Model behaviour: How do you work out the fallout from a new tax?

Guernsey’s top political committee wants to introduce a range of tax reforms, including a 3% GST, to help plug the island’s financial black hole of more than £50m.

As you’d expect with any new tax, not everyone is happy about it – but one of the biggest political rows has been around the sums behind the proposals.

After months of asking, Policy and Resources (P&R) has finally told Express a bit more about its modelling.

We now know it was put together by just five civil servants, with international accounting firm Deloitte providing some additional “economic analysis”.

But P&R has repeatedly refused to release the modelling – or even the equations it’s using – to journalists or elected deputies, claiming the data the models use is too sensitive to share.

So, in the absence of the models themselves, Express decided to ask an expert how governments typically work out the knock on effects of a new tax.

Microsimulation and macroeconomics

Martin Mikloš knows a thing or two about modelling the economic impact of tax reform.

Mr Mikloš is a Research Economist at the Centre for Tax Analysis in Developing Countries (TaxDev) with the Institute of Fiscal Studies (IFS), an independent economics research institute described as the “most influential voice in the UK’s economic debate” by the Guardian.

He was previously a Tax Policy Analyst for Slovakia’s Finance Ministry and has a degree in Philosophy, Politics and Economics from Oxford University and a Masters in Economics from Cambridge.

He explained to us that governments typically model the “shock” of a new tax using a combination of bottom-up “microsimulation” and “macroeconomic modelling” – a “picture of the entire economy”.

Looking at individual households

Mr Mikloš said: “In the microsimulation model, you calculate how much each individual household pays in GST as a result of that policy.”

Microsimulation was a “bottom-up approach” that takes information on individuals and households and “calculates the impact of different policy changes” on them.

This gives economists a “distributional analysis” showing the income and spending patterns of different households, and helping to predict how the proposed tax change would affect them.

The assumptions that go into your model matter a lot.

Martin Mikloš, Research Economist

“So poorer households will pay this. Richer households will pay that,” Mr Mikloš added.

Using large datasets for this kind of modelling is not unusual, he said.

The UK’s revenue service, HMRC, for example, uses data from individual taxpayers when modelling income tax, while household spending is often modelled using survey data because governments generally don’t hold detailed records of everything people buy.

While there were a range of inputs and a “lot of assumptions you need to make along the way”, a microsimulation would then spit out “one number”, he explained.

“The assumptions that go into it matter a lot.”

Economic shock

But working out how a new tax would affect individual households and understanding the effect it could have on the wider economy are different.

To understand the “shock” a new tax would have on things like inflation, GDP, growth and employment needed a separate “macroeconomic model”.

“The microsimulation model will basically tell you if you’re paying X in taxes today, and I change the rates tomorrow, you will be paying X plus F in taxes,” he said.

Then it looked at “how that affects inflation and all these things down the line”.

“You basically need to build on top of the microsimulation model using some sort of macroeconomic modelling,” Mr Mikloš explained.

A range of possibilities

When it comes to macroeconomics, the key is to consider a range of possibilities and check your model can deal with them, Mr Mikloš said.

“You might want to model the impacts using different assumptions, different parameters, and see how sensitive your results are to different assumptions.”

Mr Mikloš said macroeconomic models were “inherently uncertain” so there was a range of “confidence intervals”.

The important thing, he said, was that your model should be able to cope with that uncertainty.

“If you look at charts by the Office for Budget Responsibility (OBR) or the Bank of England, they usually produce these fan charts.”

The idea is that there is a central estimate, but a widening range of possible outcomes around it.

Rather than one number – whether for inflation, employment or growth – a macroeconomic model should produce a series of ‘most likely’ numbers with a margin of error.

The past is the key to the future

So how do macroeconomic models predict the future – even if it’s only a range?

“The relationships are derived from these historical relationships,” Mr Mikloš said.

He gave the example of comparing an economic “shock” from changing exchange rates in the 1980s, with ones in the 1990s and 2010s to understand how the “relationships had changed”.

While you might find one event had a 0.5% impact and another had 2%, you would “average across these shocks, and that’s why you get a range of options”.

