n?u=RePEc:bdi:opques:qef_357_16&r=sog

Questioni di Economia e Finanza

(Occasional Papers)

Tax evasion, firm dynamics and growth

by Emmanuele Bobbio

**September** 2016

Number

357

Questioni di Economia e Finanza

(Occasional papers)

Tax evasion, firm dynamics and growth

by Emmanuele Bobbio

Number 357 – **September** 2016

The series Occasional Papers presents studies and documents on issues pertaining to

the institutional tasks of the Bank of Italy and the Eurosystem. The Occasional Papers appear

alongside the Working Papers series which are specifically aimed at providing original contributions

to economic research.

The Occasional Papers include studies conducted within the Bank of Italy, sometimes

in cooperation with the Eurosystem or other institutions. The views expressed in the studies are those of

the authors and do not involve the responsibility of the institutions to which they belong.

The series is available online at www.bancaditalia.it .

ISSN 1972-6627 (print)

ISSN 1972-6643 (online)

Printed by the Printing and Publishing Division of the Bank of Italy

TAX EVASION, FIRM DYNAMICS AND GROWTH

by Emmanuele Bobbio

Abstract

Italy's growth performance has been lacklustre in the last two decades. The economy

has a low R&D intensity; firms are smaller and less likely to grow or exit than firms in other

advanced countries; the shadow economy is large. I show how these features arise

simultaneously in a Schumpeterian growth model with heterogeneous firms where the tax

auditing probability increases with firm size. Tax evasion confers a cost advantage over

competitors. In equilibrium, small firms invest less in innovation because growing entails a

(shadow) cost of fiscal regularization. Unfair competition forces other firms to lower the markup

they charge for their new products, reducing the incentive to innovate. Market selection is

hampered, further lowering the aggregate growth rate along the extensive margin. I calibrate the

model on Italian firm-level data for the period 1995-2006 and find that enforcing taxes would

have increased the long-run growth rate from 0.9% to 1.1%. The market share of high type

firms would have been 6 percentage points higher and average firm size 20% higher. Also, I

find that lowering the tax burden can have a significant impact on growth when the shadow

economy is large, while the effect is negligible when taxes are enforced.

JEL Classification: O30, O43,H26.

Keywords: growth, innovation, selection, firm dynamics, tax evasion, size dependent

policies.

Contents

1. Introduction ......................................................................................................................... 5

2. The model ........................................................................................................................... 7

3. Identification and calibration strategy .............................................................................. 17

4. Results ............................................................................................................................... 21

5. Robustness ........................................................................................................................ 24

6. Conclusions ....................................................................................................................... 28

References ............................................................................................................................... 30

Appendix A Stackelberg game: pricing, tax evasion and innovation effort choices ............... 32

Appendix B The case with no tax evasion: existence, uniqueness and identification ............. 37

Appendix C Additional material .............................................................................................. 39

Bank of Italy, Directorate General for Economics, Statistics and Research

1 Introduction

Italy has been experiencing a prolonged period of anaemic growth, even prior

to onset of the financial crisis. Labor productivity per hour worked in the nonagricultural

business sector has increased at an annual rate of 0.9% between

1995 and 2006. The Italian business sector is characterized by a low R&D intensity

and it is populated by a prevalence of small firms displaying a weak

“up-or-out” dynamics. R&D expenditure by private companies in Italy was

approximately 0.5% of GDP (as opposed to 1.3% in France and 1.8% in the

U.S.). Regarding firm demographics, businesses with fewer than 50 employees

accounted for 63.5% of employment between 2001 and 2010 (35.9% and 29.9%);

83.4% of entering firms were still in the market after three years (78.7% and

71.7%) and only 3.6% of them had grown in size (4.0 and 6.2%) – Criscuolo,

Gal, and Menon (2014) and, with specific regard to Italy, Manaresi (2015).

In this paper I show how tax evasion may explain these features of Italian firms

along with a weak aggregate economic performance. I use a model of endogenous

growth – Romer (1990), Aghion and Howitt (1992) and Grossman and

Helpman (1991) – accounting for firm dynamics – Klette and Kortum (2004) –

and firm heterogeneity – Lentz and Mortensen (2008) and Acemoglu, Akcigit,

Bloom, and Kerr (2013) – and augment it with a game of tax evasion, price,

quantity and innovation decisions.

Tax evasion in Italy is high, both in absolute terms and by international comparison.

According to estimates in Schneider and Enste (2000) and Schneider

and Williams (2013) the size of the shadow economy in Italy stands at approximately

25% of GDP as opposed to 15% in France and 8% in the U.S.. A similar

number for Italy is found by Ardizzi, Petraglia, Piacenza, and Turati (2014)

using the currency demand approach and detailed cash withdrawal data. Istat

(2011) reports an official estimate of around 18% for the period 2000-2008 and

12% regarding the share of irregular workers on total employment. More recently

and with reference to the period 2011-2013, official estimates have been

revised downward to approximately 13% for the size of the shadow economy as

a fraction of GDP and upward to 15% for the share of irregular workers – Istat

(2015). Cannari and d’Alessio (2007) use survey data and find evidence that

the propensity to evade taxes has increased between 1992 and 2004.

The crucial assumption underlying the results of the paper is that small firms

are less likely to be monitored by the tax enforcement authority than large

firms, other things equal. According to data published by the Italian Revenue

Agency (IRA) the fraction of firms with a turnover above 100mln euros

(approximately 300 employees) monitored in 2015 was 39%, as opposed to 2%

for sole-proprietorship and small firms – Agenzia delle Entrate (2015). Also,

sole-proprietorship enterprises concealed approximately 1/3 of their turnover,

according to estimates by IRA based on tax audit data for the period 2007-

2008 and corrected for possible biases due to the non-random nature of audits –

Ministero dell’Economia e delle Finanze (2014). Finally, the Italian economy is

characterized by a remarkably high self-employment rate, 28.3% between 1995

and 2006 according to the OECD, as opposed to 9.5% in France and 7.7% in the

5

U.S.. Torrini (2005) provides evidence that differences in self-employment rates

across countries are partly explained by differences in tax evasion opportunities.

The fundamental source of distortions in the model is that tax evasion confers

a cost advantage to firms with a greater scope for evading taxes – an advantage

which is greater the higher the statutory level of taxes. A small firm that innovates

and grows gives up this cost advantage; thus, she incurs a shadow cost

of “fiscal regularization”. In addition, the “unfair competition” brought about

by firms with a greater scope tax evasion forces innovative firms to lower the

markup on their products. Both, the regularization cost and unfair competition

reduce the incentive to innovate for incumbent firms and the aggregate rate of

technological progress with it. This outcome is reinforced via two general equilibrium

channels. The low innovation effort exerted by incumbent firms weakens

selection and small, less productive and less innovative firms tend to stay in the

market. In so doing, tax evasion also reduces the aggregate growth rate along

the extensive margin, because innovative firms account for a smaller share of

resources and economic output in equilibrium. Finally, the higher prevalence

of small firms in equilibrium increases the degree of unfair competition in the

economy. Such circularity may disrupt the growth process, when the scope for

tax evasion varies significantly across differently sized firms and statutory tax

rates are high. In this case the benefits from tax evasion may be large enough

that incumbent firms stop innovating, there is no selection and there are no

large firms in equilibrium.

I calibrate the model using statistics for the aggregate growth rate and for the

size of the shadow economy in Italy over the period 1995-2006 and targeting firm

demographics that I compute from micro-data. I find that enforcing taxes would

have raised the growth rate permanently from 0.9 to 1.1%, both by increasing

individual incentives to innovate and by shifting the composition of the economy

towards more innovative firms. Entry would have been lower contributing

negatively to growth instead. Also, enforcing taxes would have enhanced firm

dynamics: the probability of leaving the market within one year of entry would

have been 1 percentage point higher; the employment growth rate of surviving

firms would have increased by 2/3 and mean firm size would have been a quarter

higher. Enforcing taxes increases the revenue stream for the government

that can cut statutory rates. This cut has a negligible effect on growth, when

tax evasion has been already eliminated – consistently with mixed findings in

the empirical literature regarding the relationship between taxes and growth.

However, I find that when the size of the shadow economy is large instead, cutting

statutory rates does boost growth, because it reduces the benefits from tax

evasion: in the calibrated version of the model reducing the tax burden by 1

percentage point by cutting corporate taxes raises the long run growth rate by

4 bases points.

The paper is related to the recent body of literature that emphasizes the importance

of firm heterogeneity and efficient resource allocation for the level of productivity

– Bartelsman, Haltiwanger, and Scarpetta (2012), Hsieh and Klenow

(2009), Restuccia and Rogerson (2008) – and its evolution over time – Acemoglu

et al. (2013), Lentz and Mortensen (2008) – as well as of size related distortions

6

– Guner, Ventura, and Xu (2008), Hopenhayn (2014). The literature has developed

towards a more granular exploration of the sources of inefficiencies in

economies with heterogeneous firms. Particular attention has been devoted to

studying the role of credit frictions for the process of economic development,

Buera, Kaboski, and Shin (2011), Midrigan and Xu (2014), Moll (2014). In the

context of the Klette and Kortum (2004) framework, Aghion, Akcigit, Cagé,

and Kerr (2016) have considered the relationship between taxation, corruption

and aggregate growth when investment in a public good can improve the innovation

process, while Akcigit, Alp, and Peters (2016) have analyzed how limits

to delegation can help explaining differences in firm demographics between India

and the U.S.. Lentz and Mortensen (2015) study optimal taxation within

this framework, while Acemoglu et al. (2013) evaluate the impact of innovation

policies.

The paper is organized as follows: in section 2 I outline the model, characterize

the solution to the game determining price, quantity, innovation and tax evasion

decisions and characterize the balance growth path equilibrium of the economy.

In section 3 I discuss identification of the model parameters and outline the

calibration strategy. Results are discussed in section 4 and their robustness is

tested in section 5. Section 6 concludes.

2 The model

I extend the Schumpeterian growth model – Aghion and Howitt (1992), Grossman

and Helpman (1991), Romer (1990) – with heterogeneous firms – Klette

and Kortum (2004), Lentz and Mortensen (2008) – to account for tax evasion.

A sketch of the basic framework is as follows: there is a representative household

consuming a final good that is competitively produced by combining a

continuum of intermediate products. The productivity of an intermediate product

is the result of past innovations. A firm realizing an innovation prices out

the producer that was supplying the previous vintage of the intermediate good

and charges the final good producer a markup. The firm enters the intermediate

product market if it was a potential entrant, or grows in size, if it was already an

incumbent, adding a market niche to her portfolio of leading-edge technologies.

Tax evasion alters competition between producers of different vintages of the

same intermediate good, distorting the markup and changing the perspective

gains from investing in innovation and growing.

