## wp1738

## Source details

**Canonical URL:** [wp1738](https://www.imf.org/-/media/files/publications/wp/2017/wp1738.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp1738.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp1738.pdf.json)

---

### Key findings on employment effects of credit shocks
- Job losses attributable to the credit shock are mostly borne by temporary employees; wages not observed so wage response marginally analyzed.
- Analysis explains about a fourth of the fall in aggregate employment in Spain between 2007 and 2010 (comparative literature cited).
- Financing constraints reduce employment growth in small U.S. firms by 5 to 10 percentage points relative to large firms (Siemer (2016) evidence cited).
- Veneto sample, 2008–2012:
  - Number of employees declines by nearly 90,000 units.
  - Average job loss for the average firm = 2.1 employees.
  - Credit declined by 1.6 percent for the average firm.
  - One fourth of firms experienced a negative change in employment.
  - Credit contracted for more than half of the firms in the sample.
- Baseline 2SLS elasticity: coefficient on ∆LOAN = 0.36 (column 4 baseline). Interpretation: a 10 percent contraction in bank lending over two quarters → a 3.6 percent reduction in employment.
- Aggregate extrapolation: credit drop (1.6 percent) and employment drop (2.1 percent) imply lending-supply orientation explains about 28 percent of employment loss (simple algebra, caveat on GE effects).

### Data scope, representativeness, and sample construction (Veneto case study)
- Region characteristics:
  - Population = 4.9 million.
  - Workforce = 2.2 million.
  - Veneto ≈ 9 percent of Italian value added and of total employment.
  - 97 percent of workers resident in the region have workplace within the region; migration to other regions = 0.4 percent of the population per year.
- Firm size distribution in Veneto:
  - 94 percent of firms have less than 10 employees.
  - 57 percent of firms have at most one employee.
- Sectoral employment shares: services = 56 percent; industry = 43 percent.
- Banking system in Veneto (2012): about 120 banks; small local banks ≈ 20 percent of business loans.
- Final cleaned sample (matching via VAT/fiscal code):
  - Nearly 440,000 firms (436,311 after filters), of which about 200,000 (204,301) have bank relationships.
  - Sample excludes public sector, primary (agriculture), financial sector firms; removes temp agencies, care givers, house cleaners.
- PLANET coverage: difference with official labor force < 7 percent in 2012 for employed workers in non-financial private sector.
- Credit Register (CR) reporting thresholds:
  - 75,000 euro until December 2008.
  - 30,000 euro afterwards.

### Data sources and matching
- Primary administrative sources:
  - PLANET: daily labor market flows; worker-level variables include gender, age, nationality, occupation (5-digit ISCO), contract type (44 categories), educational attainment (13 categories), time schedule, separation reasons.
  - ASIA: archive of active firms (yearly firm-level stock; used to reconstruct quarterly stock).
  - Credit Register (CR): credit to each firm each quarter; monthly reporting of loans above thresholds and all nonperforming loans.
  - Bank of Italy Supervisory Report (SR): confidential outstanding loans by banks to provinces (used to estimate bank lending policies).
  - CADS (Cerved): balance-sheet and income statement data for a sub-sample (used in Section 5).
- Matching identifier: VAT/fiscal code.
- Final cleaned sample sizes:
  - 436,311 firms (after cleansing).
  - 204,301 firms matched with Credit Register (have credit relationships).

### Identification strategy and empirical model
- Main specification:
  - ∆EMPLOYMENT_it = β ∆LOAN_it + δ_i + (γ_{s(i)} × τ_t) + (η_{c(i)} × τ_t) + (θ_{p(i)} × τ_t) + e_it
- Change formula (equation 2):
  - ∆X_it = (X_it1 − X_it0) / (0.5×X_it1 + 0.5×X_it0)
  - Variations are symmetric and bounded between −2 (exiters) and +2 (entrants).
- Baseline loan regressor: average change in used loans over two quarters (∆LOAN_t,t−1); robustness uses contemporaneous ∆LOAN_t and three-quarter average ∆LOAN_t,t−2.
- Fixed effects:
  - Firm fixed effects (δ_i) and quarter fixed effects.
  - Industry (2-digit) × quarter FE, Province × quarter FE, Firm class size × quarter FE.
  - Robustness with more granular industry-province-size-quarter FE to absorb borrower-level time-varying demand shocks.
- Endogeneity and instrument:
  - Bank lending endogenous to firm conditions; IV approach constructs time-varying firm-specific Credit Supply Index (CSI_it) following Greenstone et al. (2014) extension.
  - CSI_it uses banks’ loan shares to a firm (w_bit^0, end-2007) and bank-time fixed effects (ˆδ_bt) estimated from ∆L_bpst = α + δ_bt + γ_pst + e_bpst.
  - CSI_it = ∑_b w_bit^0 × ˆδ_bt used as instrument for ∆LOAN_it to isolate credit supply shocks.

### Credit Supply Index (CSI) — construction, validation, and stylized facts
- Bank-time fixed effects estimated from:
  - ∆L_bpst = percentage change in outstanding business loans by bank b, province p, sector s, time t.
  - δ_bt = bank-time fixed effects (nationwide bank lending policies); γ_pst = province-sector-quarter FE (local demand).
- CSI_it = ∑_b w_bit^0 × ˆδ_bt where w_bit^0 = bank b market share for firm i at end-2007.
- Stylized facts and validation:
  - Evolution of bank lending policies mimics growth rate of business loans; stronger correlation early in crisis.
  - Banks split by ˆδ_bt median: tight banks show more negative loan patterns; large supply contraction around 2009 driven by banks with lowest ˆδ_bt.
  - (Inverse of) CSI_it closely follows: ECB Bank Lending Survey diffusion index, share of rationed firms (Bank of Italy firm survey), corporate credit rationing indicator (Burlon et al. 2016). Correlations between CSI and these measures vary between 0.5 and 0.6.
  - Regressing average ˆδ_bt (2008–2012) on end-2007 bank characteristics: worsening supply higher for larger banks, banks with larger funding gap (deposit-to-loan ratio), and banks with lower capital.

### Exogeneity concerns and checks
- Exogeneity relies on w_bit^0 and ˆδ_bt:
  - w_bit^0: assume initial bank market shares not correlated with firm-level employment trend after firm FE and granular time-varying FE. Table 3 balancing: average exposure to credit shocks not significantly correlated with firm size at beginning-of-period.
  - ˆδ_bt: bank-time FE purged of province-sector-quarter factors; unlikely firm-level unobservables drive nationwide bank policies.
- Empirical checks:
  - Table 3: no systematic correlation between exposure to credit supply shocks and firm characteristics (size, financial dependence, banking relationships, leverage, bad credit history, location, sector).
  - CADS sub-sample: instrument not correlated with labor productivity, leverage, riskiness.
- Additional design choices:
  1. Exclude Veneto provinces when estimating bank lending policies (Veneto excluded; exclusion affects only one bank accounting for < 0.1 percent of loans to Veneto firms).
  2. Translate bank-time fixed effects to firm level rather than aggregate county level.
  3. Control for province-sector-time unobservables (more granular than Greenstone et al. (2014)).
  4. Limited government interventions in Italy reduces bias from public support conditionalities.

### Main empirical results and magnitudes
- First-stage: CSI positively associated with change in used loans; 1st-stage F-statistics range between 156 and 180 (well above critical 10).
  - Example first-stage coefficient: CS I_t,t−1 = 0.0751 to 0.0786*** (standard errors ~0.00609–0.00622).
- Second-stage baseline (column 4, firm & quarter FE; industry×quarter, size×quarter, province×quarter):
  - ∆LOAN coefficient = 0.363*** (standard error 0.0689) reported across several tables; baseline rounded as 0.36 in summaries.
  - Interpretation: 10 percent credit contraction → 3.6 percent employment reduction.
  - A standard deviation of predicted change of used loan explains 18% of the standard deviation of employment.
- Aggregate/sample-level: credit fell 1.6 percent and employment fell 2.1 percent (quarterly; Table 1); credit supply orientation explains about 28 percent of employment loss.
- Comparative magnitudes: elasticity ~0.36 roughly twice Cingano et al. (2016) on larger Italian firms; comparable to Bentolila et al. (2016) (Spain) and Chodorow-Reich (2014) (U.S.) with caveats on sample composition.

