CC BY
29№ 1(25) 2012
L. Carbonari1
Quasi-fixed inputs in the Italian manufacturing: The case of the pharmaceutical industry
The aim of this paper is to study the demand for inputs in the Italian manufacturing, using firm-level data on pharmaceutical industry. The Italian pharmaceutical industry is characterized by the existence of long-term labor contracts, and this fact suggests to consider labor as quasi-fixed input. In order to characterize firms behavior we base our analysis on the restricted Generalized Leontief cost function. The choice of this flexible functional form is due to its ability to capture the input substitution patterns in presence of more than one quasi-fixed input. Therefore demand and substitution elasticities are estimated with respect to two different theoretical models: the first, QFI (1), with capital as quasi-fixed input and the second, QFI (2), with two quasi-fixed inputs, capital and labor. The choice among the two alternative specifications is based on an elasticity comparison criterion, since the two models are not nested. The results suggest a rigid productive structure during the period under observation. Our results confirm the a priori on the labor market rigidity and point out the high heterogeneity between the firms, even controlling for size and nationality. Keywords: quasi-fixed inputs; restricted Generalized Leontief; pharmaceutical industry. JEL classification: C33; D24; L65.
In recent times, most surveys have underlined the Italian transition towards a more flexible labor market, e. g. (OECD, 2004, 2005). Starting from early 90's, a series of legal reforms has been introduced to raise labor market efficiency and to reduce the unions' bargaining power2. The main result of such a policy is that now flexible workers are the overwhelming majority of overall new hired. From an economic standpoint, such a growing flexibility in hiring could offer a further argument to model labor as a variable input. Nevertheless, especially in the manufacturing, available data suggest that labor demand, at least in the short-run, is irresponsive to the change
1 I am very grateful to Bristol-Myers Squibb for the research grant that made this research possible. I acknowledge Fabrizio Gianfrate, and Gabriele Mazzoletti for their generous support. I also would like to thank Domenico Depalo, Alessia Isopi, Luisa Carpinelli, Valentina Meliciani, Amelia Perea, and Rocco Ciciretti for their helpful suggestions. I am deeply indebted to Vincenzo Atella for his patience in reading several versions of this paper and for providing the computer code. The usual disclaimer applies.
Corresponding address: lorenzo.carbonari@uniroma2.it.
2 In 1994 the use of the training on the job contracts — originally introduced ten years before to reduce youth unemployment — was extended to a wider range of situations. In 1995 there was the introduction of the ongoing collaboration contracts while in 1997 the temporary work agencies broke the monopoly of public employment agencies. In 2001 the use of fixed-term contracts for subordinate workers as well became legal. Finally, in 2003, the so-called Biagi Law has given the variety of atypical labor contracts a common framework (Russo, Veredas, 2000).
1. Introduction
№ 1 (25) 2012
in the production scale and shows an adjustment pattern very close to the one of a quasi-fixed input (Carbonari, 2007)3.
In this paper, we examine the productivity performance and the demand for inputs of the Italian pharmaceutical industry during the period 1991-2004. There are only few papers in the literature that focus on this issue, referring to the Italian case and using firm-level data. The majority of those moves from a labor economics perspectives and investigates the effectiveness of the deregulation policies or the employment dynamics, (Bertola, Ichino, 1995; Russo, Veredas, 2000). Differently, this paper belongs to the strand of the empirical literature that applies flexible functional forms to characterized productive behavior (Morrison, Berndt, 1981; Morrison, 1986, 1993; Atella, Quin-tieri, 1998a, b; Pierani, Rizzi, 2003). Precisely, our contribution builds upon Morrison (1988) in specifying a restricted cost model that accommodates multiple quasi-fixed inputs and maintains the consistency of the estimated function with microeconomic theory and approximation properties. We base our analysis on the restricted Generalized Leontief cost function proposed by Morrison (1988, 1993). The choice of this flexible functional form is due to its ability to describe factors demand, when both physical capital and labor behave as a quasi-fixed input. The productive technology consists of one aggregate output obtained using:
• three variable inputs (hired labor, intermediate inputs and productive services) and one quasi-fixed factor (physical capital) in the first model QFI(1);
• two variable inputs (intermediate inputs and productive services) and two quasi-fixed factors (physical capital and hired labor) in the second model QFI(2).
|T In order to test the hypothesis of labor rigidity, we will use a new dataset on the Italian pharmaceutical industry. The case of pharmaceutical sector is suitable for our purposes because, as ~ the rest of the Italian manufacturing, it is highly characterized by the existence of long-term labor 'I centralized contracts, which exacerbate labor rigidity4.
o The paper proceeds by presenting the main features of the Italian pharmaceutical sector (sec-| tion 2). Section 3 defines the econometric model used for empirical implementation and the input-output equations. In section 4 we briefly discuss about the dataset and the variables construc-£ tion and provide the short and the long-run elasticities of the two alternative models. Section 5 o concludes.
<u £
^ 2. The Italian pharmaceutical industry: figures and trend
o The Italian economy has historically been characterized by a large pharmaceutical sector. From | the end of World War II until the end of mid-sixties many Italian entrepreneurs contributed to dels velop a modern, fast-growing industry, able to respond to the increasing demand of medicines. c
■S -
S 3 For example, focusing on the period 1993 - 2003, the Italian Central Statistical Office reports that while the
« deseasonalized index of production in industry has grown by an estimated 1.7%, the number of workers has remained
■S substantially unchanged (ISTAT, 2008). •!2
= 4 This rigidity arises also from the Charter of Workers 'Rights that have tightly regulated employment relationships
■S in Italy since 1970. Particularly, the Statuto dei Lavoratori sets the procedures for hiring and firing, and defines the com-
<b pensation structure and the rules for workers mobility. Most of all, it prevents the dismissal of workers in absence of
¡5 appropriately motivation and prescribes that guilty employers must «compensate dismissed workers in kind, restoring
ra their employment status and paying back wages for all the period of litigation plus other monetary penalties» (Bertola,
<3 Ichino, 1995).
