## 1. It shows how the opposite sign patterns for the interest rate differential and expected

## Source details

**Canonical URL:** [1. It shows how the opposite sign patterns for the interest rate differential and expected](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026162-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026162-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026162-source-pdf.pdf.json)

---

### Identification strategy and sign restrictions
- Single structural shock identified: only the sign restrictions associated with the financial shock—the first row of Table1—are imposed on the SVAR.
- Minimal set of sign restrictions imposed on impact responses of observables:
  - UIP premium
  - interest rate differentials (r_t − r*_t)
  - expected depreciation Et(e_{t+1} − e_t)
  - capital inflows (Net Capital Inflows)
  - nominal exchange rate (e_t − e_{t−1})
- Restrictions distinguish:
  - Financial shocks: rise in the UIP premium, depreciation of the exchange rate, capital outflows.
  - Fundamental shocks: different comovement patterns.
- Identification hinges on the opposite comovement between the interest rate differential and expected depreciation under financial shocks (versus same-sign comovement under fundamental shocks); this comovement is a robust qualitative implication across models and does not depend on specific calibration.
- Clarification: UIP defined as UIP_t ≡ (r_t − r*_t) − Et(∆e_{t+1}); UIP can rise or fall depending on relative magnitudes of (r_t − r*_t) and Et(∆e_{t+1}) and calibration.

### Model impulse-response patterns (Figure 1) and Table 1 sign matrix
- Model-implied qualitative patterns (baseline calibration in Itskhoki and Mukhin (2021)):
  - Financial Shock: UIP rises; r − r* shows opposite sign pattern relative to expected depreciation; nominal exchange rate depreciates; net capital inflows fall.
  - Productivity Shock and Monetary Shock: comovements differ from the financial shock across UIP, capital inflows, and exchange-rate measures.
- Table 1 (Sign Restrictions to Identify Financial Shock) — imposed signs for a positive (adverse) financial shock and model-implied signs for fundamental shocks:
  - Financial Shock: Net Capital Inflows ↓ ; UIP ↑ ; Interest Differential (r_t − r*_t) ↑ ; Expected Depreciation Et(e_{t+1} − e_t) ↓ ; Depreciation (e_t − e_{t−1}) ↑
  - Productivity Shock: Net Capital Inflows ↑ ; UIP ↑ ; Interest Differential ↑ ; Expected Depreciation ↑ ; Depreciation ↓
  - Monetary Shock: Net Capital Inflows ↓↓ ; UIP ↑↑ ; Interest Differential ↑ ; Expected Depreciation ↑ ; Depreciation ↓

### Narrative restrictions and addressing the "shock masquerading problem"
- Sign restrictions provide set identification rather than point identification; identified set can be large.
- Shock masquerading problem (Wolf, 2022): linear combinations of other shocks can satisfy sign restrictions and masquerade as the shock of interest.
- Strategies to tighten identification:
  - Impose further restrictions on dynamic responses of key variables.
  - Impose the full row of restrictions corresponding to the financial shock, including nominal depreciation and net capital inflows.
  - Exploit narrative restrictions: anchor shocks using historical episodes where the sign of a given structural shock is known with high confidence (Antolin-Diaz & Rubio-Ramirez, 2018).
- Narrative constraints act as anchors that substantially mitigate the masquerading problem and enhance credibility of identification.

### Narrative restrictions — Case study: Chile’s October–November 2019 Social Uprising
- Contemporaneous central bank diagnostics and market evidence supporting a positive (adverse) financial shock in October–November 2019:
  - Sharp and rapid depreciation of the peso; December 2019 Monetary Policy Report: “the movements in the peso were essentially idiosyncratic” and documented “a substantial decoupling since the start of the crisis” between observed exchange rate and an empirical model based on usual fundamentals.
  - Conditions posing a tangible threat to financial stability: deteriorating liquidity and market stress; FX market volatility peaked near CLP 830 per US dollar on November 28.
- Documented financial stress and policy actions:
  - FX volatility: about 15 percent depreciation by end-November relative to previous month (December MP Report panels).
  - Liquidity deterioration: widening onshore spread (proxy of CIP deviations) and deposit-swap spread; spike in bond yields and term spreads; unwindings by residents.
  - Central Bank liquidity and FX interventions:
    - November 13–14: 30- and 90-day FX swap program; peso repo facility at floating rate linked to MPR (traditionally the MPR plus 25 bp); authorization of new repo collateral.
    - November 28: announced FX intervention program—spot dollar sales via auctions and forward operations, each for up to USD 10 billion, scheduled between December 2019 and May 2020—an amount equivalent to roughly half of the stock of FX reserves at the time. Operations sterilized as in prior interventions.
  - Central Bank assessment: liquidity provision measures “had a significant effect on long-term rates and bank and corporate bond spreads. In the case of the onshore spread, the program managed to reduce the spread from the recorded peaks to around 200 basis points”; “exchange rate volatility has fallen significantly since the start of the program”.
- Narrative conclusion: peso sharp depreciation, decoupling from usual fundamentals, spike in funding spreads, and deterioration of liquidity conditions are consistent with an adverse financial shock; BCCh’s contemporaneous assessments justify imposing a narrative restriction that the financial shock in October and November 2019 is positive.

### Extending narrative anchors across time and countries (Chile 2000–2023)
- Rule for identifying sharp rapid depreciations: months in which the change in the nominal exchange rate exceeds its average monthly change by more than 1.5 standard deviations.
- Appendix overview (period 2000–2023):
  - Chile: 12 episodes identified (statistical criterion).
  - Chile: 5 financial shock episodes identified by narrative-evaluation procedure.
  - Brazil: 11 episodes meet the statistical criterion; 3 financial shock episodes identified by narrative-evaluation procedure.
- Narrative selection procedure: keyword pre-selection (exchange rate, peso, depreciation, currency, devaluation, FX, forward, spot, swap); zero-shot BART large MNLI classification with entailment threshold 0.75; manual independent evaluation by four co-authors with binary scoring (1/0) on two criteria (large rapid depreciation; evidence consistent with non-fundamental drivers). Episode classified as financial shock if all evaluators assign 1 to Criterion 1 and majority assign 1 to Criterion 2.

### Illustrative Chile episodes classified as financial shocks (five episodes)
- 1. June–Aug 2001 — Argentina default spillovers (peso depreciated 20% from December 2000 to mid-August).
- 2. Sep–Nov 2008 — Global Financial Crisis (Lehman collapse); peso’s sharp overreaction, FX market dysfunction, central bank liquidity interventions and dollar swaps.
- 3. Oct–Nov 2019 — Social Uprising (peso plunged by around 6% over the week in mid-November; central bank FX swap program and interventions).
- 4. Feb 2020–Aug 2020 — COVID pandemic (corporate and bank bond spreads increased over 75 bp in March and 40 bp in the year).
- 5. Jun–Aug 2022 — FX volatility episode (July 2022 intervention announced citing idiosyncratic market tensions).
- Notes: appendix presents full narratives; robustness check using UIP premia captured all exchange-rate-based episodes and additional ones.

### Quantitative counts and procedural thresholds (preserved exactly)
- Chile: 12 episodes identified for the period 2000 to 2023 (statistical identification).
- Chile: 5 financial shock episodes identified by the narrative-evaluation procedure.
- Brazil: 11 episodes meet the statistical criterion for exchange rate depreciation (period 2000–2023).
- Brazil: 3 financial shock episodes identified by the narrative-evaluation procedure.
- Keyword relevance threshold for paragraph entailment: 0.75.
- Excerpt scoring: binary (1/0).
- Example spread and spread-change figures cited in narratives:
  - Peso plunged by around 6% against the USD over the week (J.P. Morgan, 2019, p. 1).
  - From December 2000 to mid-August, the observed exchange rate depreciated by 20% (Central Bank of Chile, 2001, p. 26).
  - Corporate and bank bond spreads increased over 75 bp in March and 40 bp in the year (Central Bank of Chile, 2020b, p. 51).

---

### Monthly-frequency SVAR: variables, estimation, priors, and identification
- Frequency and sample:
  - Monthly frequency to capture both macro fundamentals and high-frequency financial frictions.
  - Sample: monthly from January 1998 to June 2024.
  - Missing values handled via Kalman filter latent-state estimation.
- SVAR observable vector y_t (ten variables):
  1. UIP Premium (i_t − i^*_t − (E_t(S_{t+1}) − S_t))
  2. Interest rate differential (i_t − i^*_t)
  3. Inflation (% change in CPI)
  4. Nominal Depreciation (% change (m-o-m) in bilateral nominal exchange rate vs USD, increase means depreciation of peso)
  5. Output (year on year change in monthly output index, seasonally adjusted)
  6. CIP deviation (%) (i_t − i^*_t − (F_t(S_{t+1}) − S_t))
  7. Net capital inflows (nominal, USD)
  8. FX bid-ask spread (spot, %)
  9. Foreign exchange intervention (FXI): net purchase of foreign currency (Nominal, USD)
  10. Policy rate (%)
- Model configuration:
  - Included 12 lags in VAR.
  - Output proxies: IMACEC non-mining index for Chile and Brazilian counterpart; inflation by CPI.
  - CIP deviations and FX bid–ask spreads aggregated monthly from daily sources.
- Estimation and priors:
  - Bayesian methods with conjugate Minnesota-type normal–inverse-Wishart prior via dummy observations.
  - Uniform (Haar) prior over orthogonal rotation matrices mapping reduced-form to structural shocks.
  - Draws retained when satisfying sign restrictions; narrative restrictions implemented by importance-weighted reweighting of admissible draws.
  - Inference based on 2,000 retained draws (robust to 10,000+).
  - Prior hyperparameters preserved exactly:
    - Minnesota overall tightness λ = 0.20
    - Decay parameter α = 2
    - Dummy-initial-observation tightness θ = 0.01
    - Tightness of sum-of-coefficients prior μ = 1
    - Prior on Σ: Inverse-Wishart with scale matrix S_0 = v_0 · Ψ, degrees of freedom v_0 = n + 2
  - All variables except those in levels (bid-ask spread, net capital flows and policy rate) assigned IID priors instead of random walk.
  - Posterior reporting: posterior medians and 68 percent credible intervals for impulse responses; medians for FEVDs and historical decompositions.

