (MIRANDA-AGRIPPINO, S. AND G. RICCO, 2021).
  - High-Frequency Identification of Monetary Non-Neutrality: The Information Effect (NAKAMURA, E. AND J. STEINSSON, 2018).
  - Male-Female Wage Differentials in Urban Labor Markets (OAXACA, R., 1973).
  - The Global Dollar Cycle (OBSTFELD, M. AND H. ZHOU, 2022).
  - The Time-Varying Effect of Monetary Policy on Asset Prices (PAUL, P., 2020).
  - Identifying Uncertainty Shocks Using the Price of Gold (PIFFER, M. AND M. PODSTAWSKI, 2017).
  - Global Cycles: Capital Flows, Commodities, and Sovereign Defaults, 1815-2015 (REINHART, C. M., V. REINHART, AND C. TREBESCH, 2016).
  - Capital Flow Bonanzas: An Encompassing View of the Past and Present (REINHART, C. M. AND V. R. REINHART, 2009).
  - Dilemma not Trilemma: The Global Financial Cycle and Monetary Policy Independence (REY, H., 2013).
  - A New Measure of Monetary Shocks: Derivation and Implications (ROMER, C. D. AND D. H. ROMER, 2004).
  - How Important are Terms-of-Trade Shocks? (SCHMITT-GROHÉ, S. AND M. URIBE, 2018).
  - Macroeconomic Forecasting Using Diffusion Indexes (STOCK, J. H. AND M. W. WATSON, 2002).
  - Financial Dampening (WIELAND, J. F. AND M. YANG, 2020).
  - Discussion of “Monetary Policy for Commodity Booms and Busts” (XIONG, W., 2019).

### A — Data (sample and coverage)
- Sample:
  - Our sample comprises a group of 54 emerging and low income countries.
  - The split by income category is summarized in Table A1.
  - Tables A2 and A3 provide a comprehensive summary of the macro data coverage for each country and variable considered.
  - Figure A1 shows the probability distribution of primary commodity exports for all the countries in our sample.

### A.1 Data Sources
- Real GDP in local currency units.
  - Source: World Bank’s World Development Indicators (WDI) database.
  - Indicator code: NY.GDP.MKTP.KN
- Capital flows are scaled by trend GDP using the following sources:
  - Capital inflows. Source: International Financial Statistics.
  - Capital outflows. Source: International Financial Statistics.
  - Nominal GDP in dollar terms. Source: World Economic Outlook.
- Spreads are obtained combining data based on:
  - EMBI. Description: JP Morgan EMBI global diversified index and JP Morgan EMBI global index, in bps. Sources: Datastream, Bloomberg, and JP Morgan.
  - Interest rate spreads. Description: domestic rate over U.S. rate (lending rate or t-bill), in %. Source: International Financial Statistics.
- BAA Spread.
  - Description: Moody’s seasoned BAA corporate bond yield relative to yield on 10-year treasury constant maturity, bps.
  - Source: Federal Reserve Bank of St. Louis FRED.
- Policy rate.
  - Description: Central Bank policy rate, in bps.
  - Source: International Financial Statistics, Haver, and CEIC Data Global Database.
  - Footnote on alternatives: For a number of countries the Central Bank policy rate is not available. In such cases, we use alternative interest rates as follows. For Mozambique, Pakistan, Sri Lanka, Tanzania, Trinidad and Tobago, Uruguay, and Venezuela, we use treasury rates. For El Salvador and Panama, we use the money market rates; and for Cameroon, Ecuador, and Gabon, we use the discount rates.
- Export Prices.
  - Export price index, 2010=100.
  - Sources: Authors’ calculation using as inputs data from MIT Observatory of Economic Complexity, World Bank, Federal Reserve Economic Data (FRED, Federal Reserve Bank of St. Louis).

### A.2 Emerging Market Spreads
- Sovereign spreads:
  - Mainly derived from the J.P. Morgan EMBI global diversified index, which measures the spread over U.S. Treasuries.
  - To expand coverage for certain countries, supplemented with the J.P. Morgan EMBI global index.
  - To further extend coverage, interest rate spreads were also used, calculated as the difference between the domestic t-bill and the U.S. t-bill.
  - In cases where t-bill data was not available, the domestic lending rate over the U.S. lending rate was used instead.
  - A comprehensive breakdown of calculations is provided in Table A4.

