## 2.1    Digitizing Historical Unemployment Claims

## Source details

**Canonical URL:** [2.1    Digitizing Historical Unemployment Claims](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022117-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022117-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022117-print-pdf.pdf.json)

---

### Digitization process and data sources
- Monthly state-level unemployment insurance claims available in digital form dating back to January 1971 from the Department of Labor’s website.
- Scanned printed reports from the DOL and SSA were used to digitize monthly Initial Claims (IC) and Continued Claims (CC) back to January 1947 for all 50 states and the District of Columbia.
- Primary historical sources: Employment Security Activities; The Labor Market and Employment Security; Unemployment Insurance Statistics; Unemployment Insurance Review.
- Access methods: HathiTrust, Google Books, Interlibrary Loan requests, and scans from the Department of Labor’s internal library; duplicate scans retrieved when needed to resolve image quality issues.
- Crosschecks and aids: national aggregates reported in the reports and reported percentage changes from the prior (or subsequent) month.
- Total newly digitized monthly observations: just over 36,000.

### Merging, seasonal adjustment, and outlier treatment
- Newly digitized claims merged with DOL’s publicly available state-level IC and CC for regular state programs to construct a full time series backdated to January 1947.
- IC digitized through December 1972 and CC digitized through December 1977 to validate smooth merges with DOL data; alignment: IC seamless throughout 1971–72; CC align between 1971–1977 depending on state.
- Full IC and CC series for January 1947 through December 2019 seasonally adjusted using the Census Win-X-13 seasonal adjustment program.
- Win-X-13 outlier tests identified roughly 200 potential outliers from approximately 91,000 observations; outliers roughly evenly distributed between newly digitized historical data and existing DOL data.
- Manual checking of each potential outlier to distinguish legitimate changes (e.g., surge after Hurricane Katrina) from data coding errors (e.g., implausible IC spike in Missouri in June 1974 exceeding the state’s population).
- Sample deliberately truncated ahead of the pandemic to avoid confounding influence of unprecedented IC spike when estimating seasonal factors.

### Construction of claims-based unemployment rates (method)
- Concept: claims-based unemployment rate analogous to BLS official unemployment rate but uses IC and CC as the measure of unemployed workers and uses employed workers plus unemployed workers receiving these benefits as the labor force.
- State-level employment measure: nonfarm payroll employment from Current Employment Statistics (CES) — only monthly state-level employment series back to 1947.
  - CES start-date exceptions: MN begins in January 1950; MI begins in January 1956; HI begins in January 1958; AK begins in January 1960; claims-based unemployment rate constrained to these later start dates for these four states.
- Nonfarm payroll employment for each state seasonally adjusted using Census Win-X-13.
- Monthly claims converted to average weekly claims (ICi,t and CCi,t) to avoid double-counting; monthly data weighted by split number of five-day workweeks in the month.
- Claims-based unemployment rate formula for state i in month t:
  - URClaims_i,t = (IC_i,t + CC_i,t) / (NP_i,t + IC_i,t + CC_i,t)
  - where NP_i,t denotes nonfarm payroll employment.
- National aggregation:
  - URClaims_US,t = (IC_US,t + CC_US,t) / (NP_US,t + IC_US,t + CC_US,t)
  - where IC_US,t = sum_{i=1}^{51} IC_i,t and CC_US,t defined analogously.
- Smoothing: series smoothed using a three-month centered moving average for figures and results.

### Empirical comparisons and validation
- Representative state example (Ohio): claims-based unemployment rate and BLS official unemployment rate over January 1976–December 2019 are highly correlated; correlation for Ohio is 0.84.
- Claims-based series provides nearly three additional decades of monthly state-level data relative to official series, spanning the first six post-war national recessions identified by the NBER.
- Expected level difference: claims-based measure should be strictly lower than the official unemployment rate because of narrower eligibility and benefit exhaustion beyond maximum duration for regular state UI programs.
- Seasonality/smoothing: plotted comparisons use three-month centered moving average.

### Comparison with official unemployment measures and data constraints
- Conceptual differences:
  - BLS official unemployment requires availability for work in the reference week and either active search in prior four weeks or expectation of recall; official unemployment rate computed from CPS.
  - Claims-based unemployment restricted to individuals actively claiming regular state UI benefits as reported weekly by state unemployment offices to ETA or predecessors.
- Strengths: administrative records avoid respondent misunderstanding or CPS nonresponse bias.
- Drawbacks: excludes exhausted-benefit workers and groups excluded from state UI (agriculture, domestic, independent contractors); potential for state reporting or transcription errors to ETA.
- Relation to Insured Unemployment Rate (IUR): claims-based rates lie between headline unemployment rate and IUR; IUR is ratio of weekly continued claims divided by covered employment.
  - National IUR availability: publicly available starting in 1971.
  - Official state-level IUR data availability: publicly available starting in January 1986.
  - Official state-level unemployment rates availability: starting in January 1976.
- Agreement: U.S. claims-based unemployment rate and official U.S. unemployment rate (January 1948–December 2019) highly correlated and identify broadly consistent business cycle features for the nation.
- Agricultural employment effect: use of nonfarm payroll employment downwardly biases labor force relative to official unemployment when farm share larger, pushing claims-based rates higher in earlier decades.
  - Ratio of total farm employment to nonfarm payroll employment: 23.6% in 1947, 13.0% in 1960, and 6.4% in 1970.

### Policy-relevant time-variation in program parameters
- Maximum duration of regular state benefits:
  - Mean maximum duration approximately 22 weeks in 1950; approximately 26 weeks by 1960.
  - From 1960 to 2011 remains quite stable around 26 weeks.
  - Declines slightly to 25 weeks when a handful of states reduced benefits during the recovery from the Great Recession.
  - Conclusion: mean maximum duration quite stable; share of unemployed out of work for 27 weeks or longer generally quite low.
- Implication: legislative changes to maximum duration of regular programs should have limited influence on time variation in total claims.
- Robustness: computing claims-based unemployment using IC only yields strong correlation with the IC+CC series.

