## 1. Introduction

## Source details

**Canonical URL:** [1. Introduction](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023126-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023126-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023126-print-pdf.pdf.json)

---

### Background and motivation
- Expansions in aggregate credit are associated with boom-bust cycles in macroeconomic outcomes; credit buildups in the boom phase are linked to elevated macro‑financial stability risks and subsequent declines in economic activity.
- The paper studies the dynamic relationship between firm leverage and real outcomes using aggregate and firm-level data from 24 European economies over 2000-2018.
- Theoretical mechanisms cited: Myers 1977; Bernanke and Gertler 1989; Kiyotaki and Moore 1997; Bernanke et al. 1999; Lorenzoni 2008; Jeanne and Korinek 2010; Bianchi 2011; Korinek and Simsek 2016 — leverage can relax financial constraints short-term but raise balance-sheet risks and tighten constraints medium-term.

### Contribution and data
- Scope:
  - Countries: 24 European economies.
  - Period: 2000-2018.
- Data sources:
  - Aggregate macro data on credit to firms and employment.
  - Firm-level panel from ORBIS (non-farm, non-financial activities; listed and unlisted firms), constructed following Kalemli-Ozcan et al. (2015), Gopinath et al. (2017), Diez et al. (2021a).
- ORBIS advantages:
  - Includes SMEs (about 99 percent of businesses in Europe; in the sample SMEs account for around 98 percent of observations).
  - Median (mean) number of employees: 5 (17).

---

### 2. Data and Methodology

### Country-level data and processing
- Key variables:
  - Credit to non-financial corporations and households: ECRI 2021 Statistical Package (as percent of GDP).
  - Aggregate employment: OECD.
  - Real GDP: World Bank WDI.
  - Long-term (10-year) government bond yields and forecasts: OECD.
- Forecast errors: computed as (the rate minus its forecast) and used as a measure of unexpected tightening in financing conditions.
- Country-level variables winsorized at the 1st and 99th percentiles.

### Firm-level data and definitions
- ORBIS coverage: non-farm, non-financial industries (NACE 2-digit codes 5-82); sample size in largest sample: about 2,5 million firms and 15,7 million observations.
- Main variables:
  - Firm leverage: total liabilities as percent of total assets.
  - Alternative leverage: net liabilities (total liabilities minus cash) to total assets.
  - Employment: number of employees.
  - Investment: change in fixed assets (net investment).
  - Debt service ratio (DSR): interest payments as percent of EBITDA.
- Firm-level variables winsorized at the 2.5th and 97.5th percentiles; raw data cleaned per Kalemli-Ozcan et al. (2015), Gopinath et al. (2017), Diez et al. (2021a).

### Empirical approach (3-year sliding windows)
- Uses 3-year “sliding windows”; explanatory variable is change in firm debt or firm leverage between t−3 and t.
- Dependent variables: employment growth and investment measured over windows p = 0,...,5 corresponding to periods:
  - p=0: t−3 to t
  - p=1: t−2 to t+1
  - p=2: t−1 to t+2
  - p=3: t to t+3
  - p=4: t+1 to t+4
  - p=5: t+2 to t+5
- Firm-level estimations include firm fixed effects and country-industry-year fixed effects (narrow 4-digit NACE × country × year cells); standard errors clustered at the country-industry-year level.

---

### 3. Main Empirical Findings

### Aggregate-level results — credit to firms and households
- Firm credit (regression coefficients on ΔFirm credit (t-3,t) for Δlog(Employment)c,t):
  - (t-3,t): 0.073** (0.029)
  - (t-2,t+1): -0.002 (0.027)
  - (t-1,t+2): -0.065*** (0.024)
  - (t,t+3): -0.122*** (0.023)
  - (t+1,t+4): -0.158*** (0.023)
  - (t+2,t+5): -0.169*** (0.023)
- Economic magnitudes:
  - One standard deviation increase in change in firm credit = 11.2 percentage points of GDP → associated with a 0.8 percentage points higher employment growth between t−3 and t and a 1.9 percentage points lower employment growth between t+2 and t+5.
- Household credit (coefficients on ΔHousehold credit (t-3,t)):
  - (t-3,t): 0.004 (0.030)
  - (t-2,t+1): -0.083*** (0.031)
  - (t-1,t+2): -0.160*** (0.030)
  - (t,t+3): -0.207*** (0.030)
  - (t+1,t+4): -0.212*** (0.032)
  - (t+2,t+5): -0.157*** (0.035)
- Economic magnitudes for household credit:
  - One standard deviation increase in credit to households = 9.1 percentage points of GDP → associated with 1.9 percentage points lower employment growth over t+1 to t+4 (fifth column).
- GDP dynamics:
  - Change in firm credit: (t-3,t): 0.094** (0.045); medium-term negative e.g., (t+2,t+5): -0.202*** (0.054).
  - Change in household credit: (t-3,t): -0.182*** (0.051).

