1

Export Behaviour of Italian Food Firms Across Destinations:

Does Product Quality Matter ?

Daniele Curzi and Alessandro Olper∗

Università degli Studi di Milano

18 August, 2010

[Preliminary draft]

Abstract. Using firm-level data we investigate the export behaviour of Italian food firms during 2001-2006, focusing on the link between total factor productivity (TFP), products quality, and export across destinations. Specifically, we test the main propositions of an international trade model based on firm heterogeneity in product quality and non-homothetic preferences. Under this setting, the export intensity toward low income destinations should be negative related to TFP, as an effect of firm heterogeneity in products quality. Using different measures of revenue-TFP and several proxies for product quality we find support for the main model predictions, showing also that the elasticity of export intensity across destinations is systematically higher (in absolute value) for a sub-sample of Italian firms producing typical ‘Made in Italy’ products.

Keywords: firm heterogeneity, TFP, product quality, export destinations.

JEL Classification: F12, F14. Q17

1. Introduction

Media often emphasizes that performances of Italian food products on international markets are

largely driven by their high quality level. However and quite surprising, formal evidence of a positive

link between product quality and export performance for Italian food products is limited, and not

always in line with common intuition.

For example, Ninni, Raimondi and Zuppiroli (2006) showed that while the image of Italian food

offers protection for some traditional products, overall this protection is not strong enough to

counteract price competition. Mixed evidence about the nexus between product quality and export

performance is reported by Fischer (2007). This author found some positive link between quality and

export performance for France, Italy and Spain, (but not for UK and Germany), although these results

applied only to specific products, and are conditional to specific export destinations. This and other

facts may suggest that the relation between productivity, quality and export success is not a simple one

∗ Corresponding author: alessandro.olper@unimi.it

2

(see Manova and Zhang, 2009), and/or that this relationship is difficult to study using common proxy

for quality and export performance based on ‘aggregated’ trade data.

The relationship between product quality and export performances of Italian foods is, indeed,

normally investigated studying the evolution of unit values of the exported products computed from

trade statistics (see Scoppola, 2003; Ninni, Raimondi and Zuppiroli, 2006; Fischer, 2007). However, it

is well known that the use of unit values to infer information about quality of exported goods across

time and space, can introduce noises in the analysis, because unit values capture also aspects that are

not directly attributable to the level of product quality (see, e.g., Knetter, 1997).1

In this paper, we use an alternative approach to study the relationship between product quality and

food export performance exploiting the richness of a representative sample of Italian food firms.

Specifically, we make use of a (unbalanced) panel of roughly 750 Italian food firms observed in the

period 2001-2006, drawn from the 9th and 10th Survey on Manufacturing Firms (Indagine sulle

Imprese Manifatturiere) carried out by Unicredit-Capitalia. Although proxies for product quality are

not easy to obtain, the main advantage of this dataset is that it allows to construct a large number of

firm-level variables that are likely to be correlated with product quality, such as investment intensity,

R&D expenditure, and product and process innovations.

To rationalize our empirical exercise we rely on the analytical model of Crinò and Epifani (2010),

that extend an heterogeneous-firms model a la Meltiz (2003), by incorporating heterogeneity in

products quality and non-homothetic preferences in consumption. Under this setting, they are able to

derive clear and testable predictions about the relationship between productivity, quality and export

performance across destinations. Specifically, they show that the relation between productivity and

export intensity to low (high) income destinations, should be negative (positive), and largely driven by

across firms differences in product quality.

We investigated these predictions on a sample of Italian food firms involved in the international

trade. Specifically, the logic of the study moves in two steps. First, after computing different measures

of firm-level revenue-TFP, we test if there exists a robust relationship between the export intensity to

specific market destinations and firm productivity. Recent trade models and empirical evidences based

on firm-level data have indeed showed that the link between TFP and firm export behaviour is

conditional on the characteristics of destination countries, especially in terms of income and

geographic distance (see Crozet, Head and Mayer, 2009; Manova and Zhang, 2009; Crinò and Epifani,

2010, among others). Then, in a second step, after having established a clear link between TFP and our

1 For example, in presence of non tariff barriers to trade (NTBs), if the exporting firm has market power, it may be able to extract the quota revenue for itself by raising the import price. Knetter (1997) argues that this is the case for exports to Japan. He finds that German exporters charge substantially higher prices when exporting to Japan than to the United States, United Kingdom and Canada, and argues that this is the result of a variety of NTB’s in Japan. Note that this potential issue is particular severe for food products, as a consequence of the large diffusion of (different) public and private standards across countries in the food sectors.

3

firm-level proxies for product quality, we test if the relationship between TFP and export intensity to

specific destinations, is indeed driven by the quality nature of exported products.

Our paper is related to a growing literature that has documented systematic variation in export

performance across firms. Key stylized facts from this literature showed that more productive firms

are more likely to export, have higher export revenues, and serve more and distant markets (Clerides,

Lach and Tybout, 1998; Eaton, Kortum and Kramarz, 2004; Bernard, Jensen and Schott, 2009). These

and other patterns are consistent with firm-heterogeneity models that focus on firm productivity as the

main determinants of export performances. In this setting, more productive firms have better export

success because they have lower marginal costs and charge lower price (e.g. Melitz, 2003; Melitz and

Ottaviano, 2008). However, other evidences have also documented several stylized facts that are at

odd with the last mechanism and interpretation. For example, larger exporters are more skilled

intensive, pay higher wages and input prices, as well as they charge higher prices (see, e.g.,Verhoogen,

2008; Hallack and Sivadasan, 2008). Thus, recent models with the aim to reconcile these apparent

contradictory facts have extended the seminal productivity-heterogeneity framework by incorporating

heterogeneous quality across firms. In this setting, more efficient firms have higher export

performances as they use more expensive and quality inputs to sell higher-quality goods at higher

prices (see, e.g., Baldwin and Harrigan, 2007; Verhoogen, 2008; Manova and Zhang, 2009).