### Key statistics and data facts
- Digital availability starting point: January 1971.
- Backdating target: January 1947 for all 50 states and the District of Columbia.
- Newly digitized monthly observations: just over 36,000.
- Seasonal adjustment period: January 1947 through December 2019.
- Outlier tests: roughly 200 potential outliers from approximately 91,000 observations.
- Ohio correlation (January 1976–December 2019): 0.84.
- Nonfarm payroll employment start dates: MN January 1950; MI January 1956; HI January 1958; AK January 1960.
- CES seasonal caveat: seasonally adjusted state-level nonfarm payroll employment data is only available from BLS for January 1990 onwards (seasonal adjustment performed using Win-X-13).
- National IUR availability: back to 1971.
- State-level IUR availability: from January 1986.
- State-level official unemployment rates availability: starting in January 1976.
- Historical monthly state-level IC and CC availability in Employment Security Activities reports: back to February 1943.
- Farm employment shares: 23.6% in 1947; 13.0% in 1960; 6.4% in 1970.
- Mean maximum duration of benefits: approximately 22 weeks in 1950; approximately 26 weeks by 1960; stable around 26 weeks from 1960 to 2011; declines slightly to 25 weeks after the Great Recession.

*Source: wpiea2022117-print-pdf - 2.1    Digitizing Historical Unemployment Claims.*

### 2.4    Fitted CBUR

### Statistical model and construction
- Model estimated Jan 1976–Dec 2019 for each state:
  - UROfficial_i,t = β0,i + β1,i (URClaims_i,t − URClaims_US,t) + β2,i UROfficial_US,t + ε_i,t
- Purpose:
  - Capture that a state’s unemployment exceeds the national rate when the state’s claims-based rate exceeds the national claims-based rate.
  - Leverage predictive power of the national unemployment rate for state unemployment rates (long-term unemployed, benefit exhaustion, UI exclusions).
- Back-casting:
  - Fitted models generate predicted state unemployment rates for January 1948 through December 1975.
  - Monthly national unemployment rate only dates back to January 1948 — fitted series starts in 1948 (raw claims-based series begins in 1947).

### Model performance and diagnostics
- Goodness of fit:
  - Average R2 is 0.85.
  - Average correlation coefficient of official and predicted unemployment rates is 0.92.
  - Maximum correlation: 0.97 (AZ, FL, ID, IN, OH).
  - Minimum correlation: 0.83 (SD).
- Empirical behavior:
  - Fitted series captures state business-cycle inflection points nearly identical to official unemployment rates over 1976–2019 and to raw claims-based series over 1948–2019.
  - Claims-based indicators tend to lead the official unemployment rate (peaks and troughs often identified earlier).

### Advantages and drawbacks of fitted unemployment rates
- Advantages:
  - Anchored to national official unemployment rate so level differences disappear while inflection points remain virtually identical.
  - Anchoring smooths the series because claims data tend to be noisier.
- Drawbacks:
  - Out-of-sample observations rely on assumption of stable empirical relationship.
  - Fitting truncates claims-based unemployment data availability from starting in 1947 to starting in 1948.
- Usage:
  - Both raw claims-based and fitted series are included in the dataset; users can choose which is more appropriate.
  - Both series generate similar results when examining pace and timing of economic recoveries.

### Role relative to survey data
- Motivation:
  - Monthly CPS micro-data does not exist pre-1976; CES nonfarm employment data is highly correlated with CPS but detailed CES begins in 1990.
  - Fitted model approximates BLS state-level unemployment rate without CPS micro-data.

*Source: wpiea2022117-print-pdf - 2.4    Fitted CBUR*

### 3.3    State Recession Dates and Related Robustness Checks

### Methodology: state-level recession dating
- Applied the DNS recession dating algorithm to the fitted claims-based unemployment rate for each state.
- Computed share of states recorded as being in a recession each month.
- DNS parameter choices and scaling:
  - DNS set parameter “X” = 1.5 for baseline national interpretation.
  - For states, computed average ratio of a state’s unemployment rate to the nation’s over entire dataset and applied this ratio to scale X for each state; denote state-level parameters as Y_i.
  - Scaled all Y_i’s down by 25% to reduce Type 2 errors.
  - For national (unfitted) claims-based unemployment rate alternative setting X = 1.0 matches NBER recession dates well.
  - For states, ratio of state-level and national claims-based unemployment rates used to scale each state’s “X” parameter, then scaled down by 25%.
- Data coverage notes:
  - Sample period discussed: 1948–2019 for Figure 9 plotting.
  - Not all states included early due to nonfarm payroll employment data limits; states added when feasible.
  - Data start dates by state where limited: MN January 1950; MI January 1956; HI January 1958; AK January 1960.
- Robustness: repeated recession dating on raw (unfitted) claims-based unemployment rates and computed alternative shares (Appendix Figure B.4).

### Findings: share of states in recession over time
- Peak share of states in recession roughly same across three pre-1960s recessions and seven subsequent recessions.
- High shares of states in recession are shorter lived in earlier recessions relative to later ones, consistent with NBER-measured shorter national recessions earlier.
- Disparate recovery rates in pre-1960s recessions are not simply due to many states not being in recession.
- Several national “recovery” periods show numerous states remaining in recession, notably between:
  - 1953-54 and 1957-58 recessions,
  - 1957-58 and 1960-61 recessions,
  - 1969-70 and 1973-75 recessions,
  - 2001 and 2007-09 recessions.

### Robustness checks and alternative specifications
- Repeating state-level dating on raw claims-based unemployment yields broadly consistent state recession shares with fitted-series results.
- Two notable differences when using raw series:
  - More pronounced “double-dip” recession in the 1980s.
  - Slightly fewer states coded as in recession during recoveries between 1953-54 and 1957-58 and between 1957-58 and 1960-61 relative to fitted results.
- Conclusion: both fitted and raw claims-based unemployment rates capture consistent features of state and national business cycles.

### Comparison with existing state recession estimates (validation)
- Used Owyang et al. (2005) state recession probabilities (February 1979–June 2002) as benchmark.
- General assessment:
  - Claims-based unemployment rates identify similar business cycle dynamics for most states, particularly larger ones, during 1979–2002 overlap.
  - Differences expected because Markov-switching probabilities and DNS algorithm identify related but different objects; state coincident indexes are broader measures than unemployment rates.
- Advantages of DNS/claims-based approach:
  - Fewer erratic, short-lived recessionary spikes and fewer dubiously long recessionary periods compared with Markov regime-switching probabilities.
  - No need to choose a cutoff for recession probabilities to identify recovery dates and durations.
  - Ability to identify inflection points in state business cycles for more than 30 additional years relative to off-the-shelf state coincident indexes.

