## Supply and Demand Effects of Unemployment Insurance Benefit Extensions: Evidence from U.S. Counties

## Source details

**Canonical URL:** [Supply and Demand Effects of Unemployment Insurance Benefit Extensions: Evidence from U.S. Counties](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021070-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021070-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021070-print-pdf.pdf.json)

---

### Abstract and central findings
- Uses three decades of county-level data to estimate effects of federal unemployment benefit extensions on economic activity; identification via differences-in-differences (DiD) exploiting within-state cross-county exposure variation.
- Major empirical findings:
  - Overall impact of benefit extensions on activity is positive, pointing to strong demand effects.
  - No negative supply-side effects in tradable sectors; positive effects are driven by non-tradable sectors.
  - Fiscal multiplier estimate: 1.92.
  - Per additional dollar in UI benefits from federal extension programs, a county’s private sector labor earnings increase by 1.925 dollars.
- JEL Classification Numbers: E62; J65; J68
- Keywords: Automatic stabilizers; Fiscal multiplier; Labor markets

### Research design, data, and identification
- Data and policy variation:
  - County-level panel spanning more than three decades.
  - Policy episodes: EB program (permanent) and discretionary federal extensions: 1991-94 Emergency Unemployment Compensation, 2002-04 Temporary Extended Unemployment Compensation, 2008-13 Emergency Unemployment Compensation.
- Treatment construction:
  - Bartik-style interaction: state-level de-jure policy changes (Z_{s(c);t}) × county-level lagged historical average UI benefit receipts θ_{c;t-5} (15-year average, lagged 5 years).
  - θ_{c;t-L} formula: (1/15) * Σ_{j=1}^{15} (benefits_{c;t-L-j} / personal income_{c;t-L-j}) / (benefits_{s(c);t-L-j} / personal income_{s(c);t-L-j}).
  - Z_{s(c);t} defined as year-on-year change in maximum aggregate dollar amount unemployed households in state s(c) can receive (formula reported in source).
- Identification and controls:
  - Primary variation from interaction θ_{c;t-5} · Z_{s(c);t}; full set of state-by-year fixed effects μ_{s(c);t} included.
  - Controls for other transfers and state effective individual income tax rate changes (θ_{c;t-5} · ΔT_{s(c);t}).
  - Standard errors clustered at the state level; regressions weighted by average county population.
- Measurement notes:
  - Z_{s(c);t} correlates with actual benefit growth at 69 percent; regression of benefit growth on Z_{s(c);t} yields R-squared = 0.48.
  - θ_{c;t} accounts for one-third of within-state variation in unemployment rates (within R-squared = 0.347 at five-year lag).

### Reduced-form results (employment, participation, earnings, prices)
- Aggregate effects:
  - Estimated effects on employment, labor force participation, and labor earnings are positive and statistically significant; demand boost offsets labor market disincentives.
- Key coefficients (Table 4, coefficient on θ_{c,t-5} Z_{s(c),t}; standard errors in parentheses):
  - ∆employment: 0.269*** (0.048)
  - ∆labor force: 0.439*** (0.057)
  - ∆population: 0.027* (0.015)
  - ∆wages: 0.009 (0.045)
  - ∆earnings: 0.166*** (0.039)
  - ∆home values: 0.222** (0.085)
- Magnitude and counterfactuals:
  - Baseline reduced-form: a 1-percentage point increase in the de-jure benefit-to-income ratio → 0.275 percent increase in employment (Table 2, column (i)).
  - An extension of maximum benefit duration by three months increases the average county’s aggregate benefit receipts by less than 0.02 percent of annual household income.
  - Counterfactual: nationwide employment in 2014 would have been higher by about 560,000 jobs if the Emergency Unemployment Compensation program had not expired on January 1st of that year (partial-equilibrium estimate).
- Labor force and migration:
  - Effect on labor force participation larger than effect on employment; migration effects small and less statistically significant.
- Real wages:
  - No effect found on average real wage growth (coefficient on ∆wages = 0.009 (0.045)).
- Housing:
  - Positive and highly significant effect on residential real estate prices (Zillow Home Value Index).

