The Effect of VAT Threshold on the Behavior of Small Businesses: Evidence and Implications

The Effect of VAT Threshold on the Behavior of Small Businesses: Evidence and Implications Jarkko Harju, Tuomas Matikka and Timo Rauhanen The Eect ...
Author: Roberta Nichols
24 downloads 2 Views 1MB Size
The Effect of VAT Threshold on the Behavior of Small Businesses: Evidence and Implications Jarkko Harju, Tuomas Matikka and Timo Rauhanen

The Eect of VAT Threshold on the Behavior of Small Businesses: Evidence and Implications ∗

Jarkko Harju, Tuomas Matikka and Timo Rauhanen March 30, 2015

Preliminary version

Abstract Small businesses are often regarded as important determinants of economic growth. Simultaneously, many tax rules and regulations are size-dependent, which might decrease eciency and the economic activity of growing rms. We study the eects of the value-added tax (VAT) threshold on the behavior of small businesses. In Finland, rms with annual sales below 8,500 euros are not liable to pay VAT. We use detailed register data on the universe of Finnish businesses and the bunching method to provide robust and clear evidence of behavioral eects of the threshold. We nd that the VAT threshold has notable eects among small business. Firms bunch actively just below the threshold, which implies notable eciency implications. We nd that changing tax incentives at the threshold does not have a signicant eect on the extent of the response. This implies that compliance costs are important in explaining observed responses. We nd no evidence of tax avoidance or evasion, which suggests that rms respond by reducing output. Also, we nd that bunching behavior is relatively permanent, which implies that the threshold decreases the growth of small businesses. Keywords: Small businesses, value-added tax, VAT threshold, bunching JEL codes: H21, H25, H32 ∗ Government Institute Timo.Rauhanen@vatt.

for

Economic

Research

VATT.

1

Jarkko.Harju@vatt.,

Tuomas.Matikka@vatt.,

1

Introduction

Small and especially young businesses are often regarded as important determinants of economic growth (see e.g. Haltiwanger et al. 2013, Decker et al. 2014). Simultaneously, many tax rules and regulations are size-dependent. These rules might reduce the eciency of taxation and decelerate economic activity, in contrast to widespread objectives to enhance the growth of small businesses. Value-added tax (VAT) is a commonly applied form of consumption taxation, and a crucial component of tax revenue in many countries. Most VAT systems include varying thresholds below which rms are exempt from remitting VAT. For example, in the EU, VAT threshold varies between 0-100,000 euros. Despite the potential detrimental eects of size-dependent thresholds, there is only limited evidence on the eects of VAT threshold on the behavior of small businesses. In Finland, rms with annual sales below 8,500 euros are not liable to pay VAT and separately report sales to the Tax Administration. Relatively low VAT thresholds are common. Half of the EU countries apply thresholds below 25,000 euros, including for example Germany, Belgium and Denmark. In general, small rms comprise a large share of all businesses. In Finland, one third of all registered rms have turnover below 25,000 euros. Among young and potentially growing rms, the share of businesses with small turnover is even larger. Over 40% of rms that are younger than three years have turnover below 25,000 euros. In this study we present comprehensive evidence on the eects of the VAT threshold among small businesses. We utilize detailed data on the universe of Finnish businesses, including also rms below the VAT threshold. We use the bunching method in order to provide clear and robust evidence on behavioral eects. To understand the implications of the VAT threshold, it is important to know both why and how rms respond to it. By utilizing changes in VAT rules at the threshold, we analyze the role of both tax incentives and compliance costs. We study the anatomy of the response to learn whether rms react by changes in real economic activity or by tax avoidance and evasion. In order to illustrate the important dynamic aspects of the VAT threshold, we analyze how the threshold aects growth and development of small businesses. First, rms could respond to the VAT threshold both because of increased taxation and increased compliance costs above it. We utilize variation in tax incentives and compliance costs to analyze why rms react to the threshold.