### A.3 Country Selection
- Selection criteria:
  - Restrict to emerging countries according to the definition of the IMF World Economic Outlook.
  - Narrow the sample to countries with data available for the variables analyzed.

*Source: wpiea2024082-print-pdf - References*

### 2007.  Third, we drop large economies such as China and India.  Finally, we drop economies

### wpiea2024082-print-pdf - 2007.  Third, we drop large economies such as China and India.  Finally, we drop economies

### Sample selection and country coverage
- After applying filters (dropping advanced economies, dropping large economies such as China and India, and dropping economies classified as emerging but part of the European Union such as Poland) the sample consists of 54 emerging economies.
- Table A1: Country Classification by Income Group (as presented)
  - Upper Middle Income:
    - Argentina
    - Armenia
    - Azerbaijan
    - Belarus
    - Belize
    - Brazil
    - Chile
    - Colombia
    - Costa Rica
    - Cˆote d’Ivoire
    - Dominican Republic
    - Ecuador
    - Gabon
    - Georgia
    - Guatemala
    - Iraq
    - Jamaica
    - Kazakhstan
    - Kuwait
    - Malaysia
    - Mexico
    - Panama
    - Peru
    - Qatar
    - Russia
    - Serbia
    - South Africa
    - Thailand
    - Turkey
    - Trinidad and Tobago
    - Uruguay
    - Venezuela
  - Lower Middle Income:
    - Algeria
    - Angola
    - Bolivia
    - Egypt
    - El Salvador
    - Indonesia
    - Lebanon
    - Mongolia
    - Morocco
    - Nigeria
    - Pakistan
    - Philippines
    - Sri Lanka
    - Tunisia
    - Ukraine
  - Low Income:
    - Cameroon
    - Cˆote d’Ivoire
    - Ghana
    - Kenya
    - Mozambique
    - Tanzania
    - Vietnam
    - Zambia
- Note: The low-income classification is from the IMF while the breakdown between upper middle-income and lower middle-income is sourced from the World Bank.

### Macro data coverage (Table A2) — key aspects
- Macro variables covered (and their country-level sample periods) include:
  - Real GDP
  - EMBI Spreads
  - Export Prices
  - Interest Rate
  - Real Exchange Rate
- Example country coverage entries (exact year ranges as listed):
  - Algeria: Real GDP 1990-2019; EMBI Spreads 1999-2019; Export Prices 1990-2019; Interest Rate 1990-2019; Real Exchange Rate 1990-2019
  - Angola: Real GDP 1990-2019; EMBI Spreads 2006-2019; Export Prices 1990-2019; Interest Rate 2005-2019; Real Exchange Rate 1990-2019
  - Argentina: Real GDP 1990-2019; EMBI Spreads 1993-2019; Export Prices 1990-2019; Interest Rate 2002-2019; Real Exchange Rate 1990-2019
  - Mongolia: Real GDP 1990-2019; EMBI Spreads 2012-2019; Export Prices 1990-2019; Interest Rate 2008-2021; Real Exchange Rate 1992-2019
  - Venezuela: Real GDP 1990-2014; EMBI Spreads 1993-2019; Export Prices 1990-2019; Interest Rate 2001-2016; Real Exchange Rate 2008-2016
- Notes: Table A2 presents the macro data coverage for each country included in the sample.

### Capital flows data coverage (Table A3) — structure and acronyms
- Capital flows variables and acronyms:
  - DIO: capital outflows for direct investment
  - PDO: capital outflows for portfolio debt
  - PEO: capital outflows for portfolio equity
  - OIO: capital outflows for other investment
  - FXR: flows of foreign exchange reserves
  - DII: capital inflows for direct investment
  - PDI: capital inflows for portfolio debt
  - PEI: capital inflows for portfolio equity
  - OII: capital inflows for other investment
- Example country coverage entries (exact year ranges as listed):
  - Algeria: DIO 1990-2019; PDO 1990-2013; PEO 1990-2019; OIO 1990-2019; FXR 1990-2019; DII 1990-2019; PDI 1990-2013; PEI 1990-2013; OII 1990-2019
  - Angola: DIO 1990-2019; PDO 2002-2019; PEO 2002-2019; OIO 1990-2019; FXR 1990-2019; DII 1990-2019; PDI 1990-2019; PEI 1990-2019; OII 1990-2019
  - Argentina: DIO 1990-2019; PDO 1990-2019; PEO 1990-2019; OIO 1990-2019; FXR 1990-2019; DII 1990-2019; PDI 1990-2019; PEI 1990-2019; OII 1990-2019
  - Brazil: DIO 1990-2019; PDO 1990-2019; PEO 1990-2019; OIO 1990-2019; FXR 1990-2019; DII 1990-2019; PDI 1990-2019; PEI 1992-2019; OII 1990-2019
  - Qatar: DIO 2011-2019; PDO 2011-2019; PEO 2011-2019; OIO 2011-2019; FXR 1990-2019; DII 2011-2019; PDI 2011-2019; PEI 1990-2019; OII 2011-2019
- Notes: Table A3 presents the data coverage for each capital flow series and each country.

