Express has been asking questions for months about the modelling behind GST and the rest of Guernsey’s tax reforms, from who built it and what data it uses to who has actually been allowed to see it.

We’ve finally had some answers from Policy and Resources (P&R).

And perhaps the most striking is this: P&R hasn’t seen the modelling used to understand the knock-on effects of its own proposals. Neither have any other deputies.

However, five civil servants have… along with the private firm brought in to check their workings.

But what else have we learned about P&R’s GST modelling?

Deputies in the dark about the numbers

No current or former P&R member has been given access to the underlying modelling used to calculate the impact of the tax reforms, P&R President Deputy Lindsay de Sausmarez told Express.

That includes former P&R members Deputy Gavin St Pier and jailed former Chief Minister Jonathan Le Tocq, as well as any P&R members from the last term between 2020 and 2025.

In fact, not one deputy has seen it.

Only five civil servants have had access to the model itself, either now or in the past, including the current States’ Senior Economist.

Dorey Financial Modelling has now joined them, with its officers able to access the underlying data because it had already had the legal clearance to see the census data.

So what exactly is in the model – and what have we learned about how those headline GST figures were produced?

Deloitte saw the answers, but not the workings

The five civil servants and Dorey were not the only people to get a look at the modelling’s output.

International accounting firm Deloitte was also given “detailed outputs” to help with its “economic analysis”.

But it did not get access to the underlying model itself.

Deloitte’s work looked at what could happen to consumer behaviour, employment and GDP if taxes changed.

It found that GST could reduce consumption in the short term, while cuts to direct taxes could increase it, particularly where those cuts benefited lower-income households.

The current modelling makes an important assumption about those competing effects.

It assumes they effectively cancel each other out.

So why can’t deputies see it?

One of the main reasons given by P&R was that the data sitting underneath the model is sensitive.

It uses detailed, “pseudonymised” information from the States’ Rolling Electronic Census, along with other government data, including tax data.

That means there are no names attached to the records, but in theory it would be possible for someone to work backwards from the data, and work out at least some of the people whose data was being used.

P&R said the sheer amount of information could potentially allow some households or individuals to be identified where their circumstances were unusual, especially given the small size of the Bailiwick.

It gave examples including households with large numbers of children, people living in shared hotel staff accommodation and residents of Alderney.

Access is therefore restricted under the Electronic Census (Guernsey) Ordinance 2013, P&R said.

Only people who have been sworn in as census officers and have a “defined and justifiable” reason to access the information can see it.

That is why P&R said deputies have only ever been given the outputs from the modelling, rather than the model itself.

What’s actually in the model?

The model draws on the Rolling Electronic Census, the Corporate Address File and information held in various States administrative systems.

That includes data on:

  • employment and taxable income;
  • pensions and pension contributions;
  • benefits received;
  • mortgage and other interest;
  • age and sex;
  • island and parish;
  • property value and number of bedrooms; and
  • housing tenure.

The data is pseudonymised before being included in the model.

But because it is being used to build detailed profiles of individuals and households, P&R said some people could still potentially be identified from unusual combinations of information.

The model then uses household spending patterns to work out how much GST could be raised.

The spending data is 7 years out of date (still)

The modelling currently being used by P&R relies on household spending patterns from the 2018-19 Household Expenditure Survey – something we already knew.

That means the spending data is now around seven years old – and importantly from before the Covid pandemic.

P&R has now said it was carrying out “significant work” to update the modelling to use the newer 2023-24 Household Expenditure Survey, which was published in July, but this hadn’t been completed yet.

However, it said the updated output is due to be published before the tax reform debate resumes – in two weeks.

That matters because those spending patterns are used to estimate how much GST the island could raise.

It also matters because the deadline for amendments is 22 September, giving deputies little time – if any – to come up with any alternative ideas based on up-to-date modelling.

The spending patters currently being relied on therefore pre-date the pandemic, the cost-of-living crisis and the huge changes in household spending and prices that followed.

The updated modelling based on the 2023-24 survey should give deputies a much more recent picture.

But they still won’t get to see the workings behind it.

There are some big assumptions

This is where some of the more interesting answers emerged.

P&R assumes GST will be passed on in full

For goods and services that are subject to GST, the model assumes businesses pass the cost on to customers in full.

Anything that’s exempt from GST is treated differently, because businesses cannot necessarily recover all of the GST they pay on their own costs.

P&R said the model therefore assumes there will be some net price impact, depending on the nature of the service.

Wage inflation isn’t modelled

The modelling does not incorporate wage inflation, including public service wages.

P&R said wages are ultimately negotiated between employers and employees.

It also argued that changes to income tax and social security could affect those negotiations.

Bad debt is within the margin of error

The potential impact of bad debt was considered, but was not modelled separately.

