Methodology
How the Termination Cost Index is built, scored, and bounded. This is the published methodology for the 2026 edition.
What the index measures
The index ranks countries and territories by one quantity: the total statutory cost of ending an employment relationship, expressed in weeks of the employee’s salary. It adds two figures drawn directly from law: the minimum notice period an employer must give, and the minimum severance an employer must pay, for the redundancy dismissal of a standard full-time permanent employee. Rank 1 is the most expensive.
Both figures are already denominated in weeks of salary at the same tenure basis, so the index adds them rather than weighting them. That is a deliberate choice: any weighting would be an editorial judgment about whether a week of notice hurts more than a week of severance pay, and the index does not make one. Where we say “countries” for short, the set includes a number of territories and special administrative regions that set their own employment rules.
It measures statutory floors, not total cost. Collective agreements, enhanced contractual packages, protected-category rules, court awards for unfair dismissal, and the practical cost of the process itself all sit on top of the number shown here, and in many jurisdictions they are larger than the floor.
The metric
composite = percent_rank(total_weeks) Ă— 100
Notice and severance scale with length of service in most jurisdictions, so both are measured at a defined tenure basis: the value used is the average of the statutory amount due at 1, 5, and 10 years of service. This is the convention of the World Bank Employing Workers dataset, and values from other sources are aligned to it before they enter the index. A specific case with very short or very long service will differ from the figure shown, sometimes by a lot. In a country whose severance schedule is back-loaded, a ten-year employee can be worth several times the average.
The composite is the percent-rank of the total within the scored set, scaled to 0 to 100, where 100 is the highest total in the edition. It exists to make position legible at a glance. The weeks figure, not the composite, is the quantity to quote: the composite is relative to this edition’s set of countries, the weeks are not.
The two component percentiles shown in the table (notice and severance separately) are reported for context. They are not combined and they do not feed the composite.
Countries are ordered by total weeks, descending. Ties share a rank: countries with an identical total receive the same rank, and the next distinct total takes the next rank number (dense ranking). Because many countries share round statutory values, the last rank number is considerably smaller than the number of scored countries. The ranking is complete, not truncated.
Two conventions that change the numbers
Two rules sit between the statute text and the figure in the table. Both were settled in the August 2026 re-verification and both are stated here, because a reader reproducing a number straight from the law will otherwise get a different answer and conclude the data is wrong.
End-of-service gratuity regimes
Several countries discharge separation money through a service-based end-of-service gratuity rather than a payment labelled severance. Six of them are in this edition: the United Arab Emirates, Kuwait, Saudi Arabia, Qatar, Oman, and South Korea. The awkward feature they share is that the gratuity is payable on any exit, resignation included, so it can be argued to be deferred pay rather than a dismissal cost.
The rule this index applies is to count all of them, consistently. A gratuity is statutory, it is triggered by the termination, and an employer running a redundancy cannot avoid it, which is the same test every other severance entitlement in the index has to pass. Excluding them would remove real money that is genuinely owed on the day. What matters more than the choice is that one choice is applied to all six.
The UAE was previously the exception. It was scored at zero on severance, inherited from the 2019 World Bank vintage, while the other five carried their gratuity in full. That was not a judgment, it was an inconsistency, and it was corrected on 5 August 2026: the UAE now carries 18.1 weeks of severance under Federal Decree-Law 33/2021 art.51, which sets 21 days of basic wage per year for the first five years and 30 days per year beyond, averaged over the 1, 5 and 10 year tenure basis. One caveat travels with that figure: the UAE gratuity is computed on the basic wage, not on gross pay, so as a share of total remuneration it lands lower than the same number would in a country that pays severance on gross.
Band boundaries take the higher band
Most notice and severance schedules are banded by length of service, and the tenure basis lands on 1, 5 and 10 years, which are exactly the years the bands tend to break at. Where a schedule changes at five years, a worker at five years can be read either way. The World Bank figures apply the higher band: at an exact boundary the worker is treated as having passed it. We back-solved this from five independent statutes during re-verification, in Argentina, Cambodia, Albania, Morocco, and the Dominican Republic, and in every case only the upper-band reading reproduces the stored value.
The index keeps that convention, so its own values stay comparable with the archive they mostly come from. The honest consequence is worth stating: for a country whose notice or severance is banded, the figure shown is the cost at the point the next band engages, which sits at the top of the plausible range rather than in the middle of it. Countries with a flat per-year rate are unaffected.
Inclusion rule
A country is scored only if it has an approved, sourced value for both statutory notice and statutory severance, and is not on one of the two published exclusion lists below. If either input is missing, the country is excluded and listed by name along with which field it is missing. We do not estimate, interpolate, or use AI to fill gaps. This edition scores 190 countries and territories and excludes 60.
