2

Research Policy 30 (2001) 1185–1201

Export behaviour of Italian manufacturing firms over the nineties: the role of innovation Roberto Basile∗ ISAE (Istituto di Studi e Analisi Economica — Institute for Studies and Economic Analyses) Piazza Indipendenza, 4 Rome, Italy Received 6 March 2000; received in revised form 28 April 2000; accepted 2 October 2000

Abstract This paper analyses the relationship between innovation and export behaviour of Italian manufacturing firms in different exchange rate regimes. The paper is based on a sample including firms which have carried out an innovative activity through R&D investments and/or investments in new plants or equipment. Export behaviour is defined in a dual way: as a probability for a firm to export and as the propensity to export for the exporting firms. An empirical model of the determinants of export behaviour is estimated using the Cragg’s specification of the Tobit model. The results suggest that innovation capabilities are very important competitive factors and help explain heterogeneity in export behaviour among Italian firms. However, the exchange rate devaluation reduces the importance of technological competitiveness in affecting exports because it allows also non-innovating firms to enter foreign markets. Moreover, once new firms have entered the market, they continue to be exporters also when the exchange rate returns to its previous level (hysteresis). The export intensity of innovating firms is systematically higher than that of non-innovating firms. The paper provides also specific evidence on export behaviour of firms localised in the south of the country (Mezzogiorno). © 2001 Elsevier Science B.V. All rights reserved. JEL classification: D21; O31; F10 Keywords: Firm behaviour; Innovation; Trade

1. Introduction Models of international trade generally assume that, within a particular industry, all firms in a particular country are symmetrical. 1 That is, they all ∗

Present Address: Roberto Basile, Via Citt`a di Castello, 14, Roma CAP 00191. E-mail address: [email protected] (R. Basile). 1 The presence of asymmetries among firms in export behaviour has recently been introduced by Venables (1994) in a model of monopolistic competition a` la Dixit-Stiglitz to investigate the way in which the symmetry assumption may bias our assessment of the gains from economic integration. He observed that, by reducing trade barriers, large firms export while small firms supply only

face the same demand function and have the same technology. As a result, all firms set the same price and produce and supply the same quantities. Thus, if there are exports, all firms in the industry are exporters. Yet, a growing body of empirical work has documented the considerable inter-firm differences in export performance for a variety of countries, both developed and less developed. This variation seems to be related to technological advantages, as well as the domestic market. So, he concludes, ‘by replacing many small firms with fewer large firms, integration may have an adverse effect on product variety, and this may reduce the (still positive) gains from integration’ (p. 131).

0048-7333/01/$ – see front matter © 2001 Elsevier Science B.V. All rights reserved. PII: S 0 0 4 8 - 7 3 3 3 ( 0 0 ) 0 0 1 4 1 - 4

1186

R. Basile / Research Policy 30 (2001) 1185–1201

to other characteristics of the firm (e.g. employment, shipment, labour cost and capital intensity), the sector and the country in which it is localised. Wakelin (1998), Kumar and Siddharthan (1994), Enthorf and Pohlmeier (1990) and Hirsch and Bijaoui (1985) have analysed the relationship between innovation and exports at firm level. They generally conclude that innovation, measured by proxies of input (e.g. R&D expenditure) or of output (e.g. number of innovations), is an important factor in explaining export performance. Roberts and Tybout (1997), Bernard and Wagner, 1997, 1998 and Bernard and Jensen (1999) have considered the role of entry costs (e.g. investments in export marketing, new commercial networks, etc.) in the export decision. These studies generally found that entry costs, measured by prior exporting experience, increase the probability of exporting. This paper is concerned with the export behaviour of firms in Italy. During the 1990s, Italian export flows have been growing very fast. Certainly, this phenomenon has to be connected to the exit of the Lira from the Exchange Rate Mechanism (ERM) in September 1992. Generally, indeed, large changes in exchange rates are strongly correlated with large changes of export flows of a country. However, the magnitude of these changes in the short run is in part determined by the ability of firms to adjust their output to the changes in relative prices. Aggregate output changes result from two related but distinct activities by firms. First, existing exporters can increase their export intensity. This may result either from a redirection of output destined to the domestic market to foreign customers or from an overall increase in production coupled with an expansion of exports. The alternative mechanism for the export response is through entry of existing or new firms into the export market. In this paper we will show that both of these mechanisms took place. The main focus of the paper is to analyse and compare the relationship between export behaviour and innovation capabilities of Italian firms in three different periods: (a) in 1991, that is a period characterised by a fixed exchange rate regime; (b) in 1994, that is after the exit of the Lira from the ERM and its strong devaluation and (c) in 1997, that is after a strong appreciation of the real exchange rate and the return of the Italian currency to the ERM. Thus,

a model of export behaviour is estimated for these 3 years and then the results are compared. Following Wakelin (1998), export behaviour is defined in a dual way: both as a probability for a firm to export and the propensity to export for the exporting firms. Different factors are considered as potential determinants of export behaviour: (a) innovation activity, provided by new product developments and cost saving technical processes; (b) other firm characteristics (labour cost per unit of product, firm size and ownership structure); (c) industry and (d) geographical localisation of the firm. The paper also analyses the specificity of export behaviour of firms localised in the south of the country over the same period. The south of Italy (Mezzogiorno) plays a marginal role in Italian international trade. During the late 1980s and the early 1990s, Mezzogiorno’s share in manufacturing export of Italy was about 8%, while its share in manufacturing valueadded of the country was about 13%. It is frequently noted that it is in the area of non-price competitiveness that Mezzogiorno’s economy is particularly weak, reflecting, among other factors, poor product design and quality, after sales service and reliability. Recently, however, southern regions experienced a higher international involvement, mainly due to a stronger price competitiveness and a depressed domestic demand. Since 1992, indeed, the growth of export flows from the south was particularly sharp. While total national exports increased 12.1% per year from 1992 to 1997, total southern exports increased 13.7% per year over the same period. As a result, the share of the south in total export of Italy started to grow. According to Viesti (1998) and Bodo and Viesti (1997), the improved relative position of southern firms on international markets has also to be connected to an increased non-price competitiveness, with southern firms gradually adapting product quality to foreign demand. The present work tries to shed some light on this issue. Section 2 of the paper presents the theoretical framework and the a priori expectations of the results. Section 3 describes the data set used. Section 4 considers the econometric specification used. Section 5 gives the results for the probability for a firm to export and for the propensity of firms to export in the case of the country as a whole. Section 6 reports the results for the south. Finally, Section 7 tries to give some conclusions.

