## L’ouverture de l’Europe vers l’est

|# Chapitre VII. Adjustment Costs in Labor Demand : A Comparison between France and the Czech Republic

## Texte intégral

# INTRODUCTION

1Since the collapse of the Soviet economic system, an important literature has been devoted to the process of transition of central and eastern European countries toward a market economy. One of the main conclusions that can be drawn from this literature is that the transition process to a market economy requires large reallocation of workers from noncompetitive industries to competitive ones. This reallocation has substantial social costs as they generally induce a high level of unemployment and a dramatic fall in production, at least in the early years of the transition. At the same time, efforts of most eastern European countries to lower the costs of transition by encouraging early retirement, maintaining high unemployment benefits or pensions and creating severance pay may have increased rigidities on the labor markets (see e.g. Noorkoiv et al. (1997)) and may have slowed down the transition process. Labor-market rigidities hamper job reallocation, limit firing by unprofitable firms while they reduce hiring by new corporations. indeed, they face adjustment costs, such as transaction, planning and/or training costs (see Maccini (1987)), when they want to modify their labor demand in response to exogenous shocks. Since Oi (1962)’s seminal paper, economists have shown that this kind of costs may be high in industrialized countries, accounting for several months of pay (see e.g. Nickell (1986)). In the process of transition encountered by the eastern European countries, an important issue is to determine whether firms also incur such adjustment costs and, consequently, to which extent their labor demand is constrained. Shedding light on this problem may allow us to contribue to the debate about the relative degree of flexibility of the labor market in Eastern Europe with respect to the one observed in the European union. Moreover, our contribution may give some elements about the construction of an optimum currency area in Europe. Indeed, if western and estern European countries have labor markets with different structures, expressed in particular through different labor-demand adjustment costs, the impact of exogenous shocks may be asymmetric and then the European currency area could not be optimal.

2In the last few years, a number of empirical studies on labor-demand adjustment costs have confirmed the validity of one of the main conclusions that could be drawn from Oi’s contribution, i.e., hiring and firing costs are not symmetric, more especially they are not quadratic. These recent developments largely rely on the Generalized Method of Moments (GMM ; see Hansen (1982) and Hansen and Singleton (1982)) which makes it possible to estimate models with more flexible forms of adjustment costs.

3The aim of this paper is thus to provide an empirical comparison of labor demand functions stemming from different adjustment costs functional forms, across two countries, one from Western Europe, France, and the other one from Central Europe, the Czech Republic.

4This paper is structured as follows. In section 1, we present and discuss seven adjustment costs functions we have chosen and the corresponding dynamic labor demand models. Section 2 is devoted to the statistical description of our data. In section 3, we analyze the econometric results on which is based our comparison of structures of the French and the Czech labor demands.

# I. THEORETICAL ANALYSIS OF LABOR ADJUSTMENT COSTS

5We present here the general assumptions of our dynamic labor demand model and the seven specifications of adjustment costs we have considered. This last choice is justified since these costs can be a priori of different forms and because we compare countries that have a priori different labor market and productive structures.

## 1. General assumptions of the dynamic labor demand model

6The common framework to all the dynamic models we consider in this paper is that of an inter-temporally optimizing firm. We assume that firms evolve on product markets characterized by imperfect competition *à la *Cournot. Firms are assumed to be risk neutral and to form rational expectations about the future. In this context, a given firm *i *(*i *= 1, *n *) seeks to maximize the discounted expected value of the infinite sum of its current and future profits :

- 1 When firm
*i*hires,*HF*_{it}*>*0 and, alternatively, if it fires,*HF*_{it}*<*0 . In the same vein, if*I*_{it}*>*(...)