2.1 Final good: consumption, production

Time is continuous, the demand side of the economy consists of a representative

household supplying labor inelastically, L, and choosing consumption of a final

7

good, C, and asset holdings, H, to maximize utility:

∫ ∞

U 0 (H 0 ) = max ln C t e −ρt dt

{C t,H t} t≥0

s.t.: Ḣ t =w t L − P t C t + r t H t

ρ is time discounting, w is the equilibrium wage level. Asset holdings consist of

an exhaustive portfolio of all ownership titles of firms populating the economy,

potential and incumbent. The solution to the household problem is characterized

by the Euler equation:

0

Ċ t

P

= r t − ρ − ˙ t

(2)

C t P t

The final good is produced under perfect competition. The production technology

is Cobb-Douglas and requires a continuum of intermediate products –

indexed on the unit interval:

ln Y t =

∫ 1

0

ln(A it x it )di (3)

x i is the quantity of intermediate good i used into production and A i denotes

its productivity level. Profit maximization and perfect competition imply that

the demand function for intermediate good i is equal to:

x d it =

P t Y t

= 1

(4)

(1 + τ va )p it p it

with P where p i is the price of the intermediate good and τ va is the tax rate

on value added. The second equality follows from normalizing the price of the

final good so that aggregate expenditure equals 1+τ va and P Y/(1+τ va ) = 1 ∀t.

Substituting for (4) into (3) I rewrite aggregate output as:

where:

ln Y t = ln A t −

ln A t ≡

∫ 1

0

∫ 1

0

ln(p it )di (5)

ln(A it )di

2.2 Intermediate goods: pricing, tax evasion, innovation

The productivity of a particular intermediate good is the result of past innovations,

q 1i , q 2i , . . ., q Iiti, where I it is the number of successive innovations realized

in product line i up to time t:

∏I it

A it =

j=1

q ji

8

Innovation is the fruit of the investment activity carried out at each intermediate

good producer (whether potential or incumbent) and is modeled as a Poisson

process. The instantaneous probability of arrival depends on the amount of

resources invested in innovation and the incremental step depends on the innovation

ability of the firm, which is discovered upon entry. The final good

producer is indifferent between paying a price (1 + τ va )p for vintage I it − 1 of

variety i and paying (1 + τ va )pq Iiti for vintage I it . As argued below and shown

formally in appendix A, the payoff function is strictly increasing in p. Then, in

a Stackelberg equilibrium the firm that has the know-how to produce the latest

vintage, which I assume moves first and I refer to as the leader, prices out the

follower and becomes the monopolist producer in market niche i. The firm then

enters the market as the supplier of that single product line if it was a potential

entrant; or grows to size n + 1 if it was a size n incumbent, n denoting the

number of market niches where the firm is a technological leader. Production

of each variety the firm supplies is carried out at a different establishment and

requires labor only, in one to one proportion. The firm must pay taxes on value

added, labor, profits and turnover at rates τ va , τ l , τ pf and τ to respectively. Or

it can conceal part of the output. For simplicity suppose that the establishment

has an overground and an underground part, and that the firm decides

the fraction of output located in the underground part, λ ∈ [0, 1], where it does

not pay taxes. The tax enforcement authority visits the establishment with

a certain probability, this probability growing (at an increasing rate) with the

amount of output the firm attempts to conceal and with the size of the firm:

Pr ′ (λx|n) > 0, Pr ′′ (λx|n) ≥ 0 and Pr(λx|n + 1) ≥ Pr(λx|n). Innovation is carried

out in the overground section of the establishment and requires labor only. 1

In the robustness section I also consider the case where innovation requires the

final good instead of labor, capturing the idea that investments in goods such

as ICT can raise productivity or increase the quality of the intermediate good.

I assume that innovation is arbitrarily costly at a plant that has become a technological

lagger. Consider a particular market niche. Let v S , S ∈ {L, F } be the

expected value of an innovation for the leader (L) and for the follower (F ) and

w be the aggregate wage level and assume that when indifferent the final good

producer buys the latest vintage. Given the demand function 4, the leader and

the follower choose the price p S , tax evasion λ S and the amount of resources to

invest in innovation ι(γ S |S), or equivalently the innovation rate γ, to maximize

1 Alternatively I could assume that a fraction λ of the investment in innovation is concealed

in the underground part of the establishment. In this case the firm pays more profit taxes

but has a lower labor cost. The analysis is somewhat simplified and results are fundamentally

unaffected.

9

the payoff function Ω:

Ω(p S , λ S , γ S |n S , v S , S, q L , p −S , w) =π(p S , x(p S , p −S |q L ), λ S , γ S |n S , S, w)

where π is the profit flow:

x(p L , p F |q L ) =

x(p F , p L |q L ) =

+ γ S v S (6)

{

1

p L

{

1

p F

0 o.w.

0 o.w.

if p L ≤ p F q L

if p F q L < p L

π(p, x, λ, γ|n, S, w) =(1 − τ pf 1 Π

λ,γ

c >0 )Πλ,γ c (p, x|S, w) − (1 − λ)τ to px

+ λΠ nc (p, x|w) − (1 + ν) Pr(λx|n)Υ(p, x, λ, γ|S, w)

and Υ is tax gap, ν is the fine the firm must pay on top of taxes if it is caught

by the tax enforcement authority (in Italy the fine is indeed a proportional

constant):

Υ(p, x, λ, γ|S, w) =λ(τ va px + τ to px + τ l wx)

+ τ pr

[1 Π

0,γ

c

>0 Π0,γ c

(p, x|S, w) − 1 Π

λ,γ

c >0 Πλ,γ c (p, x|S, w)

and Π λ,γ

c (p, x|S, w) and Π nc (p, x|w) are the profit flows generated in the overground

(gross of turnover taxes) and in underground part of the establishment

respectively - c and nc standing for “compliant” and “not compliant”:

Π λ,γ

c (p, x|S, w) =(1 − λ)Π c (p, x|w) − ι(γ|S, w)

Π c (p, x|w) =px − xς c (w)

Π nc (p, x|w) =(1 + τ va )px − xς nc (w)

ς c and ς nc are the constant marginal costs in the overground and underground

part of the establishment and are equal to (1 + τ l )w and w respectively.

Proposition 1. The subgame perfect Nash equilibrium of the Stackelberg game

where the leader and the follower have payoff functions as specified in (6) and

where the tecnological leader moves first is characterized as follows:

]

i The firm never picks a price-tax evasion combination (p S , λ S ) such that

Pr(λx(p S , p −S |q L )|n S ) > 1;

ii The payoff functions of the leader Ω(·|L) and of the follower Ω(·|F ) are

strictly increasing in the price, provided that p L ≤ p F q L and p F q L < p L ,

respectively, and that the pairs (p S , λ S ), S ∈ {L, F } are sensible, in the

sense specified in part i;

iii The follower chooses γ F = 0 and p F = p L /q L and λ L ∈ [0, 1];

10

iv The leader charges the limit price p L : max λF π(p L /q L , q L x(p L , p L /q L |q L ),

λ S , 0|n F , F, w) and picks (λ, γ) to maximize the payoff π(p L , x(p L , p F |q L ), λ L ,

γ L |n F , F, w) + γ L v L .

Proof. See appendix A

A sketch of the proof follows. If the firm were to choose (λ, p) such that

Pr(λx(p S , p −S |q L )|n S ) > 1 then the expected cost of tax evasion would be

higher than paying taxes outright and (0, p) must yield a higher payoff. As for

part, this result hinges on the assumption that the production function in the

final good sector is Cobb-Douglas so that revenues do not depend on the price

the firm charges while the total cost increases in x and decreases with the price

p, assuming that p remains below the price per efficiency unit charged by the

competitor. If λ is not too high, the immediate gains from concealing production

are greater than the expected cost from tax enforcement and increasing profits

must yield a net payoff gain. In addition I assume that the probability that the

firm is monitored and caught by the tax enforcement authority decreases the

lower amount of output it conceals. This assumption is made so that the tax

evasion choice always depends on the price and viceversa – see cases A and C in

appendix A – and not only in marginal cases – case B. Since the payoff function

increases with the price, the leader charges the maximal price such that, even

if the follower tries to exploit the tax evasion margin, it cannot undercut the

leader and still make a positive profits. Since I assume that the follower cost

of innovation is arbitrarily large, there is no reason for it to stay in the market

while earning a negative profit flow, therefore it quits market niche i. The

price level the leader charges depends on his innovation ability captured by q L ,

which allows the leader to charge a markup to the final good producer, and is

constrained by the operating cost of the follower which the follower can reduce

by evading taxes. Finally given the price p L the leader picks the combination

of tax evasion rate λ L and innovation effort γ L that maximizes profits.

2.3 Intermediate goods: values

I assume that there two types of firm differing in terms of their innovation ability,

z =∈ {b, g}, with q g = q > q b = 1. Type b firms (“bad”) are imitators, they are

unable to generate improvements that actually enhance the productivity of an

intermediate product, while type g firms (“good”) are innovators. There is a

mass m of potential entrants having access to the same innovation technology

as incumbents, ι(γ). The type of a firm is realized upon entry and it is equal

to g with probability φ and to b with probability 1 − φ. Also, it is permanent.

An incumbent firm adds to her technology portfolio by investing in innovation.

Innovation is undirected and whenever a firm realizes an innovation in some

market niche, some other firm is displaced from that market niche. 2 Therefore,

all firms are subject to the same instantaneous probability of loosing a market

2 In facts, in the model some market niches offer a higher expected value than others and

it would more profitable for a firm to target market niches where the leader is more intensely

monitored by the tax enforcement authority. I prevent firms from doing so by assumption.

11

niche and become an n − 1 size firm. I denote such common destruction rate δ.

A firm that has one market niche and looses it exits the market.

As mentioned above the auditing probability increases with the size of the firm.

In particular, I assume that a firm is more intensely scrutinized if it operates

more than one plant, Pr(λ/p|1) < Pr(λ/p|n) = Pr(λ/p|n+1), ∀λ/p > 0 and n >

1 and that once a firm is subject to a high monitoring probability she remains

so, even if her size subsequently drops below 2. I refer to a firm whose size is

(or has been) equal to or greater than n = 2 as “large” and to other firms as

“small” and use the letters l and s to indicate the degree of scrutiny a firm is

subject to.

The value of a firm is the sum of the values of the establishments it operates, or

equivalently of the market niches where she is a leader. 3 Let v f zd

be the value

of an establishment operated by a type z ∈ {b, g} firm having size d ∈ {s, l}

and competing against a size f ∈ {s, l} firm. If the firm is large, the discounted

expected stream of net profits generated by an establishment is equal to:

rv f zlt − ˙vf zlt = max

λ,γ

v zlt ≡ζ st v s zlt + ζ lt v l zlt

〈

π f zlt (λ, γ) + γv zlt − δv f zlt

where ζ f is the fraction of product market niches in the economy where the

leader has size equal to f and π f zlt (λ, γ) is short for π t(p L , 1/p L , λ, γ|l, z, w),

given that the limit price p L is known from the solution to the Stackelberg game

illustrated above – up to the equilibrium wage level, w. Profits are discounted

at the nominal interest rate r; the flow value of the establishment net of capital

gains due to the flowing of time equals net profits, plus the expected capital

gains from expending into a new market niche, minus the expected capital

losses from loosing the technological leadership in the market niche currently

The assumption can be justified to a certain extent by appealing to the uncertainty intrinsic to

research effort. More prosaically, I make this assumption out of convenience. Note, however,

that relaxing it would strengthen the argument in this paper: large, more productive and more

innovative firms would be targeted and be more likely to be pushed out of the market than

small, less productive, less innovative firms, thus increasing the shadow cost of regularization,

decreasing selection and further heightening unfair competition along the extensive margin.