### Identification details implemented
- Baseline sign restrictions based on UIP decomposition and expected-appreciation restriction.
- Narrative sign restrictions anchored on documented episodes (Chile: 2001, 2008, October–November 2019, 2022; Brazil: country-specific episodes).
- Additional imposed restriction: financial shock must raise UIP premium by more than the interest-rate differential on impact (implementing expected-appreciation restriction via UIP identity).
- Importance-weighted accept–reject and reweighting algorithm implemented (Gibbs sampling with sampling–importance–resampling).

---

### Results — Chile (sign-only vs narrative-augmented)
- Sign-only VAR (Chile):
  - UIP premium, interest rate differential, and nominal depreciation move in expected directions after one standard deviation financial shock.
  - Responses of output, inflation, and FX bid-ask spread muted and largely indistinguishable from zero under sign-only identification.
- Adding single narrative restriction for October/November 2019:
  - Core financial responses broadly consistent.
  - Output exhibits a clear and sizable decline (contrasting with near-zero effect under sign-only).
  - Narrative restriction eliminates rotations that combine other shocks and masquerade identification.
- Adding full set of narrative restrictions (Chile episodes: 2001, 2008, 2019, 2022):
  - Negative effect on output remains robust and sizable.
  - FX bid–ask spread increases following the shock, consistent with tightening liquidity and FX market frictions.
  - Shock entails a rise in inflation and the policy rate, consistent with exchange rate pass-through and an active monetary response.
  - FXI response suggests use of foreign exchange intervention to lean against depreciation (sell foreign exchange reserves).
- Financial block dynamics (four components focus):
  - Exchange rate depreciates sharply following the shock, generating a positive UIP premium.
  - Interest rate differential widens, signaling tighter domestic financial conditions.
  - Expectations of further depreciation increase (expected depreciation term).
  - Mechanism: greater demand for foreign-currency bonds relative to domestic bonds, implying home-currency bonds must offer a positive expected return via interest differential and expected appreciation terms reinforcing rather than offsetting.
- FEVD (preferred specification with narratives, Chile):
  - Non-fundamental shock accounts for approximately one third of the variance of the UIP premium.
  - Accounts for about one half of nominal exchange rate (depreciation) fluctuations.
  - Accounts for less than ten percent for macroeconomic aggregates such as output and inflation.
- Historical decomposition (Chile, preferred specification):
  - Financial shock exhibits episodic dominance with large positive contributions during stress episodes: 2008/9 Global Financial Crisis, 2019 social uprising, 2022 volatility spike.
  - Near-zero or negative contributions during tranquil periods.
  - Decomposition shows the identified shock captures disorderly market dynamics rather than systematic fundamentals.
- Robustness to global factors:
  - Identified shock comoves positively but only moderately with standard global financial cycle drivers (VIX, Miranda-Agrippino–Rey global factor, and the dollar); no correlation exceeds 0.5.
  - Essentially uncorrelated with identified U.S. monetary-policy and policy-uncertainty shocks.
  - Conditioning on global factors by adding them as exogenous regressors leaves impulse responses and variance decompositions essentially unchanged.

### Results — Brazil and cross-country comparison
- Brazil impulse responses (sign-only vs sign + narrative):
  - Narrative restrictions sharpen inference and deliver a more pronounced short-run decline in output relative to sign-only, though less stark than Chile.
- Similarities across Chile and Brazil (preferred specification with full narratives):
  - Comparable magnitudes of UIP premium increases, nominal depreciation, and short-run declines in output following a financial shock.
  - Output response negative and persistent in both countries.
- Key cross-country differences:
  - Chile: shock is inflationary on impact; policy rate rises promptly.
  - Brazil: shock displays a mildly deflationary response on impact; policy rate eases.
  - Financial stress indicators differ:
    - Chile: FX bid–ask spread widens sharply.
    - Brazil: most pronounced reaction in CIP deviations.
- FEVD comparison at one-month horizon:
  - In both economies, the financial shock explains a substantially larger share of the variance of the UIP premium and the nominal exchange rate than of macro aggregates.
  - Contrast sharper in Chile; Brazil shows a less pronounced gap.
  - Similar pattern holds at 20 months horizon.

### Mechanism and theoretical mapping
- Pattern maps to intermediary–constraint models:
  - Limited risk-bearing capacity of FX intermediaries raises shadow cost of supplying currency risk → higher UIP premium (Gabaix and Maggiori, 2015).
  - Financial shock tightens balance sheet constraints and induces portfolio rebalancing toward foreign-currency debt → interest differential and expected depreciation reinforce one another (Itskhoki and Mukhin, 2021).
- Empirical findings relative to literature:
  - Financial shocks dominate exchange rate dynamics but have limited average influence on macro aggregates (consistent with “exchange rate disconnect” literature), though less extreme than some structural models.
  - Identified shocks are episodically dominant for FX variables, especially during stress and intervention windows.

### Policy-relevant findings and implications
- Financial shocks account for substantial shares of FX premia and nominal exchange rate fluctuations while being modest contributors to average macro aggregate variance:
  - Chile: roughly one-third of UIP variance and about one-half of monthly nominal depreciation variance are due to identified non-fundamental shocks.
- Tightening financial conditions that raise currency premia induce statistically and economically meaningful short-run contractions in economic activity.
- Monetary policy responses vary by country:
  - Chile: monetary tightening (policy rate rises) to contain price pressures following depreciation-driven inflationary shock.
  - Brazil: policy easing in face of demand-driven deflationary pressure following financial shock.
- Foreign exchange intervention (FXI) empirically observed as response to shocks (selling reserves to reduce depreciation) in Chile.
- Narrative sign restrictions materially sharpen identification and can change inference on the real effects of financial shocks.

### Caveats and directions for future research
- Trade-offs and limitations:
  - Approach trades deep structural DSGE mapping for disciplined sign restrictions and narrative anchors; delivers shock-level counterfactuals but not full general-equilibrium welfare mapping.
  - Identification remains set-valued; results sensitive to variable selection, horizon, and lag length.
  - Quality of narrative classification is crucial and can limit scaling to additional countries.
  - SVAR is linear with time-invariant coefficients; cannot separately identify a "pure shallowness" shock or accommodate state-dependent amplification.
- Promising extensions:
  - Enhanced narrative extraction via advanced NLP and machine learning.
  - Integrating higher-frequency financial data and exploring alternative identification schemes (e.g., max-share or agnostic approaches).
  - Linking identified financial shocks to welfare-relevant policy interventions (FX swaps, liquidity backstops).

---

### Robustness checks — Is the identified financial shock a copper (terms-of-trade) shock?
- Data: global monthly price of copper (USD per metric ton); ∆m pCu t monthly log change and zt standardized log level. Sample: 1999m2–2024m6 (n = 305 months).
- Subsample construction:
  - (a) UIP up & copper down: months with ∆UIPt > 0 and ∆m pCu t < 0 (n months 80).
  - (b) UIP down & copper up: months with UIP falling and copper rising (n months 99).
- Model-implied FEVD, UIP from financial shock:
  - h = 1 — 0.283 [0.146, 0.446]
  - h = 12 — 0.236 [0.128, 0.370]
  - h = 60 — 0.215 [0.118, 0.328]
- Empirical impact-only shares imp_fin (anchored to FEVD h = 1):
  - Full sample, 1999m2–2024m6, n months 305 — 0.280 [0.130, 0.453]
  - (a) UIP up & Copper down, n months 80 — 0.292 [0.129, 0.499]
  - (b) UIP down & Copper up, n months 99 — 0.288 [0.134, 0.473]
- Empirical cumulative shares cum_fin (anchored to FEVD h = ∞):
  - Full sample, 1999m2–2024m6, n months 305 — 0.193 [0.106, 0.305]
  - (a) UIP up & Copper down, n months 80 — 0.196 [0.096, 0.337]
  - (b) UIP down & Copper up, n months 99 — 0.189 [0.101, 0.303]
- Interpretation:
  - Financial shock’s share of UIP movements essentially unchanged on copper-consistent subsamples (impact shares 0.292 and 0.288 vs full-sample 0.280; cumulative shares 0.196 and 0.189 vs full-sample 0.193).
  - Conclusion: conditional subsample evidence does not support a copper (terms-of-trade) shock interpretation.