№ 1(25) 2012
Since then, despite this encouraging beginning, national firms started to decline in terms of com- "g petitiveness and innovation. .g
Currently, the Italian pharmaceutical market displays the same structure of other developed <3 ones: high entry barriers, a strong governmental intervention, and high profits for the sellers. As J shown in table 1, it is the third largest in Europe by total sales and by number of workers, and the fifth largest with respect to the number of companies. Moreover, it has historically attracted a relevant flow of foreign direct investments. The reason of these inflows of investments is twofold. On one hand, until the late 70s, foreign investors had been attracted by the large amount of funds granted by the Cassa del Mezzogiorno. On the other hand, the increasing demand for drugs led multinational enterprises to locate plants and commercial activities in Italy (Carbonari, 2007). These facts contribute to explain why, during the last thirty years in Italy, the world leaders' market share has been constantly increased5.
Table 1. European pharmaceutical industry (2004)
Country Total sales, million $US PPP Number of workers Number of firms
(powering power parity)
France 25392 98900 256
Germany 23242 119800 313
Italy 21989 73550 241
UK 16850 73000 362
Spain 13042 39000 245
Greece 7141 11200 —
Netherlands 4250 15500 130
Sweden 3117 21600 60
Switzerland 2953 29613 230
Portugal 2852 10691 —
Denmark 1839 15131 41
Norway 1684 4603 156
Finland 1651 7032 69
Source: Author's calculations on (Farmindustria, 2005).
The most relevant features of this industry, with respect to the rest of the Italian manufacturing sector, are the following:
• it exhibits a high segmentation due to the low substitutability among the drugs belonging to different therapeutic classes. The overall market is highly concentrated: in 2005, the top ten vendors represented the 50% of the total market, while the top hundred cover the 96.5% of the entire market (Farmindustria, 2006);
• it operates within a tightly regulated environment. In fact, the regulatory system affects the drug's safety/effectiveness, and its pricing6. There is also a unique institutional buyer, the National Health System, which is managed at regional level;
5 See (Carbonari, 2009) for further details.
6 As in other EU countries price regulation includes the entry price and any posterior increases.
№ 1 (25) 2012
• it is characterized by high investment in R&D. Even though there is an extreme lack in terms of research facilities in Italy, in 2004 the pharmaceutical industry invested 839 billions euros into R&D and the drug companies based in Italy employed 4314 individuals in R&D (Farmindustria, 2005)7. Despite this positive element, only few firms develop new drugs in Italy. The reason for this apparent contrast is dual. First, Italy has historically suffered a lack of incentives to innovation.
Second, R&D expenditure, derived from balance sheets, is only a proxy (often unreliable) of the research led by firms, because it also includes large resources invested on advertising, marketing and lobbying. Drugs' patent protection was introduced in Italy only in 1978. Since then, the Italian pharmaceutical industry has started to face an increasing number of difficulties, especially because of the high costs associated to the research in this sector. Hence, keeping aside the new generation of biotech firms, the majority of Italian companies usually prefers co-marketing techniques. Moreover, multinational companies find it more productive to invest in R&D abroad rather than in Italy (Carbonari, 2007).
Another important feature of this sector is the labor demand evolution. In the period 1985 - 2005, the number of workers employed in the Italian pharmaceutical industry has grown only at 0.8% per year, whereas the growth rate of fixed assets accumulation has been around 7.8% (Farmindustria, 2005). The comparison with the time evolution of both fixed assets and production volume s suggest that the a pattern of the employment by pharmaceutical firms cannot be interpreted as the result of optimizing behavior with flexible inputs and to test the hypothesis that labor behaves as a quasi-fixed input.
b
^ 3. The theoretical model
3
o The existing literature has shown the usefulness of considering firm's behavior in terms of cost
| rather than production function8. This choice allows to avoid the main problem that arises when
1Ü production function specification is implemented for econometric purposes: endogeneity versus
£ exogeneity of right hand side variables. Moreover, by using cost function, input demand can be eas-
o ily derived through Shephard's lemma. This section offers the theoretical structure of our analysis
w and the specification chosen in order to address the issues highlighted in the introduction. In this
® analysis we will consider an industry composed by H technologically independent firms facing a competitive input market. Suppose also there are Kquasi-fixed inputs xk (k = 1,...,K) subjected to
■[= increasing internal cost as in (Morrison, Berndt, 1981) and J variable inputs vj (j = 1,..., J), which
0 are available at constant price. Let p = {pj} be the vector of (exogenous) prices of variable inputs | and let C (xk) be the change total costs due a variation of quasi-fixed input. Furthermore, the cost Ü function exhibits the following properties:
| C(0) = 0, C'(|Xk|) > 0, C(|Xk|) > 0, (1)
1 where xk (net investment) is the first derivative of xk with respect to time. The firm's optimization problem consists in to choosing the sequences of variable inputs {vj, t} and quasi-fixed inputs
g. {xk, t} that minimize the present value of the stream of future cost subject to the production con-^ straint Y (t) = F (v, x, x, t):
s _
w 7 Around 5% of the entire national expenditure in R&D and the 9% of the whole manufacturing.
a 8
O 8 For a survey see (Morrison, 1993).