A statutory zero is a value, not a gap. Where the law genuinely sets no minimum notice or no minimum severance, the country is scored at zero for that component and stays in the ranking. Excluding those countries would quietly overstate the global picture. A zero we cannot source is a different thing, and it is handled by the second exclusion class below.
Two reasons beyond missing data
A figure can exist in the fact store and still not describe current law. The August 2026 re-verification found two situations where that is true, and both now remove a country from the ranking rather than leave a number standing that we cannot defend. Excluded countries are named on the study page with the reason, and the reason travels with the row in the downloadable dataset.
- No statutory formula.The governing law does not set a redundancy multiplier at all. Sierra Leone’s Employment Act 2023 and Ghana’s Labour Act 2003 both leave redundancy pay to negotiation between employer and worker, and Gambia’s Labour Act 2023 replaced the Act the stored figures rest on without a retrievable rate schedule. The values we held for all three were World Bank imputations from 2019, not entitlements, and all three sat in the top five of the ranking on them. An imputation at the top of a ranking is the one place it does the most damage, so they are excluded rather than published.
- Not verified. The stored value is zero for both inputs and no authoritative source establishes a statutory regime either way. Marshall Islands, Palau and Micronesia are unresearched rather than verified-zero. San Marino is a stronger case: dismissal indemnities plainly do exist under Law 7/1961 and Law 23/1977, but no rate schedule could be extracted, so the stored zero is more likely wrong than right. Absence of evidence is not evidence of a statutory zero, and the alternative would be to let four unresearched rows inflate a headline count of countries that cost nothing.
| Exclusion class | What it means | Countries |
|---|---|---|
| No sourced value | The fact store carries no approved, sourced value for statutory notice, statutory severance, or both. | 53 |
| No statutory formula | No statutory redundancy formula can be established from current law, so any weeks figure would be an imputation rather than a legal entitlement. | 3 |
| Not verified | No authoritative source establishes a current statutory notice or severance regime, so the stored zero cannot be read as a verified zero. | 4 |
A country is scored only if it has approved, non-null values for both statutory inputs (notice_weeks and severance_weeks), and is not on a published exclusion list. No gap-filling. Excluded countries are recorded with the reason they were excluded.
Data provenance
Every value comes from the same sourced-and-dated statutory dataset that powers our country pages and our Global Employer Burden Index. The great majority of values come from the World Bank Employing Workers dataset, with ILO EPLEX and national statutes supplying corrections and the long tail. This edition rests on 380 sourced statutory values across the two fields. Every row in the downloadable dataset carries its source and as-of date, so any number in the index can be traced back to where it came from.
| Source | Fields covered | Values |
|---|---|---|
| World Bank Employing Workers | notice weeks, severance weeks | 306 |
| ILO EPLEX | notice weeks, severance weeks | 43 |
| National statutes and government sources | notice weeks, severance weeks | 31 |
When more than one source carries the same figure, a fixed source-priority order decides which is used. A national statutory figure outranks a harmonised aggregator. Among the aggregators, the newer effective date wins.
On the World Bank data’s age:most notice and severance values come from the World Bank Employing Workers dataset, whose final vintage is May 2019. The Doing Business programme that produced it was discontinued in September 2021 and the data is preserved as a static archive. We use it because it remains the broadest notice and severance dataset collected under a single comparable methodology. Where we have verified that a law has changed since 2019, for example Egypt’s Labour Law 14/2025 and Indonesia’s Omnibus Law, the archive value is replaced with one sourced from the current national statute or ILO EPLEX, and the replacement is visible in the per-value source labels.
The August 2026 re-verification
Before this edition was published we re-checked the countries a reader is most likely to quote: the top of the ranking, every country the index scored at zero, and the largest economies. That is roughly 35 countries, each checked against the national statute in force rather than against the archive, using the same 1, 5 and 10 year tenure convention the rest of the index uses.
Six values were wrong and were corrected: Sierra Leone’s notice period under the Employment Act 2023, Mozambique’s severance under Labour Law 13/2023, Egypt’s notice period under Labour Law 14/2025, Thailand’s severance under the amended Labour Protection Act, Vietnam’s job-loss allowance once unemployment-insurance-covered service is excluded, and Puerto Rico, which the archive recorded as zero when Act 80 of 1976 in fact requires three months’ salary plus two weeks per year of service. Three countries were removed for having no statutory formula and four for an unverified zero, as set out above.