### Robustness exercises (selected)
- Saturating fixed effects (industry×size×quarter, industry×province×quarter, province×size×quarter; four-way FE): ∆LOAN coefficient stable between 0.34 and 0.37.
- Clustering SEs at province-industry-class size: SE marginally increases; coefficient remains highly significant.
- Restrict sample to firms with total debt > 75,000 euros throughout sample: ∆LOAN coefficient increases to 0.49 (precisely estimated).
- Adding municipality-level house prices: inclusion does not change loan variable coefficients; HOUSE PRICE_t coefficient = −0.0341** (0.0163) in some specifications.
- Alternative ∆LOAN timing:
  - ∆LOAN_t coefficient = 0.229*** (0.0510) (contemporaneous).
  - ∆LOAN_t,t−2 coefficient = 0.620*** (0.0976) (three-quarter average).
- Instrument relevance robust: first-stage F-statistics consistently between 156 and 180 in main specifications.

### Job contract heterogeneity (Section 4.3 / Table 6)
- Employment adjustment primarily via temporary contracts rather than open-ended contracts.
  - Temporary contracts account for slightly more than one tenth of contracts but bear almost half of the effect: 0.191/0.363 = 0.53.
- Coefficients on ∆LOAN positive and significant for both contract types; outflows drive results:
  - Inflows coefficient = 0.0834* (0.0447).
  - Outflows coefficient = −0.277*** (0.0617).
- Within outflows, adjustment due to non-renewal/expiry:
  - Expiry coefficient = −0.158*** (0.0330).
  - Dismissal: −0.0357 (0.0222) — not significant for driving main effect.
- Firms exposed to negative credit shocks are less likely to convert temporary contracts into open-ended ones.
- Financing constraints do not appear to affect transitions between part-time and full-time jobs.

### Worker heterogeneity (Section 4.3 / Tables 7–9)
- By education (∆LOAN_t,t−1 coefficients):
  - Low-educated: 0.186*** (0.0403)
  - Medium-educated: 0.151*** (0.0371)
  - High-educated: 0.0261* (0.0144)
  - Low-educated elasticity = 0.48 (=0.186/0.387) relative measure reported.
  - Low-educated changes account for 0.186/0.363 = 0.51 of total effect despite <40 percent workforce share.
- By skill (occupation-based):
  - Low-skill: 0.156*** (0.0397) — low-skill ≈15 percent of jobs but ≈43 percent of total effect (0.156/0.363 = 0.43).
- By gender/age/nationality:
  - Male: 0.144*** (0.0421)
  - Female: 0.220*** (0.0419) — females ≈40 percent of employment but account for 61 percent of employment change.
  - Under 30: 0.105*** (0.0295) — under 30 represent <20 percent of workforce but contribute slightly less than a third of overall effect.
  - Foreign: 0.0885*** (0.0245) — foreigners <10 percent of labor force but explain >24 percent of total change.
- Interaction contract × education (Table 9):
  - Open-ended by education: Low 0.0683*** (0.0248); Medium 0.0816*** (0.0247); High 0.0214*** (0.00821).
  - Temporary by education: Low 0.115*** (0.0301); Medium 0.0716*** (0.0275); High 0.00420 (0.0121) — high-educated temporary not significantly affected.

### Firm heterogeneity (Section 4.5, Table 10–11)
- Size:
  - Micro (<10): ∆LOAN = 0.327*** (0.0719)
  - Small (10–49): ∆LOAN = 0.618*** (0.226)
  - Med-Large (≥50): point estimate 3.7370 (7.980) — not significant; weak first-stage identification for medium-large.
- Sector:
  - Industry: 0.1670 (0.135) — not significant.
  - Services: 0.427*** (0.0803) — employment reacts to credit shocks only in services.
- Financial characteristics:
  - Debt per employee:
    - Low debt: 0.269*** (0.0682)
    - High debt: 0.498*** (0.145)
  - Credit line utilization:
    - Low utilization: 0.276*** (0.0760)
    - High utilization: 0.460*** (0.129)
  - Relationship lending:
    - Single-bank: 0.369*** (0.114)
    - Multi-banks: 0.773*** (0.260) — point estimate for multi-bank twice that for single-bank.
- Interpretation: more indebted firms, intensely using credit lines, and firms with multiple-bank relationships show larger employment sensitivity to credit supply.

### Intensive vs. extensive margins (Section 5.1 / Table 12)
- Intensive margin (excluding firm closures):
  - Excluding exits, a 10 percent contraction in credit → 2.5 percent fall in employment (∆LOAN coefficient = 0.252*** (0.0572)).
  - Intensive-margin reduction concentrated on temporary workers: temporary contracts account for ~three quarters of fall in employment in restricted sample (vs ~50 percent in whole sample).
  - Active firms reduce employment mostly among low- and medium-educated workers; high-educated temporary workers relatively insulated.
- Extensive margin (firm exit):
  - Shortfall in credit supply increases firm exit probability.
  - Given average credit contraction 1.6 percent, estimated coefficient implies a 0.1 percent increase in firm closure probability, accounting for ≈ one seventh of average exit rate.

### Matching firm balance sheets (CADS sub-sample, Section 5.2 / Tables 15, 14)
- CADS-matched sample trade-offs:
  - Move from quarterly to yearly frequency; lose coverage of micro firms.
  - Restricted sample average (median) firm size = 16.4 (4.8) employees vs whole sample ~6.3 (median 2).
- Validation checks:
  - CSI not systematically correlated with labor productivity, capital intensity, age, leverage, riskiness in CADS sub-sample (Table 14 orthogonality).
- Yearly-frequency results:
  - Whole sample annual ∆LOAN coefficient = 0.347*** (0.0504), Observations = 715,920, 1st-stage F = 421.87.
  - CADS sample annual ∆LOAN coefficient = 0.0952** (0.0459), Observations = 169,881, 1st-stage F = 1.267 (weak instrument in CADS).
  - CADS sample ∆CAPITAL_t coefficient = 1.210*** (0.315) — investment effect stronger than labor effect.
- Heterogeneity in CADS sample:
  - Age: effects limited to young firms; Young (Age) coefficient = 0.247** (0.110); Old = −0.00888 (0.0464).
  - Risk (Altman z-score median split): high-risk firms elasticity ≈ three times that of low-risk firms (High 0.179* (0.101); Low 0.0445 (0.0462)).
  - Leverage (median split): high-indebted firms elasticity ≈ four times low-indebted firms (High 0.195** (0.0897); Low 0.0293 (0.0487)).
  - Productivity: low-productive firms: 10 percent reduction in credit → 1.8 percent employment reduction reported (annual comparison); low-productive firms more affected than high-productive (High −0.0162 (0.0681) not significant).

### Limitations and methodological caveats
- Wages not observed — analysis focuses on quantity (employment) responses, not price (wage) effects.
- Full-sample lack of balance-sheet information prevents controlling for certain drivers of employment decisions and assessing capital accumulation effects (addressed via CADS sub-sample).
- Interest rates largely excluded because data available only for sub-sample of banks (excluding majority of small/local banks); utilized loans used to capture quantity and implicit price effects via utilization.
- Identification assumptions:
  - Firm unobserved heterogeneity driving labor demand is time invariant.
  - Firms in same 2-digit industry, province, and class size face same demand/productivity shock in each quarter.
- CADS sub-sample exhibits weak first-stage in some annual specifications (1st-stage F-statistics ~1.26), limiting causal interpretation in that restricted sample.

### Mechanisms, interpretation, and implications from results
- Employment adjustment occurs via:
  - Intensive margin reductions (especially temporary contracts).
  - Elevated firm exit probabilities (extensive margin).
- Credit crunch disproportionately affects:
  - Micro and small firms,
  - Younger firms,
  - Less productive firms,
  - High-debt firms (debt overhang),
  - Firms with weaker bank-firm relationships.
- Labor-market composition effects:
  - Less educated and less skilled workers with temporary contracts bear most of employment losses.
  - Evidence consistent with firms using temporary contracts as buffer stock and with skill-upgrading dynamics (firms dismiss less skilled workers first).
- Potential reallocation/cleansing channel:
  - Financial shocks may foster aggregate productivity gains by disproportionately affecting less productive firms and unskilled jobs (consistent with observed heterogeneity).