(25) 2012
œ \
C(0) = fie~ri 2WjVj + 2qkZk dt, (2)
è J k 0
where: r is the real discount rate; qk = pk / (r + d) is the asset price of xk and d the depreciation rate;pk is the rental price of new quasi-fixed input; zk = xk (t) + dxk (t — 1) is the gross investment in terms of k-th quasi-fixed input xk.
As in (Morrison, Berndt, 1981) and (Morrison, 1986), in order to minimize equation (2), first we derive the restricted cost function G that incorporates the solution of the short run optimization problem, and then we find the optimality conditions with respect quasi-fixed inputs. It is worth noticing that G (p, x, x ,Y, t), which is dual with 7(t), exhibits the following properties:
dG / dp, = v, is the optimal short run for variable input j (i. e. the Shephard's lemma);
j j
• d G / dxk = 1k is the shadow value of the quasi-fixed input k.
Hence, to define the pattern of quasi-fixed inputs accumulation, we must take into account the Euler equation corresponding to the following optimal control problem:
min C (0) + ^qkxk (0) = e~rt G (p, x, X, Y, t) + J
(3)
dG d dG
which implies:
dxk dt dXk
(4)
с о
£ и о
-J
where xk (0) is the initial stock of quasi-fixed input. Using the calculus of variation to solve problem (3), we obtain the following first order conditions:
(Gx + u) - (-G + GxX + GxX + Gxt )= 0, (5)
where: u is the vector ofpk; x is the second derivative of xk with respect to time9.
In order to solve the second order differential equation (5), we expand linearly around (x, X, t) = (x* (t), 0, t) at time t, where x* (t) — in the simplest case of only one quasi-fixed input (e. g. the physical capital) is the unique value that solves:
-Gx (•) - rGx (•) = pk, (6)
where the LHS is the shadow value and the RHS is the market rental.
k
k
The flexible functional form chosen for our purpose is the Generalized Leontief used by Morrison (1988) to compare firms' behavior in U. S. and Japanese manufacturing. The advantages of this cost function is that it is relative parsimonious of parameters and, above of all, it allows an analytical computation of the «desired» level of input stocks x* . The Generalized Leontief for multiple quasi-fixed inputs can be expressed as:
9 Notice that the assumption of static expectations imply that both p and Y are set equal to zero. Therefore we omit the term Gxpj> + GxyY in equation (5).
№ 1 (25) 2012
/ \ g=y YZ^j+2pi22у~«5
\ i j i m i m n 0
+
+Y05| HdkP,4-5 + 2Pillr^xr 1 + 2p22gkix05xi05, (7)
\ i k i m k / i k l
where:
p is the price vector of variables inputs identified by the subscripts i, j; x is the stock of quasi-fixed input identified by the subscripts k, l; Y stands for the level of output;
s captures other exogenous argument of G not included in the return to scale specification; m, n are the subscripts denoting the exogenous arguments of G not included in the return to scale specification (e.g. state of technology, net investment in quasi-fixed inputs Dxk, etc...).
Notice that in equation (7) G is linear homogeneous in price, and the global convexity inxk , as it happens with many other flexible functional forms, is not guaranteed. Furthermore, differently from Morrison (1986,1988), we do not impose any restriction on the parameter in order to allow for non-constant returns to scale10.
Following (Morrison, 1988), we can easily obtain the explicit shadow value of capital Zk by deriving equation (7) with respect to xk :
<n
Zt =-0.5
3
y 05 х-0-512**Pi+2 p, IrmkC I+2 p Em-5 x?
(8)
8 Hence, at the equilibrium:
^k =-°x (•) -rGx (•) = pk, (9)
■c a <u ■c
с
and the «desired» stock of the quasi-fixed input is given by:
x*= F (p, Pk, Y, t, Xk). (10)
Demand equations suitable for empirical analysis may be derived from equation (7) simply by applying Shephard's lemma and by dividing the I input demand for Y in order to reduce possible heteroskedasticity:
1 Y ° IPG 1=2 a j (Pj / pi Г+2«5+22g mn^m5 i5+
-i i j m m n
Ф
•¡2 3
a
Js
+H2«5+22gmk«5j+y-1 22gkixfx-. (ID
k 1
The system of estimable equations comprised in equation (11) represents the firm demand behavior in the short-run.
| 10 In fact, as explained in section 1, pharmaceutical industry exhibits high entry barriers and profits that suggest sig-<3 nificant returns to scale.
56
m
(25) 2012
Given the restrictions made on the cost function, short-run elasticities are defined as:
d ln v, dv. p,
dlnp, dp, vt
с
о ■p
о
(12) J
£SR =
iY °
d ln v dv Y
i _ i
d ln Y dY vi while the long-run elasticities are defined as:
flr = P, '
1 V.
1 ]
dv±_+y dv,. dXk dp, & дч dp
, /
(13)
(14)
4. Empirical implementation
Given the traditional rigidity exhibited by the Italian labor market, we test the dynamic model described above in two different versions: one where only physical capital is quasi-fixed, and the other where also labor is treated as a quasi-fixed input. This section provides the results of the empirical implementation of these two models carried out on firm level panel data. Our database was built from balance-sheet files collected by Bureau van Dijk Electronic Publishing (BvDEP), that sets up computer-readable files from the original balance-sheet reports. It contains information on the top hundred companies, ranked by revenue, which represent the 96.5% of the entire Italian market (Farmindustria, 2006). The data cover the period from 1991 through 2004 — not all firms are present in all years — and are relative to gross production, labor, materials, services, fixed assets, and investments11. We also use additional information on the real interest rate, the depreciation rate, and consumer price indexes12. It is worth noticing that, while fixed assets are deflated using the one-digit industry specific deflator from ISTAT, labor cost is obtained through a firm-specific index (pL) that is equal to the average wage per worker. The technological change is represented by a non-firm-specific time trend (t) and finally, in order to capture the effects of firm size and nationality, specific dummy variables are defined13.