Among this literature, two recent papers are particular relevant for our analysis, sharing a new

strategy to capture direct information about quality and/or product differentiation. Crozet, Head and

Mayer (2009), studied the effect of quality differences on heterogeneous export success matching

exporting firms of Champagne to produce ratings from two wine guides, to solve for the lack of direct

measures of quality. They showed that high quality producers export to more markets, charge higher

prices, and sell more items to each market, all evidences in line with a quality sorting model. On the

other hand, Altomonte, Colantone and Pennings (2010), extended the Melitz and Ottaviano’ (2008)

model by introducing different market segments in the demand system. In this setting, they derived

more complex relations between export status, productivity, size and mark-ups, which ultimately

depend on the degree of product differentiation. The predictions are tested at the empirical level by

comparing the performance of French wine producers, exploiting the product differentiation in the

"Denomination of Con-trolled Origin" areas.

Overall, our analysis using a different approach to extract ‘direct’ information about product

quality, reach similar conclusions showing that product quality matters a lot to explain the export

behaviour of Italian food firms, and not surprising this is especially true for firms that produce

typically ‘Made in Italy’ foods.

The organization of the paper is as follow. Section 2, summarize the heterogeneous-firm model of

Crinò and Epifani (2010) with the aim to rationalize our empirical strategy. Section 3, presents the

data, our measures of revenue-TFP, and the approach to study the relationship between productivity,

4

product quality and export performance. In Section 4 econometric results will be presented and

discussed. Finally, in section 5 we will derive some concluding comments.

2. Theoretical framework To rationalize our empirical work this Section summarize the main results of Crinò and Epifani (2010)

theoretical model. Specifically, adding in a simple firm heterogeneity model non-homothetic in

consumer preferences, they show that there exists a negative relationship between export intensity to

low income destination, firm productivity and the quality of the exported goods.

Consider a representative consumer characterized by the following utility function:

[ ] ρρρ 11 )()(∫−

∈= dvvcvqU Vv , 0<ρ<1, (1)

where V is a continuous set of varieties available for consumption, indexed by v and represents a

Cobb-Douglas bundle of physical quantity; c(v) is consumption and q(v) is quality of variety v, as

perceived by the representative consumer.

Maximizing the consumer’s utility (1) subjected to the usual budget constraint, ∫∈

=Vv

dvvcvpy )()(

with y the exogenously given per capita income, the demand for v can be written as

σσ −−= 1)()()( PRvpvqvc , where R is total income, p(v) is the price of variety v, 1)1( 1 >−= −ρσ

is the constant elasticity of substitution among varieties, and P is the ideal price index.

The first key assumption of the model is about the preferences of a representative consumer.

Unlike Melitz (2003) seminal firm heterogeneity model, where the preferences are homothetic, in this

model the preferences for quality by the representative consumer are non-homothetic with respect to

per capita income (y). Assume that )()()( yvvq αλ= , where 1)( ≥vλ denotes true product quality and

0)( >yα captures the elasticity of demand with respect to product quality. The relative demand for

higher-quality products is higher in high-income countries, if and only if, the following relation holds:

)()( yy ′′>′ αα for yy ′′>′ .

Consider now a partial equilibrium model of one sector economy open to international trade,

where firms produce differentiated products under monopolistic competition and are heterogeneous in

productivity and quality. Under this setting, it is possible to study the relationship between firm

revenue and product quality with respect to the per capita income. Let d a domestic market and f a

foreign market. Consider therefore a market { }fdz ,∈ , where θ measures firm productivity and θ/1

is the marginal cost to produce v. In this first part of the model product quality is exogenous, but this

assumption will be relaxed later.

The profit maximizing price is ρθτ zzp = , where 11 −= σσρ is a constant price-marginal

cost mark-up, and 1>zτ is an iceberg trade cost. Using the expression for zq , zp and for

5

consumer’s demand for variety, c(v), it is possible to yield the firms’ revenue in market z as a function

of productivity and product quality:

)(

1

1),( zy

z

zzz

PRr α

σσ λ

τρθθλ

−−

= , { }fdz ,∈ (2)

which imply that the elasticity of firm revenue to product quality is increasing in per capita income of

destination z, From (2) it is possible to study the ratio of exports to destination f over domestic sales:

)()(

1

1

)/(

)/(df yy

ddd

fff

d

f

PR

PR

r

r αασ

σ

λττ −

−

−

= which implies → )()(ln

)ln(df

df yyd

rrdαα

λ−= (3)

Relation (3) shows that the elasticity of the ratio df rr to product quality, for df yy > , is increasing

in per capita income of the foreign destination.

Consider now the export intensity of two foreign destinations indexed by { }hlf ,∈ , with

differences in the per capita income, with hdl yyy << . The export intensity to the lower income

destination can be written as: dhdl

dl

hld

ll rrrr

rr

rrr

rEXP

++=

++≡

1 . The assumption of non-

homothetic preferences will affect lEXP, because a rise of product quality reduces dl rr and

increases dh rr , causing the reduction of lEXP. Thus, using (3) it is possible to study the elasticity of

export intensity to low-income destination respect to product quality:

[ ]( ) [ ] 0)()(1)()(ln

ln <−−−−−= hdhlldl EXPyyEXPyy

d

EXPd ααααλ .

