CROSS-COUNTRY DIFFERENCES IN BUSINESS DYNAMICS AND IN A ALLOCATION - - PowerPoint PPT Presentation

CROSS-COUNTRY DIFFERENCES IN BUSINESS DYNAMICS AND IN A OCATION OF RESOURCES TO ALLOCATION OF RESOURCES TO PATENTING FIRMS: NEW EVIDENCE FROM MICRO DATA AND THE ROLE OF PO ICIES OF POLICIES

Chiara Criscuolo

OECD Science Technology and Innovation Directorate OECD, Science, Technology and Innovation Directorate Treasury Guest Lecture The Treasury, Wellington 13th March 2015

Based on joint work with:

h “ ” ( l l i

Based on joint work with:

• The “DYNEMP team” (Carlo Menon; Flavio

Calvino and Peter Gal)

• The “Multiprod team” (Giuseppe Berlingieri

and Patrick Blanchenay) y)

• Dan Andre s Polic St dies Branch
• Dan Andrews, Policy Studies Branch

Economics Department OECD and Carlo M St t l P li Di i i S i Menon, Structural Policy Division, Science, Technology and Industry Directorate, OECD

Thanks to the network:

Mariagrazia Squicciarini

Responsible for the Working Party of Industry Analysis, Economic Analysis and Statistics (EAS) Division, STI OECD

Javier Miranda

Chair of the Working Party of Industry Analysis, Census Bureau United States

Werner Hölzl

WIFO Institute (Austrian Institute of Economic Research) Austria

Hilde Spinnewyn, Chantal Kegels, Michel

Federal Planning Bureau Belgium

Dumont Carlos Henrique Leite Corseuil, Gabriel Lopes de Ulyssea, Fernanda de Negri

Instituto de Pesquisa Econômica Aplicada Brazil

Pierre Therrien, Jay Dixon, Anne-Marie

Industry Canada, Statistics Canada Canada

Pierre Therrien, Jay Dixon, Anne Marie Rollin, John Baldwin

y ,

Mika Maliranta

ETLA Finland

Lionel Nesta, Flora Bellone

National Center for Scientific Research (CNSR) and OFCE France

Gabor Katay, Peter Harasztosi

Central Bank of Hungary, Central Statistical Office of Hungary Hungary

Stefano Costa

Italian National Institute of Statistics Italy

Kyoji Fukao, Kenta Ikeuchi

Institute of Economic Research Hitotsubashi University Japan Economic Research, Hitotsubashi University

Leila Ben-Aoun, Anne Dubrocard, Michel Prombo

STATEC Luxembourg

Michael Polder

Statistics Netherlands (Centraal Bureau voor de Statistiek) Netherlands

Lynda Sanderson, Richard Fabling

New Zealand Treasury, Motu Economic and Public Policy Research New Zealand

Arvid Raknerud, Diana Cristina Iancu

Ministry of Trade and Industry, Statistics Norway Norway

Arvid Raknerud, Diana Cristina Iancu

y y, y y

Jorge Portugal

Presidencia da Republica Portugal

Valentin Llorente Garcia

Ministry of Industry, Energy and Tourism, Spanish Statistical Office Spain

Eva Hagsten Jan Selen

Statistics Sweden Sweden

Motivation

• Sluggish productivity growth and stalled job

Motivation

gg p y g j

• creation. Increasing policy interest in:

Job creation/destruction; creative destruction – Job creation/destruction; creative destruction and productivity growth; allocative efficiency; new sources of growth new sources of growth – Firm dynamics and heterogeneous impact of li i policies – Contribution of Innovation to aggregate (long‐ run) productivity growth – Role of patents and IPR systems p y

Motivation

C t l l f fi k d i f

Motivation

• Central role of young firms as key drivers of

job creation:

– “Up‐or‐out” dynamics: high rates of job creation and destruction – Secular decline in start‐up rates

• Heterogeneous impact of Great Recession:
• Heterogeneous impact of Great Recession:

– Even though young firms have been hit most they remained net job creators during the crisis

• Heterogeneous impact of Policies

g p

Motivation

• The allocation and the flow of resources across firms

Motivation

• The allocation and the flow of resources across firms

contributes to aggregate differences in productivity l l d th level and growth.

