Independent validation confirms GST reform package revenue estimates are reasonable
- Independent actuarial firm validates States of Guernsey's 3% GST modelling, confirming calculations are accurate across all 66,021 individuals and 25,529 households
- Household GST estimated at £38.3 million (range £36-41 million), visitor GST at £3.5 million, and International Services Entity fees at £11 million, totalling approximately £53 million
- Two refinements identified during review resulted in net 4% reduction from earlier £41 million household GST estimate to £38.3 million
- Six different validation methods all support the States' figures, and model successfully reproduced Jersey's actual GST receipts within 10% when rescaled
- Review praises Guernsey's register-based census linked to tax records as leading approach among Crown Dependencies, eliminating sampling error and enabling full-population modelling
An independent actuarial firm has validated the States of Guernsey's tax reform modelling for a proposed 3% Goods and Services Tax, confirming the calculations are accurate and the revenue estimates reasonable.
Dorey Financial Modelling (DFM) conducted a comprehensive review of the States' modelling, rebuilding every calculation independently for all 66,021 individuals and 25,529 households in Guernsey. The firm achieved exact agreement with the States' model.
The review estimates household GST would generate £38.3 million at a 3% rate, with a range of £36 million to £41 million. Visitor GST would contribute £3.5 million, and International Services Entity fees would add £11 million, bringing total GST to approximately £53 million.
The Quarry
These figures represent revisions from earlier estimates, which had household GST at £41 million and visitor GST at £4.7 million, totalling £57 million. The reduction follows two refinements identified during the review process.
Martyn Dorey FIA, the lead reviewer, said: "The States' tax reform calculations are fair. Future receipts will differ from any estimate, because the economy, people's behaviour and tax collection cannot be known in advance."
The review identified two corrections during its assessment. The first involved how households should be ranked for expenditure mapping. Households should be ranked on equivalised income rather than gross income, which increased household GST estimates by 4%. The second correction involved replacing 2018-19 expenditure survey data with more recent 2023-24 figures, which reduced estimates by 8%. The combined net effect was a 4% reduction, approximately £4 million, explaining the revision from £41 million to £38.3 million.
DFM tested the States' figures using six different approaches, including top-down calculations from GDP, scaling from Jersey's receipts, survey aggregates, income-based estimates, and OECD revenue ratio methods. All five simpler methods bracketed the States' figure, supporting its validity.
The model was also rescaled to Jersey's population and spending levels. It reproduced Jersey's actual GST receipts within approximately 10%, described as "a good result" given the model was built for Guernsey and the jurisdictions have different tax bases and spending patterns.
The £36 million to £41 million range reflects combined uncertainties from multiple sources. Expenditure survey vintage contributes approximately ±4% uncertainty, economic cycle effects add ±2.3%, data uprating from 2022 to 2026 adds ±2%, estimated incomes contribute ±2%, survey cell sizes add ±3%, and income distribution effects contribute ±1%. These combine to approximately ±6% around the central estimate.
The modelling uses Guernsey's Rolling Electronic Census linked to tax records from 2022, uprated to 2026 prices. DFM compared this register-based approach favourably to Nordic systems, considered world-leading. Among Crown Dependencies and Gibraltar, Guernsey's approach is described as "leading."
Dorey said: "Bottom-up household microsimulation is the standard tool of finance ministries and research institutes for personal taxes... The full-population version the States have built is rarer, with the Nordic register-based models as its precedent."
For approximately 12% of total income where no administrative record exists, values were estimated using statistical techniques. Testing showed that if every estimate were 20% wrong, GST revenues would move by only 1.3%.
The model's own calibration against actual receipts is close, within 1.4% for income tax and contributions combined against 2026 outturn.
Several factors not included in the base model were separately examined. The model assumes 100% collection, but in reality 3% to 5% of tax typically goes uncollected, which would be £1 million to £2 million for household GST. A low, simple rate like 3% should result in better compliance than higher VAT rates, placing the gap at the lower end.
When prices rise, people buy slightly less. The policy team allows approximately £0.7 million for this behavioural effect, which DFM considers reasonable. The package also cuts income tax and contributions for some households, giving them more to spend. An estimated £0.8 million of additional GST would come from this increased spending.
In a worst-case scenario, with everything against it, household GST could be approximately £33.5 million, described as a floor for planning rather than an expectation.
Analysis by income quintile shows GST distribution across household income levels. The lowest quintile would pay £3.8 million (8% of total household GST), the second quintile £5.9 million (13%), the third quintile £8.6 million (19%), the fourth quintile £11.2 million (24%), and the highest quintile £16.6 million (36%). These figures are for GST alone and do not account for the whole package including income tax cuts, contribution changes, and benefit measures.
The policy team estimates the package would increase RPIX (Retail Prices Index excluding mortgage interest) by approximately 1.9%. DFM's analysis shows this sits within a defensible range of 1.5% to 2.5%, depending on assumptions about price pass-through.
The package includes a personal allowance increase from £15,200 to £15,800, a new 15% tax band on the first £28,000 of taxable income, and changes to social security contributions including new allowances and rates.
DFM reproduced income tax measures within £1 million and employer contributions within £1 million of published costings. Personal contributions show a £3 million difference, attributed to professional judgement on owner-manager reclassification rather than calculation error.