The relation above shows the existence of a negative correlation between export intensity to low

income destination, lEXP, and the quality of the exported products, λ . Moreover, the model tend to

also predict a positive relationship between product quality and the export share to higher-income

destinations, )/( hlhh rrrEXS +≡ , as well as an ambiguous effect of product quality on the overall

export intensity, )/( hldlh rrrrrEXP +++≡ .

Next, after have studied the relationship between export intensity and product quality, it will be

analyzed the implications of the second key assumption of the model, namely that there exists a

positive relationship between products quality and fixed cost. In particular, Crinò and Epifani (2010),

assume that higher quality products require higher fixed costs, due to the idea that quality upgrading is

linked to more intensive products’ development activities that require higher fixed costs, such as for

R&D and marketing activities. To do this we study the relationship between endogenous product

quality and technical efficiency, the latter captured by revenue-TFP.

6

The model assumes that firms produce a variety of qualityλ paying a fixed cost ( ) ηλη1 , where η >

0 is the elasticity of the fixed costs to product quality. An important assumption is that firms produce

goods with different quality depending on the destination market, therefore firms choose the quality of

their product based on the characteristics of each market.

As said before, technical efficiency, and therefore fixed costs, are captured by revenue-TFP,

thereby the following formulation allows us to investigate the relationship between product quality

and fixed costs, simply as the elasticity of product quality to productivity. The following expression

shows how it is possible to choose the optimal product quality for a destination market z:

−−−

zy

zzM φλ

ηλθ ηασ

λ

1max )(1

, { }lhdz ,,∈ (4)

where

11

−

=σ

ρτσ z

zzz

PRM represents a measure of market size, and zφ is a fixed cost of entry into

the destination market z. By solving this problem, the optimal product quality for market z,*zλ , will be:

[ ] )(

11* )( zy

zzz My αησθαλ −−= (5)

where 0)( >− zyαη , by the second order condition for a maximum. Relation (5) says that more

productive firms produce higher-quality products for all market destinations. This is possible because

they get greater revenue from selling high-quality products in these markets, that allows them to

spread the higher fixed costs paid for upgrading products' quality over a greater revenue. Using the

expression for optimal product quality (5) into )(*1 zy

zzz Mr ασ λθσ −= , it is possible to yield the ratio of

export to domestic sales:

[ ][ ] )(

)(1

)(

)(1

)(*

)(*

)(

)(

d

d

f

f

d

f

y

y

dd

y

y

ff

d

f

ydd

yff

d

f

My

My

M

M

M

M

r

r

αηα

σ

αηα

σ

α

α

θα

θαλλ

−−

−−

== (6)

Finally, it is possible to study the elasticity of the ratio df rr to productivity, using the log of (6) and

differentiating, yields:

( )

−−

−−=

)(

)(

)(

)(1

ln

)ln(

d

d

f

fdf

y

y

y

y

d

rrd

αηα

αηα

σθ (7)

Relation (7) implies that, conditional on exporting destination f, the export intensity to low-income

destination is inversely related to productivity, 0ln

ln<

θd

EXPd l. As seen before for product quality,

from (7) emerges that the elasticity of export intensity to productivity is increasing in per capita

7

income of the foreign destination. The intuition is that high-productivity firms produce higher-quality

goods, for which relative demand is lower in low-income destinations.2

3. Empirical strategy The model summarized above predicts two main testable hypotheses. First, conditional on export

destinations, it suggests the existence of a negative correlation between firm level TFP and export

intensity to low income destinations. Second, and most importantly, as a consequence of a positive

link between product quality and production cost (as well as revenue-TFP), it suggests that the link

between TFP and export intensity should be due to product quality.

To test these predictions we need firm-level data with information on firm export behaviour and

proxies for product quality. The next sections will introduce the dataset and our strategy to estimate

TFP and the main model propositions.

3.1 Data description

In this paper we make use of a (unbalanced) panel of Italian food firms drawn from the 9th and 10th

Survey on Manufacturing Firms (Indagine sulle Imprese Manifatturiere) carried out by Unicredit-

Capitalia. The sample contains three-years firm level data on roughly 750 food firms with more than

10 employees in the period 2001-2006. The panel is stratified and rotating, so there is an overlapping

of just 40 firms between the two surveys. The sample stratification is based on the 3-digit ISTAT

ATECO 91 nomenclature (equivalent to NACE), size class and geographic area, therefore it is

representative of the population of Italian food industry. Firm level data are drawn from a very

detailed questionnaire submitted to each firm, that allows us to have information about their business,

employment, R&D activity, internationalization and management. According to standard cleaning

procedures, it was dropped firms that presents negative values for sales, material purchases, labour

cost and capital stock.

In order to calculate firm productivity with a Cobb-Douglas production function, we use a

revenue-based measure of output, that equals the values of shipments plus changes in stock of finished

goods and capitalised costs, deflated with the corresponding three-digit producer price index. As input

we use the labour cost deflated with a wage index, the book value of capital deflated with a common

price index for investment goods, and materials, defined as the difference between purchases and

change in inventories of intermediate goods, deflated with a common price deflator for intermediate

2 Crinò and Epifani (2010) highly also that, although revenue-TFP is closely related to product quality and productivity, it may also capture variation across firms in markups, which in this model are instead constant. Although markups may reflect pure demand shocks and pricing power, they are likely to be positively correlated with productivity and product quality, which may strengthen the positive correlation of revenue-TFP with both our key parameters.