### Firm-level results — leverage buildups and employment
- Baseline firm-level coefficients on ΔLeverage (t-3,t) for Δlog(Employment)j,t (Table 5):
  - (t-3,t): 0.025*** (0.001)
  - (t-2,t+1): -0.030*** (0.001)
  - (t-1,t+2): -0.055*** (0.001)
  - (t,t+3): -0.058*** (0.001)
  - (t+1,t+4): -0.025*** (0.001)
  - (t+2,t+5): -0.003*** (0.001)
- Economic magnitudes:
  - One standard deviation of change in leverage = 21.3 percentage points → associated with 0.5 percentage points higher employment growth during t−3 to t.
  - Moving from 5th to 95th percentile is a 76 percentage points increase → associated with a 1.9 percentage points boost in same period and a 4.4 percentage points lower employment growth between t and t+3 (example reported in source).
- Stylized sample medians and quartiles:
  - Firm-level sample median change in leverage = -1.6 percentage points.
  - Firms above median: 1.6 percentage points higher employment growth between t-3 and t; mean employment growth (3-year) in ORBIS sample = 1.3 percent.
  - Firms above median: 8.5 percentage points higher investment rate on average in t-3 to t; mean investment rate (3-year) = 7.2 percent.
  - Medium-term: firms with high leverage buildups see a 1.2 percentage points lower employment growth between t and t+3 and a 4 percentage points lower investment rate between t and t+3.

### Investment dynamics and balance-sheet pressures
- Investment (Table 16) — coefficients on ΔLeverage (t-3,t) for Δlog(Fixed assets):
  - (t-3,t): 0.227*** (0.004)
  - (t-2,t+1): 0.042*** (0.003)
  - (t-1,t+2): -0.077*** (0.002)
  - (t,t+3): -0.173*** (0.003)
  - (t+1,t+4): -0.081*** (0.003)
  - (t+2,t+5): -0.029*** (0.003)
- Economic magnitudes:
  - One standard deviation increase in leverage → predicts 4.8 percentage points higher investment rate during t−3 to t and 3.7 percentage points lower investment rate between t and t+3.
- Debt service ratio (Table 15):
  - Coefficients on ΔLeverage (t-3,t) for DSR:
    - p=0: -0.054*** (0.001)
    - p=1: 0.039*** (0.001)
    - p=2: 0.042*** (0.001)
    - p=3: 0.035*** (0.001)
    - p=4: 0.032*** (0.001)
    - p=5: 0.029*** (0.001)
  - Interpretation: short-term lower interest burden relative to earnings; medium-term higher interest burden consistent with mounting financial pressures and subsequent employment declines.

### Volatility and amplification by financial conditions
- Volatility of firm employment growth (Table 14) — coefficients on ΔLeverage (t-3,t):
  - (t-3,t): 0.008*** (0.000)
  - (t-2,t+1): 0.013*** (0.000)
  - (t-1,t+2): 0.012*** (0.000)
  - (t,t+3): 0.009*** (0.000)
  - (t+1,t+4): 0.006*** (0.001)
  - (t+2,t+5): 0.002*** (0.001)
- Role of aggregate financial conditions (interaction, Table 17):
  - Coefficients on ΔLeverage (t-3,t) for p = 1,...,5:
    - (t-2,t+1): -0.058*** (0.004)
    - (t-1,t+2): -0.070*** (0.003)
    - (t,t+3): -0.069*** (0.003)
    - (t+1,t+4): -0.022*** (0.004)
    - (t+2,t+5): -0.013*** (0.004)
  - Interaction ΔLeverage × Xc,t p (forecast errors of long-term rates):
    - (t-2,t+1): -0.007*** (0.001)
    - (t-1,t+2): -0.004*** (0.001)
    - (t,t+3): -0.003*** (0.001)
    - (t+1,t+4): -0.001** (0.001)
    - (t+2,t+5): -0.001* (0.001)
  - Interpretation: tighter-than-expected financial conditions amplify medium-term employment declines for firms with larger leverage buildups.

---

### 4. Robustness and Alternative Explanations
- Robustness checks preserve core pattern (short-term boost, medium-term decline) across:
  - Dummy-variable approaches (sample median, country-industry median, firm median): e.g., Panel A (sample median dummy) (t-3,t): 2.316*** (0.037); (t,t+3): -1.751*** (0.034).
  - Alternative leverage measure (net liabilities): (t-3,t): 0.033*** (0.001); (t,t+3): -0.042*** (0.001).
  - Controlling for convergence (initial employment), firm expansion (sales growth), and mean reversion — ΔLeverage patterns remain materially unchanged.
  - Subsamples by sector and firm characteristics (service vs non-service; firms with consistently ≥5 employees; firms with ≥10 years of data) — similar coefficient patterns reported.
- Conclusion: alternative mechanisms (catch-up, overexpansion, mean reversion) are relevant but do not overturn the central leverage–employment dynamics documented.

---

### 5. Conclusions and Policy Implications

### Summary of empirical regularities
- A rise in firm debt predicts boom-bust cycles in real outcomes in Europe:
  - Short-term: leverage expansions boost employment growth and investment.
  - Medium-term: leverage expansions associate with lower employment growth and investment rates, higher interest burdens, and greater volatility in employment growth.
- Household debt shows limited short-term boost to employment and predicts medium-term declines.

### Mechanism
- Evidence supports a financial-channel interpretation:
  - Leverage buildups increase DSRs over time, draining resources and hindering medium-term employment.
  - Tightening aggregate financial conditions amplifies negative medium-term outcomes.

### Policy recommendations
- Monitor and address firm leverage buildups to balance short-term growth benefits and medium-term macro‑financial costs.
- Consider well-designed and targeted macroprudential tools to strengthen firm balance sheets when large leverage buildups occur in the real sector.
- Tightening monetary policy can be an option to lean against credit booms under some conditions but should be used cautiously because it is less targeted and related economic costs can outweigh benefits (examples noted: Brandao-Marques et al. 2020; Biljanovska et al. 2023).