7where *E*_{t}*Z*_{t}+_{τ} denotes the expectation about *Z*_{t}+_{τ}, formed at time t, 9 is the discount rate, *Q*_{it}* *is the individual production, *Q*_{t}* *is the global production of the industry ( *Q*_{t}* *= Σ^{n}_{i=1} *Qi*_{t}* *) and P(Q_{t} ) is the inverse demand function of the product market. *r*_{t}* *is the user’s capital cost, *K*_{t}* *is the capital stock, *w*_{t}* *is the labor cost (wage plus payroll taxes), *L*_{t}* *is the number of employees. Firm being assumed price taker on both input markets, *w*_{t}* *and *r*_{t}* *are exogenous. *AC*(•) is an adjustment cost function that depends on number of the employees hired or fired at each period, *HF*_{it}* *, on the investment/disinvestment, *I*_{it}1*, *and on the stock of both inputs. As mentioned in the literature, different shapes of adjustment cost functions can be met by firms (see e.g. Hamermesh and pfann (1996)). Here, we assume for *AC*(•) that,

8The adjustment costs increase or are fixed/lumpy with respect to the hiring or firing of employees as with respect to investment or disinvestments :

9The stock of both inputs can create a scale effect. This can be either an experience/ »learning » effect or an « inertia » effect. In the first case, we have

10the sign of both partial derivatives being opposite in the second case.

11The adjustment cost function may be linear, concave or convex to changes in both inputs:

12Of course, the adjustment costs may be asymmetric depending on whether firm *i *hires (invests) or fires (disinvests).

13Hiring/firing and investment are linked with labor and capital by both following dynamic equations :

- 2 We will assume that, at any date, a constant and exogenous proportion
*s*e [0,1[ of worker leaves th (...)

14where *s *is a rate of exogenous quit (natural wastage)2 and ô is a depreciation rate of capital. Here, we assume that quits and capital depreciation induce no direct costs and are constant over time. Moreover, firm *i *is subject to a technological efficiency constraint:

15where *g *is a production function that will be a Cobb-Douglas. The *D*_{t}* *-s are time dummies standing for the technological evolution and possible macroeconomic shocks.

16From this constrained optimization program, we get the following first order conditions:

17^{for labor, &}

18for capital, and

19for production, where *λ*_{it+}_{τ}* *is the Lagrange multiplier associated with the technological efficiency constraint. From (2.3) and using Lerner’s equation without conjectural variations, we can rewrite the first order conditions for both production factors as

20where *ms*_{it}*+*_{τ}* *is the market share of firm *i *on product market and *ε*_{ Q}^{D}*/*_{P}* *the elasticity of demand with respect to the output price.

- 3 Such a hypothesis implies that cross-adjustment costs between labor and capital are low or non-exis (...)

21It must be noted that, unless both the production function and the adjustment costs are quadratic, it is not possible to derive from these Euler equations an explicit (i.e., a closed form) dynamic demand model for labor and capital without making some linear approximations (Meese (1980)) or assuming one of both inputs is fixed (Nickell (1986)). With more usual production functions, we have a few useful and relatively complex system of Euler equations. Assuming that adjustment costs are additively separable between labor and capital3, the first order condition for each input does not depend on the adjustment dynamic of the other input even if they depend on the stock of the latter. The system of simultaneous equations becomes simpler and then we can restrict our attention to the Euler equation for labor, the main object of our analysis.

- 4 Which will be log-linearized for estimation.

22Considering now the technology, we have considered a Cobb-Douglas production fonction defined as4 :

23From which we get the marginal productivity of labor as

## 2. Functional forms of labor-demand adjustment costs

24Different specifications of dynamic labor demand models can be obtained, depending on the assumptions made about the adjustment costs *AC*(•). We will consider seven specifications of the labor-demand adjustment costs fonction (or equivalently seven models), together with a Cobb-Douglas production function. Our objective is to consider different adjustment costs functions among the most usual in the literature but also as flexible as possible. Obviously these specifications must have the different properties presented above.