### Implications and contribution
- Dataset expands panel availability for labor market slack back to January 1947 (raw) and January 1948 (fitted), providing nearly three additional decades of seasonally adjusted monthly state-level data prior to BLS official state unemployment rates beginning January 1976.
- Dataset doubles number of post-war U.S. recessions that can be studied at state level and supports findings that:
  - Slower national recovery rates since the 1960s associated with more uniform pace of recovery across states.
  - Faster national recoveries in the 1940s and 1950s associated with greater dispersion of recovery rates across states.
  - States with greater manufacturing shares tended to see faster recoveries, particularly in the 1940s and 1950s and to a smaller but significant extent since the 1960s.
- Enables new research: constructing backdated state coincident indexes and analyzing evolving network of economic integration across U.S. states over a longer horizon.

*Source: 3.3 State Recession Dates and Related Robustness Checks (wpiea2022117-print-pdf).*

### Appendix A    Data Appendix

### A.1 Data Construction
- Monthly state-level unemployment claims digitized from primary reports to backdate DOL online series:
  - January 1947–October 1949: Employment Securities Activities (ESA) report.
  - November 1949–October 1963: Labor Market and Employment Security (LMES) report.
  - November 1963 onward: Unemployment Insurance Statistics (UIS) report.
- Digitization coverage: regular state program IC and CC back to January 1947 for all 50 states and Washington, D.C.; Alaska and Hawaii available before statehood.
- Supplementary sources: HathiTrust, Interlibrary Loan, Google Book scans, Unemployment Insurance Review reports.
- Data merging for 1947–2019:
  - IC: digitized January 1947–December 1970 merged into DOL online data January 1971–December 2019.
  - CC (benchmark): digitized January 1947–June 1977 merged into DOL online data July 1977–December 2019.
  - Alternative CC: digitized January 1947–December 1970 merged into DOL data January 1971–December 2019 used as robustness check.
- Rationale for CC benchmark: UIS and DOL CC aligned for some states over 1971–72 but UIS CC higher for others; discrepancies disappear by mid-1977; UIS-digitized CC for January 1947–June 1977 judged to capture business-cycle fluctuations with stable level effect versus DOL data.
- Seasonal adjustment and outlier detection:
  - Neither digitized nor DOL data were originally seasonally adjusted.
  - Seasonal adjustment applied for full 1947–2019 sample using U.S. Census Bureau’s X-13 ARIMA-SEATS.
  - Win-X13 tests identified roughly 200 potential additive outliers (AO) and temporary change (TC) outliers from approximately 91,000 observations.
  - Manual verification: cross-check primary sources, compare IC and CC contemporaneous/lags, and examine nonfarm employment.
- Missing observations and revisions:
  - Used reported national aggregates to cross-check state sums.
  - Imputed missing monthly observations guided by reported monthly or annual changes when necessary.
  - When multiple sources reported a month, digitized data from the most recently published report used.
- Rules for correcting “fat thumb” coding issues and implausible observations:
  - Adjust when observation off by an order of magnitude or contradicted related indicators (IC, AWIU, nonfarm employment).
  - Replacement strategies: use primary source observation; adjust first digit when off by order of magnitude; linear interpolation when UIS reflected EB and DOL excluded EB.
- Specific manual adjustments (each adjusted value reported exactly):
  - DE May 1974 CC: 42,850 to 24,850
  - DE June 1981 CC: 6,433 to 36,433
  - FL March 1972 CC: 74,478 to 143,979
  - KY February 1974 CC: 1,218,070 to 121,807
  - MA January 1978 IC: 53,954 to 50,829
  - MA February 1978 IC: 90,507 to 86,580
  - MI February 1973 CC: 546,984 to 255,264
  - MO February 1974 CC: 31,088 to 201,743
  - MO June 1974 CC: 7,132,843 to 132,843
  - NY September 1973 CC: 76,674 to 642,675
  - NY August 1977 CC: 1,762,353 to 1,162,353
  - RI May 1984 CC: 6,796 to 56,796
- Interpolation for IL and MI around 1977 merge:
  - Used AWIU to interpolate CC for IL in March–April 1977 and MI in April–May 1977 using average ratio of CC to AWIU across the two previous and two subsequent months.
- Final processing:
  - After adjustments re-ran seasonal adjustment (without hard coding outliers).
  - Seasonally adjusted total U.S. regular state-program claims constructed by summing seasonally adjusted series for all 50 states plus Washington, D.C. (instead of seasonally adjusting aggregate).

### A.2 Data Validation and Robustness Checks
- Conceptual caveats:
  - UI claims proxy omits long-term unemployed who exhausted benefits.
  - Concern material only if large, state-specific changes in maximum benefit durations or waiting periods occurred.
- Maximum-duration and waiting-period policy variation:
  - Figure A.1 reports mean of each state’s maximum benefit duration with one-standard-deviation bands; relatively little variation across states in a given year.
  - Since mid-1950s, all states implemented either a one-week waiting period or no waiting period; a handful had two-week waiting periods at sample start but phased out by late 1940s/early 1950s.
  - Twenty four states never changed waiting period policies throughout 1947–2019; plurality consistently imposed a one-week waiting period.
  - States with two-week waiting periods in late 1940s: CO, GA, MN, MS, MT, NE, OH, WI, and WY. Colorado and Montana reduced to one-week between 1954 and 1955.
  - States with one-week waiting periods throughout 1947–2018: AK, AR, AZ, CA, FL, HI, ID, IL, IN, KS, LA, MO, ND, NM, NY, OK, OR, RI, SD, TN, UT, WA, and WV. Maryland never had a waiting period over this sample.
  - Only 58 waiting period policy changes over 1948–2018, 1.6% of the 3,550 state-year observations.
- Long-term unemployment shares:
  - Figure A.2 reports share of unemployed out of work for 27 weeks or longer.
  - Except for the Great Recession, long-term unemployed typically account for 5% to 25% of unemployed workers.
  - Over January 1948–December 2019, correlation between log level of unemployed workers and log level of unemployed workers who have been out of work for 26 weeks or fewer is 0.98.
  - Implication: excluding long-term unemployed has very little effect on unemployment dynamics and inflection points at national level.
- Take-up and denial rates concerns:
  - Slow-moving, uniform changes across states pose little threat (differenced out in URClaims_i,t − URClaims_US,t and corrected by inclusion of national unemployment).
  - Abrupt, state-specific changes in take-up or denial rates could undermine inference.
  - Racially differential take-up/denial effects are difficult to assess given LMES and UIS rarely report claims by race.
  - Kuka and Stuart (2021) for 1986–2015: racial take-up gaps relatively stable; observed characteristics explain 66% of take-up gap and 81% of benefit receipt gap; South fixed effects predictive for racial UI gaps.
- Regional checks:
  - Figure A.3 plots IC per capita by Census region for 1948–2019.
  - IC per capita in the South (Region III) consistently lower than other regions, follows same inflection points, and shows no discernible break after the Civil Rights Act.
- Overall validation:
  - Image quality and revisions generally posed minimal complications; alternative scans used when needed.
  - Manual checks reduced fat-thumb errors and outliers; instrumentation (IC, AWIU, nonfarm employment) used to validate spikes.
  - Adjustments for 1976 onwards have negligible effect on fitted out-of-sample state unemployment rates for January 1948–December 1975.
  - Official state unemployment rate estimated on left-hand side of equation (3) also constructed partly from UI claims subject to similar coding issues.
  - Seasonally adjusted series and robustness checks reported separately for benchmark and alternative CC specifications.

