## Financial Stress and Economic Activity: Evidence from a New Worldwide Index

## Source details

**Canonical URL:** [Financial Stress and Economic Activity: Evidence from a New Worldwide Index](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023217-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023217-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023217-print-pdf.pdf.json)

---

### Introduction and motivation
- Builds on Romer and Romer (2017) with three modifications:
  - Source: uses Economist Intelligence Unit (EIU) country reports to extend coverage to 110 countries and change frequency from semi-annual to quarterly for 1967-2018.
  - Method: mechanistic approach using search algorithms and word counts plus expert judgement to measure intensity, enabling semi-automatic updates and improved cross-country/time comparability.
  - Exogeneity: identifies episodes of financial stress stemming from other countries to construct an external financial stress series for causality identification.
- Motivation: measure intensity and duration of financial stress beyond binary crisis chronologies and extend continuous narrative measures to developing economies.

### Data source and index construction
- Data source and rationale:
  - Sole narrative source: EIU country reports (average about 12,000 words; quarterly availability; EIU covers about 180 countries).
  - Coverage for index: 110 countries, 1967-2018 (sample restricted to countries with population above 2 million).
- Conceptual definition:
  - Episodes where an economy experiences an increase in the cost of credit intermediation or disruptions to the credit supply, excluding credit reductions due to “normal” cyclical monetary policy tightening.
- Four-step construction:
  1. Keyword search for paragraphs containing set (i): credit, financial, bank, lending, fund; and set (ii): crisis, crunch, squeeze, bailout, rescue, tight, contract, reluctant.
  2. Human reading to confirm contemporaneous financial stress and to classify domestic vs externally originated stress; text-search produced over 50 percent false positives requiring manual evaluation (about half of substantive work involved careful reading and validation).
  3. Cross-validation by IMF country economists (corrections in a few cases: Ecuador, Nepal, Venezuela).
  4. Summation of verified signals each period, scaled by total number of words in each report to account for report length variation.

### Comparisons and validation
- Pairwise correlations with existing measures:
  - Correlation with RR for OECD countries: 0.9.
  - Correlation with Laeven and Valencia (LV): 0.4.
  - Correlation with Reinhart and Rogoff (ReRo): 0.4.
- Coverage and frequency advantages relative to RR, ReRo, LV; single-source standardization improves cross-country comparability.
- Global movements captured:
  - Rises during Latin America debt crisis (1980s), Mexican Peso Crisis (mid-1990s), late 1990s Asia/Russia/Latin America crises, Global Financial Crisis (GFC) and Europe’s sovereign debt crisis (2008-2013).
- FSI positively correlated with IMF Financial Condition Indices (FCIs): correlation about 0.45.
- Validation note: for OECD countries the new index essentially mimics RR (correlation 0.9), supporting the effectiveness of the search algorithm and word-count approach.

### Stylized patterns, heterogeneity, and regional findings
- Averages over 1967-2018 (Panel A, Figure 3):
  - advanced economies: 0.033
  - emerging economies: 0.025
  - low-income economies: 0.010
- Number of quarters with financial stress (normalized by number of countries) (Figure 3c):
  - emerging economies: 20.7 quarters
  - advanced economies: 15.7 quarters
  - low-income economies: 13.7 quarters
- Example cross-sectional observation (2008Q4):
  - FSI close to global average in emerging economies;
  - below global average in low-income economies;
  - about three times global average in advanced economies.
- Regional patterns:
  - Africa and Middle-East and Central-Asia: little financial stress.
  - Asia-Pacific: stress during the Asian crisis and during the GFC.
  - Western Hemisphere: elevated FSI during early 1980s, late 1990s, and GFC.
  - Europe: captures GFC and European sovereign debt crisis.
- FSI vs binary chronologies:
  - FSI captures intensity and duration better than binary indices in many cases (e.g., Argentina, Philippines, Nigeria).
  - Discrepancies exist (e.g., Honduras severe stress 1980-1985 detected by FSI but not LV; Korea stress identified by ReRo 1985-1989 not detected by FSI).

### Baseline empirical methodology (country-level)
- Method: local projection method (Jordà, 2005).
- Baseline regression:
  - y_{i,t+k} = α_i^k + γ_t^k + sum_{j=0}^2 β_j^k ΔF_{i,t−j} + sum_{j=0}^2 θ_j^k y_{i,t−j} + ε_t^k
  - i indexes countries; t quarters; k horizon; y is log of output; ΔF denotes change in financial stress.
- Samples:
  - Quarterly baseline: unbalanced panel of 49 countries, 1996Q1–2018Q4.
  - Annual robustness/extension: unbalanced panel of 110 countries, 1967–2018.
- Estimation: OLS by horizon k = 0,...,20; standard errors clustered at country level.

### Baseline country-level estimates and main macro findings
- Lagged output coefficient close to 1, consistent with non-stationary level of GDP; country fixed effects capture average growth rates.
- One-standard deviation increase in FSI (equivalent to 0.1 changes in the index) associated with:
  - reduction in the level of output by 0.35 percent one year after the increase in financial stress;
  - reduction in the level of output by 0.2 percent 5 years after.
- Quarterly baseline point estimates (selected, Table 3):
  - F_{i,t}:
    - k=0: -0.103*** (0.0363)
    - k=5: -0.338*** (0.0675)
    - k=10: -0.405*** (0.0818)
    - k=15: -0.393*** (0.0827)
    - k=20: -0.235*** (0.0782)
- Interpretation example: peak increase in US financial stress observed during the GFC (1.7 in the fourth quarter of 2018) would be associated with a reduction in US GDP by about 6 percent in 2019.
- Annual data (110-country sample) find larger effects:
  - one-standard deviation increase in FSI associated with:
    - reduction in level of output by 0.8 percent one year after;
    - reduction in level of output by 0.6 percent 5 years after.
  - Note: larger annual estimates may reflect larger reverse causality with annual data.