“So basically, if you ask the question, ‘What is the relationship between an exchange rate and inflation?’ you will get an average, but there is a confidence interval around that because you observe some different outcomes over time.”

What does this mean for Guernsey?

Mr Mikloš was talking to us about ‘best practice’ for modelling major economic changes, not specifically about GST and the rest of Guernsey’s proposed tax reforms.

But it does give us a clearer picture of the kind of projections governments typically produce when they model tax changes.

At a household level, microsimulations can estimate who pays more, who pays less and how much revenue a new tax could raise.

Then there’s the effect of the changes on the wider economy and the potential effects on things like inflation, employment and economic growth.

For those, economists create a separate macroeconomic model which takes the “single number” the microsimulation spits out and produces a range of possibilities, based on the different assumptions.

As those wider estimates are inherently uncertain, different assumptions can produce different results.

That leaves some fairly important questions about Guernsey’s own modelling.

1. How does the States’ models work?

The first – and probably most important – question is ‘Did Guernsey use a similar approach?’

We don’t know.

We’ve asked P&R repeatedly to explain how its models work and if it can share the equations or formula the civil servants used, even if it can’t release the underlying data. The committee has repeatedly ignored the question.

What now know five civil servants produced a model using Microsoft’s Power BI (a more-powerful alternative to Excel) using some data from the rolling e-census, a seven-year-old household spending survey and some tax return data.

We also know Deloitte did some “economic analysis” using the “detailed outputs”.

Is that consistent with a microsimulation and macroeconomic modelling approach?

Possibly, though the fact P&R talked about a series of “detailed outputs” rather than the single number Mr Mikloš described could suggest otherwise.

Deloitte’s analysis considered the effects of GST on “consumer behaviour, employment and GDP”, P&R President Deputy Lindsay de Sausmarez told Express.

That could, potentially, fit Mr Mikloš’ description of a macroeconomic model.

But if it was created in 2021 and 2022, then it would predate P&R’s current proposals and even the previous GST+ proposals (with GST at 5%), which first emerged at the end of 2024.

Does that mean the States hasn’t done any macroeconomic modelling on GST+ or P&R’s latest 3% ‘GST with mitigations’ package?

We simply can’t tell.

2. Where’s the ‘fan diagram’?

One of the biggest things that jumped out from our conversation with Mr Mikloš was his description of a “range” of possibilities around a “central estimate”.

That’s not what we’ve been given by P&R.

We’ve been told the States expects 3% GST to increase inflation by 1.9% – but that’s one number, not a range.

It might be that the ‘Famous Five’ civil servants produced a range of numbers but they – or P&R – decided to publish just the 1.9% number.

If so, does that create a false sense of certainty? And where’s the fan diagram?

And if not, why didn’t they model a range of outcomes?

3. What assumptions were made and were alternative scenarios considered?

Mr Mikloš said there were lots of “assumptions you need to make” which could significantly affect the results.

Because there was “no one right answer”, economists would often test a range of “behavioural assumptions”.

“You might want to model the impacts using different assumptions, different parameters, and you [want] to see how sensitive your results are to different assumptions,” he told us.

Because P&R won’t share its models or explain how they work, we don’t know what assumptions the five civil servants made or whether they tested alternative scenarios.

And that matters, because – even if P&R’s modelling did follow a standard methodology like the one Mr Mikloš described – if the civil servants made assumptions that differ too much from reality and didn’t test alternative scenarios their projections could be off – possibly by a little, possibly a lot.

We also don’t know if the civil servants carried out any sensitivity analysis, to see how well the models coped with uncertainty around the inputs.

That’s why so many people including deputies, members of the public and this publication have been calling for P&R to release as much as it can of its modelling.

Because even well-meaning, honest, intelligent people could have made flawed assumptions that make a big difference to our understanding the effect of GST and the other tax proposals.

And if all the assumptions are reasonable and the models are robust, then we’ll find out.

With the pressure around the modelling growing louder, many will be asking one question: ‘What has P&R got to lose by releasing the methodology behind its modelling?’

Related