### Sectoral results and interpretation (supply vs demand)
- Tradable vs non-tradable:
  - Classification: tradable industries = top quartile Herfindahl index; non-tradable = bottom quartile.
  - Key coefficients (Table 5, θ_{c,t-5} Z_{s(c),t}):
    - all sectors: 0.325*** (0.081)
    - tradable: 0.109 (0.452) — not statistically significant
    - non-tradable: 0.277** (0.104)
- Sector-specific findings (selected Table 6 estimates on θ_{c,t-5} Z_{s(c),t}, standard errors in parentheses):
  - Manufacturing: 0.957*** (0.287)
  - Retail trade: 0.715*** (0.109)
  - Hotels etc: 0.896* (0.510)
  - Construction and retail trade show highly statistically significant positive effects; construction result consistent with housing-price stabilization channel.
- Interpretation:
  - Positive effects concentrated in non-tradable sectors, consistent with Keynesian demand channel rather than supply-side disincentives.
  - DiD design with state-by-year fixed effects filters out national/foreign demand for tradables, implying local estimates may understate aggregate demand spillovers through tradables.

### Fiscal multiplier estimation (IV approach) and mechanisms
- IV specification:
  - First-stage: ∆benefits_{c;t} / income_{c;t-1} = δ · θ_{c;t-5} · Z_{s(c);t} + χ_c + κ_{s(c);t} + ε_{c;t}
  - Second-stage: ∆employed_{c;t} = γ · (∆benefits_{c;t} / income_{c;t-1}) + ψ_c + η_{s(c);t} + ζ_{c;t}
- First-stage and IV results (Table 7):
  - First-stage coefficient on θ_{c,t-5} Z_{s(c),t}: 0.047*** (0.004)
  - First-stage: a one-dollar increase in maximum benefits → increase in actual benefits of less than 5 cents.
  - First-stage F-statistic = 158.3.
  - Second-stage IV: increasing benefits by one percent point of total household income → 2.451 percent more employment.
  - Per-dollar multiplier: for every dollar paid in additional UI benefits, private sector labor earnings increase by 1.925 dollars.
  - Primary reported fiscal multiplier: 1.92.
- Mechanisms consistent with large multiplier:
  - High marginal propensity to consume among unemployed workers.
  - Reduced precautionary savings among employed workers due to a more generous safety net.
  - Housing-market stabilization and wealth effects (consistent with strong construction and home-value effects).
- IV interpretation caveats:
  - γ captures overall per-dollar impact of the benefit-extension policy, beyond narrow cash-transfer effect (includes insurance effects reducing mortgage defaults, MPC heterogeneity, etc.).
  - OLS shows strong negative relationship between benefit payments and activity absent instrumentation, motivating IV approach.

### Robustness, controls, and limitations
- Controls and robustness:
  - Extensive controls for other government transfers and sectoral government wages; controls for changes in effective state individual income tax rates.
  - θ_{c;t-L} and controls winsorized at top and bottom 2.5th percentiles; dependent-variable outliers dropped.
  - Alternative weighting (Annex B, Table 8) yields similar coefficients for θ_{c,t-5} Z_{s(c),t} across weighting by population, equal, and by p area.
- Limitations and interpretation nuances:
  - α (direct Z effect at θ=0) cannot be separately estimated due to collinearity with state-by-year fixed effects.
  - State-by-year fixed effects may absorb concurrent policies, possibly understating total demand effects that cross state lines.
  - DiD does not fully disentangle channels; sectoral patterns used to infer demand vs supply dominance.
  - Local estimates likely understate aggregate effects because demand for tradable goods purchased across counties and states is differenced out.

### Policy implications and conclusion
- Overall assessment:
  - Temporary extensions of UI benefit duration have, on average over the past three decades, helped stimulate economic activity.
  - Local fiscal multipliers for UI benefits are as large as those for other government spending measures.
  - Labor-market disincentive effects appear negligible relative to positive employment effects from stimulated aggregate demand.
- Policy recommendations:
  - UI benefit extensions are an attractive stabilization tool given estimated multiplier magnitudes.
  - Discretionary nature of extensions creates policy uncertainty and implementation lags.
  - Reform suggestion: reintroduce a national trigger for EB program extensions (as existed until 1981) to automate benefit extensions in recessions, reduce policy uncertainty, and strengthen automatic stabilizer role.

*Source: IMF Working Paper WP/21/70, “Supply and Demand Effects of Unemployment Insurance Benefit Extensions: Evidence from U.S. Counties” by Klaus-Peter Hellwig (March 2021), Sections 1–4.*

### Section 1

### Supply and Demand Effects of Unemployment Insurance Benefit Extensions: Evidence from U.S. Counties

### Abstract and central findings
- Uses three decades of county-level data to estimate effects of federal unemployment benefit extensions on economic activity.
- Identification: differences-in-differences (DiD) approach exploiting within-state variation in county exposure to state-level policy changes.
- Major empirical findings:
  - Overall impact of benefit extensions on activity is positive, pointing to strong demand effects.
  - No negative supply-side effects in tradable sectors; positive effects are driven by non-tradable sectors.
  - Fiscal multiplier estimate: 1.92.
  - Per additional dollar in UI benefits from federal extension programs, a county’s private sector labor earnings increase by 1.925 dollars.
- JEL Classification Numbers: E62; J65; J68
- Keywords: Automatic stabilizers; Fiscal multiplier; Labor markets

### Research question and motivation
- Core question: Do more generous UI benefits stabilize the business cycle through demand effects, or do they create supply-side distortions via reduced job search and labor supply?
- Context: Renewed policy debate following the COVID-19 shock in 2020; contrasting public statements from Senators Rick Scott and Dick Durbin highlight policy trade-offs.
- Empirical focus: Whether federal extensions of UI benefit duration explain within-state, cross-county variation in economic outcomes.