25Since its introduction by Eisner and Strotz (1963), the quadratic adjustment costs fonction *AC (**Δ**L*_{it}* *) = *(c/*2)(Δ*L*_{it}* *)^{2} -becoming

26if one assumes a rate of exogenous quits - has been used over more than twenty-five years as the « canonical » form for most dynamic labor demand studies. The convexity of this fonction forces the adjustment costs to be everywhere increasing at the margin. Its symmetric form implies that the hiring costs a firm faces for hiring a given number of workers are equal to those it would face for firing the same number of workers. As we already mentioned this hypothesis does not seem to fit correctly the observed facts and, at least, there is no reason to assume it to be true a priori given the very different origins of these costs. Indeed hiring costs essentially account for recruitment and training costs while firing costs essentially consist in severance payments, administrative constraints and possible reorganization costs. This symmetry assumption was made mainly for analytical convenience since together with some approximations, it allowed to get closed form models that could be estimated. This specification of the adjustment costs leads to the following Euler equation :

27Assuming this condition to be satisfied for all current period t, we can set t= 0. This leads to the following (almost) estimable equation :

- 5 Because of the replacement of the expectations by the observed values, one faces a problem of error (...)

28Indeed, to make this equation estimable, we have to use our hypothesis about rational expectations of the firms and to express these expectations in terms of observable variables. Assuming rational expectations, the realized values at time *t ***+**1 in labor input demand can be substituted for the expectations5. The estimated equation then becomes

29Note that with the presence of a rate of exogenous job destruction, *s, *(9) is not directly a function of the absolute variation **(****Δ***L*_{it} ) of employment.

30The previous specification assumes that adjustment costs depend on *HF*_{it}* *only. Then, hiring or firing ten employees in a firm of one hundred workers or in one employing a thousand workers is assumed to generate the same costs. This can be considered as unlikely. As suggested by Whited (1998), it may be preferable to specify an adjustment costs function as depending on relative variations of the workforce :

31This is our second model for adjustment costs. With this functional form, the Euler equation becomes

- 6 In fact the specification proposed by Summers was

32Along the same lines we can follow Meghir et al. (1996) who transposed to labor demand the adjustment cost function initially proposed (for capital) by Summers (1981)6 and defined as

33This specification allows to take into account simultaneously the relative variations and the level of capital stock. Thus in addition to investment, the adjustment cost function now depends on an experience or inertia or size effect relative to the capital input. Transposition to labor demand, that is

34leads to the following Euler equation :

35As it has already been mentioned, the assumption of symmetric adjustment costs appears to be rather restrictive since it implies that hiring and firing costs are equivalent. In order to allow for asymmetry, Pfann and Verspagen (1989) and Pfann and Palm (1993) proposed to define adjustment costs as

36This fourth functional form encompasses the usual quadratic functional form as a special case (*b = *0). The parameter *b *represents the measure of asymmetry of the adjustment costs fonction. Indeed, if *b > *0 (resp. *b < *0), hiring costs are higher (resp. lower) than firing costs i.e., an upward (resp. downward) adjustment is more costly than a downward (resp. upward) movement. The resulting Euler equation is an extension of (6) augmented by two terms in *b. *Indeed we have

37Even if this specification integrates a possible asymmetry in the adjustment costs function, it presents two limits. First, adjustment costs do not depend on the relative hiring or firing with respect to the work force of firm *i. *Second, the effect of experience (or inertia) is not taken into account.

38Considering that it is preferable to use adjustment costs in terms of relative variations than in terms of absolute variations, Bresson, Kramarz and Sevestre (1996) proposed the following functional form :

39This is our fifth adjustment cost function.