### Heterogeneity: AEs vs EMDEs and non-linearities
- Advanced Economies (AE) vs Emerging Market and Developing Economies (EMDE):
  - Response of GDP to an equal increase in financial stress is more than twice larger in EMDEs than in AEs.
  - Heterogeneity attributed to differences in economic resilience, including fiscal and monetary policy space.
- Non-linearities by stress severity (terciles: low, medium, high):
  - Effects small and not statistically significant for lower levels of financial stress.
  - Effects large and more precisely estimated for medium-to-high levels.
  - Differences between low vs. medium and low vs. high are statistically significant up to k = 11.
  - Non-linearity more pronounced in EMDEs than in AEs; in AEs significance appears only at high stress levels.
- Alternative specifications confirm state-dependence:
  - Binary split at median; smooth transition functions of level of financial stress.

### Endogeneity, external instrument, and IV results
- Instrument: external financial stress indicator (EF_it) identified from EIU narratives as episodes of stress originating outside each country.
- IV system:
  - First stage: ΔF_it = π_i + τ_t + ϖ ΔEF_it + Σ ρ_j,k ΔF_it−j + Σ σ_j,k y_it−j + μ_t
  - Second stage: predicted ΔF_it̂ used in main equation.
- First-stage diagnostics:
  - Kleibergen–Paap rk Wald F statistic ≈ 1200 (about 75 times the Stock-Yogo critical value 16.38) — instrument is strong and statistically significant.
- IV findings:
  - IV results support OLS findings: significant negative and persistent effect of financial stress.
  - IV coefficient magnitudes are smaller in absolute value than OLS estimates, suggesting OLS coefficients are downward biased (simultaneity with weak economic activity intensifying financial stress).
- Instrument validity checks:
  - Including EF as additional control: coefficients on EF small and not significant; baseline FSI coefficients remain negative and significant (e.g., F_{i,t} k=5: -0.421*** (0.102)).
  - Regression of baseline residuals on ΔEF yields coefficients indistinguishable from zero.
  - Excluding the United States and other G7 economies one at a time: results confirm instrument validity.
  - Controlling for GDP growth in major trading partners: results corroborate instrument validity.
  - Re-estimating using external financial stress directly as main regressor yields qualitatively unchanged results.

### Channels: macro outcomes beyond GDP
- Re-estimated local projections for alternative dependent variables; key channel findings (Figure 19):
  - Labor productivity: declines by about 2 percent after one year following a one-standard-deviation increase in financial stress.
  - Employment: declines about 1 year after the shock.
  - Unemployment rate: increases by about 0.1 percentage point 1 year after the shock.
  - Financial markets and uncertainty: short-term decline in stock market growth and short-lived increases in stock market volatility and uncertainty.

### Firm-level analysis: data, methodology, and findings
- Data:
  - Main dataset: S&P Capital IQ (CIQ) quarterly firm balance sheet and income statement information.
  - Coverage: 150 countries from 1950Q1 to 2021Q2; analysis sample restricted to 2001Q1 onwards and to 75 advanced and emerging market economies.
  - Sample restrictions: non-financial corporations only; remove firms with negative assets or debt; exclude real estate and insurance companies.
  - Investment measure: capital expenditures (IQ_CAPEX-2021).
- Empirical methodology:
  - Two-step approach using Jordà (2005) local projections at firm level (equations (6) and (7)).
  - Difference-in-differences via interactions ΔF * D_n where D_n indicates firm below/above median (profitability, leverage, size, age).
  - Fixed effects: country-sector-quarter and firm-quarter; standard errors two-way clustered on firm and country-time.
  - IV: external financial stress indicator used as instrument in firm-level IV estimations.
- Main firm-level quantitative findings:
  - A one-standard deviation increase in financial stress is associated with an average reduction in firms’ investment of about 30 percent after 12 quarters.
  - Heterogeneity: less profitable and more financially constrained firms experience an additional investment loss; peak differential effect for constrained/low-profit firms about 10 percent.
  - Results robust to IV estimation using external financial stress indicator.

### Robustness checks and alternative specifications
- Robustness to sample exclusions (Figure 9): outliers; high inflation episodes (inflation above 20 percent); observations from the Americas; Asian and Sub-Saharan African economies; small countries (population below two million); episodes of large changes in financial stress (99th percentile); excluding observations after 2008Q3.
- Robustness to additional controls (Figure 10): changes in monetary policy rates; changes in cyclically adjusted budget balance; stock market growth and volatility; economic uncertainty.
- Alternative frequency and chronologies:
  - Annual data produces larger estimated effects (discussed above).
  - Re-estimation with alternative chronologies (RR, LV, ReRo) yields results in same ballpark for RR; FSI effects about one-tenth smaller than those from LV and ReRo (binary chronologies capture more extreme events).

### Key quantitative magnitudes (selected, preserved exactly)
- Correlation with RR for OECD countries: 0.9.
- One-standard deviation increase in FSI (equivalent to 0.1 change in the index) → output level reductions:
  - 0.35 percent one year after;
  - 0.2 percent 5 years after.
- Quarterly baseline coefficients (F_{i,t}):
  - k=0: -0.103*** (0.0363)
  - k=5: -0.338*** (0.0675)
  - k=10: -0.405*** (0.0818)
  - k=15: -0.393*** (0.0827)
  - k=20: -0.235*** (0.0782)
- Annual baseline (full sample, F_{i,t}):
  - K=0: -0.473*** (0.154)
  - K=1: -0.830*** (0.203)
  - K=4: -0.829*** (0.190)
- External financial shock coefficients (Table 5, EF_{i,t−1}):
  - k=0: -1.470*** (0.0376)
  - k=5: -0.976*** (0.0367)
  - k=10: -0.963*** (0.0204)
  - k=15: -1.035*** (0.0127)
  - k=20: -1.011*** (0.0106)
- Instrument strength: Kleibergen–Paap rk Wald F statistic ≈ 1200 (Stock-Yogo critical value 16.38).
- Firm-level effects:
  - one-standard deviation increase in financial stress → about 30 percent reduction in firms’ investment after 12 quarters.
  - Peak additional differential investment loss for constrained/low-profit firms → about 10 percent.

