Research and analysis

Evaluation of the change to UK Deemed domicile policy 2017

Published 30 October 2024

1. Executive summary

In 2017 the UK Government introduced reforms which meant that some non-domiciled taxpayers were to be deemed as domiciled in the UK through either of 2 routes:

• being born in the UK with a domicile of origin of the UK and resident in the UK in tax year 2017 to 2018 or later (condition A)
• being UK-resident for at least 15 of the 20 tax years immediately before the relevant tax year (condition B)

This brought more people into scope of UK taxation on their income and capital gains. When the policy was announced the Office for Budget Responsibility (OBR) expected it to raise around £300 million per year, with around 3,000 people becoming UK domiciled. Using a range of evaluation methods, this paper reports quantitative estimates of the impact of the measure, drawing on administrative tax data from HM Revenue and Customs (HMRC). The high-level results are as follows:

1.1 Exchequer revenue

The 2017 reforms led to more tax being paid by deemed domiciled taxpayers than originally expected:

• those deemed domiciled are now contributing more than £3 billion of Income Tax, Capital Gains Tax and National Insurance contributions each year with additional tax collected of over £700 million in tax year ending 2018 and over £1 billion in tax year ending 2020 and this increase in tax is observed specifically for the condition B deemed domiciled taxpayers who were previously on the remittance basis
• condition B deemed domiciled taxpayers have paid more tax on average since they became deemed with average tax paid roughly doubling in the first year and rising further in the third year
• the increase in tax paid for deemed domiciled taxpayers has partly been driven by those who had historically been higher taxpayers and they are paying even higher tax after the policy change

1.2 Mobility

The majority of those affected by the deemed domicile reforms remained in the United Kingdom:

• more than 9,800 (perhaps as many as 15,600) non-domiciled taxpayers have remained in the UK after becoming deemed and their numbers have increased each year since the policy change
• after the policy change around 17,000 new non-domiciled taxpayers who claim the remittance basis are continuing to come to the UK each year
• those formerly UK-domiciled taxpayers who had used the non-domicile regime but become deemed (condition A) were 27% more likely to remain within the UK since the policy change
• non-domiciled taxpayers on the remittance basis were 4% to 5% more likely to leave the UK after the policy change
• deemed domiciled (condition B) taxpayers were 10% to 12% more likely to leave the UK after the policy change (in the tax year ending 2018) compared with before the change (in the tax year ending 2015)

Our findings are consistent with academic literature which shows that tax rates are just one of a broad range of factors that influence mobility of high-income individuals such as family and employment ties, work prospects and cultural ties (see Annex 1).

2. Introduction

This report evaluates the impact of a 2017 tax policy change affecting non-domiciled taxpayers using a range of analytical methods. Its purpose is to provide quantitative evidence of the impact of the 2017 reforms to guide future policy.

It is worth noting that an individual’s domicile is distinct from their nationality or country of residence and is broadly defined as the territory that ‘on the balance of probabilities’ is an individual’s permanent home. Establishing domicile is not straightforward and can involve a detailed examination of an individual’s background, lifestyle, habits, and intentions over their life.

2.1 UK system prior to 2017 reforms

An individual newly resident but not domiciled in the UK paid tax on income and capital gains which had arisen in the UK. For income and gains outside the UK they could either pay UK tax on all their foreign income and capital gains (the arising basis of taxation), or they could pay UK tax on their new foreign income and capital gains only when it was brought into the UK (the remittance basis of taxation).

If an individual chooses the remittance basis and has been resident in the UK for longer than 7 years, they must pay a remittance basis charge. These charges have varied since their introduction and are summarised in Figure 1:

2.2 Remittance basis charge changes

Figure 1: Remittance basis charge changes

Date	Remittance basis charge(s)	Residency Length
April 2008	£30,000	UK resident in at least 7 of the previous 9 UK tax years.
April 2012	£30,000, £50,000	7 of the 9 UK tax years, UK resident in at least 12 out of the preceding 14 UK tax years.
April 2015	£30,000, £60,000, £90,000	7 of the 9 UK tax years, 12 out of 14 UK tax years, UK resident in at least 17 of the preceding 20 UK tax years.
April 2017	£30,000, £60,000	7 of the 9 UK tax years, 12 out of 14 UK tax years (Deemed domicile policy in effect after 15 of 20 tax years so £90,000 charge scrapped)

If an individual’s unremitted foreign income and gains in the tax year were less than £2,000, they could use the remittance basis without completing a Self Assessment tax return and not have to pay the remittance basis charge.

2.3 UK system after the 2017 reforms

After the deemed domicile policy change began on 6 April 2017, long-term UK residents meeting either of the following 2 conditions could no longer access the remittance basis of taxation, meaning that they would pay UK tax on their worldwide income and gains:

• those born in the UK with a domicile of origin of the UK and resident in the UK in 2017 to 2018 or later (condition A)
• those UK resident for at least 15 of the 20 tax years immediately before the relevant tax year (condition B)

The policy change included the following ‘transitional protections’ to ease the transition temporarily. Condition B individuals could use these to protect some of their foreign income and gains from UK tax:

1. A 2-year period from the tax year ending 2017 to tax year ending 2019 in which individuals were able to segregate their accounts, separating their ‘prior to their UK residence’ funds (termed ‘clean capital’) from their post-UK residence accounts; these ‘prior to UK residence’ finances can be brought into the UK without paying additional tax.

2. For individuals who became deemed domiciled in the tax year ending 2018, there was an optional revaluation of foreign assets to their value on 5 April 2017. This means that if those individuals were to sell those assets while subject to UK tax, any capital gains would be calculated using this value rather than the value at which they were originally bought.

3. Those who set up non-resident trusts before becoming deemed domiciled would not pay UK tax on the income and gains of those trusts if that income and gains remained within the structure.

2.4 Estimates of impact

In July 2015, HMRC published a costing note for the deemed domicile policy change. The forecast estimated that around 3,000 people would become UK domiciled from April 2017, and in the following years raise around £300 million per year of additional revenue. The modelling assumed a loss of revenue from some of the individuals affected, more than offset by a gain in revenue from those newly eligible and those that chose to stay.

2.5 Evaluation of 2017 Reforms

There are 2 stages to this evaluation. In Section 1, we describe the tax, National Insurance contributions and behavioural changes of individuals affected by the policy change. This includes tracking sub-populations of the non-domiciled taxpayers and looking at their behaviour each year. We examine the tax and National Insurance contributions (NICs) before and after the policy change.

In Section 2, we use a difference-in-differences regression analysis to attempt to isolate the causal impact of the policy change. We examine the average change in tax paid and the change in tax paid along the distribution of tax paid. We also examine the change in likelihood of remaining in the UK due to the policy change.

We recognise that there were also changes to the Inheritance Tax liabilities for deemed domiciled taxpayers due to this policy change. We do not examine this change in this publication.

2.6 Data

This evaluation draws on data provided by taxpayers in their Self Assessment tax returns. Self Assessment, introduced in tax year ending 1997, is a business system operated by HMRC where individuals, trusts and unincorporated businesses (for example, partnerships) return information on income and capital gains. All non-domiciled taxpayers must complete a Self Assessment return if they meet the criteria.

Taxpayers who are non-domiciled will complete the SA109 form titled ‘Residence, remittance basis etc’. Most information in this publication comes from data that has been provided by taxpayers on this form. We use additional information from the Self Assessment return to determine estimated liabilities for Income Tax, Capital Gains Tax and National Insurance contributions.

The NICs figures are annualised estimates of NICs derived from the Self Assessment annual return data. The NICs estimates are the sum of Class 1 primary, Class 1 secondary, Class 2 and Class 4 contributions. Class 2 and Class 4 NICs are actual annual liabilities directly extracted from Self Assessment. Class 1 primary and Class 1 secondary NICs are modelled annually from Self Assessment income. This may differ from actual NICs due to differences between modelling methods of NICs and the way some classes of National Insurance contributions are charged.

3. Results

Section 1 of the results looks first at the simple top-down evidence from aggregate statistics before tracking individual taxpayers to see what happened to their tax receipts before and after the reform. Section 2 uses regression analysis and difference-in-differences methodology to estimate causal relationships between the policy reform and changes in location choices.

4. Section 1: Descriptive analysis of non-domiciled taxpayer population and tax paid

4.1 Changes in official statistics on non-domiciled taxpayers

‘Non-domiciled taxpayers’ refers to those taxpayers who have informed HMRC that they are domiciled outside the UK and it is relevant to their Income Tax or Capital Gains Tax liability.

HMRC’s annual official statistics on non-domiciled taxpayers showed that the total number of non-domiciled taxpayers fell by 12,000 between the tax years before and after the policy change (tax years ending 2017 and 2018), slightly lower than the 14,000 to 15,000 implied in the OBR’s 2015 costing of the policy change. Over the same period, receipts from Income Tax, Capital Gains Tax and National Insurance Contributions from non-domiciled taxpayers fell by £2 billion (Figure 2), as one would expect with a move from this group. The number of non-domiciled UK residents stabilised at a lower level after the deemed domicile policy change.

4.2 Non-domiciled taxpayer numbers, Income Tax, Capital Gains Tax and National Insurance contributions

Figure 2: Non-domiciled taxpayer numbers, Income Tax, Capital Gains Tax and National Insurance contributions

Over the same period, the number of non-domiciled taxpayers on the remittance basis fell by 8,000 and tax revenue by £1.7 billion. Most of the overall change in non-domiciled taxpayers is explained by the fall in the number of remittance basis taxpayers.

Of course, this is only one half of the story. The expected fall in the number of non-domiciled taxpayers did not result in an overall fall in revenues since the tax received from those becoming deemed domiciled and continuing to pay tax in the UK and the tax received from new non-domiciled taxpayers offset the tax lost from those that no longer pay tax in the UK. This is shown in the next sections.

4.3 Changes in official statistics on deemed domiciled taxpayers

The numbers of deemed domiciled taxpayers declaring on their SA109 form rose in the period after the reform (Figure 3). This suggests that overall, individuals were not materially deterred from UK tax residence by the policy change. Nearly 9 in 10 of these deemed domiciled individuals declared that they were condition B deemed domiciled.

4.4 Number of deemed domiciled taxpayers declared on the SA109 form

Figure 3: Number of deemed domiciled taxpayers declared on the SA109 form

	2017-18	2018-19	2019-20
Condition A	911	931	935
Condition B	7,365	7,990	8,618
Both	182	264	288
Total	8,458	9,185	9,841

Data extracted November 2021.