### Spreads data (Table A4) — coverage and construction notes
- Overview:
  - Table A4 lists country-by-country spreads coverage (years) and notes on indices/methods used (EMBI G, EMBI GD, EMGI GD, MECI, splicing approaches).
- Aggregate summary from table:
  - Some country spreads series start as early as 1990 (e.g., Zambia 1990-2019) while many countries have coverage starting in the 1990s or 2000s (examples provided below).
  - Several series use EMBI GD or EMBI G indices; where EMBI indices are unavailable the series are spliced or constructed using interest rate spreads (domestic lending rate over U.S. lending rate or domestic t-bill over U.S. t-bill) or alternate indices (e.g., CACI for Costa Rica and Guatemala, Sub-Saharan Africa GD index for Cameroon/Kenya splicing).
- Example country entries (exact year ranges and notes preserved):
  - Algeria: 1999-2019 — 1999-2002 uses EMBI GD index. Coverage extended by splicing using the African EMBI GD index for 2003-2019.
  - Angola: 2006-2019 — 2012-2019 uses EMBI G and GD indices. Spliced backwards using interest rate spread based on domestic t-bill over U.S. t-bill for 2006-2011.
  - Argentina: 1993-2019 — EMBI G and EMGI GD indices
  - Kenya: 1993-2019 — 2014-2019 uses EMBI GD index. Spliced backwards using the Sub-Saharan Africa GD index for 2003-2013.
  - Thailand: 1997-2005 — EMBI GD index.
  - Zambia: 1990-2019 — 2012-2014 uses EMBI GD index. Spliced backwards using interest rate spread based on domestic t-bill over U.S. t-bill for 1990-2011.
- Notes: Table A4 displays the coverage for the country spreads data along with the specific indices used for each country.

### Commodity events: methodology and instrument construction
- Methodological criteria for identifying commodity-price events used to build the export-prices instrument:
  1. The event has to be important enough to affect a commodity market at a global level (examples: natural disasters or weather-related shocks in key producing areas, major geopolitical events, unanticipated news on global production or demand).
  2. The event should have an unambiguous effect on the price of the commodity.
  3. The event has to be unrelated to important macroeconomic developments such as the global financial crisis or a U.S. recession (to eliminate endogenous responses of commodity prices to the state of the economy).
- Using these criteria the authors identified 24 episodes of exogenous commodity price shocks that are unrelated to business cycle fluctuations:
  - 16 favorable (positive) commodity price shocks
  - 8 negative commodity price shocks
- Sources used to construct series: FAO reports, IMF and World Bank publications, newspaper articles, academic papers, and online sources.

### Selected agriculture commodity events (B.1) — documented episodes and key figures
- Coffee
  - Year of Event: 1994.
  - Type of Event: Positive price shock.
  - Evidence: International Coffee Organization (ICO) reports climate shocks in Brazil in 1994.
  - Data point: Arabica coffee prices increased from 1.56 dollars per kilo in 1993 to 3.31 in 1994.
  - Newspaper evidence (New York Times, July 11, 1994) documents frost in Brazil; quoted lines include:
    - “Frost struck in Brazil’s biggest coffee-growing state early today, and farmers said the effects were harsher than a freeze that hit two weeks ago.”
    - “(...)Coffee prices soared after the previous cold snap late last month, which destroyed one-third of next year’s crop.  Brazil is the largest coffee producer, accounting for about a quarter of world production. A threat to its crop can drastically affect world coffee prices(...).”
- Cereal
  - Year of Event: 1997.
  - Type of Event: Negative price shock.
  - Evidence: FAO forecast and global production increased by 7.8 percent in 1996, translating into lower prices; cereal price index fell from 83.61 in 1996 to 64.76 in 1997.
  - Year of Event: 2010.
  - Type of Event: Positive price shock.
  - Evidence: De Winne and Peersman (2016) and FAO report cite adverse weather in key producers (heatwave/droughts in Russian Federation, Kazakhstan, Ukraine; floods in Moldova) and export restrictions (Russian Federation ban) driving grain price surges.
- Cocoa
  - Year of Event: 1999.
  - Type of Event: Negative price shock.
  - Evidence: FAO attributes 1999 cocoa price drop to supply surplus from increased production; New York Times (November 3, 1999) reported cocoa for December delivery fell $38, to $840 a metric ton.
  - Year of Event: 2002.
  - Type of Event: Positive price shock.
  - Evidence: International Cocoa Organization reports attempted coup on 19th September in Cˆote d’Ivoire and associated uncertainty pushed prices to a 16-year high at 2.44 dollars per tonne in October 2002; authors note cocoa prices increased between 2001 and 2002.