*Source: wp1738 - IMF working paper (PDF chapters and tables provided).*

### 2.8 percentage points more than firms that were borrowing from healthier lenders, and results

### wp1738 - 2.8 percentage points more than firms that were borrowing from healthier lenders, and results

### Key findings on employment effects of credit shocks
- Job losses attributable to the credit shock are mostly borne by temporary employees, while wages adjusted only marginally.
- Analysis is able to explain about a fourth of the fall in aggregate employment in Spain between 2007 and 2010 (as reported in comparative literature cited).
- Financing constraints reduce employment growth in small U.S. firms by 5 to 10 percentage points relative to large firms (Siemer (2016) evidence cited).
- In the Veneto sample, over the sample period 2008-2012:
  - Number of employees declines by nearly 90,000 units.
  - Average job loss for the average firm is equal to 2.1 employees.
  - Credit declined by 1.6 percent for the average firm.
  - One fourth of firms experienced a negative change in employment.
  - Credit contracted for more than half of the firms in the sample.

### Data scope and representativeness (Veneto case study)
- Veneto region: population of 4.9 million individuals and a workforce of 2.2 million workers.
- Veneto accounts for roughly 9 percent of Italian value added and of total employment.
- About 97 percent of workers resident in the region have their workplace in a municipality within the region; migration to other regions is 0.4 percent of the population per year.
- Firm size distribution in Veneto:
  - 94 percent of firms have less than 10 employees.
  - 57 percent of firms have at most one employee.
- Sectoral employment shares: service sector accounts for 56 percent and industrial sector for 43 percent of total employment.
- Banking system in Veneto (2012): about 120 banks; small local banks account for nearly 20 percent of business loans.
- Aggregate lending to non-financial corporations followed a similar dynamic in Veneto and Italy.
- Italy-specific context noted: Italian firms rely more on bank credit than most of the Euro area; small firms (less than 10 employees) are the most indebted.

### Data sources and matching
- Primary administrative sources:
  - PLANET: daily labor market flows from regional public employment service (Veneto Lavoro). Includes hires, firings, transformations of employment arrangements; worker-level variables include gender, age, nationality, occupation (5-digit ISCO), contract type (44 categories), educational attainment (13 categories), time schedule, and reasons for separation.
  - ASIA: archive of active firms from ISTAT; provides yearly firm-level stock data for firms active at least six months in a calendar year; used to reconstruct quarterly stock data.
  - Credit Register (CR) database (Bank of Italy): credit extended to each firm each quarter; monthly reporting of loan amounts for loans exceeding a minimum threshold and all nonperforming loans.
  - Bank of Italy Supervisory Report (SR) database: confidential data on outstanding loans by banks to provinces (used to estimate time-varying bank lending policies).
  - CADS database (Cerved Group S.p.a.): used to match a sub-sample of relatively larger firms for balance-sheet and income statement data (used in Section 5).
- Loan reporting thresholds in CR:
  - 75,000 euro until December 2008.
  - 30,000 euro afterwards.
- Matching is performed using VAT numbers as univocal firm identifiers; final cleaned sample:
  - Nearly 440,000 firms, of which about 200,000 have bank relationships.
  - Sample excludes public sector, primary (agriculture), and financial sector firms; also removes temp agencies, care givers and house cleaners for stated reasons.

### Sample characteristics and labor market structure
- The sample is predominantly micro and small enterprises, consistent with Census data.
- Temporary contracts:
  - Account for more than 10 percent of all contracts.
  - Average duration of temporary contracts is 9.4 months.
  - About two third of temporary contracts end within a quarter.
- Indebted-firm subsample:
  - Average firm has 6.3 employees (median is 2 employees).
  - Two third of firms are in the service sector.
  - Geographical distribution across seven provinces: Padua 20 percent, Verona 19 percent, Venice 16 percent.
  - Most firms borrow from one bank; 29 percent of firms obtain credit from more than one bank.

### Identification strategy and empirical model
- Main regression specification:
  - ∆EMPLOYMENT_it = β ∆LOAN_it + δ_i + (γ_{s(i)} × τ_t) + (η_{c(i)} × τ_t) + (θ_{p(i)} × τ_t) + e_it
  - Changes ∆X_it computed as: ∆X_it = (X_it1 − X_it0) / (0.5×X_it1 + 0.5×X_it0)
  - These variations are symmetric and bounded between −2 (exiters) and +2 (entrants).
- Baseline uses the average change in used loans over two quarters (∆LOAN_it and ∆LOAN_it−1 averaged); robustness includes contemporaneous change and three-quarter averages.
- Fixed effects strategy to control demand and productivity shocks:
  - Firm fixed effects (δ_i) and time (quarter) effects.
  - Industry (2-digit) × quarter fixed effects (γ_{s(i)} × τ_t).
  - Province × quarter fixed effects (θ_{p(i)} × τ_t).
  - Firm class size (micro, small, medium-large) × quarter fixed effects (η_{c(i)} × τ_t).
  - Additional robustness with more granular industry-province-size-quarter fixed effects to absorb borrower-level time-varying demand shocks.
- Endogeneity of bank lending:
  - Bank lending is endogenous to firms’ economic conditions and employment choices.
  - Instrumental variable (IV) approach: construct a time-varying firm-specific credit supply index (CSI_it), following the extension of Greenstone et al. (2014), using banks’ loan shares to a given firm as weights and time-varying nationwide bank lending policies purged of local loan demand.
  - CSI_it is used as an instrument for ∆LOAN_it to isolate credit supply shocks and measure the firm-level ‘aggregate’ bank lending channel (accounting for possible substitution across banks).

### Limitations and methodological considerations
- Wages not observed in the dataset; analysis focuses on quantity responses (employment) rather than price (wage) effects.
- Lack of firm balance-sheet information for the full sample prevents:
  - Controlling for a number of potential drivers of employment decisions.
  - Exploring additional sources of firm-level heterogeneity.
  - Assessing effects of the credit crunch on capital accumulation.
- To address balance-sheet limitations, a sub-sample of larger firms is matched with CADS for additional analysis (Section 5).
- Interest rates largely excluded because:
  - Interest rate data are collected only for a sub-sample of banks (excluding majority of small/local banks), which would reduce sample drastically.
  - Utilized loans are used as they capture both quantity (granted loans) and price effects (through utilization reflecting unobserved price changes).
- Identification relies on the assumptions that:
  - Firm unobserved heterogeneity driving labor demand is time invariant.
  - Firms in the same 2-digit industry, province, and class size face the same demand or productivity shock in each quarter.

*Source: wp1738 - 2.8 percentage points more than firms that were borrowing from healthier lenders, and results (IMF working paper PDF).*

### 3.2    Credit supply index

### 3.2    Credit supply index

### Methodology
- Estimate equation (3): ∆L_bpst = α + δ_bt + γ_pst + e_bpst, where:
  - ∆L_bpst is the percentage change in outstanding business loans by bank b, in province p, in sector s at time t.
  - γ_pst: province-sector-quarter fixed effects capturing local demand.
  - δ_bt: bank-time fixed effects capturing (nationwide) bank lending policies (parameters of interest).
- Identification of γ_pst and δ_bt relies on:
  - Multiple banks in each province-sector market (multiple banks exposed to same demand).
  - Each bank active in multiple province-sector markets (multiple markets exposed to same bank supply).

### Construction of firm-level credit supply index
- Firm-level credit supply index (CSI_it) defined as equation (4):
  - CSI_it = ∑_b w_bit^0 × ˆδ_bt
  - ˆδ_bt are the bank-time fixed effects estimated from equation (3).
  - w_bit^0 is bank b market share for firm i at the beginning of the sample period (end-2007).
- By construction, CSI_it captures time-varying credit supply at firm level driven by:
  - Heterogeneity in changes in business lending across banks.
  - Variation in bank market shares across firms.

### Stylized facts and validation
- Nationwide patterns:
  - Evolution of bank lending policies mimics growth rate of business loans; correlation stronger in first part of crisis and weaker in more recent years.
- Bank heterogeneity:
  - Dividing banks each period into two groups (ˆδ_bt below vs above median): tight banks recorded more negative loan patterns than easy banks.
  - Large contraction in supply around 2009 driven by banks with lowest ˆδ_bt.
- Comparison with other aggregate indicators:
  - (Inverse of) CSI_it closely follows:
    1) Diffusion index from the ECB Bank Lending Survey on Italian banks,
    2) Share of rationed firms from Bank of Italy firm survey,
    3) Corporate credit rationing indicator by Burlon et al. (2016).
  - Correlation of CSI with these three measures varies between 0.5 and 0.6.
- Bank-characteristics correlations (average ˆδ_bt over 2008-2012 regressed on end-2007 bank characteristics):
  - Worsening in credit supply conditions was higher for larger banks, banks with larger funding gap (measured with deposit-to-loan ratio), and banks with lower capital.