### Data, policy variation, and identification strategy
- Data: County-level panel spanning more than three decades.
- Policy episodes studied: permanent Extended Benefits (EB) program and three discretionary federal extension programs:
  - 1991-94 Emergency Unemployment Compensation program
  - 2002-04 Temporary Extended Unemployment Compensations program
  - 2008-13 Emergency Unemployment Program
- Treatment construction:
  - Bartik-style interaction: state-level policy changes (de-jure benefit generosity) × county-level lagged historical average UI benefit receipts (as a share of income).
  - Treatment intensity is continuous; counties with higher historical benefit receipts are more exposed to state-wide extensions.
  - Long lags used so local variation in treatment intensity reflects structural factors (workforce or industry composition) that are persistent over time.
- Controls: full set of state-by-year fixed effects to control for drivers of changes in benefit extensions.
- Strategy to distinguish channels: estimate models separately for tradable and non-tradable sectors to isolate supply versus demand effects.

### Reduced-form results (employment, participation, earnings)
- Aggregate outcomes:
  - Estimated effects of UI benefit extensions on employment, labor force participation, and labor earnings are positive and statistically significant.
  - The demand boost from additional benefit weeks more than offsets any labor market disincentives.
- Sectoral disaggregation:
  - Non-tradable sectors (e.g., retail trade, construction) drive the positive effects.
  - Tradable sectors: no significant treatment effect detected.
  - No effect found on average wages.
- Magnitude and interpretive notes:
  - An extension of maximum benefit duration by three months increases the average county’s aggregate benefit receipts by less than 0.02 percent of annual household income.
  - Back-of-the-envelope counterfactual: total nationwide employment in 2014 would have been higher by half a million million jobs if the Emergency Unemployment Compensation program had not expired on January 1st of that year. (Estimate presented as partial-equilibrium; may omit general equilibrium spillovers.)
  - Local estimates likely understate full demand effects because spillovers through tradable goods across counties and states are differenced out by state-time fixed effects.

### Fiscal multiplier estimation (IV approach)
- Instrumental strategy: use Bartik-type treatment variables (de-jure benefit generosity) as instruments for actual benefit transfer payments (de-facto generosity) to estimate local fiscal multipliers.
- Key multiplier estimates:
  - Primary reported fiscal multiplier: 1.92.
  - Complementary statement: per additional dollar in UI benefits from federal extension programs, a county’s private sector labor earnings increase by 1.925 dollars.
- Comparison to literature: multiplier magnitude similar to local fiscal multiplier estimates for other government spending and to general-equilibrium transfer multipliers cited in the literature.

### Mechanisms and interpretation
- Three candidate channels for large multiplier:
  - High marginal propensity to consume among unemployed workers, producing large Keynesian demand effects.
  - More generous safety net reduces precautionary savings among employed workers.
  - UI benefits help stabilize housing prices, producing a wealth effect that raises demand; consistent with strong effects in construction.
- No evidence that supply-side disincentives dominate:
  - If disincentives were strong, employment in tradable sectors would decline; results do not support this.
- Spillovers:
  - Local estimates exclude demand for tradable goods purchased across county lines, which likely causes local estimates to understate aggregate effects.

### Contribution to literature
- Provides the first empirical county-level estimate comparing supply and demand effects of UI benefit duration extensions.
- Adds to literature on local fiscal multipliers and on labor market implications of UI generosity.
- Reconciles micro studies with macro outcomes by showing demand effects can offset search/disincentive effects at the local aggregate level.

### Background on federal UI extension programs and state UI system
- State UI system: joint federal-state program established in 1935; largest public UI scheme in the US.
- Program variation:
  - Extended Benefits (EB) program, created in 1970, provides an additional 13 weeks of benefits in states with high unemployment, financed equally by state and federal funds.
  - Discretionary federal extensions enacted in each nation-wide recession since 1958.
- State-level heterogeneity: eligibility rules, benefit formulae, and weekly benefit caps vary across states; state UI funds can borrow from the federal government when needed.