*This content unit is from the provided IMF PDF content.*

### 1.07 dollars per kilo to 1.78 dollars per kilo.

### wpiea2024082-print-pdf - 1.07 dollars per kilo to 1.78 dollars per kilo.

### Cocoa
- Year of Event: 2017.
- Type of Event: Positive price shock.
- Drivers:
  - Decline in cocoa prices in 2017 driven by favorable weather conditions in major producing countries such as Cˆote d’Ivoire and Ghana.
- Data points and citations:
  - Our data show that cocoa prices declined around 30 percent in 2017.
  - Newspaper evidence: "War Inflates Cocoa Prices But Leaves Africans Poor, New York Times (October 31, 2002)" documents a cocoa price increase originating in Cote d’Ivoire in 2002 and notes prices reached 17-year highs during civil war-related disruptions.

### Rice
- Year of Event: 2008.
- Type of Event: Positive price shock.
- Drivers:
  - 2008 rice prices nearly doubled.
  - A report from the United States Department of Agriculture explains that the price increase was driven by trade restrictions of major suppliers.

### Sugar
- Year of Event: 2006.
- Type of Event: Positive price shock.
- Drivers:
  - Sugar price increase in 2006 caused by severe draughts in Thailand, the second largest sugar producer.

### Soybean
- Year of Event: 2008.
- Type of Event: Positive price shock.
- Drivers:
  - High soybean prices in 2008 originated in the expectation of a reduction in supply.
- Data:
  - We observe an increase of 40 percent in soybean prices in our data.

### Agriculture: Raw Materials — Cotton
- Year of Event: 1994.
- Type of Event: Positive price shock.
- Drivers:
  - 1994 cotton price increase driven by a decline in production in key production areas such as China and India; decline in China due to bad weather and a bollworm infestation.
  - Recovery in world cotton consumption following stagnation after the dissolution of the Soviet Union contributed.
- Data:
  - Cotton prices declined from 1.28 dollars per kilo in 1993 to 1.76 dollars per kilo in 1994.
- Year of Event: 2003.
- Type of Event: Positive price shock.
- Drivers:
  - Severe weather damage to cotton crops in China in 2003.
  - USDA forecast put world stocks at their lowest level since 1994/95.
- Data:
  - Cotton prices increased from 1.02 dollars per kilo in 2002 to 1.40 dollars per kilo in 2003.
- Year of Event: 2010.
- Type of Event: Positive price shock.
- Drivers:
  - 2010–2011 cotton scarcity due to lower than average planted crops and negative weather shocks in the USA and Pakistan; appears cotton-specific rather than broad commodity-driven.
- Data:
  - Cotton prices increased from 1.38 dollars per kilo in 2009 to 2.28 dollars per kilo in 2010.

### Agriculture: Raw Materials — Timber
- Year of Event: 1993.
- Type of Event: Positive price shock.
- Drivers:
  - 1993 price spike driven by environmentally friendly policies issued by President Clinton to protect forests which limited timber harvests.
- Data:
  - Timber price index increased from 72.41 in 1992 to 100.58 in 1993.
- Contemporary reporting:
  - Washington Post (July 2, 1993) quoted projected average timber harvests of 1.2 billion board feet per year, versus past highs up to 5.2 billion board feet.

### Agriculture: Raw Materials — Tobacco
- Year of Event: 1993.
- Type of Event: Negative price shock.
- Drivers:
  - Worldwide increase in competition for exports in 1993 led to a substantial fall in tobacco prices.
- Data:
  - Tobacco prices declined 22 percent between 1992 and 1993.