### Identification and exogeneity concerns
- Exogeneity relies on two terms: w_bit^0 and ˆδ_bt.
  - For w_bit^0: assumption that bank market shares at firm level, once controlling for firm-fixed effects, are not correlated with firm-level employment trend.
    - Addressed by including industry×quarter and class size×quarter fixed effects; parameter of interest β is stable.
    - Table 3 balancing: average exposure to credit shocks (CSI_it averaged 2008-2012) is not significantly correlated with firm size at beginning-of-period.
  - For ˆδ_bt: bank-time fixed effects are purged of province-sector-quarter factors; unlikely that firm-level unobservables affect nationwide bank lending policies.
- Potential violations and checks:
  - Concern: banks with negative supply shocks might specialize in lending to firms hit harder by crisis (demand confounding) or endogenous sorting of weak banks to weak firms.
    - Empirical evidence: Table 3 shows no systematic correlation between exposure to credit supply shocks and firm characteristics (size, financial dependence, banking relationships, leverage, bad credit history, geographical location, sector).
    - For sub-sample with balance sheet data: instrument not correlated with labor productivity, leverage, and riskiness (proxies for firm quality).
- Additional design choices reinforcing exogeneity:
  1. Drop Veneto provinces when estimating bank lending policies to exclude region-specific demand/supply effects (Veneto excluded from estimation; exclusion affects only one bank accounting for less than 0.1 percent of loans to firms in Veneto).
  2. Translate bank-time fixed effects to firm level rather than aggregate county level.
  3. Control for province-sector-time unobserved factors (more granular than Greenstone et al. (2014)).
  4. Limited government interventions in Italy reduces bias from public support conditionalities.

### Main results: impact on employment
- Visualization: Firms grouped by exposure over 2008-2012 to tighter or easier lending (CSI below or above median) show divergent employment dynamics; firms with less favorable lending conditions exhibit employment decreases.
- 2SLS baseline estimates (Table 4):
  - First-stage: CSI positively associated with change in used loans; first-stage F-statistic ranges between 156 and 180.
  - Second-stage (baseline specification = column 4: firm and quarter fixed effects, time-varying industry, size, and province fixed effects):
    - Coefficient on ∆LOAN = 0.36.
    - Interpretation: a 10 percent contraction in bank lending over two quarters translates into a 3.6 percent reduction in employment.
    - A standard deviation of the predicted change of used loan explains 18% of the standard deviation of employment.
- Aggregate extrapolation:
  - In sample, credit extended to firms diminishes by 1.6 percent while employment by 2.1 percent (both quarterly, see Table 1).
  - Credit drop attributable to lending supply orientation over 2008-2012 explains about 28 percent of the employment loss (simple algebra, noting caveat on general equilibrium effects).
- Comparative magnitudes:
  - Elasticity (0.36) is roughly twice as large as Cingano et al. (2016) on larger Italian firms (different time span); comparable to Bentolila et al. (2016) for Spain and Chodorow-Reich (2014) for U.S., with caveat that current sample includes micro and small firms and covers almost the universe of private non-financial firms and employment of the region.

### Robustness checks
- Saturating fixed effects:
  - Interact quarter dummies with tighter borrower cells: industry×size, industry×province, province×size (columns 1-3, Table 5).
  - Fully saturate model with firm cluster×time (cluster = all firms in same industry, province, and class size) approximating quarterly firm fixed effects (column 4).
  - Coefficient on ∆LOAN remains stable between 0.34 and 0.37 in columns 1-3; four-way fixed effects do not significantly alter magnitude.
- Clustering standard errors at province-industry-class size level (column 5): standard error marginally increases; coefficient remains highly significant.
- Sample restricted to firms with total debt above 75,000 euros throughout sample (column 6): coefficient on ∆LOAN increases to 0.49 and remains precisely estimated.
- Housing net worth channel:
  - Add time-varying house prices at municipality level (column 7); inclusion does not change coefficients on loan variable.
  - Note: high home ownership rates in Italy (76 percent of households own their house in Veneto) and potential collateral use by micro entrepreneurs; but housing boom-and-bust in Italy was limited, more so in Veneto.
- Alternative constructions of ∆LOAN:
  - Consider contemporaneous change at t and average over three quarters (t, t−1, t−2) (columns 7 and 8).
  - Results: contraction in credit supply reduces employment; effect smaller for contemporaneous measure and larger when allowing more lags.
- Instrument relevance robust: first-stage F-statistics between 156 and 180 (well above critical value of 10).

*Source: wp1738 - 3.2 Credit supply index (PDF chapter).*

### 4.3    Job contract heterogeneity

### 4.3    Job contract heterogeneity

### Employment adjustment by contract type
- Employment adjustment happens primarily through variation of temporary contracts (fixed term direct hires, project workers, temporary agency workers, trainees and apprentices, and seasonal workers) rather than open-ended contracts.
- Temporary contracts, while accounting for only slightly more than one tenth of total contracts in the workforce, bear almost half of the effect of the change in credit supply: 0.191/0.363 = 0.53 (where 0.363 is the estimated coefficient of credit supply variation for the entire workforce).
- The coefficient on ∆LOAN is positive and statistically significant for both open-ended and temporary contracts.
- Outflows drive the results: firms more exposed to the credit supply shock have higher outflows; the effect on inflows is marginally significant.
- Within outflows, the adjustment is exclusively due to non-renewal of expired contracts; there is no evidence that adjustment works through dismissal or quit.
- Firms more exposed to negative credit shocks are less likely to convert temporary contracts into open-ended ones.
- Financing constraints do not appear to affect transitions between part-time and full-time jobs.

### Worker heterogeneity (summary of linkages to contracts)
- Firms adjusted labor by reducing temporary contracts and dismissing low- and medium-educated workers.
- High-educated workers were relatively insulated, even when hired with temporary contracts.
- Low-educated workers with temporary contracts account for 32 percent of the total employment effect, despite representing less than 4 percent of the workforce.
- Low-educated workers with open-ended contracts account for 19 percent of the total effect, while representing 33 percent of the workforce.
- High-educated workers with open-ended contracts account for 6 percent of the total effect and are 11 percent of the workforce.
- The effect is not significant for high-educated workers with temporary contracts.

### Interpretation
- Findings are consistent with firms using temporary contracts as a buffer stock to absorb employment volatility and with lower termination costs for temporary contracts.
- Evidence aligns with theories of skill upgrading: in a downturn firms dismiss less skilled, less profitable workers first.

---

### Worker heterogeneity (detailed findings)
- By education (ISCED-based categories):
  - Low-educated workers: elasticity of employment to credit supply equal to 0.48 (=0.186/0.387).
  - Medium-educated workers: elasticity equal to 0.31.
  - High-educated workers: elasticity equal to 0.19 (smaller and only marginally significant).
  - Changes in employment within low-educated workers account for more than half of the total effect of ∆LOAN: 0.186/0.363 = 0.51, even though low-educated workers account for less than 40 percent of the workforce.
- By skill (ISCO/occupation-based classification):
  - Effect concentrated on low-skill occupations, which represent about 15 percent of jobs but account for about 43 percent of the total effect: 0.156/0.363 = 0.43.
- By gender, age, nationality:
  - Female workers represent around 40 percent of total employment but account for 61 percent of the total change in employment.
  - Foreign workers are less than 10 percent of the labor force but their employment dynamics explains more than 24 percent of the total change.
  - Workers under 30 contribute slightly less than a third of the overall employment effect while they represent less than 20 percent of the workforce.

---

### 4.5    Firm heterogeneity

### Size and sectoral heterogeneity
- Results hold only for micro and small firms:
  - Micro firms: less then 10 employees.
  - Small firms: between 10 and 49 employees.
  - Medium-large firms: 50 or more employees — coefficient on ∆LOAN is positive but not statistically significant; first-stage F-statistic suggests weak identification for medium-large firms.
- Employment reacts to credit shocks only in services; no evidence that industrial firms reduce employment in response to a credit crunch.
- Possible explanations: industrial firms use more open-ended contracts and are larger (lower dependence on bank credit).