R. Basile / Research Policy 30 (2001) 1185–1201

2. Theoretical framework To organise the empirical analysis, a simple short run microeconomic model of export behaviour has been developed. The market structure assumed is characterised by monopolistic competition. Each firm can sell its product on the domestic market and on the foreign
market. On the first one, total demand Qd = ni=1 qid will be met by n firms, where qid is the output of firm i sold on the domestic market. The demand on the foreign market is modelled equiva
f f lently by Qf = m i=1 qi , where qi is the output of firm i sold abroad. The total output qi of firm i is f qid + qi . In addition to the output decision and the decision of how much to sell on the two markets, firms have to decide on their level of product-innovative activity. Now, these three decision levels are imagined not to be simultaneous: firms choose the level of output and the realisation of new products regardless of the intended destination. Only after new products have been developed and the level of output has been decided, firms decide which market, domestic or foreign, will yield the highest profits. 2 Furthermore, it is assumed that recently introduced product innovations (Ii ) lead to higher revenues (R) on both markets: δR d ( )/δIi > 0, δR f ( )/δIi > 0. Following Enthorf and Pohlmeier (1990), revenues are assumed to be additively separable in the two components: revenues due to unchanged products and revenues due to product innovations Ri = {p d (Qd ) + a d Ii }(1 − yi )qi f

f

f

+{p (Q ) + a Ii }yi qi

(1)

2 Obviously, one cannot exclude that a firm may innovate or increase the output in order to export, especially when it faces an increasing foreign demand. Such a “demand pull” point of view of the firm’s innovation capability appears, however, to be too limited. Indeed, innovation capabilities are also determined by factors different from the demand. In particular, a firm’s innovation capability has to be partly considered as the result of the past innovation experience of the firm itself: firms that innovated in the past are more likely to innovate today and in the future (path dependence; see, for example, Dosi, 1988). Thus, in a model of export behaviour, innovation capability might be considered as predetermined.

1187

where the parameters ad and af give the relative profitability of product innovations in the two markets and f yi = qi /qi is the export share. Production costs are given by C(Zi |qi ), where Zi is a vector of firm specific factors, such as labour cost, productivity, firm size and ownership structure (see infra, p. 7). These factors together with exogenous factors, such as the exchange rate (X), also affect the costs of exported output G(X, Zi |yi ). There are also specific costs associated with penetrating foreign markets (Pi ). These might include “i) the costs of advertising, identifying appropriate trade partners and obtaining information about market conditions in export markets; ii) the costs of constructing and maintaining marketing networks; iii) the costs of developing new marketing techniques; iv) the costs of negotiating, writing and enforcing contracts between the parties; v) the costs of information on government regulations and other policies in both foreign and domestic markets” (Yhee et al., 1998, p. 5). These costs are considered as sunk in nature: they recur in full if the firm exits the export market for any amount of time (Bernard and Wagner, 1998; Yhee et al., 1998). Now, the firm is assumed to be always able at producing at the profit-maximising level of export (qf ∗ ), that can also be equal to zero. In the short run, the firm receives profits Πi = {p d (Qd ) + a d Ii }[(1 − yi∗ )qi ] + {p f (Qf ) +a f Ii }yi∗ qi − C(Zi |qi ) −G(X, Zi |yi∗ ) − Pi

(2)

Starting from the objective function (2), a number of hypothesis on export behaviour determinants may be intuitively derived. First, if product innovation has a higher return on the foreign market than on the domestic market (a f > a d ), innovating firms are expected to be also exporters and to have a higher export intensity (export/sales ratio). Firm specific factors affecting export behaviour include labour cost per unit of product (w/q), 3 process innovation (T), firm size, ownership structure (group) and the geographical localisation of the firm (south). y = F (a f − a d , w/q , T , size, group, south) +

−

+

+

+

−

3 Labour is assumed to be the only variable input in this short run model.

1188

R. Basile / Research Policy 30 (2001) 1185–1201

Labour costs per unit of product are a measure of cost/price competitiveness and thus, are likely to have a negative impact on exports in cost sensitive export markets. Higher process innovations (T) are assumed to reduce production costs and thus, to increase the firm’s competitiveness abroad. Firm size is expected to have a positive relationship with export as international trade may be a way of extending the market and allowing exploitation of scale economies (Krugman, 1979). Firms belonging to a business group are expected to be more likely to export, since the group allows firms to overcome the problem of lacking resources necessary to export, such as finance, marketing, physical and human capital. For example, subsidiaries may exploit export channels of other firms belonging to the same group. Finally, the geographical localisation may strongly affect both efficiency and export propensity of firms. In particular, firms localised in the less developed regions of a country (south) do not benefit from market and technological knowledge spillovers also, they face higher transport costs than firms localised in the north when trying to reach foreign markets. Obviously, the relative importance of innovation, labour costs and other firm specific factors is expected to change over time because of the change of an exogenous factor, such as the exchange rate. For example, if a f > a d (product innovation has a higher return on the foreign market than on the domestic market), then all innovating firms will always export, while a large exchange rate shock may induce some non-innovating and non-exporting firms to enter foreign markets. In other terms, innovation capabilities reduce their role of non-price advantages when a strong exchange rate shock occurs. However, the reverse is not necessarily true. As emphasised before, in order to enter foreign markets, firms might incur in penetration costs. Since these costs are sunk, when the exchange rate appreciates and returns to its original level, not all of the new entrants exit (Baldwin and Krugman, 1989). This phenomenon has come to be called hysteresis in trade. Thus, once non-innovating firms have become exporters because of a large exchange rate shock and have also invested in trade penetration (P), as the currency returns to its previous level, they will find it profitable to remain in the foreign market at least in the short term.