*Source: Appendix A    Data Appendix (wpiea2022117-print-pdf).*

### Appendix B    Additional Empirical Results

### B.1 Robustness Tests
- Claims-based unemployment rates smoothed with a three-month centered moving average.
- Comparisons plotted:
  - CBUR from IC+CC (red) versus IC only (green).
  - Cross correlations: official UR vs CBUR for 1948–2019; official UR vs IUR (available 1971 onwards); official UR vs CBUR over 1971–2019 to match IUR sample.
- State-level recession coding also constructed by applying DNS algorithm to state-level raw (unfitted) CBUR.
- Recovery dating: recovery dates from DNS algorithm generated from CBUR; the 1980 recovery excluded from claims-based recoveries.

### B.2 Claims-Based Recession Dates vs. Recession Probabilities
- Coverage and methods:
  - Claims-based state unemployment rates for January 1948–November 2019 (exceptions where nonfarm payroll data begins in 1950s).
  - State recession dates estimated from CBUR using Dupraz et al.(2019) algorithm.
  - State recession probabilities for February 1979–June 2002 from Owyang et al.(2005).
  - Y-axis measures unemployment rate in percentage points and recession probabilities in five-percentage point increments (20=100%).
- Similarities (1979–2002 overlap):
  - Both identify national recessions in 1980, 1981-82, 1990-91, and 2001 in many larger states, with slight date differences or double-dip discrepancies.
  - Near-identical non-national recessions: examples include Mississippi (claims-based Feb 1986–Jun 1986; Owyang Feb 1986–Jul 1986 with probabilities > 80%) and Wyoming and Idaho timing comparisons.
- Differences:
  - Owyang et al.(2005) identify short-lived mid-1990s recessions in certain states not found in claims-based dates.
  - Claims-based identifies short-lived recessions in smaller states not found in Owyang et al.(2005).
  - Owyang et al.(2005) do not identify 1990-91 recession in several states where claims-based does (ID, LA, OK, SD, TX, UT, WY).
  - Owyang et al.(2005) do not identify 2001 recession in KS, OK, or WY, unlike claims-based dates.
  - Extremes in Owyang et al.(2005): Alaska shown in continuous recession Aug 1989–Jun 2002 (average probability 97.5%); Hawaii shown in slump through almost all of 1990s (average probability 98.3% over Nov 1991–Dec 1999).
  - Claims-based contrasts: Alaska shown in shorter recession Aug 1989–Jul 1992; Hawaii shows double-dip recession July 1990–Dec 1992 and Oct 1993–Apr 1994.
- Comparative assessment:
  - Claims-based recession dates exhibit fewer erratic short-lived spikes and fewer unusually long recessionary periods than Owyang et al.(2005).
  - Claims-based approach avoids choosing a cutoff for recession probabilities to identify recoveries.
  - Principal advantage: identify inflection points in state business cycles for more than 30 additional years versus off-the-shelf state coincident indexes.

*Source: Appendix B    Additional Empirical Results (wpiea2022117-print-pdf).*

### 2.1    Digitizing Historical Unemployment Claims

### 2.1    Digitizing Historical Unemployment Claims

### Digitization process and data sources
- Monthly state-level unemployment insurance claims are available in digital form dating back to January 1971 from the Department of Labor’s website.
- Using scanned versions of printed reports previously published by the DOL and SSA, monthly data on Initial Claims (IC) and Continued Claims (CC) were digitized back to January 1947 for all 50 states and the District of Columbia.
- Primary historical sources include: Employment Security Activities, The Labor Market and Employment Security, Unemployment Insurance Statistics, and the Unemployment Insurance Review.
- Access methods: HathiTrust, Google Books, Interlibrary Loan requests, and scans from the Department of Labor’s internal library; duplicate scans were retrieved when needed to resolve image quality issues.
- Crosschecks and aids used in digitization: national aggregates reported in the reports and reported percentage changes from the prior (or subsequent) month.
- Total newly digitized monthly observations: just over 36,000.

### Merging, seasonal adjustment, and outlier treatment
- Newly digitized claims data were merged with DOL’s publicly available state-level IC and CC data for regular state programs to construct a full time series backdated to January 1947.
- IC digitized through December 1972 and CC digitized through December 1977 to validate smooth merges with DOL’s data; the two datasets align seamlessly for IC throughout 1971–72 and align for CC somewhere between 1971–1977, depending on the state.
- After merging, the full IC and CC series for January 1947 through December 2019 were seasonally adjusted using the Census Win-X-13 seasonal adjustment program.
- Win-X-13 outlier tests identified roughly 200 potential outliers from approximately 91,000 observations (newly digitized and existing data combined); outliers were roughly evenly distributed between newly digitized historical data and existing DOL data.
- Each potential outlier was manually checked to determine whether it reflected legitimate changes in claims (e.g., a surge in Louisiana following Hurricane Katrina) or data coding errors (e.g., an implausible IC spike in Missouri in June 1974 that exceeded the state’s population).
- The sample is deliberately truncated ahead of the pandemic to avoid the confounding influence of an unprecedented spike in IC when estimating seasonal factors.