### Post‑Covid-19 context
- Nonfinancial sector leverage had been increasing before the pandemic and rose further following policymakers’ credit-support responses to Covid-19 (IMF 2021), underscoring a trade-off between short-term support and medium-term macro‑financial stability risks.

---

### 6. Selected summary statistics (Appendix, Table A.1)

- Panel A. Firm-level variables (25th ptile / Median / Mean / 75th ptile)
  - Leverage, 3-year growth (pp): -10.998 / -1.571 / -1.150 / 6.415
  - Net leverage, 3-year growth (pp): -14.988 / -1.598 / -1.485 / 10.306
  - Debt service ratio (%): 0.448 / 6.100 / 13.048 / 22.292
  - Number of employees (#): 2 / 5 / 16.873 / 14
  - Employment, 3-year growth (%): -14.310 / 0 / 1.318 / 18.232
  - Investment, 3-year (%): -10.721 / -1.640 / 7.209 / 38.870
  - Sales, 3-year growth (%): -25.640 / 1.242 / 4.328 / 32.249

- Panel B. Country-level variables (25th ptile / Median / Mean / 75th ptile)
  - Firm credit, 3-year growth (pp): -4.042 / 0.342 / 0.192 / 4.150
  - Household credit, 3-year growth (pp): -2.400 / 3.370 / 2.378 / 7.780
  - Employment, 3-year growth (%): 0.045 / 2.627 / 2.181 / 4.575
  - GDP, 3-year growth (%): 1.694 / 5.332 / 5.442 / 8.977

*Source: 1–5. Introduction, Data, Methodology, Results, Conclusions and Implications — wpiea2023126-print-pdf (IMF Working Paper).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Background and motivation
- Expansions in aggregate credit are associated with boom-bust cycles in macroeconomic outcomes; credit buildups in the boom phase are linked to elevated macro‑financial stability risks and subsequent declines in economic activity.
- The paper focuses on the dynamic relationship between firm leverage and real outcomes, using both aggregate and firm-level data from Europe over the period of 2000-2018.
- Theoretical mechanisms (Myers 1977; Bernanke and Gertler 1989; Kiyotaki and Moore 1997; Bernanke et al. 1999; Lorenzoni 2008; Jeanne and Korinek 2010; Bianchi 2011; Korinek and Simsek 2016) suggest firm leverage can relax financial constraints short-term but raise balance-sheet risks and tighten constraints medium-term.

### Contribution and data
- Scope:
  - Countries: 24 European economies.
  - Period: 2000-2018.
- Data sources:
  - Aggregate macro data on credit to firms and employment.
  - Firm-level panel data from the ORBIS database (non-farm, non-financial activities; includes listed and unlisted firms; constructed following Kalemli-Ozcan et al. (2015), Gopinath et al. (2017), Diez et al. (2021a)).
- Motivation for ORBIS:
  - Includes small and medium-size enterprises (SMEs) and regulated reporting for many European countries, capturing firms likely facing greater financial constraints.

### Empirical approach (key features)
- Uses 3-year “sliding windows” comparable to Mian et al. (2017) and Giroud and Mueller (2021).
- Explanatory variable: change in firm debt between year t−3 and t.
- Dependent variables: employment growth and investment measured over windows:
  - employment growth over t−3 to t; t−2 to t+1; t−1 to t+2; t to t+3; t+1 to t+4; t+2 to t+5 (six regressions).
- Firm-level estimations include:
  - firm fixed effects;
  - country-industry-year fixed effects (narrow 4-digit NACE × country × year cells) to absorb common shocks.

### Main empirical findings — aggregate level
- A rise in credit to firms predicts boom-bust cycles in aggregate employment growth:
  - A one standard deviation increase in credit to firms (as share of GDP) is associated with a 0.8 percentage points boost in aggregate employment growth within the same time span (i.e., between t−3 and t).
  - The same increase predicts a 1.9 percentage points lower growth in aggregate employment between t+2 and t+5.
- These aggregate patterns remain similar when changes in household debt are accounted for.
- Household debt findings (aggregate):
  - An increase in household debt does not show much evidence of a short-term boost to aggregate employment growth in this sample.
  - Aggregate employment growth declines in the medium-term following a rise in credit to households.
- Related note: Similar patterns reflected in GDP dynamics (rise in credit to firms associated with boom-bust cycles in GDP growth; household debt negatively associated with GDP growth short- and medium-term).

### Main empirical findings — firm level
- Firm leverage buildups predict boom-bust cycles in firm employment:
  - A one standard deviation increase in leverage is associated with a 0.5 percentage points boost in employment growth within the same time span (i.e., between t−3 and t).
  - Mean employment growth in the ORBIS sample (over a 3-year period) is 1.3 percent.
  - The relationship switches to negative in the medium-term: a one standard deviation rise in leverage (between t−3 and t) is associated with a 1.2 percentage points lower employment growth between t and t+3.
- Leverage buildups increase the volatility of firm employment growth both in the short- and medium-term.

### Investment and balance-sheet channels
- Investment dynamics:
  - A one standard deviation increase in leverage is associated with a 4.8 percentage points increase in investment rate in the same period (i.e., between t−3 and t).
  - Mean investment rate in the sample is 7.2 percent.
  - The same increase in leverage predicts a 3.7 percentage points lower investment rate between t and t+3.
- Debt service / balance-sheet pressures:
  - Firms with larger leverage buildups persistently use a larger fraction of their earnings for interest payments (debt service ratio), indicating worsening balance sheets over time.
  - This increased drag on finances is consistent with a financial channel that can hinder medium-term employment growth.

