ON THE ROLE OF PROCESS INNOVATIONS ON SMES PRODUCTIVITY

Universitat de Valencia and ERI-CES

1

Mañez, J. A. ; Rochina, M.E. ; Sanchis, A. and Sanchis, J. A.

Outline

1. Introduction2. Are process innovators more productive than non-

process innovators?3. On the relationship between process innovations

and productivity for SMEs?1. Are the ex-ante more productive SMEs those that start

introducing product innovations?2. Does process innovations boost SMEs productivity

growth? A matching approach.

2

Introduction

1. In order to introduce the paper we use Griliches (1979) seminal work that outlined that the relationship between R&D and productivity encompasses two different and complex processes:

3

R&D activities

Obtention of (process)

innovations

Productivity gains

Introduction

The focus of our paper is to analyze in depth the effects of process innovation in productivity growth for SMEs. We aim to analyze both: The extent of the productivity gains The life span of the productivity gains

4

Introduction

Further, we aim to take into account an endogeneity problem that characterizes this relationship: It could be true that process innovation increase

productivity But it could also be true that only the most productive

firms are able to generate the resources to implement process innovations

We use techniques coming from public policy evaluation to solve this endogeneity problem

5

Introduction6

Database : Encuesta sobre Estrategias Empresariales (ESEE) from 1991 to 2002 Two different sampling procedures according to size: All firms with more than 200 workers (participation rate 70%)

Firms from 10 to 200 workers: random sampling holding arround 5% of the population

The panel data nature of the data set allows to classify firms according to their process innovation patterns over time.

SMEs

Introduction

To measure total factor productivity (TFP) we use the Good et al (1996) extension of the firm level multilateral productivity index developed by Caves et al (1982) and adapted to the ESEE by Delgado et al (2002).

7

Are process innovators more productive than non-process innovators?

To get a first picture of the effects of process innovations on the productivity levels of SMEs we check whether SMEs that introduce process innovations have higher productivity levels than SMEs that do not introduce them. We compare year-by-year from 1991 to 2002 the TFP distributions

of process innovators (PI) and non-process innovators (NPI)using stochastic dominance techniques (Kolmogorov-Smirnov one and two tailed tests).

The results of these tests that the TFP distribution of PI dominates that of NPI for every year except 1991 and 1992.

8

Are process innovators more productive than non-process innovators?

These results can be summarized in Figure I that shows the relative distribution functions of TFP for PI in t and NPI in t.

9

0.1

.2.3

.4.5

.6.7

.8.9

1P

roce

ss in

nova

tors

qu

antil

es

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1Non-process innovators quantiles

U(0,1)

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

Are process innovators more productive than non-process innovators?

Result 1: In general for SMEs, PI are more productive in terms of TFP than NPI

10

On the relationship between process innovations and SMEs productivity

11

As a first step of the analysis of the relationship between process innovations and productivity, we check whether among non-process innovators today those that will introduce process innovations in the future are already more productive than those that will not If future process innovators are ex-ante more productive, one

would find that these firms would perform better in the future even without introducing process innovations

On the relationship between process innovations and SMEs productivity

12

Thus, if selection into introducing process innovations is not a random a process, the effects of the introduction of process innovations on SMEs productivity growth cannot evaluated simply by comparing the productivity growth of SMEs that start introducing process innovations and SMEs not introducing process innovations

we need to use techniques to explicitly take into account this non-random selection into the introduction of process innovations (matching)

Are the ex-ante more productive SMES those that start introducing process innovations?

13

To test whether among non process innovators in t-1 those that introduce process innovations in t are more productive in t-1, we should compare for each year the TFP distributions:

First time process innovators (FTPI) in t: SMEs that implement for the first time a process innovation in t (for t=1991,…,2002)

Non process innovators in t (NPI): firms that have not implemented a process innovation until t-1 and do not implement it at time t either.

Are the ex-ante more productive SMES those that start introducing process innovations?

Year Number of FTPI

1992 47

1993 51

1994 36

1995 30

1996 21

1997 28

1998 43

1999 26

2000 24

2001 33

2002 12

Total 1992-2002 351

14

Problem: small size of the cohorts of FTPI from 1991 to 2002 suggest not to carry out the KS tests year-by-year as they would be scarcely reliable.

Solution: apply this test jointly for the full sample

Are the ex-ante more productive SMES those that start introducing process innovations?