Summary statistics are presented in table 2.
11 Services include the outside labor and and/or materials for specialized or overflow work. Physical capital is given by the sum of equipment, machineries and structures.
12 The source for the interest rate data is the Bank of Italy; depreciation rate has been set equal to 10%; the source for consumer price index is the Italian Institute of Statistics (ISTAT).
13 Dummy variables are used in order to evaluate the effects of firms' size and their nationality. The former is given by the (average) total revenues, the latter is given by the ownership nationality. The number of workers has also been used in order to capture the size, but any statistically significant difference in the results has been found. Following (Morrison, 1988) and consistently with the Shephard's lemma, these dummy variables have been integrated and added as terms in the restricted cost function equation (7). See the table 9 in the Appendix for further details on data construction.
№ 1 (25) 2012
Table 2. Summary statistics of selected variables (for 695 observations)
Variable Definition Unit Mean Standard deviation Min Max
Y production l06 current euros 142.119 221.081 0.081 1812.924
Vs services demand l06 current euros 27.926 38.576 0.001 272.800
vI materials demand l06 current euros 77.567 120.210 0.010 900.718
XK capital demand l06 current euros 19.113 29.409 0.000 223.189
VL labor demand 102 number of workers 6.827 12.467 0.010 110.420
Ps services price index 2000=1 1.004 0.082 0.729 1.136
Pi materials price index 2000=1 0.993 0.045 0.801 1.081
Pl labor price index 2000=1 1.071 0.703 0.014 14.137
Pc capital price index 2000=1 0.993 0.042 0.787 1.050
4.1. Descriptive analysis
Table 3 provides a brief sketch of the characteristics of the firms included in our sample through the most relevant indicators of firm performance. The existence of high heterogeneity for many of
the variables of interest suggests to focus on median values.
с
■с a <u ■с
Table 3. Overall sample, relevant indicators (in millions of current euros)
» с
3 t> .2 э с w S с
ф
;C .С
•¡2 з a с
Year Total Revenues Value Added Ebit Cash flow Equity Working capital
2004 mean 252.163 81.679 9.151 21.597 18.813 86.682
median 106.615 29.731 2.453 7.636 7.748 31.672
2003 mean 226.108 77.309 7.204 20.336 16.588 70.636
median 89.096 26.930 2.857 6.973 7.006 24.515
2002 mean 203.132 66.038 11.102 21.852 13.472 55.997
median 80.319 24.107 2.563 6.123 5.168 16.255
2001 mean 166.606 55.159 9.147 18.060 12.975 61.832
median 74.368 24.149 2.571 7.662 5.170 17.554
2000 mean 166.156 55.183 8.382 16.500 13.836 58.138
median 69.730 22.347 2.162 4.866 5.165 17.174
Note. Value added is defined as total revenues less non-labor costs of inputs. Net result is defined as total revenues minus total costs (business, depreciation, interest, and taxes). Cash flow is equal to net result plus amounts charged off for depreciation, and amortization. Working capital is given by the difference between current assets and current
"g liabilities.
An interesting feature emerging from descriptive analysis is the evolution of capital and labor <3 demands, that corroborates our a priori on the labor market rigidity. Figure 1 plots the evolution
58
№ 1(25) 2012
of physical capital demand and labor demand with respect to the aggregate production. This graph shows that the pattern of labor demand is very close to the one of physical capital14.
physical capital labor
Fig. 1. Do capital and labor behaves as QFIs?
Note. Values are represented on a logarithmic scale.
Our data confirms also that, keeping aside the early 90's when pharmaceutical aggregate production was hit by a strong exogenous shock (the so-called pharma bribery scandal), the labor demand exhibits a substantial degree of rigidity (with respect to the scale of production) until the late 90's. Another interesting issue arising from the descriptive analysis concerns the distribution of capital across workers. Table 4 provides the correlation between capital/labor ratio and price ratio pK/pL. As suggested by the theory all values are negative, though statistically insignificant.
Table 4. Capital & Labor (1991-2004)
C°rrfe 1 xL, Pk 1Pl)
Overall sample - 0.027
Small size - 0.034
Medium size - 0.088
Large size - 0.075
Note. xK is the stock of physical capital, xL is equal to the number of workers, pK is the rental price of xK and pL is the average wage.
The upper graph of figure 2 plots the evolution of the dispersion of the logarithm of the capital/labor ratio and of the logarithm of the average wage (not deflated by the CPI). The remarkable fact documented in figure 2 is that, while the cross-firm dispersion of the capital/labor ratios has increased significantly during the period 1991-2004, the dispersion of wages remains almost constant. As mentioned before, this phenomenon is likely due the existence, in pharma-
14 Annual average values for input demands and production level have been computed and than properly rescaled to the interval [0,1].
№ 1 (25) 2012
<л з ■ö с
ceutical sector as well as in the entire Italian manufacturing, of long-term labor contracts that provide a source of rigidity in wage. In fact, as measured by the intra-quartile range, the dispersion of the capital/labor ratio within the pharmaceutical industry appears fairly stable throughout the first half of the 90s', then it increases sharply and reaches a peak in 2000. In contrast, consistently with the transition towards a more flexible labor market, the path followed by the average wage, which is stable along the nineties, started to exhibit more volatility only from 2002. In the period 2002-2004 we found an increased inequality both in wages, and in capital/labor. The lower panel shows that the average capital/labor ratio, for small firms in particular, has sharply increased. Moreover, we find an increasing heterogeneity not only between different size groups but also within groups.