### Financial characteristics and bank relationships
- Firms that were more indebted at the beginning of the sample (higher debt per employee relative to similar firms in same industry, province, and class size) experience a larger employment reaction to credit supply restrictions.
- Firms using granted credit lines more intensively at the beginning of the sample have higher elasticity of employment to credit (less able to cope using existing lines).
- Bank relationships:
  - Positive and significant elasticities for firms borrowing from one bank and for firms with multiple lending banks.
  - Point estimate for firms with multiple bank relationships is twice as large as that for firms with one bank, suggesting stronger lending relationships mitigate the effect of the credit crunch on employment outcomes.

---

### 5.1    Adjustment at the extensive and intensive margins

### Intensive margin (excluding firm closures)
- Re-estimating the model excluding firms that close in a given quarter yields a smaller but precisely identified elasticity: a 10 percent contraction in credit translates into a 2.5 percent fall in employment.
- The intensive-margin adjustment falls disproportionately on temporary workers: temporary contracts account for about three quarters of the fall in employment in the restricted sample (compared with about 50 percent in the whole sample).
- Results on worker heterogeneity in the restricted sample are qualitatively similar but stronger: active firms that experienced reductions in credit supply reduced employment mostly among low- and medium-educated workers.
- Intensive-margin adjustment has exclusively affected less educated workers with temporary contracts; high-educated temporary workers (about a fifth of temporary contracts) were not hit by the financial shock.

### Extensive margin (firm exit)
- A shortfall in the supply of credit increases the likelihood of firm exit.
- Magnitude: given the average contraction of bank credit of 1.6 percent in the sample period, the estimated coefficient implies a 0.1 percent increase in the probability that a firm closes down, which accounts for about one seventh of the average exit rate.

---

*Source: wp1738 - 4.3    Job contract heterogeneity*

### 5.2    Matching firm balance sheets

### 5.2 Matching firm balance sheets

### Sample and data trade-offs
- Match with CADS (Company Accounts Data Service) yields annual balance-sheet information for a restricted sample of larger firms.
- Trade-offs:
  - Lose coverage of the universe of firms (including small and micro enterprises).
  - Move from quarterly to yearly frequency in empirical analysis.
- Sample size characteristics (restricted sample):
  - Average (median) firm size is 16.4 (4.8) employees, nearly three times the respective values in the whole sample.

### Validation of identification and balancing properties
- Balance-sheet information allows testing exogeneity of the instrument via balancing checks.
- Credit supply index is not systematically correlated with:
  - labor productivity,
  - capital intensity,
  - age,
  - leverage,
  - riskiness.
- Moving to yearly frequency:
  - Coefficient on ∆LOAN in the whole sample: from 0.363 to 0.347 (compare quarterly and yearly; Table 4, column 4, and Table 15, column 1).
  - Coefficient in the restricted CADS sample: 0.095 (significantly smaller).

### Implications of using the restricted (larger-firm) sample
- Large drop in elasticity of employment to credit supply in restricted sample further confirms the importance of analyzing the universe of firms to precisely estimate employment effects.
- If analysis is limited to relatively large firms, the employment effect is likely to be under-estimated.
- Effect on capital accumulation:
  - Negative impact of credit supply on investment, stronger than the effect on labor (consistent with recent evidence cited).

### Heterogeneity: which firms reduced employment more?
- Age (young vs. old firms):
  - Split around median age within each industry, province and class size cluster.
  - Effects on employment are limited to relatively young firms.
  - Interpretation: young firms have higher borrowing needs and are more exposed to a financial shock.

- Risk (Altman z-score at December 2007):
  - z-score ranges from 1 (safest) to 9 (most likely to default); split at the sample median.
  - Ex-ante riskier firms are more sensitive to a credit crunch.
  - Estimated elasticity for high-risk firms is about three times larger than for low-risk firms.

- Leverage (financial debt / (financial debt + equity), median split at December 2007):
  - Role of leverage is much stronger in the restricted sample of larger firms than in the whole sample.
  - Elasticity of employment to the supply of credit is almost four times higher in high-indebted than in low-indebted firms.

- Productivity (labor productivity = value added per employee as of December 2007):
  - A 10 percent reduction in the supply of credit translated into a reduction of employment of 1.8 percent in low-productive firms—twice the effect estimated for the average firms.
  - No statistically significant effect in high-productive firms.
  - Interpretation: consistent with evidence that less productive establishments are more likely to exit during the Great Recession.

### Comparison to literature
- Results align with recent evidence on real effects of large credit contractions (Acharya et al., 2016; Amiti and Weinstein, 2017; Bottero et al., 2016; Cingano et al., 2016; Degryse et al., 2016).
- Comparable findings to Bentolila et al. (2016) for Spanish firms.
- Complements Kalemli-Ozcan et al. (2015) on higher debt overhang leading to larger investment contractions.

### Key quantified findings
- Baseline elasticity of employment to a credit supply shock: 0.36.
- Elasticity excluding firm closures (intensive margin): 0.25.
- Yearly-frequency coefficient in whole sample: 0.347.
- Yearly-frequency coefficient in restricted CADS sample: 0.095.
- High-risk firms’ elasticity ≈ three times that of low-risk firms.
- High-indebted firms’ elasticity ≈ four times that of low-indebted firms.
- A 10 percent reduction in credit supply → 1.8 percent employment reduction in low-productive firms.
- Contraction in bank lending can explain about one fourth of the reduction in employment (aggregate effect).

### Mechanisms and interpretation
- Employment adjustment occurs both via intensive margin reductions and higher probability of firm closure.
- Credit crunch disproportionately affects:
  - smaller firms,
  - younger firms,
  - less productive firms,
  - firms with higher debt overhang,
  - firms with weaker bank-firm relationships.
- Labor-market composition effects:
  - Less educated and less skilled workers with temporary contracts were mainly affected.
  - Firms’ adjustments are consistent with skill upgrading of labor force, moderated by labor market regulation.
- Potential “cleansing” effect:
  - Financial shocks may foster aggregate productivity gains by disproportionately hitting less productive firms and unskilled jobs.

*Source: wp1738 - 5.2 Matching firm balance sheets (PDF chapter).*

### References

### wp1738 - References

### Major bibliographic themes
- Core topics covered by the cited literature:
  - Monetary policy, credit supply, and loan application effects (e.g., "Monetary Policy in a Developing Country; Loan Applications and Real Effects", IMF Working Paper 15/270).
  - Real effects of sovereign and bank distress, including syndicated loans and sovereign-bank-firm linkages (e.g., works on the sovereign debt crisis, Debt Overhang on Europe).
  - Bank lending channel, credit crunches, and firm-level employment responses (multiple studies: Chodorow-Reich 2014; Greenstone, Mas, and Nguyen 2014; Jiménez et al. 2014).
  - Financing constraints, credit rationing, and firm adjustment during recessions (e.g., Buera et al. 2015; Cingano, Manaresi, and Sette 2016; Burlon et al. 2016).
  - Labor-market responses to financial shocks, including contract types, wage rigidity, and heterogeneous impacts by age, education, and skill (e.g., Bentolila et al. 2016; Devicienti et al. 2007; Mian and Sufi 2014).
  - Bank heterogeneity, interest rate setting, and bank-level determinants of lending policy (Gambacorta and Mistrulli 2014; Auer and Ongena 2016).
- Frequent empirical methods and data sources:
  - Matched bank-firm datasets, Bank Lending Survey (BLS), Survey on the Access to Finance of Enterprises (SAFE), Bank of Italy data, ISTAT census and Labour Force Survey, SSRN working papers, and NBER/WP outputs.