Before 2004, VAT liability increased sharply at the threshold, as rms

marginally above the threshold were liable to fully pay the VAT on all sales. In 2004, Finland introduced a VAT relief scheme, in which remitted VAT increases gradually above the threshold. Thus the reform drastically changed tax incentives at the threshold, which allow us to disentangle the eects of tax incentives and compliance costs. Second, rms can respond to the threshold by reducing sales, or by engaging in various tax avoidance activities or underreporting of sales. We analyze the nature of the response by studying how the production factors of the rms, such as equity and expenses, develop around the VAT threshold. Potential discontinuous changes in production factors exactly at the VAT threshold indicate changes in behavior caused by this regulation, and shed light on how rms respond to the threshold. Third, in terms of welfare eects, it is essential to know how the VAT threshold aects the growth of small businesses. The threshold could signicantly hinder growth if rms avoid exceeding the threshold for a prolonged period of time. The panel structure of the data allow us to follow rms over time, which enable us to characterize the eect of the threshold on growth and the scale of business activity. We nd that the VAT threshold has notable eects among small business. just below the threshold, which implies signicant eciency implications.

Firms bunch actively

We nd that changing tax

incentives at the threshold does not signicantly decrease the eect, which implies that compliance costs are important in explaining observed behavior.

We nd no clear evidence of tax avoidance or

evasion, which suggests that rms respond by changes in real economic activity. Finally, we nd that

2

bunching behavior is relatively permanent, which implies that the threshold decreases the growth of small businesses. Despite the scal importance of VAT and the generally applied sales thresholds, only a few previous papers study the eects of these thresholds. The theoretical literature has focused on determining the rules for optimal VAT threshold. Keen and Mintz (2004) and Kanbur and Keen (2014) show that the optimal VAT threshold depends on administrative and compliance costs, and the extent of the eect of the threshold on rm behavior. Empirically, Onji (2009) was the rst to detect clear eects of the VAT threshold on the distribution of rms in Japan. He shows that relatively large Japanese rms reacted to the introduction of a VAT threshold by splitting into smaller entities, reecting clear tax avoidance behavior. Li and Lockwood (2014) show that rms in the UK bunch actively at the relatively large VAT threshold (approx.

¿90,000).

Also, Waseem (2015) observes a strong clustering of rms at the VAT

threshold in Pakistan. This paper proceed as follows: Section 2 describes the VAT system and the VAT threshold in Finland. Section 3 presents the methodology and Section 4 describes the data. Section 5 oers the results and Section 6 concludes the study.

2

Institutions

2.1 Value-added taxation Most developed countries use the value-added tax (VAT) as their primary consumption tax system. VAT is usually a broadly based tax assessed on the value added to goods and services. The amount of value added is calculated by subtracting the amount of externally purchased goods and services from the value of goods and services produced. In short, the VAT assessment process is the following:

each trader in the chain of supply from

manufacturers to retailers charges VAT on the sales. Firms are entitled to deduct from this amount the VAT paid on purchases. VAT is remitted to the tax authorities by the seller of the goods and services. VAT is the main source of tax revenue in many developed countries. For example, among all OECD countries almost one-fth of all tax revenue is collected by VAT. However, the variation in VAT revenues is large across countries. A common feature in many VAT systems is that rms with annual sales under a certain threshold are not required to register and remit VAT. Figure 1 depicts these annual sale thresholds among OECD countries in 2014. The Figure shows thresholds vary notably across countries. While some countries levy VAT on all turnover without any threshold (e.g. Sweden and Turkey), some countries apply relatively high thresholds around 100,000 euros (e.g. Switzerland and the UK).

2.2 VAT in Finland Finland, as a member of the EU, applies the general EU VAT legislation. All members of the EU apply a standard rate of at least 15%. The EU allow member countries to use a maximum of two reduced VAT rates for specic products and services, such as food and pharmaceuticals. The standard VAT rate in Finland is 24% in 2014 that applies to most of the goods and services sold. Finland uses two reduced rates: 14% is applied to e.g. food and restaurant services, and 10% is applied to e.g. books and pharmaceuticals. VAT registered rms are obliged to regularly le periodic tax returns to the Finnish Tax Administration. The ling and reporting obligation covers all VAT on sales at dierent rates, input purchases, zero-rated sales, imports and exports.

