A jont natve of udwg-maxmlans Unversy s Center for Economc Studes and the Ifo Instute for Economc Research Ifo / CESfo & OECD Conference on Regulaton: Polcal Economy, Measurement and Effects on Performance 9 30 January 00 CESfo Conference Centre, Munch Temporary Workers and Seasoned Managers as Causes of ow Productvy Francesco Daver and Mara aura Pars CESfo GmbH Phone: 49 0 89 94-40 Poschngerstr. 5 Fax: 49 0 89 94-409 8679 Munch E-mal: offce@cesfo.de Germany Web: www.cesfo.de
Ths draft: January 5, 00 Temporary workers and seasoned managers as causes of low productvy Francesco Daver Unversà d Parma, Iger and CESfo Mara aura Pars Unversà d Bresca Abstract We employ company mcro data to study the determnants of low productvy growth n the Italan economy. We show that the acute productvy slowdown wnessed n the last few years has manfested self n parallel wh two factors seemngly stfg nnovaton: the presence of a hgh share of temporary workers and the old age and hgh senory of top managers and members of the boards. These results may be seen as provdng elements consstent wh both a labor supply vew and a labor demand vew of Italy s productvy slowdown. An earler draft of ths paper was presented n Ax-en-Provence. We are grateful to Glbert Cette and our dscussant Remy ecat for ther useful comments.
. Introducton After decades of fast growth that led Italan lvng standards to converge towards European levels by the end of the 980s, GDP growth n Italy has markedly slowed down snce the md 990s. Fgure below shows that, snce 995, Italy s GDP has cumulatvely lost some twenty percentage ponts vs-à-vs the GDP of the other four bg countres n the European Unon Germany, France, the UK and Span. Hence Italy s prolonged growth slowdown has been a mystery to explan long before the Great Recesson of 008 and 009. Fgure. - GDP growth n Italy and n the other bg four European countres, 995-008 Source: Francesco Daver, Italy, before and after ehman Brothers, VoxEu.org, June 5, 009 Italy s decng growth performance the contnuaton of a secular trend only partally shared wh other OECD countres - has been the combnaton of two dstnct though as we wll dscuss possbly nterrelated - labor market developments. There has been a marked dece - from.% n 980-95 to 0.5% after 995 - n the growth rate of labor productvy. Yet ths productvy slowdown has been paralleled by a marked ncrease n the growth rate of total hours worked, whch turned from negatve to posve fgures. Both developments have been que at odds wh what had happened n prevous decades. Table - Decomposng Italy s per-capa GDP growth, 970-004 Growth rates of Per-capa GDP GDP per Hours per workng Workng age populaton hour worked age person over total populaton 970-80 3. 3.9-0.8 0.0 980-95.8. -0.7 0.4 995-04.3 0.5.0-0. Source: Daver and Jona asno 005 As shown n Table, productvy started s stagnaton after 995 n both manufacturng and
servce ndustres. Yet, whle the presence of stagnatng or even decng productvy s not surprsng n the servces ndustres, especally n a country lke Italy whose servces are stll plagued wh extensve monopolstc provson, the partcularly strkng features of these developments has been the zerong of productvy growth n the Italan manufacturng ndustres. Ths s notable because manufacturng ndustres have been the productvy-leadng ndustres n Italy n the past decades. Ths decng path has been partcularly pronounced durng the 00-03 aggregate slowdown, when labor productvy growth has gone down below zero, but was there even n 995-00 when the economy was dong relatvely well. Italy s productvy slowdown s not the consequence of unfortunate busness cycle fluctuatons. Table - Growth of labor productvy n Italy, 970-003, man ndustry groups 970-80 980-95 995-03 995-00 000-03 Economy.4.8 0.6. -0. Agrculture 3. 4.3.7 5. -.5 Manufacturng.8 3.0 0..0 -.0 -- non-durables.7 3. 0.3 0.7-0. -- durables.9.7 0.0.7 -.7 Utles -0.4 0.8 5.5 3.7 8.7 Constructon.9.0 0. 0.5-0.5 Busness sector servces.8. 0. 0.5-0.5 Publc servces.4 0.7 0.4 0.8-0. Source: Daver and Jona-asno, 005 In ths paper, we take a look backwards to 00-03, a perod of partcularly acute slowdown n the Italan economy and less so elsewhere n Europe, to learn about the possble determnants of productvy developments at the frm level. Our mcro data evdence ndcates that productvy growth was partcularly low n frms wh dsproportonately hgh shares of temporary workers to start wh. Ths result s robust to all changes of specfcatons. Yet decng productvy s also slghtly less robustly - assocated to such manageral features as the average age and senory of the managers runnng the companes. Our estmates separately run for nnovatve and non-nnovatve companes ndcate that senory do not seem to play a major role, whle an older age of managers s assocated wh lower productvy n nnovatve frms and wh hgher productvy n non-nnovatve frms. Ths s consstent wh common sense that suggests a more posve role of experence n frms wh relatvely standardzed and stable busness practces, whle old age and senory s presumably more damagng for nnovatve frms that would be supposed to swftly adopt new technologes as they become avalable. Ths s presumably more tghtly correlated wh schoog and less wh experence on the job. Our -stage model results, whle confrmng the robustness of the partal correlaton between the share of temporary workers and productvy, shows that both product and process nnovaton are posvely related wh productvy growth. For the nnovatve frms age s negatvely correlated wh productvy growth and zero or posvely correlated wh non nnovatve frms, throughout all estmaton methods used.