40Although this functional form is quite flexible, it does not retain the possibility of an experience/inertia effect like in Bresson, Sevestre and Teurlai (1996). This effect is taken into account in our sixth model with the following functional form :

41This last specification is in accordance with the recent empirical literature showing that the form of adjustment costs may change according to the scale and the way of the adjustments (see e.g. Abowd and Kramarz (2000)). However, it is impossible here to detect the presence of fixed/lumpy adjustment costs. Then, from the contribution of Lundgren and Sjôstrôm (2001), we propose the last following functional form:

42which is the seventh adjustment cost function estimated. This specification has two important differences with respect to those of Lundgren and Sjôstrôm. Firstly, we remove their term -1. Indeed, otherwise, if the exponential fonction is lower than 1, the adjustment cost fonction becomes negative and positively contributes to the profit of the firms. Moreover, having an adjustment cost positive without firing or hiring during one period is not a strong assumption since a personnel department must be maintained for the human resources management. Then, in model no. 7, if α** > **0, a part of adjustment costs is lumpy. Secondly, the adjustments of the workforce enter in our cost function by means of an exponential rather than an absolute value as in Lundgren and Sjôstrôm (2001). This choice is explained by the fact that our specification allows also to identify to which extent the relative variation of hiring and firing is relevant, i.e. if β is near to -y** **. Otherwise, β may be interpreted as an indicator of a learning/size effect (β **< **0) or of a inertia/size effect **(**β **> **0). Of course, we have tested the Lundgren-Sjôstrôm model. Unfortunately with our data, GMM failed to converge.

43The seven Euler equations together with equation (4) form the different systems we have estimated. Before analyzing our econometric results we present below the characteristics of our data.

# II. STATISTICAL DESCRIPTION OF DATA

44Our empirical work concerns the Czech Republic and France. We have used the AMADEUS database which gives, to a certain extent, homogeneous accounting information on firms in different European countries.

45Initially this database included the nine following central and eastern European countries: Bulgaria, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Romania, the Slovak Republic and Slovenia (years 1993 to 1998). Our horizontal (between firms) and vertical (time) trimming and the need for some specific variables for our estimates unfortunately reduced our sample to only two countries : the Czech Republic and Estonia. Firstly, to be more precise, Latvia, Lithuania and Slovenia have disappeared from the sample when we have constrained the value added, the fixed assets, the costs of employees, and the number of employees to be positive and the duration of accounting years to be of 12 months. Secondly, Hungary had to be discarded from our sample because only three observations remained for this country after our first trimming step. Third, Romania and the Slovak Republic have been also removed from the sample because, since we use instrumental-variable estimation methods, we have had to keep only firm-level data with five consecutive years. We are aware that this procedure can lead to a « survival bias » in a sense that we favor the old firms, the new companies aged less than five years getting out of our data base. Fourth, Bulgaria has been suppressed due to the erratic evolution of prices during the 1993-1998 period which induced strange evolutions of the labor productivity. More precisely, Bulgaria has known an hyperinflation episode at the end of 1996/beginning of 1997. In mid-1997, there was a major policy reform : Bulgaria adopted a currency board arrangement, and, in the years after, inflation subsided dramatically. Lastly, Estonia had to be discarded from the sample because only six Estonian firms verified our criterions. Finally, we have thus been led to keep only the data from the Czech Republic.

46Our sample concerns also France during the same period and we have adopted an identical trimming procedure. This procedure is the one proposed by Tukey (1977) which has been taken up recently by Kremp (1995) on a panel of French firms. Tukey’s method is based on the definition of a inter-quartile gap in order to pick out outliers. The trimming procedure has been implemented on the average productivity variables for capital and labor.

- 7 For the French firms this measure is produced by the « DARES » (French Department of Labor) while i (...)

47The variables used for our estimates are defined in Appendix 1 while Tables 1 and 2 provide descriptive statistics about our two samples. The comparison of both tables leads to three comments. First, the average unit cost of labor in the Czech Republic equal 13.5 per cent of that in France. This figure is in accordance with the conclusions of recent contributions (see e.g. Belke and Hebler (2000, p. 225)). Secondly, we observe that, during the 1993-97 or 1994-98 periods, French firms have globally hired (although slightly in absolute terms) whereas the Czech firms have known a (substantial) decline of their workforce. Indeed, both in terms of absolute and relative variations, employment variations observed in the Czech sample are significantly higher than those observed in the French sample. This fact will influence greatly our estimates of labor demand models with adjustment costs as we shall see later on. In the third place, taking into account a measure of the exogenous « natural » quit rate *(s) *induces larger individual employment variations for the Czech firms than for the French ones.7 One possible and certainly the most plausible explanation is a « pull » effect of the new sectors which offer higher wages and therefore attract workers who happily leave the low-paying old-sector jobs (Jurajda and Terrell (2002)).