### Construction of claims-based unemployment rates (method)
- Concept: claims-based unemployment rate is conceptually similar to the BLS official unemployment rate (ratio of unemployed workers to the labor force) but uses IC and CC as the measure of unemployed workers and uses employed workers plus unemployed workers receiving these benefits as the labor force.
- State-level employment measure: nonfarm payroll employment from the Current Employment Statistics (CES), selected because it is the only monthly state-level employment series going back to 1947.
  - Exceptions in CES start dates: MN begins in January 1950, MI begins in January 1956, HI begins in January 1958, and AK begins in January 1960; the claims-based unemployment rate is constrained to these later start dates for these four states.
- Seasonal adjustment: nonfarm payroll employment for each state is seasonally adjusted using the Census Win-X-13 seasonal adjustment program.
- Claims-to-weekly conversion: monthly claims are converted to average weekly claims for ICi,t and CCi,t to avoid double-counting individuals, conceptually similar to the BLS reference week and consistent with DOL data for average weekly insured unemployment; monthly data are weighted by the split number of five-day workweeks in the month.
- Claims-based unemployment rate formula for state i in month t (from January 1947 onwards):
  - URClaims_i,t = (IC_i,t + CC_i,t) / (NP_i,t + IC_i,t + CC_i,t) (equation (1))
  - where NP_i,t denotes nonfarm payroll employment.
- National aggregation:
  - URClaims_US,t = (IC_US,t + CC_US,t) / (NP_US,t + IC_US,t + CC_US,t) (equation (2))
  - where IC_US,t = sum_{i=1}^{51} IC_i,t and CC_US,t defined analogously.
- Smoothing: claims-based unemployment rate series are smoothed using a three-month centered moving average for figures and results.

### Empirical comparisons and validation
- Representative state example (Ohio): claims-based unemployment rate (red) and BLS official unemployment rate (blue) over January 1976–December 2019 are highly correlated; the correlation between the two series for Ohio is 0.84.
- Benefits of claims-based series: provides nearly three additional decades of monthly state-level data relative to official series, spanning the first six post-war national recessions identified by the NBER.
- Expected level difference: the claims-based measure should be strictly lower than the official unemployment rate because of the narrower pool of workers eligible for benefits and exhaustion of benefits for workers unemployed beyond the maximum duration for regular state UI programs.
- Seasonality and smoothing note: the claims-based unemployment rate is smoothed with a three-month centered moving average in plotted comparisons.

### Comparison with official unemployment measures and related data constraints
- Conceptual differences:
  - BLS official unemployment: requires availability for work in the reference week and either active search in the prior four weeks or expectation of recall; official unemployment rate is computed as unemployed divided by the labor force from CPS.
  - Claims-based unemployment: restricted to individuals actively claiming regular state UI benefits as reported weekly by state unemployment offices to ETA or its preceding federal agencies.
- Strengths of claims-based approach: no margin for respondent misunderstanding or CPS nonresponse bias; uses administrative records.
- Drawbacks: excludes workers who exhaust regular state benefits (they drop from the claims measure but remain in BLS unemployed if search requirements are met); state UI exclusions (agriculture, domestic workers, independent contractors) mean some unemployed workers counted by CPS are omitted from claims-based series; potential for state reporting or transcription errors to ETA.
- Relation to Insured Unemployment Rate (IUR): claims-based rates lie conceptually between the headline unemployment rate and the IUR; IUR is the ratio of weekly continued claims divided by covered employment and is conceptually similar (omits exhausted and ineligible workers).
  - National IUR availability: publicly available starting in 1971.
  - Official state-level IUR data availability: publicly available starting in January 1986.
  - Official state-level unemployment rates: available starting in January 1976.
- Empirical concordance: U.S. claims-based unemployment rate and the official U.S. unemployment rate (January 1948–December 2019) are highly correlated and identify broadly consistent business cycle features for the nation.
- Agricultural employment effect: use of nonfarm payroll employment downwardly biases the size of the labor force relative to the official unemployment rate when a larger share of workers are employed in agriculture, pushing claims-based unemployment rates higher in earlier decades.
  - Ratio of total farm employment to nonfarm payroll employment: 23.6% in 1947, 13.0% in 1960, and 6.4% in 1970.

### Policy-relevant time-variation in program parameters
- Maximum duration of regular state benefits:
  - Mean maximum duration begins at approximately 22 weeks in 1950, rises to approximately 26 weeks by 1960.
  - From 1960 to 2011 it remains quite stable around 26 weeks.
  - Declines slightly to 25 weeks when a handful of states reduced benefits during the recovery from the Great Recession.
  - Overall conclusion: mean maximum duration of benefits is quite stable; the share of unemployed workers out of work for 27 weeks or longer is generally quite low.
- Implication: given the relative stability of maximum durations and the typically small share of long-term unemployed, legislative changes to maximum duration of regular programs should have limited influence on time variation in total claims.
- Robustness check: computing the claims-based unemployment rate using IC data only (which is not impacted by changes in maximum duration policies) yields strong correlation with the IC+CC series (Appendix reference).

### Key statistics and data facts (preserved exactly)
- Digital availability starting point: January 1971 (DOL website).
- Backdating target: January 1947 for all 50 states and the District of Columbia.
- Newly digitized monthly observations: just over 36,000.
- Seasonal adjustment period: January 1947 through December 2019.
- Outlier tests: roughly 200 potential outliers from approximately 91,000 observations.
- Ohio correlation between claims-based and BLS official series (January 1976–December 2019): 0.84.
- Nonfarm payroll employment start dates for select states: MN January 1950; MI January 1956; HI January 1958; AK January 1960.
- CES-based nonfarm payroll seasonal series availability caveat: seasonally adjusted state-level nonfarm payroll employment data is only available from BLS for January 1990 onwards (seasonal adjustment performed using Win-X-13).
- National IUR availability: back to 1971.
- State-level IUR availability: from January 1986.
- State-level official unemployment rates availability: starting in January 1976.
- Historical monthly state-level IC and CC availability in Economic Security Activities reports: back to February 1943.
- Share of farm employment to nonfarm payroll employment: 23.6% in 1947, 13.0% in 1960, 6.4% in 1970.
- Mean maximum duration of benefits: approximately 22 weeks in 1950; approximately 26 weeks by 1960; stable around 26 weeks from 1960 to 2011; declines slightly to 25 weeks after the Great Recession.

*Source: wpiea2022117-print-pdf - 2.1    Digitizing Historical Unemployment Claims.*

### 2.4    Fitted CBUR

### 2.4    Fitted CBUR

### Statistical model and construction
- Model specification (estimated Jan 1976–Dec 2019 for each state):
  - UROfficial_i,t = β0,i + β1,i (URClaims_i,t − URClaims_US,t) + β2,i UROfficial_US,t + ε_i,t
- Purpose:
  - Capture that a state’s unemployment rate is higher than the national rate when the state’s claims-based unemployment rate is higher relative to the national claims-based rate.
  - Leverage predictive power of the national unemployment rate for state unemployment rates (long-term unemployed, benefit exhaustion, exclusions from UI eligibility).
- Back-casting:
  - Fitted models used to generate predicted state unemployment rates for January 1948 through December 1975.
  - Monthly national unemployment rate only dates back to January 1948—fitted series starts in 1948 (raw claims-based series begins in 1947).