### Role of aggregate financial conditions and heterogeneity
- Exploiting cross-country heterogeneity in financial conditions:
  - Financial conditions proxied by forecast errors of long-term interest rates.
  - As financial conditions tighten, firms with initially larger leverage expansions experience larger declines in employment growth in the medium-term.
  - This interaction supports a financial-channel interpretation of the medium-term negative effects.

### Tests for alternative explanations and robustness
- The paper examines and addresses alternative explanations:
  - Employment convergence: firms with initially higher employment tend to grow less.
  - Firm expansion / overexpansion: larger leverage increases may reflect expansions that later reverse.
  - Mean reversion in employment growth.
- Results indicate these channels are important for employment dynamics but do not alter the core dynamic relationship between firm leverage buildups and employment growth.
- Firm-level specifications control for firm fixed effects and country-industry-year fixed effects to alleviate omitted-variable concerns.

### Synthesis and relation to existing literature
- Differences relative to Mian et al. (2017):
  - Mian et al. (2017) (30 countries, 1960-2012) found household debt mainly drove boom-bust cycles, with limited evidence for firm debt; present results differ, possibly due to sample/country/time-span differences.
- Consistency with other findings:
  - Medium-term negative patterns align with Greenwood et al. (2022).
  - Firm-level findings parallel Giroud and Mueller (2021) (US firms, 1976-2011): leverage buildups raise employment short-term and lower it medium-term.
- Novel contributions:
  - Demonstrates that firm leverage buildups predict boom-bust cycles in employment and investment for European firms (2000-2018).
  - Documents an associated rise in volatility of employment growth.
  - Provides suggestive evidence that balance-sheet deterioration and tighter aggregate financial conditions amplify medium-term declines.

### Conclusions (summary)
- A rise in firm debt predicts boom-bust cycles in real outcomes in Europe:
  - Short-term: leverage expansions boost employment growth and investment.
  - Medium-term: leverage expansions are associated with lower employment growth and lower investment rates, higher interest burdens, and greater volatility in employment growth.
- Evidence points to a financial-channel mechanism: deleterious medium-term effects are larger when aggregate financial conditions tighten.

*Source: 1. Introduction, wpiea2023126-print-pdf (IMF Working Paper).*

### 2. Data

### 2. Data

### Country-level Data
- Data sources and variables:
  - Credit to non-financial corporations (as percent of GDP): European Credit Research Institute database (ECRI 2021 Statistical Package).
  - Credit to households: ECRI 2021 Statistical Package.
  - Aggregate employment: OECD database.
  - Real GDP: World Bank World Development Indicators database.
  - Long-term (10-year) government bond yields and yields data: OECD database.
  - Forecasts for yields: fall issue of the OECD Economic Outlook in the previous year.
- Data processing:
  - Country-level variables are winsorized at the 1st and 99th percentiles to reduce the effect of outliers.
  - Forecast errors of long-term (10-year) government bond yields are computed as (the rate minus its forecast), following Ahn et al. (2020).
  - Forecast errors are used as an intuitive measure of a more-than-expected tightening in financing conditions in a country for a given year.
- Sample:
  - The sample is the same as the firm-level regressions (described below).
- Notes:
  - An advantage of long-term rates is to capture financing conditions for firms in a broader sense compared to short-term policy rates, and to envisage the effect of both conventional and unconventional monetary policy measures (Ahn et al. 2020).

### Firm-level Data
- Data source and coverage:
  - ORBIS database compiled by Bureau van Dijk Electronic Publishing (BvD).
  - Provides harmonized firm-level information on employment, sales, liabilities, assets, etc.
  - About 99 percent of firms in the dataset set are private.
  - The main sample covers non-farm, non-financial industries restricted by NACE 2-digit codes 5-82.
  - Sample size in the largest sample: about 2,5 million firms and 15,7 million observations.
  - Geographic and time coverage: 24 advanced European economies over the period of 2000-2018: Austria, Belgium, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Ireland, Iceland, Italy, Latvia, Lithuania, Luxembourg, Netherlands, Norway, Portugal, Slovakia, Slovenia, Spain, Sweden, Switzerland, and the United Kingdom.
- Importance of ORBIS and SME representation:
  - ORBIS includes both listed and unlisted firms and captures SMEs (about 99 percent of businesses in Europe; SMEs correspond to about 99 percent of all businesses in Europe and account for more than half of Europe’s GDP).
  - In the sample, the majority of firms are SMEs (accounting for around 98 percent of all observations) based on the Eurostat definition (less than 250 employees).
  - Median (mean) number of employees: 5 (17) as illustrated in the Appendix.
- Variable definitions and robustness:
  - Firm leverage: total liabilities as percent of total assets.
  - Alternative leverage measure: ratio of net liabilities (total liabilities minus cash) to total assets.
  - Employment: number of employees.
  - Sales growth: used in robustness checks as a proxy for firm expansion.
  - Investment: change in fixed assets (net investment; total fixed assets = tangible + intangible fixed assets).
  - Debt service ratio (DSR): percentage of interest payments to EBITDA, following Kalemli-Ozcan et al. (2022).
- Data cleaning and winsorization:
  - Raw ORBIS data is cleaned following Kalemli-Ozcan et al. (2015), Gopinath et al. (2017), and Diez et al. (2021a), including correcting reporting errors (e.g., negative total assets or employment) and merging vintages.
  - All firm-level variables are winsorized at the 2.5th and 97.5th percentile levels to reduce the influence of outliers.
- Additional notes:
  - The Appendix provides summary statistics and industry listings.
  - The majority of firms in ORBIS are private, which distinguishes it from datasets focusing on large/listed firms (e.g., Compustat, Compustat Global, Worldscope).