### Empirical implications and policy-relevant takeaways
- Financial stress measured by the FSI is a robust predictor of subsequent output declines across multiple horizons and samples.
- Effects are state-dependent and non-linear: shocks matter more when initial financial stress is medium-to-high.
- EMDEs are more vulnerable: the same increase in financial stress produces more than twice the GDP response in EMDEs relative to AEs.
- External/global financial stress matters: the constructed external financial stress indicator is a strong instrument and highlights international spillovers.
- Firm-level transmission confirms a causal channel: financial stress leads to persistent and economically meaningful declines in firm investment, with larger effects for less profitable and more financially constrained firms.
- Robustness: results hold under multiple robustness checks, alternative specifications, frequencies, samples, and IV approaches.

*IMF Working Paper: Financial Stress and Economic Activity: Evidence from a New Worldwide Index (content as provided in the source unit).*

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

### Financial Stress and Economic Activity: Evidence from a New Worldwide Index

### Introduction and motivation
- Financial crises have large impacts but are difficult to predict; measuring financial stress and defining crises is challenging because statistical indicators may misidentify episodes and data are often limited to advanced economies.
- Previous widely used crisis indexes are largely binary and focus on occurrence rather than intensity; examples include Caprio and Klingebiel (1996), Reinhart and Rogoff (2009), and Laeven and Valencia (2013, 2014, and 2020).
- Romer and Romer (2017; RR) developed a narrative-based index classifying financial stress on a scale to capture intensity and duration for 24 advanced countries over 1967-2012 (with a later update adding 6 countries and extending to 2017).
- This paper introduces a new index that builds on RR with three main modifications:
  - Source: uses Economist Intelligence Unit (EIU) country reports instead of OECD reports to extend coverage to 110 countries and change frequency from semi-annual to quarterly for 1967-2018.
  - Method: employs a mechanistic approach using search algorithms and word counts in addition to expert judgement to measure intensity, enabling semi-automatic updates and improved cross-country/time comparability but risking missed information an expert reader might catch.
  - Exogeneity: identifies episodes of financial stress stemming from other countries to construct an external financial stress series for causality identification.

### Key empirical findings (summary)
- Correlation with RR for OECD countries: 0.9.
- First main finding: a one-standard deviation increase in the financial stress index is associated with:
  - a reduction in the level of output by 0.35 percent one year after the increase in financial stress, and
  - a reduction in the level of output by 0.2 percent 5 years after.
- Second finding: effects of crises are quantitatively larger for emerging markets and developing economies than for advanced economies.
- Third finding: the effect of financial stress on economic activity is non-linear:
  - small and not statistically significantly different from zero for lower levels of financial stress,
  - large and more precisely estimated for medium-to-high levels of financial stress,
  - non-linearity is markedly more significant and robust in emerging markets than in advanced economies.
- Fourth finding: using the external financial stress series as an instrumental variable shows financial stress has a statistically significant exogenous effect on economic activity, and simultaneous causality biases OLS coefficients downward (weaker economic activity tends to intensify financial stress).
- Fifth finding: firm-level difference-in-differences evidence indicates increases in financial stress lead to persistent declines in firms’ investment, with larger effects for firms that are less profitable (lower profits, revenues and return on assets) and more financially constrained (higher debt-to-asset ratios, smaller, and younger).

### Literature context
- Two broad strands of measures:
  - Binary chronologies of crises (systemic vs non-systemic) covering many countries and long time spans: examples include Bordo et al. (2001), Caprio and Klingebiel (2003), Demirgüç-Kunt and Detragiache (2005), Reinhart and Rogoff (2009; ReRo), Schularick and Taylor (2012), Laeven and Valencia (2013, 2014, 2020; LV).
    - Laeven and Valencia (2020) compiled a dataset for 165 countries over 1970 to 2017 and identified 151 banking crises, 236 currency crises, and 74 sovereign crises.
  - Continuous, narrative- or market-based measures capturing intensity (e.g., Jalil 2015; Romer and Romer 2017; Baron et al. 2021).
    - RR classify financial stress on a scale of 0 to 15 using OECD Economic Outlook narrative text.
    - Baron et al. (2021) construct bank equity crash measures using bank equity prices and dividends for 46 advanced and emerging economies from 1870 to 2016.
- This paper contributes by extending continuous narrative measures to a much larger sample (adding about 80 developing economies) and increasing frequency to quarterly data, while constructing an instrument for exogenous international stress.

### Data source and rationale
- Sole narrative source: Economist Intelligence Unit (EIU) country reports.
  - Reports average about 12,000 words in length and are available on a quarterly basis going back to the 1950’s.
  - EIU covers about 180 advanced, emerging markets, and low-income countries.
  - Reports are prepared by designated country experts, peer-reviewed, subedited, and put through data-quality checks to deliver transparency, accuracy and consistency.
- Advantages of EIU:
  - high frequency (minimum quarterly),
  - extended time period (current work covers 1967 to 2018),
  - broad coverage enabling focus on emerging markets and developing economies,
  - standardized format, topics and level of analysis across countries and time.
- Limitations noted:
  - single-source approach may miss information present in other sources; single-source narrative-based indexes are viewed as a complement rather than a substitute for comprehensive zero-one historical datasets such as LV.
  - EIU is preferred over OECD for coverage of developing economies and over IMF Article IV reports for the desired quarterly frequency.

### Index construction methodology
- Sample and scope:
  - FSI constructed for 110 countries for the period 1967-2018 (sample restricted to countries with population above 2 million).
- Conceptual definition:
  - Following Bernanke (1983) and RR, episodes are classified where an economy experiences an increase in the cost of credit intermediation or disruptions to the credit supply, excluding credit reductions due to “normal” cyclical monetary policy tightening.
- Four-step construction process:
  1. Keyword search of EIU reports for paragraphs/lines containing two sets of keywords:
     - set (i): credit, financial, bank, lending, and fund;
     - set (ii): crisis, crunch, squeeze, bailout, rescue, tight, contract, and reluctant.
  2. Human reading of extracted paragraphs to confirm the text describes contemporaneous financial stress and to exclude false positives; identify domestically originated vs externally originated stress.
     - Example domestic classification: United States (2009Q4) quote indicating recovery support interpreted as ongoing domestic financial stress.
     - Example external classification: Denmark (2008Q2) quote referencing a response to the global credit crunch interpreted as externally driven stress.
     - The text-search produced over 50 percent false positive signals which had to be manually evaluated and eliminated; it is estimated that about half of the substantive work involved careful reading and validation by a human expert.
  3. Cross-validation by IMF country economists led to corrections in a few cases (examples: Ecuador, Nepal, and Venezuela).
  4. Summation of verified signals in each period, scaled by the total number of words in each report to account for variation in report length and make the index comparable across countries.
- Additional comparability features:
  - single-source basis and standardized report preparation process improve cross-country comparability and mitigate concerns about ideological bias or inconsistency.