3 I assume that each establishment pays taxes separately, that innovation is carried out at

the establishment level and that innovation costs cannot be shifted from one establishment to

another. A firm making negative profits at an establishment and positive profits at another

establishment would have an incentive to shift innovation costs at the establishment yielding

positive profits so as to decrease the overall corporate tax bill. These are cases that turns

out to be irrelevant for the empirical application implemented below. Also, in the empirical

application a firm with two or more product market niches exerts the same innovation effort

at all establishments, regardless of follower characteristics. Therefore I could simply assume

parameter values such that the equilibrium has these features and prove that the value function

is modular across market niches, as in Klette and Kortum (2004) and Lentz and Mortensen

(2008), rather than working at the establishment level and aggregating establishments up to

the firm level. Another approach delivering the same result is to assume that profit taxes are

calculated on profits gross of innovation costs, braking the dependency of γ on p and λ. In this

case the result regarding the effect of tax evasion on aggregate growth would be fundamentally

unaffected, though the counterfactual implications of changing corporate taxes might differ,

as it should be clear from the analysis in appendix B for the case with no tax evasion.

〉

12

dominated by the firm. Since vzlt s < vl zlt , then v zlt < vzlt l . Thus, the stronger

the cost advantage for the follower due to tax evasion (vzlt l − vs zlt

) and the more

widespread tax evasion (ζ s ), the lower incentives to invest in innovation for large

firms. I refer to the quantity vzlt l − v zlt as the “extent of unfair competition”.

If the firm is small and makes an innovation it grows into a 2 product lines firm,

becoming large and subject to a higher degree of scrutiny by the tax enforcement

authority. As a result, while the firm acquires a new market niche, she also must

regularize part of her business at the plant she was already operating, which

entails a shadow cost:

〈

〉

rv f zst − ˙v f zst = max πzst(λ, f γ) + γ[v zl − (v f zst − v f zlt )] − λ,γ

δvf zst

where v f zlt − vf zst > 0 follows from comparing this expression with that for a

large firm and from the fact that, since a small firm is subject to lower scrutiny,

it can implement the same strategy as a large firm, i.e. pick (λ f zlt , γf zlt

, ), and

still make higher profits than a large firm. I refer to the difference v f zlt − vf zst

as the “regularization cost”. Comparing the value function for a small and for

a large firm one notices that the regularization cost reduces the expected value

of an innovation. Thus, tax evasion reduces the incentives for small firms to

innovate and grow in size.

A type b firm, i.e. a firm such that q b = 1, makes positive profits only if she is

up against a follower that is subject to a higher degree of fiscal enforcement. If

this firm innovates and grows she looses this cost advantage. Thus, she pays the

regularization cost and does not rip any benefit from entering a second market

niche, where she will earn zero profits at most, i.e. π f blt (λf blt , γf blt

) ≤ 0. Therefore

a type b firm will always choose not to invest in innovation, γ f bdt

= 0, and to

remain small. It follows that in equilibrium all large firms must be type g.

Finally, as it is clear from the labor market clearing condition outlined below,

on a balanced growth path the aggregate wage level, w t , is constant, thus the

value of a market niche is constant as well. All this considered I can rewrite the

system of value functions more compactly:

πf b

v f b = , ∀f ∈ {s, l}

ρ + δ (12a)

〈

π

f

vgs f gs (λ, γ) + γ[v l − (vgs f − v f gl

= max

)]

〉

, ∀f ∈ {s, l} (12b)

λ,γ

ρ + δ

〈

π

f

v f gl = max gl (λ, γ) + γv 〉

l

, ∀f ∈ {s, l} (12c)

λ,γ ρ + δ

where I renamed v l = v gl since only type g firm can be large and I have used

the Euler equation from the household problem (2) which along with the normalization

P t Y t = 1 + τ va implies r = ρ.

13

2.4 General equilibrium

As mentioned above, there is a fix mass of potential entrants, m; firms in the

entry pool have access to the same innovation technology as incumbents, ι(γ).

Upon realizing an innovation the firm observes her type, g with probability φ

and b with probability 1−φ, and enters the market with one product line, n = 1.

The optimal innovation effort exerted by potential entrants, γ 0 , is such that the

marginal cost equals the expected value of entering the intermediate product

market and the aggregate entry flow is η = mγ 0 :

⎛

⎞

η = mγ 0 = mι ′−1 ⎝φ

∑

∑

ζ f vgs f + (1 − φ) ζ f v f ∣

b ∣w⎠ (13)

f∈{s,l}

f∈{s,l}

The aggregate destruction rate is the result of the innovation activity carried

out by all firms in the economy, new entrants and incumbents, small or large:

δ = η +

∑ ∑

γ f gd ζf gd

(14)

d∈{s,l} f∈{s,l}

The steady state flow equations determining the share of types and leader and

follower sizes across product lines are:

ζ f b , f ∈ {s, l} : η(1 − φ)ζ f =δζ f b

(15a)

ζ f gs, f ∈ {s, l} : ηφζ f =(γ f gs + δ)ζ f gs (15b)

ζ f gl , f ∈ {s, l} : (δ − η)ζ f + γ f gl ζf gl =δζf gl

(15c)

In steady state the inflow equals the outflow. The inflow of product lines supplied

by type b firms and competing against a follower with size f equals to the

entry flow η times the probability that the entering firm turns out to be type

b, 1 − φ, times the probability that the firm ends up competing against a size

f follower, which is equal to ζ f , since innovation is undirected. With regard

to ζgs, f the outflow is augmented with the probability that the an innovation is

realized and the firm becomes large, γgs. f Finally, with regard to large firms, all

innovation that is not realized by new entrants results in the creation of establishments

owned by large firms, accruing to the stock ζ f gl

. In addition plants

which were operated by small, type g firms become part of a large firm further

contributing to ζ f gl .

The mass of intermediate products supplied by small type firms is equal to

ζ s = ζb s + ζl b + ζs gs + ζgs. l Using eqs. (15a) and (15b) and ζ l = 1 − ζ s – which

follows from the fact that the measure of product lines is normalized to 1 –

I can express ζ s as a function of rates characterizing the birth-death process

{η, δ, γgs, s γgs} l and then solve the system of equations (15) recursively:

ζ s = η (γgs l + δ)(γgs s + δ) − φγgs(γ l gs s + δ)

δ (γgs l + δ)(γgs s + δ) + φη(γgs s − γgs)

l

14

The wage rate must clear the labor market. Labor supply is exogenous and

equal to L. Labor demand is the sum of labor hired for production and for

innovation. One unit of labor produces one unit of output and demand for an

intermediate good equals the inverse of the price, eq. (4). The price that a firm

charges for an intermediate product is equal to the type specific innovation step

– q z , z ∈ {b, g} – times the limit price that lowers the follower profits to zero –

see proposition 1. This price depends on the ability of the follower to conceal

production, i.e. on its size – f ∈ {s, l} – and on the cost of production, w (see

appendix A). Labor is also used for innovation both by potential entrants and

incumbents, a firm requiring ι(γ)/w units of labor at a plant for that plant to

generate an innovation at rate γ. Thus, the equilibrium wage rate solves:

L = m ι ( η

m |w) + ∑

w

∑

d∈{s,l} f∈{s,l}

ζ f d

ι(γ f gd |w)

+ ∑

w

∑

1

ζz

f p f z∈{b,g} f∈{s,l} z

(16)

Finally, from eq. (5) the aggregate growth rate is equal to:

⎛

A˙

A = ⎝ηφ +

∑

∑

d∈{s,l} f∈{s,l}

γ f gd ζf gd

⎞

⎠ ln q (17)

Definition 1. Given the normalization P t Y t = 1+τ va , a Balanced Growth Path

Equilibrium consists of:

• an aggregate state {w, r, {P t , A t } t≥0 , ζ} characterized by a constant wage

w, a constant interest rate r, a final good price sequence {P t } t≥0 , a TFP

sequence {A t } t≥0 and a constant vector of market niche shares ζ = {ζ s b , ζl b ,

ζ s gs, ζ l gs, ζ s gl , ζl gl }

• individual consumption {C t } t≥0 and production decisions {Y t } t≥0 by the

household and by the final good producer respectively, as well as constant

decisions by intermediate good producers for prices p = {p s b , pl b , ps gs, p l gs,

p s gl , pl gl } quantities x = {xs b , xl b , xs gs, x l gs, x s gl , xl gl

} tax evasion rates λ =

{λ s b , λl b , λs gs, λ l gs, λ s gl , λl gl } and innovation rates γ = {γs b , γl b , γs gs, γgs, l γgl s , γl gl }

and associated values v = {vb s, vl b , vs gs, vgs, l vgl s , vl gl }

such that:

i. given {P t } t≥0 , factor demand in the final good sector is optimal, i.e.

{{Y t } t≥0 , p, x} satisfy eq. (4);

ii. given w and v, individual choices for {p, x, λ, γ} sustain the the subgame

perfect Nash equilibrium of the Stackelberg game, i.e. they satisfy the

conditions in Proposition 1.iv;

iii. given {w, r, δ, p, x, λ, γ}, individual values v satisfy the value functions

eq. (12);

15

iv. given {w, ζ, v} the innovation effort exerted by potential entrants is optimal,

i.e. η satisfies eq. (13);

v. given the vector of market niche shares ζ, the entry rate η and incumbent

innovation rates γ yield the destruction rate δ, i.e. eq. (14) holds;

vi. given the birth-death process as characterized by the tuple {η, δ, γ}, the

vector of market niche shares ζ satisfies the system of stock-flow equations,

eq. (15);

vii. the wage w clears the labor market, i.e. {w, x, η, γ} satisfy eq. (16);

viii. the market for the final good clears C t = Y t and, under the chosen normalization,

P t = (1 + τ va )/Y t , ∀t ≥ 0;

ix. {r, {P t , C t } t≥0 } satisfy the Euler equation eq. (2), i.e. r = ρ;

x. {{Y t , A t } t≥0 , p satisfy eq. (5), {Y t } t≥0 and {1/P t } t≥0 grow at the same

rate as technology {A t } t≥0 which evolves as in eq. (17).

2.5 Firm size distribution

Let m gl [n] be the mass of good type, large firms supplying n product lines.