8

inputs. As shown in table 1, where we report descriptive statistics on firms’ inputs and output, the

median firm in the sample is characterized as follows: produces roughly 10 million Euros worth of

output, employees about 30 workers, with a labour productivity (value added per worker) equal to 54

thousand Euros.

Our data contains also detailed information on firm internationalization for the year 2003 in the

first survey and for the year 2006 in the second survey. We use data on firms’ exports in order to study

the relation between export behaviour and firm productivity. Following Crinò and Epifani (2010) we

divide firms’ export into two groups, depending on the per capita income of destination market: high

income and low income destination. The former group includes firms’ exports to EU15, North

America and Oceania, the latter group includes export to Latin America, Africa, New EU member and

China.

As shown in table 1, roughly 60% of the firms in the sample are exporters. Considering the

exporter firms in our sample, it is possible to see that Italian food firms export mostly to high income

destination, roughly 90%, while exporting firms to low income destination are almost 30%. Similarly,

considering export intensity in the two groups (the ratio of exports to these areas over total sales), it is

possible to see that export intensity to high income destinations is higher than export intensity to low

income destinations.

3.2 TFP estimation

The first step of our econometric strategy is the estimation of the total factor productivity (henceforth

TFP). We estimate revenue-based measure of TFP because it captures not only technical efficiency,

but also product quality. In order to calculate TFP we start from a standard Cobb-Douglas production

function:

mllititititit MKLAY βββ= (8)

Where itY is revenue-based output of firm i in period t, itL , itK and itM are, respectively, labour,

capital and materials inputs, lβ , kβ and mβ the input coefficients, anditA is the Total factor

productivity. While itL , itK and itM are all observable by the econometrician, itA is unobservable to

the researcher.

Considering the log-linearization of (8) yields:

ititmitkitlit mkly ηββββ ++++= 0 (9)

where:

ititA ηβ += 0ln . (10)

9

In the relation (10), 0β represents a measure of the mean efficiency level across firms and over time

and itη is the time- and producer-specific deviation from that mean.

In order to calculate TFP, our variable of interest in (9) is the error term, itη . Note that, to get a

consistent OLS estimator of the production function, therefore extracting TFP as the residual, itη must

be uncorrelated with the input variables. However, using OLS to estimate our production function, itη

results correlate with the input variables, generating the well known simultaneity problems. Following

Griliches and Mareisse (1995), it is possible to explain this problem considering that profit-

maximizing firms immediately adjust their inputs each time they observe a productivity shock,

consequently input levels will be correlated with the same shocks. As we said before, while firm

productivity shocks are normally observable and observed by firms, they are unobservable by the

econometrician. Because of this, productivity shocks enter in the error term of the regression, hence

inputs turn out to be correlated with the error term, causing a bias OLS estimation of the productivity

function.

Olley and Pakes (OP, 1996) and Levinsohn and Petrin (LP, 2003) have proposed two similar

methods to solve this problem, based on a semi-parametric estimation in which the error term itη can

be decomposed into two parts. The equations (9) becomes:

itititmitkitlit mkly εϖββββ +++++= 0 . (11)

Therefore the error term in (11) has two component: itϖ , that represents the transmitted productivity

component and itε , an error term that is uncorrelated with input choices. The key difference between

the two components is that itϖ is a state variable that impacts the productivity shocks and it is

observed by the firm but not by the econometrician. Hence OP and LP propose an estimation method

to make observable the productivity shocks, finding an observable proxy for the productivity term

itϖ . In particular, the OP methodology uses investment as proxy, while the LP methodology uses

material costs.

OP and LP assume that, respectively, investment demand function and materials demand function,

depend on the firm’s state variables itk and itϖ . Assuming that these demand functions are

monotonically increasing in TFP, it is possible to invert them to express TFP in terms of observables.

Solving (11) for itϖ , productivity can be calculated as follows:

itmitlitkitit mlky βββϖ ˆˆˆˆ −−−= (12)

where itϖ̂ is the (log of) TFP.

10

Table 2 shows the estimated coefficients of our production function using the different techniques

previously described. In particular, all output elasticities are positive and, excluding the capital

coefficient in the LP procedure, precisely estimated. For each specification, the table also reports

estimated returns to scale. As can be see, all estimates are close to constant returns to scale.

Using the three estimation methods, we plot in Figure 1 the evolution of aggregate TFP indices,

computed as the ratio between the yearly un-weighted average of the firm level TFP and its initial

(2001) value. Results point to a high correlation of the three estimates and a declining trend for our

sample of firms from 2001 to 2003, followed by an increasing trend from 2004 to 2006. Figure 2,

shows the evolution of the TFP index according to its industrial dimension, using OP estimates as our

baseline. Across food industries the patterns are quite heterogeneous. Indeed, we find sectors with a

declining TFP level in the observed period, such as Conserved and preserved meat (15130), Fish

preparation (15200), Oil and fat (15400) and Processed grains (15600); sectors with an increasing

TFP level, such as Conserved fruit&vegetables (15300) and Cheese (15500), and, finally, sectors that

do not display any relevant trend in the observed period, such as Meat (15100), Beverage (15900) and

Wine (15930).