*Source: IMF Working Paper WP/21/70, “Supply and Demand Effects of Unemployment Insurance Benefit Extensions: Evidence from U.S. Counties” by Klaus-Peter Hellwig (March 2021), Section 1.*

### Section 2

### wpiea2021070-print-pdf - Section 2

### Historical emergency UI programs
- Emergency Unemployment Compensation (EUC) Act of 1991:
  - Effective in November of 1991.
  - Temporarily replaced the EB program de facto.
  - Extended benefit duration for all states by at least 13 weeks, with higher extensions for high-unemployment states.
  - Included a “reachback” provision for workers who exhausted regular state UI benefits after March 1991.
  - Program amended several times and expired in April of 1994.
- Temporary Extended Unemployment Compensation (TEUC) program:
  - Signed into law on March 9, 2002 and expired in March of 2004.
  - Provided up to either 13 or 26 weeks of extended benefits to workers who exhausted regular benefits, depending on the state unemployment rate.
  - Entitlements under the EB program were left unaffected; state governors had discretion over payment order (TEUC before EB).
- Emergency Unemployment Compensation program of 2008:
  - Passed in June of 2008, initially granting extensions of up to 13 weeks.
  - November 2008: extensions increased up to 20 weeks for all states and a second tier added providing further 13-week extensions in high-unemployment states.
  - November 2009: second tier expanded to all states and third and fourth tiers created; tier generosity changed over time.
  - At its peak in mid-2012, allowed for EUC extensions of up to 63 weeks in states with all tiers activated.
  - Program expired on January 1, 2014.
- Aggregate implications:
  - Emergency discretionary programs produced considerable fluctuations in aggregate benefit payments, whereas aggregate payouts of extended benefits under EB have been relatively small.

### State-level treatment variable
- Definition (de-jure benefit changes):
  - Z_{s(c);t} = (#unemployed_{s(c);t-1} * WBA^{max}_{s(c);t-1} * Δx_{s(c);t}) / income_{s(c);t-1}
  - Interpretation: year-on-year change in the maximum aggregate dollar amount unemployed households in state s(c) can receive, assuming unemployment numbers and WBA^{max}_{s(c);t-1} remain constant.
  - Z_{s(c);t} can be interpreted as the change in the state’s de-jure maximum benefit-to-income ratio caused by federal benefit extensions.
  - Changes in unemployment or WBA^{max}_{s(c);t-1} are deliberately not included in Z_{s(c);t}.
- Empirical variation:
  - Correlation coefficient between Z_{s(c);t} and actual benefit growth is 69 percent.
  - Regressing benefit growth on Z_{s(c);t} yields an R-squared of 0.48.
  - Slope less than one because extensions define maximum duration, not average duration.

### County-level treatment intensity
- Baseline relationship (county level):
  - Δemployed_{c;t} = α * Z_{s(c);t} + β * θ_{c;t-5} * Z_{s(c);t} + φ_c + μ_{s(c);t} + ω_{c;t}
  - Regression used in practice (to address collinearity of Z with μ):  
    Δemployed_{c;t} = β * θ_{c;t-5} * Z_{s(c);t} + φ_c + μ̃_{s(c);t} + ω_{c;t}, where μ̃_{s(c);t} = α * Z_{s(c);t} + μ_{s(c);t}
- Treatment intensity θ_{c;t-L}:
  - θ_{c;t-L} = (1/15) * Σ_{j=1}^{15} (benefits_{c;t-L-j} / personal income_{c;t-L-j}) / (benefits_{s(c);t-L-j} / personal income_{s(c);t-L-j})
  - Measures 15-year average of each county’s UI transfer receipts (as a share of income), lagged by 5 years and normalized by the state average.
  - Captures structural county characteristics: industry composition, expected unemployment duration, UI eligibility shares, salary differences.
- Predictive power for local unemployment:
  - Regression of per-capita county unemployment on lags of θ shows within R-squared of 0.347 at a five-year lag, indicating θ_{c;t} accounts for one-third of within-state variation in unemployment rates.

### Identification strategy and interpretation
- Identification:
  - Interaction term θ_{c;t-5} * Z_{s(c);t} is the primary source of variation; state-by-year fixed effects μ_{s(c);t} control for state-wide cyclical conditions causing policy changes.
  - Strategy corresponds to difference-in-differences comparing high-benefit and low-benefit counties, with continuous treatment intensity.
  - Standard errors clustered at the state level.
- Limitations and interpretation nuances:
  - α (the direct effect of Z in a county with θ=0) cannot be separately estimated due to collinearity.
  - If counties are open economies, α may be > 0 because local demand can affect firms in other counties; focusing on β may understate total demand effects.
  - State-time fixed effects may absorb offsetting concurrent national or state policies (e.g., government spending, monetary policy, consumer expectations of future taxes).
  - DiD framework does not disentangle channels (labor market disincentives, housing market stabilization, Keynesian demand stabilization, reduced precautionary savings).
  - Sectoral analysis can help infer relative importance of demand vs supply channels (tradable vs non-tradable exposure).