By equating the inflow and outflow I obtain the following system of difference

equations:

n > 2 :

n = 2 :

n = 1 :

m gl [n]n(δ + γ l ) =m gl [n + 1](n + 1)δ + m gl [n − 1](n − 1)γ l

m gl [2]2(δ + γ l ) =m gl [3]3δ + m gl [1]1γ l + γ s gsζ s gs + γ l gsζ l gs

m gl [1]1(δ + γ l ) =m gl [2]2δ

where γ l ≡ ∑ f∈{s,l} ζf gl γf gl / ∑ f∈{s,l} ζf gl

is the innovation rate chosen by large

firms across product lines, on average. By inspection it can be verified that

m gl [n] = aθ n−2

l

/n solves the first equation with θ l ≡ γ l /δ and a equal to some

constant. Next I use the first and third expressions to write m gl [3] and m gl [1]

in terms of m gl [2], substitute out for m gl [3] and m gl [1] in the second expression,

solve for a and obtain the closed form solution:

m gl [n] =

{

θs(1+θ l )θ n−2

l

n

, n ≥ 2

θ s , n = 1

(18)

θ s ≡ζgs

s γgs

s

δ

+ γ ζl gs

l

gs

δ

θ l ≡ γ l

δ

16

The overall firm size distribution µ [ n] is then equal to:

µ[n] =

{

θs(1+θ l )

Θ

θ n−2

l

n

, n ≥ 2

θ η

Θ , n = 1 (19)

Θ ≡ θ η + θ s (1 + θ l ) ln 1

1−θ l

− θ l

θl

2

where Θ is the total mass of firms in the economy, θ η ≡ η/δ and I made use of the

stock-flow equation for ζ s , which dividing through by δ becomes θ η = ζ s + θ s . 4

The firm size distribution is entirely determined by three quantities: θ η ≡ η/δ,

θ s ≡ ζ s gsγ s gs/δ + ζ l gsγ l gs/δ and θ l ≡ γ l /δ. Such quantities reflect the tension

between the sullying effect of entry, θ η , and the cleansing effect of the innovation

effort exerted by incumbents, whether small or large, θ s and θ l . The intensity

of the creative destruction process, δ, is irrelevant for the steady state degree

of selection in the economy; what matters is instead how δ brakes down into η

on one hand and γ s gs, γ l gs, γ l on the other. Tax evasion lowers θ s relative to θ η –

both due to the regularization cost and to unfair competition – compressing the

size distribution to the left. Furthermore it reduces θ l relative to θ η because of

unfair competition.

3 Identification and calibration strategy

I assume that the monitoring probability function and innovation cost function

have the power form:

with a s ≤ a l and a ≥ 1, and:

Pr (λx, n) = a n (λx) a (20)

a n =

{

a s if n = 1

a l if n > 1

ι(γ|w) = wι 0 γ 1+ι1 (21)

with ι 0 > 0 and ι 1 > 0.

In appendix B I characterize the model solution in the case with no tax evasion

a s → ∞ and show that under the assumption that ι is a power function and

4 The total mass of firms is computed noting that:

∞∑ θ n−2 [

] [

l

= 1

∞∑ θl

n

n θ 2 −θ l + = 1

∞∑

∫ ]

θl

n=2

l

n θ 2 −θ l + x n−1 dx

n=1

l

n=1 0

[

= 1 ∫ ]

θl ∞∑

θl

2 −θ l + x n−1 dx = 1 [

]

1

0

θ 2 ln − θ l

n=1

l 1 − θ l

17

given that the risk free rate r is known, the model is identified only up to one

of the three parameters {m, ι 0 , ι 1 }, if no information on innovation spending is

available:

Proposition 2. Assume that ι(·) is as in (21) and that there is no tax evasion

(i.e. a s → ∞, if Pr is as in eq. 20) then:

i. there is one and only one balance growth path as defined in Definition 1;

ii. if the risk free rate can be observed, then the model is identified only up to

one of the three parameters {m, ι 0 , ι 1 }, if no data on innovation spending

is available.

Proof. (i.) See appendix B. (ii.) In appendix B I show that the equilibrium

conditions can be rewritten as η = ξ 0 (m, φ, ι 1 )γ, δ = ξ 1 (φ, ξ 0 )γ, ζ g = ξ 2 (φ, ξ 0 , ξ 1 )

and γ = ξ 3 (ι w 0 , ι 1 , q, ξ 1 ), where ι w 0 ≡ (1−τ pf )wι 0 , and that the aggregate growth

rate is (ξ 0 + ξ 2 )γ ln q. Then, given {φ, ξ 0 , q}, demographic characteristics and

their evolution and the aggregate growth rate are fully determined up to two

parameters while there remain three unknowns {m, ι w 0 , ι 1 }. Then, for example,

given knowledge of ι 1 I can invert ξ 0 and ξ 3 (which are invertible) and recover

m and ι w 0 respectively. As for L, note that any w can be rationalized by picking

the appropriate level of L and w determines the number of labor units used

in the innovation process, given a value for ι(·), or equivalently γ. Thus w is

one-to-one with the relative size of a type g vs. a type b firm.

Instead, with tax evasion and different degrees of tax enforcement across

firms (i.e. a s finite, a s < a l and ζ l > 0) the model is fully identified, because

at least two (and up to four) realizations of the innovation intensity choice are

observed, for example γ l gs = ι ′−1 (·|ι 0 , ι 1 ) and γ l gl = ι′−1 (·|ι 0 , ι 1 ) 5 . Then ι 1 can

be identified in theory by comparing the growth rate of small and large firms, for

instance. In practice, small firms tend to be younger and younger firms tend to

grow faster, possibly for reasons other than innovation. A possibility that I do

not explore here might be to compare the growth rate of small and large firms

conditional on age. Instead I tune this parameter based on values found in the

literature, as discussed below, and evaluate the robustness of counterfactuals to

changes in ι 1 .

In appendix B I provide a formal argument and describe a practical strategy

for calibrating the model without tax evasion, given ι 1 . The target statistics

are reported in table 1. In short, to the extent that incumbent firms invest in

innovation, the exit probability declines with age; the entry rate and a point on

the hazard curve provide information on η and δ – or, analogously, two points

on the hazard curve, since in steady state the entry rate equals the exit rate

and a particular hazard function maps only into one exit rate. The relationship

between average firm size and age is indicative of the magnitude of γ, given δ

and φ – these three quantities fully describing the evolution of an incumbent

firm in terms of market niches. Finally, any aggregate growth rate can always

5 In the case of potential entrants one only observes mγ 0 and not γ 0 directly.

18

e rationalized by picking an appropriate level for q – see eq. (17). As for the

model with tax evasion, I set a = 1 in (20) and then pick a s and a l to replicate

the incidence of tax evasion in the economy. Finally, regarding ι 1 , Acemoglu

et al. (2013) estimate a model similar to Lentz and Mortensen (2008) on a sample

of innovative firms using R&D and patent data and find ι 1 = 1.75. They

also discuss various micro-econometric estimates which are obtained either by

examining the relationship between R&D expenditure and patents or the response

of R%D expenditure to changes in taxes and subsidies. The empirical

results in these studies point to a value for ι 1 in the range [0.7, 2.3]. Instead

Lentz and Mortensen (2008) structurally estimate their model with three types

on a panel of Danish firms with at least 20 employees using data on wages and

value added and find ι 1 = 3.73. I set ι 1 = 2 and then in the robustness section

consider how results are affected by changing the value of this parameter.

I compute firm demographics using social security data covering the universe of

Italian employer businesses between 1990 and 2013. The data contains information

on the number of employees and the wagebill, sector and province, along

with entry and exit dates. 6 I aggregate observations at the firm level, using the

fiscal code as the definition for what constitutes a firm. I restrict attention to

the non-agricultural business sector (NACE R1.1 sector C to K) and focus on

the years between 1995 and 2006, i.e. after the recovery that followed the 1992

recession and before the onset of the financial crisis. Descriptive statistics are

reported in table 5 of appendix C.

I consider the 5 cohorts born between 1995 and 1999 and follow them through

to 2006. The left panel of figure 3 displays the exit probability derived from life

table estimates of the survival functions for each of the 5 cohorts along with the

growth rate of average size by age – right panel. I average across cohorts and

take as targets for the calibration the average hazard between age 6 and 8 and

between age 9 and 11 and the average growth rate between age 9 and 11. The

reason for disregarding earlier years is that other mechanisms, such as learning

or time-to-build, might be more important in explaining the exit rate or firm

growth in the first few years of a firm life-cycle – Jovanovic (1982) and Ericson

and Pakes (1995). Indeed, average firm size doubles within one year of entering

(not reported).

Regarding the level of tax evasion, I assume that large firms do not evade taxes

(a l → ∞) and target the fraction of underground full time equivalent employees

estimated by the National Statistical Institute for the period under consideration

in Italy – Istat (2011). The assumption that large firms do not evade is

a simplification which is dictated by the lack of empirical evidence regarding

the propensity to evade taxes across different size classes. In facts, the Italian

Revenue Agency does report recovering unpaid taxes from small as well

6 The data covers all legal entities making social security contributions for at least one

employee worker during at least a month in a given year. Entry (Exit) dates are defined as

the earliest entry (latest exit) date of the legal entities sharing the same fiscal code. With

regard to exit I limit the attention to exits which are flagged in the same year when the exit

is supposed to occur. See Adamopoulou, Bobbio, De Philippis, and Giorgi (2016a) for a more

through description of the data.

19

Figure 1: Firm dynamics, cohorts 1995-1999 (INPS)

Hazard of exit

Size by age (avg. gr.)

0.12

0.25

0.11

0.20

0.10

0.09

0.08

0.15

0.10

0.07

0.05

0.06

0 2 4 6 8 10 12

0.00

2 3 4 5 6 7 8 9 10 11

Exit probability and growth rate of average size by age, private business sector, cohorts 1995-

1999. Source: INPS data

from large firms – Ministero dell’Economia e delle Finanze (2014). However,

as discussed in the introduction firms with a turnover above 100mln euros (approximately

300 employees) are twenty times more likely to be monitored than

small and individual firms – 39% and 2% respectively, Agenzia delle Entrate

(2015). Furthermore the official estimate on the fraction of unreported turnover

by individual firms between 2007 and 2008 is 34.7% – Ministero dell’Economia

e delle Finanze (2014). Finally, the number I use as a target for the incidence

of tax evasion is the lowest among available estimates – see the discussion in

the introduction – and what matters is the difference between how much small

and large firm evade. In the robustness section I also consider the case where

large firms evade same taxes and the overall level of tax evasion is higher than

imputed here.

As a measure of economic growth I consider chained value added at basic prices

per hour worked in the private business sector, Eurostat national accounts. Finally,

I set tax rates on value added, profits, turnover and the penalty parameter

{τ va , τ pf , τ to , ν} based on statutory rates. For the labor wedge τ l I consider the

difference between the labor cost and take home income pay from the OECD

tax database – OECD (2016) – and take averages for the years 2000-2006, 2000

being the first year covered by that data. This approach reflects the view that

the labor cost in the underground sector is cheaper because it is not burdened

by income taxes and social security contributions; also numbers are consistent

with the findings underlying the correction of the Italian national accounts implemented

by Istat to account for the shadow economy, Istat (2014).

20

Table 1: Calibration: targets and resulting model parameters

Target statistics

hazard a hazard a size x age a VA/hour shadow

age 6-8 age 9-11 gr.th 9-11 growth b FTE c

.06804 .06533 .02643 .00921 0.12

Parameters set ex-ante

ν d

d

τ va

τ pf

d

τ to

d

τ l

e

r ι 1 a

0.30 0.20 0.275 0.039 0.86 0.05 2.0 1.0

Parameters calibrated based on target statistics

m φ q ι 0 a s

0.668 0.467 1.175 60.95 2.706

a Averages over age intervals 6 to 8 and 9 to 11, see note to figure 3 b Chain linked Value

added at basic prices per hour worked, average growth rate over the period 1995-2006, source

Eurostat. c Fraction of FTE working underground approximate number of the period 1995-

2006, Istat (2011). All quantities refer to the private business sector. d Statutory rates.

e Labor cost/take home pay − 1, average over the period 2000-2006, OECD (2016).

4 Results

Any wage can be rationalized by picking the appropriate labor supply level,

L. Therefore I set w = 1 and look for the combination of model parameters

{m, φ, q, ι 0 , a s } minimizing the distance with the the data. The target statistics

and the resulting model parameters are reported in table 1. The calibrated

model perfectly matches the targets.