3. Data and variables The empirical work draws on firm-level data collected by Mediocredito Centrale in 1992 for the period 1989–1991, in 1995 for the period 1992–1994 and in 1998 for the period 1995–1997. Mediocredito surveys cover a sample of manufacturing firms operating in Italy with more than 10 employees and all the firms with more than 500 employees. 4 The sample (more than 4000 firms) is random and stratified according to the size of the firm, in terms of the number of employees, the sector and the region. Unfortunately, for the third survey, the code number of the firm is not available. This precluded a full exploitation of the panel nature of the data. The dependent variable is the export performance, calculated by the ratio of export and sales in 1991, 1994 and 1997. The currency devaluation occurred in September 1992, coupled with a strong reduction of the labour cost dynamics, determined a strong depreciation of the Lira’s real effective exchange rate. 5 This large exchange rate shock allowed a huge increase of Italian exports until the 3rd quarter of 1995 (Fig. 1). Coherently with this macro evidence and according to Mediocredito data, in 1994 both the percentage of exporting firms and the share of output shipped abroad by the average exporting firm (the export intensity) were higher than in 1991 in all sectors and in all size classes, with very few exceptions 6 (Table 1). The relative presence of exporting firms increased more among 4 The sample procedure used to collect data was the same over the three surveys. 5 The currency depreciated twice during this period. The first depreciation occurred between the 3rd quarter of 1992 and the 1st quarter of 1993, while the second one occurred between the 2nd quarter of 1994 and 1995. 6 Sectors are defined according to Pavitt’s classification. As well known, Pavitt (1984) grouped sectors according to different characteristics of the firms: innovation behaviour (product and process innovation, sources of knowledge, appropriability regimes and the like), production organisation (e.g. degree of vertical integration) and competitive factors. Thus, he individuated four categories of sectors: 1) traditional or supplier dominated sectors (e.g. footwear, clothing, food); 2) specialised supplier sectors (e.g. machine tools); 3) scale intensive sectors (e.g. cars, television set) and 4) science based sectors (e.g. electronics, pharmaceuticals). Technological interdependencies among these four categories are very complex and shape a circular process of production and use of technology. Pavitt’s taxonomy has showed to be very useful in many empirical studies on trade and competitiveness both at macro and micro level.

R. Basile / Research Policy 30 (2001) 1185–1201

1189

Fig. 1. Italy — export growth and real exchange rates (“export” is the cyclical component of the index of export at constant prices, obtained removing the trend with the Hodrick–Presctt’s filter. Data are normalised [X − mean (X)/standard deviation (X)]. Source: elaboration on Istat (National Institute of Statistics) (export) and Bank of Italy (exchange rate) data.

scale intensive and science based sectors, rather than among traditional and specialised suppliers sectors, that is sectors in which Italy is highly specialised. Thus, after the currency devaluation occurred in 1992, Italian firms have adjusted their output to the changes in relative prices in different ways. First, existing exporters increased their export intensity, either through a redirection of output destined for the domestic market to foreign customers or through an overall increase in production coupled with an expansion of exports. Second, new firms entered into the export market. Since the 2nd quarter of 1995, the Lira’s real effective exchange rate strongly appreciated and returned to the level reached in 1993. In spite of this lower cost competitiveness, the reduction of the export volumes was contained. This evidence may be interpreted in terms of a persistent effect of the large exchange rate shock (see Baldwin and Krugman, 1989). Moreover, in 1997 an expansion of the world demand allowed a further increase of exports. Coherently, in 1997 both export participation and intensity of Italian firms remained at the level reached in 1994. The existence of sunk costs, necessary to penetrate the foreign market since 1992, has probably induced exporting firms to reduce their profit margins in order to remain into the market. Differently from 1994, the fraction of firms exporting increased in the sectors of Italian trade specialisation (traditional and

specialised suppliers), while export intensity increased in specialised suppliers and in science based sectors. This simple statistical evidence suggests to compare the export behaviour of Italian firms over the three different phases of the exchange rate and to analyse both the export participation and the export intensity, instead of confining the research only to the participation decision, as usually done in recent studies of export behaviour (Roberts and Tybout, 1997; Bernard and Wagner, 1997, 1998; Bernard and Jensen, 1999). Different independent variables are included in the empirical model as suggested by the theoretical framework (see Table 2). The innovation variables, as well as the other firm specific variables, refer to the entire periods 1989–1991, 1992–1994 and 1994–1996, respectively, that is the 3 years before those of the dependent variable (1991, 1994 and 1997). 7 With this time-lag, possible simultaneity bias may be avoided. In analysing the impact of innovation on export behaviour, different firm level studies have used R&D expenditure as proxy of innovation (e.g. Kumar and Siddharthan, 1994; Hirsch and Bijaoui, 1985). Yet, many small firms and firms operating in traditional sectors do not have a separate R&D department or even an R&D budget (Pavitt et al., 1987). Nevertheless, they innovate through acquiring knowledge 7 Continuous variables, such as SIZE and LCUP, are calculated as averages over each period.

1190

R. Basile / Research Policy 30 (2001) 1185–1201

Table 1 Italy — percentages of exporting firms and average export intensity by sector and sizea Sector

Average export intensity (export/total sales × 100)

Percentage of exporting firms 1991

1994

1997

1991

1994

1997

Traditional 11–20 21–50 51–100 101–250 251–500 >500

63 41 59 72 77 80 60

68 40 60 75 85 91 93

70 58 69 77 82 90 98

31 32 29 33 32 31 28

37 33 38 38 37 33 32

38 38 37 38 37 41 40

Scale intensive 11–20 21–50 51–100 101–250 251–500 >500

58 33 50 73 70 72 65

69 39 57 77 84 92 93

65 45 59 73 83 88 91

29 22 25 28 31 29 34

34 24 34 32 36 38 36

33 28 28 35 34 44 41

Specialised suppliers 11–20 21–50 51–100 101–250 251–500 >500

70 38 74 76 82 77 68

77 44 71 90 86 89 85

81 68 80 87 95 96 97

42 35 34 41 45 51 44

47 32 41 48 52 56 44

44 35 41 45 53 59 55

Science based 11–20 21–50 51–100 101–250 251–500 >500

60 33 40 64 73 76 61

71 56 51 77 74 74 86

70 61 66 74 89 71 77

23 11 12 14 41 14 19

30 38 27 34 33 28 26

34 28 30 35 48 39 36

All sectors 11–20 21–50 51–100 101–250 251–500 >500

63 37 58 73 76 76 64

70 40 60 79 85 90 91

72 57 69 79 86 90 91

32 29 28 32 35 34 33

38 30 37 38 39 39 35

38 35 36 39 41 48 43

a

Source: elaboration on Mediocredito data.