### Controls and robustness concerns
- Omitted variable bias risks:
  - Other countercyclical transfers and tax changes correlate with UI extensions (retirement benefits, food stamps, education and training assistance, medical benefits, income maintenance, veterans’ benefits, other transfers).
  - Controls included for changes in benefit payments from a range of other programs and federal/state/local salary payments.
- Taxes:
  - Control for θ_{c;t-5} * ΔT_{s(c);t}, where ΔT_{s(c);t} is the annual change in the state’s effective individual income tax rate (computed as total individual income taxes divided by total personal income).
- Data and sample:
  - Economic data primarily from Bureau of Labor Statistics’ Regional Economic Accounts.
  - State income tax receipts from Annual Surveys on State and Local Government Finances (pre-1995) and Quarterly Summary of State and Local Government Finances (post-1995).
  - Sector-level employment data from County Business Patterns.
  - UI benefit generosity data compiled from various sources (Annex A).
  - Nominal series deflated using national CPI.
  - Largest estimation sample spans 1987 to 2019 using all counties.
  - Regressions weighted by average county population; robustness checks explore alternative weights.
  - θ_{c;t-L} and control variables winsorized at top and bottom 2.5th percentiles; dependent-variable outliers (top/bottom percentile) dropped.
  - θ_{c;t-L} constructed using data back to 1969; for 1987 and 1988 observations, θ uses 13 and 14 years respectively.

### Reduced-form results — Employment growth
- Baseline estimate (Table 2, column (i), minimal controls):
  - A 1-percentage point increase in the de-jure benefit-to-income ratio leads to a 0.275 percent increase in employment.
  - Reported standard errors indicate >99% confidence that β is positive.
- Sensitivity to fixed effects and controls:
  - Omitting state-by-year fixed effects reverses sign (negative) due to reverse causality.
  - Including θ_{c;t-5} (column (iii)) barely affects β; θ_{c;t-5} coefficient is highly significant and positive (mean reversion).
  - Extensive controls (column (iv)) and changes in effective state income tax rates (column (v)) leave the estimated treatment effect robust; tax rates are highly significant.
- Economic significance:
  - In 2014, a uniform 13-week benefit extension for all states would have increased average employment that year by more than 200,000 additional jobs nationwide.
  - If Great-Recession-era EUC extension levels from 2013 had remained in place instead of expiring on January 1st, 2014, employment in 2014 would have been higher by about 560,000 jobs.

### Reduced-form results — Other outcomes
- Labor force participation:
  - Estimated effect on labor force participation is larger than the effect on employment, suggesting unemployed workers are more likely to remain in the labor force and search.
  - Staying in the labor force may be directly required for continued eligibility or indirectly encouraged by stimulated economic activity.
- Migration:
  - Column (iii) indicates migration effects are small and less statistically significant; labor force growth in one county does not appear to come at the expense of others.
- Real wages:
  - Column (iv) finds a regression coefficient of zero for average real wage growth, providing no evidence that extensions reduce search effort or raise reservation wages.
- Private-sector labor earnings:
  - Column (v) shows significant positive effects on private-sector labor earnings, consistent with higher employment growth.
- Residential real estate prices:
  - Column (vi) reports a positive and highly significant effect on percentage change in residential real estate prices (Zillow Home Value Index), consistent with housing-market stabilization channels.

*Source: wpiea2021070-print-pdf - Section 2*

### Section 3

### wpiea2021070-print-pdf - Section 3

### Results by sector
- Data source: Census Bureau’s County Business Pattern data; sectors defined at the 4-digit SIC/NAICS level.
- Tradable vs non-tradable classification:
  - Tradable industries: sectors whose Herfindahl index is in the top quartile.
  - Non-tradable industries: sectors whose Herfindahl index falls into the bottom quartile.
- Key empirical findings:
  - Column (i) baseline: estimated treatment effect for total county employment similar to Table 2 (reduced-form coefficient reported in Table 2: 0.275*** with standard error (0.049)).
  - Tradable sectors (column (ii)): employment does not respond to UI benefit extensions; no statistically significant negative effect on manufacturing employment is found.
  - Non-tradable sectors (column (iii)): treatment effect is positive and statistically significant.
  - Broader-sector results (Table 6):
    - Construction (column (iii)) and retail trade (column (vi)) show highly statistically significant positive effects.
    - Positive impact on construction employment is consistent with the stabilizing effect on real estate prices reported in Table 4 and in Hsu et al. (2018).
- Identifying considerations:
  - The DiD identification strategy filters out changes in tradable-sector employment that originate from national or foreign demand via state-by-year fixed effects.
  - The asymmetry between tradable and non-tradable sectors is partly owed to the DiD design; it is unlikely additional consumer demand only covers non-tradables.
  - Observations where the sector of interest accounts for less than 1 percent of a county’s total employment are dropped; top and bottom 5-th percentiles are dropped for sectoral employment analyses.