### Global movements and validation
- Global FSI movements highlight rises during:
  - the Latin America debt crisis in the 1980s,
  - the Mexican Peso Crisis in the mid-1990s,
  - the late 1990s crises in Asia, Russia, and Latin America (coinciding with the Long-Term Capital Management episode),
  - the Global Financial Crisis (GFC) and Europe’s sovereign debt crisis between 2008-2013.
- The index remained relatively stable up to the last observation in 2018 (an update to current times would likely show activity during the COVID crisis).
- Validation:
  - For OECD countries the new index essentially mimics RR (correlation 0.9), supporting the effectiveness of the search algorithm and word-count approach.

*IMF Working Paper: Financial Stress and Economic Activity: Evidence from a New Worldwide Index (content as provided in the source unit).*

### Appendix Table A2 reports country-by-country and period-by-period examples of EIU narratives identified as related to fi

### Appendix Table A2 reports country-by-country and period-by-period examples of EIU narratives identified as related to financial distress

### Heterogeneity across country income groups
- The magnitude of financial stress varies significantly across income and regional country groups and across events.
- Example: in 2008Q4, the level of the FSI is
  - close to the global average in emerging economies,
  - below it in low-income economies,
  - about three times it in advanced economies.
- Explanation: the GFC has been described as a crisis of advanced economies.

### Averages vs. episodes
- Average FSI over 1967-2018 (Panel A, Figure 3):
  - advanced economies: 0.033
  - emerging economies: 0.025
  - low-income economies: 0.010
- Excluding 2008 to 2012 (GFC) (Figure 3, Panel B): the average FSI is higher in emerging economies, followed by advanced economies, and low-income economies.
- Number of quarters with financial stress (normalized by number of countries) (Figure 3c):
  - emerging economies: 20.7 quarters
  - advanced economies: 15.7 quarters
  - low-income economies: 13.7 quarters
- Interpretation: low FSI values for low-income economies likely reflect less developed and interconnected financial sectors—an explanation for why these economies survived the GFC better than richer countries.

### Regional heterogeneity
- FSI patterns by region (Figure 4):
  - Africa and the Middle-East and Central-Asia: little financial stress; regions characterized by lower income per capita and financial development.
  - Asia-Pacific: financial stress during the Asian crisis and during the GFC.
  - Western Hemisphere: elevated FSI during financial crises in early 1980s, late 1990s, and during the GFC.
  - Europe: FSI captures financial stress during the GFC and the European sovereign debt crisis.

### Comparison of FSI with existing chronologies
- Coverage and comparison:
  - Table 1 reports key characteristics of FSI and those by RR, ReRo, and LV; FSI generally compares favorably on country coverage, frequency, and time coverage.
  - Pairwise correlations (Table 2):
    - correlation between FSI and RR in the overlap: 0.9
    - correlation between FSI and LV: 0.4
    - correlation between FSI and ReRo: 0.4
    - correlation among RR, ReRo, and LV: in the range of 0.5 to 0.7
- Country-case comparisons (Figure 5) for United States, South Korea, Honduras, Argentina, the Philippines, Nigeria, Costa Rica, and Rwanda (data restricted to sample periods of LV, ReRo, RR):
  - Observation 1: FSI closely aligned with RR in many countries; both capture severity better than binary indicators (LV, ReRo).
  - Observation 2: Broader country coverage of FSI allows examination of heterogeneity in severity and duration for countries not covered by RR (e.g., Argentina, Philippines, Nigeria). Binary indices fail to distinguish severity across episodes.
  - Observation 3: Some discrepancies where intensity measures and binary chronologies diverge:
    - Honduras: FSI identifies severe stress 1980-1985 (narratives: “liquidity crisis”, “credit squeeze”, “financial crisis”); LV and ReRo do not pick up this period. LV requires both signs of banking stress and significant banking policy intervention to identify a banking crisis.
    - Korea: ReRo identifies stress 1985-1989 not identified by FSI; ReRo’s broader definition and view of banking-system risks (financial liberalization, increase in banks) differs from FSI’s stricter credit-supply shortage definition.
  - Observation 4: Some cases show non-matching identification between FSI and binary chronologies:
    - Costa Rica: LV and ReRo identify crises in 1987 and 1995 driven by non-performing loans and bank closure; EIU narratives in these periods barely mention FSI signals, so FSI moves little. FSI spikes in 1967 due to EIU mention “economic and financial crisis that now seems endemic in Costa Rica”.
    - Rwanda: FSI identifies stress periods not in LV (1985: “tight liquidity”; 2009: “liquidity squeeze” related to withdrawals, losses on foreign investments, lower domestic saving).
- General conclusion: measures are complements rather than substitutes; FSI has high signal-to-noise ratio and compares favorably with existing narrative measures, but is not always the preferred one.

### Statistical correlation with Financial Condition Indices (FCIs)
- FSI is positively and statistically significantly correlated with IMF FCIs: correlation about 0.45.
- Note on FCI computation (IMF, 2017): computed as the principal component of several financial variables; covers an unbalanced sample of 43 advanced and emerging market economies from 1996Q1.

### Empirical analysis: the effect of FSI on economic activity — overview
- Two-pronged approach:
  - Country-level analysis using quarterly data to estimate baseline effects of FSI on GDP for a panel of 49 countries; examine nonlinearities and variation across Advanced Economies (AEs) vs. Emerging Market and Developing economies (EMDEs); conduct robustness checks including alternative samples, specifications, and frequencies; construct a novel instrumental variable to address simultaneous causality.
  - Firm-level analysis using a comprehensive quarterly firm-level dataset for sixty-three AEs and EMDEs over 20 years (over 20,000 firms) to estimate effects with greater precision and to implement a difference-in-differences framework based on firms’ exposure to external finance, including country-sector-time fixed effects.