• At the firm level productivity is driven by innovation

and in turn, the incentives to invest in patents and innovation are affected by the efficiency of resource y y reallocation mechanisms P bli li i i ifi tl ff t th d f

• Public policies can significantly affect the degree of

static and dynamic allocative efficiency

• Very limited cross‐country evidence.

Motivation: data needs and challenges

• Data needs: based on firm level data; cross‐country;

g

Data needs: based on firm level data; cross country; longitudinal; representative; detailed information on sector of activity; age and size dimensions; sector of activity; age and size dimensions;

• Commercial data repositories have well known

shortcomings shortcomings

• Lack of “timely” cross‐country harmonized and

“representative” (micro‐aggregated) firm‐level longitudinal data on job flows and productivity across OECD countries

– National Statistical Offices surveys and Business Registers – Access / Confidentiality – Comparability

Our contribution Our contribution

• Unveil enterprise dynamics and contribution to

Unveil enterprise dynamics and contribution to employment growth using new cross‐country evidence from firm level micro data (Dynemp).

• Unveil differences in returns to patenting across

Unveil differences in returns to patenting across countries and role of policies in explaining these differences (Matched Orbis‐Patstat database)

• Ongoing work…

g g

Ongoing Work Ongoing Work

• Understand the role of policies in explaining

Understand the role of policies in explaining differences in post‐entry growth performance.

• Analyse Long run productivity growth
• Look at frontier growth

Look at frontier growth

• MULTIPROD project to analyse micro driver of

differences in aggregate productivity performance using new cross‐country evidence from firm level g y micro data. D i t diff i IPR t

• Depict differences in IPR systems

The Data

• DynEmp provides new cross‐country evidence on

The Data

DynEmp provides new cross country evidence on employment dynamics using microaggregated data data

– Coordinated by the DynEmp‐team at the OECD and condcuted by delegates from the Working Party of condcuted by delegates from the Working Party of Industry Analysis (WPIA) – Phase I ‐ Dynemp Express: Data for 18 countries (17 Phase I Dynemp Express: Data for 18 countries (17 OECD + Brazil) over 2001‐2011 – Phase II – Dynemp v.2: in the field, data for 17 Phase II Dynemp v.2: in the field, data for 17 countries received, up to 28 countries (OECD and non‐ OECD) )

The DynEmp Express Database

Dynemp express :

The DynEmp Express Database

y p p

• Covers 18 countries over 2001‐2011 and 3 macro

sectors, i.e. manufacturing, non financial services sectors, i.e. manufacturing, non financial services and construction

• Contribution to aggregate job flows (creation and

Contribution to aggregate job flows (creation and destruction) of firms of different size and age

• transition dynamics: follows firms over 3 years

transition dynamics: follows firms over 3 years period in 2001; 2004 and 2007

• Caveats:

Caveats:

– De alio vs de novo entrants – Failures vs acquisitions vs restructuring a u es s acqu s o s s es uc u g – Definition of “inactivity”

The DynEmp v 2 Database

In addition to Dynemp express Dynemp v.2 includes:

The DynEmp v.2 Database

y p p y p

• transition dynamics of start‐ups: follows cohorts of

entrants after 3, 5 and 7 years after entry

• more granular analysis at industry level, as data are now

aggregated up to 2‐digit sectors,

• additional variables and moments of the distribution,

e.g. employment growth volatility; average growth rate; gross job creation by the top 10% of the employment gross job creation by the top 10% of the employment growth distribution

• “distributed regressions”, i.e. regressions conducted at

distributed regressions , i.e. regressions conducted at the unit level within each separate country following the same estimation method and model and over the same ti i d time period

The OECD HAN Database

Matching commercial (Orbis ‐unconsolidated accounts) and d i i i d (P ) id C l i di l

The OECD HAN Database

administrative data (Patstat) provides Cross‐country longitudinal microdata on firms patenting activity and performance over 2003‐2010

• Main measure of firms’patent stock: cumulative sum of EPO‐PCT‐USTPO granted

Main measure of firms patent stock: cumulative sum of EPO PCT USTPO granted patents since 1980, depreciated at 15% p.a.