Table 2 displays equilibrium values for some equilibrium variables along with

relevant statistics computed on the data generated by the model. The fraction

of firms that are able to produce innovation is φ = 46.7% at entry. By investing

in innovation these firms tend to displace less productive one, increasing their

output share by 28 percentage points to 74.7% in equilibrium. However the

innovation rate chosen by small type g firm is roughly half that of large type

g firms hinting to an imperfectly functioning selection mechanism, as further

discussed below. Revenues represent 49.7% of output, which is higher than the

value reported for example in the OECD tax database for the period 1995-2006,

40.1%. Also, the model does not account other sources of government revenues,

such as property taxes or excise duties. On the other hand the latter statistic

refers to the overall economy, while the number displayed in table 2 is for the

private sector only and value added taxes are not levied on public services such

as publicly provided education or healthcare; also, rates can be lower for certain

goods, such as food, books medicines, or zero for example in the case of financial

services. The investment in innovation activity represents 2.6% of GDP. Private

sector R&D expenditure in Italy is approximately 0.7% of GDP; the service

sector contribution to R&D expenditure is negligible and industry represent approximately

1/4 of GDP. In addition, innovation investment in the model has a

broader interpretation and may encompass investments in equipment and skills

21

improving the productivity of the firm, both in terms of production efficiency

or product quality. All in all, I interpret the model outcome of a 2.6% innovation

investment share of GDP as a reasonable number, perhaps on the low side.

As shown below, reducing the value of ι 1 – towards the quadratic case as in

Acemoglu et al. (2013) – increases this figure along with the estimated cost of

tax evasion, in terms of a lower long-run aggregate growth rate. The entry rate

(or equivalently the exit rate) is 5.8% accounting for 2/3 of the corresponding

figure reported in official statistics, approximately 8.5%. As remarked above

the calibration aims at capturing the component of firm dynamics fueled by innovation

and it abstracts from other mechanisms such as imperfect information

which are likely to play an important role especially in the first few years of a

firm life-cycle, or from other shocks supply and demand shocks. 10.6% of firms

that are born during a given year t exits by the end of year t + 1. Average

employment size – including underground employment – of surviving businesses

grows by 1.6% a year between age 9 and 11. 7 Average firm size in the economy

in terms of product lines, or equivalently turnover or value added, is 1.62.

4.1 Policy experiment: enforcing taxes

I now consider the effect of curbing tax evasion. Results are reported in the

second row of table 2. As a s grows to infinity tax evasion decreases. In the limit

the long run aggregate growth rate increases by 0.2 percentage points from 0.92

to 1.13%. Such a change comes about through several channels. As vgs f → v f gl

in eq. (12b) the cost of regularization disappears and small innovative firms

increase innovation spending to the level of that of large firms, ι(γ gs ) → γ gl .

Because unfair competition also fades, vgd s → vl gd

in eqs. (12b) and (12c), both

small and large innovative firms increase their innovation expenditure further.

The aggregate destruction rate (δ) increases despite the weakening entry flow

(η) which is due to the decline of the value of being a small firm, when the

scope for tax evasion vanishes – all firms are small when they enter the market.

The increased innovation expenditure by innovative firms and the rise in the destruction

rate both contribute to strengthening selection. Small, less productive

firms are pushed out of the market and the share of value added produced by

innovative firms increases from 78.9 to 85.0%. Thus, the aggregate growth rate

increases both along the intensive and along the extensive margin. The negative

contribution from the decline of entry is mitigated by the fact that only a

fraction φ = 0.47 of entries is associated with an improvement in productivity,

eq. (17). Finally, note that the general equilibrium effect via the labor market

is also positive. Since more productive firms require less labor for production,

the compositional shift towards more productive firms frees up resources, the

wage declines lowering the cost of innovation and part of the labor is reabsorbed

into innovative activities whose share rises from 3.0 to 4.1% of employment (not

reported).

7 This is the growth rate of actual size in terms of the total number of employees, including

those underground, as apposed to the figure reported in table 1 which refers to employees for

which the firm pays social security contributions.

22

Table 2: Main results: calibrated model and counterfactuals

λs γgl γgs η δ ζg τpf τva τto τl w gr.th rev. inn. shad. ent. ex.1y gr.9-11 E(sz.)

Calibr. 28.4 0.0593 0.0327 0.0364 0.0765 0.747 27.5 20.0 3.9 86.0 1.00 0.92 49.7 2.6 11.6 5.8 10.6 1.6 1.62

Lower evasion

as = ∞ 0.0 0.0694 0.0694 0.0282 0.0854 0.824 27.5 20.0 3.9 86.0 0.96 1.13 55.1 3.5 0.0 5.6 11.6 2.6 2.03

Ev./2 17.2 0.0636 0.0516 0.0335 0.0818 0.782 27.5 20.0 3.9 86.0 0.98 1.03 52.5 2.9 5.8 5.8 11.2 2.1 1.77

No evasion (as = ∞) taxes ↓

τpf ↓ 0.0 0.0684 0.0684 0.0326 0.0874 0.801 0.0 20.0 3.9 86.0 0.98 1.13 52.5 3.6 0.0 6.0 11.9 2.5 1.86

τva ↓ 0.0 0.0694 0.0694 0.0282 0.0854 0.824 27.5 7.0 3.9 86.0 0.96 1.13 49.7 3.5 0.0 5.6 11.6 2.6 2.03

τto ↓ 0.0 0.0694 0.0694 0.0282 0.0854 0.824 27.5 20.0 0.0 86.0 1.02 1.13 52.6 3.7 0.0 5.6 11.6 2.6 2.03

τl ↓ 0.0 0.0694 0.0694 0.0282 0.0854 0.824 27.5 20.0 3.9 62.3 1.10 1.13 49.7 3.5 0.0 5.6 11.6 2.6 2.03

τ ↓ 0.0 0.0694 0.0694 0.0282 0.0854 0.824 27.5 20.0 3.9 62.3 1.10 1.13 49.7 3.5 0.0 5.6 11.6 2.6 2.03

Lower taxes (as as in benchmark)

τpf ↓ 28.3 0.0597 0.0411 0.0376 0.0796 0.748 12.5 20.0 3.9 86.0 1.01 0.96 48.7 2.8 11.3 5.9 11.0 1.8 1.61

τva ↓ 28.3 0.0599 0.0361 0.0360 0.0776 0.753 27.5 16.6 3.9 86.0 1.00 0.94 48.7 2.6 11.2 5.8 10.8 1.7 1.64

τto ↓ 28.3 0.0607 0.0376 0.0361 0.0786 0.755 27.5 20.0 1.8 86.0 1.02 0.96 48.7 2.8 11.1 5.9 10.9 1.8 1.65

τl ↓ 28.4 0.0596 0.0342 0.0362 0.0771 0.750 27.5 20.0 3.9 80.4 1.03 0.93 48.7 2.6 11.4 5.8 10.7 1.7 1.63

23

As for firm demographics, a higher degree of tax enforcement is also associated

with a higher early exit probability (the probability for a firm born in t of not

surviving to the end of t+1 increases by 1 percentage point) and with a marked

increase in the rate of firm expansion (average firm size rises with each year of

age by 2.6 as opposed to 1.6%, between age 9 and 11). As a result mean firm

size increases by 24.8%.

The third row displays results for the case where a s is increased so to halve the

fraction of underground employment in the economy. In this case the growth

rate increases by 0.1 percentage point to 1.03. Note that with a higher level of

tax enforcement, the fraction of market niches supplied by large firms grows from

59.1% to 66.5% (not reported). Thus not only selection contributes to growth by

increasing the share of economic activity commanded by firms engaging in innovation,

but it also decreases the scope for unfair competition, further sustaining

innovative activity and long-run growth, v l ≡ ζ s vgl s + (1 − ζ s)vgl l in eqs. (12b)

and (12c).

4.2 Policy experiment: lowering taxes

With no tax evasion the proceeds collected by the government increase by 5.4

percentage points. In the row from 5 to 9 of table 2 I report results for the case

where such resources are used to lower taxes. The effect on aggregate growth is

nil. In facts, the analysis in appendix B indicates that value added taxes have

no effect on aggregate growth, while a cut in corporate, turnover taxes or the

labor wedge have an ambiguous effect, when labor is the only input entering

the innovation process and there is no tax evasion – eq. (41a) shifts downward

while 42 shifts upward in the (w, γ) plane.

Results are different when there is tax evasion. As displayed in the bottom part

of table 2, a cut in statutory tax rates such to reduce government revenues by

1 percentage point has a significant impact on the long run rate of economic

expansion which increases by approximately 0.04 of a percentage point in the

case of a cut in corporate and turnover taxes. This is essentially because a

cut in taxes reduces the benefits from tax evasion and thus both, the cost of

regularization and the extent of unfair competition, in a similar manner as

enforcing taxes.

5 Robustness

5.1 Innovation via the final good

Suppose innovation requires investing into the final good instead of labor and

suppose that the amount of the final good that a firm must purchase at a plant

to generate a given arrival rate of innovation grows with the level of technology

attained by the economy up to that point, then:

ι(γ|P, A) = P Aι 0 γ 1+ι1 (22)

24

Table 3: Case where innovation requires investing in the final good (m = .733, φ = .581, q = 1.163, ι0 = 44.6, as = 2.88)

λs γgl γgs η δ ζg τpf τva τto τl AP gr.th rev. inn. shad. ent. ex.1y gr.9-11 E(sz.)

Calibr. 27.2 0.0561 0.0283 0.0389 0.0772 0.789 27.5 20.0 3.9 86.0 2.55 0.92 49.8 2.4 11.6 5.9 10.7 1.7 1.54

Lower evasion

as = ∞ 0.0 0.0649 0.0649 0.0309 0.0860 0.850 27.5 20.0 3.9 86.0 2.54 1.11 55.2 3.3 0.0 5.6 11.7 2.6 1.86

Evas./2 16.8 0.0598 0.0475 0.0359 0.0828 0.818 27.5 20.0 3.9 86.0 2.54 1.03 52.6 2.8 5.7 5.9 11.3 2.2 1.66

No evasion (as = ∞) taxes ↓

τpf ↓ 0.0 0.0641 0.0641 0.0359 0.0892 0.832 0.0 20.0 3.9 86.0 2.54 1.12 52.6 3.5 0.0 6.1 12.1 2.5 1.72

τva ↓ 0.0 0.0680 0.0680 0.0324 0.0901 0.850 27.5 6.9 3.9 86.0 2.26 1.16 49.7 3.3 0.0 5.9 12.2 2.7 1.86

τto ↓ 0.0 0.0664 0.0664 0.0316 0.0880 0.850 27.5 20.0 0.0 86.0 2.54 1.13 52.7 3.5 0.0 5.8 11.9 2.6 1.86

τl ↓ 0.0 0.0649 0.0649 0.0309 0.0860 0.850 27.5 20.0 3.9 62.1 2.54 1.11 49.7 3.3 0.0 5.6 11.7 2.6 1.86

τ ↓ 0.0 0.0680 0.0680 0.0324 0.0901 0.850 27.5 6.9 3.9 86.0 2.26 1.16 49.7 3.3 0.0 5.9 12.2 2.7 1.86

Lower taxes (as as in benchmark)