### Figures (titles and key notes)
- Figure 1: External validity: firm distribution across size and sectors in Veneto and Italy
  - Notes: elaborations on ISTAT data (census 2011).
  - Panel (a) distribution across firm size classes: categories shown include 0-1, 2, 3-5, 6-9, 10-49, 50-249, > 250.
  - Panel (b) distribution across sectors with percentage axes up to 25.
- Figure 2: External validity: bank penetration and lending in Veneto and Italy
  - Panel (a) Branches per 100,000 inhabitants (locations: Verona, Vicenza, Belluno, Treviso, Venice, Padua, Rovigo, North Italy); vertical scale 50–90.
  - Panel (b) Bank loan growth to nonfinancial corporations (q-o-q) series for 2008:Q1 through 2012:Q3; compares Italy and Veneto.
  - Notes: elaborations on data from Bank of Italy.
- Figure 3: Bank financing in Italy across firm size
  - Panel (a) % of firms using credit for financing (Average 2009-2012, SAFE) comparing Euro area and Italy across all, Micro, SME.
  - Panel (b) Debt per employee (measured in thousands of euro) by size: Micro (<10), Small (10-49), Medium-large (>49).
  - Notes: elaborations on SAFE, Bank of Italy, PLANET, and ASIA.
- Figure 4: Sample representativeness, comparison with the Census
  - Panel (a) Firms’ distribution (%) by size: Micro (<10), Small (10-49), Medium-large (>49).
  - Panel (b) Employees’ distribution (%).
  - Notes: data from ISTAT (2011 census), PLANET, and ASIA.
- Figure 5: Sample representativeness, dynamics of firms and employment
  - Panel (a) Sample, non-financial private sector (deseasonalized data): Number of employees range shown 1,480,000 to 1,560,000; Number of firms range 398,000 to 403,000; quarterly points 2008:Q4 to 2012:Q4.
  - Panel (b) Universe, total economy (Source: ISTAT, RFL): Number of employees range 1,550,000 to 1,700,000; Number of firms range 450,000 to 465,000; quarterly points 2008:Q1 to 2012:Q3.
- Figure 6: Bank lending policies and credit supply index: descriptive statistics
  - Panel (a) plots bank lending policies (vertical range -.04 to 0) and credit growth rate (-.02 to .03) over 2008–2012.
  - Panel (b) credit growth across tight and ease banks (tight/ease definition relative to median).
  - Panel (c) bank lending policies across banks at the 25th, 50th, and 75th percentiles.
  - Panel (d) standardized comparison of: 1) the inverse of the CSI; 2) the Diffusion Index from ECB BLS (range -1 to 1); 3) share of rationed firms (INVIND); 4) corporate credit rationing measure (Burlon et al. 2016).
  - Notes: CSI and bank lending policy estimates follow Greenstone et al. (2014); data from Bank of Italy SR, CR, BLS, and INVIND, ECB, and Burlon et al. (2016).
- Figure 7: Credit supply and employment dynamics
  - Plots averages of residuals from regression of log employees on firm and quarter fixed effects for 2008:Q4–2012:Q4.
  - Compares firms with lower credit supply vs higher credit supply (defined by average CSI over 2008-2012 above/below the median).

### Tables (titles, key statistics and exact numeric values)
- Table 1: Summary statistics
  - Variable, Mean, St. Dev., Share in total employment (%)
  - ∆EMPLOYMENT - Total: Mean -0.0211, St. Dev. 0.2610, Share 100.0
  - ∆EMPLOYMENT - Open-ended: Mean -0.0173, St. Dev. 0.1890, Share 88.7
    - low-education: Mean -0.0080, St. Dev. 0.1000, Share 3.9
    - medium-education: Mean -0.0075, St. Dev. 0.1020, Share 5.4
    - high-education: Mean 0.0016, St. Dev. 0.0355, Share 2.0
  - ∆EMPLOYMENT - Temporary: Mean -0.0039, St. Dev. 0.1660, Share 11.3
    - low-education: Mean -0.0016, St. Dev. 0.1060, Share 32.8
    - medium-education: Mean -0.0017, St. Dev. 0.0993, Share 45.0
    - high-education: Mean 0.0006, St. Dev. 0.0464, Share 11.0
  - ∆EMPLOYMENT - Under 30: Mean -0.0025, St. Dev. 0.1080, Share 17.9
  - ∆EMPLOYMENT - Over 30: Mean -0.0186, St. Dev. 0.2090, Share 82.1
  - ∆EMPLOYMENT - Male: Mean -0.0126, St. Dev. 0.1690, Share 59.9
  - ∆EMPLOYMENT - Female: Mean -0.0085, St. Dev. 0.1470, Share 40.1
  - ∆EMPLOYMENT - Italian: Mean -0.0179, St. Dev. 0.2260, Share 91.4
  - ∆EMPLOYMENT - Foreign: Mean -0.0033, St. Dev. 0.0919, Share 8.6
  - ∆EMPLOYMENT by education and skill:
    - Low-education: Mean -0.0098, St. Dev. 0.1480, Share 38.7
    - Medium-education: Mean -0.0088, St. Dev. 0.1420, Share 48.0
    - High-education: Mean 0.0022, St. Dev. 0.0569, Share 13.3
    - Low-skill: Mean -0.0042, St. Dev. 0.1390, Share 15.2
    - Medium-skill: Mean -0.0113, St. Dev. 0.1500, Share 49.3
    - High-skill: Mean -0.0056, St. Dev. 0.0874, Share 35.5
  - ∆LOAN: Mean -0.0163, St. Dev. 0.3151
  - CSI: Mean -0.0085, St. Dev. 0.0404
  - EXIT: Mean 0.0066, St. Dev. 0.0811
  - Notes: Sample is universe of firms conditional on having bank debt; contract and education definitions provided.
- Table 2: Credit supply and bank heterogeneity (OLS cross-section, dependent variable: CS I at the bank level (average 2008-2012))
  - Coefficients (standard errors in parentheses)
  - Bank size: -0.0047*** (0.0009); alternative specification -0.0014** (0.0007)
  - Funding gap: -0.0102*** (0.0018); alternative -0.0085*** (0.0017)
  - Tier 1 capital ratio: 0.0470*** (0.0073); alternative 0.0383*** (0.0074)
  - Share of NPLs: -0.0237 (0.0388); alternative -0.0494 (0.0365)
  - Observations: 536 across specifications
  - R-squared values: 0.084, 0.113, 0.148, 0.001, 0.240 (presented across five columns)
  - Notes: bank-level variables measured as of end-2007; robust standard errors; significance levels: *** p<0.01, ** p<0.05, * p<0.1.
- Table 3: Orthogonality conditions (average firm-specific variables by quintile of CS I)
  - Credit supply index (CS I) by quintile: -0.040, -0.018, -0.007, 0.000, 0.022; correlation with credit supply (CS I) = 1,000
  - % industry: 0.323, 0.329, 0.342, 0.278, 0.298; correlation -0.024
  - % services: 0.677, 0.671, 0.658, 0.722, 0.702; correlation 0.024
  - # employees: 4.578, 7.728, 9.308, 4.825, 3.919; correlation -0.007
  - % main province: 0.236, 0.238, 0.222, 0.173, 0.144; correlation -0.055
  - Debt per employee: 128,460; 175,598; 164,244; 165,833; 114,852; correlation -0.003
  - Utilized/granted credit: 0.194, 0.311, 0.396, 0.237, 0.194; correlation 0.003
  - Multi-banks: 0.870, 0.855, 0.823, 0.987, 0.884; correlation -0.017
  - NPLs: 0.035, 0.032, 0.048, 0.044, 0.041; correlation 0.014
  - Notes: Definitions for % industry, % services, Utilized/granted credit, Multi-banks, NPLs; CS I definition in Section 3.2 and equation 4.
- Table 4: Baseline regressions – IV estimates (2SLS results for equation 1)
  - First-stage (Dep var: ∆LOANt,t−1)
    - CS It,t−1 coefficients across four specifications: 0.0786*** (0.00609), 0.0779*** (0.00609), 0.0758*** (0.00609), 0.0751*** (0.00622)
    - R-squared: 0.160, 0.161, 0.162, 0.162
  - Second-stage (Dep Var: ∆EMPLOYMENTt)
    - ∆LOANt,t−1 coefficients: 0.437*** (0.0672), 0.438*** (0.0676), 0.445*** (0.0699), 0.363*** (0.0689)
    - Observations: 2,459,948 across all specifications
    - 1st-stage F-statistic: 179.6, 176.6, 166.9, 156.3
    - Fixed effects: Firm FE Yes in all; Quarter FE Yes in first, others vary; Industry×quarter FE and Size×quarter FE and Province×quarter FE included progressively across specifications as listed.
  - Notes: ∆EMPLOYMENTt and ∆LOAN variables bounded between −2 and +2; fixed effects constructed on 30 industries, 7 provinces, 3 class sizes.
- Table 5: Robustness exercises
  - Provides (second-stage) 2SLS regression results variations for dependent variable ∆EMPLOYMENTt and alternative loan change definitions:
    - ∆LOANt,t−1: average change in used loans in quarters t and t−1
    - ∆LOANt: change in used loans in quarter t
    - ∆LOANt,t−2: average change in used loans in quarters t, t−1 and t−2
  - (Table content truncated in source excerpt; table documents robustness setups and alternative instrument/configuration checks.)