### 3. Methodology

### 3.1. Aggregate Credit and Employment Dynamics
- Objective:
  - Examine the dynamic relationship between credit expansions and employment growth at the aggregate level using panel regressions with fixed effects and 3-year sliding windows.
- Baseline specification (as presented in the source):
  - Δlog(Employment)_{c,t}(t+p−3,t+p) = α ΔFirm credit_{c,t}(t−3,t) + θ_c + θ_t + ε_{c,t}
- Variables and interpretation:
  - c and t denote country and year.
  - Explanatory variable: change in credit to nonfinancial firms (in percentage points of GDP) from t−3 to t.
  - Dependent variable: change in log employment (number of employees) for periods (t+ p −3, t+ p), expressed in percent.
  - The regression is run for p = 0,...,5, producing six regressions that map short- and medium-term relationships:
    - p=0: captures short-term relationship between buildups in firm credit and employment growth (between t−3 and t).
    - p increases: examines relationship over successive medium-term horizons (e.g., t−2 to t+1; t−1 to t+2; t to t+3; t+1 to t+4; t+2 to t+5).
  - A positive (negative) estimate for α implies an increase (decrease) in employment growth associated with an increase in credit to firms for the corresponding period.
- Controls and inference:
  - Country (θ_c) and year (θ_t) fixed effects included.
  - Standard errors are robust to heteroskedasticity.
- Extensions and robustness:
  - Replace right-hand-side variable with change in credit to households ΔHousehold credit_{c,t}(t−3,t).
  - Include both firm and household credit changes simultaneously.
  - Use log change in real GDP ΔGDP_{c,t}(t+p−3,t+p) as the dependent variable to investigate effects on economic growth.

### 3.2. Firm Leverage Buildups and Employment Dynamics
- Objective:
  - Investigate the dynamic relationship between firm leverage buildups and employment growth exploiting firm-level heterogeneity using panel regressions with fixed effects and 3-year sliding windows.
- Baseline specification (as presented in the source):
  - Δlog(Employment)_{j,t}(t+p−3,t+p) = α ΔLeverage_{j,t}(t−3,t) + θ_j + θ_{c,i,t} + ε_{j,t}
- Variables and interpretation:
  - j, c, i, and t denote firm, country, (4-digit NACE) industry, and year respectively.
  - Explanatory variable: percentage points change in firm leverage from t−3 to t.
  - Dependent variable: change in log employment for periods (t+ p −3, t+ p), expressed in percent.
  - Regressions run for p = 0,...,5 to trace effects from short- to medium-term horizons.
- Fixed effects and standard errors:
  - Firm fixed effects (θ_j) absorb firm-level time-invariant characteristics.
  - Country-industry-year fixed effects (θ_{c,i,t}) absorb factors common to all firms in a country and 4-digit industry in a given year (e.g., supply or demand shocks).
  - Standard errors are clustered at the country-industry-year level.
- Extensions and additional tests:
  - Add initial level of firm employment (at t−3) to examine convergence.
  - Control for firm sales growth (t−3 to t) to test whether firm expansion/over-expansion affects the leverage–employment link.
  - Control for firm employment growth between t−3 and t to explore mean reversion.
  - Examine volatility of firm employment growth using a time-variant measure of volatility as the dependent variable.
  - Investigate association between firm leverage buildups and debt service ratio where dependent variable is DSR_{j,t+p} (interest payments as percent of EBITDA).
  - Investigate relationship with investment by replacing dependent variable with Δlog(Fixed assets)_{j,t}.
- Role of aggregate financial conditions (interaction specification):
  - Specification (as presented in the source):
    - Δlog(Employment)_{j,t}(t+p−3,t+p) = α ΔLeverage_{j,t}(t−3,t) + β ΔLeverage_{j,t}(t−3,t) × X_{c,t}^p + θ_j + θ_{c,i,t} + ε_{j,t}
  - X_{c,t}^p denotes the average forecast errors of long-term interest rates (proxy for surprise component of financial tightening) during each period (e.g., for p=3, average of forecast errors over t+1, t+2, and t+3).
  - Country-industry-year fixed effects absorb the direct role of X_{c,t}^p on firm employment growth; the interaction captures differential effects by firm leverage buildups.
  - Estimated for p = 1,...,5 to test whether future tightening in financial conditions amplifies medium-term employment declines for firms with larger leverage buildups.
  - Interpretation example from source: if α < 0 for p=3 and β < 0, then (i) firms with larger leverage buildups experience a decline in employment between t and t+3, and (ii) this decline becomes larger if financial conditions tighten more than expected during that period.