3.3 Relationship between productivity, quality and export destinations

With our firm level TFP estimates in hand, we can now study the relationship between firm

productivity, quality and export destinations. First, using the information about firms’ export to

different destinations, we build an index of the export intensity to low income destination, measured as

the ratio of exports to these areas over total sales, henceforth lEXP. The first key proposition of the

model suggests the existence of a negative relationship between TFP and export intensity to low

income destinations. We test this prediction by running the following cross-sectional OLS regression:

jijlj uTFPEXP +++= ηαα lnln 10 (13)

where lEXP represents export intensity to low income destination for a firm j, TFP is our measures of

total factor productivity, iη are 3-digit industry fixed effects, and u is an error term.

The second proposition of the model asserts that the key channel through which there exists a

correlation between TFP and export intensity, conditional on export destinations, is product quality.

To test this hypothesis we follow two different strategies.

The first strategy is indirect, and exploit the specific nature of Italian food production. Indeed,

several Italian food sectors are characterized by an high quality tradition and image. Thus, in line with

the model predictions, we expect that firms producing typical ‘Made in Italy’ products, should display

a greater (absolute) TFP elasticity to export destinations. If this is the case then, we have a first

roughly support to the idea that the relationship between TFP and the export intensity across

destinations, is indeed due to products quality. To implement this strategy we generate a dummy

11

variable called dMIT, equals to 1 if a firms belongs to one of the 4-digit sectors of Made in Italy (zero

otherwise).3 Hence, in order to test this hypothesis we simple add in (13) an interaction term between

the Made in Italy dummy and TFP.

The second empirical strategy follows the approach proposed by Crinò and Epifani (2010), who

tests the main model’s implications, by estimating directly the relationship between product quality

and firm productivity, generating variable proxies for product quality. To do this, we first explore the

richness of the dataset, with the aim to find variables linked to product quality. According to the

empirical literature (e.g., Sutton 1998), product quality differentiation is generally associated with

some firm features, such as an intensive investment activities, an innovative behaviour and an

intensive R&D and marketing activities. We choose the following variables to proxy for some of these

firms features linked to product quality differentiation: total investment expenditure, sales of

innovative products, ICT investments, a dummy variable for R&D investments, average wage, number

of employees and a dummy variable for product innovation.

Next we first regress each of these proxies on our TFPs controlling for 3-digit industry dummies,

to test if there exists the expected positive relationship between them. Then, in order to generate a

synthetic variable that proxies for firm quality, we extract their principal component by factor analysis.

The idea is that the principal component may capture the common link of these variables with product

quality. Using this strategy, we generate 3 alternative proxies for quality. The first one, AQ , is the

principal component of the following variables: total investment expenditure, sale of innovative

products, ICT investments, a dummy variable for R&D investments and a dummy variable for product

innovation. The second proxy for quality, BQ , is generated adding a variable proxy for firm size, the

number of employees, to the variables used to generate AQ . Finally, the third proxy for quality, CQ ,

is generated adding also a variable proxy for input quality, the firm average wage, to the variables,

used for BQ .Then, once again, we regress our three proxies for quality on TFP and other control to

study the relationship between TFP and product quality.

Finally we test the main proposition of the model replacing in (13) our three proxies for quality in

place of TFP, in order to verify whether also in this case there exists a negative relationship between

export intensity to low income destination and product quality.

4. Results Table 3 reports regression results of the relationship between export intensity and TFP. As it is clear

from the figures, the results strongly confirm that the TFP elasticity of export intensity to low income

destination is negative, large in magnitude and significantly different from zero at 5% level. This

3 According to the INEA classification, food sectors typical of Made in Italy in the 4-digit ATECO 91 nomenclature are: 15130, 15300, 15411, 15512, 15520, 15610, 15620, 15810, 15811, 15812, 15820, 15840, 15850, 15930

12

result holds irrespective to the different TFP estimation method used. Thus, as suggested by the model,

the TFP elasticity is increasing in the per capita income of export destinations. The estimated elasticity

is also large in magnitude, implying that a 100% increase in TFP is associated with about 55% fall in

the export intensity to low income destinations, an effect very close with the finding obtained for the

overall manufacture sector from Crinò and Epifani (2010).

Regressions from 4 to 6, test the proposition that the TFP elasticity to export intensity is higher in

absolute value for firms producing typical Made in Italy products. As explained before, this is done by

adding to the specification the interaction between the Made in Italy dummy (dMIT) and TFP. In line

with our expectation, the estimated TFP elasticity to export intensity for firms producing Made in Italy

is higher in (absolute) magnitude (0.80 vs. 0.55). This evidence gives a preliminary confirmation to

the idea that product quality may effectively represent a first order explanation for the observed link

between productivity and export behaviour.

In order to test the relationships between TFP and product quality directly, Table 4 reports

regression results of each of the selected quality proxy on the TFP, controlling for 3-digit industry and

time dummies. the results shows that the TFP is indeed positively related with all these variables.4

Next, as describe in Section 3, from these quality proxies we have construct three synthetic

product quality variables, by extracting their principal component through factor analysis. Then, we

regressed each of them on TFP, controlling for industry and time dummies. The finding are reported in

Table 5, and confirm the expected positive and significant relation between TFP and product quality.

Moreover, once again, as it is clear from results reported in column from 4 to 6, food firms belonging

to sectors classified as typical of Made in Italy, present a stronger positive link between product

quality and TFP. Specifically, Made in Italy firms display an estimated coefficient on TFP that is from

2 to 3 time larger than the average firms, giving a further confirmation of the link between TFP and

product quality. The last evidence, indirectly, corroborate our simple approach of testing the main

model prediction through the ‘Made in Italy’ dummy.