### Agriculture: Raw Materials — Rubber
- Year of Event: 2010.
- Type of Event: Positive price shock.
- Drivers:
  - Severe draughts in Thailand and India, major rubber producers.
- Data:
  - In 2010 rubber prices almost doubled.
- Contemporary reporting:
  - Financial Times (March 31, 2010) attributed the surge to the worst drought in north Thailand in a decade and drought impacts in India.

### Energy Commodities — Combined Energy Commodities
- Year of Event: 2015.
- Type of Event: Negative price shock.
- Drivers:
  - Booming U.S. shale oil production significantly contributed to the oil price plummet in 2015 and affected other fossil fuels.
- Data:
  - Crude oil prices declined 47 percent between 2015 and 2015.
  - Coal prices contracted 16 percent between 2015 and 2015.
  - Natural gas prices contracted 26 percent between 2015 and 2015.
- Year of Event: 2019.
- Type of Event: Negative price shock.
- Drivers:
  - United States became a net energy exporter following development of shale technology; effect partially offset in crude oil by attacks to Saudi Arabia oil facilities.
- Data:
  - Natural gas prices declined 25 percent in 2019.
  - Coal prices declined 15 percent in 2019.
- Contemporary reporting:
  - New York Times (October 5, 2015) described global oil price near lowest since 2009 as a result of a supply glut and battle for market share between OPEC and the United States.

### Crude Oil
- Year of Event: 1998.
- Type of Event: Negative price shock.
- Drivers:
  - Oil supply expectations played a significant role in the 1998 oil price plunge.
- Data:
  - Oil prices declined 32 percent in 1998.

### Natural Gas
- Year of Event: 2000.
- Type of Event: Positive price shock.
- Drivers:
  - 2000–2001 California energy crisis resulted from sudden demand increase, impaired physical capacity, natural gas diversion, and inadequate storage fill; low hydroelectric availability and increased gas-fired generation usage were key factors.
- Data:
  - Natural gas price index jumped from 39.78 in 1999 to 73.85 in 2000.
- Year of Event: 2005.
- Type of Event: Positive price shock.
- Drivers:
  - Effects of Hurricanes Katrina and Rita compounded already tight supply and demand conditions, increasing fuel prices in the United States.
- Data:
  - Natural gas index went from 95.39 in 2004 to 142.40 in 2005.

*Source: wpiea2024082-print-pdf - 1.07 dollars per kilo to 1.78 dollars per kilo.*

### Introduction\%20to\%20North\%20American\%20Natural\%20Gas\%20Markets_0.pdf.

### Introduction%20to%20North%20American%20Natural%20Gas%20Markets_0.pdf

### B.4 Metals and Mineral Commodities — event summaries and magnitudes
- Gas price surge associated with Hurricane Katrina: “Gas prices in cities across the United States soared by as much as 40 cents a gallon from Tuesday to Wednesday, a surge blamed on disruptions by Hurricane Katrina in Gulf of Mexico oil production.”
- Aluminum
  - Year of Event: 1994.
  - Type of Event: Positive price shock.
  - Finding: Aluminum prices went up 30 percent in 1994.
  - Source note: Price rise attributed to reduction in stocks and cutbacks in production by major producers.
- Iron ore
  - Year of Event: 2019.
  - Type of Event: Positive price shock.
  - Finding: Iron ore prices increased around 35 percent in 2019.
  - Source note: Price increase linked to the collapse of a mining dam in Brazil.
- Lead
  - Year of Event: 2017.
  - Type of Event: Negative price shock.
  - Finding: Lead prices declined 32%.
  - Source note: Decline linked to rising stocks and expectation that suspended production from the Magellan mine in Australia will be allowed to resume in the first quarter of 2008 (as reported by the World Bank “Commodity Markets Review”).
- Nickel
  - Year of Event: 2001.
  - Type of Event: Positive price shock.
  - Finding: Nickel prices increased by 44%.
  - Source note: Supply problems noted, including technical problems bringing on new capacity in Australia and labor strikes in Canada.