### Robustness: including copper directly in the VAR
- Specification: included contemporaneous standardized log copper price zt as exogenous regressor in every equation; sign and narrative restrictions, lag length, and priors held identical.
- Findings:
  - Impulse responses lie on top of each other across all variables and horizons; posterior bands essentially indistinguishable.
  - FEVD change at h = 1: UIP FEVD baseline 0.283 → copper-augmented 0.267 (a 1.6 percentage-point change).
  - Shares for nominal depreciation and output virtually identical across specifications.
- Interpretation: identified financial shock picks up variation not driven by copper prices; conditioning on copper leaves impulse responses and variance shares essentially unchanged.

### Role of global factors (comovement and conditioning)
- Comovement (Chile baseline specification, sign + five narrative restrictions, sample 1999m1–2024m6):
  - Positive contemporaneous correlations with global risk/credit and the dollar:
    - VIX (∆), n = 306: Pearson 0.313 ∗∗∗ (Spearman 0.270 ∗∗∗)
    - Global factor, Miranda-Agrippino–Rey (∆), n = 306: Pearson 0.277 ∗∗∗ (Spearman 0.237 ∗∗∗)
    - Dollar index, DXY (%∆), n = 306: Pearson 0.280 ∗∗∗ (Spearman 0.243 ∗∗∗)
    - Broad nominal dollar (%∆), n = 221: Pearson 0.467 ∗∗∗ (Spearman 0.420 ∗∗∗)
  - Essentially uncorrelated with identified U.S. monetary-policy and policy-uncertainty shocks (every correlation |·| < 0.08; examples: policy surprise, 6-month, n = 305: 0.045; policy-uncertainty shock, n = 305: −0.000).
  - Interpretation: comoves positively and significantly with broad indicators of global financial cycle but correlations moderate (around 0.28–0.31); not dominated by U.S. monetary policy.
- Conditioning on global factors within the VAR:
  - Two principal components used as exogenous regressors: first PC from global financial-conditions block; first PC from monetary-policy block.
  - Re-estimated VAR with both components included.
  - Findings:
    - Impulse responses nearly identical with and without the two global-financial-variable factors.
    - FEVD at h = 1: UIP baseline 0.28 → conditioned 0.26; nominal depreciation baseline 0.49 → conditioned 0.47; other shares remain stable.
  - Interpretation: conditioning on global financial variables and U.S. monetary policy leaves the identified shock essentially unchanged.

Final synthesis of robustness checks
- Multiple checks (copper-consistent subsample analysis; copper included as exogenous regressor; comovement analysis with global variables; conditioning the VAR on global-factor and monetary-policy factors) show:
  - The identified financial shock overlaps only modestly with copper-price fluctuations and is not mechanically a terms-of-trade (copper) shock.
  - The shock comoves with global financial conditions but is not dominated by them and is essentially uncorrelated with identified U.S. monetary-policy surprises.
  - Conditioning on copper or on global-factor and monetary-policy factors leaves impulse responses and FEVD shares of the financial shock essentially unchanged.

*Source: wpiea2026162-source-pdf — Financial Shocks in Currency Markets: Evidence from UIP Premia — Working Paper No. WP/2026/162*

### 1. It shows how the opposite sign patterns for the interest rate differential and expected

### 1. It shows how the opposite sign patterns for the interest rate differential and expected

### Identification strategy and sign restrictions
- Single structural shock is identified: only the sign restrictions associated with the financial shock—the first row of Table1—are imposed on the SVAR.
- Productivity and monetary rows in Table1 report model-implied sign patterns of fundamental disturbances but are not imposed as identifying restrictions on additional shocks.
- Minimal set of sign restrictions imposed on impact responses of observables:
  - UIP premium
  - interest rate differentials (r_t − r*_t)
  - expected depreciation Et(e_{t+1} − e_t)
  - capital inflows (Net Capital Inflows)
  - nominal exchange rate (e_t − e_{t−1})
- These restrictions distinguish:
  - Financial shocks: rise in the UIP premium, depreciation of the exchange rate, capital outflows.
  - Fundamental shocks: different comovement patterns.
- Identification hinges on the opposite comovement between the interest rate differential and expected depreciation under financial shocks (versus same-sign comovement under fundamental shocks); this comovement is a robust qualitative implication across models and does not depend on specific calibration.

### Model impulse-response patterns (Figure 1)
- Figure 1 panels show model-implied impulse responses to a one-standard-deviation shock of the indicated type; responses are expressed in percentage deviations from steady state and based on the baseline calibration in Itskhoki and Mukhin (2021).
- Qualitative patterns illustrated:
  - (a) Financial Shock: UIP rises; r − r* shows opposite sign pattern relative to expected depreciation; nominal exchange rate depreciates; net capital inflows fall.
  - (b) Productivity Shock: different UIP and capital inflow patterns; comovements among variables differ from financial shock.
  - (c) Monetary Shock: UIP and expected depreciation move differently than under financial shock.
- Appendix Figure17: financial shock has similar qualitative pattern in terms of components of the UIP premium in Adrian et al. (2021).

### Table 1 — Sign restrictions to identify financial shock
- Table 1 (Sign Restrictions to Identify Financial Shock) reports imposed signs for a positive (adverse) financial shock and model-implied signs for fundamental shocks:
  - Financial Shock: Net Capital Inflows ↓ ; UIP ↑ ; Interest Differential (r_t − r*_t) ↑ ; Expected Depreciation Et(e_{t+1} − e_t) ↓ ; Depreciation (e_t − e_{t−1}) ↑
  - Productivity Shock: Net Capital Inflows ↑ ; UIP ↑ ; Interest Differential ↑ ; Expected Depreciation ↑ ; Depreciation ↓
  - Monetary Shock: Net Capital Inflows ↓↓ ; UIP ↑↑ ; Interest Differential ↑ ; Expected Depreciation ↑ ; Depreciation ↓
- Notes: sign restrictions imposed to identify a positive (adverse) financial shock based on Itskhoki and Mukhin (2021).
- Clarification: UIP defined as UIP_t ≡ (r_t − r*_t) − Et(∆e_{t+1}); UIP can rise or fall depending on relative magnitudes of (r_t − r*_t) and Et(∆e_{t+1}) and calibration.

### Narrative restrictions and the "shock masquerading problem"
- Sign restrictions provide set identification rather than point identification; identified set of admissible structural impulse responses can be large.
- Shock masquerading problem (Wolf, 2022): linear combinations of other shocks can satisfy sign restrictions and masquerade as the shock of interest.
- Strategies to tighten identification:
  - Impose further restrictions on dynamic responses of key variables (reduce dimensionality of identified set).
  - Impose the full row of restrictions corresponding to the financial shock in Table 1, including nominal depreciation and net capital inflows.
  - Exploit narrative restrictions: anchor shocks using historical episodes where the sign of a given structural shock is known with high confidence (Antolin-Diaz & Rubio-Ramirez, 2018).
- Narrative constraints act as anchors that substantially mitigate the masquerading problem and enhance credibility of identification.