embodied in new technical equipment or other external sources. In order to avoid the bias of R&D expenditure as an innovation proxy, other studies have used either proxies of innovation output (Wakelin, 1998; Enthorf and Pohlmeier, 1990) or a mix of proxies of both internal (R&D) and external (e.g. purchase of advanced equipment) sources of knowledge (Brower and Kleinknecht, 1993). Here the second strategy is

followed by using: a) two dummy variables (‘R&D strategies’), which indicate, respectively whether the firm realised a product innovation (whether or not combined with a process innovation), or only a process innovation through R&D investments, if any and b) five categorical variables (‘Investment strategies’) defined in an ordinal scale (from zero to three), which indicate the importance of different objectives of new

R. Basile / Research Policy 30 (2001) 1185–1201

1191

Table 2 Description of variables Variables

Description

Y SIZE LCUP GROUP

Total exports/total sales × 100 (1991, 1994 and 1997) Number of employees Labour cost per unit of product GROUP = 1 if the firm belongs to a business group

R&D strategies PRODPROC

PRODPROC = 1 if the firms realised a product innovation (whether combined or not with a process innovation) through R&D investments ONLYPROC = 1 if the firms realised only a process innovation through R&D investments

ONLYPROC Investment strategies PRDQUAL MOREPROD NEWPROD RAWMAT LABOUR MEZZ TRAD SI SS SB

Investments in capital equipment oriented to improve firm’s product quality (intensity from 0 to 3) Investments in capital equipment oriented to improve firm’s productivity (intensity from 0 to 3) Investments in capital equipment oriented to develop new products (intensity from 0 to 3) Investments in capital equipment oriented to reduce the use of raw material (intensity from 0 to 3) Investments in capital equipment oriented to employ less labour (intensity from 0 to 3) MEZZ = 1 if the firm is localised in a southern region TRAD = 1 if the firm belongs to a ‘traditional’ sector SI = 1 if the firm belongs to a ‘scale intensive’ sector SS = 1 if the firm belongs to a ‘specialised supplier’ sector SB = 1 if the firm belongs to a ‘science based’ sector

investments in advanced equipment (PRDQUAL, MOREPROD, NEWPROD, RAWMAT, LABOUR), if any. However, as easily expected all these innovation variables are not independent among each other. In order to avoid multicollinearity problems, some

non-parametric measures of association between these variables have been calculated (Table 3). The index proposed by Agresti (1981) has been used as measure of association between ordinal variables (the five ‘Investment strategies’) and dichotomous variables

Table 3 Coefficients of association between innovation variablesa PRDQUAL

MOREPROD

NEWPROD

RAWMAT

LABOUR

1991

1994

1997

1991

1994

1997

1991

1994

1997

1991

1994

1997

1991

1994

1997

All firms PRODPROC ONLYPROC MOREPROD NEWPROD RAWMAT LABOUR

0.31 0.31 0.71 0.76 0.71 0.67

0.14 0.14 0.67 0.56 0.54 0.51

0.01 0.09 0.36 0.56 0.56 0.56

0.23 0.35

0.13 0.14

0.09 0.06

0.26 0.16

0.38 −0.05

0.42 −0.19

0.03 0.15

0.08 0.09

0.05 0.05

0.21 0.25

0.05 0.10

0.09 −0.03

0.62 0.67 0.74

0.53 0.57 0.53

0.48 0.59 0.55

0.64 0.48

0.48 0.43

0.44 0.49

0.74

0.83

0.73

Exporting firms PRODPROC ONLYPROC MOREPROD NEWPROD RAWMAT LABOUR

0.26 0.35 0.64 0.72 0.64 0.61

0.13 0.14 0.56 0.44 0.47 0.48

0.02 0.03 0.55 0.16 0.25 0.28

0.19 0.31

0.12 0.12

0.10 0.05

0.22 0.17

0.35 −0.09

0.41 −0.24

0.01 0.12

0.08 0.08

0.05 0.01

0.18 0.23

0.08 0.07

0.08 −0.05

0.53 0.60 0.69

0.41 0.55 0.51

0.13 0.16 0.26

0.58 0.42

0.46 0.40

0.43 0.39

0.62

0.82

0.75

a

Source: elaboration on Mediocredito data.

1192

R. Basile / Research Policy 30 (2001) 1185–1201

(the two ‘R&D strategies’), 8 while Goodman-Kruskal coefficients have been used to measure association between the ordinal variables (‘Investment strategies’). 9 Coefficients above 0.50 are considered to be rather high and are given in bold letters. Only some coefficients of ‘Investment strategies’ are higher than 0.50. These results suggest to consider, within the multivariate analysis, only the two ‘R&D strategies’ and the two variables NEWPROD and LABOUR. 10 Finally, LCUP is the ratio between wage per employee and labour productivity. This is a measure of cost competitiveness. Normalising wages per employee by the productivity level is very important: lower wages per employee, by themselves, might not imply a competitive advantage if they are accompanied by lower labour productivity. Thus, the inclusion of this variable is considered as a good alternative to the separated inclusion of wages per employees and labour productivity variables, as often done in other studies of export behaviour (e.g. Bernard and Jensen, 1999; Bernard and Wagner, 1998).