# III. THE ESTIMATION RESULTS

- 8 All estimations have been realized in levels.

48Tables A-2 and A-3 (in Appendix 2) present the different GMM estimates obtained for the seven theoretical functional forms outlined above. Here, the discount rate, 6 , may be constrained to be equal to 0.05 or be free. In the same line, we consider two variants of our models depending on whether a rate of natural wastage *(s) *is introduced in the hiring and firing adjustment cost functions. The different Euler equations depend also on the marginal productivity of labor which is a fonction of the output elasticity to labor. Before estimating the parameters of the adjustment costs fonctions, we have estimated the Cobb-Douglas production fonction for both samples (see Table A-1 in Appendix 2). The GMM estimates of the elasticities of output to labor have then been used for the calculation of *marginal *productivities of capital and labor.8

49The technology available in both countries can be characterized by constant returns to scale as confirmed by the Wald test. This result is in accordance with recent empirical papers with panel data for France (see Blanchard et al. (1996)) as for eastern European countries (see Korôsi (2001)). Under the assumption that the markets are not too imperfectly competitive, the elasticity of each production factor is the share of its remuneration in the total value of the production. The share of capital (labor) remuneration is higher (lower) in the Czech Republic than in France. This result can be interpreted as a differential of factor endowment between the two countries : France, being better endowed in capital than the Czech Republic.

- 9 The model no. 4 with
*s*= 0 and θ = 0.05 is not consistent for the Czech Republic because for low va (...)

50While we have quite similar characteristics of the production functions for France and the Czech Republic, there are important differences between both countries as regards the adjustment costs functions. Sticking to the models for which all coefficients are statistically significant and for which the Sargan statistics does not lead to a rejection of the specification, no models are unfortunately comparable between the two countries. In fact, the adjustment costs are either zero or negative (inconsistent) for the Czech Republic, contrary to France. The only case for which they are significantly positive but very low in the Czech Republic is for quadratic form and for a constrained discount rate (model no. 1 with *s > *0 and **6= **0.05 )9. For this model in France, the *c *estimate appears however insignificant. Then, in the Czech Republic the labor adjustment costs do not seem to exist or to be very low. This conclusion is important since it leads to think that the effective level of rigidities present on the Czech labor market is not necessarily very strong. This result is a bit surprising as this market is often considered as one of the less flexible eastern and central European labor markets. Again, one possible explanation of this result is that our sample comprises firms in which employment variations have been rather important over the period, both in absolute and relative terms, which mechanically induce low adjustment costs.

51On the French side, the estimates give very different values of the parameters according to the specification used. This heterogeneity can be explained by three factors: the way the variation of the workforce is taken into account, either in relative or absolute terms; the introduction of an experience/inertia effect or not; the degree of asymmetry of the firing and hiring costs. More problematic is the heterogeneity due to the great positive time preference of the firms. Indeed, when this parameter **(**θ) is unconstrained, we get an infinite estimate at least for the symmetric adjustment cost functions (models no. 1 to 3). Moreover, we can note that *c *is systematically lower when θ **= **0.05 than when θ is unconstrained. This last result is consistent with the fact that with a great positive time preference, the firms adjust closely to the exogenous demand and supply shocks and then support higher adjustment costs (see Hamermesh and Pfann (1996)).