Figure 4 shows that the Exchequer revenue from these deemed domiciled individuals was over £2 billion in the tax year ending 2018 (the first year of the reforms). This rose to nearly £3 billion in the tax year ending 2020. This more than offsets the fall of £2 billion receipts from those considered non-domiciled (Figure 2).

4.5 Deemed domiciled taxpayer numbers, Income Tax, Capital Gains Tax and National Insurance contributions

Figure 4: Deemed domiciled taxpayer numbers, Income Tax, Capital Gains Tax and National Insurance contributions

The guidance for the SA109 form explains that individuals need to complete the deemed domicile sections of the form if their domicile or deemed domicile status is relevant to their tax affairs.

The status is relevant if the individual is using the transitional protection measures. There may be some deemed domiciled individuals that are no longer filling in the SA109 form as they do not use the transitional protections. This means that these totals of deemed domiciled taxpayers and the revenue may be an undercount.

4.6 Tracking of deemed domiciled taxpayers

In this part of the analysis, we track the Income Tax and Capital Gains Tax paid by the deemed domiciled individuals and the residual non-domiciled taxpayers on the remittance basis before and after the policy change.

We cannot include NICs in this tracking exercise because this data cannot be extracted directly from Self Assessment. This means that here we are understating receipts paid by non-domiciled taxpayers. However, NICs should not be impacted by the deemed domicile reforms as this tax is charged on UK employment income and so is taxed regardless of domicile status. Section 2 considers the net impact of the policy change including foregone remittance basis charges.

We also track taxpayers’ residency and domicile status, and whether they are on the arising or remittance basis. There are 5 status outcomes that individuals can take:

• be on the remittance basis while claiming non-domiciled status
• be on the arising basis while claiming non-domiciled status
• start paying tax in the UK and pay it on the arising basis
• be not UK tax resident
• leave the UK and potentially not pay any tax in the UK; this is the same as being not UK tax resident but also not having any UK income or gains that they need to state on their Self Assessment forms

Figure 5 shows that, overall, taxpayers who became deemed domiciled paid £470 million higher Income Tax and Capital Gains Tax (31%) after the policy change. While Income Tax and Capital Gains Tax payments had generally increased, most of this extra tax (£420 million) was raised from the one-third of deemed domiciles who had previously used the remittance basis. This is to be expected as this group is now taxed on their worldwide income and gains.

This does not include the loss of RBC payments from this group, so we need to net off at least the £222 million RBC payments made by 2,500 individuals paying the £90,000 remittance basis charge in 2017. This leads to an estimate of around £200 million of Exchequer revenue gain in the first year from those formerly using the remittance basis.

4.7 Tracking the status of deemed domiciled taxpayers before the policy change

Figure 5: Tracking the status of deemed domiciled taxpayers before the policy change

Due to the amount of data presented, only part of the table below is visible. Please use the scrollbar at the bottom of the table to view all the columns.

Status before change	Numbers before change	Percentage before change	Tax Paid before change (£million)	Status (after change)	Numbers after change	Tax Paid after change (£million)	Change in tax revenue
Remittance basis	2,869	34%	1,010	Deemed domiciled	2,869	1,430	42%
Arising basis (non-dom)	2,659	31%	290	Deemed domiciled	2,659	310	8%
Arising basis (UK domiciled)	2,121	25%	230	Deemed domiciled	2,121	240	6%
Non-resident	369	4%	2	Deemed domiciled	369	7	293%
Not in the UK	440	5%	0	Deemed domiciled	440	6
Total	8,458		1,530	Total	8,458	2,000	31%

Tax paid by those previously on the arising basis increased by 6% or around £20 million.

The remaining sub-group in Figure 5 (9% of individuals) were either non-resident (if they stated that they were non-resident on their SA109 form) or presumed not in the UK (where we had no Self Assessment record). Non-residents only pay tax on their UK income and gains. Receipts from Income Tax and Capital Gains Tax were also up more modestly by £11 million.

4.8 Tracking of non-domiciled taxpayers on the remittance basis from 2013-14 to 2019-20

We are mainly interested in remittance basis taxpayers because we can track how revenues change within this group when their worldwide income and gains become taxed.

We track the remittance basis population and their Income Tax and Capital Gains Tax from tax year ending 2014 to tax year ending 2019, covering pre-announcement, announcement and policy starting. Deemed domiciled taxpayers are included in this analysis.

4.9 Numbers and tax paid by the remittance basis population from tax year ending 2014 to tax year ending 2019

Figure 6: Numbers and tax paid by the remittance basis population from tax year ending 2014 to tax year ending 2019

Looking at the overall stock in each year, Figure 6 illustrates that the number of remittance basis non-domiciled taxpayers remained broadly constant until the policy change and then the population restabilised after the policy change. The Income Tax and Capital Gains Tax paid by these remittance basis taxpayers increased prior to the policy change and fell to £4.5 billion in 2018, reflecting the decrease in population.

Figure 6 also shows that the proportional decrease in population was smaller than the decrease in tax paid. The average tax paid by remittance basis taxpayers peaked in 2017 but remained at around £100,000 both before and after that point.

4.10 Index of numbers and Income Tax and Capital Gains Tax paid of the new remittance basis taxpayers from the tax year ending 2015 to the tax year ending 2019 (2015=100)

Figure 7: Index of numbers and Income Tax and Capital Gains Tax paid of the new remittance basis taxpayers from the tax year ending 2015 to the tax year ending 2019 (2015=100)

Looking at the flow of new entrants each year, Figure 7 shows that the number of new non-domiciled taxpayers opting into the remittance basis each year fell slightly after the policy change. By 2019 it had dropped to 87% of the 2015 level, representing a drop from 19,500 to 17,000. The average tax paid by the new non-domiciled taxpayers who opted into the remittance basis increased after the policy change.

Figure 8 shows year-on-year behavioural changes of the remittance basis taxpayer population over time. It shows that year-on-year transitions in statuses remained broadly constant except for the year of the policy change (tax year ending 2018).

This analysis only considers those arising basis users that have returned domicile information on a SA109 form. It is likely that there are arising basis users that are not returning this information if they do not consider it relevant to their tax return. For example, they may consider that they are following the tax rules as though they are common law UK-domiciled.

4.11 Tracking the percentage of the remittance basis taxpayers undertaking year-on-year behavioural changes from the tax year ending 2015 to tax year ending 2020

Figure 8: Tracking the percentage of the remittance basis taxpayers undertaking year-on-year behavioural changes from the tax year ending 2015 to tax year ending 2020

In the year of the policy change, Figure 8 shows that the percentage remaining on the remittance basis dipped and a larger percentage moved onto the arising basis (taxed on their UK and worldwide income and gains).

Approximately 15% of the remittance basis taxpayers move onto the arising basis each year. This increased to 22% in the year that the policy change began. This is likely due to the inclusion of individuals who are deemed domiciled but are not claiming to be so on their SA109 form.

The percentage of taxpayers who became non-resident remained broadly constant at 5% to 6%. The percentage of taxpayers who left the UK increased from 13% in 2015 to 16% in 2020. The combined rates of becoming non-resident and leaving the UK increased from 18% in 2015 to 22% in 2020. This does not represent causal evidence that the policy has affected exit rates as there are many concurrent factors that may have affected UK residency (not least the UK’s exit from the EU, as will be discussed later).

4.12 Tracking year-on-year change in tax paid by the remittance basis population due to behavioural changes from the tax year ending 2015 to tax year ending 2020

Figure 9: Tracking year-on-year change in tax paid by the remittance basis population due to behavioural changes from the tax year ending 2015 to tax year ending 2020

Figure 9 shows that the year-on-year change in Income Tax and Capital Gains Tax due to individuals leaving the UK and becoming non-resident stayed broadly constant giving a total loss of £0.5 billion each year. However, it was at its largest in the first year of the reforms.

Figure 9 also shows that, prior to the reforms, the change in tax from individuals moving onto the arising basis while remaining non-domiciled led to losses of £0.05 billion or less per year. However, this substantially increased in the year of the policy change.

The finding that moving from the remittance basis to the arising basis often leads to a loss in revenue is consistent with some non-domiciled taxpayers flexibly moving between the arising basis and the remittance basis depending on how much foreign income and gains they expect to realise in a specific tax year.

Non-domiciled taxpayers can claim the remittance basis when they want to realise foreign income and gains and hence it is tax-appropriate to pay the remittance basis charge.

The change in tax revenue for the individuals who remained on the remittance basis increased from £0.21 billion to £0.74 billion in 2015 and 2017 respectively. This effect could be due to 2 reasons.

Firstly, non-domiciled taxpayers claim the remittance basis when they plan to realise higher foreign income and gains.

Secondly, these taxpayers could have brought forward some of their tax liability to reduce future tax payments in response to the policy change. This could include selling foreign assets prior to becoming deemed to avoid having to pay any UK Capital Gains Tax that might otherwise have been payable, or liquidating UK assets prior to the policy change in preparation for leaving the UK.

After the policy change, the change in tax revenue for remittance basis taxpayers remained positive but at a lower level of around £0.4 billion.

5. Section 2: Using regression analysis to attempt to identify causal impacts

Moving beyond simple descriptive analysis of aggregates and tracking individual taxpayers, we look now at trying to draw causal inferences about what is driving these changes – is it the 2017 policy change, or some other concurrent factor? To do so we use a statistical technique called ‘difference-in-differences’ to estimate the absolute and percentage changes in taxes paid by deemed domiciled taxpayers because of the 2017 reforms.

‘Taxes paid’ in this section is defined as the sum of Income Tax, Capital Gains Tax and the remittance basis charge.

We compare a treatment group (those taxpayers affected by the reforms) with a ‘control’ group of non-domiciled taxpayers on the remittance basis who are similar in terms of UK tax residence length. We can use the difference in the change in these 2 groups to estimate the Average Treatment Effect on the Treated (ATET). This is the impact on taxes paid by deemed domiciled taxpayers. We are also interested in how rates of leaving the UK differ between the groups.

5.1 Improving the identification of deemed domiciled taxpayers

Taxpayers do not need to declare their deemed domiciled status on their SA109 form if it is not relevant to their UK tax. That means that just using information from SA109 forms to create the treatment group would result in the exclusion of some deemed domiciled taxpayers.

To capture the largest treatment group possible, we selected any individual who claimed deemed domiciled status on the SA109 form regardless of residency or domicile status for each year after the policy change began (tax year ending 2018 to tax year ending 2020, as shown in Figure 10). We require that they state on their SA109 form that they are condition A or condition B deemed domiciled.