Dispersion of capital per worker and pay per worker (logs)
1.6 1.4 1.2 1
0.8 0.6 0.4 0.2
VAR[log(xr /*,)]
VAR[log( p)]
■c a <u ■c
0.0 -0.2
Capital/labor ratio (logs) — disaggregation by size
С 3
о
.2
с w S с
ф ■с
3
а .с
тз
Js
-0.4 -0.6 -0.8 -1.0 -1.2 -1.4
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 h h small medium large
Fig. 2. Capital per worker
Note. K is the stock of physical capital; L is the number of workers.
0
№ 1(25) 2012
4.2. Estimation 'g
In order to evaluate the demand factors patterns of the Italian pharmaceutical industry, the re- <3 stricted Generalized Leontief cost function is employed under two different specifications. The J first, QFI(1), with one quasi-fixed input and the second, QFI(2), with two quasi-fixed inputs. Technology describes how variable inputs (e. g. services, materials, and labor in the first model), quasi-fixed inputs (capital, and labor in the second model), and a proxy to capture the impact of technical change are used to create output.
In line with the existing literature, we apply the technique of iterated seemingly unrelated regressions (Zellner, 1962) in order to estimate the system of equations (11) and the Euler equation (9) presented in section 3.
Full results from the two alternative dynamic models with static expectations are presented in the Appendix (see tables 10 -11).
According to the economic theory, relative prices parameters ap are positive, even if sometimes non-significant. The parameters 51 of the technological change are generally negative. This result shows that most of the firms in the sample, during the period 1991-2004, have applied factor saving procedures, except for those of small size and Italian nationality15. Actually, in an environment where physical capital complements labor, innovative procedures presumably require increasing demand for (high skilled) workers and services, at least in the short-run. The interaction between input prices and the stock of quasi-fixed input, captured by dy, presents different results according to the model implemented.
The estimation of the two models used to test the effective rigidity of labor provides results consistent with our set of assumptions. The regularity conditions hold under both specifications, yet under the QFI(2) model we obtain better results with respect to dG /dxk and d2G /dx2k (see tables 5-6). The global fit is higher in the QFI(2) model, and the parameter jLAL, that catches directly the labor adjustment effects, is significant and negative.
Table 5. QFI (1) model, regularity conditions
Overall sample, % Small size,% Medium size, % Large size, %
G > 0 100.0 100.0 100.0 100.0
dG / dpL > 0 79.4 67.6 91.0 80.5
dG / dpS > 0 78.7 66.6 93.4 72.6
dG / dp, > 0 83.0 67.2 99.0 83.2
dG / dY> 0 51.8 16.7 95.2 31.9
dG / dt < 0 17.0 0.0 33.0 18.0
not all dxK / dY > 0 100.0 100.0 100.0 100.0
dG / dxK < 0 100.0 100.0 100.0 100.0
d 2G / dx2 K > 0 50.9 46.1 45.0 78.8
15 The 55.4% of the Italian companies in the sample are small size.
№ 1 (25) 2012
Table 6. QFI(2) model, regularity conditions
Overall sample, % Small size, % Medium size, % Large size, %
G > 0 100.0 100.0 100.0 100.0
dG I dpS > 0 56 63 47 61
dG I dp, > 0 69 68 66 81
dG I dY> 0 58 17 86 91
dG I dt < 0 80.6 99.7 66.8 66.4
dxL I dpL < 0 88 86 91 84
not all dxK I dY > 0 100.0 100.0 100.0 100.0
dG I dxK < 0 100.0 100.0 100.0 100.0
dG I dxL < 0 100.0 100.0 100.0 100.0
d 2GI dx2 K > 0 70.1 99.7 74.4 91.2
d 2GI dx2 k > 0 90.2 57.7 99.7 41.6
Note. Percentages in tables 5-6 indicate the times in which regularity conditions have been satisfied along the period 1991-2004.
The best global fit and the lower values of B. I. C. (see table 7) are not the only reasons to consider the QFI(2) model the better to interpret the Italian case. Further useful insights arise from the assessment carried out through the elasticity estimates (table 8).
<n a ■ö с
■с a <u ■с
Table 7. Comparing models
Model Schwarz B. I. C.
QFI(1) short-run 5970.41
QFI(1) long-run 6182.84
QFI(2) short-run 2885.07
QFI(2) long-run 2834.32
Table 8. Short-run and long-run selected elasticities
» с
3 t> .2 э с w S с
ф
;C .С
•¡2 з a .с
Js
QFI(1) model QFI(2) model
Mean Median Standard deviation Mean Median Standard deviation
Short-run
eLL -1.341 - 0.143 2.437
e LS 0.318 0.300 0.501
eLI 0.962 - 0.066 2.283
ess - 6.263 - 0.711 21.780 0.689 4.086 85.843
eiI - 4.286 - 0.591 9.601 - 0.575 1.908 7.387
eLY -13.911 -3.155 24.725
SSY - 4.876 0.210 16.346 3.857 0.952 57.764
eiY -5.398 - 0.262 12.389 0.019 0.847 3.474
62
№ 1(25) 2012
Ending the table 8 ...