52In our first three models, the adjustment costs functions are quadratic. Under this assumption, the dynamics of labor demand may be approximated by a partial adjustment process. Our estimates of this last model are summarized in Table A-4. They are not so different from those obtained in previous empirical works (see e.g. Dormont and Sevestre (1986), Hamermesh (1990), Bresson, Kramarz and Sevestre (1996)). Note that the labor adjustment process is quicker in the Czech Republic than in France. This result confirms our previous conclusions about the differences in adjustment costs.

53However, as mentioned above, the adjustment costs must be represented by a more general form than a quadratic function. This is largely verified by the econometric estimates of our seventh model. Thus, a part of adjustment costs is lumpy. These costs depend on an experience effect rather than an inertia effect (β is negative whatever the assumption on *s). *At last, they are asymmetric with respect to the direction of the workforce adjustment (γ is significative). In fact, it is more expensive to hire than to fire (γ **> **0). These estimates are globally close to those obtained by Lundgren and Sjôstrôm (2001) using Swedish panel data.

54In order to give a better appreciation of the importance of the adjustment costs, we have calculated in Table A-5 (Appendix 2) the marginal costs of adjustment in percent of labor unit cost. Of course we have restricted our calculations for France because the calculation of marginal costs of adjustment is not relevant for the Czech Republic. For France we can retain only six models and then calculate the marginal costs of adjustment. These models are model no. 1 **(**θ unconstrained, *s = *0 ), model no. 2 **(**θ unconstrained, *s > *0 ), model no. 3 **(**θ unconstrained, *s = *0 ), model no. 3 **(**θ unconstrained, *s > *0 ), model no. 7 **(**θ **= **0.05 , *s = *0 ) and model no. 7 **(**θ **= **0.05 , *s > *0 ). Two main conclusions can be drawn from our results. Firstly, marginal costs are rather weak with a maximum of 3.13 months of labor unit cost. This level is low relatively to the results of Lundgren and Sjôstrôm (2001) but it is more in accordance with the French case (see e.g. Blanchard et al. (1996)). Secondly, using the most flexible model (i.e. model no. 7), the experience effect leads to create unambiguous decreasing marginal costs of labor adjustment. Thus, at margin, it is less expensive to hire or to fire for a big size firm than for a low size firm, other things being hequal.

# CONCLUSION

55A comparison of the structure of labor-demand adjustment costs between France and the Czech Republic shows some differences. More precisely, our econometric estimates suggest that the dynamics of employment depends on non-trivial adjustment costs in France while these costs appear to be very weak or even non-existent in the Czech Republic. From this viewpoint, the labor-market structures for both countries appear different.

56For France, we have found that the adjustment costs have a fixed part and are asymmetric. To a large extent, they also exhibit a scale effect (which we interpret as an experience effect). The marginal adjustment costs (a maximum of 3.13 months of labor unit cost) appear to quite weak but are rather in accordance with other French studies.

57For the Czech Republic, the period of our study (1993-1998) leaves the first crucial years of transition uncovered. However our data on workforce movements show that the transitional process was still at work in the mid-1990s. Then our results show that restructuring would have not been thwarted by the high adjustment costs brought about by possible transitional difficulties. However, we consider our work as a first and partial analysis of the issue. Of course, our conclusions have to be confirmed by future studies which could use larger databases.

## Bibliographie

# REFERENCES

Abowd J. M. and Kramarz F. (2000), « The costs of hiring and separations », CREST *Working Paper *9543, January, 39 pages.

Belke A. and M. Hebler (2000), « EU enlargement and labour markets in the CEECs », *Intereconomics, *35, 5, Sept./Oct., 219-230.

Blanchard P. et al. (1996), « Coûts d’ajustement de l’emploi, structures de marché et gestion de la main-d’œuvre : Analyses théoriques et estimations économétriques sur données d’entreprises », Rapport au Commissariat Général du Plan, June.

Bresson G., F. Kramarz and P. Sevestre (1992), « Heterogeneous labour and the dynamics of aggregate labour demand : some estimations using panel data », *Empirical Economics, *17-1, 153-168.