### Narrative restrictions — Case study: Chile’s 2019 Social Uprising
- Two distinctive features identified in contemporaneous central bank accounts:
  - Sharp and rapid depreciation of the peso.
    - December 2019 Monetary Policy Report: “the movements in the peso were essentially idiosyncratic” (Central Bank of Chile, 2019, p. 22).
    - Report documented “a substantial decoupling since the start of the crisis” between observed exchange rate and an empirical model based on usual fundamentals (Central Bank of Chile, 2019, p. 25).
  - Conditions posing a tangible threat to financial stability (deteriorating liquidity and market stress).
- Chile monetary policy context:
  - Inflation-targeting regime with a floating exchange rate; MPR set so inflation forecasts converge to 3 percent within policy horizon.
  - Inflation historical average of 3.8%.
  - Central Bank may intervene in FX market under exceptional circumstances; interventions aim at limiting overreactions that could jeopardize price stability or payments functioning, not at setting exchange-rate levels.
- Social Uprising timeline and features:
  - Protests began with subway-fare hike of 3 percent on October 6, 2019, escalated into nationwide riots on October 18.
  - Unrest persisted into November; political agreement on November 15 to draft new constitution.
  - Aruoba et al.,2024 identify three characteristics: unexpected but relatively short-lived (Google searches surged almost one hundredfold between October 17 and 19, before subsiding by mid-November), nationwide with varying intensity, sharp rise in uncertainty in surveys and market indicators.
  - FX market volatility remained elevated after the political agreement, peaking near CLP 830 per US dollar on November 28.
- Central Bank’s contemporaneous diagnostics (December 2019 MP Report):
  - Panel a) documented sharp depreciation in bilateral and multilateral exchange rates beginning mid-October—about 15 percent by end-November relative to previous month.
  - Panel b) showed peso depreciation idiosyncratic to Chile vs. other Latin American currencies and commodity exporters.
  - Panel c) showed widening gap between observed peso and model-implied exchange rate (model includes copper prices, oil prices, domestic and U.S. price levels, CDS spreads on Chilean sovereign bonds, and one-year interest rate differential). Model estimation uses data from July 15th - December 3rd, 2019.
  - Central Bank characterized the episode as “excessive volatility or sudden movements in the exchange rate, which could affect the healthy adjustment of the economy and trigger market turmoil”.
- Financial conditions deterioration documented:
  - Increase in stock market volatility, peso depreciation, rise in fixed-income rates, money market interest rates in both pesos and dollars; EMBI and CDS spreads rose mildly.
  - Deterioration in liquidity in dollars and pesos: increase in onshore spread (proxy of CIP deviations) and deposit-swap spread (banks’ funding costs in pesos net of MPR expectations) driven by portfolio movements.
  - Figures 15a and 15b in MP report depict deterioration in onshore and time deposit-swap spreads.
  - Spike in bond yields and term spreads; widening of corporate bond spreads across credit-risk categories; unwindings of local fixed-income positions by residents (pension funds and mutual funds).
- Central Bank policy reactions:
  - November 13-14: liquidity provision measures including a 30- and 90-day FX swap program; peso repo facility through collateralized window at floating rate linked to MPR (traditionally the MPR plus 25 bp); authorization of repo operations backed by securities not previously eligible.
  - November 28: announced FX intervention program—spot dollar sales via auctions and forward operations, each for up to USD 10 billion, scheduled between December 2019 and May 2020—an amount equivalent to roughly half of the stock of FX reserves at the time. Operations sterilized as in prior interventions.
  - December MP Report noted liquidity provision measures “had a significant effect on long-term rates and bank and corporate bond spreads. In the case of the onshore spread, the program managed to reduce the spread from the recorded peaks to around 200 basis points” (Central Bank of Chile, 2019, p. 26). The CBCh reported that “exchange rate volatility has fallen significantly since the start of the program” (Central Bank of Chile, 2020a, p. 29).
- Narrative conclusion:
  - Peso sharp depreciation, decoupling from usual fundamentals, spike in funding spreads, and deterioration of liquidity conditions are consistent with an adverse financial shock in the framework.
  - The BCCh’s contemporaneous assessments justify imposing a narrative restriction that the financial shock in October and November 2019 is positive, anchoring identification in policy assessments of market conditions.

### Extending the narrative approach across time and countries
- Approach to extend across Chile’s history since 2000:
  - Identify months of sharp and rapid exchange rate depreciation defined as months in which the change in the nominal exchange rate exceeds its average monthly change by more than 1.5 standard deviations.

*Source: wpiea2026162-source-pdf - 1. It shows how the opposite sign patterns for the interest rate differential and expected (IMF working paper content provided).*

### Appendix lists all identified episodes. Focusing on the period from 2000 to 2023, we identify

### wpiea2026162-source-pdf - Appendix lists all identified episodes. Focusing on the period from 2000 to 2023, we identify

### Overview
- Focusing on the period from 2000 to 2023, we identify 12 episodes for Chile.
- The narrative search focuses on depreciation episodes because:
  - Documentary records are more informative for sharp depreciations (liquidity strains, disorderly trading, overshooting).
  - Anchoring narrative restrictions to depreciation episodes does not restrict the sign of the shock in the linear SVAR; the same identified shock can take negative values and generate appreciations when financial conditions move in the opposite direction.
- Central Bank of Chile (BCCh) reports and private market analyst reports published during or soon after identified episodes were collected to capture contemporaneous assessments and investors’ perceptions and reactions.

### Documentary selection and automated filtering procedure
- Manual subsample review to identify informative paragraphs for the research question.
- Keyword-based pre-selection set (final): exchange rate, peso, depreciation, currency, devaluation, FX, forward, spot, swap.
  - A conservative approach to keyword expansion was adopted to limit unrelated paragraphs.
- Zero-shot classification using a pre-trained transformer model (BART large MNLI) trained on natural language inference tasks.
  - Each paragraph evaluated against eight guiding questions (see below).
  - A paragraph is classified as relevant if the associated entailment score exceeds a threshold of 0.75 for at least one guiding question.
- Guiding questions used for paragraph entailment evaluation:
  1. Does the paragraph discuss a change in the exchange rate?
  2. Does it explain the reasons behind exchange rate movements?
  3. Does it assess economic fundamentals?
  4. Does it explore global and local drivers of currency fluctuations?
  5. Does it highlight FX market imbalances?
  6. Does it examine interest rate differentials?
  7. Does it consider divergence in monetary policies?
  8. Does it explain the reasons behind exchange rate depreciation?
- Iterative refinement of keyword set and guiding questions based on comparisons with manually identified excerpts to align automated classification with economically meaningful narrative content while limiting false positives.

### Manual evaluation and classification into financial-shock episodes
- Each selected excerpt was manually and independently evaluated by all four co-authors using two criteria:
  - Criterion 1: The excerpt discusses the relevant episode and indicates a large and rapid exchange rate depreciation.
  - Criterion 2: The excerpt provides evidence consistent with a financial shock driven by non-fundamental drivers (i.e., it does not reject the non-fundamental nature of the drivers).
- Excerpts scored binary: 1 indicates presence of the criterion, 0 indicates absence.
- An episode was classified as a financial shock if:
  - All evaluators assigned a score of one to Criterion 1, and
  - The majority assigned a score of one to Criterion 2.
- This procedure identified five financial shock episodes for Chile (including the 2019 Social Uprising) and three financial shock episodes for Brazil.

### Chile — Episodes identified (financial shocks) and illustrative narrative excerpts
- Five financial shock episodes for Chile (summarized in Table 2, left column):
  1. June–Aug 2001 — Argentina default spillovers
     - Excerpt (Central Bank of Chile, 2001, p. 26):
       - “During the present year, the peso has depreciated substantially against the dollar. Thus, from December 2000 to mid-August, the observed exchange rate depreciated by 20%. This was mainly due to economic instability in the region’s countries (Argentina and, to a lesser degree, Brazil) and the drop in the terms of trade, particularly the depressed price for copper. Thus, the nominal exchange rate has been the main adjustment variable in dealing with recent turbulence abroad, which is exactly what is to be expected when using a free floating exchange rate regime such as the one operating in Chile since 1999”
       - “The substantial weakening of the peso during this period disturbed financial markets. The speed and quick succession of movements that led to a decline in conditions abroad exacerbated peso volatility and depreciation, despite the strength of Chile’s economy and the coherence of its macroeconomic policies” (Central Bank of Chile, 2001, p. 26).
  2. Sep–Nov 2008 — Global Financial Crisis (Lehman’s Collapse)
     - Narratives emphasize peso’s sharp overreaction, FX market dysfunction, and central bank liquidity interventions and dollar swap programs (Central Bank of Chile, 2008a; 2008b).
     - Excerpts note abnormal onshore spread between Libor and domestic dollar rates and that forward exchange market did not adequately reflect expectations.
  3. Oct–Nov. 2019 — Social Uprising
     - Market commentary (J.P. Morgan, 2019, p. 1) excerpt:
       - “Uncertainty weighed heavily on asset prices as violent social protests in Chile extended for another week. The protests are having an increasingly pronounced impact on financial markets, particularly the exchange rate. The peso plunged by around 6% against the USD over the week. The BCCh, a central bank averse to discretionary intervention in the FX market, intervened verbally first and later announced a preemptive FX swap program to provide liquidity to the market. The monetary authority pointed out that in a context of great uncertainty it makes sense to see heightened volatility in the FX market. Yet, this must be contrasted with the fundamentals of the economy: a solvent financial system, limited currency exposure of economic agents, a solid fiscal position (low public debt), adequate international reserves and sovereign wealth funds, and anchored inflation expectations.”
  4. Feb 2020–Aug 2020 — COVID pandemic
     - Narratives emphasise pronounced exchange-rate volatility, heightened stress in financial markets, and provision of dollar and peso liquidity.
     - Excerpts (Central Bank of Chile, 2020b, pp. 16, 51) note reduction in commodity prices, increase in risk aversion, and spreads increases (corporate and bank bond spreads increased over 75 bp in March and 40 bp in the year).
  5. Jun–Aug 2022 — FX volatility episode
     - BCCh rationale for July 2022 intervention (Central Bank of Chile, 2022, p. 25):
       - “By mid-July, the depreciation process was accompanied by a sharp increase in exchange rate volatility at a time when agents were finding it difficult to determine the equilibrium price. The abrupt instability in the FX market did not occur in other countries in the region, suggesting that local market tensions were of idiosyncratic nature. This behavior, and the risks it posed to the performance of other markets, such as fixed income, motivated the decision to announce an intervention and liquidity provision program on 14 July.”
     - Market-analyst perspective (Goldman Sachs, 2022, p. 14):
       - “The sheer magnitude of the CLP depreciation in recent weeks and rapidly rising risk that distressed FX market price dynamics could destabilize the broader financial system (beyond adding more fuel to an already very hot inflation backdrop) led the central bank to overcome the initial reluctance and move to intervene in the FX market. This is a welcome development, particularly if coupled with a decisive conventional monetary policy strategy.”
- Notes:
  - The appendix presents the full extent of narratives identified.
  - Robustness check: analysis repeated using uncovered interest parity (UIP) premia instead of exchange rate changes. The exchange rate-based approach captured all episodes identified by the UIP premia, as well as additional ones, providing a less conservative measure of depreciation episodes.