• Size threshold: 20 employees in 2003 or at the time of first appearance if the

dataset if >2003 (~582k firms) dataset if >2003 ( 582k firms)

Main Caveats:

• ORBIS is a commercial databases which is unlikely to be representative at aggregate level
• ORBIS is a commercial databases which is unlikely to be representative at aggregate level

– Assumption: sample selection is uncorrelated with conditional (on size, country, sector, etc.) patenting probability

• Patents are allocated to the firms based on fuzzy matching, we expect substantial

measurement error – We restrict the analysis to countries where the matching ratio is good y g g

• There is not a 1:1 correspondence between patents and innovation
• Measurement of policies is based on synthetic indicators (OECD and World Bank)

Key results on firm dynamics

• Employment weight of small firms and their age profile

Key results on firm dynamics

– Significantly different across countries; the business dynamics of an economy with a large number of small old firms is different from one with many young (small) firms many young (small) firms

• Small/young firms’ contribution to job creation:

– Net job creation does not come from all small firms, but only from those Net job creation does not come from all small firms, but only from those that are young.

• Growth dynamics of young firms:

– Cross‐country differences in up or out dynamics

• Growth potential of young firms

– Large differences in the extent of upscaling across countries

• Heterogeneous impact of Great Recession within and across

countries: countries:

– Young hit most but most jobs destroyed in large incumbents

SMEs are important for job creation d j b d i and job destruction ...

Source: Criscuolo, Gal and Menon, 2014

…but young SMEs are those hi h t j b which create jobs…

Source: Criscuolo, Gal and Menon, 2014

…across sectors

Source: Criscuolo, Gal and Menon, 2014

The share of start-ups is declining in most countries…

Share of start‐ups (less than 3 year old) in all firms ‐ average over the period

Source: Criscuolo, Gal and Menon, 2014

In some countries upscaling is limited In some countries upscaling is limited

Manufacturing Services 80 Startups (0-2) Old (>10) 80 Startups (0-2) Old (>10) 60 70 es 60 70 es 30 40 50 Emplyoee 30 40 50 Employee 10 20 30 E 10 20 30 E USA LUX CAN BEL FRA HUN GBR AUT SWE NOR BRA FIN NLD PRT ESP NZL ITA JPN USA GBR BEL CAN LUX FRA NOR BRA AUT HUN PRT SWE NLD NZL FIN ESP ITA JPN

Source: Criscuolo, C, P. Gal, C. Menon (2014), “The Dynamics of Employment Growth: New Evidence from 17 Countries,” OECD STI Policy Papers No. 14.

...as can be seen by the growth dynamics of start-ups

Share and growth of surviving micro (<10 emp.) start‐ups over 3, 5, and 7 years

75 80 250

Growth rate Survival rate

65 70 75 200 55 60 150 40 45 50 50 100 30 35 40 50 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7

Source: Dynemp v.2.

3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 3 5 7 BEL AUT SWE FIN HUN ITA NOR NZL

Differences in performance at the top

Mean growth index, average size, median age and share of total employment

• f top 10% firms

Mean growth index (scale on right axis) Mean employment 1.6 1.8 2.0 16 18 20

% 5 4 7 6 9 4 7 20 6 17

1.0 1.2 1.4 10 12 14 wth index ployment 0.4 0.6 0.8 4 6 8 Grow Emp SWE FIN NZL PRT NOR BEL HUN NLD AUT ITA 0.0 0.2 2

Source: OECD DynEmp v2 database. Preliminary data. Note: Growth index = (empt+1‐empt)/(0.5*(empt+empt+1))

Few firms go “up” but many jobs created

Micro (<10) start-ups (<3 year old) across countries, average 2001-4; 2004-7; 2007-10.

Share of firms Share of jobs affected

Micro ( 10) start ups ( 3 year old) across countries, average 2001 4; 2004 7; 2007 10.