QFI(1) model QFI(2) model
Mean Median Standard deviation Mean Median Standard deviation
Long-run
eLL 0.316 0.460 0.942 0.027 0.000 0.439
eLY -1.060 0.467 1.642 0.723 0.000 16.590
Sjy 1.466 0.987 1.714 - 4.908 1.416 70.994
eCL 0.991 0.000 0.696 0.327 0.000 4.720
ecc 0.053 0.000 0.459 - 0.417 - 0.336 0.367
e CY 0.594 0.177 0.135 0.153 0.100 1.100
eGY 0.052 0.000 0.504 0.501 0.364 1.442
In both models, the intermediate input demand and the labor demand are much more responsive to the scale of production than to their own prices, regardless of the model implemented. In contrast, short-run and long-run changes in demand of services seem to depend more on their price then on production level. These results are consistent with the existence of economy of scale especially for small-size companies. Comparison of tables 5 and 6 yields a remarkable difference between the two models concerning the presence of productivity improvements. QFI(1) model suggests no productivity improvements at all for small enterprises, whereas QFI(2) model implies almost 100% of the small firms had productivity improvements. Similar rather different findings (although not to the same extent) can be seen for the medium and large enterprises16. This is an argument in favor of modeling labor as a quasi-fixed input. In fact, this specification is able to reproduce the high productivity of small size firms which is one of the main features of this sector (Bobulescu, Soulas, 2006)17.
When considered as a variable input, labor exhibits a short-run own-price elasticity that indicates a relative degree of responsiveness that tends to disappear in the long-run under both specifications. The elasticity of the demand for physical capital with respect to its own price exhibits values less than 1 under both specifications. Only the one evaluated when labor is treated as quasi-fixed input is consistent with economic theory.
As the scale of production increases, interesting results are provided by the values of eLY that emphasizes the differences between the two specifications implemented. In the QFI(1) model, labor appears very sensitive to changes in output: in the short run, a unit increase in Y reduces significantly the labor demand (around 13%). The reason of such a high and non-realistic level of eLY is that in QFI(1) model labor is a variable input, therefore when firms experimented a negative shock — i. e. the pharma bribery scandal — that remarkably reduced the output level (see upper
16 Short-run price and substitution elasticities of capital in QFI(1) model and capital and labor in QFI(2) model with respect to the other inputs are zero by definition because they are evaluated for a constant value of the quasi-fixed input.
17 The majority of the Italian small size pharmaceutical companies are basically research-oriented often located in spatially concentrated districts or technological clusters. In such a context, firm size is of secondary importance and the innovative process is carried out by inter-firm cooperation and/or benefits from technological transfers inside clusters.
№ 1 (25) 2012
graph of fig. 3 in Appendix) the effect on labor demand is considerably overestimated. At the opposite, in the long run, values of eLY strengthen the idea that Italian firms have applied labor saving procedures and that labor demand adjusts slowly.
The QFI(2) model moving from the assumption that labor is a quasi-fixed input displays a slightly different pattern. Indeed, even though positive (the mean of the eLY is less than unit while the median value is null), the elasticity witnesses the labor input relative rigidity.
Production level exerts a slight effect also on capital demand. Indeed the mean value of eKY is equal to 0.594 (median 0.177) in the QFI(1) model and is close to zero, 0.153 (median 0.100), in the QFI(2) model. These results are presumably due to the numbers of years taken into account. In fact, thirteen years may be not enough to capture physical capital adjustments in manufacturing.
Cross-price elasticities indicate a high substitutability between materials and labor in the short-run. Finally, according to the theory, we have found that the intermediate input and services own-price elasticities e n and eSS how negative values.
5. Concluding remarks
In this paper we argue that labor rigidity is a crucial to represent properly the technology and the firm behavior of the Italian pharmaceutical sector. We estimate a restricted Generalized Leontief function with multiple quasi-fixed inputs using a novel firm-level panel data. Two different specifi-is cations have been tested in order to analyze the productive behavior of the top hundred pharmaceutical companies operating in Italy, ranked by revenue. The econometric analysis provides important ~ insights on the relationship between input substitution patterns and firms' performances. Although 'I the two models can not be directly compared (statistically), since they are not nested, important 8 questions are addressed by comparing the elasticities. Particularly, three main results emerge: | i) significant returns to scale (eGY < 1) have been found under both specification while only the QFI(2) model is able to capture the correlation between size and productivity experienced within £ this sector;
o ii) cross-prices elasticities suggest a general substitution pattern among capital and labor that SS is more slight in the QFI(2) model;
® iii) estimations confirm the a priori on the labor market rigidity.
These results lead to the conclusion that QFI(2) model provides a better interpretation of the ■[= labor demand in this sector and elasticities more consistent with both the economic theory and the
0 existing literature on this sector. Finally, from an econometric standpoint, the QFI(2) model exhibits | a better global fit and lower values of Schwarz B. I. C. both in the short run and in the long run.
to
1 References
<u
■S Atella V., Quintieri B. (1998a). Cambiamento tecnologico e domanda dei fattori nell'industria manifat-
<0
a turiera italiana. Rivista di Politica Economica, 2, 3 - 42. a
Atella V, Quintieri B. (1998b). Productivity growth and the effects of recessions. Giornale degli Econo-
£ misti e Annali di Economia, 57 (3-4), 359 - 386. «is
Bertola C., Ichino A. (1995). Crossing the river: A comparative perspective on Italian employment dy-<3 namics. Economic Policy, 21, 359 - 420.
№ 1(25) 2012
Bobulescu R., Soulas C. (2006). Innovation and firm size in the pharmaceutical industry. International ...
Journal of Business Environment, 1 (2), 253 - 264. |
-is
Carbonari L. (2007). I fattori di sviluppo dell'industria farmaceutica italiana — un'analisi storica. Man- fe uscript. J
Carbonari L. (2009). How variable is labor input in the Italian manufacturing: the case of the pharmaceutical industry. CEIS Research Paper, 140, Tor Vergata University, CEIS.