For those who do not declare their deemed domiciled status, we additionally selected those that paid the £90,000 remittance basis charge in the tax years ending 2016 or 2017 or who changed their UK domicile of origin (box 25 of SA109). We believe these characteristics are clear indicators of condition B deemed domiciled taxpayers and condition A deemed domiciled taxpayers respectively.

To identify this wider group, we looked beyond the subset of taxpayers that fill in the SA109 form to the wider SA cohort. This enabled us to identify which taxpayers had remained tax resident in the UK. We did not include taxpayers only recorded in PAYE.

5.2 SA109 population and updated population of deemed domiciled taxpayers

Figure 10: SA109 population and updated population of deemed domiciled taxpayers

Due to the amount of data presented, only part of the table below is visible. Please use the scrollbar at the bottom of the table to view all the columns.

	2017-18	2017-18	2018-19	2018-19	2019-20	2019-20
	SA109 Population	Updated Population	SA109 Population	Updated Population	SA109 Population	Updated Population
Condition A	911	5,380	931	5,591	935	5,764
Condition B	7,365	8,242	7,990	8,926	8,618	9,493
Both	182	214	264	294	288	315
Total	8,458	13,836	9,185	14,811	9,841	15,572

Figure 10 shows that the expanded population is around 5,000 individuals larger for each tax year, largely driven by condition A taxpayers. This supports our hypothesis that some deemed domiciled taxpayers no longer complete the SA109 form when they become deemed as they can no longer opt for the remittance basis and respond to SA as though they are UK domiciled taxpayers.

Many of the condition A deemed domiciled taxpayers did not continue to fill in the SA109 form. This makes sense as these taxpayers cannot use the transitional protections. Stating that they are deemed domiciled is not directly relevant to their UK tax affairs as they are treated as any other UK-domiciled and resident taxpayer.

Figure 11 plots the total and average tax paid by condition A deemed domiciled taxpayers in the tax year ending 2018 before and after the policy change.

5.3 Tracking of the average and total tax paid by deemed domiciled taxpayers from the tax year ending 2018, where ‘tax year’ refers to the year in which the tax year ended

Figure 11: Tracking of the average and total tax paid by deemed domiciled taxpayers from the tax year ending 2018, where ‘tax year’ refers to the year in which the tax year ended

Figure 11 shows an increase in the total and average tax paid in the tax year ending 2017. This suggests that these individuals are remitting more money to the UK or realising gains on UK assets and liquidating them prior to leaving.

The number of condition B deemed domiciled taxpayers increases by about 1,000 each year when individuals who do not declare on their SA109 form are included. These individuals can benefit from the transitional protections we have described so their status as common law non-domiciled taxpayers remains relevant to their tax affairs even though they are ‘deemed’ UK-domiciled.

Figure 12 shows that for the updated population of deemed domiciled taxpayers in 2018, 29% were on the remittance basis, 59% state that they were non-domiciled and 13% claimed that they were non-resident in the tax year ending 2018. This is similar to the composition of statuses seen in Figure 5, but as in Figure 10 we see an increase in deemed domiciled taxpayers who became domiciled before the policy change was applied.

5.4 Comparing the distributions of ‘initial Self Assessment’ deemed domiciled taxpayers with the updated population

Figure 12: Comparing the distributions of ‘initial Self Assessment’ deemed domiciled taxpayers with the updated population

	2017-18	2017-18
	SA109 Population	Updated Population
Non-domiciled	65%	59%
Remittance basis	33%	29%
Non-resident	4%	13%
Total	8,458	13,836

5.5 Selecting the treatment and control groups

For our treatment group, we selected condition B deemed domiciled taxpayers who were on the remittance basis before the policy change.

Unlike condition B taxpayers who do not become deemed until they have been resident for several tax years, condition A taxpayers are immediately deemed if they have been born in the UK or have the UK as their domicile of origin.

That means that all individuals in these circumstances if seen in the tax data immediately become affected by the policy change (unless they have left the UK and hence will not be seen in the tax data). We therefore do not have a control group to compare to.

Furthermore, 90% of condition A deemed domiciled taxpayers were on the arising basis before the policy change and would have faced no substantial change in their tax liabilities because of the policy change.

Similarly, we do not include condition B deemed domiciled taxpayers who were on the arising basis of taxation, since the policy change will not have caused any substantial change to their tax liabilities.

Under the difference-in-differences approach, one of the key assumptions is that the treatment and control groups showed parallel trends prior to the change.

Comparing 2 groups of taxpayers who were on the remittance basis before the policy change is important as their tax treatment for foreign income and gains would have been the same. Taxpayers on the remittance basis are likely to have foreign income and gains in order to make paying the remittance basis charge worthwhile.

However, a key difference between our treatment and control groups is that those in our treatment group have been resident in the UK for longer. They are more likely to have built up strong roots in the UK and may have higher sunk costs of location such as established family and work in the UK. We might expect them to be less inclined to leave the UK than those who have been resident for less time.

This is supported by international evidence. In France, the 2012 personal income tax reform changed the top rate of French income tax from 41% to 45% above €150,000, with additional exceptional contribution for individuals with labour income over €1,000,000. Additionally, the tax base was changed to include dividend income and other savings income, and rules around tax deductions were tightened.

Advani and others (2024) consider the French case and find that migration responses of foreigners to increases in top tax rates are small from a fiscal standpoint and those paying the most tax are the least responsive. They find that the migration response decreases with time and drops off quite sharply around 5 years since arrival in the country; similarly, we may expect that those in our treatment group have a slightly lower likelihood of emigration as by definition they will have been in the UK for at least 15 of the preceding 20 tax years.

It is also important to consider whether any substantial tax changes or policy reforms in neighbouring European countries may have affected our treatment and control groups differently.

In 2017 Italy introduced a ‘new resident tax regime’ for those moving their tax residence to Italy and who had not been resident in Italy for at least 9 out of the 10 preceding tax periods. This provided the opportunity for a tax payment of €100,000 on non-Italian sourced income. However, given that reporting suggests there were 98 applicants in the first year, rising to 549 in 2020, it is unlikely to have affected either population in the early years of Italy’s reforms.

In 2020 Greece introduced a regime for those not tax resident in Greece to pay a flat tax rate of €100,000 on foreign-sourced income if they became tax resident in Greece and were not tax resident for 7 of the 8 prior tax periods. The applicant (or a close relative) would also need to invest a minimum of €500,000 in Greece within 3 years of the residency transfer. This change may have occurred too late to have a substantial impact on our study period.

While the UK’s exit from the EU was a significant event in terms of international mobility, we have no reason to think that that it would impact our treatment and control groups differently.

As we will come to in the ‘checking the parallel trends assumption’ section below, we also restrict the control group to those with longer residency lengths of between 9 and 15 years, which should further mitigate this risk.

5.6 Difference-in-differences model

Our difference-in-differences methodology compares the change in tax paid before and after the policy between the treated (condition B deemed domiciled) and control groups (non-domiciled taxpayers on the remittance basis who have not yet been deemed).

The control group provides us with a ‘counterfactual difference’ – the change we would expect in the absence of this policy change. This is valid on the assumption that, without the policy, the tax paid would change at the same rate, called the ‘parallel trends assumption’.

This is the regression equation. Tax revenue is the dependent (left hand side (LHS)) variable which consists of the Income Tax, Capital Gains Tax and remittance basis charges paid by the individual as defined at the start of Section 2.

On the right hand side (RHS), β₀ shows the average tax paid by the non-domiciled taxpayers on the remittance basis (the control group). β₁ shows the average change in tax paid from before the policy change to after the policy change for the non-domiciled taxpayers on the remittance basis (the control group). β₂ shows the average difference in tax paid between the control group and the deemed domiciled group before the policy change.

β₃ estimates the change in tax paid due to the policy change. It shows the average change in tax paid for the deemed domiciled taxpayers, comparing before the policy change to after it. Hence shows the average effect among those individuals who were treated.

The following equation estimates the ATET by considering treatment and control groups:

D=1 indicates treatment and D=0 indicates non-treatment (control). Y represents tax paid for an individual, where the subscript 0 denotes the untreated state, subscript 1 denotes the treated state, subscript t denotes the post-treatment period and subscript t’ denotes the pre-treatment period.

5.7 Checking the parallel trends assumption

The estimator above identifies the ATET if the following assumption holds:

This is the parallel trends assumption.

We examined the average tax paid by the control group and the treatment group over the whole period of the available data.

We have calculated the residency length of our proposed control group. We did this because non-domiciled taxpayers who have been resident in the UK for longer are more similar to the deemed domiciled in terms of tax paid.

We calculated residency length by summing the number of years that a taxpayer was UK resident in the 20 tax years immediately preceding the policy change (the tax year ending 1997 to the tax year ending 2017). We count a taxpayer as resident in the UK if they fill in a Self Assessment tax return and they do not declare that they are non-resident on their SA109 form.

When non-domiciled taxpayers on the remittance basis are grouped by similar residency lengths, we see in Figure 13 that the parallel trends assumption holds for those with residency length between 9 and 15 years. These individuals pay similar levels of tax to the deemed domiciled treatment group and the trend of tax paid is similar before the policy change was announced.

Therefore, the use of a control group consisting of individuals with a residency length between 9 and 15 years should give a suitable estimation of the treatment effect.

5.8 Comparing the average tax paid by the deemed domiciled treatment group and subpopulations of remittance basis taxpayers from 2015 with residency lengths of 9 to 15 years

Figure 13: Comparing the average tax paid by the deemed domiciled treatment group and subpopulations of remittance basis taxpayers from 2015 with residency lengths of 9 to 15 years

5.9 Regression specification

We take the tax year ending 2015 as a baseline before the policy change because this is before the policy change was announced. It is possible for those who would have been affected to modify their tax affairs prior to the policy coming into effect in April 2017.

We calculate the residency length of the control group up to the tax year ending 2017. The tax paid data is extracted from the tax year ending 2015. This is to ensure we are selecting individuals that are most like the treatment group but who have not yet altered their tax affairs.

We have also looked at what might happen if we took the tax year ending 2017 as a baseline. This is after the announcement, but before the policy change took place. We observe an increase in tax revenue from the deemed domiciled group between the tax years ending 2017 and 2018 in comparison to the control group.

This could support a hypothesis that deemed domiciled taxpayers started to modify their tax affairs once the policy change was announced to reduce the impact after the policy began. There could be a divergence in behavioural response to the policy change due to the range in ability to plan affairs for tax purposes. This is why we observe wide confidence intervals for this treatment effect coefficient.