Share of firms Share of jobs affected

80 % 0-9 employees 10-19 employees more than 20 employees exit missing 100 % 0-9 employees 10-19 employees more than 20 employees exit missing 20 40 60 40 60 80

• 40
• 20
• 20

20

• 80
• 60
• 40

AUT BEL DNK FIN HUN ITA NLD NOR NZL PRT SWE

• 80
• 60
• 40

AUT BEL DNK FIN HUN ITA NLD NOR NZL PRT SWE

The growth funnel: countries The growth funnel: countries

Average growth index at different percentiles of the growth distribution

Source: OECD DynEmp v2 database. Preliminary data.

Growth index

The growth funnel: sectors The growth funnel: sectors

Average growth index at different percentiles of the growth distribution

Source: OECD DynEmp v2 database. Preliminary data.

Do resources flow to patenting firms?

• Innovation (patenting) has positive effects at firm level (e.g. Hall et al.,

Do resources flow to patenting firms?

2005; Balasubramanian and Sivadasan, 2011; Kogan et al., 2012)

–

Evidence from the US indicates that:

• increases in patent stock are associated with increases in firm size, scope, skill

p , p , and capital intensity;

• news about patents are associated with a strong stock market response
• using ORBIS firm‐level data matched with PATSTAT for the whole

g economy (manufacturing and services) of 20 countries:

• We look at what happens to employment; capital; output and

productivity when firms patent productivity when firms patent

• We address causality building an IV based on litigation data

–

We look at patents granted by different offices; patent families; group‐ l l level patents

–

We compare the different magnitude of these estimates across countries

• We assess the role of national policies and framework conditions in
• We assess the role of national policies and framework conditions in

explaining cross‐country differences

Descriptive evidence

Non patentees Patentees

Descriptive evidence

p

Differ.

Mean Std Dev Number Mean Std Dev Number

Log em ploym ent

4.011 0.001 2,354,738 5.044 0.003 157,246 1.034***

Log real

6 6 6 8 68 6 ***

Log real capital stock

6.6 0.005 2,320,796 8.221 0.001 155,268 1.621***

Log total turnover

9.005 0.001 2,295,936 10.497 0.004 152,673 1.493***

Log labour productivity (turnover)

3.633 0.001 1,650,046 4.117 0.003 111,940 0.484***

Log labour productivity (value added)

4.96 0.001 2,268,045 5.403 0.003 150,577 0.443***

)

Base model Base model

P tS Y  ) l ( l

• where: Yijct is the input/output for firm i, in industry j, country c

i i d P S i h k (EPO USTPO PCT) i

isct sct i isct isct

PatS Y         ) ln( ln

1

in period t, PatSijct is the patent stock (EPO, USTPO, PCT), ηi is a firm F.E., μjct are industry‐country‐year F.E. β i h i d i i i f fi l ( i l

• β is the estimated sensitivity of firm employment (capital;

Capital intensity; sales; value added; Labour Productivity; TFP etc )to the patent stock relative to the country‐industry‐year etc.)to the patent stock, relative to the country industry year

• average. Β s significant and positive for all but TFP.

isct sct i j ct isct j j isct isct

P PatS PatS Y          



* ) ln( ) ln( ln

2 1

• Pct refers to policy j of country c at time t

IV: addressing potential endogeneity g p g y

• Unobserved heterogeneity or omitted variables might drive

the positive association between patenting and firms’ activity, e g : e.g.:

– firms endowed with a more skilled workforce or better management might patent more, and at the same time grow more g p , g – reverse causality: firms that manage to attract more resources might be able to convey some of these resources into their patenting activity

• Measurement error may lead to attenuation bias
• A valid IV must be strongly correlated with patenting activity

at firm level conditional on firm FEs, and have no additional effect on firm activity

IV: addressing potential endogeneity g p g y

Key results on patenting firms

• A 10% increase in patent stock is associated with a

Key results on patenting firms

A 10% increase in patent stock is associated with a 1% increase in labour, 1.3‐1.5% increase in capital, 1 2% in turnover 0 5% in value added 1.2% in turnover, 0.5% in value added

• the positive impact of patenting on firm size is

likely to be causal

• There are significant differences across OECD
• There are significant differences across OECD

countries in the magnitude of these estimates

– 4x‐6x bigger in the US, SE, BE as compared to JP, FI, DN

• National framework conditions matter

Country specific OLS effects (capital) Country-specific OLS effects (capital)

Note: Country-specific sensitivity of employment to patent family stock, obtained by interacting the patent stock variable with a full set of country dummies. Bars report 10% confidence intervals.