### Fiscal multipliers
- IV approach specification:
  - First-stage: ∆benefits_{c;t} / income_{c;t-1} = δ · θ_{c;t-5} · Z_{s(c);t} + χ_c + κ_{s(c);t} + ε_{c;t}
  - Second-stage: ∆employed_{c;t} = γ · (∆benefits_{c;t} / income_{c;t-1}) + ψ_c + η_{s(c);t} + ζ_{c;t}
  - Fixed effects: same set as reduced-form regression; DiD in first stage isolates variation driven by state-wide cyclical conditions and county-specific structural factors.
- Interpretation caveat for γ:
  - γ captures the overall per-dollar impact of the benefit extension policy, not only the narrow cash-transfer effect.
  - UI programs also reduce idiosyncratic household income risk, which can affect aggregate demand via:
    - Prevention of mortgage defaults and housing-market stabilization (Hsu et al., 2018).
    - Reduced precautionary savings due to anticipated insurance (McKay and Reis, 2016).
    - Higher marginal propensity to consume (Ganong and Noel, 2019) and amplification via MPC heterogeneity (Kekre, 2016).
- First-stage and IV results (Table 7):
  - First-stage: a one-dollar increase in maximum benefits is associated with an increase in actual benefits of less than 5 cents.
  - F-statistic of first-stage: 158.3.
  - Second-stage:
    - Increasing benefits by one percent point of total household income leads to 2.451 percent more employment.
    - For every dollar paid in additional UI benefits, private sector labor earnings increase by 1.925 dollars.
- Comparison and context:
  - Estimated multiplier in column (iv) is almost exactly the same as Di Maggio and Kermani (2016) and broadly in line with other county-level spending multipliers:
    - Suarez-Serrato and Wingender (2016): effect on aggregate personal incomes between 1.7 and 2.
    - Shoag (2016): effect of state government spending on in-state income of 2.1.
    - Chodorow-Reich (2019): output multiplier around 1.8.
  - Explanations for large multipliers:
    - Keynesian demand effect from cash inflows.
    - Reduced precautionary savings of employed workers.
    - Wealth effect via housing market stabilization.
    - Targeting: UI benefit extensions flow disproportionately to regions with high unemployment.
  - OLS (columns (v) and (vi)) highlights a strong negative relationship between benefit payments and economic activity absent instrumentation, underscoring the need for IV to isolate exogenous policy variation.

### Conclusion
- Overall assessment:
  - Temporary extensions of UI benefit duration have, on average over the past three decades, helped stimulate economic activity.
  - Local fiscal multipliers for UI benefits are as large as those for other types of government spending.
  - Effects likely underestimated by the methodology because demand effects in tradable sectors are filtered out by state-by-year fixed effects.
- Labor-market disincentive findings:
  - No negative effect on tradable sectors is detected.
  - No evidence that benefit extensions affect average wages.
  - Labor market disincentive effects and search externalities appear negligible relative to positive employment effects from stimulated aggregate demand.
- Policy implication:
  - The estimated size of the multipliers makes UI benefit extensions an attractive alternative to other, less targeted, stabilization policies.
  - The discretionary nature of UI extension programs creates policy uncertainty and implementation lags.
  - Reform suggestion: reintroducing a national trigger for EB program extensions (as existed until 1981) to automate benefit extensions in recessions would eliminate policy uncertainty and enhance the capacity of UI benefit extensions to act as automatic stabilizers.