The frequency of the required reports depends on the annual

sales of a rm: Firms with annual sales below 25,000 euros are allowed to report annually, rms with

3

VAT thresholds in OECD countries in 2014 (in euros) Chile Mexico Spain Sweden Turkey Netherlands Greece Belgium Norway Iceland Denmark Finland Portugal Estonia Israel Germany Hungary Korea Canada Luxembourg Average Austria Italy Poland Czech Republic New Zealand Slovak Republic Slovenia Australia Japan Ireland France Switzerland United Kingdom

0

20,000

40,000 60,000 Sales (in euros)

80,000

100000

Source: OECD Statistics

Figure 1: Annual sale thresholds for VAT registration among OECD countries in 2014 (in euros)

turnover 25,000-50,000 euros must report quarterly, and rms with sales above 50,000 euros have to report monthly. Some sectors and industries are exempt from VAT or have other special rules for paying VAT. These include nancial and insurance activities, letting and operation of dwellings, education, human health and social work activities. A rm that sells solely these goods or services are not liable to pay VAT.

VAT threshold before 2004.

In Finland, the VAT threshold for rms is 8,500 euros of annual sales.

Below this threshold rms are exempt from VAT. The threshold has remained constant from 1995, even in nominal terms. Albeit small businesses below the threshold are exempted from VAT, they need to pay other taxes and report their income to the Tax Administration. Firms that exceeded the threshold paid VAT for sales, including sales below the threshold. Thus the average tax rate jumps at the threshold before 2004. Firms that do not exceed the thresholds can voluntarily register and pay VAT. There are logical reasons for registering even when it is not necessary. A rm can only deduct VAT from purchases and costs if registered. For example, voluntary registration could be important for businesses that have large start-up costs. Also, rms below the threshold that have a large share of business-to-business sales have an incentive to register, as the VAT rebate is only possible from purchases of VAT registered rms.

VAT relief scheme from 2004 onwards.

The VAT rate at the threshold changed in 2004 although

the threshold itself remained at 8,500 euros. The reform introduced a VAT relief scheme for annual sales below 20,000 euros in 2004 and 22,500 euros 2005 onwards. After the reform, rms can apply for a VAT relief that gradually decreases (above 8,500 euros) with the increase in sales. Figure 2 shows the VAT remittance (above) in euros and average tax rates (below) for dierent levels of total annual sales of rms for dierent years. The data is split to 100 euro turnover bins in the Figure. The Figure shows the introduction of the relief region in 2004 and post-2005 in comparison to pre-2003

4

period for a representative rm that is subject to the standard VAT rate. The Figure clearly depicts that the pre-reform system created a salient VAT notch, implying a jump in remitted VAT and the average VAT rate from 0 to 22% at the threshold (standard VAT was 22% rate in 2003-2009 in Finland). After the reform the notch was replaced by a VAT kink, implying a gradually increasing remitted VAT and average VAT rates above the threshold. Within the relief scheme, gradually increasing average VAT rate implies an increasing marginal VAT rate up to the point in which the average VAT rate equals 22%. This leads to marginal VAT rates between 13-57% within the relief region, which was 8,500-20,000 euros in 2004 and 8,500-22,500 euros from 2005 onwards. An additional important detail of the VAT relief is that it is not automatically granted by the Tax Administration. Firms needed to apply for the VAT relief using a separate tax form before 2010. From 2010 onwards, rms can apply for the VAT relief with the same periodic tax form they use to declare VAT. This can have important implications for the salience of the VAT relief.