Key results on policy and framework conditions

Policy Em ploym ent Capital

conditions

Policy Em ploym ent Capital

Enforcing contracts cost

– *** – **

Closing business cost

– ** – **

Product market regulation

– * –

• Empl. protection legislation

– *** – *

Stock market capitalization

+ ** + **

Early stage finance

+ ** + ***

Expansion finance

= + *

Note: the table reports the sign of the coefficient of the patent stock interacted with the policy variable. Regressions are run separately.

Key results on policy

• Sensitivity of employment or capital to patent stock

Key results on policy

Sensitivity of employment or capital to patent stock changes is higher in economies characterised by:

– Less stringent EPL and bankruptcy legislation Less stringent EPL and bankruptcy legislation – Better investors protection More efficient judicial systems – More efficient judicial systems – More developed seed and early stage venture capital markets markets

• The effect is generally much stronger for young firms

( )

• Robust to controlling for openness (outward FDI) and

size (GDP)

Propensity score matching Propensity score matching

• Following B&S (2011), we implement a matching

exercise aimed at

– Assessing the extensive margin effect – establishing the timing of the patent effect – testing for “pre‐treatment” differences

• We match every patentee with the “most similar”

We match every patentee with the most similar non‐patentee within the same year, 3‐digits sector, and country and country

– Similarity based on treatment propensity estimated on employment, capital, and turnover the year before the 1st employment, capital, and turnover the year before the 1 patent

Propensity score matching: employment

Propensity score matching: capital

Wrapping up

• Empirical regularities

Wrapping up

– Net job creation does not come from all small firms, but only from those that are young – Growth dynamics of firms differs across countries; in some countries firms hardly Growth dynamics of firms differs across countries; in some countries, firms hardly scale after entry – Growth of young innovative firms means “up” or “out”; entrepreneurs need flexibility to experiment with new technologies and new business models – Co‐existence of success and failure (experimentation) – Decline in start‐up rates

• A range of policies influence the reallocation of resources to innovative

g p firms and has impacts on the scope for experimentation

– Enable experimentation: Reduce barriers to the entry (e.g. red tape), growth (e.g. size‐specific regulations), and exit/failure of firms (e.g. penalising bankruptcy l i l i l i l i l i l i ) legislation, overly strict employment protection legislation). – Level the playing field for new and innovative firms: Some policies favour incumbents and MNEs (e.g. R&D tax credits). Strengthen the innovation system for young and innovative firms e g through – Strengthen the innovation system for young and innovative firms, e.g. through enhanced access to (risk) capital and developed financial markets, etc.

Ongoing work

E l iti th D 2 0 d t b l

Ongoing work

• Exploiting the Dynemp 2.0 database we explore :

– the role of policies in driving differences in size distribution (e.g. size contingent policies) (e.g. size contingent policies) – How finance and other framework conditions affect post‐ entry growth d f h b d d l – drivers of the observed decline in entry rates across countries – Employment growth volatility Employment growth volatility – Differences in IPR systems

• Looking more closely at productivity…

– WP1‐CIIE project on long term productivity and Multiprod

• Policy Evaluation (both for firm dynamics and productivity)

Next steps

• include investigating:

Next steps include investigating:

– role of policies in explaining differences in post‐entry growth g – drivers of the observed decline in entry rates across countries – Differences in IPR systems – Employment growth volatility

• Looking more closely at productivity…

– WP1‐CIIE project on long term productivity and MUltiprod

• Policy Evaluation (both for firm dynamics and

y ( y productivity)

MultiProd

• Cross country differences in firm‐level productivity performance:

MultiProd

– Schumpeterian process of creative destruction across countries – Characterization of entire firm‐level productivity distribution by industry, and refined by size, age, and ownership categories – Measures of allocative efficiency Descriptive statistics of firms’ characteristics at different segments of – Descriptive statistics of firms characteristics at different segments of the productivity level and growth distributions – Firms at the frontier: differences across countries; contribution to aggregate productivity

• Estimates of misallocation and market inefficiency
• Drivers of wage dispersion and relationship with productivity

dispersion i b f d i d f h G i

• Time: before, during and after the recent Great Recession

Policy Evaluation

• Ex‐ante policy evaluation at time of implementation

Policy Evaluation

Ex ante policy evaluation at time of implementation (e.g. randomization)

• Distributed microdata approach to evaluate R&D tax
• Distributed microdata approach to evaluate R&D tax

credit (joint with NESTI) E t t i l ti f

• Ex‐post econometric evaluation of:

– Training and apprenticeship programmes – Innovation policies – Regional Policies (based on the approach developed in Criscuolo et al., 2012 and Criscuolo and Polat, 2014) – Also exploiting the distributed microdata project approach

Policy Evaluation

• Ex‐post econometric evaluation of:

Policy Evaluation

Ex post econometric evaluation of:

– Regional Policies (based on the approach developed in Criscuolo et al., 2012 and Criscuolo and Polat, 2014) , , ) – The idea is to exploit “exogenous” variation in the rules designating supported areas to identify the causal impact of an area becoming/being/ losing eligibility for support. – “exogenous”:

• defined at an upper administrative level (e.g. EU rather than member

country); i hift h h l iti l d i f ti

• using a shift‐share approach; exploiting lagged information on area

conditions used to determine eligibility and exploit changes in rules

• Accidental randomization: areas that were ineligible become eligible

g g (and viceversa) for reasons other than economic conditions and in a short time span.

Thank you for your attention

Chiara.Criscuolo@oecd.org dynemp@oecd org dynemp@oecd.org multiprod@oecd.org

ADDITIONAL SLIDES ADDITIONAL SLIDES

The methodology

• Methodology:

The methodology

gy

– Metadata collection – Confidential national business registers – Flexible micro‐aggregation along different dimensions using a distributed microdata (DMD) approach. Si l th hl t t d St t ti – Single, thoroughly tested Stata routine:

• Flexible to adapt to differences in data setup
• Extensive confidentiality checks and blanking
• Extensive confidentiality checks and blanking
• Internal bridging of different sectoral classifications
• Easily extendable over time and countries

y

• Programmed in a modular way: flexible to updates in

methodology and policy issues

• Country notes
• Country notes

Phase I: DynEmp Express

Annual panel data on

The database

–Annual panel data on

• job flows (creation, destruction)

l t d b f fi

• employment and number of firms

–By:

18 countries (17 OECD + Brazil) × 3 broad sectors (Manufacturing, construction and non‐

)

financial services)

× 5 age classes (0; 1‐2; 3‐5; 6‐10; 11+) × 6 size classes (Thresholds: 1, 10, 50, 100, 250, 500) × 11 years (2001‐2011) 3 ( ) × 3 status (incumbent, entrant, exiting firm)

…but most of young firms are small …but most of young firms are small

Note: Young firms are defined as 5 years old or younger Source: Criscuolo, Gal and Menon, 2014

SMEs are important for job creation d j b d i and job destruction ...

Source: Criscuolo, Gal and Menon, 2014

…but young SMEs are those hi h t j b which create jobs…

Source: Criscuolo, Gal and Menon, 2014

…and not all SMEs

Source: Criscuolo, Gal and Menon, 2014

The share of start-ups is declining in most countries

Share of start‐ups (less than 3 year old) in all firms ‐ average over the period

Source: Criscuolo, Gal and Menon, 2014

Young firms suffered relatively more f h i i from the crisis…

Yearly growth rate of young and old firms expressed as difference y g y g p from the 10‐year trend

Source: Criscuolo, Gal and Menon, 2014

…but most jobs were destroyed by th d i i f ld i b t the downsizing of old incumbents

Contributions to aggregate net job creation by entrants, young/old exitors, and young/old incumbents.