15

Therefore, we compare using stochastic dominance techniques:

1991,...,2002 1991,...,2002 1991,...,2002 1991,...,2002.F z vs G z

1991,...,2002

1991,...,2002

where: is the previous TFP distribution of the 12 cohorts of FTPI

G is the yearly average TFP distribution over 1991-2002 for NPI

F

A NPI is now a firm that does not introduce any innovation any of the years it is in the sample

To get the previous TFP distributions of FTPI, we follow two different approaches:

1. TFP in t-1 for FTPI in t

2. Average TFP: calculated from the first year the firm appears in the sample up to t-1

Are the ex-ante more productive SMES those that start introducing process innovations?

The results of the KS of stocahstic dominance can be summarized in the two following figures:

16

0.2

5.5

.75

1C

um

ula

tive

pro

ba

bili

ty

-2 -1 0 1 2Productivity

Non-process innovat. First-time process innovat.

(a):TFP. en t-1

0.2

5.5

.75

1C

um

ula

tive

pro

ba

bili

ty

-2 -1 0 1 2Productivity

Non-process innovat. First-time process innovat.

(b): Mean TFP up to t-1

Are process innovators more productive than non-process innovators?

Therefore, the previous productivity distribution of FTPI dominates that of NPI

17

Result 2: Among non-process innovators SMEs those that start introducing process innovations are the most productive ones

Non-random selection into the introduction of process innovations that should be taken into account when analyzing the effects of process innovation in SMEs productivity growth using of matching procedures

Does process innovations boost SMEs productivity growth: a matching approach?

With non-random selection, it is not correct to compare the TFP growth of FTPI and NPI, as the already better FTPI could perform better in the future even without introducing any PI.

We need to compare the actual TFP growth of FTPI after introducing the PI with the TFP growth of the same firm if it would not have introduced a PI Problem: we do not have information about the counterfactual situation: TFP growth of FTPI if it would not have introduced process innovations

18

Does process innovations boost SMEs productivity growth: a matching approach?

Matching techniques provide a way to construct a control group from the pool of NPI that provides for each FTPI a matched NPI unit that is as similar as possible to the FTPI

Differences in TFP growth between FTPI and matched-NPI can be attributed to the fact that FTPI introduce a PI.

19

How does matching techniques work?

To construct the counterfactual (TFP growth of FTPI if it would not have introduced process innovations) matching techniques identify among the control group of NPI those with a distribution of observable characteristics affecting TFP growth and the probability of becoming a FTPI as similar as possible to that of FTPI in t.

It is assumed then, that conditional on X, firms with the same characteristics are randomly exposed to start introducing process innovations.

Problem: the number of variables that can potentially affect the probability of becoming a FTPI is quite large

Which ones should we choose? If we choose more than one which is the appropriate weight for each one of

them?

20

How does matching techniques work?

To solve this problem we use propensity score techniques. Before performing the matching itself we run a probit model using as

explanatory variables those that may potentially affect the probability of becoming a FTPI

21

Probit regression of the probability of becoming a first-time process innovator

Export Intensityt-1 0.1562 (0.056)***Significant market sharet-1 0.0131 (0.022)log(employment)t-1 0.0343 (0.014)**log(TFP)t-1 0.1070 (0.041)***Advertising intensityt-1 1.0021 (0.680)Foreign capital participationt-1 0.0002 (0.036)Legal formt-1 0.0454 (0.022)**R&D Intensityt-1 3.2809 (1.245)***Complementary R&D activitiest-1 0.0888 (0.024)***

Number observations 1667

Notes:(1)* significant at 10%, ** significant at 5% and *** significant at 1%.(2)Robust standard errors in parentheses. (3)The regression includes year dummies and 2-digit year dummies.

How does matching techniques work?

We use two matching methods: One-to-one: nearest neighbour each FTPI is matched with the NPI with

closest propensity score (we impose common support) Kernel matching (Epachenikov) all FTPI are matched with a weighted average

of all (some) NPI with weight that are inversely proportional to the distance between the propensity score of FTPI and NPI (we impose common support)

In order to get an idea both of size and life-span of the effect of introducing process innovations we repeat the process for:

t-1/t ; t/t+1 ; t+1/t+2 ; t+2/t+3 ;t+3/t+4 Results and quality of the matching are quite similar using

both methods, we explain here those of one-to-one matching

22

Results

Results can be summarized in the following figure:23

Results

Result 4: The poductivity gains of FTPI over NPI have an inverted U-shaped form: They start 1 year after implementing the innovation Peak after 2 year Start decreasing after 3 years Vanish after 4 years

24

Possible interpretation

Process innovators make the effort to create a new process innovation that confers them a temporary “extra productivity growth”

However, as this process innovation become “common knowledge” progressively and are incorporated by other firms the path of productivity growth of process and non-process innovators converge. This is quite likely if we take into account that, in general Spanish

SMEs process innovations imply mainly the introduction of new machines that could be copied or acquired by competitors, and worse appropriability conditions.

25

Thanks for your attention

26