*Source: wpiea2021070-print-pdf - Section 3*

### Section 4

### wpiea2021070-print-pdf - Section 4

### Control variables (Table 3)
- Macroeconomic conditions and demographics:
  - farm proprietor earnings*: year t-1 level and change, year t-2 change (I71)
  - log(population): year t-1 level and change, year t-2 change (I20)
  - log(personal income per capita): year t-1 level and change, year t-2 change (I10)
  - dividend, interest, and rental income*: year t-1 level and change, year t-2 change (I46)
- Government salaries and wages:
  - state and local government wages and salaries*: year t change, year t-1 level and change (S930/N2010)
  - federal government wages and salaries*: year t change, year t-1 level and change (S910N2001)
  - military wages and salaries*: year t change, year t-1 level and change (S920/N2002)
- Government transfers (year t change, year t-1 level): other UI benefits*, retirement benefits*, income maintenance benefits*, medical benefits*, veteran benefits*, education and training assistance*, refundable tax credits*, other current transfers* (T2420, T2100, T2300, T2200, T2500, T2600, T5000, T2700)
- Contribution rates:
  - effective employer contribution rate to social insurance and pension and insurance funds: year t-1 level, year t-2 level (I60)
  - effective employee contribution rate to social insurance: year t-1 level, year t-2 level (I37)
- Notes:
  - Series marked with * are scaled by lagged personal income when the dependent variable is change in earnings. For all other dependent variables, they are scaled by lagged population.
  - Codes refer to BEA Regional Economic Accounts series codes. I=Income, T=current transfer, S/N = SIC/NAICS industry-level data.

### Reduced-form results across outcome variables (Table 4)
- Specification details:
  - County FE: yes; State-by-year FE: yes; Additional controls: yes (see Table 3)
  - Observations weighted by average county population.
- Coverage and sample sizes:
  - ∆employment: N = 100,533; Coverage 1987-2019
  - ∆labor force: N = 85,297; Coverage 1991-2019
  - ∆population: N = 100,530; Coverage 1987-2019
  - ∆wages: N = 100,530; Coverage 1987-2019
  - ∆earnings: N = 100,532; Coverage 1987-2019
  - ∆home values: N = 41,813; Coverage 1997-2019
- Key coefficient estimates (standard errors in parentheses):
  - Coefficient on _{c,t-5} Z_{s(c),t}:
    - ∆employment: 0.269*** (0.048)
    - ∆labor force: 0.439*** (0.057)
    - ∆population: 0.027* (0.015)
    - ∆wages: 0.009 (0.045)
    - ∆earnings: 0.166*** (0.039)
    - ∆home values: 0.222** (0.085)
  - Coefficient on _{c,t-5}:
    - ∆employment: 0.390*** (0.073)
    - ∆labor force: 0.288*** (0.066)
    - ∆population: 0.100** (0.038)
    - ∆wages: 0.149*** (0.054)
    - ∆earnings: 0.303*** (0.070)
    - ∆home values: 0.918*** (0.210)
  - Coefficient on _{c,t-5} ∆T_{s(c),t}:
    - ∆employment: -0.363*** (0.094)
    - ∆labor force: -0.468*** (0.151)
    - ∆population: 0.085** (0.042)
    - ∆wages: -0.165* (0.095)
    - ∆earnings: -0.213* (0.109)
    - ∆home values: 0.212 (0.236)
- Notes: standard errors clustered by state; significance: *, **, *** = 10%, 5%, and 1%. ∆employment in percent.

### Results for tradable and non-tradable sectors (Table 5)
- Sample and weighting:
  - Sample: 1987-2017; observations weighted by average county population.
  - N (all sectors) = 90,747; N (tradable) = 86,365; N (non-tradable) = 88,193
- Key coefficients (∆employment dependent):
  - _{c,t-5} Z_{s(c),t}:
    - all sectors: 0.325*** (0.081)
    - tradable: 0.109 (0.452)
    - non-tradable: 0.277** (0.104)
  - _{c,t-5}:
    - all sectors: 0.343*** (0.127)
    - tradable: 0.685 (0.481)
    - non-tradable: 0.146 (0.106)
  - _{c,t-5} ∆T_{s(c),t}:
    - all sectors: -0.244 (0.285)
    - tradable: -2.419** (1.043)
    - non-tradable: -0.027 (0.264)
- Notes: standard errors clustered by state; significance: *, **, *** = 10%, 5%, and 1%.

