Joint Research Centre the European Commission's in-house science service

Serving society

Stimulating innovation

Supporting legislation

Measuring the fiscal and equity impact of tax

evasion in the EU: Cross-country evidence using

microsimulation modelling

Sara Riscado XXIV Encuentro de Economía Pública

Toledo, 26-27 January 2017

This is a joint work with Salvador Barrios (EC-JRC), Bent Greve and M. Azhar Hussain (Roskilde University), Alari Paulus (ISER-University of Essex) and Fidel Picos (EC-JRC).

The views expressed are purely those of the authors and may not in any circumstances be regarded as stating an official position of the European Commission.

26/1/2017 2

Outline

1. Motivation

2. Data

3. Estimation: methodology and results

4. Fiscal and distributional effects

5. Conclusion

26/1/2017 3

1.Motivation

“Si el IVA lo pagaran los que tienen que pagarlo, no habría que subirlo

tanto”

Cristóbal Montoro, Spanish Minister of Finance, 9th July 2012

Tax evasion:

o Shifts the tax burden from evaders to non-evaders, distorting

consumption and labour supply decisions

o Undermines the social contract between the state and the

taxpayers

o Weakens the redistributive nature of the tax and benefit system

Effects especially relevant in times of severe economic crisis,

when strong fiscal consolidation is required

26/1/2017 4

1.Motivation Our goal is threefold:

1. To estimate individual measures of employment income underreporting

2. To measure the fiscal and distributional impact of eliminating tax non-compliance

3. To progressively extend EUROMOD coverage of tax evasion:

o EUROMOD is a tax-benefit calculator for the 28 EU MS

o Uses survey data from EU-SILC as input microdata

o Need of accurate measures of true income

o Up to now: Italy, Greece and Bulgaria can be corrected for accounting for non-compliance in a very aggregate way

26/1/2017 5

This paper…

Combination of micro-econometric and

microsimulation approaches

Use of administrative data together with survey

data

Focus on two countries: DK and EE

Simulations using EUROMOD

26/1/2017 6

2. Data

Quantification of tax evasion directly linked to the availability and

quality of survey and administrative data:

o EE: Exact respondent matching between SILC data and

individual tax records, additionally pre-populated by third

parties (employers)

o DK: SILC data drawn from tax records need of

complementary information (hidden economy surveys)

and national aggregates

26/1/2017 7

2. Data: EE

2008 wave of national SILC for EE exactly matched with tax

records

o 14,942 individual observations in SILC, 99.5% linked with

tax records complete employment information for

10,237 observations

o No consent bias (no consent required to link datasets)

o Information complemented with information from

employers

o Income from tax records is on average 87.5% of survey

income (excluding zero earnings)

26/1/2017 8

2. Data: DK

2011 wave of national SILC for DK, based on administrative data

+

Cross-section studies on the hidden economy (Hvidtfeldt et al.

2010, Skov 2014, Skov et al. 2015) 'black activities' but also

free exchanges of services:

o Representative sample population aged between 18 and 74

o Covering the period 1994-2009, with a final total of

respondents of 23000 in the final set

o Includes also individual and household information on

demographic, education, income, and labour market

26/1/2017 9

3. Estimation – Methodology: EE

o Estimation of true earnings/propensity to comply as a

latent variable (Paulus, 2015)

o Identification hypothesis:

i. Civil servants do not lie, i.e. they cannot evade, so any

difference between what they declare to the interviewer

and the tax agency is due to measurement error

ii. Measurement errors are distributed across all the

sample, in the same way as they are distributed across the

sub-sample of civil servants

o Survey earnings-measurement error = true earnings

=register earnings + non-reported earnings 26/1/2017 10

3. Estimation – Methodology: EE

𝑦𝑖𝑟 =

0 if 𝑦𝑖𝑇 = 0 (no earnings)

0 if 𝑦𝑖𝑇 > 0 and 𝑟𝑖

∗ ≤ 0 (full non−compliance)

𝑟𝑖∗ ⋅ 𝑦𝑖𝑇 if 𝑦𝑖

𝑇 > 0 and 0 < 𝑟𝑖∗ < 1 (partial compliance)