τpf ↓ 27.1 0.0566 0.0383 0.0406 0.0818 0.792 10.9 20.0 3.9 86.0 2.55 0.98 48.8 2.8 11.0 6.1 11.3 1.9 1.53

τva ↓ 27.2 0.0574 0.0329 0.0389 0.0797 0.795 27.5 16.3 3.9 86.0 2.47 0.96 48.7 2.5 11.0 6.0 11.0 1.8 1.56

τto ↓ 27.1 0.0582 0.0349 0.0391 0.0809 0.798 27.5 20.0 1.6 86.0 2.55 0.98 48.8 2.7 10.8 6.0 11.1 1.9 1.57

τl ↓ 27.2 0.0564 0.0302 0.0387 0.0779 0.792 27.5 20.0 3.9 80.3 2.55 0.93 48.7 2.5 11.3 5.9 10.8 1.7 1.55

25

The model is exactly as above except that the labor market clearing condition

simplifies to:

L =

∑ ∑ 1

(23)

ζz

f p f z∈{b,g} f∈{s,l} z

and given the normalization P t Y t /(1 + τ va ) = 1 and eq. (5):

ln(AP ) =

∫ 1

0

ln[(1 + τ va )p it ]di =

∑

∑

z∈{b,g} f∈{s,l}

ζ f z ln[(1 + τ va )p f z ] (24)

In table 3 I report the results for this case. The table contains the same information

as table 2 – except that it displays the (constant) product AP instead of w,

which does not play any role here, as apparent from the labor market clearing

condition, eq. (23). The effect of enforcing taxes is essentially the same, both

regarding the long-run growth rate (which increases from 0.92 to 1.11%) and

firm dynamics (the entry rate slightly declines, the exit probability within one

year of entry increases by 1 percentage point, average employment grows with

age at 2.6 vs. 1.7%, mean firm size increases by 20.7%). However the impact of

lowering taxes is higher. Under eq. (23), eq. (41a) in appendix B implies that

the labor wedge has no effect on growth, while corporate, turnover and value

added taxes lower the long-run growth rate, when there is no tax evasion. Table

3 shows that using all the extra proceeds from tax enforcement to reduce the

value added tax rate to 6.9% would boost aggregate growth by another 0.05 of a

percentage point to 1.16%. A similar result would also be obtained by reducing

turnover taxes as the elasticity is approximately the same. Value added taxes

have an effect on the long run growth rate when the final good is necessary

for the innovation process bacause they make it more expensive. For the same

reason, the labor wedge becomes irrelevant. The impact of reducing taxes at the

calibrated value of the tax enforcement parameter is also stronger: a reduction

in corporate or turnover taxes resulting in government revenues 1 percentage

point lower boost aggregate growth by 0.06 percentage point (by 0.04 and 0.01

in the case of an equivalent reduction of value added taxes and the labor wedge

respectively). Similarly to the case where innovation requires labor, the policy

maker can boost growth either by enforcing taxes or by reducing statutory tax

rates; both approaches lower the cost advantage of evading taxes, leveling the

field out between in the Stackelberg game between the leader and the follower.

5.2 Other robustness checks

Finally I test the robustness of the results with respect to changes in the elasticity

of the innovation cost function, 1 + ι 1 , and with respect to different target

statistics. In table 4 I report results for the cases where the model is calibrated

on the same targets as in table 1 but ι 1 = 1.5 or ι 1 = 3.5. The other

calibrated parameters are similar to those obtained when ι 1 = 2 along with

equilibrium outcomes. As mentioned above and implicit in the identification

argument, spending on innovation as a fraction of GDP varies significantly, decreasing

from 3.0 to 1.6% when ι 1 is increased from 1.5 to 3.5. The impact on

26

Table 4: Other robustness exercises: different innovation elasticities and target statistics

λs γgl γgs η δ ζg τpf τva τto τl w gr.th rev. inn. shad. ent. ex.1y gr.9-11 E(sz.)

ι1 = 1.5

ι1 = 1.5 29.1 0.0603 0.0194 0.0349 0.0753 0.828 27.5 20.0 3.9 86.0 1.00 0.93 49.5 3.0 11.8 5.7 10.5 1.6 1.67

as = ∞ 0.0 0.0726 0.0726 0.0251 0.0902 0.897 27.5 20.0 3.9 86.0 0.97 1.20 55.0 4.1 0.0 5.4 12.1 3.1 2.18

a ∞ s , τ ↓ 0.0 0.0709 0.0709 0.0304 0.0927 0.879 0.0 13.0 3.9 86.0 0.98 1.21 49.5 4.3 0.0 5.8 12.5 3.0 1.95

ι1 = 3.5

ι1 = 1.5 26.1 0.0515 0.0370 0.0428 0.0789 0.758 27.5 20.0 3.9 86.0 1.00 0.92 49.9 1.6 11.6 6.0 10.9 1.7 1.44

as = ∞ 0.0 0.0570 0.0570 0.0386 0.0839 0.795 27.5 20.0 3.9 86.0 0.96 1.03 55.6 2.0 0.0 6.0 11.4 2.1 1.58

a ∞ s , τ ↓ 0.0 0.0565 0.0565 0.0419 0.0862 0.783 0.0 20.0 0.0 81.3 1.04 1.04 49.9 2.3 0.0 6.3 11.7 2.0 1.53

hz4-7,hz8-11,gr.8-11

ι1 = 1.5 28.3 0.0579 0.0272 0.0379 0.0776 0.799 27.5 20.0 3.9 86.0 1.00 0.92 49.7 2.5 11.7 5.9 10.8 1.8 1.58

as = ∞ 0.0 0.0680 0.0680 0.0302 0.0886 0.860 27.5 20.0 3.9 86.0 0.96 1.13 55.2 3.3 0.0 5.7 12.0 2.8 1.92

a ∞ s , τ ↓ 0.0 0.0668 0.0668 0.0348 0.0911 0.842 0.0 20.0 3.9 72.6 1.06 1.14 49.7 3.5 0.0 6.1 12.3 2.6 1.78

Shad./GDP = 25%, Shad./VA at large firms= 8%

ι1 = 1.5 42.1 0.0564 0.0197 0.0403 0.0745 0.763 27.5 20.0 3.9 86.0 1.00 0.93 43.1 2.7 24.3 5.9 10.4 1.4 1.50

as = ∞ 0.0 0.0665 0.0665 0.0312 0.0872 0.843 27.5 20.0 3.9 86.0 0.92 1.21 55.1 3.5 0.0 5.7 11.8 2.6 1.87

a ∞ s , τ ↓ 0.0 0.0653 0.0653 0.0360 0.0898 0.824 0.0 0.5 3.9 86.0 0.93 1.22 43.0 3.7 0.0 6.1 12.2 2.5 1.73

27

the long-run growth rate of shutting down tax evasion also depends on the value

of ι 1 and it is stronger the lower the value of this parameter. When lowering ι 1

from 2 to 1 the impact on the long-run growth rate increases from 0.2 to 0.3 of

a percentage point. When raising it to 3.5 it is lower but remains economically

significant at 0.1 of a percentage point. A similar pattern holds for firm demographic

statistics.

I then check the robustness of results with respect to changes in the target statistics

used for the calibration. In the main calibration exercise I ignore the early

years of a firm life-cycle, because other factors may play a more important role

than innovation in driving firm demographics, and consider the hazard function

after age 6 and the employment growth rate after age 9. Here I vary the age

range and consider the average exit probability age 4-7= .0709, the average exit

probability age 8-11= .0657 and the growth rate of average size age 8-11= .0280.

Results are reported in the third part of table 4 and are broadly unaffected relative

to the benchmark. The long-run growth rate in the case with no tax evasion

is 1.14. Finally in the last part of the table I display results for the case where

the model is calibrated based on Schneider and Williams (2013)’s estimate of

the size of the shadow economy as a percentage of GDP which I set at 25%. If

large firms are not allowed to evade taxes the cost advantage associated with

tax evasion is high enough that there is no innovation by incumbent and the

model cannot replicate the data. I then allow large firms to evade taxes, which

is consistent with Agenzia delle Entrate (2015) as discussed above, and set a

target of 8% for the fraction of output they conceal underground. Under this

calibration the effect of shutting down tax evasion is stronger and the long-run

growth rate rises to 1.21% when shutting down tax evasion.

6 Conclusions

I showed that in a Schumpeterian model of growth with heterogeneous firms tax

evasion reduces the long-run growth rate, if smaller firms are less likely to be

monitored by the tax enforcement authority. Under these circumstances small

firms spend less on innovation and remain small so as to stay under the “radar”

– or, formally, not to incur the (shadow) cost of tax regularization associated

with growth. The cost advantage enjoyed by firms with a higher scope for

tax evasion results in unfair competition, lowering the incentives to innovate

for all firms. Both these channels depress the innovative activity of incumbent

firms in the economy. As a result there is less selection in equilibrium and

the economy is populated by higher fraction of small, less productive and less

innovative firms than it would be in the absence of tax evasion, further reducing

the aggregate growth rate along the extensive margin. In addition a larger

fraction of small firms with a higher scope for tax evasion increases the degree

of unfair competition, potentially triggering a vicious cycle where the growth

process brakes down and incumbent firms stop innovating.

Counterfactual exercises based on a calibrated version of the model suggest

that enforcing taxes would have increased the long-run growth rate from 0.9 to

28

1.1% in Italy, with reference to the period between 1995 and 2006. Lowering

taxes would have also increased growth, because it would have reduced the

cost advantage from tax evasion, which is substantial when both the shadow

economy is large and statutory rates are high. Enforcing taxes also would have

affected firm dynamics: the entry rate would have been lower in equilibrium

and the exit probability higher in the first few years following entry, while the

employment growth rate of surviving firms would have increased, resulting in a

higher average firm size.

29

References

D. Acemoglu, U. Akcigit, N. Bloom, and W.R. Kerr. Innovation, reallocation

and growth. 2013.

E. Adamopoulou, E. Bobbio, M. De Philippis, and F. Giorgi. Allocative efficiency

and aggregate wage dynamics in italy, 1990-2013. Banca d’Italia,

Quaderni di Economia e Finanza, forthcoming, 2016a.

Agenzia delle Entrate. Risultati 2015 e strategie 2016, 2015.

P. Aghion and P. Howitt. A model of growth through creative destruction.

Econometrica, 60(2):323–351, 1992.

P. Aghion, U. Akcigit, J. Cagé, and W.R. Kerr. Taxation, corruption, and

growth. European Economic Review, 2016.

U. Akcigit, H. Alp, and M. Peters. Lack of selection and limits to delegation:

firm dynamics in developing countries. University of Chicago, mimeo, 2016.

G. Ardizzi, C. Petraglia, M. Piacenza, and G. Turati. Measuring the underground

economy with the currency demand approach: a reinterpretation of

the methodology, with an application to italy. Review of Income and Wealth,

60(4):747–772, 2014.

E. Bartelsman, J. Haltiwanger, and S. Scarpetta. Cross country differences

in productivity: The role of allocation and selection. American Economic

Review, forthcoming, 2012.

F.J. Buera, J.P. Kaboski, and Y. Shin. Finance and development: A tale of two

sectors. The American Economic Review, 101(5):1964–2002, 2011.

L. Cannari and G. d’Alessio. Le opinioni degli italiani sull’evasione fiscale. Temi

di Discussione, 618, 2007.