Remitted VAT and average tax rates before and after the reform

0

Remitted VAT 3000 6000

Remitted VAT

0

5000

10000

15000

20000

25000

30000

20000

25000

30000

Average tax rate (%) 0 5 1015202530

Average tax rates

0

5000

10000

15000 Annual turnover

VAT pre−2003 VAT post−2005

VAT 2004

Figure 2: VAT remittance and average VAT rates for dierent levels of sales before and after the introduction of VAT relief region

3

Methodology

3.1 Bunching at the VAT threshold We use the bunching methodology introduced in Saez (2010) to analyze responses to the VAT threshold. The intuition behind the bunching approach is that if a discontinuous change in tax liability at the threshold aects the behavior of rms, we should nd an excess mass of rms located at the threshold. Consider a rm which is owned and managed by a single entrepreneur

1

that maximizes the following

function

1 As the VAT threshold in Finland is low, most rms around it are managed and owned by a single owner. Therefore, it is reasonable to assume that individual owners make the relevant decisions on s. However, for conceptual simplicity, throughout the paper we denote that rms respond to the VAT threshold, not individual owners.

5

π = (s − d(s))(1 − τp ) − c(s) − τvat d(s) − [T (s)vat − τvat d(s) + δ(s)] · 1(s > s∗ ) where

s

denotes annual sales, and

d(s)

(1)

is a concave funtion of tax-deductible costs needed to generate

s.

We assume that the marginal unit of sales produces positive net income for the rm, which implies that

d0 (s) ≤ 1.

Net income from the rm

(s − d(s))

is taxed at a at income tax rate

function of the cost of eort of the owner, which is not tax-deductable. where

τvat

T (s)vat

denotes the at VAT rate.

τ vat d(s)

τp . c(s)

is a convex

is VAT paid on

denotes the convex VAT function, and

δ(s)

d(s),

represents

compliance costs related to VAT reporting. The rm is not liable to report and pay VAT below a sales threshold

0

if



s≤s

, and thus exceeding



s

s∗ .

Therefore,

[T (s)vat − τvat d(s) + δ(s)] =

creates a jump in both remitted VAT and compliance costs. However,

above the threshold the rm can deduct the VAT on tax-deductible costs from remitted VAT. Below

s∗

VAT paid on purchases is not tax-deductable.

(s∗ −,s∗ ) below the VAT threshold. 0 0 0 Maximizing π with respect to s below the threshold implies that c (s) = (1 − d (s))(1 − τp ) − τvat d (s). ∗ 0 0 0 0 At s , maximization yields c (s) = (1 − d (s))(1 − τp ) − T (s)vat − δ (s). Let us assume that d(s) is ∗ ∗ 0 approximately linear in sales within (s − ,s ), which implies that d (s) is approximately equal within Let us consider rm decision making within a small sales interval

this region. Firms have incentives not to exceed the threshold because the marginal cost of additional sales is larger at the threshold than just below it,

τvat d0 (s) ≤ T 0 (s)vat + δ 0 (s).

Intuitively, an additional

unit of sales is less valuable at the VAT threshold because the rm needs to pay both VAT and compliance costs if

s∗

is exceeded.

First, we study the eect of the change in the VAT rate at the threshold. For now we ignore compliance costs, which we will study in Section 3.3. In Finland, there has been two kinds of changes in the VAT rate at the threshold: a VAT notch and a VAT kink.

To start with the VAT notch, consider a VAT

schedule where sales are not taxed until the notch point

s∗ .

applied to all sales. Thus the VAT liability jumps discretely at the sales below

s∗

If sales exceed



s

s∗ ,

the VAT rate will be

, as the rm needs to pay VAT also for

if the threshold is exceeded. More formally, the VAT function in equation (1) in the

notch schedule is of the form

Tvat = s ∗ τvat · 1(s > s∗ ),

where

τvat

is the at VAT rate.

Bunching behavior at the VAT notch is illustrated in the upper graph of Figure 3.

The vertical

axis denotes the net-of-tax sales, and horizontal axis denotes sales before taxes. The straight blue lines illustrate the tax rates, and curvy red lines the indierence curves of dierent rms (type A and type B).

4τvat

represents the VAT paid from sales below the threshold once the VAT threshold is exceeded.