The control group are non-domiciled taxpayers who were on the remittance basis before the policy change with UK residency of between 9 to 15 years (in the last 20 years).

We applied a logarithmic transformation to the tax paid variable to ensure a symmetrical distribution. This is required to give a good interpretation of the average effect of a variable through regression (further details can be found in Annex 4).

5.10 Methodology for testing mobility

We test whether deemed domiciled taxpayers were more likely to remain in the UK rather than leaving the UK compared to non-domiciled taxpayers.

We flag whether a taxpayer remained in the UK by creating a dummy variable for whether they were filling in a Self Assessment tax return form (SA100) for 2 consecutive tax years with the following flag:

We use this flag as the dependent variable on the LHS in the difference-in-differences regression as used previously. The equation is stated below.

On the RHS, β₀ shows the average tax paid by the non-domiciled taxpayers on the remittance basis (the control group). β₁ shows the average change in tax paid from before the policy change to after the policy change for the non-domiciled taxpayers on the remittance basis (the control group). β₂ shows the average difference in tax paid between the control group and the deemed domiciled group before the policy change.

β₃ estimates the change in tax paid due to the policy change. It shows the average change in tax paid for the deemed domiciled taxpayers, comparing before the policy change to after it. Hence shows the average effect among those individuals who were treated.

The Remain in UK_i.t variable on the LHS is binary. That means that we use logistic (logit) and probit regressions.

We use the deemed domiciled taxpayers from the tax year ending 2018 as the treatment group.

We add the taxpayers who had a residency length greater than 15 years from the remittance basis taxpayers ‘control’ group to the deemed domiciled treatment group for the regressions that compare the deemed domiciled condition B taxpayers to the remittance basis taxpayers.

This is to include individuals who might have become deemed domiciled but left the UK or became non-resident to avoid this status. Otherwise, the treatment group would only include people who had remained UK tax resident.

We use the tax years ending 2013 and 2014 as the 2 consecutive tax years before the policy change as there were no policy changes occurring to the non-domicile regime in these years.

We use the tax years ending 2018 and 2019 as the 2 consecutive tax years after the policy change. We expect an instant migration elasticity once the policy change was enforced as the policy change was announced in advance.

5.11 Baseline regression results of our modelling - tax revenue

We examine our headline results in Figure 14. The values in the parentheses show the 95% confidence intervals for each of the coefficients. This means that in 95% of our estimates the average value of the coefficient is in this range.

We create a control group from the remittance basis population from 2015 who have a residency length of between 9 and 15 years (out of the last 20 years).

The treatment group are the condition B deemed domiciled taxpayers who were on the remittance basis before the policy change.

We examine 3 post-policy change tax years separately by using different deemed domiciled treatment groups from each of these tax years.

Regressions 1 and 2 use the deemed domiciled taxpayers from the tax year ending 2018 as the treatment group. Regressions 3 and 4 use the deemed domiciled taxpayers from the tax year ending 2019 as the treatment group. Regressions 5 and 6 use the deemed domiciled taxpayers from the tax year ending 2020 as the treatment group.

This means that we compare the base tax year (tax year ending 2015) to each of the tax years ending 2018, 2019 or 2020.

5.12 Baseline regression results table

Figure 14: Baseline regression results

Due to the amount of data presented, only part of the table below is visible. Please use the scrollbar at the bottom of the table to view all the columns.

	N	Regression type	β0	β1posti,t	β2deemeddomi,t	Treatment effect β3posti,tdeemeddomi,t
1	5,994	Linear regression, 2018 deemed doms	264,070 [211,088, 317,052] (Note 3)	-5,734 [-71,861, 60,394]	18,627 [-68,599, 105,853]	197,797 [59,297, 336,296] (Note 3)
2	5,371	Log-linear regression, 2018 deemed doms	10.72 [10.55, 10.89] (Note 3)	0.28 32% [0.05, 0.52] (Note 2)	-0.41 -34% [-0.63, -0.19] (Note 3)	0.64 90% [0.35, 0.93] (Note 3)
3	5,831	Linear regression, 2019 deemed doms	224,073 [192,387, 255,760] (Note 3)	43,199 [-26,008, 112,406]	11,248 [-37,102, 59,600]	107,161 [12,553, 201,770] (Note 2)
4	5,226	Log-linear regression, 2019 deemed doms	10.62 [10.45, 10.79] (Note 3)	0.19 0% [-0.07, 0.44]	-0.36 -30% [-0.58, -0.15] (Note 3)	0.76 114% [0.45, 1.07] (Note 3)
5	5,877	Linear regression, 2020 deemed doms	224,073 [192,387, 255,760] (Note 3)	47,802 [-28,034, 123,638]	16,119 [-31,760, 63,998]	241,667 [100,774, 382,560] (Note 3)
6	5,230	Log-linear regression, 2020 deemed doms	10.62 [10.45, 10.79] (Note 3)	0.16 0% [-0.10, 0.42]	-0.35 -30% [-0.57, -0.13] (Note 3)	0.82 127% [0.52, 1.13] (Note 3)

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

Figure 14 shows that the average treatment effect on income for the treated is positive and significant for both the linear and log-linear regressions for all the post-policy change tax years. The average treatment effect fluctuates for these 3 tax years. It is lowest for the tax year ending 2019 and highest for the tax year ending 2020.

We examine the linear regressions and find that deemed domiciled taxpayers paid more tax due to the policy change in all the tax years after the policy change compared to the tax year ending 2015.

The difference in tax paid due to the policy change fluctuates from £198,000 more in the tax year ending 2018, to £107,000 more in the tax year ending 2019 and £242,000 more in the tax year ending 2020 (the Average Treatment Effect for the Treated in Figure 14).

The log-linear regressions suggest that deemed domiciled taxpayers increased their tax paid by 90%, 114% and 127% for the tax years ending 2018, 2019 and 2020 respectively when compared to the tax year ending 2015. Following this policy change deemed domiciled taxpayers paid approximately double the tax they paid before the change.

5.13 Robustness checks

To test the robustness of the results, we consider different versions of the regression to check that a similar treatment effect is still observed. These different versions of the regression take different choices in how to form the control group. They also explore picking a different ‘treated’ group from after the introduction of the reform.

We again examine 3 post-policy change tax years. Most of the treatment effects remain positive and significant which suggests that we are observing a real increase in tax revenue due to this policy change, rather than the result being due to a chance selection of particular control or treatment groups.

The results of the robustness check regressions are shown in Annex 8.

5.14 Placebo regressions

We consider the change in the tax paid by deemed domiciled taxpayers and by remittance basis taxpayers in the control group in the tax years before the policy change was announced. This is another check to ensure that the results are capturing a real effect and are not by chance.

We find no significant treatment effect in the tax years before the policy change was announced. This is further evidence to suggest that we are capturing a real increase in tax revenue due to this policy change. The results of the placebo regressions are shown in Annex 6.

5.15 Estimating the treatment effect along the tax paid distribution

We examine the distribution of the tax paid by deemed domiciled individuals and find that it is positively skewed. This means that the number of taxpayers paying less tax than the average is higher than the number of taxpayers paying more tax than the average.

Our baseline results show wide 95% confidence intervals for the treatment effect coefficient. This means that it is likely that individuals are affected differently due to the policy change depending on how much tax they paid before the change.

The earlier regression results provide only an overall average treatment effect for the deemed domiciled taxpayer population. Due to the skewness in the tax paid distribution, it is more informative to look beyond the overall average.

We examine the change in the treatment effect depending on the level of tax paid before the policy change using quantile regressions. This means that we can estimate the treatment effect at different points in the tax paid distribution.

This means that we compare the treatment group of people who became deemed in the tax year ending 2018 to a control group of those who had not been resident long enough to become deemed, given that both groups were non-doms on the remittance basis in the tax year ending 2015. The control group had a residency length of between 9 and 15 years in the tax year before the policy change.

Figure 15 shows the coefficient of the treatment effect at different percentiles of the tax paid distribution. We estimate the change in tax paid caused by the policy change or the ATET for all deemed domiciled taxpayers. We display the treatment effect at each percentile ordered by ascending magnitude.

For example, the median effect selects the middle value in the range of treatment effects estimated from the regression. In Figure 15 this treatment effect is £70,000.

5.16 Quantile regression results for 2018 deemed domiciled taxpayers

Figure 15: Quantile regression results for 2018 deemed domiciled taxpayers

Percentile of the tax paid distribution	Change in tax paid due to the policy change
Bottom 25%	£12,000 (Note 3)
Median	£70,000 (Note 3)
Top 25%	£122,000 (Note 3)
Top 10%	£258,000 (Note 2)
Top 5%	£748,000 (Note 3)
Top 1%	£1,302,000 (Note 1)

Note 1: 95% Confidence error  p<0.01

Note 2: 95% Confidence error p<0.05

Note 3: 95% Confidence error p<0.1

The results show that the size of the treatment effect is different at different points in the tax paid distribution. We see that individuals who paid less tax before the change had a smaller increase in their tax paid, whereas taxpayers who paid more beforehand had a much larger increase in their tax paid following the policy change.

In Figure 16 below we examine the different treatment effects along the tax paid distribution for the deemed domiciled taxpayers from the tax year ending 2019 as the treatment group. We compare the base tax year (ending 2015) to the treatment tax year (ending 2019) for these quantile regressions.

5.17 Quantile regression results for 2019 deemed domiciled taxpayers

Figure 16: Quantile regression results for 2019 deemed domiciled taxpayers

Percentile on the tax paid distribution	Change in tax paid due to the policy change
Bottom 25%	£10,000 (Note 3)
Median	£64,000 (Note 3)
Top 25%	£125,000 (Note 3)
Top 10%	£413,000 (Note 2)
Top 5%	£268,000
Top 1%	£1,294,000

Note 1: 95% Confidence error  p<0.01

Note 2: 95% Confidence error p<0.05

Note 3: 95% Confidence error p<0.1

In Figure 17 below we set out the treatment effects along the tax paid distribution for the deemed domiciled taxpayers from the tax year ending 2020 as the treatment group. We compare the base tax year (ending 2015) to the treatment tax year (ending 2020).