### Model performance and diagnostics
- Goodness of fit:
  - Average R2 is 0.85.
  - Average correlation coefficient of official and predicted unemployment rates is 0.92.
  - Maximum correlation: 0.97 (AZ, FL, ID, IN, OH).
  - Minimum correlation: 0.83 (SD).
- Empirical behavior:
  - Fitted series captures state business-cycle inflection points nearly identical to official unemployment rates over 1976–2019 and to raw claims-based unemployment rate over 1948–2019.
  - Claims-based indicators tend to lead the official unemployment rate (peaks and troughs often identified earlier in claims-based series).

### Advantages and drawbacks of fitted unemployment rates
- Advantages:
  - Anchored to the national official unemployment rate so level differences disappear while inflection points remain virtually identical.
  - Anchoring helps smooth the series because claims data tend to be noisier than unemployment data.
- Drawbacks:
  - Out-of-sample observations rely on the assumption of a stable empirical relationship.
  - Fitting truncates claims-based unemployment data availability from starting in 1947 to starting in 1948.
- Usage:
  - Both raw claims-based and fitted series are included in the dataset; users can choose which is more appropriate.
  - Both series generate similar results when examining pace and timing of economic recoveries.

### Role of fitted series relative to survey data
- Motivation:
  - Monthly CPS micro-data does not exist pre-1976; CES nonfarm employment data is highly correlated with CPS but detailed CES begins in 1990.
  - Fitted model approximates BLS’s state-level unemployment rate without CPS micro-data.

*Source: wpiea2022117-print-pdf - 2.4    Fitted CBUR*

### 3.3    State Recession Dates and Related Robustness Checks

### 3.3    State Recession Dates and Related Robustness Checks

### Methodology: state-level recession dating
- Applied the DNS recession dating algorithm to the fitted claims-based unemployment rate for each state.
- Computed the share of states recorded as being in a recession in each month.
- DNS algorithm parameter choices and scaling:
  - DNS set the algorithm parameter “X” to be 1.5 for the baseline national interpretation referenced.
  - For state-level application, computed the average ratio of a state’s unemployment rate to the nation’s over the entire dataset and applied this ratio to scale the X parameter for each state; denote these state-level parameters as Y_i.
  - Scaled all the Y_i’s down by 25% to reduce Type 2 errors in recession dating.
  - For the national (unfitted) claims-based unemployment rate the alternative setting was X = 1.0, which generates a good match with NBER recession dates.
  - For states, the ratio of the state-level and national claims-based unemployment rates was used to scale each state’s “X” parameter, and these were again scaled down by 25% to be conservative.

- Data coverage notes:
  - Sample period discussed: 1948–2019 for Figure 9 plotting.
  - Due to data limitations in nonfarm payroll employment, not all states are included early in the sample; states are added when feasible.
  - Data start dates by state where limited: MN begins in January 1950, MI begins in January 1956, HI begins in January 1958, and AK begins in January 1960.
- Also repeated recession dating on raw (unfitted) claims-based unemployment rates as a robustness check and computed alternative shares (plotted in Appendix Figure B.4).

### Findings: share of states in recession over time
- The peak share of states experiencing a recession is roughly the same across the three pre-1960s recessions associated with faster recoveries and the seven subsequent recessions.
- High shares of states in recession are shorter lived in earlier recessions relative to later ones, consistent with these earlier national recessions being relatively short-lived as measured by NBER.
- The disparate recovery rates in the pre-1960s recessions documented elsewhere are not simply a product of many states not being in recession.
- Several national “recovery” periods show numerous states remaining in recession; in particular, a sizable share of states remains coded as still experiencing a recession throughout the national recoveries between:
  - the 1953-54 and 1957-58 recessions,
  - the 1957-58 and 1960-61 recessions,
  - the 1969-70 and 1973-75 recessions,
  - the 2001 and 2007-09 recessions.

### Robustness checks and alternative specifications
- Repeating state-level recession dating on raw (unfitted) claims-based unemployment rates yields broadly consistent state recession shares with the fitted-series results (Appendix Figure B.4).
- Two notable differences when using raw claims-based series:
  - A more pronounced “double-dip” recession in the 1980s, possibly because claims capture shorter-duration changes in unemployment and because the national UR (included as a regressor in the fitted series) does not prominently display the double-dip.
  - Slightly fewer states coded as in recession during the recoveries between 1953-54 and 1957-58 and between 1957-58 and 1960-61 relative to the fitted-series results.
- Conclusion from robustness checks: both fitted and raw claims-based unemployment rates capture consistent features of state and national business cycles.

### Comparison with existing state recession estimates (validation)
- Used Owyang et al. (2005) state recession probabilities (February 1979–June 2002) as a benchmark for validation; that sample is limited by availability of state coincident indexes and official state unemployment data starting in January 1976.
- Appendix Figure B.5 depicts claims-based unemployment rates (blue lines), state recession dates (gray bars), and Owyang et al. (2005) state recession probabilities (red lines) for all 50 states over the overlapping sample.
- General assessment:
  - The claims-based unemployment rates identify similar business cycle dynamics for most states, particularly larger ones, during the overlapping 1979–2002 sample.
  - Differences are expected because:
    - Markov-switching models and the DNS algorithm identify related but fundamentally different objects.
    - State coincident indexes are a broader measure of economic activity than state unemployment rates (the latter being the exclusive focus here).
- Advantages of the DNS/claims-based approach highlighted:
  - Recession dates derived here exhibit fewer erratic, short-lived recessionary spikes or dubiously long recessionary periods compared with Markov regime-switching probabilities.
  - No judgment required regarding a cutoff for recession probabilities to identify recovery dates and durations.
  - Ability to identify inflection points in state business cycles for more than 30 additional years relative to off-the-shelf state coincident indexes.

### Implications and contribution
- The claims-based state unemployment dataset expands panel data availability for measuring labor market slack back to January 1947 (raw series) and January 1948 (fitted series), providing nearly three additional decades of seasonally adjusted monthly state-level data prior to BLS official state unemployment rates beginning in January 1976.
- The dataset doubles the number of post-war U.S. recessions that can be studied at the state level and supports analysis showing:
  - Slower national recovery rates since the 1960s have been associated with a more uniform pace of recovery across states.
  - Faster national labor market recoveries in the 1940s and 1950s were associated with greater dispersion of recovery rates across states.
  - States with greater manufacturing shares of economic output tended to see faster recoveries, particularly in the 1940s and 1950s, and to a smaller but significant extent since the 1960s.
- The dataset enables new research avenues, including constructing backdated state coincident indexes and analyzing the evolving network of economic integration across U.S. states over a longer horizon.