### Brazil — Episodes identified (financial shocks) and illustrative narrative excerpts
- Statistical identification: 11 episodes meet the statistical criterion for exchange rate depreciation for Brazil (period 2000–2023).
- Applying the narrative criteria identified three financial shock episodes (Table 2, right column):
  1. May–Jun 2002 — Lula election uncertainty / Political Uncertainty
     - Goldman Sachs (2002, p. 1) excerpt:
       - “Over the last two weeks, Brazilian financial markets sold off strongly. The turbulence was caused in part by a rise in political risk, with local financial institutions growing increasingly uncomfortable about holding domestic public debt maturing after January 2003, when the new administration takes office. Investors worry about fiscal sustainability and the lack of specific economic policy proposals from the presidential candidates for permanently resolving this problem. The turbulence was also due to financial and regulatory developments that altered the balance in local financial markets, a balance shaken by higher political risk and a complex domestic public debt structure.”
     - Banco Central do Brasil (2003, p. 9) excerpt:
       - “In the second half of 2002, the Brazilian economy faced a crisis of confidence in the domestic and international scenarios of relevant magnitude. Internally, the uncertainties about the evolution of the electoral process and the future of the economic policy contaminated the financial markets and the capital flows. The deterioration in the perception of credit risk and the increase in prices volatility had affected public securities and the exchange rate, promoting stress.”
  2. Sep–Nov 2008 — Global Financial Crisis (Lehman Collapse)
     - Banco Central do Brasil (2008a, p. 142) excerpt:
       - “The prospects for the monetary policy in emerging countries became even more complex in the last weeks, when several currencies, not only from commodity-exporting countries, significantly depreciated in relation to the US$, in the midst of an increase of volatility and risk aversion indicators in international financial markets.”
     - Banco Central do Brasil (2008b, p. 163) excerpt:
       - “After Lehman Brothers declared insolvent, instability increased in the international financial markets, once risk aversion and lack of liquidity affected more decisively real economy funding, mainly in the US. The financial crisis assumed a more generalized aspect worldwide, spreading geographically. Credit channels were obstructed by uncertainty and liquidity and solvency problems that emerged subsequently.”
  3. Feb–Mar 2020 — COVID (Financial instability and dysfunctional FX market)
     - Market commentary (Goldman Sachs, 2020, p. 3) excerpt:
       - “Overall, given the Copom cautious posture, already accommodative Selic level, pressure on the BRL and rising risk-premia, our call is for the Copom to cut, but significantly less than FOMC cumulative easing up to the March 18 meeting date. Extraordinary high uncertainty, dysfunctional FX market pricing dynamics in recent years, and broken markets in general, make it particularly difficult to make decisions with a high degree of confidence.”

### Quantitative counts and procedural thresholds (preserved exactly)
- Chile: 12 episodes identified for the period 2000 to 2023 (statistical identification).
- Chile: 5 financial shock episodes identified by the narrative-evaluation procedure.
- Brazil: 11 episodes meet the statistical criterion for exchange rate depreciation (period 2000–2023).
- Brazil: 3 financial shock episodes identified by the narrative-evaluation procedure.
- Keyword relevance threshold for paragraph entailment: 0.75.
- Excerpt scoring: binary (1/0).
- Example spread and spread-change figures cited in narratives:
  - Peso plunged by around 6% against the USD over the week (J.P. Morgan, 2019, p. 1).
  - From December 2000 to mid-August, the observed exchange rate depreciated by 20% (Central Bank of Chile, 2001, p. 26).
  - Corporate and bank bond spreads increased over 75 bp in March and 40 bp in the year (Central Bank of Chile, 2020b, p. 51).

*Source: wpiea2026162-source-pdf - Appendix lists all identified episodes. Focusing on the period from 2000 to 2023, we identify (source PDF).*

### 2. We work at a monthly frequency and estimate

### 2. We work at a monthly frequency and estimate country-specific models for Chile and Brazil

### Motivation for monthly frequency and sample
- Monthly frequency chosen to capture both macroeconomic fundamentals and high-frequency financial frictions:
  - Quarterly data would obscure high-frequency dynamics of financial shocks.
  - Daily data would preclude inclusion of key macro aggregates.
- Monthly aggregation mechanically attenuates financial disturbances that mean-revert within the month; identification isolates disturbances large enough to leave a measurable monthly footprint.
- UIP premium and CIP premia constructed at a three-month horizon but sampled monthly; consecutive observations refer to overlapping windows.
- Sample: monthly from January 1998 to June 2024.
- Missing values handled via a Kalman filter based approach to estimate and fill in missing values as latent states.

### Variables and transformations (SVAR observable vector y_t stacks ten macro-financial variables)
- 1. UIP Premium (i_t − i^*_t − (E_t(S_{t+1}) − S_t))
- 2. Interest rate differential (i_t − i^*_t)
- 3. Inflation (% change in CPI)
- 4. Nominal Depreciation (% change (m-o-m) in bilateral nominal exchange rate vs USD, increase means depreciation of peso)
- 5. Output (year on year change in monthly output index, seasonally adjusted)
- 6. CIP deviation (%) (i_t − i^*_t − (F_t(S_{t+1}) − S_t))
- 7. Net capital inflows (nominal, USD)
- 8. FX bid-ask spread (spot, %)
- 9. Foreign exchange intervention (FXI): net purchase of foreign currency (Nominal, USD)
- 10. Policy rate (%)
- Definitions and sources provided (Table 3); IMF/central bank/Haver/Refinitiv/Consensus Forecast used as data sources for specific series.
- Included 12 lags in VAR to allow sufficient flexibility in dynamic relationships.
- Output proxies: IMACEC non-mining index for Chile and Brazilian counterpart; inflation by CPI.
- CIP deviations and FX bid–ask spreads included as market-functioning barometers; available at daily frequency though aggregated monthly here.
- Policy reactions captured via short-term policy rate and foreign exchange intervention.

### Estimation approach and role of priors
- Model estimated using Bayesian methods, following Rubio-Ramírez et al. (2010) and Antolin-Diaz and Rubio-Ramirez (2018).
- Conjugate Minnesota-type normal–inverse-Wishart prior on reduced-form parameters implemented through dummy observations.
- Uniform (Haar) prior over orthogonal rotation matrices mapping reduced-form innovations into structural shocks.
- Draws retained when they satisfy sign restrictions (Table 1); narrative restrictions imposed by reweighting admissible draws with importance weights.
- Unbalanced series handled with a data-augmentation step; inference based on 2,000 retained draws (results robust to increasing retained draws to 10,000 or more).
- Prior hyperparameters (computational details):
  - Minnesota overall tightness λ = 0.20
  - Decay parameter α = 2
  - Dummy-initial-observation tightness θ = 0.01
  - Tightness of sum-of-coefficients prior μ = 1
  - Prior on Σ: Inverse-Wishart with scale matrix S_0 = v_0 · Ψ, degrees of freedom v_0 = n + 2
- All variables except those in levels (bid-ask spread, net capital flows and policy rate) are assigned IID priors instead of random walk.
- Posterior reporting: posterior medians and 68 percent credible intervals for impulse responses; medians for FEVDs and historical decompositions.
- Identification is set-valued; discussion of the uniform rotation prior's role:
  - Haar prior is necessary and sufficient for the joint prior/posterior over the identified set of impulse responses to be uniform.
  - Narrative restrictions affect admissibility and importance weights and thus the posterior over rotations, limiting the rotation prior's influence.

### Identification via sign and narrative restrictions
- Baseline sign restrictions based on UIP decomposition and expected-appreciation restriction; include sign restrictions to isolate financial shocks.
- Narrative sign restrictions anchored on documented episodes (Chile: 2001, 2008, October–November 2019, 2022; Brazil: country-specific episodes listed in Appendix tables).
- Narrative restrictions implemented as additional constraints that sharpen identification and eliminate rotations that masquerade the true shock.
- Identification imposes that the financial shock must raise the UIP premium by more than the interest-rate differential on impact (implementing the expected-appreciation restriction via the UIP identity).
- Importance-weighted accept–reject and reweighting algorithm implemented (Gibbs sampling with sampling–importance–resampling).