A second pass then took the rest of the top of the ranking, covering every country entry in ranks 11 to 42 the first pass had not reached. It produced no value corrections at all: every leg that could be re-sourced reproduced the figure already stored, so the order of the ranking down to rank 42 stands as published and the top five are untouched. It did produce the two conventions set out above, and one correction that was a matter of consistency rather than of sourcing, the UAE gratuity.
Seventeen country-legs could not be verified either way and are named rather than quietly counted as checked. They are SĂŁo TomĂ© and PrĂncipe, Somalia, Yemen, Guinea-Bissau, Sudan, Timor-Leste, Afghanistan, Libya and Dominica, on both inputs, plus Paraguay’s notice period and Tunisia’s severance. Each still rests on the 2019 archive or on a statute whose current authority we could not establish. For Afghanistan, Libya, Somalia, Sudan and Yemen the obstacle is that the applicable labour authority is itself contested.
Tunisia is under review. Its stored severance figure is higher than the ceiling we can reconstruct from the statute: the indemnity appears to run one day of wages per month of service and to be capped at three months of salary, which no tenure can push above 13.0 weeks. We have not changed the value, because only vendor-grade sources were reachable and a correction needs the Code du Travail text itself. Until that lands the figure stands as published with this note against it, and it is the one number in the top 42 we would not defend as sourced.
The countries outside that set still rest on the World Bank 2019 vintage. We have not silently upgraded their status: every value in the index carries its own source and as-of date, in the table on the study page and in the downloadable dataset, so a 2019 figure is visible as a 2019 figure. Re-verification continues by cohort, and corrections are applied to the current edition as they land rather than held for the next one.
Source dates refer to when a figure was published or a law enacted, not to validity: long-standing labour codes remain in force as amended. Bolivia’s severance and notice, for instance, are set by the General Labour Law of 1942, as amended, and still the governing statute today.
Editions and versioning
The index is published in annual editions (2026is the current one). The weeks figures are absolute and comparable across editions. The composite is not: it is a percentile within an edition’s set of included countries, so the same raw total can produce a different composite once the set changes. Compare weeks between editions, and positions within one.
Corrections policy:when a sourcing error is found within an edition, we correct the underlying value, rebuild the edition, and note the correction in the affected value’s provenance. We do not leave known-wrong numbers standing until the next edition. The dataset’s generated_at timestamp reflects the most recent rebuild.
Limitations
- Redundancy, not every exit route. The index prices a dismissal on economic or redundancy grounds. Dismissal for cause, resignation, mutual termination, and the end of a fixed term carry different entitlements, often much lower ones.
- Statutory, not negotiated. The number is the legal floor. Collective agreements, enhanced contractual severance, and settlement practice routinely push the real figure well above it, and in several European markets the negotiated package is the norm rather than the exception.
- Tenure basis. Notice and severance scale with service; the index uses the average of the amount due at 1, 5, and 10 years. A specific case with very short or very long service may differ substantially.
- Money, not friction.Weeks of salary do not capture how hard a dismissal is to carry out: works-council consultation, labour authority approval, mandatory reinstatement remedies, and litigation risk are real costs the index does not price. Two countries at the same number of weeks can be very different places to run a redundancy. The dataset carries the OECD’s own employment-protection indicators, where they exist, as an independent cross-check.
- Severance follows the statute’s label. Where a country channels end-of-service money through a deferred-compensation scheme rather than statutory severance, Italy’s TFR being the clearest example, it scores zero on severance, because the payment is accrued salary rather than a dismissal cost. The money is real, but it is not a statutory termination obligation. End-of-service gratuity regimes sit on the other side of that line and are counted in full, for the reasons set out under the two conventions above.
- End-of-service money is not always severance. Three cases are worth stating outright, because each one looks like an error until the definition is read. Germany has no general statutory severance entitlement at all; its figure reflects the customary half-month-per-year practice, which is not a legal floor. Vietnam’s headline job-loss allowance is near-nil in practice because service covered by compulsory unemployment insurance is excluded from the count. Mozambique’s rate depends on which national-minimum-wage band the salary falls in, and the index uses the standard band.
- Excluded, not estimated. Countries without a sourced value for both inputs are left out and listed by name, rather than scored on incomplete data. So are countries whose stored figure does not describe current law, under the two exclusion classes set out above.
Relationship to the Burden Index
Our Global Employer Burden Index scores the ongoing cost of employing someone and includes severance and notice as two of its three weighted pillars. This index isolates the exit: it uses the same two statutory inputs, at the same tenure basis, from the same fact store, but adds them into one absolute quantity instead of percent-ranking each pillar and weighting the result. A country can sit high on one index and low on the other: heavy ongoing social contributions with a cheap exit, or the reverse. For reference, the median scored country here owes 15.3 weeks of salary to end one employment relationship.