*Source: 3.3 State Recession Dates and Related Robustness Checks (wpiea2022117-print-pdf).*

### Appendix A    Data Appendix

### Appendix A    Data Appendix

### A.1 Data Construction
- Monthly state-level unemployment claims digitized from primary reports to backdate DOL online series:
  - January 1947–October 1949: Employment Securities Activities (ESA) report (Bureau of Employment Security, Federal Security Agency Social Security Board).
  - November 1949–October 1963: Labor Market and Employment Security (LMES) report (Bureau of Employment Security, U.S. Department of Labor).
  - November 1963 onward: Unemployment Insurance Statistics (UIS) report (Bureau of Employment Security, U.S. Department of Labor Manpower Administration).
- Digitization coverage: regular state program IC and CC back to January 1947 for all 50 states and Washington, D.C.; Alaska and Hawaii available before statehood.
- Supplementary sources used where needed: HathiTrust scans, Interlibrary Loan, Google Book scans, Unemployment Insurance Review (UIR) reports.
- Data merging strategy for complete 1947–2019 series:
  - IC: digitized January 1947–December 1970 merged into DOL online data January 1971–December 2019.
  - CC (benchmark): digitized from UIS and earlier sources January 1947–June 1977 merged into DOL online data July 1977–December 2019.
  - Alternative CC series: digitized January 1947–December 1970 merged into DOL data January 1971–December 2019 (used as robustness check).
- Rationale for CC benchmark choice:
  - UIS and DOL CC series aligned for some states over 1971–72, but for other states UIS CC were significantly higher; discrepancies generally disappear by mid-1977.
  - UIS-digitized CC series for January 1947–June 1977 judged to capture business-cycle fluctuations with a stable level effect versus DOL data.
- Seasonal adjustment and outlier detection:
  - Neither digitized nor DOL data were seasonally adjusted originally.
  - Seasonal adjustment applied for full 1947–2019 sample using U.S. Census Bureau’s X-13 ARIMA-SEATS.
  - Win-X13 tests identified roughly 200 potential additive outliers (AO) and temporary change (TC) outliers from approximately 91,000 observations.
  - Manual verification process: cross-check primary sources, compare IC and CC contemporaneous/lags, and examine nonfarm employment for corroboration.
- Handling of missing observations and revisions:
  - Used reported national aggregates to cross-check state sums.
  - Imputed missing monthly observations guided by reported monthly or annual changes when necessary.
  - When multiple sources reported a month, digitized data from the most recently published report used.
- Examples and rules for correcting “fat thumb” coding issues and implausible observations:
  - Adjustments made when an observation was off by an order of magnitude or contradicted related indicators (IC, AWIU, nonfarm employment).
  - Replacement strategies: use primary source observation, adjust first digit when off by order of magnitude, or linear interpolation between prior and subsequent month when UIS reflected EB and DOL series excluded EB.
- Specific manual adjustments to identified “fat thumb” outliers (each adjusted value reported exactly):
  - DE May 1974 CC: 42,850 to 24,850 (UIS report reads “24,850” not “42,850”)
  - DE June 1981 CC: 6,433 to 36,433 (off by an order of magnitude)
  - FL March 1972 CC: 74,478 to 143,979 (UIS report reads “148,845” not “74,478’)
  - KY February 1974 CC: 1,218,070 to 121,807 (off by an order of magnitude)
  - MA January 1978 IC: 53,954 to 50,829 (UIS report reads “50,829” not “53,954”)
  - MA February 1978 IC: 90,507 to 86,580 (UIS report reads “86,580” not “90,507”)
  - MI February 1973 CC: 546,984 to 255,264 (UIS report reads “413,526” not “546,984”)
  - MO February 1974 CC: 31,088 to 201,743 (off by an order of magnitude)
  - MO June 1974 CC: 7,132,843 to 132,843 (off by an order of magnitude)
  - NY September 1973 CC: 76,674 to 642,675 (UIS report reads “671,981” not “76,674”)
  - NY August 1977 CC: 1,762,353 to 1,162,353 (UIS report reads “1,162,353” not “1,762,353”)
  - RI May 1984 CC: 6,796 to 56,796 (off by an order of magnitude)
- Interpolation for Illinois and Michigan around the 1977 merge:
  - Used monthly average weekly insured unemployment (AWIU) to interpolate CC for IL in March–April 1977 and MI in April–May 1977.
  - Method: compute average ratio of CC to AWIU across the two previous and two subsequent months, then multiply by AWIU in months in question.
- Final processing:
  - After adjustments re-ran seasonal adjustment (without hard coding outliers).
  - Seasonally adjusted total U.S. regular state-program claims constructed by summing seasonally adjusted series for all 50 states plus Washington, D.C. (rather than seasonally adjusting aggregate).