### Country-level analysis — empirical methodology
- Method: local projection method (Jordà, 2005) to examine dynamics of output following changes in financial stress; does not impose VAR dynamic restrictions and suits estimation of nonlinear dynamic responses.
- Baseline regression (Equation (1)):
  - y_{i,t+k} = α_i^k + γ_t^k + sum_{j=0}^2 β_j^k ΔF_{i,t−j} + sum_{j=0}^2 θ_j^k y_{i,t−j} + ε_t^k
  - i indexes countries, t refers to quarters, k denotes horizon (quarter after change in financial stress), y is log of output, α_i are country fixed effects, γ_t are time fixed effects, ΔF denotes change in financial stress.
- Sample and estimation:
  - Baseline: unbalanced sample of 49 countries with quarterly data from 1996Q1 to 2018Q4.
  - Robustness/extension: larger sample of 110 countries using annual data from 1967 to 2018.
  - Equation (1) estimated by OLS for horizon k = 0,...,20.
  - Coefficient β_k denotes impact of changes in FSI on output at horizon k.
  - Impulse response functions computed using estimated β_k; confidence bands use standard errors clustered at the country level.
- Identification considerations:
  - Baseline does not distinguish between foreign-originated (more exogenous) and domestic-originated changes in financial stress.
  - Later sensitivity checks use foreign-originated changes either as an instrument for overall changes in financial stress or directly as regressors.

### Data sources
- Quarterly and annual macroeconomic series for GDP, employment, labor productivity (GDP/employment), unemployment, policy rates, cyclically adjusted balance: IMF World Economic Outlook.
- Country classification (AEs vs. EMDEs; regions Africa, Asia-Pacific, Europe, MENA, Americas): IMF World Economic Outlook.
- Data for uncertainty: Ahir, Bloom and Furceri (2022).
- Stock returns and return volatility: Baker, Bloom and Terry (2021).

_© IMF — https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023217-print-pdf.pdf_

### 20. The lagged output coefficient, as expected, is close to 1, suggesting that the level of GDP is non-

### wpiea2023217-print-pdf - 20. The lagged output coefficient, as expected, is close to 1, suggesting that the level of GDP is non-

### Baseline estimates and main findings
- The lagged output coefficient is close to 1, suggesting that the level of GDP is non-stationary and that the country fixed effects capture average GDP growth rates.
- Panel cointegration tests reject the null hypothesis that the estimated residual of Equation (1) is non-stationary.
- A one-standard deviation increase in FSI (equivalent to 0.1 changes in the index) is associated with:
  - a reduction in the level of output by 0.35 percent one year after the increase in financial stress;
  - a reduction in the level of output by 0.2 percent 5 years after the increase in financial stress.
- The peak increase in financial stress observed in the United States during the GFC (1.7 in the fourth quarter of 2018) would have been associated with a reduction in US GDP by about 6 percent in 2019.
- Results indicate increases in financial stress are associated with sizeable and persistent reductions in the level of output, and transitory effects on the growth rate.

### Heterogeneity: Advanced Economies (AEs) vs Emerging Market and Developing Economies (EMDEs)
- Regression specification incorporates dummy D = 1 for AEs and D = 0 for EMDEs to estimate separate βk coefficients for AEs (βk_AE) and EMDEs (βk_EMDE).
- The response of GDP to an equal increase in financial stress is more than twice larger in EMDEs than in AEs.
- The heterogeneity is attributed not to different severity of financial stress (responses are to the same increase) but to differences in economic resilience, including fiscal and monetary policy space.

### Nonlinearities: severity of financial stress
- Specification using indicator I[Fit ∈ G] for stress terciles (low, medium, high) estimates β0_kG for each group.
- Estimates show nonlinearity:
  - Effects of financial stress on output are small and not statistically significant for lower levels of financial stress.
  - Effects are precisely estimated and larger than baseline for medium-to-high financial stress.
  - Differences between low vs. medium and low vs. high financial stress are statistically significant up to k = 11, but become insignificant in the medium-term because of large confidence bands for the low-stress regime.
- Nonlinearity is more pronounced and significant in EMDEs than in AEs (in AEs significance appears only at high stress levels).
- Alternative specifications confirming non-linearity:
  - D = 1 if financial stress above median (binary split);
  - smooth transition function of the level of financial stress.
- Conclusion: the response to a given increase in financial stress is large and statistically significant when initial stress is already high, and small and typically not statistically significant when initial stress is low.

### Robustness checks
- Alternative sample exclusions tested (results reported robust in Figure 9):
  - Outliers (bottom and top percentiles of output regression residuals)
  - High inflation episodes (inflation above 20 percent)
  - Observations from the Americas
  - Asian and Sub-Saharan African economies
  - Drop small countries
  - Episodes of large changes in financial stress (99th percentile)
  - Observations after the third quarter of 2008 (post-GFC)
- Alternative specification checks:
  - Restrict change in FSI to enter only with a lag (no contemporaneous effect).
  - Add controls: changes in monetary policy rates, changes in cyclically adjusted budget balance, stock market growth and volatility, economic uncertainty.
  - Results (Figure 10) not statistically different from baseline.
- Alternative data frequency:
  - Annual data for an unbalanced panel of 110 countries (1950-2018) finds larger effects: a one-standard deviation increase in FSI associated with:
    - reduction in level of output by 0.8 percent one year after;
    - reduction in level of output by 0.6 percent 5 years after.
  - Larger annual estimates persist when constraining sample to that used for quarterly data; potential explanation is larger reverse causality with annual data, suggesting annual analyses may overestimate macroeconomic effect of financial stress.