5.18 Quantile regression results for 2020 deemed domiciled taxpayers

Figure 17: Quantile regression results for 2020 deemed domiciled taxpayers

Percentile of the tax paid distribution	Change in tax paid due to the policy change
Bottom 25%	£10,000 (Note 3)
Median	£77,000 (Note 3)
Top 25%	£160,000 (Note 3)
Top 10%	£346,000 (Note 2)
Top 5%	£447,000
Top 1%	£2,257,000

Note 1: 95% Confidence error  p<0.01

Note 2: 95% Confidence error p<0.05

Note 3: 95% Confidence error p<0.1

We find that deemed domiciled taxpayers from the tax years ending 2019 or 2020 paid statistically significantly more tax (p<0.05) due to the policy change up to the top 10% of the tax paid distribution.

There is a positive but not statistically significant difference in the tax paid between the deemed domiciled and non-domiciled taxpayers in the top 5% and top 1% of the tax paid distribution in the tax years ending 2019 and 2020.

The lack of statistical significance could be due to the small size of the population in the top 5% and top 1% of the tax paid distribution. This means that only a relatively large difference in the tax paid between the deemed domiciled taxpayers and the control group would give a statistically significant result.

It could also mean that for the taxpayers paying the most tax, there are some deemed domiciled taxpayers who pay a lot more tax and others for whom there is no difference in tax paid compared to the control group.

5.19 Event study analysis: estimating the treatment effect for all the treatment years

We examine the treatment effect for the 3 post-policy change tax years by selecting the deemed domiciled population that remains deemed domiciled for the tax years ending 2018, 2019 and 2020. This is the treatment group.

We use the remittance basis population from the tax year ending 2015 as the control group when their residency length is between 9 and 15 years in the tax year before the policy change.

We estimate the treatment effect for each post-policy change tax year using the tax year ending 2015 as the base year. This means we are comparing the difference in tax paid in the tax year ending 2015 to either of the tax years ending 2018, 2019 or 2020.

5.20 Baseline regression results of the post-policy change treatment effect

Figure 18: Baseline regression results of the post-policy change treatment effect

Due to the amount of data presented, only part of the table below is visible. Please use the scrollbar at the bottom of the table to view all the columns.

N	Regression type	β0	β1posti,t	β2deemeddomi,t	β3posti,tdeemeddomi,t
5,059	Linear regression, 2018 treatment effect	224,073 [192,384, 255,763] (Note 3)	16,963 [-35,389, 69,315]	85,548 [-7,078, 178,174] (Note 1)	207,244 [47,760, 366,727] (Note 2)
4,971	Linear regression, 2019 treatment effect	224,073 [192,384, 255,763] (Note 3)	43,199 [-26,015, 112,412]	85,548 [-7,079, 178,175] (Note 1)	81,121 [-42,226, 204,467]
4,936	Linear regression, 2020 treatment effect	224,073 [192,384, 255,763] (Note 3)	47,802 [-28,042, 123,646]	85,548 [-7,080, 178,176] (Note 1)	194,677 [26,398, 362,956] (Note 2)

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

5.21 Treatment effect, post-policy change

Figure 19: Treatment effect, post-policy change

We find from Figures 18 and 19 that taxpayers who became deemed domiciled paid more tax both before and after the policy change than those who remained non-domiciled.

Figures 18 and 19 show the treatment effect oscillates in size between tax years ending 2018 and 2020. There are increases in tax paid by deemed domiciled taxpayers in comparison to non-domiciled taxpayers or no difference in tax paid in the years after the policy change. We will explain this outcome further shortly.

Log-linear regression is an alternative way to estimate a treatment effect as a percentage change in tax paid rather than a change in average tax paid.

5.22 Logarithm baseline regression results of the post-policy change treatment effect

Figure 20: Logarithm baseline regression results of the post-policy change treatment effect

Due to the amount of data presented, only part of the table below is visible. Please use the scrollbar at the bottom of the table to view all the columns.

N	Regression type	β0	β1posti,t	β2deemeddomi,t	β3posti,tdeemeddomi,t
4,528	Log-linear regression, 2018 treatment effect	10.62 [10.45, 10.79] (Note 3)	0.34 40% [0.10, 0.58] (Note 3)	-0.25 -22% [-0.48, -0.02] (Note 2)	0.76 114% [0.45, 1.06] (Note 3)
4,486	Log-linear regression, 2019 treatment effect	10.62 [10.45, 10.79] (Note 3)	0.19 0% [-0.07, 0.44]	-0.25 -22% [-0.48, -0.02] (Note 2)	0.73 108% [0.42, 1.05] (Note 3)
4,403	Log-linear regression, 2020 treatment effect	10.62 [10.45, 10.79] (Note 3)	0.16 0% [-0.10, 0.42]	-0.25 -22% [-0.48, -0.02] (Note 2)	0.90 146% [0.58, 1.22] (Note 3)

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

5.23 Percentage change in treatment effect, post-policy change

Figure 21: Percentage change in treatment effect, post-policy change

Figures 20 and 21 show that deemed domiciled taxpayers paid more tax in percentage terms after the policy change compared to those remaining non-domiciled. This increase in tax revenue is sustained in the years following the policy change.

We see the difference in maximum tax revenue values in Figure 22, defined as the largest amount of tax paid by anyone in either the treatment or control group. Figure 22 shows that deemed domiciled taxpayers pay consistently more in maximum tax and that their maximum tax revenue is more volatile than the remittance basis taxpayers from the tax year ending 2015.

It is possible that since deemed domiciled taxpayers have stayed in the UK longer than the control group, they have had more time to build up assets in the UK. Deemed domiciled taxpayers would therefore be more likely to pay Capital Gains Tax and the infrequency of the disposal of assets and the range in value of these assets could result in greater volatility in tax liabilities. This volatility may lead to variable treatment effect sizes and significances.

5.24 Comparison of the maximum tax revenues for the treatment and control groups

Figure 22: Comparison of the maximum tax revenues for the treatment and control groups

We observe a drop in the maximum tax revenue value for the deemed domiciled taxpayers in the tax year ending 2019 compared to tax years ending 2018 and 2020. This corresponds to the lower and insignificant treatment effect in the case of the linear regressions for the tax year ending 2019. We suggest that the volatility of the treatment effect reflects the volatility of maximum tax revenue for deemed domiciled taxpayers.

Figure 23 plots the average tax paid by the non-domiciled taxpayers who became deemed, and the average tax paid by those remaining non-domiciled on the remittance basis in the tax year ending 2015. This means we can track the average difference in tax paid by the treatment and control groups over time.

We see a sustained increase in tax paid by the deemed domiciled taxpayers compared to the remittance basis taxpayers. This increase begins in the tax year ending 2017 but is much larger once the policy change begins.

5.25 Average tax paid by deemed domiciled taxpayers and remittance basis taxpayers

Figure 23: Average tax paid by deemed domiciled taxpayers and remittance basis taxpayers

5.26 Results for condition B deemed domiciled taxpayers who were on the remittance basis - mobility

We compare the mobility of condition B deemed domiciled taxpayers who were on the remittance basis prior to the policy change to the remainder of the remittance basis population from the tax year ending 2015.

Similarly to the regressions that examine the tax revenue, we use remittance basis taxpayers with UK tax residency of between 9 and 15 years in the 20 years before the year of the policy change (the tax year ending 2017).

It is important that both the control and treatment groups have similar residency lengths to ensure that factors that are not captured in the regression are more likely to impact both groups in a similar way.

We examine the marginal effects of the coefficients. This means we do not examine the intercept coefficient β₀. We estimate the change in likelihood that an individual leaves the UK due to a specific change. A positive coefficient means that individuals were less likely to leave the UK. A negative coefficient means that individuals were more likely to leave the UK.

β₁ shows the average change in likelihood of remaining in the UK from before the policy change to after the policy change for non-domiciled taxpayers remaining on the remittance basis (the control group).

β₂ shows the average change in likelihood of remaining in the UK between the control group and the deemed domiciled group before the policy change.

β₃ estimates the average change in likelihood of remaining in the UK due to the policy change. It shows the average change in likelihood of remaining in the UK for the deemed domiciled taxpayers, comparing before the policy change to after it. Hence β₃ shows the ATET.

We have taken a definition of the treatment group which includes taxpayers who had the potential to be deemed (remittance basis taxpayers who had a residency length greater than fifteen years), so it includes some who left the UK rather than becoming deemed.

We feel that it is important to include this broader group, otherwise we would not be including those who may have left the UK because of the reforms and our results would be misleading (they would show no increased likelihood of leaving the UK).

5.27 Mobility regression results for condition B deemed domiciled taxpayers

Figure 24: Mobility regression results for condition B deemed domiciled taxpayers

Due to the amount of data presented, only part of the table below is visible. Please use the scrollbar at the bottom of the table to view all the columns.

N	Regression type	Marginal effect of β1posti,t	Standard Error	Marginal effect of β2deemeddomi,t	Standard Error	Marginal effect of the treatment effect β3posti,tdeemeddomi,t	Standard Error
10,209	Logit	-0.0417 (Note 3)	0.00583	0.123	0.0195	-0.0855 (Note 3)	0.0206
10,209	Probit	-0.0545 (Note 3)	0.00596	0.0998 (Note 3)	0.0128	-0.0583 (Note 3)	0.014

Note 1: 95% Confidence error p<0.01

Note 2: 95% Confidence error p<0.05

Note 3: 95% Confidence error p<0.1

We use the logit and probit results as they are more appropriate for regressions using binary dependent variables.

Figure 24 shows, using β₁, non-domiciled taxpayers on the remittance basis (the control group) were 4% to 5% more likely to leave after the policy change.

The marginal effect of the treatment effect term β₃ is negative and significant. Deemed domiciled taxpayers were an additional 5% to 8% more likely to leave the UK due to the policy change than the control group. This means that deemed domiciled taxpayers were 10% to 12% more likely to leave the UK due to the policy change.

We found no issues in our robustness checks, which are summarised in Annexes 5 and 8. Standard errors reported are unadjusted.

5.28 Results for condition A deemed domiciled taxpayers - mobility

We compare the mobility of condition A deemed domiciled taxpayers to the control group of individuals who had previously changed their UK domicile of origin to a non-UK domicile.

5.29 Mobility regression results for condition A deemed domiciled taxpayers

Figure 25: Mobility regression results for condition A deemed domiciled taxpayers

N	Regression type	Marginal effect of β1posti,t	Marginal effect of β2deemeddomi,t	Marginal effect of β3posti,tdeemeddomi,t
41,884	Logit	-0.0298	-0.0201	0.270 (Note 3)
41,884	Probit	-0.0940	-0.0613	0.271 (Note 3)

Note 1: 95% Confidence error p<0.01

Note 2: 95% Confidence error p<0.05

Note 3: 95% Confidence error p<0.1

Figure 25 shows that the marginal effect of the interaction term is positive and significant. Condition A deemed domiciled taxpayers were 27% more likely to remain in the UK after the policy change.