### A.2 Data Validation and Robustness Checks
- Conceptual caveats when using UI claims as proxy for unemployment:
  - Claims-based unemployment rate omits long-term unemployed who exhausted benefits (as does official IUR).
  - Concern only material if there were large, state-specific changes in maximum benefit durations or waiting periods.
- Maximum-duration and waiting-period policy variation:
  - Figure A.1 reports mean of each state’s maximum benefit duration with one-standard-deviation bands; there is relatively little variation across states in a given year.
  - Since mid-1950s, all U.S. states implemented either a one-week waiting period or no waiting period; a handful had two-week waiting periods at sample start but were phased out by late 1940s or early 1950s.
  - Twenty four states never changed waiting period policies throughout 1947–2019; plurality consistently imposed a one-week waiting period.
  - States with two-week waiting periods in the late 1940s: CO, GA, MN, MS, MT, NE, OH, WI, and WY. Colorado and Montana reduced to one-week between 1954 and 1955.
  - States with one-week waiting periods throughout 1947–2018: AK, AR, AZ, CA, FL, HI, ID, IL, IN, KS, LA, MO, ND, NM, NY, OK, OR, RI, SD, TN, UT, WA, and WV. Maryland never had a waiting period over this sample.
  - There were only 58 waiting period policy changes over 1948–2018, 1.6% of the 3,550 state-year observations.
- Long-term unemployment shares:
  - Figure A.2 reports share of unemployed who have been out of work for 27 weeks or longer.
  - Except for the Great Recession, long-term unemployed typically account for 5% to 25% of unemployed workers.
  - Over January 1948–December 2019, correlation between log level of unemployed workers and log level of unemployed workers who have been out of work for 26 weeks or fewer is 0.98.
  - Implication: excluding long-term unemployed has very little effect on unemployment dynamics and inflection points at national level.
- Take-up rates and denial rates concerns:
  - Slow-moving, uniform changes across states pose little threat (differenced out in URClaims i,t − URClaims US,t in equation (3) and corrected by inclusion of national unemployment).
  - Abrupt, state-specific changes in take-up or denial rates could undermine inference.
  - Racially differential take-up/denial effects (e.g., post–Civil Rights Act of 1964) are difficult to assess given LMES and UIS rarely report claims by race.
  - Kuka and Stuart (2021) evidence for 1986–2015: racial take-up gaps in UI relatively stable; observed characteristics explain 66% of gap in take-up and 81% of gap in benefit receipt; South fixed effects have predictive power for racial UI gaps.
- Regional checks:
  - Figure A.3 plots IC per capita by Census region for 1948–2019.
  - IC per capita in the South (Region III) is consistently lower than other regions, follows same inflection points, and shows no discernible break after the Civil Rights Act—supporting stability of regional IC dynamics across entire sample.
- Overall validation notes:
  - Image quality and revisions generally posed minimal complications; alternative scans used when needed.
  - Manual checks reduced fat-thumb errors and outliers, and instrumentation (IC, AWIU, nonfarm employment) used to validate legitimacy of spikes.
  - Adjustments for 1976 onwards have negligible effect on fitted out-of-sample state unemployment rates for January 1948–December 1975.
  - The official state unemployment rate being estimated on the left-hand side of equation (3) is also constructed partly from UI claims data subject to similar coding issues.
  - Seasonally adjusted series and robustness checks reported separately for benchmark and alternative CC specifications.

*Source: Appendix A    Data Appendix (wpiea2022117-print-pdf).*

### Appendix B    Additional Empirical Results

### Appendix B    Additional Empirical Results

### B.1 Robustness Tests
- Claims-based unemployment rates are smoothed with a three-month centered moving average.
- Figure comparisons:
  - Claims-based unemployment rates computed from IC+CC data (red line) versus IC data only (green line) are plotted with left and right axes, respectively.
  - Cross correlations plotted between:
    - official unemployment rate (UR) versus claims-based unemployment rate (CBUR) for 1948-2019;
    - official unemployment rate versus insured unemployment rate (IUR), available from 1971 onwards;
    - official unemployment rate versus CBUR over 1971–2019 to match IUR's sample.
- State-level recession coding:
  - Constructed by applying the DNS algorithm to state-level raw (unfitted) claims-based unemployment rates.
  - DNS algorithm parameter is adjusted for each state proportionate to its average level of unemployment over the entire time period.
- Recovery dating:
  - Recovery dates from DNS algorithm are generated from the claims-based unemployment rate.
  - The 1980 recovery is excluded from claims-based recoveries.

### B.2 Claims-Based Recession Dates vs. Recession Probabilities
- Data coverage and methods:
  - Claims-based state unemployment rates (blue) are for January 1948–November 2019, with exceptions for states where nonfarm payroll employment data begins in the 1950s.
  - State recession dates (gray bars) are estimated from claims-based unemployment rates using the Dupraz et al.(2019) algorithm.
  - State recession probabilities (red) for February 1979–June 2002 are from Owyang et al.(2005).
  - The y-axis measures the unemployment rate in percentage points and recession probabilities in five-percentage point increments (20=100%).
- Key similarities across datasets (1979–2002 overlap):
  - Both approaches identify the national recessions in 1980, 1981-82, 1990-91, and 2001 in many larger states, though with slightly different peak/trough dates or differences regarding double-dip versus longer early-1980s recessions (examples: AZ, CA, CT, FL, IL, MA, NC, NY).
  - Examples of near-identical non-national recessions:
    - Mississippi: claims-based recession over February 1986–June 1986; Owyang et al.(2005) identify February 1986–July 1986 with probabilities exceeding 80% for each month.
    - Wyoming: Owyang et al.(2005) identify February 1986–March 1987 with probabilities exceeding 80%; claims-based identifies December 1984–October 1986.
    - Idaho: Owyang et al.(2005) identify July 1985–August 1986 with probabilities exceeding 80%; claims-based identifies December 1984–October 1986.
- Key differences across datasets:
  - Owyang et al.(2005) identify short-lived mid-1990s recessions in Maine, Maryland, New Mexico, Washington that are not identified in claims-based dates.
  - Claims-based dataset identifies short-lived recessions in smaller states not found in Owyang et al.(2005), e.g., Delaware, Hawaii, North Dakota in the 1980s and 1990s.
  - Owyang et al.(2005) do not identify the 1990-91 recession in several states where claims-based dates do (ID, LA, OK, SD, TX, UT, WY).
  - Owyang et al.(2005) do not identify the 2001 recession in Kansas, Oklahoma, or Wyoming, unlike claims-based dates.
  - In some states where both identify recessions around 1990-91 and 2001, Owyang et al.(2005) show considerably shorter downturns than claims-based unemployment recession dates (examples: KY, MN, NV, WI).
  - Extremes in Owyang et al.(2005):
    - Alaska shown continuously in recession from August 1989–June 2002, with recession probabilities averaging 97.5% and never falling below 50% for this sample.
    - Hawaii shown in a slump throughout almost all of the 1990s, with recession probabilities averaging 98.3% and never falling below 60% over November 1991–December 1999.
  - Claims-based contrasts:
    - Alaska shown in a much shorter recession, over August 1989–July 1992, reflecting a persistent recovery in claims-based unemployment rates in the early 1990s.
    - Hawaii identified as experiencing a double-dip recession: July 1990–December 1992 and October 1993–April 1994, followed by a persistent recovery in unemployment.
- Comparative assessment:
  - Claims-based recession dates exhibit fewer erratic, short-lived recessionary spikes and fewer unusually long recessionary periods compared to Owyang et al.(2005).
  - Claims-based approach avoids the need for choosing a cutoff for recession probabilities to identify recovery dates and durations.
  - Principal advantage of claims-based approach: ability to identify inflection points in state business cycles for more than 30 additional years versus existing off-the-shelf state coincident indexes.

*A New Claims-Based Unemployment Dataset: Application to Postwar Recoveries Across U.S. States Working Paper No. WP/2022/117*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022117-print-pdf.pdf_