### Comparison with other chronologies
- Re-estimated Equation (1) using alternative measures: (i) RR (annualized); (ii) LV; (iii) ReRo.
- Findings:
  - Responses for FSI over overlapping samples confirm robustness.
  - Magnitudes vary across chronologies:
    - Output effects of a one standard deviation increase in FSI are in the same ballpark as those from RR.
    - FSI effects are about one-tenth smaller than those associated with crises identified in LV and ReRo.
- Interpretation: supports non-linearity and that binary measures (chronologies) capture more extreme events and thus larger effects.

### Addressing endogeneity and IV strategy
- Identified episodes of financial stress originating outside each country (external financial stress indicator EFit) from EIU report narratives; these episodes are treated as plausibly exogenous to domestic GDP.
- IV system:
  - First stage: ΔF_it = π_i + τ_t + ϖ ΔEF_it + Σ ρ_j,k ΔF_it−j + Σ σ_j,k y_it−j + μ_t
  - Second stage uses predicted ΔF_it̂ in main equation for y_i,t+k.
- First-stage diagnostics:
  - Kleibergen–Paap rk Wald F statistic ≈ 1200, about 75 times the associated Stock-Yogo critical value (16.38).
  - Instrument is “strong” and statistically significant.
- IV results:
  - Support OLS findings: significant negative and persistent effect of financial stress.
  - IV coefficient magnitudes are smaller in absolute value than OLS estimates, suggesting OLS estimates are downward biased and should be considered an upper bound of the negative effect.
- Instrument validity checks:
  - Including instruments as additional controls: coefficients not negative and significant, suggesting exclusion restriction holds (Table 6).
  - Direct test of association between baseline residuals and instrument: relationship indistinguishable from zero (Table 7).
  - Excluding the United States and other G7 economies one at a time: results confirm instrument validity (Figure 14).
  - Controlling for GDP growth in major trading partners: results corroborate instrument validity (Figure 15).
  - Re-estimating using external financial stress directly as main regressor yields qualitatively unchanged results.
- IV extensions:
  - Equation (2) re-estimated with annual data; Equations (2) and (3) re-estimated with IV approach (Figures 16-18): results qualitatively consistent with OLS though smaller in magnitude.

### Channels
- Re-estimated Equation (1) with alternative dependent variables to probe channels: (i) labor productivity; (ii) employment; (iii) unemployment; (iv) growth rate of stock market returns; (v) stock market volatility; (vii) economic and policy uncertainty (World Uncertainty Index).
- Key channel findings (Figure 19):
  - Labor productivity declines by about 2 percent after one year following a one-standard-deviation increase in financial stress.
  - Employment declines about 1 year after the shock.
  - Unemployment rate increases by about 0.1 percentage point 1 year after the shock.
  - Financial stress associated with a short-term decline in stock market growth and short-lived increases in stock market volatility and uncertainty.

*Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023217-print-pdf.pdf*

### references therein). Similar results are obtained when using the IV approach (Figure 20).

### Firm-level Analysis

### Data
- Main dataset: S&P Capital IQ (CIQ) providing quarterly firm balance sheet and income statement information.
- Coverage: 150 countries from 1950Q1 to 2021Q2; sample restricted to 2001Q1 onwards and to 75 advanced and emerging market economies.
- Sample restrictions:
  - Non-financial corporations only.
  - Firms with negative assets or debt in any year removed.
  - Observations with incorrect sign for revenue, capital expenditure, cash, tangible assets, and interest expenditure set to missing.
  - Real estate and insurance companies excluded.
- Sector and country detail: Number of firms across countries and 20 economic sectors reported in Tables A7 and A8 (source).
- Investment measure: capital expenditures (IQ_CAPEX-2021), defined as funds used by firms to acquire assets such as property, plant, or equipment.

### Empirical Methodology
- Two-step approach:
  1. Estimate the average (unconditional) effect of financial stress on firm investment using Jordà’s (2005) local projections. Specification summarized as equation (6) with dependent variable y_{n,i,t+k} (investment ratio), ∆F_{i,t} (change in financial stress), firm fixed effects γ_n^k, and country-sector-quarter dummies α_{isq}^k.
  2. Estimate heterogeneity via a difference-in-differences approach by expanding equation (6) into equation (7). Interaction term ∆F_{i,t-j} * D_n identifies differential effects for firms that are less profitable or more financially constrained. D is a dummy equal to one if the firm (or firm country characteristic) is below (above) the median of the country; the average profitability over the entire sample is used to define this dummy to reduce endogeneity.
- Fixed effects and clustering:
  - α_{ist}^k are country-sector-time fixed effects.
  - γ_{nq}^k are firms-quarter dummies.
  - Equations estimated using OLS and IV; standard errors two-way clustered on firm and country-time.
- Instrumental variable approach: external financial stress indicator used as instrument in IV estimations to address endogeneity and simultaneity.

### Results
- Average response:
  - A one-standard deviation increase in financial stress is associated with an average reduction in the level of firms’ investment of about 30 percent after 12 quarters.
  - The effect is statistically significant and economically sizeable.
- Heterogeneity (difference-in-differences):
  - Firms with relatively low profitability and high financial constraints experience a statistically significant and precisely estimated additional investment loss across most horizons.
  - Peak differential effect for low-profitability/high-constraint firms is about 10 percent.
  - Results robust to IV estimation using the external financial stress indicator as instrument.
- Presentation details:
  - Figure 21: response of (log) investment to an increase in financial stress; solid line = average estimated response; dashed and dotted areas denote 90 and 68 percent confidence bands respectively.
  - Figure 22: differential response between low-profitability/high-constraint firms and high-profitability/low-constraint firms.
  - Figure 23: IV-based robustness of differential responses.

### Key statistics and specifications preserved from the source
- Dataset coverage: 150 countries; 1950Q1 to 2021Q2; analysis sample from 2001Q1; 75 advanced and emerging market economies.
- Investment variable: IQ_CAPEX-2021.
- Local projections methodology: Jordà (2005); equations (6) and (7) as specified.
- Clustering: two-way clustered on firm and country-time.
- Main quantitative findings:
  - "one-standard deviation increase in financial stress" → "about 30 percent" reduction in firms’ investment after 12 quarters.
  - Peak additional differential investment loss for constrained/low-profit firms → "about 10 percent".