### B.5 Country-Specific Assumptions — exclusions from the analysis
- The following country-episode exclusions were implemented because events were due to unique weather conditions or political incidents exclusive to a specific country:
  - The cocoa price surge of 2002 in Côte d’Ivoire (attempted coup and civil war) is omitted for that country.
  - The sugar price shock in 2006 (drought in Thailand) is not considered for Thailand.
  - The 2019 disruption to iron ore prices (collapse of a mining dam in Brazil) is excluded for Brazil.
  - The 2010 spike in cereal prices (weather in Russia, Kazakhstan, and Ukraine) results in these countries’ exclusion from that event.
  - The cotton price shock in 2010 (weather-related incidents in Pakistan) is disregarded for Pakistan.
  - The rubber price shock in 2010 (droughts in Thailand) leads to Thailand’s exclusion for that event.

### C Additional Results — first-stage statistics and inference
- Table C1: First-stage F-Statistic for Alternative Instrument Sets
  - Commodity Events: 60.157 (No Controls); 10.68 (With Controls)
  - U.S. Monetary Policy: 35.422 (No Controls); 56.38 (With Controls)
  - Global Risk Appetite: 38.756 (No Controls); 73.75 (With Controls)
  - Note: All statistics are significant at the 1% level, with F >10, indicating that the instruments are not weak (Staiger and Stock, 1997).

### C Additional Results — impulse-response evidence (summary)
- General notes on impulse-response figures (Figures C1–C12):
  - The impulse responses present LATE estimates for one standard deviation shocks:
    - Increase in export prices driven by commodity price shocks (Px).
    - Decline in the BAA spread driven by U.S. monetary policy shocks.
    - Decline in the BAA spread driven by shifts in global risk appetite.
  - Capital flow measures are presented as a ratio with respect to the (quadratic) trend of GDP denominated in U.S. dollars.
  - Confidence intervals reported: 68% and 90% (gray areas, dashed lines, or equivalent).
- Specific figure families and robustness checks:
  - Figures C1–C4: Baseline LATE impulse responses; comparisons with OLS and exclusions of potentially contaminated events (1998 and 2008).
  - Figures C5–C6: Sensitivity to subset of events (Table C2) and single-country drop procedures.
  - Figures C7–C11: Effects of declines in the BAA spread (U.S. monetary policy and global risk appetite) including single-country drop robustness and full sample vs. post-2000 comparisons.
  - Figure C12: Effects of increase in export prices driven by energy-commodity-specific shocks; alternative instrument constructed only including events related to energy commodities.
- Table C2: Subset of Events (list of episodes used by Di Pace et al. (2020))
  - Selected entries (year — commodity — sign — source of shock):
    - 1993 — Timber — + — Clinton’s environmentally friendly policies
    - 1993 — Tobacco — - — Worldwide increase in competition for exports
    - 1994 — Aluminum — + — Reduction in stocks of major producing countries
    - 1994 — Coffee — + — Frost in Brazil
    - 1994 — Cotton — + — Decline in production due to bad weather in key producing countries
    - 1997 — Cereals/Food — - — Favorable production forecast
    - 1998 — Crude oil — - — Expectations of higher supply
    - 2000 — Natural gas — + — California gas crisis
    - 2002 — Cocoa — + — Attempted coup in Cote d’Ivoire
    - 2003 — Cotton — + — Severe weather damage in China
    - 2005 — Natural gas — + — Effects of hurricanes Katrina and Rita
    - 2010 — Cereals/Food — + — Adverse weather conditions in key producing countries
    - 2010 — Cotton — + — Negative weather shocks in the U.S. and Pakistan
    - 2015 — Energy — - — Booming in U.S. shale oil production
    - 2019 — Energy (excluding crude oil) — - — The U.S. became a net energy exporter

### Appendix references (selected)
- De Winne, J. and G. Peersman (2016). Macroeconomic Effects of Disruptions in Global Food Commodity Markets: Evidence for the United States. Brookings Papers on Economic Activity 47(2 (Fall)), 183–286.
- Di Pace, F., L. Juvenal, and I. Petrella (2020). Terms-of-Trade Shocks are Not all Alike. CEPR Discussion Papers 14594, C.E.P.R. Discussion Papers.
- Janzen, J. P., A. Smith, and C. A. Carter (2018). Commodity Price Comovement and Financial Speculation: The Case of Cotton. American Journal of Agricultural Economics 100(1), 264–285.
- Staiger, D. and J. H. Stock (1997). Instrumental Variables Regression with Weak Instruments. Econometrica 65(3), 557–586.
- World Bank, G. (2001). Global Economic Prospects. Technical report, Washington, D.C.

*Unveiling the Dance of Commodity Prices and the Global Financial Cycle — Working Paper No. WP/2024/082*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024082-print-pdf.pdf_