We find that before the policy change there is no difference in the likelihoods of remaining in the UK for the condition A deemed domiciled taxpayers and the non-domiciled control group. We would expect this as there is no reason for these 2 groups of non-domiciled taxpayers to behave differently in this regard before the policy change.

We are unable to run the same kind of robustness checks for this group as all condition A taxpayers are immediately considered deemed, so the vast majority of those deemed condition A were deemed in the first year of the reforms. Subsequently only very small numbers of new condition A taxpayers occur.

5.30 Estimating Exchequer impact

We calculate the difference in tax paid by deemed domiciled taxpayers and the counterfactual tax paid by these taxpayers according to the tax revenue regression results. We use the log-linear regressions in Figures 14 and 20 for this calculation.

We use the population of condition B deemed domiciled taxpayers in each of the post-policy change tax years who were on the remittance basis in the tax year ending 2015. We also add the taxpayers who had a residency length greater than 15 years from the remittance basis taxpayers ‘control’ group to the deemed domiciled group.

This is to include individuals who must have left the UK or become non-resident to avoid the status. This gives a total of 3,500, 3,700 and 4,000 taxpayers for the deemed domiciled taxpayer group from the tax years ending 2018, 2019 and 2020 respectively that were in the UK in the tax year ending 2015.

We apply the change in likelihood of leaving the UK from Figure 24. We use the β₀ estimate from Figure 18 and the percentage change in tax paid from Figure 20.

We include more detail about the estimation of the Exchequer impact in Annex 9.

5.31 Estimation of the Exchequer impact of the deemed domicile policy change, by tax year ending

Figure 26: Estimation of the Exchequer impact of the deemed domicile policy change, by tax year ending

	2015	2018	2019	2020
Number of counterfactual deemed domiciled taxpayers	3,446	3,291	3,580	3,804
Total tax paid by counterfactual deemed domiciled taxpayers (£million)	603	869	627	666
Number of deemed domiciled taxpayers	3,446	3,067	3,337	3,545
Total tax paid by deemed domiciled taxpayers (£million)	603	1,595	1,391	1,780
Difference in tax paid (£million)	0	726	764	1,114

Figure 26 shows there is an increase of over £700 million in tax paid from the condition B deemed domiciled taxpayers due to this policy change when comparing the tax paid by these taxpayers who remained in the UK in the tax year ending 2015 to the tax year ending 2018. This means that the increase in tax paid outweighs the loss of tax from the deemed domiciled taxpayers who left the UK.

This difference in tax paid increases to over £1 billion in the tax year ending 2020.

We have not included condition A deemed doms and condition B deemed doms on the arising basis before the policy change in the regression analysis. This means that the Exchequer impact is a lower bound estimate of the increase in tax paid by deemed domiciled taxpayers.

6. Discussion

More than 9,800 (perhaps as many as 15,600 when considering those that do not proactively indicate their deemed status) non-domiciled taxpayers have remained in the UK after becoming deemed and their number has increased since the policy change. They are contributing more than £3 billion of tax and National Insurance contributions.

We have shown that 4% to 5% of non-domiciled taxpayers who were unaffected by the change left the UK following the policy change. This provides us an estimate of normal churn.  We have seen that our analysis suggests that the policy change did cause some increase to the number of deemed domiciled taxpayers leaving the country, with departures of 10% to 12% instead.

The number of those affected by the policy change is low relative to all those who are non-domiciled (but do not formally declare as such). In the big picture, this is a minor policy change that affects only a small portion of UK non-domiciled taxpayers.

We have discussed the international research literature which suggests that while personal tax policy can form part of migration decisions, it does not form a large part. Other factors may have contributed to deemed domiciled taxpayers staying in the UK despite the policy change including family ties, their likelihood to be older and potentially in a long-term job, and their social and cultural ties to the country (see Annex 1).

The literature also suggests that career prospects and seniority in a job are linked to lower worker migration. Advani and others (2022) found that 80% of non-doms have earnings from work or pensions. This is likely to at least partially explain the fact that the majority of deemed domiciled taxpayers, especially those deemed under condition B, remain in the UK.

Advani and others (2024) confirm that there is no evidence of any substantial increase in migration elasticity when a greater number of years post-implementation are considered. This suggests that there was no substantial delayed migration response after the first year of the new policy.

It is also worth noting that we have used the population of current non-doms for this report. This means we have used the taxpayers currently claiming non-domiciled status on their SA109 form. Advani and others (2022) state the shortcomings of using current non-doms. They state that individuals who choose to no longer claim non-dom status or who change their domicile status from year to year are not consistently captured in the measured population.

Advani and others (2022) instead use the ‘ever non-doms’ population. This includes taxpayers who have claimed non-dom status in the past but are now domiciled UK residents. They found that the ‘ever non-doms’ population has increased between 2001 and 2018. This further supports our findings that non-domiciled taxpayers have not been substantially deterred from remaining in the UK due to the announcement of the deemed domicile reforms.

Using the difference-in-differences method to estimate the change in mobility controls for other factors that could impact mobility over this period. However, if other factors have different impacts on the mobility of the treatment and control groups then this would affect our analysis.

It is therefore important to consider the potential impact of events-based migration such as that driven by EU exit. However, the post-EU exit migration system has not had much impact on the higher-skilled sectors in which non-doms are predominantly found.

Portes and Springford (2023) find that in the information and communications (ICT), finance, professional and technical, and education sectors visa issuance is comparable to the net change in non-UK employment seen pre-pandemic.

These were also the sectors which relied heavily on work-related visas prior to the introduction of the new system. This can be taken to mean that the new system is as open overall as before and that there would be little impact on either the treatment or control groups.

The period we have examined to analyse mobility was before the COVID-19 pandemic, so this will not have impacted the results.

We have seen that the deemed domiciled taxpayers have paid more tax on average since they became deemed. That means despite the increased rate of departures from the UK, the additional tax collected from those that remained more than compensated for the loss from those that left.

The increase for the condition B deemed domiciled taxpayers who were previously on the remittance basis totalled over £700 million in the tax years ending 2018 and 2019 and over £1 billion in the tax year ending 2020. This was partly driven by those who were historically higher taxpayers paying higher tax still after the policy change.

The analysis suggests the policy is achieving more progressive taxation of the deemed domiciled group as they are now paying more tax on average. Those at the lower end of the prior taxpaying distribution are paying more tax and those at the top of the distribution are paying a lot more tax.

7. References

Advani, A., Burgherr, D. & Summers, A., 2024. Taxation and migration by the super-rich. [Online] Available at: AdvaniBurgherrSummers2024_TaxationAndMigrationByTheSuperrich.pdf (arunadvani.com)

Advani, A., Summers, A., Savage, M. & Burgherr, D., 2022. Reforming the non-dom regime: revenue estimates. [Online]. Available at: bn38.2022.pdf (warwick.ac.uk) and updated in 2024 by AdvaniBurgherrSummers2024_TaxationAndMigrationByTheSuperrich.pdf (arunadvani.com)

Advani, A., Summers, A., Savage, M. & Burgherr, D., 2022. The UK’s global economic elite: A sociological analysis using tax data. [online] Warwick.ac.uk. Available at: https://warwick.ac.uk/fac/soc/economics/research/centres/cage/publications/workingpapers/2022/the_uks_global_economic_elite_a_sociological_analysis_using_tax_data/ [Accessed 31 May 2022].

Advani, A., Summers, A. & Poux, C., 2024. Top Flight: Who migrates in response to top tax rates? [Online] Available at: arunadvani.com/slides/AdvaniPouxSummers2024_TopFlight.pdf

Akcigit, U., Baslandze, S. & Stantcheva, S., 2016. Taxation and the International Mobility of Inventors. American Economic Review, 106(10), pp.2930-2981.

Benoit, K., 2011. [Online] Kenbenoit.net. Available at: https://kenbenoit.net/assets/courses/ME104/logmodels2.pdf [Accessed 31 May 2022].

Borjas, G., 2013. Labor economics. 6th ed. New York: McGraw-Hill Companies, pp.321, 326-327, 350.

Del Carpio, X., Ozden, C., Testaverde, M., Wagner, M. & Del Carpio, X., 2022. Global Migration of Talent and Tax Incentives: Evidence from Malaysia’s Returning Expert Program. [Online] Papers.ssrn.com. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2867656 [Accessed 31 May 2022].

Feng, C., Wang, H., Lu, N., Chen, T., Hh, H., Lu, Y. & Tu, X., 2014. Log-transformation and its implications for data analysis. [Online] PubMed Central (PMC). Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4120293/ [Accessed 31 May 2022].

Friedman, S., Gronwald, V., Summers, A. & Taylor, E., 2024. Tax flight? Britain’s wealthiest and their attachment to place. [Online] Available at: Friedman and others (2024) Tax flight? Britain’s wealthiest and their attachment to place

Giuffrida, A., 2023. Global super-rich head to Italy for tax breaks and dolce vita. The Guardian. [Online] Available at: Global super-rich head to Italy for tax breaks and dolce vita, The Guardian

HMRC, 2016. Income Tax, Inheritance Tax and Capital Gains Tax: deemed domicile rule. [Online] Available at: https://www.gov.uk/government/publications/income-tax-inheritance-tax-and-capital-gains-tax-deemed-domicile-rule/income-tax-inheritance-tax-and-capital-gains-tax-deemed-domicile-rule.

HMRC, 2020. Non domiciled taxpayers in the UK.

HMRC, 2020. Non domiciled statistics in the UK (Table 10).

Kleven, H., Landais, C., Saez, E. & Schultz, E., 2013. Migration and Wage Effects of Taxing Top Earners: Evidence from the Foreigners’ Tax Scheme in Denmark*. The Quarterly Journal of Economics, 129(1), pp.333-378.

Kleven, H., Landais, C. & Saez, E., 2013. Taxation and International Migration of Superstars: Evidence from the European Football Market. American Economic Review, 103(5), pp.1892-1924.

Moretti, E. & Wilson, D., 2017. The Effect of State Taxes on the Geographical Location of Top Earners: Evidence from Star Scientists. American Economic Review, 107(7), pp.1858-1903.

Portes, J., & Springford, J., 2023. The impact of the post-Brexit migration system on the UK labour market. Contemporary Social Science, 18(2), 132-149.

van Dalen, H. P., & Henkens, K., 2007. Longing for the Good Life: Understanding Emigration from a High-Income Country. Population and Development Review, 33(1), 37–65. [Online] Available at: http://www.jstor.org/stable/25434584

Young, C. & Varner, C., 2011. Millionaire migration and state taxation of top incomes: Evidence from a natural experiment. National Tax Journal, 64(2.1), pp.255-283.