4. The econometric specification Before presenting the results of the multivariate analysis of export behaviour, some methodological issues have to be discussed. The dependent variable, 8 This indicator (δ) is constructed comparing the probability to observe higher levels of an ordinal variable Y when a dichotomous variable X equals 0, with that when X = 1. If the variables are independent, the two probabilities are equal. The values of δ vary between −1 and 1. If the variables are independent, δ = 0. 9 The association between two ordinal phenomena (X and Y) can be synthesised by using the concepts of concordance and discordance. A couple of observations is defined concordant when the statistical unit assuming higher levels for the first variable, assumes also higher levels for the second one; while it is defined discordant when the statistical unit assuming higher levels for the first variable, assumes lower levels for the second one. Given the total number of concordant (C) and discordant (D) couples of observations in a contingency table, a simple indicator of association between two ordinal variables may be calculated as follows: γ = (C − D)/(C + D) (Goodman and Kruskal, 1954). The values of γ vary between −1 and 1. In particular, γ = 1 if D = 0, and γ = −1 if C = 0. Thus, |γ | = 1 suggests a monotonic relation between the variables: when X grows, Y does not decrease (γ = 1), or does not increase (γ = −1). If the variables are independent, γ = 0. 10 Similar results have been found for the south.

y (export/sales) is a doubly truncated random variable: its values vary between 0 and 1 by definition. Moreover, this variable often takes the value of zero. A generally used approach to dealing with the problem of censored samples is the Tobit model (see, for example, Kumar and Siddharthan, 1994). This model uses all the available information from the explanatory variables, including those for which the dependent variable is zero. In the Tobit model, the change of the expected value of the dependent variable y, with respect to each regressor, has two components. One effect works by changing the conditional mean of y (intensity) and the other by changing the probability that an observation will be positive (participation). However, it is reasonable to maintain that the explanatory variables have different effects on the participation decision (export) and on the decision on how much to export conditional on participation. Since the Tobit constraints the participation equation and the intensity equation to have the same parameters, the Tobit may be mis-specified in this case, and this mis-specification may have profound and undesirable consequences for the estimates. This constraint may be relaxed by viewing the problem as involving two equations. Cragg (1971) proposed a two-stage specification, which weakens one of the central characteristics of the Tobit model. In Cragg’s specification of the Tobit model, the probability of a non-limit outcome is determined apart from the level of the non-limit outcome. The first stage of this specification uses the whole set of data and considers the probability to sell abroad. The dependent variable y is binary, taking a value of 1 when the firm sells abroad and zero when it does not. In this case, a Probit model is appropriate. For the second stage, only the subset of firms which sell abroad are considered. A truncated estimation procedure is used as the dependent variable is observed only if it is greater than zero (i.e. this sub-sample is truncated). This double specification can be tested as the unrestricted model against a Tobit model. In the field of export behaviour analysis, Wakelin (1998) has used the Cragg’s specification of the Tobit model. The assumption underlying the Cragg’s specification is that the two stages are independent of each other, that is the disturbances in the latent regression underlying the Probit model and those in the truncated regression are independent. If the two stages are not

R. Basile / Research Policy 30 (2001) 1185–1201

independent, the resulting estimates cannot hold true for the population as a whole, because they are based on a non-randomly selected subset. Thus, the resulting truncated estimates will be biased estimates of the true population parameters and more importantly, they are also biased estimates of the parameters among that group of the population for whom y > 0. The sample-selection model extends Cragg’s model by relaxing the assumption that the two stages are independent. The basic idea of such a model is that the outcome variable, y, is only observed if some criterion, defined with respect to a different set of variables is met. The simplest form of such a model has two stages: in the first stage, a dichotomous variable z (= 0 or 1) determines whether or not y is observed, y being observed only if z = 1; in the second stage, the expected value of y is modelled, conditional on it having been observed. Using Heckman’s (1979) procedure, the Probit results are firstly taken and for the sub-sample for whom z = 1, the estimate of φi /Φi (the inverse of Mill’s ratio, Θ i ) is computed, where Φi is the standard normal distribution function and φ i is the corresponding standard normal density function. Then, for this same sub-sample, OLS is used to regress y on xi and on the estimate of Θ i . However, OLS coefficients, although efficient, are inconsistent. In order to obtain consistent and efficient coefficients, maximum likelihood estimators have to be computed.

5. Estimation results A sample selection model was firstly applied for the analysis of export behaviour. Since the correlation between the error term in the Probit equation and in the 2SLS equation was found to be low, any problem of selection bias was rejected and concluded that the two stages were independent. Thus, the Cragg’s specification of the Tobit model was applied to the analysis of export behaviour. The Tobit restriction was rejected at the 99% probability level using a chi-square test based on the likelihood ratio test statistics. 11 All this means that: i) the subset of firms selling abroad represents an unbiased sample of exporting firms and ii) explanatory The test is the following: λ = −2[ln Lt − (ln Lp + ln Ltr )], where ln Lt is the likelihood for the Tobit model, ln Lp for the Probit model and ln Ltr for the truncated model. 11

1193

variables have different effects on the participation decision (whether to sell abroad) and on the intensity decision (the fraction of output shipped abroad). 12 Table 4 reports the results for the country as a whole. The first column shows the results of the random effect Probit model for the probability of exporting. 13 These estimates account for a large portion of cross-firm differences in export behaviour: the percentage of correct prediction is about 70%. According to the results, firms that introduce product and/or process innovations either through R&D activity or through investments in new capital equipment are more likely to export. This means that: i) Italian manufacturing firms have a higher return on their product innovations on the foreign market than on the domestic market (a f > a d ) and ii) process innovations allow Italian firms to reduce costs and compete in foreign markets. These results are not obvious. For the case of Germany, Enthorf and Pohlmeier (1990) found a negative impact of product innovation on export propensity, that is German manufacturing firms have a higher return on their product innovations on the domestic market than on the foreign market (German domestic demand is characterised by a lower price elasticity). Wakelin (1998) also found that being an innovative firm in the UK has a negative impact on the probability of exporting and concluded that innovating firms are more inclined to use their innovations to exploit the domestic (UK) market rather than to enter foreign markets. The hypothesis that firms with higher labour costs per unit of product (LCUP) are less likely to enter foreign markets is confirmed. The magnitude of the coefficient of this variable is, however, relatively low and only weakly significant (P < 0.10). As expected, firms that are part of a business group are more likely to export, while firms localised in southern regions are less likely to export. Finally, the coefficients of the three industries’ dummy variables turned out to have all a positive and significant effect on export participation, with the magnitude of TRAD and SS double 12 The hypothesis of sample selection bias as well as the Tobit restriction were rejected in the case of all cross-sections reported in Table 4 and in the case of those reported in Table 7 (the case of south Italy). 13 As it has been mentioned above, the code number of the firm is not available for the third survey. Thus, the panel used includes only the first two periods. Then, it is unbalanced: it consists of 7855 observations for 5852 firms.