### Results — Chile (baseline and narrative-augmented)
- Sign-only VAR (Chile):
  - UIP premium, interest rate differential, and nominal depreciation move in expected directions after a one standard deviation financial shock.
  - Responses of output, inflation, and FX bid-ask spread are muted and largely indistinguishable from zero under sign-only identification.
- Adding a single narrative restriction for October/November 2019 social uprising:
  - Core financial responses remain broadly consistent.
  - Output exhibits a clear and sizable decline (contrasting with near-zero effect under sign-only).
  - Narrative restriction eliminates rotations that combine other shocks and masquerade identification.
- Adding full set of narrative restrictions (Chile episodes: 2001, 2008, 2019, 2022):
  - Negative effect on output remains robust and sizable.
  - FX bid–ask spread increases following the shock, consistent with tightening liquidity and FX market frictions.
  - Shock entails a rise in inflation and the policy rate, consistent with exchange rate pass-through and an active monetary response.
  - FXI response suggests use of foreign exchange intervention to lean against depreciation (sell foreign exchange reserves).
- Financial block dynamics (four components focus):
  - Exchange rate depreciates sharply following the shock, generating a positive UIP premium.
  - Interest rate differential widens, signaling tighter domestic financial conditions.
  - Expectations of further depreciation increase (expected depreciation term).
  - Mechanism: greater demand for foreign-currency bonds relative to domestic bonds, implying home-currency bonds must offer a positive expected return via interest differential and expected appreciation terms reinforcing rather than offsetting.
- FEVD (preferred specification with narratives, Chile):
  - Non-fundamental shock accounts for approximately one third of the variance of the UIP premium.
  - Accounts for about one half of nominal exchange rate (depreciation) fluctuations.
  - Accounts for less than ten percent for macroeconomic aggregates such as output and inflation.
- Historical decomposition (Chile, preferred specification):
  - Financial shock exhibits episodic dominance with large positive contributions during stress episodes: 2008/9 Global Financial Crisis, 2019 social uprising, 2022 volatility spike.
  - Near-zero or negative contributions during tranquil periods.
  - Decomposition shows the identified shock captures disorderly market dynamics rather than systematic fundamentals.
- Robustness to global factors:
  - Identified shock comoves positively but only moderately with standard global financial cycle drivers (VIX, Miranda-Agrippino–Rey global factor, and the dollar); no correlation exceeds 0.5.
  - Essentially uncorrelated with identified U.S. monetary-policy and policy-uncertainty shocks.
  - Conditioning on global factors by adding them as exogenous regressors leaves impulse responses and variance decompositions essentially unchanged.

### Results — Brazil and cross-country comparison
- Brazil impulse responses (sign-only vs sign + narrative):
  - Narrative restrictions sharpen inference and deliver a more pronounced short-run decline in output relative to sign-only, though less stark than Chile.
- Similarities across Chile and Brazil (preferred specification with full narratives):
  - Comparable magnitudes of UIP premium increases, nominal depreciation, and short-run declines in output following a financial shock.
  - Output response is negative and persistent in both countries, indicating real effects of financial disturbances.
- Key cross-country differences:
  - Chile: shock is inflationary on impact; policy rate rises promptly, consistent with exchange rate pass-through and monetary tightening.
  - Brazil: shock displays a mildly deflationary response on impact; policy rate eases, suggesting demand-side contraction dominates near-term price dynamics.
  - Financial stress indicators differ:
    - Chile: FX bid–ask spread widens sharply, signaling liquidity strains in spot markets.
    - Brazil: most pronounced reaction in CIP deviations, pointing to stress in cross–currency funding markets; limited spot bid–ask movement possibly due to price discovery in futures markets in Brazil.
- FEVD comparison at one-month horizon:
  - In both economies, the financial shock explains a substantially larger share of the variance of the UIP premium and the nominal exchange rate than of macro aggregates.
  - Contrast sharper in Chile; Brazil shows a less pronounced gap indicating more diffuse propagation or stronger role for other drivers.
  - Similar pattern holds at 20 months horizon (Appendix Figure 19).

### Interpretation, mechanism, and theoretical mapping
- Identified impulse response pattern maps closely to intermediary–constraint models:
  - Limited risk-bearing capacity of FX intermediaries raises shadow cost of supplying currency risk → higher UIP premium (Gabaix and Maggiori, 2015).
  - Financial shock tightens balance sheet constraints and induces portfolio rebalancing toward foreign-currency debt → interest differential and expected depreciation reinforce one another (Itskhoki and Mukhin, 2021).
- Empirical findings relative to literature:
  - Financial shocks dominate exchange rate dynamics but have limited average influence on macro aggregates (consistent with “exchange rate disconnect” literature), though less extreme than some structural models.
  - Identified shocks are episodically dominant for FX variables, especially during stress and intervention windows.

### Policy-relevant findings and implications
- Financial shocks account for substantial shares of FX premia and nominal exchange rate fluctuations while being modest contributors to average macro aggregate variance:
  - Chile: roughly one-third of UIP variance and about one-half of monthly nominal depreciation variance are due to identified non-fundamental shocks.
- Tightening financial conditions that raise currency premia induce statistically and economically meaningful short-run contractions in economic activity.
- Monetary policy responses vary by country context:
  - Chile: monetary tightening (policy rate rises) to contain price pressures following depreciation-driven inflationary shock.
  - Brazil: policy easing in face of demand-driven deflationary pressure following financial shock.
- Foreign exchange intervention (FXI) observed empirically as a response to shocks (selling reserves to reduce depreciation) in Chile; this behavior is consistent with models where intervention is used to lean against financial stress.
- Narrative sign restrictions materially sharpen identification and can change inference on the real effects of financial shocks; credible, well-documented narrative anchors are valuable for policy-relevant identification.

### Caveats and directions for future research
- Trade-offs and limitations:
  - Approach trades deep structural DSGE mapping for disciplined sign restrictions and narrative anchors; delivers shock-level counterfactuals but not full general-equilibrium welfare mapping or policy rules.
  - Identification remains set-valued; results sensitive to variable selection, horizon, and lag length.
  - Quality of narrative classification is crucial and can limit scaling to additional countries.
  - SVAR is linear with time-invariant coefficients; cannot separately identify a "pure shallowness" shock orthogonal to the financial shock nor accommodate state-dependent amplification.
- Promising extensions:
  - Enhanced narrative extraction via advanced NLP and machine learning to improve credibility and precision of empirical identification.
  - Integrating higher-frequency financial data and exploring alternative identification schemes (e.g., max-share or agnostic approaches) to sharpen inference on real effects.
  - Linking identified financial shocks to welfare-relevant policy interventions (FX swaps, liquidity backstops) for normative implications.

*Source: wpiea2026162-source-pdf — 2. We work at a monthly frequency and estimate country-specific models for Chile and Brazil (PDF chapter).*

### 2019. Source: Central Bank of Chile.

### wpiea2026162-source-pdf - 2019. Source: Central Bank of Chile.

### F. Is the Identified Financial Shock a Copper (Terms-of-Trade) Shock?
- Research question: whether the financial shock identified in Section 4 is actually a commodity (copper) / terms-of-trade shock in disguise.
- Measurement: global monthly price of copper (USD per metric ton); ∆m pCu t denotes its monthly log change and zt its standardized log level. Sample: 1999m2–2024m6 (n = 305 months).

Findings (subsample comparison using historical decomposition and FEVD anchors)
- Construction: two copper-consistent subsamples:
  - (a) UIP up & copper down: months with ∆UIPt > 0 and ∆m pCu t < 0.
  - (b) UIP down & copper up: months with UIP falling and copper rising.
- Two empirical share statistics anchored to FEVD benchmarks:
  - scum_fin(S) and simp_fin(S) as defined in equation (F.1) (cumulative and impact-only shares).
- Model-implied FEVD, UIP from financial shock (Section 4):
  - h = 1 (headline impact share)—0.283 [0.146, 0.446]
  - h = 12—0.236 [0.128, 0.370]
  - h = 60 (≈∞, long horizon)—0.215 [0.118, 0.328]
- Empirical impact-only shares imp_fin (anchored to FEVD h = 1):
  - Full sample, 1999m2–2024m6, n months 305 — 0.280 [0.130, 0.453]
  - (a) UIP up & Copper down, n months 80 — 0.292 [0.129, 0.499]
  - (b) UIP down & Copper up, n months 99 — 0.288 [0.134, 0.473]
- Empirical cumulative shares cum_fin (anchored to FEVD h = ∞):
  - Full sample, 1999m2–2024m6, n months 305 — 0.193 [0.106, 0.305]
  - (a) UIP up & Copper down, n months 80 — 0.196 [0.096, 0.337]
  - (b) UIP down & Copper up, n months 99 — 0.189 [0.101, 0.303]

Interpretation
- The financial shock’s share of UIP movements is essentially unchanged on copper-consistent subsamples (impact shares 0.292 and 0.288 vs full-sample 0.280; cumulative shares 0.196 and 0.189 vs full-sample 0.193).
- A copper-shock explanation would predict markedly elevated shares precisely on these windows; this is not observed.
- Conclusion: the conditional subsample evidence does not support a copper (terms-of-trade) shock interpretation of the identified financial shock.

### F.2 Robustness: Including Copper Directly in the VAR
- Specification: re-estimated VAR with the contemporaneous standardized log copper price zt included as an exogenous regressor in every equation (equation (F.2)), holding sign and narrative restrictions, lag length, and priors identical to baseline.
- Rationale: identified financial shock is orthogonal to contemporaneous copper movements under this specification; if the financial shock were essentially copper-driven, impulse responses and FEVD shares should shrink.

Findings
- Impulse responses: the impulse responses to the financial shock lie on top of each other across all variables and horizons; posterior bands essentially indistinguishable between baseline and copper-augmented VAR (Figure 20).
- FEVD change at h = 1:
  - UIP FEVD: baseline 0.283 → copper-augmented 0.267 (a 1.6 percentage-point change).
  - Shares for nominal depreciation and output are virtually identical across specifications (Figure 21).