Annexes

Annex 1: Academic literature on international mobility decisions

This section presents a summary of the academic literature on the impact of tax on global migration of high-income individuals. Overall, the evidence shows that tax policy can have mixed results on the migration of individuals and suggests that when it comes to the decision of migration, other factors are more influential.

Van Dalen and others (2007) found that age, education, income, and social networks play key roles in triggering emigration from the Netherlands. The potential migrant is predominantly young, better educated, with a high income, with family or friends abroad. The effects of social networks both abroad and those at home (as approximated by marital status) are substantial. The results suggest that while economic considerations play some role in explaining the intent to emigrate from the Netherlands, they are not the dominant factor.

Varner and Young (2011) examined whether a small increase in the tax burden for the very rich in California caused any substantial migration out of the state. The study found that the change in tax burden did not cause an impact on outward or inward migration. Yet the study found that there was a high outward migration effect for individuals who were getting divorced. This suggests that factors other than an individual’s tax burden can be more salient.

Kleven and others (2013) examined the impact of a scheme introduced in Denmark where high earning foreigners received preferential tax treatment for the first 3 years of their residency in Denmark. Overall, the scheme doubled the number of highly paid foreigners in Denmark below the income eligibility threshold for the scheme. But it also showed that other factors are important for mobility; for example, having children was linked to a 50% increase in the likelihood of staying in Denmark after the 3-year expiration of the scheme. There was also evidence that individuals brought forward their tax bill to avoid an increase in tax.

Kleven and others (2013) examined the effects of top tax rates on international migration of football players across Europe. The study found that foreign players were sensitive to their net-of-tax rate. The sensitivity of domestic players to their net-of-tax rate was much smaller but there was still some sensitivity.

Borjas (2013) also states that other factors impact the migration of workers such as family ties, age, and seniority in their job. The paper suggests that optimal family migration may differ from optimal individual migration due to different jobs in a household and the instability that a move can cause on a family. It also found that the probability of a worker migrating decreases with age – for example, the probability of a worker migrating decreases from 4% to 1% when comparing a worker in their twenties to a worker in their fifties.

Del Carpio and others (2016) studied the impact of Malaysia’s Returning Expat Program which offered tax reductions and exemptions if educated Malaysian expats returned to work in Malaysia. It found that individuals who had a job offer in Malaysia were highly incentivised to return due to the tax reduction but individuals without a job offer were not.

Akcigit and others (2016) studied the effect of top tax rates on the top 1% of inventors registering patents in the US or Europe. They found that the top 1% of foreign inventors were significantly influenced by top tax rates.

But other factors also matter, such as whether the inventor worked in a multinational company and whether the country in which they lived at the time had a research centre for their research area. Inventors were less sensitive to tax changes due to career considerations and the presence of positive agglomeration externalities (synergies from being available in the same location as other similar businesses) dampens the negative effects of tax on migration.

Moretti and Wilson (2017) found that differences in taxes charged at the state level in the United States had a significant impact on the location of top-earning scientists. As the net-of-tax rate increases, the stock of these scientists in the state increases. This paper found that personal tax and corporation tax have similar long run impacts on mobility.

Advani and others (2022) (with updated results in 2024) also used HMRC data to evaluate the impact of the 2017 deemed domicile reforms assessing the impact of the reforms on both migration and Exchequer revenue.

Their analysis found that the 2017 reforms led to limited emigration. Migration responses were largest for those who were paying relatively little UK tax. This meant the fiscal impact of this loss of revenue was limited. Their updated analysis in 2024 showed that emigrants’ UK tax payments did not entirely cease, but instead fell by 60%.

The updated analysis also showed that those affected by the reforms who remained in the UK paid around 155% more UK tax. This effect was sustained across the period from tax year ending 2018 to tax year ending 2020.

Friedman and others (2024) undertook in-depth interviews of 35 individuals in the top 1% of income and/or wealth to explore whether they would consider migrating for tax reasons.

They found that none of the individuals were leaving the UK for tax reasons. For those who were leaving, tax reasons were not the central driving force. They also found the vast majority would never consider migrating for tax reasons. Key ‘pull’ factors of the UK included cultural infrastructure, social ties, private schools and health care, and a more general attachment to British culture and values.

To summarise, the literature suggests that personal tax can be a factor in migration but other factors such as corporation tax, employment, location or agglomeration effects and family affairs are important factors in making migration decisions.

Annex 2: Testing the potential deemed domiciled group

We know that not all deemed domiciled taxpayers will declare their deemed status on their SA109 form.

We wanted to identify at least some of those missing to correct for the potential bias in the selection of the deemed domiciled population declared on their SA109 form. We expect to see those taxpayers using transitional protections indicating deemed domiciled status. This could cause a bias in selection as it is likely that taxpayers with more substantial offshore financial affairs are incentivised to use these protections to a greater extent.

We wanted to identify a more complete potential deemed domiciled taxpayer population by examining the characteristics of the non-domiciled population who did not become deemed domiciled.

For individuals who do not indicate their deemed domiciled status, we selected individuals from the tax years ending 2014 to 2017 who had characteristics which suggested that they may become deemed domiciled. There were 4 different characteristics:

1. Individuals who were paying the £50,000, £60,000 or £90,000 remittance basis charge. This suggests that they have a residency length that means they would meet condition B.

2. Individuals who had stated that they had changed their domicile status from their UK domicile of origin to a domicile of choice. This means that if they were a resident in the UK, they would meet condition A.

3. Individuals who had stated that they had moved to the UK over 15 years before the deemed domicile policy change came into force. This suggests that they are long-term UK residents and hence will meet condition B. The live date variable was used to estimate the date they moved to the UK (box 27 of SA109).

4. Individuals who had stated that they are born non-domiciled. It is likely that they are long-term non-domiciled and would meet condition B.

We found these characteristics covered a pool of 48,000 individuals. We narrowed this group of taxpayers down, using their residency lengths to select those who were likely to reach 15 years of UK tax residence out of the previous 20. Within this we found 5,000 individuals who we did not otherwise know had become deemed domiciled.

These taxpayers can change their tax circumstances frequently. Taxpayers can change their residency status, their remittance or arising basis status and they can declare that their unremitted foreign income and gains are less than £2,000. Figure A1 below includes the most frequent transitions for these individuals.

Behavioural transitions of the potential deemed domiciled group

Figure A1: Behavioural transitions of the potential deemed domiciled group

Description of transition	Number of individuals
Arising basis, selected due to date they came to live in the UK, born non-dom or UK domicile of origin change.	3,078
Remittance basis - £90,000 charge	406
Remittance basis - less than £2,000 unremitted income and gains	683
Remittance basis - £50,000 charge	303
Remittance basis - less than £2,000 unremitted income and gains and one year on the Arising basis	47
Remittance basis - £90,000 charge in 2016 then arising basis in 2017	38

The volatility of individual statuses makes it difficult to clearly identify all potentially deemed domiciled individuals. We excluded individuals where their deemed domiciled status is not clear to avoid making excessive assumptions.

Annex 3: Reasoning behind using difference-in-differences approach

Estimating the change in tax revenue due to the deemed domicile policy is an observational study, as the treatment selection is influenced by the subjects’ characteristics. We assessed 3 possible methods to estimate the change in tax revenue caused by the deemed domicile change.

Difference in Differences

This method examines the treatment and control groups before and after the intervention, in this case a tax policy change. The observed trends across time are used to identify the counterfactual trend. This is the trend of the treatment group if it had not been treated. By comparing the counterfactual trend with the trend of the treatment group, the impact of the treatment is estimated.

However, for this method to give a suitable estimation of the treatment effect, the trend lines for the control and treatment groups need to be parallel before the treatment period. We also require a large quantity of high-quality data to ensure a high-quality difference-in-differences estimate.

Regression Discontinuity

In this method, the impact of the treatment is examined by comparing individuals above and below an assigned threshold. It is assumed that individuals above and below this threshold will be similar and hence only the impact of the treatment will be the difference between them. This method is useful when randomised control trials are not a suitable experimental method.

However, this method only examines the individuals around the assigned threshold which means the impact of the treatment for individuals further away from the threshold is not examined. Therefore, it needs to be assumed that the treatment effect does not differ between very different individuals. We have found that this is not the case for deemed domiciled taxpayers.

We did not deem the regression discontinuity method suitable because it would be difficult to assign an appropriate threshold. We are uncertain when individuals started to change their behaviour in response to the policy change and it is likely that individuals would have started their behavioural responses at differing times.

Propensity Score Matching

For this method, we would create a counterfactual group by matching individuals with similar characteristics based on their probability of being treated.

The treatment effect is estimated by comparing the counterfactual group with the treatment group. This method is useful when randomised control trials are not feasible. It usefully estimates the average effect of the treatment on all of those treated and it requires fewer assumptions about the treatment effect if there is sufficient data on factors affecting treatment participation.

Matching individuals can only be done using observable characteristics, so the estimates of the treatment effect could be biased if unobservable characteristics affect the treatment.

Therefore, extensive data on the characteristics of treated and untreated individuals is necessary for this analysis.

The sources supplying the data of these characteristics need to be comparable.

The characteristics used for the matching process need to be constant before and after the treatment. This narrows down the suitable factors that can be used. For example, the use of residency length, which is key in separating the treatment and control groups, would be unsuitable as it is time-variant.

The propensity score matching method was also regarded as unsuitable because there is not enough data to build an accurate and reliable score matching approach. The characteristics on which to base the scores need to be exogenous from the treatment, yet most of the characteristics of the individuals in the data are linked to the deemed domicile policy change. We would also need a larger sample to pursue this method effectively.

Annex 4: Using a logarithmic transformation

The distribution of the tax paid by the treatment and control groups shown in Figure A2 is positively skewed. This indicates that it is likely that the residuals of the regression will also be positively skewed. This is a concern for inference of the coefficients in the regression.

‘Positively skewed’ means that there are more observations at the lower end of the tax paid distribution. For the regression coefficients to give a good interpretation of the average effect of a variable, a normal distribution or a symmetrical distribution around a central point is preferred.

Tax revenue distribution

Figure A2: Tax revenue distribution

Benoit (2011) states that logarithmic transformations are useful in changing a highly skewed variable to a variable that is more normally distributed. The distribution of the logarithm of tax paid (Figure A3) shows that this is the case for the tax revenue dependent variable.

