Guernsey Tax Reform: Why Every Option Should Face the Same Test

Before Guernsey Settles GST, Shouldn’t Every Serious Option Face the Same Test?

Imagine Guernsey needed a new bridge.

Several engineers submit designs. One team is given access to all the geological surveys, traffic data, engineering software and technical expertise. Its bridge is stress-tested, modelled and refined.

The other teams are given only the information available to the public.

Then we ask which bridge is safest.

Whatever the answer, most people would probably agree that wasn't a particularly fair test.

That, in simplified terms, is the problem I think is worth considering in Guernsey's tax debate.

This isn't an argument that GST is wrong.

It isn't an argument that one of the alternatives is right.

It's an argument for something much simpler:

If we're going to compare them, shouldn't we test them all by the same rules?

We certainly aren't short of ideas

The Tax Reform debate has become enormous.

As of 29 September, the official States Tax Reform page contains numbered amendments all the way through to Amendment 57, although some have already been carried, lost, not laid or cover overlapping issues.

Among the ideas raised during the process are different forms of GST, income-tax alternatives, a common tax-rate framework, changes to property taxation, wealth and asset taxation, visitor-related taxation, spending controls and various targeted measures.

Some might work.

Some might not.

But surely that's what proper modelling is supposed to establish.

The problem is that everyone isn't using the same laboratory

P&R has access to an extremely detailed household model built using States information.

There is a perfectly legitimate reason why that model can't simply be published.

P&R has explained that it uses pseudonymised information from the Rolling Electronic Census and other sources, but because Guernsey is such a small population and the information is so detailed, unusual combinations of circumstances could potentially allow individuals or households to be identified.

That's a genuine privacy issue.

But it creates another problem.

If somebody proposes an alternative tax structure, they can't simply download the same data, put their proposal through it and demonstrate what happens.

So we risk comparing a fully modelled proposal with alternatives that haven't been given access to the same laboratory.

That doesn't prove the alternatives are better.

It doesn't prove they're worse either.

It means we don't necessarily know.

Recent events show exactly why modelling matters

P&R originally estimated that its reform package would generate around £39 million a year in net additional revenue.

After incorporating newer household and visitor expenditure information, that estimate fell to around £36 million.

The estimated GST raised from household spending fell from £41 million to £38.3 million.

The visitor element fell from £4.7 million to £3.5 million.

That doesn't mean modelling is useless.

Quite the opposite.

It shows why modelling needs to be constantly tested against better and more recent evidence.

Change the data and the answer can change.

Change the assumptions and the answer can change.

Change people's behaviour and the answer may change again.

That matters when we're talking about restructuring Guernsey's tax system.

The independent audit tells us something important too

Dorey Financial Modelling has now independently reviewed P&R's tax model.

There is some genuinely positive news in that report.

Dorey independently recreated the calculations across 66,021 individuals and 25,529 households and found that the calculation engine matched the States' model.

It also described Guernsey's full-population approach as sophisticated, because our Rolling Electronic Census allows modelling across the population rather than relying purely on a sample.

But this is where things get interesting.

The audit assessed 17 areas.

Twelve were rated green.

Four were amber.

One was red.

The red issue was tax collection.

The published modelling effectively assumes that all GST due is collected. Dorey noted that no tax is collected perfectly and suggested that a normal collection gap of around 3% to 5% should be anticipated.

That alone could reduce expected GST receipts by roughly £1 million to £2 million.

When behavioural effects and a possible collection gap were included, Dorey suggested household GST could have a planning floor of around £33.5 million.

This is a really important distinction.

A calculator can be mathematically correct while the eventual answer still depends heavily on the assumptions fed into it.

That's not a criticism of modelling.

That's how modelling works.

And it's exactly why the assumptions should be visible and why competing options should be tested consistently.

So why not create one common testing ground?

Keep the real Revenue Service, census and household information exactly where it is.

Secure.

Confidential.

Accessible only to people legally permitted to work with it.

Use that secure model to produce the authoritative results.

But alongside it, create something else.

A synthetic Guernsey.

That phrase probably sounds more futuristic than it really is.

It doesn't mean asking AI to invent random islanders.

Synthetic data uses statistical modelling to create artificial records which reproduce useful characteristics of a real population without those records representing actual identifiable people.

The UK Office for National Statistics already has a specific policy covering this.

ONS says synthetic data can be useful for testing and research when real information can't safely be shared, while also making clear that synthetic data must be validated and checked carefully for disclosure risk.

Imagine a model containing tens of thousands of artificial Guernsey households.

One might represent a retired couple.

Another might be a single parent.

Another could be two working parents with two children and a mortgage.

Another could represent someone renting on a lower income.

None of them would be real people.

But collectively they could be calibrated so that the synthetic population behaves statistically like Guernsey.

Then give every proposal the same exam paper

This is the bit I think really matters.

Take exactly the same population and run it through the current system.

Then run the same population through the proposed GST package.

Then an income-tax alternative.

Then a common tax-rate structure.

Then different GST exemptions.