Bresson G., F. Kramarz and P. Sevestre (1996), « Dynamic labour demand models », in Matyas and Sevestre (eds.), *The Econometrics of Panel Data : A Handbook of the Theory with Applications, *660-684.

Bresson G., P. Sevestre and J.-C. Teurlai (1996), « The specification of labour adjustment costs : A microeconomic comparative study », paper presented to the 13èmes Journées de Microéconomie Appliquée (JMA).

Dormont B. and P. Sevestre (1986), « Modèles dynamiques de demande de travail : spécification et estimation sur données de panel », *Revue Economique, *55, 3.

Eisner R. and R. Strotz (1963), « Determinants of business investment », in Commission on Money and Credit, *Impacts of Monetary Policy, Part I, *Englewood Cliffs, Prentice Hall, chapter 2.

Hamermesch D. S. (1990), « Aggregate employment dynamics and lumpy adjustment costs », NBER Working Paper No. 3229.

Hamermesh D. S. and G. A. Pfann (1996), « Adjustment costs in factor demand », *Journal of* *Economic Literature, *34, 3, 1264-1292.

Hansen L. P. (1982), « Large sample properties of generalized methods of moments estimators », *Econometrica, *50, 1029-1054.

Hansen L. P. and K. J. Singleton (1982), « Generalized instrumental variables estimation of nonlinear rational expectations models », *Econometrica, *50, 1269-1286.

Jurajda S. and K. Terrell (2002), « What drives the speed of job reallocation during episodes of massive adjustment ? », CEPR Discussion Paper No. 3212, February, 42 pages.

Korôsi G. (2001), « Labour adjustment and efficiency in Hungary », Working Paper, Institute of Economics, Budapest.

Kremp E. (1995), « Nettoyage de fichiers dans le cas de données individuelles : Recherche de la cohérence transversale », *Economie et Prévision, *119, 1995-3, 171-193.

Lundgren T. and M. Sjôstrôm (2001), « A Flexible Specification of Adjustment Costs in Dynamic

Factor Demand Models », *Economics Letters, *72, 2, August, 145-150. Maccini L. J., (1987), « Adjustment costs », entry in Eatwell et al. (eds.), *New Palgrave : A Dictionary of Economics.*

Meghir C., Ryan A. and J. van Reenen (1996), « Job creation, technological innovation and* *adjustment costs : Evidence from a panel of British firms », *Annales d’Economie et de Statistique, *0, 41-42, Jan.-June, 255-274.

Meese R. (1980), « Dynamic factor demand schedules for labor and capital under rational* *expectations », *Journal of Econometrics, *14, 1, Summer, 141-158.

Nickell S. (1986), « Dynamic Models of Labour Demand », in *Handbook of Labor Economics, Volume ***1, **Edited by O. Ashenfelter. and R. Layard, Elsevier Science Publishers, 473-522.

Noorkoiv R., P. F. Orazem, A. Puur and M. Vodopivev (1997), « How Estonia’s economic* *transition affected employment and wages (1989-95) », Policy Research Working Paper,* *no. 1837, World Bank.

Oi W. Y. (1962), « Labour as a Quasi-Fixed Factor », *Journal of Political Economy, *70, 6, December, 538-555.

Pfann G. A. and F. C. Palm (1993), « Asymmetric adjustment costs in non-linear labour demand models for the Netherlands and U.K. manufacturing sectors », *Review of Economic Studies, *60, 2, 397-412.

Pfann G. A. and B. Verspagen (1989), « The structure of adjustment costs for labour in the Dutch manufacturing Sector », *Economics Letters, *29, 4, 365-371.

Whited T. (1998), « Why do investment Euler equations fail ? *», Journal of Business and* *Economics Statistics, *16, 4, October, 479-488. Summers L. H. (1981), « Taxation and corporate investment : A q-theory approach », *Brookings* *Papers On Economic Activity, *1, 1981, 67-127.