If a variable includes zero values, a small constant is often added to the distribution so it can be logarithmically transformed. Feng, Changyong and others (2014) showed that this practice can have consequences on the significance levels in hypothesis testing.

We have removed the zero tax paid values before the logarithmic transformation so that the tax paid distribution does not need to be translated. Subsequently we see that the treatment effect remained a similar size and significance to the baseline linear regression when the zero tax paid values were removed.

After we apply a logarithmic transformation, we find the logarithm of tax paid is more normally distributed.

Log of tax revenue distribution

Figure A3: Log of tax revenue distribution

Annex 5: Robustness checks for the tax revenue regressions

We use different versions of the remittance basis control groups in terms of residency length. This allows us to check that a similar treatment effect occurs when we use a control group that is less like the deemed domiciled treatment group.

We also use the remittance basis population from the tax year ending 2017. We use different remittance basis populations due to the movement of taxpayers between the arising basis and the remittance basis.

It is important to use more than one year of remittance basis populations due to the volatility of the population. We want to ensure that a similar treatment effect holds when we compare the deemed domiciled treatment group to a different control group.

For the robustness check regressions, we compare the tax year ending 2015 as the pre-policy change tax year to the tax year ending 2018 as the post-policy change tax year so the results are comparable to Figure A3.

The treatment effect remains positive and significant for the different regression specifications when we use the 2018 deemed domiciled taxpayers as the treatment group, as shown in Figure A4.

Robustness check regression results of the 2018 deemed domiciled taxpayers

Figure A4: Robustness check regression results of the 2018 deemed domiciled taxpayers

N	Regression specification changes compared to baseline regression	Treatment Effect β3posti,tdeemeddomi,t
8,825	Linear regression using control group 6-12 years residency	170,364 [45,910, 294,819] (Note 3)
7,951	Log-linear regression using control group 6-12 years residency	0.82 [0.60, 1.04] (Note 3)
5,557	Linear regression using remittance basis 2017 for control and treatment group, 9-15 years residency	162,423 [51,529, 273,317] (Note 3)
4,969	Log-linear regression using remittance basis 2017 for control and treatment group, 9-15 years residency	0.50 [0.17, 0.82] (Note 3)

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

We completed similar robustness checks for the tax years ending 2019 and 2020. We show the results in Figures A5 and A6.

We see an insignificant effect in the tax year ending 2019 when we consider people who were on the remittance basis in 2017. This is because we observe a smaller treatment effect for the tax year ending 2019 in the baseline regression.

We explain this result when examining the event study analysis. The remittance basis population from the tax year ending 2017 may have started to respond to the policy change announcement and already changed in composition from the tax year ending 2015. We suggest that the range of behavioural responses corresponds to wide confidence intervals which result in an insignificant treatment effect.

For the other regression specifications, the treatment effect remains positive and significant.

Robustness check regression results of the 2019 deemed domiciled taxpayers

Figure A5: Robustness check regression results of the 2019 deemed domiciled taxpayers

N	Regression specification changes compared to baseline regression for 2019	Treatment Effect β3posti,tdeemeddomi,t
8,650	Linear regression using control group 6-12 years	103,063 [29,879, 176,246] (Note 3)
7,756	Log-linear regression using control group 6-12 years	0.84 [0.62, 1.07] (Note 3)
5,582	Linear regression using remittance basis 2017 for control and treatment group, 9-15 years residency	52,842 [-106,680, 212,364]
4,969	Log-linear regression using remittance basis 2017 for control and treatment group, 9-15 years residency	0.36 [0.03, 0.69] (Note 2)

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

Robustness check regression results of the 2020 deemed domiciled taxpayers

Figure A6: Robustness check regression results of the 2020 deemed domiciled taxpayers

N	Regression specification changes compared to baseline regression for 2020	Treatment Effect β3posti,tdeemeddomi,t
8,619	Linear regression using control group 6-12 years	236,003 [109,927, 362,078] (Note 3)
7,659	Log-linear regression using control group 6-12 years	0.85 [0.63, 1.08] (Note 3)
5,469	Linear regression using remittance basis 2017 for control and treatment group, 9-15 years residency.	165,872 [41,367, 290,377] (Note 3)
4,862	Log-linear regression using remittance basis 2017 for control and treatment group, 9-15 years residency	0.39 [0.06, 0.72] (Note 2)

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

Annex 6: Placebo regression results

We estimate the treatment effect on the tax years before the policy change was announced to ensure that the treatment effect we identify only appears once the policy change has been announced or begun.

We examined the changes in tax paid for the control and treatment groups comparing the tax year ending 2013 to the tax year ending 2014. We used these tax years as there was no change in policy for non-domiciled taxpayers and so there should be no treatment effect visible.

Figure A7 examines the tax paid in the tax years ending 2013 and 2014 by deemed domiciled taxpayers from the tax years ending 2018, 2019 and 2020 and the remittance basis taxpayers from the tax year ending 2015.

Placebo regression results

Figure A7: Placebo regression results

N	Regression specification	Treatment Effect
5,663	Linear baseline regression, 2018 deemed doms	6,198 [-72,337, 84,733]
5,073	Log-linear baseline regression, 2018 deemed doms	-0.15 [-0.48, 0.18]
5,844	Linear baseline regression, 2019 deemed doms	14,348 [-64,389, 93,084]
5,249	Log-linear baseline regression, 2019 deemed doms	-0.06 [-0.38, 0.27]
5,915	Linear baseline regression, 2020 deemed doms	4,344 [-42,719, 129,599]
5,319	Log-linear baseline regression, 2020 deemed doms	-0.06 [-0.39, 0.26]

Note 1: 95% Confidence errors in parentheses p<0.01

Note 2: 95% Confidence errors in parentheses p<0.05

Note 3: 95% Confidence errors in parentheses p<0.1

We see in Figure A7 that the placebo regressions demonstrate no significant evidence of a treatment effect in the years before the policy change was announced (there are no asterisks after treatment effect estimates).

Annex 7: Further analysis of the volatility in tax paid by deemed domiciled taxpayers

We observe in Figure A8 that the interquartile range (the spread of the middle of the data – the difference between the end of the first quartile and the start of the last quartile) of tax revenue values increases for deemed domiciled taxpayers once the policy change comes into effect. This reflects the variation in tax affairs for deemed domiciled taxpayers.

Comparison of the interquartile range of the tax revenue for the treatment and control groups

Figure A8: Comparison of the interquartile range of the tax revenue for the treatment and control groups

Annex 8: Robustness checks for the mobility regressions

We repeat the regressions estimating mobility effects using different control groups as robustness checks. We used the remittance basis taxpayers from the tax year ending 2017 and used the 6-to-12-year residency length as a different subsection of the control group.

We find similar marginal effects on the coefficients from these robustness checks which suggests that we are examining a real increase in the likelihood of leaving the UK due to this policy change.

For the robustness check regressions, we compare the mobility in the tax years ending 2013 and 2014 to the mobility in the tax years ending 2018 and 2019 in Figure A9. This means that these results are comparable to Figure 24.

Robustness check regression results

Figure A9: Robustness check regression results

N	Regression specification changes compared to baseline regression	Marginal effect of the treatment effect β3posti,tdeemeddomi,t
16,965	Logit using control group 6-12 years	-0.110 (Note 3)
16,965	Probit using control group 6-12 years	-0.0498 (Note 2)
7,767	Logit regression using remittance basis 2017 for control and treatment groups, 9 to 15 years residency	-0.0813 (Note 3)
7,767	Probit regression using remittance basis 2017 for control and treatment groups, 9 to 15 years residency	-0.0659 (Note 3)

Note 1: 95% Confidence error p<0.01

Note 2: 95% Confidence error p<0.05

Note 3: 95% Confidence error p<0.1

We find that the treatment effect varies depending on the control group used but remains negative and significant. Deemed domiciled taxpayers were an additional 4% to 11% more likely to leave the UK due the policy change.

Year fixed effects are included in the specification.

Annex 9: Detail of the Exchequer impact calculation

We estimate that the counterfactual group of non-domiciled taxpayers paid on average £175,000 in tax before the policy change. This is calculated by reducing £224,000 (the β₀ values from Figure 18) by 22% (the β₂  values from Figure 20).

The counterfactual situation is that 4% to 5% of the deemed domiciled taxpayers would have left the UK after the policy change. Taking the midpoint migration elasticity of 4.5%, we calculate the number of the remaining taxpayers as 3,291 for the tax year ending 2018, 3,580 for the tax year ending 2019, and 3,804 for the tax year ending 2020.

We estimate that the average tax paid by the counterfactual group increases by 40% in the tax year ending 2018 compared to the tax year ending 2015 but the tax paid remains unchanged from the tax year ending 2015 in between the tax years ending 2015 and 2019 and 2020. Results are shown in Figure A10 below.

Estimation of the average tax paid by the counterfactual deemed domiciled group

Figure A10: Estimation of the average tax paid by the counterfactual deemed domiciled group

	2015	2018	2019	2020
Number of counterfactual deemed domiciled taxpayers	3,446	3,291	3,580	3,804
Percentage difference in tax paid from the £224,000 value	-22%	18%	-22%	-22%
Tax paid by counterfactual deemed domiciled taxpayers (£)	175,000	264,000	175,000	175,000
Total tax paid by counterfactual deemed domiciled taxpayers (£million)	603	869	627	666

We estimate that the treated group of non-domiciled taxpayers also paid on average £175,000 in tax before the policy change.

The impact of the policy change is that 10% to 12% of the deemed domiciled taxpayers would have left the UK. Taking the midpoint migration elasticity of 11%, we calculate the number of remaining taxpayers as 3,067 for the tax year ending 2018, 3,337 for the tax year ending 2019, and 3,545 for the tax year ending 2020.

The treatment effect estimated shows that the tax paid by condition B deemed domiciled taxpayers increased by 114%, 108% and 146% in the tax years ending 2018, 2019 and 2020 respectively (β₃ values from Figure 20). Results are shown in Figure A11 below.

Estimation of the average tax paid by the treated deemed domicile group

Figure A11: Estimation of the average tax paid by the treated deemed domicile group

	2015	2018	2019	2020
Number of treated deemed domiciled taxpayers	3,446	3,067	3,377	3,545
Percentage difference in tax paid from the £224,000 value	-22%	132%	86%	124%
Tax paid by treated deemed domiciled taxpayers (£)	175,000	520,000	417,000	502,000
Total tax paid by treated deemed domiciled taxpayers (£million)	603	1,595	1,391	1,780