Farmindustria (2005). Fatti & Cifre. http://www.farmindustria.it/pubblico/faci2005.pdf.
Farmindustria (2006). L'offerta e la distribuzione dei farmaci. http://www.farmindustria.it/farmindus-tria/documenti/in20060l. pdf.
ISTAT (2008). http://con.istat.it/.
Morrison C. J. (1986). Structural models of dynamic factor demands with nonstatic expectations: An empirical assessment of alternative expectations specifications. International Economic Review, 27 (2), 365 - 386.
Morrison C. J. (1988). Quasi-fixed inputs in U. S. and Japanese manufacturing: A generalized Leontief restricted cost function approach. The Review of Economics and Statistics, 70 (2), 257 - 287.
Morrison C. J. (1993) A microeconometric approach to the measurement of economic performance. Springer-Verlag, New York.
Morrison C. J., Berndt E. R. (1981). Short-run labor productivity in a dynamic model. Journal of Econometrics, 16, 339 - 365.
OECD (2004). Employment outlook 2004. http://www.oecd.org/document/62/0,3746,en_2649_33927 _31935102_1_1_1_1,00.html.
OECD (2005). Employment outlook 2005. http://www.oecd.org/document/1/0,3746,en_2649_33927_ 34855489_1_1_1_1,00.html.
Pierani P., Rizzi P. L. (2003). Technology and efficiency in a panel of Italian dairy farms: An SGM restricted cost function approach. Agricultural Economics, 29, 195 - 209.
Russo G., Veredas D. (2000). Institutional rigidities and employment on the Italian labor market: The dynamic of the employment in the large industrial firms. CELPE Discussion Papers, 53, Centre of Labor Economics and Economic Policy, University of Salerno, Italy.
Zellner A. (1962). An efficient method of estimating seemingly unrelated regressions and test of aggregation bias. Journal of American Statistical Association, 57, 500 - 509.
Appendix
Table 9. Overall sample, disaggregation by size
Size
Italian
MNE
Total
Overall sample Small size,
(avg.) total revenues < 60 millions euros Medium size,
60 millions euros < (avg.) total revenues < 300 millions euros
Large size,
(avg.) total revenues > 300 millions euros
390 169
174 47
305 124
115
66
695 293
289 113
65
№ 1 (25) 2012
Table 10. Estimations
SUR estimates
<n
■ö с
■с
а ф
■с
.с С з '<3 .2 з с
W
S с
<u
.С
•¡2 з а .с
тз
Js
QFI(1) model QFI(2) model
Parameter Size* Estimate P-value Parameter Size Estimate P-value
ÖSS 1 6.662 0.810 aSS 1 8.402 0.743
2 7.141 0.620 2 15.214 0.243
3 -4.098 0.747 3 -0.548 0.962
4 0.566 0.979 4 8.781 0.646
5 5.396 0.719 5 11.336 0.401
6 -2.824 0.823 6 21.515 0.061
aS! 1 -1.743 0.941 aSI 1 1.114 0.959
2 -1.870 0.874 2 -2.222 0.832
3 10.851 0.310 3 13.710 0.152
4 -0.215 0.991 4 -0.706 0.965
5 -0.023 0.999 5 0.336 0.977
6 9.021 0.418 6 -0.636 0.949
aSL 1 -0.119 0.971 aSL 1 -0.093 0.944
2 -0.304 0.730 2 -0.868 0.054
3 0.646 0.072 3 -0.921 0.000
4 5.607 0.060 4 -1.122 0.000
5 -0.696 0.585 5 -0.583 0.056
6 2.973 0.002 6 -2.322 0.000
ÖSt 1 -2.576 0.425 dst 1 -9.020 0.005
2 -3.660 0.179 2 -9.820 0.000
3 2.838 0.286 3 -4.454 0.098
4 -1.714 0.559 4 -8.812 0.003
5 -3.465 0.195 5 -9.325 0.001
6 1.054 0.674 6 -5.203 0.041
d SY 1 -0.165 0.848 d SY 1 1.031 0.283
2 0.480 0.530 2 0.953 0.231
3 -11.269 0.000 3 -7.702 0.000
4 -0.637 0.440 4 1.231 0.187
5 0.405 0.601 5 0.847 0.293
6 -8.757 0.000 6 -9.186 0.000
dS&C -0.025 0.730 ds&c 0.010 0.000
dSC 1 0.346 0.089 dSC 1 -0.167 0.000
2 0.147 0.000 2 -0.050 0.000
3 0.182 0.000 3 -0.026 0.001
4 0.657 0.000 4 -0.144 0.000
(25) 2012
Continued the table 10
SUR estimates
с о
£ и о
■J
QFI(1) model
QFI(2) model
Parameter Size* Estimate P-value Parameter Size Estimate P-value
5 0.134 0.000 5 -0.044 0.000
6 0.191 0.000 6 -0.030 0.002
aiL 1 -0.569 0.908 aiL 1 -1.511 0.259
2 -0.243 0.860 2 -1.053 0.016
3 -0.132 0.764 3 -0.770 0.000
4 12.183 0.006 4 -0.693 0.001
5 -0.154 0.937 5 -1.279 0.000
6 8.074 0.000 6 0.194 0.279
ail 1 7.551 0.736 aII 1 10.247 0.606
2 6.757 0.566 2 14.657 0.162
3 0.659 0.950 3 2.718 0.772
4 -2.855 0.876 4 8.208 0.597
5 5.006 0.687 5 12.011 0.274