Then property or asset-based options.

Then targeted measures or combinations of measures.

And anything else that's sufficiently developed to be modelled properly.

Use the same assumptions every time.

Same population.
Same starting year.
Same economic forecast.
Same inflation assumptions.
Same visitor assumptions.
Same behavioural assumptions.
Same method for calculating administration costs.

And most importantly:

Give them the same financial target.

Because comparing different targets isn't really a comparison

Suppose Guernsey decides it needs another £40 million.

There's very little value in comparing one proposal designed to raise £40 million with another currently designed to raise £20 million and concluding that the first raises more money.

Of course it does.

The useful question is:

What would each tax structure have to look like if it had to raise the same amount?

Now we're comparing like with like.

If an income-tax option needs a particular rate to achieve the target, show us.

If a property-based option would require unrealistic rates, show us.

If targeted levies can't raise enough, show us.

If GST raises the required revenue more reliably, show us.

If a combination of measures produces similar revenue with different distributional effects, show us that too.

No slogans.

No guesswork.

Just put them through the same test.

Then show people what actually happens

For every option, publish the answers to the questions people genuinely care about.

How much does it raise?
What does it cost to introduce?
What does it cost to administer every year?
What happens to lower-income households?
What happens to middle-income households?
What happens to higher earners?
What happens to families with children?
What happens to pensioners?
What happens to renters and homeowners?
What happens to businesses?
What happens if consumer spending falls?
What happens if tourism changes?
What happens if the population gets older or smaller?
What happens if projected States savings aren't fully achieved?
How sensitive is the result to the assumptions being used?

And instead of simply saying:

"This will raise £X million."

tell us something more useful:

"Under our central assumptions the estimate is £X, but under reasonable alternative assumptions the likely range is between £Y and £Z."

That's much closer to how uncertainty actually works.

Interestingly, the States has already accepted part of this principle

Amendment 24 has already been carried.

It requires P&R, when developing the Child Responsibility Tax Allowance, to undertake fiscal and distributional analysis using Revenue Service data, including how the allowance interacts with the wider tax reform package.

Amendment 25 was also carried.

That puts the development and examination of a single common tax-rate framework into the longer-term work on Guernsey's tax system.

So we're already accepting that alternatives and changes need detailed modelling.

The question is whether we can take the next logical step and put them into a common framework.

A fair test should be capable of proving an alternative wrong

This is important.

A comparative model shouldn't exist to prove GST wrong.

It shouldn't exist to prove GST right either.

It should make every proposal prove itself.

Someone proposes an alternative to GST?

Put it through the model.

If the revenue doesn't stack up, we learn something useful.

If its administration costs are huge, we learn something useful.

If it produces unexpected consequences for particular households or businesses, we learn something useful.

If it works broadly as claimed, we learn something useful too.

The exact same standard should apply to GST.

Think of it like an exam

You wouldn't give one student the full textbook, all the revision notes and the questions in advance, then give everybody else a summary sheet and compare their marks.

You'd give everyone the same exam paper.

That's essentially what I'm suggesting here.

Same data. Same assumptions. Same financial target. Same test.

Then compare the results.

It still wouldn't tell deputies how to vote

No computer model can decide what Guernsey should value.

It can't tell us whether consumption, income, property or wealth should carry more of the tax burden.

It can't decide what level of public spending is desirable.

It can't define fairness for us.

Those are political judgements, and elected deputies ultimately have to make them.

But a good model can help establish the facts underneath those judgements.

It can estimate who pays.

It can estimate how much.

It can show which assumptions the answer depends upon.

It can identify uncertainty.

It can expose weaknesses in proposals from any side.

And it can test whether claims actually stack up.

Maybe that's the common ground we've been missing

Guernsey has spent years arguing about tax reform.

We now have dozens of amendments, competing calculations, strongly held positions and a great deal of public scepticism.

Another spreadsheet from one side probably isn't going to settle that.

Another Facebook argument certainly won't.

But perhaps a common model could at least give everyone the same starting point.

Not because models are perfect.

They're not.

Not because synthetic data can magically predict the future.

It can't.

But because a fair comparison has one very basic requirement:

Comparable evidence.

You wouldn't test one bridge and simply guess whether the others would stay standing.

You wouldn't give people different exam papers and compare their marks.

And you wouldn't call a race fair if everybody started from a different line.

Whatever tax system Guernsey eventually chooses, there is surely a worthwhile question to answer first:

Can the serious options be put on the same starting line, using the same data, the same assumptions and the same financial target?

If they can, then perhaps we can spend less time arguing over whose spreadsheet is right and concentrate on the question that genuinely belongs to our elected deputies:

Given the same evidence, what kind of tax system does Guernsey want?

Fact checked: 29 September 2026. Sources consulted include the States of Guernsey Tax Reform 2026 proposition and amendment records, the official States voting record, P&R's updated modelling reported on 23 September, the Dorey Financial Modelling external validation reported on 25 September, and the Office for National Statistics Synthetic Data Policy.

#GST