*Italic: Source — wp1738 - References (PDF chapter/section).*

### 2.  All these variables are calculated as in equation 2, so that they are bounded between

### wp1738 - 2.  All these variables are calculated as in equation 2, so that they are bounded between

### Key definitions and estimation setup
- Dependent variable: ∆EMPLOYMENT_t defined as the change in employment at the firm level over the quarter t, divided by the average stock of all firm’s workers over the quarter. Not a growth rate but a contribution to aggregate (firm-level) growth; coefficients need scaling by relative share of job contracts/workers.
- Main regressor: ∆LOAN_t,t−1 = average change in used loans at the firm level in quarters t and t−1 (or other quarter windows where noted).
- Excluded instrument (first stage): credit supply index CSI (see Section 3.2 and equation 4) with CSI defined analogously to ∆LOAN and calculated over specified quarter windows.
- ∆EMPLOYMENT_t and ∆LOAN_t,t−1 (when explicitly stated) are calculated as in equation 2 and are bounded between −2 and +2.
- Fixed effects: Firm FE always included; additional fixed effects constructed based on 30 (2-digit) industries, 7 provinces, and 3 class sizes (firms with less than 10 employees, between 10 and 49, and 50 or more). Time dimensions vary (quarter or year) by table. Robust standard errors reported; clustering where specified.
- Significance annotation: *** p<0.01, ** p<0.05, * p<0.1.

### Table 6 — Job contract heterogeneity (selected estimates and diagnostics)
- Main specification (various columns): ∆LOAN_t,t−1 coefficients (selected):
  - 0.373*** (standard error 0.0691)
  - 0.337*** (0.0675)
  - 0.361*** (0.0687)
  - 0.345*** (0.0681)
  - 0.363*** (0.108)
  - 0.491*** (0.100)
  - 0.363*** (0.0689)
- HOUSE PRICE_t coefficient: −0.0341** (0.0163)
- Additional loan timing:
  - ∆LOAN_t coefficient: 0.229*** (0.0510)
  - ∆LOAN_t,t−2 coefficient: 0.620*** (0.0976)
- Observations: 2,459,948 (except column with Drop < 75k: 1,863,812)
- 1st-stage F-statistics (selected): 157.1, 156.3, 157.0, 154.9, 169.7, 118.9, 156.3, 108.2, 142.8
- Standard errors: Robust in most columns; column (5) clustered at province-industry-class size level.

### Table 6 (panel): Job contract heterogeneity — contract types, exits, flows, transitions
- Contracts / Reasons for exit — ∆LOAN_t,t−1 coefficients:
  - Open-ended: 0.170*** (0.0449)
  - Temporary: 0.191*** (0.0471)
  - Dismissal: −0.0357 (0.0222)
  - Expiry: −0.158*** (0.0330)
  - Quit: −0.0334 (0.0283)
- Flows / Transitions — ∆LOAN_t,t−1 coefficients:
  - Inflows: 0.0834* (0.0447)
  - Outflows: −0.277*** (0.0617)
  - Fixed to open: 0.0219** (0.00864)
  - Full to part-time: −0.00155 (0.00651)
  - Part-time to full: −0.00414 (0.00590)
- Observations: 2,459,948
- 1st-stage F-statistic: 156.3

### Table 7 — Worker heterogeneity by education and skills
- Education level — ∆LOAN_t,t−1 coefficients:
  - Low: 0.186*** (0.0403)
  - Medium: 0.151*** (0.0371)
  - High: 0.0261* (0.0144)
- Skill level — ∆LOAN_t,t−1 coefficients:
  - Low: 0.156*** (0.0397)
  - Medium: 0.135*** (0.0375)
  - High: 0.0725*** (0.0215)
- Observations: 2,459,948 (each)
- 1st-stage F-statistic: 156.3

### Table 8 — Worker heterogeneity by personal characteristics
- Gender / Age / Nationality — ∆LOAN_t,t−1 coefficients:
  - Male: 0.144*** (0.0421)
  - Female: 0.220*** (0.0419)
  - Under 30: 0.105*** (0.0295)
  - Over 30: 0.257*** (0.0539)
  - Italian: 0.274*** (0.0586)
  - Foreign: 0.0885*** (0.0245)
- Observations: 2,459,948 (each)
- 1st-stage F-statistic: 156.3

### Table 9 — Interaction of contract type and education
- Open-ended contracts — ∆LOAN_t,t−1 by education:
  - Low: 0.0683*** (0.0248)
  - Medium: 0.0816*** (0.0247)
  - High: 0.0214*** (0.00821)
- Temporary contracts — ∆LOAN_t,t−1 by education:
  - Low: 0.115*** (0.0301)
  - Medium: 0.0716*** (0.0275)
  - High: 0.00420 (0.0121)
- Observations: 2,459,948 (each)
- 1st-stage F-statistic: 156.3

### Table 10 — Firm heterogeneity by size and sector
- Firm size (∆LOAN_t,t−1 coefficients):
  - Micro: 0.327*** (0.0719)
  - Small: 0.618*** (0.226)
  - Med-Large: 3.7370 (7.980) — note large point estimate with large standard error
- Sector:
  - Industry: 0.1670 (0.135)
  - Services: 0.427*** (0.0803)
- Observations:
  - Micro: 2,086,193
  - Small: 333,629
  - Med-Large: 40,126
  - Industry: 818,609
  - Services: 1,641,339
- 1st-stage F-statistics (selected): 133.5, 25.1, 30.3, 34.7, 122.6

### Table 11 — Firm heterogeneity by financial characteristics
- Debt per employee — ∆LOAN_t,t−1:
  - Low debt: 0.269*** (0.0682)
  - High debt: 0.498*** (0.145)
- Credit lines use (utilization) — ∆LOAN_t,t−1:
  - Low utilization: 0.276*** (0.0760)
  - High utilization: 0.460*** (0.129)
- Relationship lending — ∆LOAN_t,t−1:
  - Single-bank: 0.369*** (0.114)
  - Multi-banks: 0.773*** (0.260)
- Observations:
  - Debt: 1,207,961 (Low), 1,251,987 (High)
  - Credit lines: 1,222,596 (Low), 1,237,352 (High)
  - Relationship: 1,393,068 (Single-bank), 1,013,409 (Multi-banks)
- 1st-stage F-statistics (selected): 100.78, 81.05, 89.66, 72.19, 59.97, 25.55

### Table 12 — Adjustment at the intensive and extensive margins
- Dependent variables: ∆EMPLOYMENT_t (annual or quarterly variants), and EXIT_t dichotomous (firm closed in quarter t).
- Main estimates — ∆LOAN_t,t−1 coefficients:
  - Full sample ∆EMPLOYMENT_t: 0.363*** (0.0689)
  - Excluding firm closures (All contracts): 0.252*** (0.0572)
  - Excluding closures (Open-ended): 0.0653** (0.0314)
  - Excluding closures (Temporary): 0.185*** (0.0472)
  - EX I T_t (Full sample): −0.0591*** (0.0184)
- Observations:
  - Full sample ∆EMPLOYMENT_t and EXIT_t: 2,459,948
  - Excluding firm closures variants: 2,443,651
- 1st-stage F-statistics: 156.3 (or 156.0 for excluding-closures subsample)

### Table 13 — Contract type and education (intensive margin, excluding closures)
- All contracts — ∆LOAN_t,t−1 by education:
  - Low: 0.141*** (0.0373)
  - Medium: 0.100*** (0.0325)
  - High: 0.0108 (0.0136)
- Open-ended contracts — ∆LOAN_t,t−1 by education:
  - Low: 0.0261 (0.0220)
  - Medium: 0.0307* (0.0186)
  - High: 0.00897 (0.00717)
- Temporary contracts — ∆LOAN_t,t−1 by education:
  - Low: 0.114*** (0.0303)
  - Medium: 0.0694** (0.0276)
  - High: 0.00181 (0.0121)
- Observations: 2,443,651 (each)
- 1st-stage F-statistic: 156.0

### Table 14 — Orthogonality conditions, balance sheet characteristics (by CSI quintile)
- Credit supply index (CSI) by quintile: −0.037, −0.017, −0.008, −0.001, 0.018
- Selected averages by quintile (rows left-to-right across quintiles 1 to 5):
  - Labor productivity: 51.45, 55.05, 53.74, 46.44, 49.3; correlation with CSI: −0.007
  - Capital per worker: 303.1, 284.7, 386.3, 352.4, 287.6; correlation with CSI: 0.000
  - Age: 10.300, 12.300, 12.500, 10.400, 9.350; correlation with CSI: −0.036
  - Riskiness (Altman z-score): 5.407, 5.253, 5.260, 5.176, 5.319; correlation with CSI: −0.026
  - Leverage: 0.669, 0.661, 0.663, 0.632, 0.640; correlation with CSI: −0.034

### Table 15 — Effects of financial shocks on employment and capital accumulation (annual)
- Dependent variables: ∆EMPLOYMENT_t (annual) and ∆CAPITAL_t (change in stock of assets, annual).
- Main coefficients — ∆LOAN_t,t−1:
  - Column (1) All firms ∆EMPLOYMENT_t: 0.347*** (0.0504); Observations: 715,920; 1st-stage F-statistic: 421.87
  - Column (2) CADS sample ∆EMPLOYMENT_t: 0.0952** (0.0459); Observations: 169,881; 1st-stage F-statistic: 1.267
  - Column (3) CADS sample ∆CAPITAL_t: 1.210*** (0.315); Observations: 169,881; 1st-stage F-statistic: 1.26
- Fixed effects: Firm FE and industry×year, size×year, province×year FE included.