As a final step we test the crucial implication of the model, by replacing in the relation (13), our

three proxies for product quality in place of TFP, in order to verify if the negative relation between

TFP and export intensity to low income destination is indeed driven by firm heterogeneity in product

quality. Table 6 shows the results. For all our quality proxies the estimated coefficient is as expected

negative, although statistical significant only for the quality proxies that incorporate firm size (QB) and

input quality (QC). Moreover, and interesting, if we add the TFP to the regressors (see columns 4-6),

first the main results about quality effect are basically unchanged and, second, the TFP coefficient

although as expected negative is no longer a significant determinant of export intensity. Thus, this

4 Note that, unlike in (13) for EXPl,the three proxy for quality enter in level rather than in logs, because they are standardize variables with a mean zero and standard deviation 1. Hence, we can not interpret the coefficients as elasticity.

13

evidence gives a broadly confirmation to the model predictions, supporting the idea that the

correlation between TFP and firm-level export intensity is largely driven by product quality.

Finally, we have also studied the relationship between TFP, quality and firms’ export behaviour in

term of number of export destinations. Indeed, several previous firm-level evidence have shown that

more productive firms export to an higher number of destinations (Crozet et al. 2009; Bernard et al.,

2007; Crinò and Epifani, 2010). Thus, a natural extension consistent with the model discussed above is

that the positive relation between TFP and the number of export destinations could be once again

mediated by product quality. Indeed, this appear fully consistent with the idea that these firms can

spread the higher fixed costs of quality upgrading over a larger output and across an higher number of

foreign markets.

Table 7 reports the results of regressing the number of served markets on TFP and the usual 3-

digit industry dummies. The evidence shows a positive and robust correlation across all the different

TFP estimation methods considered. Moreover, when we add the interaction between the Made in

Italy dummy and TFP to the specification, we find a preliminary evidence that quality matter. Indeed,

firms exporting Made in Italy products displayed a magnitude of the estimated effect that is about two

times larger than those of average firm.

The idea that firms selling to a large number of destinations produce higher-quality products, is

directly tested in Table 8, by regressing our three proxies for quality on the number of served markets,

plus 3-digit industry dummies. In line with the previous evidence we find that firms producing higher

quality products also serve more export markets, an effect that remain strong when we add to the

specification the TFP, that still exert a significant effect on quality.

5. Conclusions

This paper examines the link between total factor productivity (TFP), products quality, and export

across destinations, to analyse the relative importance of production efficiency versus product quality

for firms’ export success. To this end we exploit the export behaviour of a sample of 750 Italian food

firms, observed during the 2001-2006 period, finding several interesting stylized facts.

Specifically, using different measures of revenue-TFP and several proxies for product quality we

find four main results. First, there exists a robust negative relationship between revenue-TFP and

export intensity to low income destinations. Second, this relation seems largely driven by our proxies

for product quality, ceteris paribus. Third we find a robust positive relationship between the number

of export destinations, TFP, and product quality. Finally, all the above results are stronger when

applied to a sub-sample of firms producing typically ‘Made in Italy’ products. All these facts are

consistent with a model based on firm-heterogeneity in product quality and cross-country

heterogeneity in consumption quality.

Understanding the determinants of firms’ export success and behaviour is important for their

implications on international trade patterns, and the welfare effects of globalization. Moreover, a

14

deeper comprehension of the factors that drives firms export performance can facilitate the design of

policies that promote trade. From this point of view, our results tend to highlight what should be a

government priority to encourage investment in R&D and technologies policy that allows firms to

produce and export higher quality products.

15

References Altomonte, C., Colantone, I. and E. Pennings (2010), International Trade with Heterogeneous Firms

and Asymmetric Product Varieties, Catholic University of Leuven (KUL) MSI Working Paper No. 1005

Baldwin, R. and J. Harrigan (2007), Zeros, Quality and Space: Trade Theory and Trade Evidence, NBER Working Paper No. 13214.

Bernard, A., Jensen, J. and P. Schott (2009), Importers, Exporters, and Multinationals: A Portrait of Firms in the U.S. that Trade Goods, in T. Dunne, J.B. Jensen, and M.J. Roberts, eds., Producer Dynamics: New Evidence from Micro Data, University of Chicago Press.

Bernard, A.B., Jensen, J.B., Redding, S.J and P.K Schott. (2007), Firms in International Trade, Journal of Economic Perspectives, vol. 21 (3): 105-130.

Clerides, S., Lach, S. and J. Tybout (1998), Is Learning by Exporting Important? Micro-dynamic Evidence from Colombia, Mexico, and Morocco. Quarterly Journal of Economics 113 (3): 903-47.

Crinò, R. and P. Epifani (2010), Productivity, Quality, and Export Behavior across Destinations, Development Working Paper No. 271, Centro Studi Luca d'Agliano, University of Milano.

Crozet, M., Head, K. and T. Mayer (2009), Quality Sorting and Trade: Firm-level Evidence for French Wine, University of British Columbia mimeo.

Eaton, J., Kortum, S. and F. Kramarz (2004), Dissecting Trade: Firms, Industries, and Export Destinations, American Economic Review Papers and Proceedings 94(2):150-54.

Fischer, C. (2007), Food quality and product export-performance – an empirical investigation of the EU situation, Journal of International Food & Agribusiness Marketing, Forthcoming, 2010.

Griliches, Z. and J. Mairesse (1995), Production Functions: the Search for Identification, NBER Working Paper No. 5067, Washington, D.C.