### Conclusions (as related to firm-level analysis and broader paper)
- Firm-level evidence supports a causal interpretation: financial stress leads to persistent and economically meaningful declines in firm investment.
- Heterogeneous effects: less profitable and more financially constrained firms bear larger investment cuts.
- IV and diff-in-diff approaches provide robustness to concerns about simultaneity and endogeneity.

*IMF Working Paper: Financial Stress and Economic Activity: Evidence from a New Worldwide Index (excerpts from the firm-level analysis section).*

### 7. Financial stress in 2012Q2 in 16 countries: Austria, Belgium, Denmark, France, Germany, Greece, Haiti, Hungary, Irela

### 7. Financial stress in 2012Q2 in 16 countries: Austria, Belgium, Denmark, France, Germany, Greece, Haiti, Hungary, Ireland, Italy, Netherlands, Nigeria, Pakistan, Portugal, Spain, and United Kingdom.

### FSI construction and coverage
- Definition and method:
  - Financial Stress Index (FSI) is summing the number of keywords identified with financial stress in EIU country reports. The index is then normalized by total number of words. A higher number means higher financial stress and vice versa.
  - Constructed at quarterly frequency covering 1967-2018 for 110 economies (see Table 1: FSI country coverage: "110 economies"; Time Horizon: "1967-2018"; Frequency: "quarterly").
- Comparisons and pair-wise correlations (Table 2):
  - Correlations: FSI vs Romer&Romer (RR) = 0.9; FSI vs Reinhart&Rogoff (ReRo) = 0.4; FSI vs Laeven&Valencia (LV) = 0.4.
- Country and regional coverage:
  - Full-country list spans Africa (32), Asia and the Pacific (19), Europe (24), Middle East and Central Asia (13), Western Hemisphere (22) (Table A1).
  - Examples of financial distress discussions and dates across countries provided in Table A2 and Table A5 (external financial stress examples).

### Stylized FSI patterns and comparisons
- Time-series patterns:
  - FSI plotted over 1967-2018 by income group (Advanced, Emerging, Low-income) and by geographic region (Africa; Asia and the Pacific; Europe; Middle East and Central Asia; Western Hemisphere) (Figure 2, Figure 4).
  - Key historical episodes visible include "Great Recession and Europe's Sovereign Debt Crisis", "Financial crisis in Asia, Russia and Latin America", "Mexican peso crisis".
- FSI vs other chronologies and measures:
  - Extensive country-level comparisons (Figure 5; Figure A3 series; Table A3) show alignment and differences with existing chronologies (Romer&Romer; Reinhart&Rogoff; Laeven&Valencia; others).
  - Table A3 provides FSI episode dates alongside many alternative datasets for 110 countries.

### Macro impact: FSI → output (quarterly evidence)
- Baseline local-projection specification (quarterly sample: 49 countries; 1996q1-2018q4):
  - Equation form used: y_{i,t+k} = α_i^k + γ_t^k + Σ_{j=0}^2 β_j^k ΔF_{i,t−j} + Σ_{j=0}^2 θ_j^k y_{i,t−j} + ε_{t}^k (notes in Figures and Tables).
- Main quarterly results (Table 3: Impact of Change in FSI on Output):
  - Coefficients (point estimates) and standard errors (in parenthesis), horizons k reported as k=0, k=5, k=10, k=15, k=20:
    - F_{i,t}:
      - k=0: -0.103*** (0.0363)
      - k=5: -0.338*** (0.0675)
      - k=10: -0.405*** (0.0818)
      - k=15: -0.393*** (0.0827)
      - k=20: -0.235*** (0.0782)
    - F_{i,t−1}:
      - k=0: -0.126*** (0.0352)
      - k=5: -0.358*** (0.0841)
      - k=10: -0.437*** (0.101)
      - k=15: -0.409*** (0.107)
      - k=20: -0.200* (0.107)
    - F_{i,t−2}:
      - k=0: -0.102*** (0.0285)
      - k=5: -0.246*** (0.0695)
      - k=10: -0.300*** (0.0828)
      - k=15: -0.306*** (0.0928)
      - k=20: -0.144* (0.0801)
    - Lagged output controls:
      - y_{i,t−1}: e.g., k=15: 0.332* (0.187)
      - y_{i,t−2}: e.g., k=15: -0.441** (0.187)
  - Observations and fit:
    - Observations: 4,211 (k=0), 3,966 (k=5), 3,721 (k=10), 3,476 (k=15), 3,231 (k=20).
    - R-squared: 0.197 (k=0), 0.433 (k=5), 0.514 (k=10), 0.586 (k=15), 0.654 (k=20).
- Annual data evidence (Table 4; full sample of 110 countries; 1967-2018):
  - F_{i,t} (K horizons K=0..K=5):
    - K=0: -0.473*** (0.154)
    - K=1: -0.830*** (0.203)
    - K=2: -0.788*** (0.192)
    - K=3: -0.794*** (0.196)
    - K=4: -0.829*** (0.190)
    - K=5: -0.763*** (0.159)
  - Other coefficients: F_{i,t−1} e.g., K=0: -0.340*** (0.0938); F_{i,t−2} often small and not significant.
  - Observations: 5,058 (K=0..K=2), declining to 4,731 (K=5). R-squared rises from 0.150 to 0.411 across horizons.