Interpretation
- The identified financial shock picks up variation not driven by copper prices; conditioning on copper leaves impulse responses and variance shares essentially unchanged.
- Conclusion: the financial shock is not mechanically a terms-of-trade (copper) shock.

### G. Identifying Financial Shocks in Chile: The Role of Global Factors
- Purpose: characterize the identified financial shock’s comovement with observable drivers of the global financial cycle and test robustness to conditioning on these drivers inside the VAR.

G.1 Comovement with global financial conditions and U.S. monetary policy
- Data: posterior-median identified financial shock from baseline Chile specification (sign restrictions + five narrative restrictions), sample 1999m1–2024m6.
- A positive realization interpreted as contractionary: raises UIP premium, depreciates the currency, coincides with net capital outflows.
- Correlations computed contemporaneously with 22 monthly series across four families: global risk and credit, the dollar, U.S. interest rates, and identified U.S. monetary-policy and uncertainty shocks. Series oriented so higher values denote tighter U.S. or global financial conditions. Table 7 reports Pearson and Spearman correlations.

Key correlation estimates (Pearson; sample sizes shown in Table 7)
- Global risk and credit:
  - VIX (∆), n = 306: 0.313 ∗∗∗ (Spearman 0.270 ∗∗∗)
  - Global factor, Miranda-Agrippino–Rey (∆), n = 306: 0.277 ∗∗∗ (Spearman 0.237 ∗∗∗)
  - Excess bond premium, n = 306: 0.108 ∗ (Spearman 0.072)
  - Gilchrist–Zakrajšek credit spread, n = 306: 0.083 (Spearman 0.072)
- U.S. dollar:
  - Dollar index, DXY (%∆), n = 306: 0.280 ∗∗∗ (Spearman 0.243 ∗∗∗)
  - Broad nominal dollar (%∆), n = 221: 0.467 ∗∗∗ (Spearman 0.420 ∗∗∗)
- U.S. interest rates (∆):
  - Federal funds rate, n = 306: −0.202 ∗∗∗ (Spearman −0.096 ∗)
  - 2-year Treasury yield, n = 306: −0.079 (Spearman −0.051)
  - 10-year Treasury yield, n = 306: −0.072 (Spearman −0.052)
  - Wu–Xia shadow rate, n = 294: 0.002 (Spearman 0.044)
- Identified U.S. monetary-policy shocks and uncertainty (various high-frequency surprises and uncertainty measures, n = 246–306):
  - Across ten high-frequency monetary-policy measures and two uncertainty measures, every correlation is smaller than 0.08 in absolute value and statistically indistinguishable from zero (examples: policy surprise, 6-month, n = 305: 0.045; policy-uncertainty shock, n = 305: −0.000).

Interpretation
- The identified financial shock comoves positively and significantly with broad indicators of the global financial cycle (VIX, global factor, DXY) but correlations are moderate (around 0.28–0.31 for VIX/global factor/DXY; 0.47 with broad nominal dollar) and do not exceed 0.5.
- The shock is essentially uncorrelated with directly identified U.S. monetary-policy and policy-uncertainty surprises (all correlations |·| < 0.08).
- Conclusion: the identified shock relates to global financial conditions without being a manifestation of U.S. monetary policy.

G.2 Robustness: conditioning on global factors within the VAR
- Method: summarize conditioning information in two principal components:
  - First principal component from global financial-conditions block (VIX, global factor, credit spreads, the dollar, U.S. interest rates).
  - First principal component from monetary-policy block (high-frequency surprise and uncertainty measures).
- Re-estimate VAR with both components included as contemporaneous exogenous regressors (equation (G.1)), under same sign and narrative restrictions.

Findings
- Impulse responses: nearly identical with and without the two global-financial-variable factors in the VAR (Figure 22).
- FEVD at h = 1 (impact shares):
  - UIP: baseline 0.28 → conditioned 0.26.
  - Nominal depreciation: baseline 0.49 → conditioned 0.47.
  - Other responses and variance shares remain stable (Figure 23).

Interpretation
- Conditioning on the global financial variables and U.S. monetary policy leaves the identified shock essentially unchanged.
- Conclusion: the identified financial shock is robust to controlling for the global financial cycle and U.S. monetary policy and is not largely a conditioning artifact of these global factors.

Final synthesis
- Multiple checks (copper-consistent subsample analysis; copper included as exogenous regressor; comovement analysis with global variables; conditioning the VAR on global-factor and monetary-policy factors) consistently show that:
  - The identified financial shock overlaps only modestly with copper-price fluctuations and is not mechanically a terms-of-trade (copper) shock.
  - The shock comoves with global financial conditions but is not dominated by them and is essentially uncorrelated with identified U.S. monetary-policy surprises.
  - Conditioning on copper or on global-factor and monetary-policy factors leaves impulse responses and FEVD shares of the financial shock essentially unchanged.

*Source: 2019. Source: Central Bank of Chile.*

### References

### References

### Methodology, Econometrics, and Identification
- Antolin-Diaz, J., & Rubio-Ramirez, J. F. (2018). Narrative sign restrictions for svars. American Economic Review, 108(10), 2802–29.
- Arias, J. E., Rubio-Ramírez, J. F., & Waggoner, D. F. (2018). Inference based on structural vector autoregressions identified with sign and zero restrictions: Theory and applications. Econometrica, 86(2), 685–720.
- Arias, J. E., Rubio-Ramírez, J. F., & Waggoner, D. F. (2025). Uniform priors for impulse responses. Econometrica, 93(2), 695–718.
- Baumeister, C., & Hamilton, J. D. (2015). Sign restrictions, structural vector autoregressions, and useful prior information. Econometrica, 83(5), 1963–1999.
- Inoue, A., & Kilian, L. (2020). The role of the prior in estimating VAR models with sign restrictions (Working Paper No. 2030). Federal Reserve Bank of Dallas.
- Rubio-Ramírez, J. F., Waggoner, D. F., & Zha, T. (2010). Structural vector autoregressions: Theory of identification and algorithms for inference. The Review of Economic Studies, 77(2), 665–696.
- Wolf, C. K. (2022). What can we learn from sign-restricted vars? AEA papers and proceedings, 112, 471–475.
- Koch, K. (2007). Gibbs sampler by sampling-importance-resampling. Journal of Geodesy, 81(9), 581–591.
- Baumeister, C., & Hamilton, J. D. (2015). Sign restrictions, structural vector autoregressions, and useful prior information. Econometrica, 83(5), 1963–1999.

### Exchange Rates, UIP/CIP, Risk Premia, and FX Intervention
- Albagli, E., Ceballos, L., Claro, S., & Romero, D. (2024). UIP deviations: Insights from event studies. Journal of International Economics, 148, 103877. https://doi.org/10.1016/j.jinteco.2024.103877
- Bacchetta, P., Benhima, K., & Berthold, B. (2023). Foreign exchange intervention with uip and cip deviations: The case of small safe haven economies (Discussion Paper No. DP18469). Centre for Economic Policy Research (CEPR).
- Backus, D. K., Foresi, S., & Telmer, C. I. (2001). Affine term structure models and the forward premium anomaly. Journal of Finance, 56(1), 279–304. https://doi.org/10.1111/0022-1082.00325
- Bocola, L., & Lorenzoni, G. (2020). Financial crises, dollarization, and lending of last resort in open economies. American Economic Review, 110(8), 2524–2557. https://doi.org/10.1257/aer.20180830
- Dao, M., & Gourinchas, P.-O. (2025). Covered interest parity in emerging markets: Measurement and drivers (Working Paper No. 2025/057). International Monetary Fund. https://doi.org/10.5089/9798229003995.001
- Dao, M., Gourinchas, P.-O., & Itskhoki, O. (2025). Breaking parity: Equilibrium exchange rates and currency premia (Working Paper No. 2025/153) (IMF Working Paper Series). International Monetary Fund. https://doi.org/10.5089/9798229015615.001
- De Leo, P., Keller, L., & Zou, D. (2024). Speculation, forward exchange demand, and cip deviations in emerging economies (Research Paper) (Posted October 11, 2024; revised March 11, 2025). The Wharton School, University of Pennsylvania. https://doi.org/10.2139/ssrn.4949594
- Du, W., & Schreger, J. (2022). Cip deviations, the dollar, and frictions in international capital markets. In G. Gopinath, E. Helpman, & K. Rogoff (Eds.), Handbook of international economics: International macroeconomics, volume 6 (pp. 147–197). Elsevier. https://doi.org/10.1016/bs.hesint.2022.03.001
- Engel, C. (2016). Exchange rates, interest rates, and the risk premium. American Economic Review, 106(2), 436–474. https://doi.org/10.1257/aer.20121365
- Fanelli, S., & Straub, L. (2021). A theory of foreign exchange interventions. The Review of Economic Studies, 88(6), 2857–2885.
- Farhi, E., & Gabaix, X. (2016). Rare disasters and exchange rates. Quarterly Journal of Economics, 131(1), 1–52. https://doi.org/10.1093/qje/qjv040
- Gabaix, X., & Maggiori, M. (2015). International liquidity and exchange rate dynamics. Quarterly Journal of Economics, 130(3), 1369–1420. https://doi.org/10.1093/qje/qjv016
- Itskhoki, O., & Mukhin, D. (2021). Exchange rate disconnect in general equilibrium. Journal of Political Economy, 129(8), 2183–2232.
- Itskhoki, O., & Mukhin, D. (2023). Optimal exchange rate policy (tech. rep.). National Bureau of Economic Research.
- Miyamoto, W., Nguyen, T. L., & Oh, H. (2023). In search of dominant drivers of the real exchange rate. Review of Economics and Statistics, 105(4), 1–14. https://doi.org/10.1162/rest_a_01342
- Verdelhan, A. (2010). A habit-based explanation of the exchange rate risk premium. Journal of Finance, 65(1), 123–145. https://doi.org/10.1111/j.1540-6261.2009.01525.x
- Bodenstein, M., Cuba-Borda, P., Gornemann, N., & Presno, I. (2024). Exchange rate disconnect and the trade balance. International Finance Discussion Paper, (1391).
- Albagli, E., Ceballos, L., Claro, S., & Romero, D. (2024). UIP deviations: Insights from event studies. Journal of International Economics, 148, 103877. https://doi.org/10.1016/j.jinteco.2024.103877
- Chahrour, R., Cormun, V., De Leo, P., Guerrón-Quintana, P. A., & Valchev, R. (2025). Exchange rate disconnect revisited (Working Paper No. w32596) (Previously circulated as Exchange Rate Disconnect Redux). National Bureau of Economic Research. https://doi.org/10.3386/w32596