*IMF Working Paper — Chapter 2 (Data) and Methods excerpt*

### 4. Results

### 4. Results

### Stylized Facts — Aggregate Patterns
- Change in credit to firms classification: observations with change above (below) sample median; sample median = 0.3 percentage points of GDP.
- Short-term aggregate employment differential:
  - Countries with higher increase in credit to firms experience a 0.8 percentage points higher employment growth between t-3 and t on average.
  - Mean employment growth (over a 3-year period) in the aggregate sample = 2.2 percent.
- Medium-term reversal:
  - Between t+2 and t+5, countries with larger expansion in firm credit see a 1.5 percentage points lower employment growth relative to the rest of the sample.
- Household credit dynamics:
  - Initial short-term boost smaller: 0.2 percentage points higher employment growth for larger expansion in household credit in the first period.
  - Relationship turns negative immediately after the first period and stays negative.
  - Between t and t+3 (fourth bar), countries with larger household credit buildups experience a 1.1 percentage points lower employment growth relative to others.
- Aggregate conclusion:
  - Greater credit buildups likely boost aggregate employment growth in the short-term, and the relationship switches to negative in the medium-term.
  - Initial boost is larger following firm credit expansion compared to household credit expansion.

### Stylized Facts — Firm-level Patterns
- Firm leverage classification: high (low) leverage growth = change above (below) sample median; sample median = -1.6 percentage points.
- Short-term firm employment and investment:
  - Firms with high leverage buildups experience a 1.6 percentage points higher employment growth between t-3 and t on average.
  - Mean employment growth (3-year) in the ORBIS sample = 1.3 percent.
  - Firms with high leverage have 8.5 percentage points higher investment rate on average in the same period.
  - Mean investment rate (3-year) in the sample = 7.2 percent.
- Medium-term reversal at firm level:
  - Firms with high leverage buildups see a 1.2 percentage points lower employment growth between t and t+3 relative to other firms.
  - Firms with high leverage see a 4 percentage points lower investment rate between t and t+3 compared to other firms.
- Quartile evidence (Figure 3):
  - Short-term: firms in first quartile of change in leverage have average employment growth -1.1 percent during t-3 to t; firms in fourth quartile have 0.9 percent.
  - Medium-term (t+1 to t+4): first quartile firms have 0.6 percent employment growth; fourth quartile firms have -0.4 percent.
- Firm-level conclusion:
  - Leverage buildups are associated with higher firm employment growth and investment in the short-term, and lower employment growth and investment in the medium-term; patterns align with aggregate evidence.

### Aggregate Credit and Employment Dynamics (Regression Evidence)
- Specification: equation 1 for p = 0,...,5.
- Firm credit (Table 1):
  - Dependent variable Δlog(Employment)c,t over periods (t-3,t) ... (t+2,t+5).
  - Coefficient on ΔFirm credit (t-3,t):
    - (t-3,t): 0.073** (0.029)
    - (t-2,t+1): -0.002 (0.027)
    - (t-1,t+2): -0.065*** (0.024)
    - (t,t+3): -0.122*** (0.023)
    - (t+1,t+4): -0.158*** (0.023)
    - (t+2,t+5): -0.169*** (0.023)
  - R-squared range: 0.486 to 0.661; Observations = 366 down to 245 across columns.
  - Economic magnitude: one standard deviation increase in change in firm credit = 11.2 percentage points of GDP → associated with 0.8 percentage points higher employment growth short-term; same increase predicts 1.9 percentage points lower employment growth in medium-term (last column).
- Household credit (Table 2):
  - Coefficient on ΔHousehold credit (t-3,t):
    - (t-3,t): 0.004 (0.030)
    - (t-2,t+1): -0.083*** (0.031)
    - (t-1,t+2): -0.160*** (0.030)
    - (t,t+3): -0.207*** (0.030)
    - (t+1,t+4): -0.212*** (0.032)
    - (t+2,t+5): -0.157*** (0.035)
  - Economic magnitude: one standard deviation increase in credit to households = 9.1 percentage points of GDP → associated with 1.9 percentage points lower employment growth over t+1 to t+4 (fifth column).
- Joint inclusion of firm and household credit (Table 3):
  - Firm credit coefficients become:
    - (t-3,t): 0.125*** (0.044)
    - (t,t+3): -0.090*** (0.027)
    - (t+1,t+4): -0.123*** (0.029)
    - (t+2,t+5): -0.128*** (0.029)
  - Household credit coefficients are negative and significant in short- and medium-term (e.g., (t-3,t): -0.093** (0.039); (t-2,t+1): -0.116*** (0.038)).
- GDP dynamics (Table 4):
  - Change in firm credit predicts initial boost in GDP growth:
    - (t-3,t): 0.094** (0.045)
    - Medium-term negative coefficients, e.g., (t+2,t+5): -0.202*** (0.054)
  - Change in household credit associated with GDP decline in short-term and medium-term, e.g., (t-3,t): -0.182*** (0.051).

### Firm Leverage Buildups and Employment Dynamics (Firm-level Regression)
- Specification: equation 2, p = 0,...,5 (Table 5).
- Coefficients on ΔLeverage (t-3,t) for Δlog(Employment)j,t:
  - (t-3,t): 0.025*** (0.001)
  - (t-2,t+1): -0.030*** (0.001)
  - (t-1,t+2): -0.055*** (0.001)
  - (t,t+3): -0.058*** (0.001)
  - (t+1,t+4): -0.025*** (0.001)
  - (t+2,t+5): -0.003*** (0.001)
- Sample sizes and fit:
  - Observations range from 15,716,519 down to 5,142,150; R-squared range 0.359 to 0.419.
- Economic magnitude:
  - One standard deviation of change in leverage = 21.3 percentage points → associated with 0.5 percentage points higher employment growth during same period.
  - Example from footnote: moving from 5th to 95th percentile is a 76 percentage points increase → associated with 1.9 percentage points boost in same period and 4.4 percentage points lower employment growth between t and t+3.