Hallak, J.C. and J. Sivadasan (2008), Productivity, Quality and Exporting Behaviour under Minimum Quality Requirements, University of San Andres mimeo.

Knetter, M.M. (1997), The Segmentation of International Markets: Evidence from The Economist, NBER Working Papers No. 5878.

Levinsohn, J. and A. Petrin (2003), Estimating Production Functions Using Input to Control for Unobservables, The Review of Economic Studies, 70 (2): 317-342.

Manova, K. and Z.Zhang (2009), Quality Heterogeneity across Firms and Export Destinations, NBER Working Papers No. 15342.

Melitz, M.J. (2003), The impact of trade on intra-industry reallocations and aggregate industry productivity, Econometrica, vol. 71 (6): 1695-1725.

Melitz, M. and G. Ottaviano (2008), Market Size, Trade, and Productivity, Review of Economic Studies 75 (1): 295-316.

Ninni, A., Raimondi, M. and M. Zuppiroli (2006), The success of “Made in Italy”: an appraisal of quality-based competitiveness in food markets, Economics Department Working Papers 2006-EP10, Department of Economics, Parma University.

Olley, S. and A. Pakes (1996), The dynamics of productivity in the telecommunications equipment industry, Econometrica, 64(6): 1263-98.

Petrin, A., Poi, B. P. and J. Levinsohn, 2004, Production Function Estimation in Stata Using Inputs to Control for Unobservables, The Stata Journal, 4(2): 113-123.

Scoppola, M. (2003). Il commercio internazionale dei prodotti agro-alimentari: la posizione e le prospettive dell’Italia in una Europa allargata, Atti del XL Convegno della SIDEA, Padova.

16

Sutton, J. (1998), Technology and Market Structure: Theory and History, MIT Press: Cambridge, Massachusetts.

Verhoogen, E. (2008), Trade, Quality Upgrading and Wage Inequality in the Mexican Manufacturing Sector, Quarterly Journal of Economics 123 (2): 489-530.

Yasar, M., Raciborski, R. and B. Poi (2008), Production Function Estimation in Stata Using the Olley and Pakes method, The Stata Journal, 8(2): 221-231.

17

.96

.98

11.

021.

04

2001 2002 2003 2004 2005 2006years

ln_tfp_OP_index ln_tfp_LP_indexln_tfp_OLS_index

Figure 1. Evolution of average TFP in the food industry under different estimation methods.

.8.9

11.

11.

2.8

.91

1.1

1.2

.8.9

11.

11.

2

2000 2002 2004 2006 2000 2002 2004 2006 2000 2002 2004 2006 2000 2002 2004 2006

Food (15000) Meat (15100) Conserved Meat (15130) Fish (15200)

Fruit&vegetables (15300) Oil and fat (15400) Cheese (15500) Processed grains (15600)

Feed (15700) Other food (15800) Beverage (15900) Wine (15930)avg_

ln_t

fp_O

P_i

ndex

yearsGraphs by ateco3

Figure 2. Evolution of average TFP (Olley and Pakes estimates) in different food sectors.

18

Table 1. Sample description

Mean Median Std. Dev. Observations

Output (€, '000) 28010 9868 68265 756VA per worker (€, '000) 95 54 486 767Labor cost per worker (€, '000) 48 31 201 769Materials per worker (€, '000) 355 196 1043 769Capital stock per worker (€, '000) 137 73 374 769Number of employees 77 30 220 782

Exporters

Mean Median Std. Dev. # (%) of firmsAll destinations 27.7 19.0 23.7 471 (60.2%)High income destination 23.1 14.3 23.6 438 (56.0%)Low income destination 8.0 4.0 10.7 144 (18.4%)

Export intensity (%)

Notes: Variables definition: Output equals values of shipments plus changes in stock of finished goods and capitalised cost. Materials are the difference between purchases and change in inventories of intermediate goods. Capital stock is the book value of capital. Export intensity is the ratio of exports to total sales. High income destinations include EU15, North America and Oceania. Low income destinations include New EU members, China, Africa and Latin America. All variables are taken by Unicredit-Capitalia surveys and are computed for the year 2003 for the IX survey and for the year 2006 for the X survey. (See text).

Table 2. TFP estimation, Olley and Pakes, Levinsohn and Petrin and OLS methods

Dependent variable: Log of Output

Olley and Pakes Levinsohn and Petrin OLSParameter (1) (2) (3)

ln labor 0.331*** 0.337*** 0.354***(0.012) (0.023) (0.010)

ln capital 0.051*** 0.040 0.040***(0.016) (0.045) (0.008)

ln material costs 0.611*** 0.610*** 0.603***(0.009) (0.214) (0.008)

Return to scale 0.99 0.99 1.00

Observations 2275 1737 2275

TFP method

Notes: In columns (1) and (2) standard error based on 100 bootstrap replications in round brackets; In columns (3) robust standard errors in round brackets; ***,** ,* = significant at 1, 5 and 10 percent level, respectively.

19

Table 3. Export intensity to low income destinations and TFP

TFP method Olley and Pakes Levinsohn and Petrin OLS Olley and Pakes Levinsohn and Petrin OLS

(1) (2) (3) (4) (5) (6)

ln TFP -0.588*** -0.566*** -0.510** -0.531*** -0.430* -0.491**

(0.173) (0.196) (0.239) (0.160) (0.285) (0.233)

ln TFP*dummy Made in Italy -0.792*** -0.857*** -0.748***

(0.154) (0.155) (0.126)

R-squared 0.15 0.15 0.14 0.29 0.27 0.27

Observations 135 135 135 135 135 135

Dependent variable: Export intensity to low-income destinations

Baseline Disentangly 'Made in Italy'

Notes: OLS regressions with robust standard errors in round brackets . ***,** ,* = significant at 1, 5 and 10 percent level, respectively. Each column in the table refers to a different TFP estimate. All specifications include a full set of industry dummies, defined at the 3-digit level classification. (See text).