GST receipts will fluctuate with the economy. Normal cycle variation would be ±2% to 2.5% for household GST. A one percentage point rise in interest rates would reduce household GST by 0.1% to 0.4% after one to two years. A finance-sector downturn would impact income tax and contributions more severely than GST.
Mortgaged households, numbering 7,294 (29% of total), would be most affected by interest rate rises, concentrated in the middle three income quintiles.
Financial services businesses serving mainly non-residents would not charge GST on transactions but instead pay fixed annual fees by category. The structure includes 21 banks at £78,300, 145 lead trust companies at £13,100, 1,552 collective investment schemes at £4,700, and various other categories at lower rates. The total is £12.1 million gross, £10.1 million after a £2 million provision for voluntary non-registration, with £11 million carried as the midpoint.
Visitor GST calculations are based on Visit Guernsey's 2025-26 Visitor Expenditure Survey showing £149 million visitor spend, less £35 million for zero-rated air and sea travel, giving £3.5 million in GST.
The review specifically examined four assumptions questioned publicly. On expenditure survey age, claims that £41 million was £9 million to £15 million too high were not supported. The update from 2018-19 to 2023-24 survey data reduces the figure by 8%, but combined with the quintile correction gives a net reduction of 4% (£2.7 million).
On inflation effects, the 1.9% RPIX estimate sits within a defensible range of 1.5% to 2.5% depending on pass-through assumptions. On behavioural response, the model uses conventional elasticities giving a reduction of £0.7 million; at double the elasticities the effect would be £1.4 million. On rent treatment, under the package rent is zero-rated, so no pass-through issues arise.
Dorey said: "This is the most sophisticated of the three approaches and the most demanding of its inputs... Used within those limits, we consider the approach sound and proportionate."
Strict data safeguards were maintained throughout the review. Only three sworn Census Officers accessed the underlying data, working in a secure environment. The data was minimised before reaching reviewers, with every unnecessary field removed. The data was pseudonymised and handled as if it could identify people. No personal data left the secure environment or appears in any report. All published figures are aggregates or averages across many households. The data will be permanently deleted as directories were deliberately excluded from backup systems.
Guernsey's Rolling Electronic Census is a continuously updated population register built from administrative records and linked to tax and social security data. This approach eliminates sampling error entirely and allows household-level distributional analysis.
In Jersey, the 2021 census was a questionnaire with 99.5% response, administrative data is used only for checking rather than as a source, and the population register is not linked to tax data. The Isle of Man runs a questionnaire census every 10 years, as does Gibraltar. The UK has no population register, runs a census every 10 years, and HMRC costings run on samples.
Dorey said: "Among the Crown Dependencies and Gibraltar the approach is unique: Jersey, the Isle of Man and Gibraltar all still run a questionnaire census once a decade... none links it to tax records for modelling, and none could carry out a full-population calculation of this kind."
The review stated several limitations. The exemption step (£7.8 million) was taken as residual to the published figure rather than rebuilt item-by-item. The Power BI build was not reviewed, though the Excel model was recomputed in full. Uprating factors were not audited line-by-line against published indices. Imputed self-employment income in two age bands was not reconciled externally, with a 2% uncertainty allowance applied. Some 44% of individuals could not be linked between datasets; sensitivity was measured on the linked 56% and scaled up.
Several matters were stated as out of scope, including whether GST or any other measure should be adopted, which combination of measures the States should choose, benefit and pension measures accompanying the package, how the package falls on different household groups, the public online tax calculator, the States' spending plans, and Pillar 2 OECD global minimum tax.
On the calculations, the review concluded: "The models compute income tax, social security contributions and GST correctly for each person and household. We recomputed every field for all 66,021 individuals and 25,529 households and agreed with the model exactly."
On assumptions, it stated: "The explicit and implicit assumptions are fair and reasonable. We tested the contested assumptions, and exemptions set out in the model."
On revenue estimates, it concluded: "The final revenue estimates, including the adjustments made after the model runs, are a reasonable estimate of what the States would collect. Household GST at 3 per cent is fairly stated at £38.3 million within a range of £36 million to £41 million, total GST of about £53 million is a fair central estimate."
The report was prepared by Martyn Dorey FIA, Fellow of the Institute and Faculty of Actuaries, and peer reviewed by Robert Whitfield AIA, Associate of the Institute and Faculty of Actuaries, for the Policy & Resources Committee, States of Guernsey.
Q&A
Q: What is the estimated total GST revenue at 3%?
A: Approximately £53 million in total, comprising £38.3 million from household GST (range £36-41 million), £3.5 million from visitor GST, and £11 million from International Services Entity fees.
Q: Why were the revenue estimates revised downward from earlier figures?
A: Two refinements were identified during the independent review: households should be ranked on equivalised income rather than gross income (increasing estimates by 4%), and expenditure survey data should be updated from 2018-19 to 2023-24 (reducing estimates by 8%). The combined net effect was a 4% reduction, approximately £4 million, explaining the revision from £41 million to £38.3 million for household GST.
Q: How does Guernsey's tax modelling approach compare to other jurisdictions?
A: Guernsey uses a register-based census continuously updated from administrative records and linked to tax and social security data, enabling full-population modelling. This approach eliminates sampling error and is compared favourably to Nordic systems. Jersey, Isle of Man, and Gibraltar all use questionnaire censuses and do not link population registers to tax records for modelling. The UK has no population register and runs HMRC costings on samples.
Comments ()