P&R said it was treated as falling within the margin of error of the calculations.

Tax cuts and GST are assumed to cancel each other out

This is one of the more significant assumptions.

The modelling assumes the opposing effects of GST and reductions in income tax and social security contributions on consumption are neutral.

In simple terms, the model assumes the reduction in direct taxation offsets the reduction in spending caused by GST.

Benefits are expected to rise with inflation

The impact analysis also factors in the existing policy of increasing benefits in line with inflation.

P&R said the resulting increase in income support spending would largely be offset by lower income support costs because people would have more take-home pay after the tax changes.

What did we ask – and was it answered?

We asked P&R a long list of questions on 27 July.

Seven weeks later we got an answer.

Here’s what the committee’s President said – and what she didn’t.

On access to the modelling

  • How many States officers have had access to the underlying modelling used to produce the GST estimates? (I realise you won’t be able to release names)

    ✅ Answered: “There are five officers who either currently have or previously had access to underlying modelling used to inform the Tax Reform proposals.”

  • How many deputies have seen the underlying modelling?

    ✅ Answered: “None: the model is only accessible to a limited number of people sworn in as census officers.

    “The modelling builds a bottom-up profile of Guernsey households using detailed income and expenditure profiles in order to calculate and map both their current and expected tax liabilities.

    • It includes a significant amount of pseudonymised data. While there are no direct personal identifiers, the extent of the data incorporated means that is it possible to isolate and identify some individual households -for example where they have a large number of children, are resident in scale shared accommodation (such as hotel staff accommodations), or are resident in Alderney. Because of the extent of the personal data included access to it is strictly controlled under the Electronic Census (Guernsey) Ordinance (2013) and access is limited to census officers with a defined and justifiable purpose for accessing and processing the data.

    • States Members have only ever had access to the outputs of the model, not the model itself which is only accessible to people sworn in as census officers.”


  • Has every current member of P&R seen it?
    ✅ Answered: “No. As above, the underlying modelling is only accessible under law to those sworn in as census officers, which States Members are not. For the avoidance of doubt, no member of P&R has seen the modelling.”

  • Has every former member of P&R during this political term (Deputies Jonathan Le Tocq and Gavin St Pier) seen it?

    ✅ Answered: “No: no former member of P&R has had access to the model itself, either this term or previously. States Members have only ever had access to the outputs of the model.”

  • Have any deputies outside P&R been given access to the modelling? If so, who?

    ✅ Answered: “No: States Members have only ever had access to the outputs of the model.”

  • Has anyone outside the States (consultants, advisers, external organisations etc.) had access to the modelling? If so, who?

    ✅ Answered: “Deloitte were provided with detailed outputs of the modelling (but not the model itself) to undertake economic analysis and were required to comply with strict data sharing protocols covering the handling of that data in accordance with the Rolling Electronic Census Ordinance.

    “Dorey Financial Modelling work with Rolling Electronic Census Data to support the production of demographic and economic forecasts. Their officers are sworn in as census officers and must comply with the same restrictions on the processing and management of that data as States of Guernsey Officers.”

On the Household Expenditure Survey and underlying data

  • How many households from the 2018-19 Household Expenditure Survey are represented in the modelling?

    ✅ Answered: “None. It is not used in this way. The Household Expenditure Survey results are used to derive a profile of spending (as a percentage of income) on different goods and services which is then applied to the individual and household data drawn and developed from the Rolling Electronic Census.”

  • We understand Deputies Camp, Curgenven and Collins were told they could view part of the spreadsheet, which contained around 66,000 rows of data. What does each row represent?

    ✅ Answered: “These deputies were shown a small screenshot of the spreadsheet with potential personal identifiers redacted to illustrate how income data was broken down.

    “The 66,000 rows in the data set represent each individual in Rolling Electronic Census data set (essentially each individual in the population) identified as on-island on the snapshot date. These are then condensed into approximately 26,000 households.”

  • Besides the Household Expenditure Survey, what other datasets are used within the model?

    ✅ Answered: “The data set is primarily drawn from the Rolling Electronic Census and the Corporate Address File. It includes data presented on an individual level drawn from States administrative systems, including the revenue service and benefit systems. This includes:

    • An anonymised personal identifier
    • Age
    • Sex
    • Island of residence
    • Parish of residence
    • Flags identifying:
      • the individual as on-island
      • the individual as alive
      • whether they have immigrated
      • whether they have emigrated
      • whether their household has been resident at the same address with the same group of other individuals for 12 months
    • Employment status
    • A breakdown of taxable income including:
      • Employment income
      • Self-employed/business income
      • Investments/dividend/bank interest
      • Rental income
      • Maintenance income
      • Loan interest income
      • Pensions/annuity income
      • Other income
    • Pensions contributions
    • Mortgage interest
    • A breakdown of benefit receipt including
      • States Pension payment
      • Income support payments
      • Unemployment benefit
      • Family allowance
      • Carers allowance/severe disability benefit
      • Unemployment benefit
      • Invalidity benefit
      • Sickness benefit
      • Other benefit payments
    • An anonymised address point identifier (from which the model maps people living at the same address)
    • The TRP value of that property and number of bedrooms
    • The housing market category of that property
    • The tenure of that property”

  • Which of those datasets contain personal information, and which are already anonymised before being incorporated into the model?