A fraction of rms originally above

s∗

will locate themselves at the VAT threshold after the introduc-

tion of a discontinuous jump in VAT liability. The extent of this bunching behavior depends on the sales elasticity with respect to VAT rate, which we will come back to in more detail below. Firms originally at

s∗

or below the threshold will not change their behavior after the introduction of the notch (type

A rm).

In the graph,

s∗ + 4s

denotes the hypothetical rm with the highest sales to bunch at the

threshold (type B rm). In other words,

s∗ + 4s

marks the last rm bunching at the notch, which we

call the marginal buncher. More formally, the fraction of rms located at is denoted as

B(∆s) =

´ s∗ +∆s s∗

h0 (s)ds,

where

h0 (s)

absence of the notch.

6

s∗

in response to the notch

denotes the counterfactual density of sales in the

Indifference curves

Net-of-tax sales

Type B

Type A Slope (1-τp- τvat)

∆ τvat

Slope (1-τp)

sB s*+∆s

s*

sales

Indifference curves Type B

Net-of-tax sales

Type A Slope (1-τp- τvat)

Slope (1-τp)

s*

s*+∆s

sales

Figure 3: Bunching at a VAT notch (upper graph) and a VAT kink (lower graph)

Bunching at the VAT kink is illustrated on the lower graph side of Figure 3. In the VAT kink system,

s∗ are taxed at the VAT rate, and the VAT function in equation (1) is of the form Tvat = (s − s ) ∗ τvat · 1(s > s∗ ). Type A rm which is located at s∗ before the introduction of the ∗ VAT kink will not respond to the kink, whereas a fraction of rms above s will locate themselves at the only sales exceeding



VAT kink. As with notches, type B rm in the graph represents the marginal rm with the largest sales



(s

+ 4s)

to bunch at

s∗

after the introduction of the VAT kink.

Intuitively, the main dierence between VAT notch and VAT kink is the size of the change in tax

7

incentives at the threshold. Compared to VAT notch, a VAT kink produces notably smaller incentives to respond. Therefore, it is presumable that less rms will bunch at the VAT kink than at the VAT notch. Figure 4 describes bunching in the sales distribution.

In the Figure, the horizontal axis denotes

the number of rms and vertical axis denotes sales levels. The solid blue line represents observed sales distribution, and the dotted red line the counterfactual density in the absence of the threshold. excess mass caused by the threshold is presented as a spike in the distribution at



s



s

The

. The excess mass at

comes from the missing above the threshold. The missing mass above the threshold is denoted as the

area between the counterfactual distribution and the obseved distribution within the region

(s∗ , s∗ +4s).

Assuming smooth and heterogenous sales elasticities across dierent rms, the observed density gradually approaches the counterfactual density above

s∗ .

Thus

s∗ + 4s

represents the rm with the largest sales

to bunch at the threshold. Intuitively, the larger the excess mass at the threshold is the further away from

s∗ comes

the last rm to bunch at the VAT threshold. We discuss the empirical estimation in more

detail below.

Number of firms

Excess mass

Observed distribution Counterfactual

Missing mass

s*

s*+∆s

Sales

Figure 4: Bunching at the VAT threshold

Abstracting from compliance costs, there are also circumstances in which a rm has no tax incentive to bunch at the VAT threshold.

The main instance is substantial VAT paid on purchases stemming

from, for example, large start-up costs. In other words, for some rms it could be that

T (s) < τvat d(s)

above the VAT threshold, and thus (marginally) exceeding the threshold does not increase tax liability.

d0 (s) > 1

Second, it could be that

for some small businesses in the short run, which might not induce

incentives to bunch at the threshold as

τvat d0 (s) > T 0 (s).

However, small businesses in our data are on

average protable and have notably larger level of sales compared to overall expenses, which indicates that incentives to bunch at the VAT threshold exist for a large proportion of small rms in Finland.