### Robustness Checks
- Dummy variable approach (Table 6):
  - Panel A (sample median dummy):
    - (t-3,t): 2.316*** (0.037)
    - (t,t+3): -1.751*** (0.034)
    - Observations: same large sample as main regression.
  - Panel B (country-industry median dummy) and Panel C (firm median dummy) show similar patterns with coefficients of similar magnitudes.
  - Interpretation: firms with relatively high leverage buildups see higher short-term employment growth and persistent medium-term declines (e.g., up to 1.8 percentage points decline as suggested by column 4 in footnote).
- Alternative measure of leverage — net liabilities (Table 7):
  - Coefficients on ΔLeverage (net of cash):
    - (t-3,t): 0.033*** (0.001)
    - (t,t+3): -0.042*** (0.001)
    - Observations range 14,969,478 down to 4,839,984; R-squared 0.364 to 0.426.
  - Conclusion: findings robust when leverage measured as total liabilities minus cash.
- Alternative explanations tested:
  - Catch-up (convergence) — Table 8:
    - Including log(Employment)j,t( t-3 ) control: coefficient on log(Employment) negative and significant across periods (e.g., -0.629*** (0.002) for (t-3,t)), but ΔLeverage pattern remains similar: (t-3,t) 0.016*** (0.001); (t,t+3) -0.063*** (0.001).
  - Firm expansion (sales growth) — Table 9:
    - Δlog(Sales) coefficients: positive short-term (0.295*** (0.002) for (t-3,t)), but negative in later periods; ΔLeverage pattern remains similar (e.g., (t-3,t) 0.050*** (0.001); (t,t+3) -0.060*** (0.001)).
  - Mean reversion — Table 10:
    - Controlling for prior employment growth, ΔLeverage coefficients remain negative in medium-term (e.g., (t-2,t+1) -0.046*** (0.001); (t,t+3) -0.037*** (0.001)).
  - Conclusion: employment convergence, firm expansion, and mean reversion do not alter the core relationship.
- Sample variations (Tables 11–13):
  - Service industries (Panel A, Table 11):
    - ΔLeverage: (t-3,t) 0.030*** (0.001); (t,t+3) -0.059*** (0.001); Observations 10,211,903 down to 3,235,367.
  - Non-service industries (Panel B, Table 11):
    - ΔLeverage: (t-3,t) 0.012*** (0.002); (t,t+3) -0.056*** (0.002); Observations 5,504,616 down to 1,906,783.
  - Firms with consistently ≥5 employees (Table 12):
    - ΔLeverage: (t-3,t) 0.018*** (0.001); (t,t+3) -0.044*** (0.001); Observations 5,506,573 down to 2,123,342.
  - Firms with ≥10 years of data (Table 13):
    - ΔLeverage: (t-3,t) 0.035*** (0.001); (t,t+3) -0.060*** (0.001); Observations 10,871,059 down to 5,114,534.
  - Findings remain similar across subsamples and alternative weighting/continuing-sample tests.

### Volatility of Employment Growth (Table 14)
- Volatility measure constructed from absolute residuals of 3-year firm employment growth regressions net of firm and country-industry-year fixed effects.
- Coefficients on ΔLeverage (t-3,t) for volatility:
  - (t-3,t): 0.008*** (0.000)
  - (t-2,t+1): 0.013*** (0.000)
  - (t-1,t+2): 0.012*** (0.000)
  - (t,t+3): 0.009*** (0.000)
  - (t+1,t+4): 0.006*** (0.001)
  - (t+2,t+5): 0.002*** (0.001)
- Observations: 15,716,519 down to 5,142,150; R-squared 0.473 to 0.549.
- Conclusion: leverage buildups predict increased volatility of firm employment growth both short- and medium-term.

### Financial Distress — Debt Service Ratio (Table 15)
- Dependent variable: DSR (interest paid as percentage of EBITDA).
- Coefficients on ΔLeverage (t-3,t):
  - p=0: -0.054*** (0.001)
  - p=1: 0.039*** (0.001)
  - p=2: 0.042*** (0.001)
  - p=3: 0.035*** (0.001)
  - p=4: 0.032*** (0.001)
  - p=5: 0.029*** (0.001)
- Observations: 11,315,551 down to 4,179,399; R-squared 0.341 to 0.384.
- Interpretation:
  - Short-term: leverage buildups associated with lower interest payments relative to earnings (possible expansion effects).
  - Medium-term: firms with higher leverage buildups spend a higher fraction of earnings on interest payments → evidence of mounting financial pressures consistent with declines in employment growth.

### Investment Dynamics (Table 16)
- Dependent variable: Δlog(Fixed assets) j,t (investment rate change).
- Coefficients on ΔLeverage (t-3,t):
  - (t-3,t): 0.227*** (0.004)
  - (t-2,t+1): 0.042*** (0.003)
  - (t-1,t+2): -0.077*** (0.002)
  - (t,t+3): -0.173*** (0.003)
  - (t+1,t+4): -0.081*** (0.003)
  - (t+2,t+5): -0.029*** (0.003)
- Observations: 14,221,120 down to 4,706,550; R-squared 0.385 to 0.417.
- Economic magnitude:
  - One standard deviation increase in leverage → predicts 4.8 percentage points higher investment rate during same period.
  - Medium-term: one standard deviation leverage buildup associated with 3.7 percentage points lower investment rate between t and t+3.
- Conclusion: leverage buildups produce boom-bust cycles in investment analogous to employment.