Tukey J. (1977), *Explorating Data Analysis, *John Wiley.

## Annexes

# APPENDIX 1 : DEFINITION OF VARIABLES AND STATISTIC

Annually firm-level data ; year *t *(1996 or 1997)*VA *: Added value, in USD.*OPERPL *: Gross operating surplus, in USD.*TM *: = *OPERPL/VA *; proxy of price cost margin ( *-ms*_{it}*/ε*_{QD/P}* *).*COSTSE *: Costs of employees, in USD. *NUME *: = *L *; number of employees.*LL *: = ln *L *.*W *: = *COSTSE/NUME *; labor cost (wage plus payroll taxes) ; 1995 th USD.*PRIXVATDC *: Added value constant price index, at 1995 prices, in USD. *PRIXFBCFTDC *: Gross fixed capital formation price index, at 1995 prices, in USD.*FIXEDAS *: Fixed assets, in USD.*K *: = *FIXEDAS/PRIXFBCFTDC *; stock of capital.*LK *: = ln *K *.*PL *: = *(VAPRIXVATDC*)*/L *; average productivity of labor.*PK *: = *(VAPRIXVATDC*)*/K *; average productivity of capital.*INTK *: = *KL *; capital-intensy. *Q *: = *VA PRIXVATDC *; individual level of production.*LQ *: = ln *Q *.*CR *: = *W PRIXFBCFTDC *; labor/capital relative cost.*S *: = rate of exogenous job destruction (natural wastage).

Table 1 - Statistic on the French labor demand (1761 firms, 1993-97 or 1994-98, annual data) a in percentage.

Table 2 - Statistic on the Czech labor demand (298 fïrms, 1993-97 or 1994-98, annually data)

Note aa

# APPENDIX 3 : ECONOMETRIC RESULTS TABLES

## Notes

1 When firm *i *hires, *HF*_{it}* > *0 and, alternatively, if it fires, *HF*_{it}* < *0 . In the same vein, if *I*_{it}* > *0 , some investment is realized while for *I*_{it}* < *0 , the firm disinvests.

2 We will assume that, at any date, a constant and exogenous proportion *s *e [0,1[ of worker leaves the firm. This assumption aims at distinguish the regular and planned rotations of labor (e.g. retirements) from redundancies. The first ones cause costs which are mainly linked to new hires they involve while the second ones add to them costs which are characteristic of redundancies. That is why we assume that separations from redundancy and all hires involve, at any date, a adjustment cost *(AC).*

3 Such a hypothesis implies that cross-adjustment costs between labor and capital are low or non-existent. We will make this assumption for all the adjustment cost functional forms we have considered.

4 Which will be log-linearized for estimation.

5 Because of the replacement of the expectations by the observed values, one faces a problem of error in variables. Then, we have to use instrumental variables or GMM estimation techniques, where instruments belong to the information set of the firm at time t.

6 In fact the specification proposed by Summers was

It was then asymmetric and discontinuous but Meghir et al. (1994) have adapted it as shown.

7 For the French firms this measure is produced by the « DARES » (French Department of Labor) while it comes from Jurajda and Terrell (2002, Table A.3, p. 39) for the Czech firms.

8 All estimations have been realized in levels.

9 The model no. 4 with *s *= 0 and θ = 0.05 is not consistent for the Czech Republic because for low values of hiring or firing the total adjustment costs may be negative.

## Notes de fin

a *in percentage*.

b *t-values are in brackets.*

c *t-values are in brackets*

d *t-values are in brackets.*

e *GMM fail to converge.*

f *t-values are in brackets.*

g *GMM fail to converge.*

h *t-values are in brackets.*

i *t-values are in brackets.*

j *GMM fail to converge.*

k *t-values are in brackets.*

l *GMM fail to converge*

m *t-values are in brackets*

## Table des illustrations

© Presses universitaires François-Rabelais, 2004

Conditions d’utilisation : http://www.openedition.org/6540