### Heterogeneity, non-linearities, and spillovers
- Advanced Economies (AE) vs Emerging Markets and Developing Economies (EMDE):
  - Separate impulse responses documented (Figure 7; Figure 17; Figure A1). Estimates obtained using the same local projection framework with AE/EMDE decompositions.
- Non-linear effects by FSI level (terciles):
  - Panel results (Figure 8; Figure 18) present responses for Low, Medium, High financial stress bins (groups defined as terciles).
  - Table A4 reports p-values for differences in responses (Low vs. Medium; Low vs. High; Medium vs. High) across horizons K=0..K=20. Selected p-values:
    - Low vs. Medium: K=1: 0.0524; K=2: 0.0157; K=3: 0.0190; K=4: 0.00730.
    - Low vs. High: K=3: 0.0349; K=4: 0.0266; K=5: 0.0281.
    - Medium vs. High: many p-values large (e.g., K=1: 0.815), indicating less significant differences between medium and high bins in some horizons.
- External financial stress and reverse causality:
  - Foreign shocks—reverse causality (Table 5): EF_{i,t−1} coefficients large and negative across horizons:
    - k=0: EF_{i,t−1} = -1.470*** (0.0376)
    - k=5: -0.976*** (0.0367)
    - k=10: -0.963*** (0.0204)
    - k=15: -1.035*** (0.0127)
    - k=20: -1.011*** (0.0106)
  - Specification used: EF_{i,t+k}=α_i^k + γ_t^k + Σ β_j^k EF_{i,t−j} + Σ β_j^k Δln gdp_{i,t−j} + ε_{t}^k (notes in Table 5).
- Instrumental variable (IV) approach:
  - IV strategy uses an indicator of external financial stress (EF) as instrument for ΔF.
  - IV baseline impulse responses shown in Figure 13; IV results excluding G7 and China shown in Figure 14; IV with controls for trading partners in Figure 15; annual IV in Figure 16.
  - Adding instrument as control (Table 6) — baseline coefficients remain negative and significant:
    - F_{i,t} (k=0..20): k=0: -0.156** (0.0703); k=5: -0.421*** (0.102); k=10: -0.493*** (0.115); k=15: -0.444*** (0.106); k=20: -0.298*** (0.105).
    - EF_{i,t} included: e.g., EF_{i,t} at k=0: 0.00896 (0.00673) — small and not significant.
  - Validity of instrument (Table 7): regression of baseline residuals on ΔEF shows small coefficients:
    - resid_{i,t} coefficients across k: 0.00319 (k=0; 0.00226), 0.00494 (k=5; 0.00595), 0.00520 (k=10; 0.00833), 0.00299 (k=15; 0.0102), 0.00382 (k=20; 0.0116).

### Robustness checks
- Alternative samples and exclusions (Figure 9):
  - Robustness to dropping: outliers; high inflation episodes (>20 percent); observations from the Americas; Asian and Sub-Saharan African economies; small economies (population below two millions); episodes of large changes in financial stress (99th percentile); excluding period after 2008q3.
- Alternative controls (Figure 10):
  - Controlled specifications include: i) without lag of FSI; ii) overall balance (% of GDP); iii) short-term interest rate (%); iv) uncertainty; v) log stock return volatility; vi) stock return (%).
- Alternative chronologies and samples (Figure 12; Table 1; Table A3):
  - Results compared across differing country/time coverage and chronologies (Romer&Romer; Reinhart&Rogoff; Laeven&Valencia; others).

### Broader macroeconomic channels and firm-level effects
- Impact on other macro variables (Figure 19; Figure 20—IV results):
  - Outcomes analyzed include: (i) stock returns %; (ii) log(stock returns volatility); (iii) World Uncertainty Index (wui); (iv) unemployment; (v) employment; and (vi) labor productivity.
  - Responses plotted with 90 percent confidence bands for horizons after ΔFSI shocks.
- Firm investment (firm-level quarterly data; 75 countries; 2001Q1–2020Q4) (Figures 21–23):
  - Baseline firm-level local-projection specification: y_{n,i,t+k} = α_{i s}^k + γ_{n q}^k + Σ β_j^k ΔF_{i,t−j} + Σ θ_j^k y_{n,i,t−j} + ε_{n,i,t}^k (up to 12-quarter horizons).
  - Impulse responses show negative impacts of ΔFSI on firm capital expenditure growth up to 12 quarters (Figure 21).
  - Heterogeneity by firm characteristics (Figure 22; Figure 23—IV):
    - Characteristics examined: Profits, Revenue, ROA, Debt to Asset, Size, Age.
    - Estimates are provided for interactions ΔF * D_n where D_n is indicator for firm characteristic groups; IV results analogous.

### Key quantitative magnitudes (selected)
- Quarterly baseline: F_{i,t} cumulative negative impact on output:
  - Point estimates range from -0.103*** (k=0) to -0.405*** (k=10) and remain negative up to k=20 (-0.235***).
- Annual baseline (full sample): F_{i,t} large negative annual impacts:
  - K=1: -0.830*** (0.203); K=4: -0.829*** (0.190).
- External financial shock instrument relevance:
  - EF_{i,t−1} large negative coefficients in reverse-causality checks (e.g., -1.470*** at k=0, Table 5).
- Firm-level sample size and sector/country coverage (Tables A6–A8; Table A7):
  - Firm-level: 75 countries (Table A6 sample list).
  - Number of firms and observations (Table A7): United States: 4,740 firms, 388,680 Obs.; China: 4,077 firms, 334,314 Obs.; Japan: 3,085 firms, 252,970 Obs.; India: 2,672 firms, 219,104 Obs.; (full table lists many countries).
  - Sectoral coverage (Table A8): Materials 5,433 firms, 445,506 Obs.; Capital Goods 4,888 firms, 400,816 Obs.; etc.

### Empirical implications and policy-relevant takeaways (from results)
- Financial stress, as measured by FSI, is a robust predictor of subsequent output declines across multiple horizons and samples:
  - Negative and statistically significant coefficients for ΔFSI on output in quarterly and annual specifications.
- External/global financial stress matters for domestic outcomes:
  - External financial stress indicator (EF) is a relevant instrument and shows large associations with domestic outcomes in reverse-causality tests.
- Non-linearities and state-dependence:
  - Output responses to ΔFSI are stronger when initial FSI levels are higher (statistically significant differences between Low vs. Medium/High in early horizons; Table A4).
- Firm-level transmission:
  - ΔFSI shocks reduce firm capital expenditure; effects vary by firm profits, revenue, ROA, leverage (Debt to Asset), size, and age—suggesting credit and balance-sheet channels.
- Robustness:
  - Results hold under multiple robustness checks: sample exclusions, alternative controls, IV approaches, and alternative chronologies.

*Financial Stress and Economic Activity: Evidence from a New Worldwide Index — Working Paper No. WP/2023/217 (Figures, Tables and notes extracted from the supplied content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023217-print-pdf.pdf_