3.2 VAT rate elasticities based on observed bunching We approximate the sales elasticity at the VAT threshold using a similar approach as Kleven and Waseem (2013). We characterize the elasticity at VAT notch and VAT kink by relating the earnings response of a



marginal buncher rm (s

+ ∆s)

to the change in tax liability caused by exceeding the threshold by

8

∆s.

This upper-bound reduced-form approximation of the sales elasticity oers a conveivable way to scale the extent of the behvioral response to the threshold with the change in the VAT rate under dierent VAT rate schedules. Elasticity at the VAT notch is calculated with the following quadratic formula

eN ≈ (4s/s∗ )2 /4tN

(2)

4tN = (4s + s∗ )τvat /4s denes the relative increase in VAT payments caused by exceeding the threshold by ∆s. Importantly, when exceeding the VAT notch, the rm needs to pay VAT also for sales ∗ below s . where

Sales elasticity associated with VAT kink can be written as

eK ≈ (4s/s∗ )2 /4tK

(3)

4tK = (4s)τvat /4s = τvat . Compared to the VAT notch, the rm needs to pay VAT only for ∗ sales above s within the VAT kink system, and thus the denominator of equation (3) reduces to the at where

VAT rate. Equations (2) and (3) imply that the change in the implicit marginal tax rate (4tN , at VAT notch compared to VAT kink.

4tK )

is larger

This is creates larger incentives to bunch at the VAT notch.

Therefore, assuming similar underlying (structural) elasticity regardless of the VAT system, we should nd larger excess bunching at the VAT notch compared to the VAT kink.

3.3 Compliance costs of VAT reporting [To be added here later]

3.4 Empirical analysis The excess mass of rms at the VAT threshold is estimated by comparing the actual density function around the threshold to an estimated smooth counterfactual density. The counterfactual density function describes what the distribution of sales would have looked like without changes in tax liability at We follow the methods in Chetty et al.

s∗ .

(2011) and Kleven and Waseem (2013) to estimate the

counterfactual density. Intuitively, the counterfactual density is estimated by tting a exible polynomial function to the observed distribution, excluding an area around we re-center income in terms of



s

s∗

from the observed distribution. First,

, and group rms into small sales bins of 100

¿.

We estimate a

counterfactual density by regressing the following equation and excluding the region around the threshold

[sL , sH ]

from the regression

cj =

p X

βi (sj )i +

i=0 In equation (4),

cj

cˆj =

Pp

i=0

p.

ηi · 1(sj = i) + εj

(4)

i=sL

is the count of rms in bin

of the polynomial is denoted by

sH X

j,

and

sj

denotes the sales level in bin

j.

The order

Thus the tted values for the counterfactual density are given by

βi (sj )i .

The excess bunching is estimated by relating the actual number of rms close to the threshold within

(sL , s∗ ) to the estimated counterfactual density within the same region. Ps∗

i=sL ˆb(s ) = P s∗ ∗

(cj − cˆj )

ˆj /Nj i=sL c

where

Nj

is the number of bins within

[sL , s∗ ].

9

We calculate excess bunching as

(5)

One important question when estimating the counterfactual density is how to determine the excluded the region

[sL , sH ].

As in earlier literature, we determine the lower limit

of the sales distribution. Intuitively,

sL

sL

based on visual observations

represents the point in the sales distribution where the bunching

behavior begins. We follow the approach of Kleven and Waseem (2013) to dene the upper limit. We determine

Ps∗ such that the estimated excess mass ˆ bE (s∗ ) = ( i=sL cj − cˆj ) equals the estimated PsH ˆj − cj ). Theoretically, this condition denes that the threshold, ˆ bM (s∗ ) = ( s>s ∗ c

sH

missing mass above rms who bunch at

the threshold come from the region directly above it. We apply this convergence condition by starting from a small value of

sH

and increasing it gradually until

ˆbE (s∗ ) ≈ ˆbM (s∗ ).

This denition for

sH

denotes

the upper bound of the excluded range, and thus the lower bound for estimated excess bunching (Kleven

2

and Waseem 2013). sales response

4s

In addition, the convergence condition implies that we can intuitively dene the

of the marginal buncher rm using the estimated excess mass and the upper limit

sH .