20

Table 4. Quality related variables and TFP (panel regressions)

Dependent variableInvestment expenditure

Dummy for product innovation

ICT investmentsSales of innovative

product(1) (2) (3) (4)

ln TFP 0.032*** 0.193*** 0.039*** 0.102*

(0.010) (0.075) (0.010) (0.06)

R-squared 0.04 0.03 0.02 0.02

Observations 1636 2221 1863 1767

Dependent variable

Dummy for R&D investments

Sales of innovative product x dummy for process innovation

Number of employees

Average Wage

(5) (6) (7) (8)

ln TFP 0.089*** 0.018* 0.026** 0.050*

(0.024) (0,010) (0,013) (0,030)

R-squared 0.03 0.02 0.02 0.02

Observations 2235 1767 2224 2219

Notes: OLS regressions with robust standard errors in round brackets. ***,** ,* = significant at 1, 5 and 10 percent level, respectively. All variables are standardized with mean 0 and variance 1. TFP is based on the Olley - Pakes estimates. All specifications include a full set of industry dummies, defined at the 3-digit level classification and time dummies. (See text).

Table 5. Product quality and TFP (panel regressions)

Dependent variable Quality A Quality B Quality C Quality A Quality B Quality C

(1) (2) (3) (4) (5) (6)

ln TFP 0.043** 0.033** 0.033** 0.032* 0.022* 0.022*

(0.018) (0.016) (0.016) (0.017) (0.015) (0.015)

ln TFP*dummy Made in Italy 0.092* 0.091** 0.093**

(0.054) (0.047) (0.046)

R-squared 0.05 0.05 0.05 0.05 0.05 0.05

Observation 1421 1416 1415 1421 1416 1415

Baseline Disentangly for 'Made in Italy'

Note: OLS regressions with robust standard errors in round brackets. ***,** ,* = significant at 1, 5 and 10 percent level, respectively. Quality A, B and C represent proxy for product quality and are obtained through factor analysis, by extracting the principal components of the quality proxy variables of table 4. All variables are standardized with mean 0 and variance 1. TFP is based on the Olley - Pakes estimates. All specifications include a full set of industry dummies, defined at the 3-digit level classification and time dummies.

21

Table 6. Export intensity and product quality Dependent variable: Export intensity to low-income destinations

Quality A Quality B Quality C Quality A Quality B Quality C

(1) (2) (3) (4) (5) (6)

Proxy for quality -0.172 -0.224* -0.223* -0.157 -0.215* -0.215*

(0.127) (0.125) (0.126) (0.131) (0.129) (0.130)

ln TFP -0.664 -0.618 -0.680

(0.576) (0.516) (0.578)

R-squared 0.18 0.19 0.19 0.18 0.19 0.19

Observation 88 88 88 80 80 80

Baseline Controlling for TFP

Notes: OLS regressions with robust standard errors in round brackets. ***,** ,* = significant at 1, 5 and 10 percent level, respectively. All variables are standardized with mean 0 and variance 1. All specifications include a full set of industry dummies, defined at the 3-digit level classification. (See text).

22

Table 7. Number of export destinations and TFP

TFP method Olley and Pakes Levinsohn and Petrin OLS Olley and Pakes Levinsohn and Petrin OLS

(1) (2) (3) (4) (5) (6)

ln TFP 0.136** 0.135** 0.145** 0.107* 0.097 0.099

(0,057) (0,070) (0,068) (0.057) (0.062) (0.066)

ln TFP*dummy Made in Italy 0.173*** 0.234*** 0.185***

(0.041) (0.033) (0.066)

R-squared 0.16 0.15 0.15 0.17 0.17 0.17

Observations 441 441 441 441 441 441

Dependent variable: Number of export destinations (from 1 to 8)

Baseline Disentangly for 'Made in Italy'

Notes: OLS regressions with robust standard errors in round brackets . ***,** ,* = significant at 1, 5 and 10 percent level, respectively. Each column in the table refers to a different TFP estimate. All specifications include a full set of industry dummies, defined at the 3-digit level classification. (See text).

23

Table 8. Product quality and export destinations

Dependent variable Quality A Quality B Quality C Quality A Quality B Quality C

(1) (2) (3) (4) (5) (6)

Number of export destinations 0.150** 0.152** 0.151** 0.147** 0.149** 0.149**

(0.060) (0.059) (0.059) (0.058) (0.060) (0.060)

ln TFP 0,059*** 0.035** 0.035**

(0.020) (0.017) (0.017)

R-squared 0.06 0.07 0.07 0.06 0.06 0.06

Observation 302 302 302 295 295 295

Baseline Controlling for TFP

Notes: OLS regressions with robust standard errors in round brackets. ***,** ,* = significant at 1, 5 and 10 percent level, respectively. Quality A, B and C represent proxy for product quality and are obtained through factor analysis, by extracting the principal components of the variables on the table 4. All variables are standardized with mean 0 and variance 1. TFP is based on the Olley - Pakes estimates. All specifications include a full set of industry dummies, defined at the 3-digit level classification. (See text).