### Table 16 — Adding balance sheet characteristics (introductory text)
- Table 16 reports 2SLS (second-stage) estimates based on the CADS-matched sub-sample with dependent variable ∆EMPLOYMENT_t defined as the change in employment at the firm level over the year t; main regressor ∆LOAN_t,t−1 defined analogously for year t−1. (Full numeric results and coefficients continue in the source beyond the provided excerpt.)

*Source: wp1738 (IMF working paper). PDF: wp1738 - 2.  All these variables are calculated as in equation 2, so that they are bounded between −2 and +2.*

### 1. Both variables are calculated as in equation 2, so that they are bounded between

### wp1738 - 1. Both variables are calculated as in equation 2, so that they are bounded between

### Regression results and interpretation
- Dependent variable: ∆EMPLOYMENT_t, defined as the change in employment at the firm level over quarter t divided by the average stock of the firm’s workers over the quarter (contribution to aggregate firm-level growth).
- Instrument in first-stage regressions: credit supply index CS I (as defined in Section 3.2 and equation 4).
- Sample splits: low and high, computed along the median value of each variable within each industry-province-size cluster, measured at December2007.
- Fixed effects included: firm FE; Industry × quarter FE; Size × quarter FE; Province × quarter FE. Fixed effects are constructed based on 30 (2-digit) industries, 7 provinces and 3 class sizes (firms with less than 10 employees, between 10 and 49, and 50 or more).
- Standard errors: robust standard errors in parentheses. Significance markers: *** p<0.01, ** p<0.05, * p<0.1. In column (5) standard errors are clustered at the province-industry-class size level.
- Key coefficient estimates for ∆LOAN_t,t−1 across sub-samples (columns (1)–(8)):
  - Column (1) Young (Age): 0.247** (0.110)
  - Column (2) Old (Age): -0.00888 (0.0464)
  - Column (3) Low (Riskiness): 0.0445 (0.0462)
  - Column (4) High (Riskiness): 0.179* (0.101)
  - Column (5) Low (Leverage): 0.0293 (0.0487)
  - Column (6) High (Leverage): 0.195** (0.0897)
  - Column (7) Low (Labor productivity): 0.177** (0.0690)
  - Column (8) High (Labor productivity): -0.0162 (0.0681)
- Observations and instrument strength:
  - Observations by column: 84,239; 85,642; 84,899; 84,982; 84,936; 84,945; 84,937; 84,944
  - 1st-stage F-statistic by column: 20.80; 49.23; 44.81; 26.96; 37.72; 38.44; 34.05; 33.16

### Data sources and sample construction
- Main data sources:
  - PLANET: administrative dataset of daily labor market flows for the Italian region of Veneto (Veneto Lavoro).
  - ASIA: firm register maintained by ISTAT (covers private sector firms active more than 6 months per calendar year; agriculture excluded).
  - Credit Register (CR): Bank of Italy information system with monthly reporting by banks/financial companies on loans exceeding minimum thresholds (75,000 euro until December 2008, 30,000 euro afterwards), plus all nonperforming loans.
- Merge and sample filtering:
  - Unique identifier used for merging: fiscal code.
  - Excluded: private financial and primary sector firms, temp agencies, care givers and house cleaners.
  - Data cleansing removed: (i) firms that closed before January 1st 2008, (ii) records with missing event date, (iii) records where end of contract is prior to start, (iv) records where start is after firm closure date in ASIA.
  - Final sample after filters: 436,311 firms (about 10 percent loss from original sample).
  - Firms matched with Credit Register: 204,301 (have credit relationships with the banking system).

### Variable creation and reconstruction of employment stocks
- Two main additions to PLANET:
  - Quarterly indicator whether a firm is alive (constructed using ASIA start/end dates with specific imputation rules).
  - Quarterly reconstruction of the stock of workers combining ASIA year-end stocks and PLANET flows.
- Imputation rules for firm start date (if missing in ASIA):
  - placed at the last quarter of the year before the firm is first observed in ASIA with positive size, or
  - at the quarter before the first movement observed in PLANET,
  - whichever comes first.
- Imputation rules for firm end date (if missing in ASIA):
  - placed at the first quarter of the year after the firm is last observed in ASIA with positive size, or
  - at the quarter of the last movement observed in PLANET, provided the variation is negative and cumulative employment variation is also negative,
  - whichever comes last.
- Construction of quarterly employment stocks:
  1. Firms observed in PLANET at the beginning of the period (2008/I): start from ASIA initial (2007) stock, add estimate of initial stock of agency workers and independent contractors (based on PLANET contract termination/renewal/transformation information), then add quarter-specific employment variations observed in PLANET.
  2. New firms first observed in PLANET after 2008/I: first observed stock = first flow augmented by one unit to include the entrepreneur; subsequent stocks obtained by adding quarter-specific observed flows.
  3. Firms observed only in ASIA: stocks remain equal to the initial stock observed in ASIA (no flows during observed period).
- Note on PLANET coverage: difference between PLANET and official labor force statistics is less than 7 percent in 2012 for employed workers in the non-financial private sector (difference due to employees with no observed transitions in 2008-2012).
- Note on Credit Register threshold: minimum threshold for monthly reporting changed from 75,000 euro until December 2008 to 30,000 euro afterwards.

### Additional tables and figures (high-level contents)
- Table A1: 2SLS second-stage results for ∆EMPLOYMENT_t by contract type (open-ended vs temporary) across sectors and firm size. Key numbers (examples):
  - Open-ended contract: coefficients reported include 0.124, 0.185***, 0.136***, 0.394**, 2.346 with standard errors (0.106)(0.0478)(0.0467)(0.153)(5.046). Observations: 818,609; 1,641,339; 2,086,193; 333,629; 40,126. 1st-stage F-statistic examples: 34.76, 122.61, 33.52, 5.13, 0.32.
  - Temporary contract: coefficients reported include 0.0444, 0.239***, 0.189***, 0.225*, 1.392 with standard errors (0.0769)(0.0581)(0.0504)(0.130)(3.036).
- Table A2: 2SLS second-stage results for ∆EMPLOYMENT_t by education level (low, medium, high) across sectors and firm size. Key numbers (examples):
  - Low education: coefficients 0.106, 0.213***, 0.166***, 0.323**, 2.091 with standard errors (0.0874)(0.0448)(0.0423)(0.127)(4.458).
  - Medium education: coefficients 0.0490, 0.184***, 0.138***, 0.237**, 1.255 with standard errors (0.0679)(0.0445)(0.0398)(0.101)(2.710).
  - High education: coefficients 0.00843, 0.0313*, 0.0229, 0.0514, 0.334 with standard errors (0.0222)(0.0180)(0.0156)(0.0349)(0.743).
- Figure A1: Regional bank lending policies — Diffusion Index plotted over 2008 II to 2015 II for North West, North East, Centre, South. Notes: Diffusion index varies between -1 and 1; computed as weighted mean of answers to Regional Bank Lending Survey question 1 with values 1, 0.5, 0, -0.5, -1 for answers (i)-(v) and weights equal to observed frequencies.
- Figure A2: Housing prices in Veneto and Italy, 2000–2013 — Index number (2005=100) plotted from 2000 to 2013 for Veneto and Italy.

*Source: wp1738 - 1. Both variables are calculated as in equation 2, so that they are bounded between (IMF working paper, content unit provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp1738.pdf_