This enables us to approximate sales elasticities with respect to VAT rate for both the VAT kink and the VAT notch by relating the marginal sales response to the implied change in the tax rate. Following Chetty et al.

(2011) and Kleven and Waseem (2013), the standard errors for all the

estimates are calculated using a residual-based bootstrap procedure. We generate a large number of sales distributions by randomly resampling the residuals from equation (4) with replacement, and generate a large number of new estimates of the counterfactual density based on the resampled distributions. The bootstrap procedure takes into account the iterative process to determine

sH .

Based on the bootstrapped

counterfactual densities, we evaluate variation in the estimates of interest. The standard errors for each estimate are dened as the standard deviation in the distribution of the estimate.

4

Data and descriptive statistics

Our data are from the Finnish Tax Administration and contain all businesses that operate in Finland, including rms that are registered to pay VAT and rms that are not included in the VAT register. The data include all information needed for tax purposes, such as sales, number of employees, taxable prots, total assets and organizational form.

Importantly, data include accurate information on total

sales also for rms below the VAT threshold. This enables us to analyze the eects of the VAT threshold on the distribution of sales. In addition, we can link owner-level variables to the rm-level data, such as personal taxable wage and capital income of the main owner of the rm. Figure 5 shows the distribution of sales for all businesses with annual sales between 1,500-20,000 euros in 2000-2011. The Figure shows a clear excess mass at the VAT threshold of 8,500 euros (marked with a vertical line in the Figure). This provides strong visual evidence that rms have responded to the threshold. The distribution seem to be otherwise rather smooth, with the exception of round-number bunching, which can be seen as spikes in the distribution at convenient round numbers such as 5,000, 10,000, and 15,000 euros. Nevertheless, bunching is much more evident at the VAT threshold compared to any of the round numbers, implying apparent behavioral responses to the threshold.

2 Kleven and Waseem (2013) apply this convergence condition to estimate the counterfactual density for an individual income tax notch in Pakistan. For individual tax rate kink points in Denmark, Chetty et al. (2011) determine the upper limit visually, and then iteratively adjust the counterfactual density above the kink point such that it includes the excess mass at the kink. This makes the estimated counterfactual density equal to the observed density. These procedures are intuitively similar, but the convergence method of Kleven and Waseem (2013) typically provides a smaller estimate for excess bunching. In addition, the convergence method provides a more justied approach to dene the upper limit of the excluded region.

10

2000

4000

Frequency 6000

8000

Annual sales, all firms 2000−2011

2500

5000

7500

10000 12500 Sales

15000

17500

20000

Note: Bin width=100 euro

Figure 5: Annual sales of all rms, 2000-2011

In the following analysis, we restrict our sample by excluding those rms for which the VAT rules or the VAT threshold are not binding. Thus all rms in sectors that produce nancial and insurance activities, letting and operation of dwellings, education, human health and social work activities are not included in our sample. In addition, we restrict the sample to include only rms with annual sales below 20,000 euros. Also, we exlude rms that are taxed by assessment of the Finnish Tax Administration, as tax record information based on assessment does not provide evidence of behavioral choices of rms. The most common reason for assessed taxation is that a rm has not declared its tax forms in time. Table 1 oers the descriptive statistics of the data. The upper panel of the Table shows rm-level statistics, and lower panel presents owner-level variables. From rm-level statistics we can see that most of the rms in our sample of small businesses do not have any employees, and have low taxable prots, expenses and assets. The lower panel of the Table shows that sole proprietor is the most common organizational form among small rms.

The average total income of the main owner (the sum of all wage and capital

income) is relatively low, less than 11,000 euros. However, it seems that many of the owners seem to fulll our denition of a full-time entrepreneur, as over 60% of all main owners have more annual turnover in their rm than they have total personal income.