### Results by sector (Table 6)
- Sample: 1987-2017; observations where sector of interest accounts for < 1 percent of total county employment are omitted; observations weighted by average county population.
- N by sector:
  - agriculture (SIC 0700-0999 / NAICS 1100-1199): N = 16,963
  - mining: N = 20,360
  - construction: N = 81,646
  - manufacturing: N = 79,878
  - wholesale trade: N = 77,881
  - retail trade: N = 83,681
  - finance and insurance: N = 82,217
  - real estate: N = 18,883
  - hotels etc: N = 35,010
  - other services: N = 81,471
  - entertainment: N = 36,600
- Selected coefficient estimates on _{c,t-5} Z_{s(c),t}, _{c,t-5}, and _{c,t-5} ∆T_{s(c),t} (standard errors in parentheses):
  - Manufacturing (iv):
    - _{c,t-5} Z_{s(c),t}: 0.957*** (0.287)
    - _{c,t-5}: 0.289 (0.376)
    - _{c,t-5} ∆T_{s(c),t}: 0.076 (0.593)
  - Retail trade (vi):
    - _{c,t-5} Z_{s(c),t}: 0.715*** (0.109)
    - _{c,t-5}: 0.183 (0.140)
    - _{c,t-5} ∆T_{s(c),t}: 0.085 (0.329)
  - Hotels etc (ix):
    - _{c,t-5} Z_{s(c),t}: 0.896* (0.510)
    - _{c,t-5}: 0.532 (0.605)
    - _{c,t-5} ∆T_{s(c),t}: -0.170 (0.948)
- Notes: standard errors clustered by state; significance: *, **, *** = 10%, 5%, and 1%.

### Fiscal multipliers and IV estimates (Table 7)
- Sample: 1987-2019; observations weighted by average county population.
- First-stage:
  - Coefficient on ∆benefits_{c,t} income_{c,t-1} (OLS first stage): 2.451*** (0.656)
  - First-stage F statistic = 158.3
  - Coefficient on _{c,t-5} Z_{s(c),t} = 0.047*** (0.004)
- Second-stage (IV) and OLS on labor outcomes:
  - IV (second stage) ∆employment: -2.339*** (0.186)
  - IV (second stage) ∆earnings: -2.069*** (0.155)
  - OLS ∆employment: 0.411*** (0.075)
  - OLS ∆earnings: 0.323*** (0.071)
  - OLS (alternate columns) ∆employment: 0.380*** (0.074); ∆earnings: 0.293*** (0.071)
- Additional coefficients:
  - _{c,t-5}: -0.008*** (0.002) in first-stage column
  - _{c,t-5} ∆T_{s(c),t}: range from -0.009 (0.008) to -0.623*** (0.115) across specifications
- Notes: standard errors clustered by state; significance: *, **, *** = 10%, 5%, and 1%. ∆employment is in percent; ∆earnings is in percent of lagged county real personal income.

### Annex A — Data on UI benefit generosity (construction details)
- Benefit extensions:
  - For each state, length of benefit extensions in a calendar year = average number of maximum benefit extension weeks over all weeks of that year.
  - Components of weekly maximum benefit extension:
    1. Extended benefits (EB): Since 2003, Department of Labor weekly reports indicate participation in EB and number of weeks added (13 or 20). For pre-2003, Weekly Claims and Extended Benefits Trigger Data used; weeks reported with EB trigger before 2003 are assumed to add 13 weeks (no 20-week opt-in).
    2. EUC, 1991-94: use Lake (2002) compilation; during EUC, all EB triggers imposed off.
    3. TEUC, 2002-04: follow Lake (2003). First-tier duration = minimum of half regular UI duration and 13 weeks. Second-tier (TEUC-X) available only if trigger on. Trigger rule excerpt: a state is high-unemployment if IUR ≥ 4% and ≥ 120% of average 13-week IUR in prior 2 years for same period; once on TEUC-X, remains classified as high-unemployment for 13 weeks, proceeding in 13-week increments.
    4. EUC, 2008-13: determine eligibility for Tier 2, Tier 3, and Tier 4 using weekly reports by U.S. Department of Labor; changes in maximum weeks from U.S. Department of Labor (2018) and referenced laws.
- Maximum weekly benefit amounts:
  - Compiled semi-annually (January and July) by U.S. Department of Labor.
  - Procedure: parse reports with R script; manual corrections for older scanned pdfs; if report gives a range (e.g., dependents), take average of upper and lower bound; average January and July numbers to arrive at annual number.

### Annex B — Robustness: alternative weighting methods (Table 8)
- Sample: 1987-2019.
- Weighting methods compared:
  - by population (baseline), equal, by p area
- Key coefficients (dependent variable: ∆employment):
  - _{c,t-5} Z_{s(c),t}:
    - by population: 0.269*** (0.048)
    - equal: 0.278*** (0.043)
    - by p area: 0.294*** (0.056)
  - _{c,t-5}:
    - by population: 0.390*** (0.073)
    - equal: 0.419*** (0.061)
    - by p area: 0.406*** (0.070)
  - _{c,t-5} ∆T_{s(c),t}:
    - by population: -0.363*** (0.094)
    - equal: -0.295*** (0.086)
    - by p area: -0.226** (0.090)
- Notes: standard errors clustered by state; significance: *, **, *** = 10%, 5%, and 1%.

*Source: wpiea2021070-print-pdf - Section 4*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021070-print-pdf.pdf_