### Role of Aggregate Financial Conditions (Table 17)
- Specification: equation 3 includes interaction ΔLeverage × Xc,t p (average forecast errors of long-term rates as proxy for tightening financial conditions).
- Coefficients on ΔLeverage (t-3,t) for p = 1,...,5:
  - (t-2,t+1): -0.058*** (0.004)
  - (t-1,t+2): -0.070*** (0.003)
  - (t,t+3): -0.069*** (0.003)
  - (t+1,t+4): -0.022*** (0.004)
  - (t+2,t+5): -0.013*** (0.004)
- Coefficients on interaction ΔLeverage × Xc,t p:
  - (t-2,t+1): -0.007*** (0.001)
  - (t-1,t+2): -0.004*** (0.001)
  - (t,t+3): -0.003*** (0.001)
  - (t+1,t+4): -0.001** (0.001)
  - (t+2,t+5): -0.001* (0.001)
- Observations: 10,300,308 down to 4,498,481; R-squared 0.401 to 0.435.
- Interpretation:
  - As financial conditions become tighter, firms with larger leverage expansions experience disproportionately larger losses in employment growth in the medium-term.
  - Provides evidence for a financial channel amplifying medium-term negative effects of leverage buildups.

*Italic: Source — 4. Results, wpiea2023126-print-pdf (IMF Working Paper).*

### 5. Conclusions and Implications

### 5. Conclusions and Implications

### Key findings on credit and employment dynamics
- Expansions in credit to firms predict a boost in aggregate employment growth initially, but employment growth declines in the medium-term.
- This result holds even when household debt dynamics are accounted for.
- Accumulation of household debt is not much associated with an initial boost in employment growth; however, similar to a rise in firm debt, it predicts a decline in employment growth in the medium-term.

### Firm-level evidence
- Firm leverage expansions predict a boost in firm employment growth in the short-term, whereas employment growth decreases in the medium-term.
- Firm-level specifications absorb effects of other factors on employment growth at a granular level; robustness tests rule out various alternative explanations for this dynamic relationship.
- Boom-bust growth cycles in firm employment predicted by leverage buildups increase the volatility of employment growth both in the short- and medium-term.

### Financial channel and mechanisms
- Firms with a larger increase in leverage persistently use a larger fraction of their earnings for interest payments, leaving fewer resources for production activities and potentially hindering employment growth in the medium-term.
- Boom-bust cycles predicted by firm leverage buildups are also pronounced for investment: firm leverage buildups promote investment in the short-term, while holding investment back in the medium-term.
- Cross-country heterogeneity in financial conditions matters: firms with an initially larger expansion in leverage face even larger declines in employment growth in the medium-term if financial conditions tighten, supporting a financial channel.

### Policy implications and recommendations
- Policies that incentivize or allow firms to increase borrowing (for example, loose macroprudential policies or low interest rates) can be growth-enhancing in the short-term but may yield undesirable medium-term outcomes due to leverage buildups in the real sector.
- The findings favor more proactive policy measures to “lean against the wind of incipient credit booms”:
  - Consider well-designed and targeted macroprudential tools to strengthen firm balance sheets when large leverage buildups occur in the real sector, to balance short-term benefits and medium-term costs of rising leverage levels.
  - Under some conditions, tightening monetary policy can be another option to lean against the wind, but it should be treated with caution because it is less targeted relative to macroprudential tools and related economic costs can outweigh benefits (examples noted: Brandao-Marques et al. 2020; Biljanovska et al. 2023).

### Implications in the post‑Covid-19 context
- Nonfinancial sector leverage had been increasing running up to the pandemic, reaching historical highs, due to a loosening in financial conditions since the Global Financial Crisis in 2008.
- Policymakers’ responses to the Covid-19 shock, aimed at supporting the flow of credit, have contributed to a further increase in nonfinancial sector leverage (IMF 2021).
- The study’s findings highlight a policy trade-off between supporting growth in the short-term through easing financial conditions and containing downside macro‑financial stability risks going forward.

### Selected firm- and country-level summary statistics (from Appendix, Table A.1)
- Panel A. Firm-level variables (25th ptile / Median / Mean / 75th ptile)
  - Leverage, 3-year growth (pp): -10.998 / -1.571 / -1.150 / 6.415
  - Net leverage, 3-year growth (pp): -14.988 / -1.598 / -1.485 / 10.306
  - Debt service ratio (%): 0.448 / 6.100 / 13.048 / 22.292
  - Number of employees (#): 2 / 5 / 16.873 / 14
  - Employment, 3-year growth (%): -14.310 / 0 / 1.318 / 18.232
  - Investment, 3-year (%): -10.721 / -1.640 / 7.209 / 38.870
  - Sales, 3-year growth (%): -25.640 / 1.242 / 4.328 / 32.249
- Panel B. Country-level variables (25th ptile / Median / Mean / 75th ptile)
  - Firm credit, 3-year growth (pp): -4.042 / 0.342 / 0.192 / 4.150
  - Household credit, 3-year growth (pp): -2.400 / 3.370 / 2.378 / 7.780
  - Employment, 3-year growth (%): 0.045 / 2.627 / 2.181 / 4.575
  - GDP, 3-year growth (%): 1.694 / 5.332 / 5.442 / 8.977

*Source: IMF Working Paper — 5. Conclusions and Implications (wpiea2023126-print-pdf).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023126-print-pdf.pdf_