C. Criscuolo, P.N. Gal, and C. Menon. The dynamics of employment growth.

2014.

R. Ericson and A. Pakes. Markov-perfect industry dynamics: A framework for

empirical work. The Review of Economic Studies, 62(1):53–82, 1995.

G.M. Grossman and E. Helpman. Innovation and growth in the global economy.

MIT press, 1991.

N. Guner, G. Ventura, and Y. Xu. Macroeconomic implications of sizedependent

policies. Review of Economic Dynamics, 11(4):721–744, 2008.

H.A. Hopenhayn. On the measure of distortions. NBER, 2014.

C.T. Hsieh and P.J. Klenow. Misallocation and manufacturing tfp in china and

india. The Quarterly Journal of Economics, 124(4):1403–1448, 2009.

30

Istat. La misura delloccupazione non regolare nelle stime di contabilità nazionale,

2011.

Istat. La revisione della stima dei redditi da lavoro dipendente. 2014.

Istat. L’economia non osservata nei conti nazionali, 2015.

B. Jovanovic. Selection and the evolution of industry. Econometrica: Journal

of the Econometric Society, pages 649–670, 1982.

T.J. Klette and S. Kortum. Innovating firms and aggregate innovation. Journal

of Political Economy, 112(5):986–1018, 2004.

R. Lentz and D.T. Mortensen. An Empirical Model of Growth Through Product

Innovation. Econometrica, 76(6):1317–1373, 2008.

R. Lentz and D.T. Mortensen. Optimal growth through product innovation.

Review of Economic Dynamics, 2015.

F. Manaresi. Net employment growth by firm size and age in italy. Banca

d’Italia, Questioni di Economia e Finanza, 298, 2015.

V. Midrigan and D.Y. Xu. Finance and misallocation: Evidence from plant-level

data. American Economic Review, 104(2):422–458, 2014.

Ministero dell’Economia e delle Finanze. Nota di aggiornamento del documento

di economia e finanza 2015, rapporto sui risultati conseguiti in materia di

misure di contrasto dell’evasione fiscale, 2014.

B. Moll. Productivity losses from financial frictions: can self-financing undo

capital misallocation? The American Economic Review, 104(10):3186–3221,

2014.

OECD. Taxing wages 2016. 2016.

D. Restuccia and R. Rogerson. Policy distortions and aggregate productivity

with heterogeneous establishments. Review of Economic Dynamics, 11(4):

707–720, 2008.

P.M. Romer. Endogenous technological change. Journal of Political Economy,

98(5), 1990.

F. Schneider and D.H. Enste. Shadow economies: Size, causes, and consequences.

Journal of Economic Literature, 38:77–114, 2000.

F. Schneider and C.C. Williams. The shadow economy. Institute of Economic

Affairs, London, 2013.

R. Torrini. Cross-country differences in self-employment rates: The role of

institutions. Labour Economics, 12(5):661–683, 2005.

31

A

A Stackelberg game: pricing, tax evasion and

innovation effort choices

A size n firm supplying x unit of output at price p(1 + τ va ), evading taxes on a

fraction λ of output, exerting innovation effort γ and facing a value of innovation

v, has an expected payoff equal to:

where π is the profit flow:

π(p, x, λ, γ|n) + γv

π(p, x, λ, γ|n) =(1 − τ pf 1 Π

λ,γ

c >0 )Πλ,γ c (p, x) − (1 − λ)τ to px + λΠ nc (p, x)

− (1 + ν) Pr(λx|n)Υ(p, x, λ, γ)

Υ is total amount of taxes unpaid by the firm and Π λ,γ

c (p) and Π nc (p) are the

profit flows generated in the regular establishment (gross of turnover taxes) and

in the shadow establishment respectively:

Υ(p, x, λ, γ) = λ(τ va px + τ to px + τ l wx) + τ pr

[1 Π

0,γ

c

Π λ,γ

c (p, x) = (1 − λ)Π c (p, x) − ι(γ)

Π c (p, x) = px − xς c

Π nc (p, x) = (1 + τ va )px − xς nc

>0 Π0,γ c

]

(p, x) − 1 Π

λ,γ

c >0 Πλ,γ c (p, x)

Substituting for the demand function, x = 1/p, and using the fact that τ l w =

ς c − ς nc when labor is the only input to production (or when there is no substitution

between labor and capital) the profit flow can be rewritten as:

π(p, λ, γ|n) =(1 − τ pf 1 Π

λ,γ

c

>0 )Πλ,γ c (p) − (1 − λ)τ to + λΠ nc (p)

− (1 + ν) Pr(λ/p|n)Υ(p, λ, γ)

+ γv

Υ(p, λ, γ) =λ[τ to + Π nc (p) − Π c (p)] + τ pr

[1 Π

0,γ

c

Π λ,γ

c (p) =(1 − λ)Π c (p) − ι(γ)

Π c (p) =1 − ς c

p

Π nc (p) =1 − ς nc

p + τ va

>0 Π0,γ c

(p) − 1 Π

λ,γ

c

]

>0 Πλ,γ c (p)

Equilibrium characterization: note that Π λ,γ

c (p) > 0 implies Π 0,γ

c (p) > 0

therefore there are three possible cases:

A: Π 0,γ

c (p) > 0 and Π λ,γ

c (p) ≥ 0

B: Π 0,γ

c (p) > 0 and Π λ,γ

c (p) < 0

32

C: Π 0,γ

c (p) ≤ 0 and Π λ,γ

c (p) < 0

Case A corresponds to the scenario where the firm reports positive profits in the

formal establishment and case C to that where it would report negative profits

even if no production where concealed from the tax authority. Case B is the

intermediate case where the firm reports negative profits on formal production

but were it compelled to report all production it would make positive profits

after paying for labor and value added taxes and would have to pay profit taxes

as well.

In cases A and C the expression for the profit flow simplifies to:

π(p, λ, γ|n) =(1 − τ pf 1 )Π˜λ,γ

Π

λ,γ

c >0 c (p) − (1 − ˜λ)τ to + ˜λΠ nc (p) (28a)

˜λ ≡λ[1 − (1 + ν) Pr(λ/p|n)]

Where we suppress the arguments in ˜λ(p, λ|n) to avoid clutter. Note that a firm

would never choose (p, λ) such that (1 + ν) Pr(λ/p|n) > 1, because in this case

the expected cost of tax evasion would be higher than paying taxes outright:

π(p, λ, γ|n) =(1 − τ pf 1 Π

λ,γ

c

>0 )Πλ,γ c (p) − (1 − λ)τ to + λΠ nc (p)

− (1 + ν) Pr(λ/p|n)Υ(p, λ, γ)

+ γv

0 )Πλ,γ c (p) − (1 − λ)τ to + λΠ nc (p)

− Υ(p, λ, γ)

+ γv

=π(p, 0, γ|n)

Therefore maximal value λ can take is ¯λ(p) = min(1, p Pr −1 (1/(1 + ν))) and

˜λ ∈ [0, λ] ⊆ [0, 1]. Also, note that ˜λ is strictly concave in λ provided that Pr is

increasing and convex in λ.

In both cases, A and C, the payoff function is strictly increasing in p: it is a

linear combination of two terms, both strictly increasing in p, where the weight

shifts towards the larger of the two terms – Π nc > Π c and ∂˜λ/∂p > 0 – minus

a positive term which declines as p increases – (1 − ˜λ)τ to .

As for case B, the expression for the profit flow can be written as:

π(p, λ, γ|n) =(1 − λ)[Π c (p) − ι(γ)] + λ[Π nc (p) − ι(γ)] − (1 − λ)τ to

− (1 + ν) Pr(λ/p|n)τ pf [Π c (p) − ι(γ)]

− (1 + ν) Pr(λ/p|n)λ [τ to + τ va + τ l w/p]

The term in the third line is strictly increasing in p. As for the terms in the first

and second line note that the effect of an increase in p via changes in Π c and Π nc

is positive, since Π nc > Π c , ∀p > 0 and (1+ν) Pr(λ/p)τ pf < (1+ν) Pr(λ/p) ≤ 1.

Also it has positive effect via changes in Pr, because Π c (p) > ι(γ) in case C by

definition. Thus, the payoff function is increasing in p in case B as well. Since

33

the function is piecewise continuous, I conclude that it is globally increasing in

p.

Because the payoff function is increasing in p, then for any admissible pair (λ, γ)

the leader optimally charges the maximal price such that the follower is priced

out of the market. A customer buying the intermediate good is indifferent between

paying a price p for the vintage offered by the follower and paying pq L for

the vintage offered by the leader where q L ≥ 1 measure the efficiency advantage

of the latest vintage. Therefore, the solution to the game is characterized by

the following conditions:

{

λ F = arg max π(p F , λ, γ|n F )

follower stage:

p L : max λ π(p F , λ, 0|n F ) = 0

leader stage: (λ L , γ L ) = arg max π(p F q L , λ, γ|n L ) + γv

Below I further characterize the solution to this system of equations.

Follower stage: the follower does not engage in innovation, γ F = 0, therefore

case B cannot occur and the follower’s payoff function, which corresponds to

the profit flow, simplifies to:

π(p, λ) = (1 − ˜λ)[(1 − τ pf 1 p>ςc )Π c (p) − τ to ] + ˜λΠ nc (p)

I first check for corner solutions and then consider interior solutions.

Suppose the firm is fully compliant and chooses λ = 0. Then ˜λ = 0 and

the zero profit condition, π(p F , 0) = 0, implies Π c (p) > 0, therefore p 0 F =

ς c (1 − τ pf )/(1 − τ pf − τ to ).

Next consider the case where the firm chooses the maximal value of tax evasion.

This boundary value is the minimum between λ = 1 and λ : (1 + ν) Pr(λ/p) =

1. In the latter case the zero profit condition becomes Π nc (p) = 0, i.e. p =

ς nc /(1 + τ va ). Therefore the maximal value of λ in the follower stage is ¯λ F =

ςnc

min〈1, ¯λ(

1+τ va

)〉 and the optimal price level conditional on this value of λ solves

π(¯p F , ¯λ F ) = 0.

Finally consider the case of an interior solution for λ. The first order necessary

condition is:

1 − (1 + ν) Pr(λ i /p i |n F ) − λ(1 + ν) ∂ Pr

∂λ (λi /p i |n F ) = 0

⇓

Pr(λ i /p i |n F ) =

1

(1 + a)(1 + ν)

⇓

˜λ i F = λ i a

1 + a = a

pi ã nF

1 + a

≡ [a nF (1 + a)(1 + ν)] −1/a

where I made use of the fact that Pr is a power function, a is the exponent

and a nF the proportionality coefficient which depends on size. Note that

ã nF

34

a

1+a

π(p i , p i ã nF ) = 0 becomes a quadratic that can be solved separately for the

two cases 1 p>ςc ∈ {0, 1}, and which I denote as p C and p A respectively. It can

be shown that in both cases the negative root implies a negative price, thus the

relevant root is the positive one. For p i , i ∈ A, C to be admissible it must be

that λ i F < ¯λ F , i.e. p i a ã nF 1+a < min〈1, ς ncã nF (1 + a) 1/a 〉, and Π λi F

c

,0 (p i ) ≥ 0 if

i = A, or Π λi F

c ,0 (p i ) ≤ 0 if i = C.