11

Firm-level statistics Sales Expenses* No. of empl. Prots Assets Mean 8,942 3,633 .195 1,705 12,600 Median 7,962 2,071 0 758 1,492 SD 5,355 14,531 1.27 9,448 75,374 N 588,505 341,754 481,407 587,677 487,047 Min 1,500 17 0 -81,852 -141,825 Max 20,000 3,716,961 90 580,561 3,111,189 Owner-level statistics Sole proprietors Corporations Partnerships Total inc. 'Full-time' Mean .688 .224 .088 11,821 .633 Median 1 0 0 4,390 1 SD .463 .417 .283 18,005 .482 N 588,505 588,505 588,505 586,710 588,505 Min 0 0 0 0 0 Max 1 1 1 177,759 1 Sample: Sales between 1,500-20,000 euros per year. Pooled data from 2000-2011. *Information only from 2002 onwards. Table 1: Descriptive statistics

5

Results

5.1 Baseline results Figure 6 shows the sales distribution around the VAT threshold for all rms in our estimation sample using pooled data from 2000-2011.

The gure plots the observed sales distribution (solid line) and

¿ in the range

counterfactual distribution (dashed line) relative to the threshold point in bins of 100 of +/- 7,000 region

¿ from the threshold.

[sL , sH ]

The threshold is marked with a dashed vertical line. The excluded

3

in the estimation of the counterfactual is marked with solid vertical lines.

The Figure

denotes the estimate for the excess mass at the threshold with bootstrapped standard errors, and the estimate for the upper limit of the excluded region,

sH ,

which is determined by the iterative process

explained above. The upper limit also denotes the sales response of the marginal bunching rm,

∆s.

Excess bunching is measured by relating the number of rms in the observed sales distribution to the counterfactual density within the region

[sL , 0].

3 The counterfactual density is estimated using a 7th-order polynomial function. Our results are not sensitive to the choice of the order of the polynomial.

12

VAT threshold, all firms 2000−2011

2000

3000

Frequency 4000 5000

6000

7000

Excess bunching: 3.195 (.179) Upper limit: 27 (2.44)

−70 −60 −50 −40 −30 −20 −10 0 10 20 30 Distance from the threshold Observed

40

50

60

70

Counterfactual

Figure 6: Bunching at the VAT threshold, 2000-2011

Figure 6 illustrates that excess bunching is striking. A signicant proportion of small rms locate themselves just below the VAT eligibility threshold. The estimate for excess bunching is notable and strongly signicant statistically.

These imply that the VAT threshold clearly aects reported sales of

small businesses. We study how excess bunching evolves over time in Section 5.2. In Table 2 we describe which types of rms bunch at the VAT threshold.

Column (1) of Table

2 shows the results from an OLS regression where we regress the dummy variable of locating in the bunching region 7,600-8,500e with rm and owner-level characteristics. We also show the results for the regressions on belonging to sales region below 6,600-7,500e in column (2) and above 8,600-9,500e the bunching region in column (3). These estimations provide benchmark information on the characteristics of small businesses close to the VAT threshold. Thus by comparing estimates in column (1) with columns (2) and (3) illustrate which chracteristics correlate with bunching behavior.

13

(1) Buncher: sales 7,600-8,500e 0.192*** 0.093*** 0.272***

(2) Placebo 1: 6,600-7,500e 0.108*** 0.058*** 0.115***

(3) Placebo 2: 8,600-9,500e 0.075*** 0.045*** 0.061***

-0.001 0.006*** -0.000

0.001 -0.003*** -0.012***

-0.003** -0.006*** -0.001**

0.006*** 0.008*** 0.011*** 0.017*** 546,277 0.081

0.002 0.003** 0.005** 0.000 546,277 0.015

0.001 -0.003*** -0.001 -0.002** 546,277 0.006

Bunches t − 1 Bunches t − 2 (t-1)*(t-2) Ref: corporation Partnership Sole proprietor 'Full-time' Industry ref: Construction Hotels and restaurants Professional activities Admin. activities Arts N R2

Note: Standard errors in parentheses *** p

Suggest Documents