Source: Criscuolo, Gal and Menon, 2014

The growth funnel: countries The growth funnel: countries

Average growth index at different percentiles of the growth distribution

Source: OECD DynEmp v2 database. Preliminary data.

Growth index

The growth funnel: sectors The growth funnel: sectors

Average growth index at different percentiles of the growth distribution

Source: OECD DynEmp v2 database. Preliminary data.

The link between churning and growth of top performers

Average growth index at different percentiles of the growth

0 35

g g p g distribution vs. churning rate

0.30 0.35 0.25 e

AUT

0.20 m churning rate

BEL FIN HUN

0.15 Firm

ITA NLD NOR

0 05 0.10

NZL PRT SWE

Source: OECD DynEmp v2 database. Preliminary data.

0.05 0.5 1 1.5 2 2.5 Average growth index top 10% firms

It’s young not small firms which are d i more dynamic

Dependent variable: Net Growth Rate Net Growth Rate Young 0 180*** Young 0.180 (0.003) Sm all 0.068***

• 0.022***

(0 005) (0 004) (0.005) (0.004) Medium 0.065***

• 0.023***

(0.005) (0.004) Macrosector F.E. YES YES Country X year F.E. YES YES Observations 1,885 1,885 R-squared 0.246 0.710

… even when entrants are excluded f h from the young-group

Dependent variable: Net Growth Rate Net Growth Rate Young 0 047*** Young 0.047 (0.003) Old Base group Sm all

• 0 008**
• 0 032***

Sm all

• 0.008
• 0.032

(0.004) (0.004) Medium 0.010***

• 0.013***

(0.004) (0.003) Large Base group Macrosector F.E.

YES YES

Country X year F.E.

YES YES

Observations 1,885 1,885 R-squared 0.489 0.567

It’s the young and small group which i h d i is the most dynamic

Dependent variable: Net Growth Rate Dependent variable: Net Growth Rate

Young-Sm all 0.171*** ( ) (0.004) Young-Medium 0.143*** (0.006) Old S ll *** Old-Sm all

• 0.035***

(0.004) Old-Medium

• 0.009**

(0 004) (0.004)

Old-Large Base group

Macrosector F.E. YES Country X year F.E. YES Ob ti 88 Observations 1,885 R-squared 0.720

The financial crisis hit hardest d ll fi young and small firms

Dependent variable Net Growth Rate Young-Sm all 0.182*** (0.005) Young-Medium 0.153*** (0.007)

rmal es”

Old-Sm all

• 0.031***

(0.004) Old-Medium

• 0.007*

(0.004)

“Nor time

Old-Large

Base group Country X year F E YES Country X year F.E. YES

Macrosector F.E.

YES Observations 1,885 R-squared 0.732

Change in firm employment associated ith 10% h i th t t t k with a 10% change in the patent stock

M i



Maximum (Sweden) Maximum (Switzerland) Minimum (United States) Minimum (United Kingdom) Minimum (Norway) 2.5 3.0

UK

(United States) 2.0

UK UK UK UK

Maximum Minimum (Greece) Minimum 1.0 1.5 Maximum (Portugal) Maximum (Poland) (Czech Rep.) ( ) (Slovak Rep.) 0.0 0.5 Stringency of Access to Early Stringency of Judicial inefficiency Stock market Stringency of Employment Protection Legislation Access to Early Stage VC Stringency of Product Market Regulation Judicial inefficiency Stock market capitalization

The estimated impact of various policies on the responsiveness of the firm employment to patenting The estimated impact of various policies on the responsiveness of the firm employment to patenting

Note: the chart shows that the sensitivity of firm employment to changes in the patent stock varies according to the policy and institutional environment.