Then, the leader picks the price p L = q L p F where p F = min(p 0 F , ¯p F , {I Ad

F }),

IF

Ad being defined as the set of admissible candidate interior solutions for the

price level and it is either empty or contain p A or p C or both. 8

Leader stage: the leader charges the limit price p L = q L p F and picks (λ, γ) to

maximize:

π(p L , λ, γ) + γv

Note that if p L < ς c then the firm makes negative profits on the production

it reports to the fiscal authority, Π λ,γ

c (p L ) < 0, ∀(λ, γ), and only case C can

occur, simplifying the analysis. This is a special case and in general solving the

leader problems involves checking for several candidate solutions. Some of these

solutions share some components and it is convenient to define:

λ i L = p L ã nL

˜λ i L = λ i a

L

1 + a

γ A : (1 − τ pf )ι ′ (γ + ) = v

γ C : ι ′ (γ − ) = v

γ 0 (λ) : ι(γ 0 (λ)) = (1 − λ)Π c (p L )

Consider first candidate corner solutions. Suppose the firm is fully compliant

and λ = 0. Since p L is known one can compute Π 0,γ+

c (p L ), if it is positive then

(λ = 0, γ = γ + ) is a candidate solution. Similarly if Π 0,γ−

c (p L ) < 0 then (λ =

0, γ = γ − ) is a candidate solution. γ 0 (0) is always a candidate solution. The case

where the tax evasion choice is maximal is characterized by the same procedure

except that λ is set equal to the boundary value ¯λ L = min〈1, p L a nL (1+ν) −1/a 〉.

Next, consider the case of an interior solution. If p L > ς c , then there are five possible

cases, three corresponding to points laying in the interior of the intervals

spanned by the conditions A, B and C above and the remaining two corresponding

to the two kinks at the intersection between these three segments. In the

case of candidate solutions laying in the interior of A and C the expression for

the profit flow is as in (28a) and the tax evasion choice and the innovation effort

choice are locally independent of one another. The corresponding candidates

are (λ i L , γA ) and (λ i L , γC ), provided that Π λi L

c

,γA (p L ) > 0 and Π 0,γC

c (p L ) < 0

respectively.

8 It should be that p A and p C cannot be both admissible at the same time though I have

not verified it.

35

Next consider a candidate solution laying in the interior of the segment spanned

by condition B. In this case the payoff function can be written as:

( ) λ

(1−˜λ)[Π c (p L )−τ to ]−ι(γ)+˜λΠ nc (p L )−(1+ν) Pr |n L τ pf [Π nc (p L )−ι(γ)]+γv

p L

Then the candidate solution (λ B L , γB L ) is characterized by the system of first

order conditions:

˜λ λ [Π nc (p L ) − Π c (p L ) + τ to ] − (1 + ν)Pr λ

( λ

B

L

p L

|n L

)

τ pf [Π nc (p L ) − ι(γ B L )] = 0

[

−ι ′ (γL B ) 1 − (1 + ν) Pr

( λ

B

L

p L

|n L

)

τ pf

]

+ v = 0

where, using of the fact that Pr takes the power form with exponent a:

( )

( )

λ

B

˜λ λ = 1 − (1 + ν) Pr L

λ

|n L − λ B B

L (1 + ν)Pr L λ |n L

p L p L

( ) λ

B

= 1 − (1 + ν)(1 + a) Pr L

|n L

p L

Then using this expression and further exploiting the functional forms of ι and

Pr, the system of f.o.c. above can be rewritten as:

( ) λ

[Π nc (p L ) − Π c (p L ) + τ to ] =(1 + ν)Pr |n L {(1 + a)[Π nc (p L ) − Π c (p L ) + τ to ]

p L

− a }

λ τ pf [Π nc (p L ) − ι(γ)]

⎧

⎨

v

ι(γ) = [

(

⎩(1 + ι 1 ) 1 − (1 + ν) Pr

)

λ

p L

|n L

⎫

⎬

]

τ ⎭ pf

Finally, I consider the possibility that the solution lays at the kinks. At the

conjunction between A and B, i.e. when Π 0,γ

c (p) > 0 and Π λ,γ

c (p) = 0, the

following conditions must hold:

1+ι 1

ι 1

(1 − λ AB

L )Π c (p L ) = ι(γL AB )

(1 − τ pf )[−˜λ λ Π c (p L )dλ − ι ′ (γ AB

L )dγ] + ˜λ λ [Π nc (p L ) + τ to ]dλ + vdγ = 0

where the second expression is obtained by totally differentiating the profit

function from case A with respect to λ and γ. Totally differentiating the first

expression as well and substituting for dγ in the second expression I obtain the

following characterization of the candidate solution (λ AB

L

, γAB L ):

(1 − λ AB

L )Π c (p L ) =ι(γL AB )

Π nc (p L ) + τ to − Π ( )

c(p L )

λ

AB

ι ′ (γ AB)v

=(1 + ν)(1 + a) Pr L

|n L

p L

L

× [Π nc (p L ) + τ to − (1 − τ pf )Π c (p L )]

36

Similarly, in the case where Π 0,γ

c (p) = 0 and Π λ,γ

c (p) < 0 it must be that:

where ι ′ (γ AB

L

Π c (p L ) = ι(γL BC )

˜λ λ [Π nc (p L ) + τ to − Π c (p L )]dλ = ι ′ (γL

AB )dγ

)dγ = 0 by total differentiation of the first therefore. The candidate

, γBC L ) = (λi L , ι−1 (Π c (p L )))

solution in this case is (λ BC

L

B

The case with no tax evasion: existence, uniqueness

and identification

Suppose a s → ∞ so that there is no tax evasion, then the model is characterized

by the following system equations:

rv = π − ι(γ|w) + γv − δv

γ = ι ′−1 (v|w)

η = mι ′−1 (φv|w)

δ = η + ζ g γ

ζ g =

φη

δ − γ

(40a)

(40b)

(40c)

(40d)

(40e)

Where π ≡ (1 − τ pf )[1 − (1 − τ to )/q] − τ to . Under the assumption that ι(·) takes

the power form, ι(γ) = ι w,τ

0 γ 1+ι1 , with ι w,τ

0 ≡ (1 − τ pf )(1 + τ l )wι 0 , the solution

to the system above can be characterized as follows:

[

ι(γ|w) r 1 + ι ]

1

+ A + ι 1 (A − 1) = π (41a)

γ

η = ξ 0 (m, φ, ι 1 )γ

δ = ξ 1 (φ, ξ 0 )γ

ζ g = ξ 2 (φ, ξ 0 , ξ 1 ) ≡ φξ 0

ξ 1 − 1

ξ 0 ≡ mφ 1/ι1

√

(ξ0 − 1)

ξ 1 ≡

2 + 4φξ 0 + ξ 0 + 1

2

(41b)

(41c)

(41d)

(41e)

(41f)

where it can be shown that ξ 1 > 1 and, given w, (41a) admits one and only one

solution, since ι 1 > 0 (so that the LHS monotonically increases from 0 and goes

to ∞ as γ increases from 0 and becomes arbitrarily large). Finally the labor

market clearing condition is:

L = 1 − ξ 2 + ξ 2 [1/q + ι(γ|w)] + mι(ξ 0 γ/m|w)

w(1 + τ l )

(42)

Note that from (41a) the higher is w the lower is γ and that for w → 0 then γ →

∞ and that for w → ∞ then γ → 0. As for the labor market clearing condition,

37

it implies a strictly positive relationship between γ and w. w ranges from to

(1 − ξ 2 + ξ 2 /q)/L to ∞, γ correspondingly varying from 0 to ¯γ : ι 0 [ξ 2 γ 1+ι1 +

m(ξ 0 γ/m) 1+ι1 ]. Then an equilibrium always exists and is unique.

The growth rate and the equilibrium mass of firms are:

Ȧ

A = (φη + ζ gγ) ln(q) = (ξ 0 + ξ 2 )γ ln q

θ = η (

)

1

1 − φ + φξ 1 ln

δ

1 − 1/ξ 1

(43a)

(43b)

As argued in section 3 and as it is clear from the equations above the model is

identified up to the parameter ι 1 only (or m or ι w,τ

0 ), if no data on innovation

expenditure is available.

A possible strategy for calibration in practice is as follows. First, note that the

fraction of firms that exit in an interval dt (small) after entering equals δ as

dt → 0 and that (43b) can be rewritten as the entry rate, η/θ. Also, the growth

rate can be used to recover ln q given knowledge of the other parameters in

(43a). Then, given these three statistics, the model can be calibrated according

to the following procedure: fix ξ 1 and recover φ from (43b). Next, use (40e) to

substitute for ζ g in (40d) and solve for η. Third, solve (41e) for m and compute

ζ g from (41d). Then q can be recovered from (43a). Finally, the growth rate

of surviving firms is a mixture of the process m b (t) = m b (0) exp −δt and a

truncated Poisson process with parameters γ and δ. Such process can be easily

implemented and ξ 1 tuned so as to match the growth rate of average size with

age. This procedure turns out to provide a remarkably good starting point for

the more general model with tax evasion.

38

C

Additional material

Table 5: INPS data, descriptive statistics

Year Nr. Firms %Firms Wage (month.) Employment Entries Exits

Manuf. mean SD mean SD

1990 1,116,992 0.32 1102 457 7.96 182.3 119,761 89,388

1991 1,120,621 0.32 1217 495 7.96 181.0 110,943 89,500

1992 1,122,468 0.31 1288 539 7.86 188.1 108,285 101,587

1993 1,084,614 0.31 1334 556 7.80 184.2 92,043 98,763

1994 1,059,329 0.31 1382 579 7.83 180.2 94,419 81,301

1995 1,063,816 0.30 1441 620 7.87 179.1 99,008 80,571

1996 1,069,946 0.30 1492 646 7.94 172.9 99,307 84,649

1997 1,058,116 0.30 1550 670 7.96 163.1 94,745 76,463

1998 1,082,872 0.29 1580 697 7.97 156.2 106,450 75,046

1999 1,136,162 0.28 1595 711 7.86 138.3 128,635 82,627

2000 1,181,332 0.27 1637 766 7.97 139.1 132,294 88,274

2001 1,222,383 0.27 1675 821 7.98 140.1 137,106 90,527

2002 1,293,290 0.26 1693 788 7.73 133.2 155,194 94,486

2003 1,325,115 0.25 1728 819 7.70 130.0 125,665 106,940

2004 1,369,569 0.24 1765 837 7.59 127.9 144,785 119,100

2005 1,380,837 0.24 1816 892 7.56 128.7 135,505 123,533

2006 1,403,806 0.23 1872 938 7.55 132.0 146,537 122,250

2007 1,474,110 0.22 1898 994 7.53 133.5 176,132 135,930

2008 1,496,808 0.22 1973 1030 7.57 129.0 155,171 149,387

2009 1,478,586 0.22 1975 1006 7.48 146.9 135,234 147,925

2010 1,471,068 0.21 2031 1055 7.43 169.6 139,684 140,797

2011 1,467,732 0.21 2068 1070 7.46 165.1 137,296 146,722

2012 1,468,611 0.21 2073 1086 7.35 167.6 139,508 162,870

2013 1,414,664 0.21 2100 1139 7.44 169.1 118,601 159,614

Non-agricultural business sector, universe of employer businesses. SD = standard deviation.

Source: social security administrative register.

39