### Financial Markets, Term Premia, and International Capital Frictions
- Greenwood, R., Hanson, S., Stein, J. C., & Sunderam, A. (2023). A quantity-driven theory of term premia and exchange rates. Quarterly Journal of Economics, 138(4), 2327–2389. https://doi.org/10.1093/qje/qjad024
- Della Corte, P., Sarno, L., Schmeling, M., & Wagner, C. (2022). Exchange rates and sovereign risk [Published online October 1, 2021]. Management Science, 68(8), 5591–5617. https://doi.org/10.1287/mnsc.2021.4115
- Du, W., & Schreger, J. (2022). Cip deviations, the dollar, and frictions in international capital markets. In G. Gopinath, E. Helpman, & K. Rogoff (Eds.), Handbook of international economics: International macroeconomics, volume 6 (pp. 147–197). Elsevier. https://doi.org/10.1016/bs.hesint.2022.03.001
- Patel, N., & Xia, F. D. (2019). Offshore markets drive trading of emerging market currencies. BIS Quarterly Review, December.
- Hofmann, B., Patel, N., & Wu, S. P. Y. (2022). Original sin redux: A model-based evaluation. Bank for International Settlements, Monetary; Economic Department.

### Macro-Financial Policy, Small Open Economies, and Crises
- Adrian, M. T., Erceg, C. J., Kolasa, M., Lindé, J., & Zabczyk, P. (2021). A quantitative microfounded model for the integrated policy framework. International Monetary Fund.
- Basu, S., Boz, E., Gopinath, G., Ünsal, F., & Adrian, T. (2025). Integrated monetary and financial policies for small open economies. Econometrica.
- Das, M., Gopinath, G., & Kalemli-Özcan,. (2022). Preemptive policies and risk-off shocks in emerging markets (tech. rep. No. 2022/003). International Monetary Fund. https://doi.org/10.5089/9781616358341.001
- Bocola, L., & Lorenzoni, G. (2020). Financial crises, dollarization, and lending of last resort in open economies. American Economic Review, 110(8), 2524–2557. https://doi.org/10.1257/aer.20180830
- Fanelli, S., & Straub, L. (2021). A theory of foreign exchange interventions. The Review of Economic Studies, 88(6), 2857–2885.
- Jara, A., & Piña, M. (2022). Exchange rate volatility and the effectiveness of fx interventions: The case of chile (Working Paper No. 962). Central Bank of Chile.
- Arenas, J., & Griffith-Jones, S. (2023). Effectiveness of foreign exchange interventions: Evidence and lessons from chile (Working Paper No. 983). Central Bank of Chile.
- Banco Central do Brasil. (2003, May). Financial stability report (No. 1). Banco Central do Brasil. Brasília.
- Banco Central do Brasil. (2008a, September). Inflation report (No. 3). Banco Central do Brasil. Brasília.
- Banco Central do Brasil. (2008b, December). Inflation report (No. 4). Banco Central do Brasil. Brasília.

### Country Studies, Central Bank Reports, and Policy Notes
- Central Bank of Chile. (2001, September). Monetary policy report. Central Bank of Chile. Santiago.
- Central Bank of Chile. (2008a). Financial stability report (Second Half 2008). Central Bank of Chile. Santiago.
- Central Bank of Chile. (2008b, September). Monetary policy report. Central Bank of Chile. Santiago.
- Central Bank of Chile. (2019, December). Monetary policy report. Central Bank of Chile. Santiago.
- Central Bank of Chile. (2020a, January). Chile’s monetary policy within an inflation-targeting framework. Central Bank of Chile. Santiago.
- Central Bank of Chile. (2020b). Financial stability report (First Half 2020). Central Bank of Chile. Santiago.
- Central Bank of Chile. (2022, September). Monetary policy report. Central Bank of Chile. Santiago.
- Medel, C. (2018). Econometric analysis on survey-data-based anchoring of inflation expectations in chile (Working Paper No. 825). Central Bank of Chile.
- Garcia, M., Medeiros, M., & Santos, F. (2014). Price discovery in brazilian fx markets (tech. rep.). Texto para discussão.
- Goldman Sachs. (2002, June 14). Brazil: A few ideas to stabilize local financial markets (Latin America Economic Analyst No. 02/12). Goldman Sachs Global Investment Research.
- Goldman Sachs. (2020, March 18). Brazil: Copom: Restarting easing with a 50bp selic cut (LATAM Today). Goldman Sachs Global Investment Research.
- Goldman Sachs. (2022, July 15). Chile: A taxing tax reform and a new constitution that pleases neither left nor right (Latin America Economics Analyst). Goldman Sachs Global Investment Research.
- J.P. Morgan. (2019, November 15). Chile and ecuador (Economic Research, Global Data Watch). J.P. Morgan Securities LLC.
- Robitaille, P., Zhang, T., & Weisberg, B. (2024, December). How well-anchored are long-term inflation expectations in latin america (FEDS Notes). Board of Governors of the Federal Reserve System.
- Joignant, A., & Garrido-Vergara, L. (2025). Revisiting the chilean social uprising: Explanations, interpretations, and over-interpretations. Latin American Research Review, 1–12.

### Recent Working Papers, NBER, and AI / New Methods in Economics
- Acosta-Henao, M., Fernández, A., Gomez-Gonzalez, P., & Kalemli-Özcan,. (2025). Firm financing during sudden stops: Can governments substitute markets? (Working Paper No. 2025/072). International Monetary Fund. https://doi.org/10.5089/9798229005128.001
- Aruoba, S. B., & Drechsel, T. (2024). Identifying monetary policy shocks: A natural language approach (tech. rep.). National Bureau of Economic Research.
- Aruoba, S. B., Fernández, A., Guzmán, D., Pastén, E., & Saffie, F. (2024). Pricing under distress (NBER Working Paper No. 32538). National Bureau of Economic Research. https://doi.org/10.3386/w32538
- Clayton, C., Coppola, A., Maggiori, M., & Schreger, J. (2025). Geoeconomic pressure (tech. rep.). National Bureau of Economic Research.
- Dell, M. (2025). Deep learning for economists. Journal of Economic Literature, 63(1), 5–58.
- Fang, H., Li, M., & Lu, G. (2025). Decoding china’s industrial policies (tech. rep.). National Bureau of Economic Research.
- Korinek, A. (2025). Generative ai for economic research: Use cases and implications [Forthcoming]. Journal of Economic Literature.
- Kalemli-Özcan,., & Varela, L. (2025). Five facts about the uip premium (Working Paper No. No. 28923) (Revised January 2025). National Bureau of Economic Research.
- Chahrour, R., Cormun, V., De Leo, P., Guerrón-Quintana, P. A., & Valchev, R. (2025). Exchange rate disconnect revisited (Working Paper No. w32596) (Previously circulated as Exchange Rate Disconnect Redux). National Bureau of Economic Research. https://doi.org/10.3386/w32596
- Ottonello, P., Song, W., & Sotelo, S. (2024). An anatomy of firms political speech (tech. rep.). National Bureau of Economic Research.
- De Leo, P., Keller, L., & Zou, D. (2024). Speculation, forward exchange demand, and cip deviations in emerging economies (Research Paper) (Posted October 11, 2024; revised March 11, 2025). The Wharton School, University of Pennsylvania. https://doi.org/10.2139/ssrn.4949594

*Financial Shocks in Currency Markets: Evidence from UIP Premia — Working Paper No. WP/2026/162*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026162-source-pdf.pdf_