Caveats on data

• ORBIS is a commercial databases which has some issues

Caveats on data

ORBIS is a commercial databases which has some issues

– Non representative at aggregate level – Likely a selected sample: need to assume that sample Likely a selected sample: need to assume that sample selection is uncorrelated with conditional (on size, country, sector, etc) patenting probability

• Patents are allocated to the firms based on fuzzy

matching, we expect substantial measurement error

– We restrict the analysis to countries where the matching ratio is good

• There is not a 1:1 correspondence between patents and

innovation

• Measurement of policies is based on synthetic

indicators (OECD and World Bank)

Key results on patenting firms Key results on patenting firms

Capital Capital Em ploym ent Capital Stock p Stock / Em ploym ent Turnover Value Added TFP Patent stock (firm )

0.102*** 0.130*** 0.0289*** 0.115*** 0.0570*** 0.00263

( )

(0.00784) (0.0119) (0.00995) (0.00964) (0.00829) (0.00641)

Firm fixed effects

Yes Yes Yes Yes Yes Yes

Country-nace2-year fixed

Yes Yes Yes Yes Yes Yes

y y effects

Yes Yes Yes Yes Yes Yes

Num ber of observations

2,412,689 2,412,689 2,412,689 2,412,689 1,498,466 1,669,235

Patent stock - Fam ilies (firm )

0.115*** 0.156*** 0.0413*** 0.120*** 0.0669***

• 0.00288

(0.00909) (0.0129) (0.0110) (0.0104) (0.00914) (0.00711)

Firm fixed effects

Yes Yes Yes Yes Yes Yes

Country-nace2-year fixed effects

Yes Yes Yes Yes Yes Yes

Num ber of observations

2,412,689 2,412,689 2,412,689 2,412,689 1,498,466 1,669,235

F ll i B&S ( ) i l t

Propensity score matching

• Following B&S (2011), we implement a

matching exercise aimed at

A i th t i i ff t – Assessing the extensive margin effect – establishing the timing of the patent effect testing for “pre treatment” differences – testing for pre-treatment differences

• We match every patentee with the “most

similar” non-patentee within the same similar non patentee within the same year, 3-digits sector, and country.

– Similarity based on treatment propensity Similarity based on treatment propensity estimated on employment, capital, and turnover the year before the 1st patent

Propensity score matching: results p y g

Em ploym ent Capital Turnover Patentees dum m y x first patent year – 3

• 0.0314
• 0.0419

0.0540

y p y 3

(0.0391) (0.0702) (0.0531)

Patentees dum m y x first patent year – 2

• 0.0421*

0.0173 0.0161 (0.0230) (0.0462) (0.0239)

Patentees dum m y x first patent year – 1

• 0.0235

0.0350 0.0358* 0.0235 0.0350 0.0358 (0.0175) (0.0397) (0.0194)

Patentees dum m y x first patent year

0.00522 0.0855** 0.0581*** (0.0170) (0.0377) (0.0194)

Patentees dum m y x first patent year + 1 Patentees dum m y x first patent year + 1

0.0223 0.143*** 0.0982*** (0.0176) (0.0419) (0.0216)

Patentees dum m y x first patent year + 2

0.0922*** 0.169*** 0.154*** (0.0230) (0.0455) (0.0259)

Patentees dum m y x first patent year + 3

0.0899*** 0.142*** 0.129*** (0.0256) (0.0508) (0.0299)

Patentees dum m y x first patent year + 4

0.134*** 0.112 0.128*** (0 0358) (0 0747) (0 0475) (0.0358) (0.0747) (0.0475)

Year fixed effects

YES YES YES

Index year fixed effects

YES YES YES

Num ber of observations

9,233 9,233 9,233

Propensity score matching: employment p y g p y

Propensity score matching: capital p y g p

Country-sector diff-in-diff

• based on the assumption that some industries have ‘naturally’

higher exposure to a given policy than other industries

• the term (lnPatS*P) is interacted with a relevant index of
• the term (lnPatS P) is interacted with a relevant index of

sectoral exposure (E) to the policy at hand, to form a triple interaction term in the following model:

isct t c s i s isct c isct j s j ct isct j j isct

S PatS C PatS E P PatS Y                  * ) ln( * ) ln( * * ) ln( ln

3 2 1

• includes country (C) and sector (S) dummies (C); firm fixed

effects; sector, country and year fixed effects; the parameter of effects; sector, country and year fixed effects; the parameter of interest is δ1

• Results are confirmed