1194

R. Basile / Research Policy 30 (2001) 1185–1201

R. Basile / Research Policy 30 (2001) 1185–1201

1195

Table 5 Export and innovationa Average export intensity (export/total sales × 100)

Percentage of exporting firms

PRODPROC No PRODPROC ONLYPROC No ONLYPROC a

1991

1994

1997

1991

1994

1997

89 58 83 62

88 65 84 69

90 68 83 70

31 18 28 20

37 23 30 26

40 26 36 27

Source: elaboration on Mediocredito data.

than that of SI. These results are not surprising, given the strong specialisation of Italy in traditional and specialised supplier sectors and its relative weakness in science based sectors (the reference category). Finally, a dummy variable for the 1994 was included and found significantly positive: it caught the effect of the large exchange rate shock. The rest of the table contains Probit 14 and truncated estimates of coefficients for each explanatory variable and for the 3 years (1991, 1994 and 1997). As far as Probit estimates are concerned, it is interesting to observe that the magnitude of the coefficients of all innovation variables included is higher in 1991 than in 1994. Moreover, the coefficients of the two proxies of ‘Investment strategies’ (NEWPROD and LABOUR) are only weakly significant in 1994. This suggests that non-innovating firms have been more able to export in 1994 than in 1991, mainly because of the stronger price competitiveness offered by the currency devaluation. Moreover, in 1997 in spite of the exchange rate appreciation, less innovative firms continued to be exporters: the coefficients of PRODPROC and ONLYPROC are lower than in 1994 and 1991. The relationship between innovation and export behaviour may be easily analysed in a univariate context (Table 5): 90% of firms that have introduced product innovations (PRODPROC) over each period were also 14 In order to explicitly test whether the Probit coefficients were stable across the two cross-sections (1991 and 1994), a random effects probit model was run, which included all the explanatory variables, plus the dummy variable D94 and the interactions between each explanatory variable and D94. A likelihood ratio test was then performed to assess the joint effect of D94 and the interaction variables. The null hypothesis (no joint effect) was rejected at the 99% probability level using a chi-square test. This suggested running cross-section estimates for each year in order to compare the marginal effects of explanatory variables.

exporters. Instead, the percentage of non-innovating firms exporting was about 58% in 1991, 65% in 1994 and 68% in 1997. Similar results were found for the variable ONLYPROC. As discussed above, the different export behaviour of non-innovative firms may be partly explained in terms of entry costs. A large exchange rate shock may induce some non-innovating firms entering foreign markets, investing in entry (sunk) costs and thus, remaining in the foreign market when the exchange rate returns to its previous level (hysteresis). Turning to Table 4, the coefficient of SIZE is significantly positive in 1994 and 1997, but not in 1991, in the Probit estimates. This does not imply, however, that in 1991 the size was not an important factor to enter foreign markets. For 1991, indeed, the coefficient of size lost significance, only after innovation variables have been included. In other words, in 1991 the ‘chumpeterian’ positive correlation between size and innovation was higher than in 1994 and 1997. Belonging to a business group increases the likelihood to export more before than after the devaluation: the coefficient of the dummy variable GROUP is higher in 1991 than in 1994 estimation. Thus, a strong improvement of price competitiveness, such as that derived from the devaluation of the Italian Lira, helps firms reduce financial as well as other resource constraints to export. Surprisingly, in 1997 the dummy variable GROUP is not significant. The hypothesis that firms with higher labour costs per unit of product are more likely to enter foreign markets after a currency devaluation is also confirmed. The coefficient of LCUP is significantly negative only in 1991 in the Probit estimates. The coefficients of sectors TRAD and SS do not show relevant changes among the 3 years considered, while SI becomes significantly different from zero in

1196

R. Basile / Research Policy 30 (2001) 1185–1201

1994 and returns to be non-significant in 1997. Finally, the negative coefficient for the south appears to be decreasing, confirming the upgraded relative position of southern firms in foreign markets. As far as truncated estimates are concerned, they show that R&D related product innovation strategies (PRODPROC) have a positive effect on the export intensity only in 1994 and 1997. This result may be interpreted as follows. As discussed above, a large exchange rate shock helps non-innovating firms enter and remain in foreign markets (at least in the short run) even when the exchange rate returns to its previous level. However, the export intensity of non-innovating firms is smaller than that of innovating firms (Table 5). Thus, the higher presence of non-innovative firms among exporters after the currency depreciation may help explain the positive relationship between innovation and export intensity in 1994 and 1997. Finally, the coefficient of LCUP is significant both in 1991 and 1994, but not in 1997.

6. The export behaviour of southern firms This section analyses the peculiarities of the export behaviour of firms localised in the south of the country. There are some reasons to carry out this specific analysis. As discussed above (Section 1), southern regions traditionally play a marginal role in total national export: about 90% of Italian exports in manufacturing industries comes from the north of the country. Since 1992, however, the growth of southern

exports was particularly strong and the share of this geographical area on total national exports increased (Fig. 2). According to some authors (Bodo and Viesti, 1997; Viesti, 1998), the recently improved relative position of southern firms on the international markets is not only the effect of the large exchange rate shock, but it has also to be connected to an increased non-price competitiveness of these firms. According to Mediocredito data, firms localised in the south show a lower export participation than northern firms (Table 6), but, after the currency devaluation, the percentage of exporting firms increased more in the south than in the rest of the country. Differences between north and south in the average export intensity are negligible. The lower export participation of southern firms is due to a number of reasons. Surely, southern firms suffer from the lack of market knowledge spillovers, while northern firms, especially those operating within industrial districts, may take advantage of the proximity relationships with other exporting firms. Moreover, many small and medium sized firms (SMEs) located in the south work mainly as subcontractors for medium and large firms, localised either in the north or in the south, which export the final product. Hence, these SMEs may be considered only as ‘indirect exporters’. Table 6 clearly shows that being a small firm in the south (that is, having less then 50 employees) represents a strong barrier to export. Hence, also in the econometric model (that is, even controlling for other factors) firm size is expected to play a stronger role in the south than in the country as a whole.