    ✅ Answered: “The Rolling Electronic Census contains personal information. While all data sets are pseudonymised before inclusion in the model, the extent of the data on individuals and its treatment (which condenses individual data into households) means that it is possible to identify individuals and households from the data set, particularly where circumstances are unusual.”

On the modelling itself

  • Is the modelling contained within a single spreadsheet, multiple spreadsheets, or another type of model?

    ✅ Answered: “Two models exist: the original model, built in an excel spreadsheet, and a more recent version operated in Power BI.”

  • What has actually been modelled as part of the GST work? For example, have you modelled:

    ✅ Answered: The response addressed each of the examples we listed, as follows.

    • changes in consumer spending behaviour: “economic analysis undertaken by Deloitte in 2021/22 considered the impact that various changes in taxation might have on:
      • consumer behaviour
      • employment
      • GDP
    • “Their conclusion was that a GST would reduce consumption in the short term, and that a reduction in direct taxes might increase consumption. Their analysis suggested that applied in combination, this could result in a slight increase in consumption, particularly if reductions in direct taxation are focused on those at the lower end of the income scale”

    • business behavioural changes: “The model under consideration deals with the application of taxes as applied to individuals and households. It builds a distribution of consumer spending from which aggregate GST revenue can be drawn. It makes an implicit assumption that businesses will pass the GST application to standard rated goods on to customers in full, and that there is some net price impact on exempt goods depending on the nature of the services provided.
    • “Calculations on corporate tax receipts (undertaken separately) do incorporate assumptions around behavioural changes for businesses”

    • bad debt: “the potential for bad debt is considered within the margin of error on the calculations”

    • inflation: “Inflation impacts have been estimated directly on the basket of goods used for the quarterly calculation of RPIX and RPI data”

    • wage inflation: “The setting of wages is a matter of negotiation between employers and employees and is not incorporated in the modelling.
    • The reduction in income tax and social security liabilities for individuals, resulting in them keeping more of their pay packet, is also likely to have a bearing on wage negotiations.”

    • changes in benefit claims: “There is a standard policy of increasing benefits in line with inflation and this is factored into impact analysis. It is worth noting that any increase to income support is largely offset by the reduction in spend on income support because of the increase in take-home pay thanks to the tax restructure.”

    • any increase in consumption resulting from reductions in income tax and social security contributions? “As described, the opposing effects on consumption of reductions in income tax and social security and the introduction of a tax on consumption are assumed to be neutral.”

  • If not, which of these have instead been held constant or assumed?

    🟠 Partially answered: The response contains assumptions about several of the factors listed above, but does not provide a separate answer identifying which factors have been held constant or assumed.

On releasing the model

  • If the underlying data cannot be released for confidentiality reasons, are you able to publish the spreadsheet structure, formulae, assumptions and calculation methodology with the data removed or replaced with dummy values?

    ❌ Not answered.

  • Has a formal assessment been carried out into whether an anonymised or redacted version of the model could be released? If so, when was this done and by whom?

    ❌ Not answered.

  • Is there any mechanism by which an independent third party could reproduce the GST revenue estimates using the information that has been published?

    🟠 Partially answered: The response did not directly address whether a third party could reproduce the estimates.

    Instead, it said: “The Policy & Resources Committee has agreed to a third-party validation of the modelling by a supplier who is already legally authorised to handle census data, is familiar with its structure and has suitable storage and processes in place to do so.”

  • Have any deputies outside P&R been able to independently verify the calculations underpinning the projected GST revenues and net fiscal benefit?

    🟠 Not answered, but implied.

    The response said: “The Policy & Resources Committee has agreed to a third-party validation of the modelling by a supplier who is already legally authorised to handle census data, is familiar with its structure and has suitable storage and processes in place to do so.”

    That does not answer whether any deputies have independently verified the calculations, however we can reasonably imply that no deputies have independently verified the calculations, given they were not given access to the models or the underlying data.

Finally

  • Will P&R be producing an updated model now the 2023-24 Household Expenditure Survey results have been released? If so, when is that expected to happen?

    ✅ Answered: “Yes: the intention is to update the modelling with the updated household expenditure data. This is a significant piece of work that has been carried out in recent weeks, but the Committee will release the output from it prior to the continuation of the tax reform debate.”