6 0.797 0.939 6 19.568 0.037
di, 1 -2.496 0.492 di, 1 -9.184 0.007
2 -3.434 0.213 2 -9.508 0.000
3 4.861 0.081 3 -2.891 0.295
4 -1.772 0.590 4 -8.745 0.004
5 -3.470 0.224 5 -9.196 0.001
6 7.693 0.005 6 -1.30 0.630
d IY 1 -0.174 0.853 dIY 1 1.075 0.278
2 0.493 0.560 2 1.033 0.217
3 -16.446 0.000 3 -11.794 0.000
4 -0.837 0.344 4 1.354 0.157
5 0.451 0.608 5 0.811 0.343
6 -22.924 0.000 6 -14.605 0.000
dIC 1 -0.480 0.022 dIY 1 0.148 0.002
2 -0.153 0.000 2 0.033 0.000
3 -0.194 0.000 3 0.0087 0.210
4 -0.867 0.000 4 0.125 0.000
5 -0.124 0.000 5 0.027 0.000
6 -0.193 0.000 6 0.012 0.173
aLL 1 5.993 0.776 gcY 0.032 0.491
2 5.417 0.616 Yly 0.144 0.000
3 22.175 0.022 У LDL -0.037 0.083
№ 1 (25) 2012
Ending the table 10
<n з ■ö с
■с
а ф
■с
.с с
3
'<3 •2 с
W
S с
<u
.С
•¡2 з а .с
Js
SUR estimates
QFI(1) model QFI(2) model
Parameter Size* Estimate P-value Parameter Size Estimate P-value
4 32.998 0.043 d LL 0.176 0.000
5 5.371 0.629 dee -0.499 0.000
6 11.866 0.225 deL 0.162 0.000
1 -2.644 0.718 d SDL 0.034 0.487
2 -3.316 0.402 dIM -0.059 0.271
3 12.942 0.001 УIDL -8.61E-03 0.002
4 1.037 0.866 Угл1 4.13E-03 0.000
Угле -2.97E -05 0.649 Угле -6.96E-06 0.000
Yyy 0.076 0.089 Угг -5.61E-03 0.828
Y,y -0.161 0.223 УtY -0.073 0.286
gtt 0.767 0.091 У„ 0.902 0.000
Уе -3.88E-03 0.000 Уе 2.85E-03 0.000
Уле 2.21E-03 0.033 У.ле 2.69E-05 0.000
Уеу 6.61E-06 0.972 УеY -1.96E-04 0.004
У еле 2.17E-03 0.356 Уеле -1.28E-04 0.004
du 5 -2.747 0.535
6 11.863 0.003
dLY 1 -0.139 0.928
2 0.279 0.852
3 -35.44 0.000
4 -8.16 0.000
5 -0.117 0.944
6 -31.045 0.000
д1ле -2.75E-03 0.741
dee -0.647 0.000
dLe 1 0.138 0.003
2 0.038 0.000
3 0.046 0.000
4 0.227 0.000
5 0.025 0.000
6 0.035 0.000
* 1 — large size national firms; 2 — medium size national firms; 3 — small size national firms; 4 — large size multinational firms; 5 — medium size multinational firms; 6 — small size multinational firms.
nPHKMffHAR 3K0H0METPMKA /-
I № 1(25) 2012
Table 11. QFI(1) model versus QFI(2) model
<0
- c
QFI(1) model |
Dependent variable: — Mean Y Standard deviation Sum of squared residuals Variance of residuals 5.21184 22.9291 221775 319.56 Standard error of regression P-squared LM het. test Durbin — Watson 17.8762 0.395179 8.86965 [0.003] 1.4235
Dependent variable: — Mean Standard deviation Sum of squared residuals Variance of residuals 1.26042 4.4179 8809.41 12.6937 Standard error of regression P-squared LM het. test Durbin - Watson 3.56282 0.456939 70.9963 [0.000] 0.757202
Dependent variable: — Mean Standard deviation Sum of squared residuals Variance of residuals 2.71918 9.4487 25567.3 36.8404 Standard error of regression P-squared LM het. test Durbin - Watson 6.06963 0.601098 153.912 [0.000] 0.59381
Dependent variable: pC Mean Standard deviation Sum of squared residuals Variance of residuals 0.993294 0.041722 5.88099 0.8474 E-02 Standard error of regression P-squared LM het. test Durbin - Watson 0.092055 0.14784 137.804 [0.000] 1.01128
QFI(2) model
Dependent variable: pL Mean Standard deviation Sum of squared residuals Variance of residuals 1.07105 0.702944 419.734 0.604804 Standard error of regression P-squared LM het. test Durbin - Watson 0.777691 0.085848 9.26978 [0.002] 1.18327
Dependent variable: — Mean Standard deviation Sum of squared residuals Variance of residuals 1.26042 4.4179 7293.56 10.5095 Standard error of regression P-squared LM het. test Durbin - Watson 3.24183 0.513907 22.5817 [0.000] 0.687794
Dependent variable: — Mean Standard deviation Sum of squared residuals Variance of residuals 2.71918 9.4487 15427.7 22.2301 Standard error of regression P-squared LM het. test Durbin - Watson 4.71488 0.751411 119.375 [0.000] 0.656393
Dependent variable: pC Mean Standard deviation Sum of squared residuals Variance of residuals 0.993294 0.041722 0.252957 3.64E-04 Standard error of regression P-squared LM het. test Durbin - Watson 0.019092 0.884992 91.8647 [0.000] 1.19063
Note. Square brackets contain P-values.