𝑦𝑖𝑇 if 𝑦𝑖

𝑇 > 0 and 𝑟𝑖∗ ≥ 1 (full compliance)

ln 𝑦𝑖𝑇 = 𝑥𝑖𝛽

𝑇 + 𝜀𝑖𝑇 , 𝜀𝑖𝑇∼ 𝑁 0, 𝜎𝑇

2 True earnings

Register earnings

ln 𝑦𝑖𝑠 = 𝜃𝑠 ln 𝑦𝑖

𝑇 ⋅ 1 𝑦𝑖𝑇 > 0 + 𝜃0

𝑠 ⋅ 1 𝑦𝑖𝑇 = 0 + 𝑥𝑖𝛽

𝑠 + 𝜀𝑖𝑠, 𝜀𝑖𝑠 ∼ 𝑁 0, 𝜎𝑠

2 Survey earnings

𝑟𝑖∗ = 𝜃𝑟𝑦𝑖

𝑇 + 𝑥𝑖𝛽𝑟 + 𝜀𝑖

𝑟 , 𝜀𝑖𝑟 ∼ 𝑁 0, 𝜎𝑟

2 Propensity to comply

Estimation of the overall probability density function for a pair of observed

individual earnings 𝑦𝑖𝑟 , 𝑦𝑖𝑠 conditional on true earnings through maximum

likelihood

ln 𝐿 = ln 𝑓 𝑦𝑖𝑟 , 𝑦𝑖𝑠

26/1/2017 11

3. Estimation – Results: EE

Estimated (true) status of employed individuals (%)

o Private employees: low full non-compliance around 4% and

partial compliance around 30%

o Including public employees: the fully and partially compliant

groups drop to about 3% and 23% of the sample

Private employees All employees

No earnings 0.8 1.0

Fully non-compliant 3.9 3.1

Partly compliant 29.0 22.8

Fully compliant 66.3 73.2

26/1/2017 12

3. Estimation – Results: EE

Misreported earnings as a share of total gross true earnings (%)

o Substantial variation across the distribution of (true) earnings U-

shaped profile

o High variation in survey mismeasurement Tendency of people to

present themselves more similar to the rest than they really

are 26/1/2017 13

-20

-10

0

10

20

30

40

1 2 3 4 5 6 7 8 9 10 All

%

Deciles of true earnings

Tax non-compliance Measurement error

3. Estimation – Methodology: DK

o “Tax evasion” behaviour assumed to have three components:

participation, number of hours per week, and the hourly

wage rate

o Regression equations dependent on individual characteristics,

such as gender, age, family status, income levels, etc

o For people participating in “tax evasion” activities, estimated

non-reported income is obtained as the product between

estimated hours spend and estimated wages earned on "hidden

activities"

26/1/2017 14

3. Estimation – Methodology: DK

• Decision to participate: logit regression and ranking of

individuals (from highest to lowest) until reaching the national

average in 2011 (Skov, 2014)

• Average number of weekly hours devoted to hidden activities,

conditional on gender and age assigned to the individuals

already found to participate in tax evading activities

• Log of weekly wages: OLS regression fitted to the national tax

evasion hourly wage in 2011 (Skov, 2014)

• For those participating: 𝑵𝒐𝒏 − 𝒓𝒆𝒑𝒐𝒓𝒕𝒆𝒅 𝒊𝒏𝒄𝒐𝒎𝒆 =

𝑯𝒐𝒖𝒓𝒔 𝒑𝒆𝒓 𝒘𝒆𝒆𝒌 𝒄𝒉𝒂𝒓𝒂𝒕𝒆𝒓𝒊𝒔𝒕𝒊𝒄𝒔 𝒆𝒗𝒂𝒅𝒆𝒓

∗ 𝒘𝒂𝒈𝒆 𝒓𝒂𝒕𝒆 𝒄𝒉𝒂𝒓𝒂𝒕𝒆𝒓𝒊𝒔𝒕𝒊𝒄𝒔 𝒆𝒗𝒂𝒅𝒆𝒓 ∗ 𝟓𝟐

26/1/2017 15

3. Estimation – Results: DK

Estimated (true) status of individuals (%)

o More than two thirds of employees are fully compliant, around

24% engaged in hidden activities.

o 6.2% of the population did not declare employment income but

were involved in “hidden activities” (fully non-compliant)

Employees Whole population

Fully non-compliant - 6.2

Partly compliant 23.5 16.7

Fully compliant 76.5 77.1

26/1/2017 16

3. Estimation – Results: DK

Misreported earnings as a share of total gross true earnings (%)

o On average unreported income accounts for around 26% for partially

compliant employees and 7% for employees.

o Decreasing patterns on both graphs, except for the first decile where

there are few “evaders” who “hide” big shares of their incomes.