Fig. 2. Annual growth rates of exports (percentages): a comparison between Italy and south Italy (source: elaboration on Istat (National Institute of Statistics) data).

R. Basile / Research Policy 30 (2001) 1185–1201

1197

Table 6 North and south — percentage of exporting firms and the average export intensity by sector and sizea Sector

North TRAD Scale intensive Specialise suppliers Science based

Average export intensity (export/total sales × 100)

Percentage of exporting firms 1991

1994

1997

1991

1994

1997

67 62 73 62

70 73 81 72

71 70 83 75

21 18 31 14

26 25 38 22

27 24 37 26

All 11–20 21–50 51–100 101–250 251–500 >500 South TRAD Scale intensive Specialise suppliers Science based

66 40 63 77 78 78 65

73 42 64 80 88 91 92

74 61 72 81 88 93 92

33 12 19 25 27 27 22

38 13 24 32 35 35 22

39 22 26 32 38 44 39

44 31 21 43

53 41 28 65

64 37 56 53

32 23 28 30

35 30 50 23

36 28 27 28

All 11–20 21–50 51–100 101–250 251–500 >500

37 21 25 46 53 56 42

47 25 35 63 49 79 80

54 38 48 65 79 69 80

29 29 17 33 34 25 37

34 24 30 30 41 44 38

32 25 28 34 33 54 61

a

Source: elaboration on Mediocredito data.

Furthermore, firms localised in the south and belonging to a business group generally have scarce commercial autonomy from the headquarter. Hence, one may expect a weaker relationship between export behaviour and group belonging in the south than that found for the country as a whole. The pattern of international specialisation of southern regions is quite different from that of the national average. While Italy is strongly specialised in traditional and machine tools (specialised suppliers) sectors, southern regions are specialised in traditional and scale intensive sectors 15 and strongly de-specialised in machine tools sectors. 16 This last 15

This is partly due to the recent industrial development of the south. In the recent past, indeed, in order to exploit public financial incentives, some big (scale intensive) firms (e.g. FIAT) installed production plants within southern regions. 16 Some evidence on the pattern of trade specialisation of Italian regions according to the Pavitt taxonomy is reported in ISVE (1993).

evidence is particularly important. The export success of Italian industrial districts localised in the north-east-centre of the country (the, so-called, Third Italy) during the 1980s and 1990s was indeed partly linked to the user–producer relationship between traditional and machine tool sectors: firms operating in these two kinds of sectors have pushed the export success of each other. Such type of inter-sector technological and market spillover did not work within the south because of the lack of specialised suppliers. Recently, however, the growth path of export flows from the south involved different sectors. In particular, after the currency devaluation some non-exporting sectors, such as machine tools, became exporters. Mediocredito data clearly show the low export participation of specialised suppliers localised in the south before the currency devaluation, but also the strong increase of their average export intensity in 1994 and of their export participation in 1997. All these considerations suggest expecting coefficients of sectors for

1198

R. Basile / Research Policy 30 (2001) 1185–1201

R. Basile / Research Policy 30 (2001) 1185–1201

the south different from those found for the country as a whole. As far as LCUP is concerned, it is important to remember that during the, 1990s labour costs per employee were lower in the south than in the north because of the public social contribution to labour costs in the Mezzogiorno. Over the same period, however, southern firms showed on average a much lower labour productivity than northern firms. As a result, LCUP were higher in the south than in the north. Mediocredito data confirm this empirical fact: the ratio between the LCUP of the south and the LCUP of the north was 1.03 in 1991, 1.08 in 1994, and 1.13 in 1997. Thus, even controlling for other factors, a strong negative relationship between LCUP and export behaviour is expected in the case of the south, especially before the currency devaluation. Finally, it is widely recognised that the participation of southern firms to innovation activity (a very important point of the present analysis) is much lower than that of northern firms. 17 In line with these considerations, the relationship between innovation and export behaviour of southern firms is expected to be much weaker than that found for the national average. Table 7 shows the results of Probit estimates for the group of firms localised in the south. 18 Two different specifications are reported in order to avoid collinearity problems between innovation variables. Again, panel estimates show that a f > a d (the coefficients of PRODPROC and NEWPROD are significantly positive), that is also southern firms have a higher return on their product innovations on the foreign market than on the domestic market, while process innovations have not a significant effect on the export participation of southern firms. Moreover, cross-section estimates suggest that, in line with the results for the country as a whole, the impact of product innovation strategies (PRODPROC and 17

Different studies have documented the extent of regional disparities within Italy in technological innovation capacity (e.g. Iammarino et al., 1995). 18 Given the small number of southern firms in the sample, truncated estimates have not been reported. Moreover, a test was performed in order to explicitly assess whether the coefficients were stable across the two cross-sections (1991 and 1994) (see footnote 14). This test confirmed the necessity to reports cross-section results.

1199

NEWPROD) was decreasing. The other coefficients of innovation variables are not significant, a part from LABOUR in 1994. Thus, the hypothesis of a weaker relationship between innovation and export behaviour for the south is partially confirmed. Interestingly, labour costs per unit of product (LCUP) have a significant and negative impact only before the currency devaluation. Hence, after devaluation, labour costs per unit of product do not help discriminate between exporters and non-exporters. The relationship between group belonging and the export participation is positive. As in the case of the country as a whole, GROUP is not significant in 1997. Surprisingly, even the coefficient of SIZE does not show relevant differences from that found for the country as a whole.