0

10

20

30

40

50

60

70

80

90

1 2 3 4 5 6 7 8 9 10 All

%

Deciles of true positive earnings

All employees Partially compliant employees

26/1/2017 17

4. Fiscal and distributional effects

Scenario Income Tax and benefits

True baseline True Based on declared income

No tax evasion True Based on true income

26/1/2017 18

4. Fiscal and distributional effects: EE

Aggregate components of disposable income (million EUR)

Tax evasion No tax evasion Difference

Total Total Total Standard

error

95% confidence interval % of

baseline

Lower

bound

Upper

bound

Original income 5,854 5,854 0 - - - 0.0

Taxes 874 995 121 6 109 134 13.9

Social Insurance

Contributionsa 105 115 9 1 8 10 8.9

Benefits 1,209 1,200 -10 2 -13 -6 -0.8

Disposable income 6,084 5,944 -140 7 -154 -126 -2.3

Inequalityb 0.332773 0.330775 -0.001998 0.000564 -0.003104 -0.000892 -0.6

a. Employees and self-employed

b. Gini coefficient of equivalised disposable income.

o Declared employment income increases taxes and social insurance

contributions increase (13.9% and 8.9% respectively) and benefits go

down (-0.8%) disposable income decreases (2.3%).

o Inequality slightly decreases.

26/1/2017 19

4. Fiscal and distributional effects: EE

Distributive impact of tax compliance on household disposable

income (change as % of household disposable income)

-4

-2

0

2

4

6

8

10

12

14

1 2 3 4 5 6 7 8 9 10 All

Ch

an

ge

as

% o

f d

isp

osa

ble

in

co

me

Deciles of equivalised disposable income

Reported original income Taxes

Social Insurance Contributions (employee) Benefits

Disposable income

26/1/2017 20

o 10% increase of reported income increase in taxes (2%) and SIC (less

than 1%), no effect on benefits 2.3% reduction of disposable income.

o Changes in disposable income mirror changes in taxes for all deciles.

4. Fiscal and distributional effects: DK

Aggregate components of disposable income (million EUR)

Tax evasion No tax evasion Difference

Total Total Total Standard

error

95% confidence interval % of

baseline Lower

bound

Upper

bound

Original income 1,029,445 1,029,445 0 - - - 0.0

Taxes 351,211 372,175 20,964 791 19,412 22,515 6.0

Social Insurance

Contributionsa 91,728 97,553 5,825 210 5,413 6,237 6.3

Benefits 312,050 307,677 -4,373 746 -5,836 -2,910 -1.4

Disposable income 898,555 867,393 -31,162 1,188 -33,491 -28,833 -3.5

Inequalityb 0.250311 0.250480 0.000169 0.000792 -0.001384 0.00172220 0.1

a. Employees and self-employed

b. Gini coefficient of equivalised disposable income.

26/1/2017 21

o Effect have the same direction than in Estonia, but smaller relative

changes in taxes (6%) and SIC (6.3%) have higher impact on disposable

income (3.5%).

o No significant change in inequality

4. Fiscal and distributional effects: DK

Distributive impact of tax compliance on household disposable

income (change as % of household disposable income)

-6

-4

-2

0

2

4

6

8

10

12

1 2 3 4 5 6 7 8 9 10 All

Ch

an

ge

as

% o

f d

isp

osa

ble

in

co

me

Deciles of equivalised disposable income

Reported original income Taxes

Social Insurance Contributions (employee) Benefits

Disposable income

26/1/2017 22

o Increase in reported original income (7%) lower than in Estonia, but

effect on disposable income higher (3.5%) due to the combined effect of

the more significant effect on taxes, social contributions and benefits.

o Taxes more relevant in higher deciles, benefits in lower.

5. Conclusion

o Data availability conditions methodologies used and results

obtained

o Linking survey and administrative data is fundamental for

disentangling tax evasion and measurement error, as in EE

o Taking tax evasion into account is important to make accurate

policy recommendations even in countries with high tax

compliance, as in DK

26/1/2017 23