7. Conclusions This paper has analysed the role of innovation and other firm specific factors in affecting export behaviour at the microeconomic level. In addition to firm characteristics (size, ownership structure, labour costs per unit of product and innovation), characteristics of the sector and of the region in which the firm is located have been included. The main advantage of examining trade behaviour at the firm level is the potential heterogeneity among firms. Studies at the aggregate or sector level abstract from variation among firms. The results of the analysis confirm the hypothesis that innovation is a very important competitive factor and helps explain firm level heterogeneity in export behaviour among Italian firms. In contrast to the cost based hypothesis, labour costs per unit of product seem to play a marginal role in export behaviour at the firm level. The main results emerging from this paper concern, however, the effect of a large exchange rate shock on the relationship between innovation and export behaviour. As far as the participation decision (whether to export or not) is concerned, Probit estimates have suggested that innovation strategies reduce their role of non-price competitive when a strong devaluation occurs. A large exchange rate shock, such as that occurred between 1992 and 1995 in Italy, allows some non-innovative firms enter foreign markets. Then, as the exchange rate returns to its previous level, as it

1200

R. Basile / Research Policy 30 (2001) 1185–1201

happened in 1996, not all of (non-innovating) new entrants exit (hysteresis). Moreover, truncated estimates have suggested that product innovation strategies have a positive effect on the export intensity (the fraction of output shipped abroad) only after the currency devaluation. A higher presence of non-innovating firms (that generally have a lower export intensity) among exporters after the currency devaluation help explain the emerging role of product innovation. Moreover, the specific case of the south of Italy (Mezzogiorno) has been considered in order to search the microeconomic determinants of the unexpected success of a less developed area on the export markets. It is important, however, to point out that, given the small number of southern firms in the sample, the econometric results are not very robust. With this caveat in mind, the emerging results suggest that the relationship between innovation strategies and export behaviour of southern firms is weaker than that found for the national average. Moreover, even in the south, after the devaluation, innovation strategies reduced their role of non-price competitive. In general, it can be concluded that the hypothesis formulated by some authors (Bodo and Viesti, 1997; Viesti, 1998) on the role of non-price competitiveness in explaining the recent success of southern firms on foreign markets are not confirmed by the present analysis. Instead, this analysis shows that the currency devaluation allowed less efficient firms from the south (that is, firms with a higher labour cost per unit of product) to enter foreign markets. Finally, different issues remain open. For example, which was the role of public support systems in affecting the export behaviour of Italian firms over the nineties? What kind of penetration strategies did Italian exporters use before and after the currency devaluation? What are the consequences of the Monetary Union on the export behaviour of Italian firms? These and other questions will be the subject for future research.

Acknowledgements I wish to thank Anna Giunta, Jeffrey Nugent, Almorò Rubin de Cervin and two anonymous referees for useful comments. I remain, however, solely

responsible for any omissions and mistakes, as well as for the views in the paper, which do not necessarily reflect those of ISAE. References Agresti, A., 1981. Measures of nominal-ordinal association. Journal of the American Statistical Association 76, 524–529. Baldwin, R., Krugman, P., 1989. Persistent effects of large exchange rate shocks. Quarterly Journal of Economics 104, 635–654. Bernard, A., Wagner, J., 1997. Exports and success in German manufacturing. Weltwirtschftliches Archiv 133, 134–157. Bernard, A., Wagner, J., 1998. Export Entry and Exit by German Firms, NBER WP, No. 6538. Bernard, A., Jensen, J.B., 1999. Exceptional export performance: cause, effect, or both? Journal of International Economics 47, 1–25. Bodo, G., Viesti, G., 1997. La grande svolta. Il Mezzogiorno nell’Italia degli anni novanta (Rome: Donzelli Editore). Brower, E., Kleinknecht, A., 1993. Thechnology and a firm’s export intensity: the need for adequate innovation measurement, Konjunkturpolitik, Vol. 39. Cragg, J., 1971. Some statistical models for limited dependent variables with application to the demand for durable goods. Econometrica 39, 829–844. Dosi, G., 1988. Sources, procedures and microeconomic effects of innovation. Journal of Economic Literature 26, 1120–1171. Enthorf, H., Pohlmeier, W., 1990. Employment, innovation and export activity: evidence from firm-level data. In: Florens et al. (Eds.), Microeconometrics: Surveys and Applications. Basic Blackwell, London. Goodman, L.A., Kruskal, W.H., 1954. Measures of association for cross-classifications. Journal of the American Statistical Association 49, 732–764. Hirsch, S., Bijaoui, I., 1985. R&D intensity and export performance: a micro view. Weltwirtschftliches Archiv 121, 138–251. Iammarino, S., Prisco, M.R., Silvani, A., 1995. On the importance of regional innovation flows in the EU: some methodological issues in the Italian case. Research Evaluation 5 (3), 189–206. ISVE, 1993. La proiezione internazionale del Mezzogiorno, Rapporto 1993, Il Sole 24ore libri. Krugman, P., 1979. Increasing returns, monopolistic competition and international trade. Journal of International Economics 9, 469–480. Kumar, N., Siddharthan, N.S., 1994. Technology, firm size and export behaviour in developing countries: the case of Indian enterprise. Journal of Development Studies 32 (2), 288–309. Pavitt, K., 1984. Patterns of technical change: towards a taxonomy and a theory. Research Policy 13, 343–373. Pavitt, K., Robson, M., Townsend, J., 1987. The size distribution of innovating firms in the UK: 1945–1983, The Journal of Industrial Economics 35 (3). Roberts, M., Tybout, J., 1997. The decision to export in Columbia: an empirical model of entry with sunk cost. American Economic Review 87, 545–564.

R. Basile / Research Policy 30 (2001) 1185–1201 Venables, A.J., 1994. Integration and the export behaviour of firms: trade costs, trade volumes and welfare, Weltwirtschaftliches Archiv, 118–132. Viesti, G., 1998. Esportatori ed esportazioni nel Mezzogiorno 1992–1996: un primo confronto (Roma: Rapporto ICE).

1201

Wakelin, K., 1998. Innovation and export behaviour at the firm level. Research Policy 26, 829–841. Yhee, S., Nugent, J.B., Hsiao, C., 1998. A censored switching regression approach to evaluating the effects of sunk costs and firm-level disequilibrium on export performance, mimeo.