The relaton between manageral senory and productvy s nstead less tghtly correlated than whn the OS framework. The correlaton between senory and productvy vares consderably across specfcatons. It appears always non correlated for nnovatve frms, whle the correlaton s ambguous for non nnovatve frms and depends on the estmaton method beng used. Our paper s structured as follows. In secton, we present the two man competng explanatons for Italy s productvy slowdown n words. In secton 3 we present the paper s conceptual underpnnngs and a theoretcal model of manageral capal and productvy that nests the two hypotheses. In secton 4 we llustrate our econometrc framework. In secton 5 and 6 we descrbe the man features of the data set and our estmaton results. Secton 7 concludes.. Italy s productvy slowdown: two explanatons Ths sudden productvy turnaround wnessed n the Italan economy after the md 990s begs an explanaton. A vew shared by many Italan entrepreneurs s the so called abor Supply vew consstent wh the deas expressed by Robert Gordon and Ian Dew-Becker 008 on the nterrelaton between labor market reform and productvy developments n Europe. In ther paper, they conjectured that the process of - mostly pecemeal - labor market reform that occurred n many European countres, whle helpng Europe reverse the past tendences towards job destructon, has been eventually detrmental to productvy growth. Smplfyng ther vew to an extreme, f labor market reform occurs and labor demand does not shft, labor supply shfts to the rght along a gven labor demand curve. No wonder that productvy deces as a result. Ths may have occurred n Italy as well, n the wake of the changes n labor market legslaton n favor of more flexbly ntroduced by the Prod Government n 997-98. These legslatve changes gave full legal recognon to a host of contractual forms of part-tme and temporary jobs, some of whch had been n place even before though restrcted to the unoffcal labor market. The flurry of cheap labor orgnated from such half hearted labor market reforms has resulted n a dece of the equlbrum capal-labor rato. It may have also dscouraged the nnovatve ably of many entrepreneurs who found themselves confronted wh the hard-to-resst temptaton to adopt technques ntensvely usng the part-tme workers now abundantly avalable n the labor market. Yet the labor supply vew s not the only game n town. Another potental explanaton would stress the role of labor demand, and notably of the nterplay between a changng world economy and a largely unchanged domestc manageral envronment. The majory of Italan frms, typcally agglomerated n ndustral dstrcts or nvolved as sub-contractors by large-scale exportng companes, smply contnued ther prevous busness conducts. Yet the mode and the opportunes of competon around the world had changed meanwhle. There was a technologcal revoluton out there whose frus could have been reaped by those companes adoptng the new technologes as well as adaptng ther manageral technques to the changed envronment. Yet the Italan economy was not well equpped for whstandng competon n ths changng world. In a recent paper, Bandera, Guso, Prat and Sadun 008 have analyzed the ncentve structure that Italan managers face, ther career profle and even ther use of tme. It turns out that only a fracton of frms especally the non-famly owned and multnatonals - adopts a performance-based model, whereby managers are hred through formal channels such as busness contacts and head-huntng actves, undergo regular assessment procedures and are rewarded, promoted and dsmssed on the bass of ther assessment results. Most frms partcularly the famly-owned ones and those manly actve n the domestc market - follow a fdely model of
manageral talent, hrng ther managers based on personal or famly contacts, whout formally assessng ther performance and promotng and rewardng them accordng to the qualy of ther personal relatons wh the frm s owners. The problem s that the type of manageral model based on performance or fdely s tghtly assocated wh the qualy, the conduct and the performance of managers as well as of the frm self. And here then comes our second explanaton of Italy s productvy slowdown: frms endowed wh managers of such poor qualy are presumably at a dsadvantage when faced wh new technologcal opportunes wh respect to foregn competors less dependent on famly-based modes of runnng a frm. Whle there may be some grans of truth n both vews, Gordon and Dew-Becker have not contrasted ther deas wh mcro data, whle Bandera, Guso, Prat and Sadun have not looked at the nteracton of labor market reform wh the productvy and nnovaton counterpart of manageral practces. So there s room for comparng a streamed verson of the two vews aganst the backdrop of the same data set. Ths s the man goal of our paper. If the abor Supply vew s correct, we expect to see n the data more pronounced productvy deces n companes wh dsproportonately hgher shares of part-tme and temporary workers at the begnnng of the perod. If the abor Demand vew s correct, we expect to detect stronger productvy deces n companes wh older and/or senor managers. Ceters parbus, older CEOs, board members and managers are possbly less prone to ntroduce new technologes and nnovaton, because ths may change ther routnes and more mportantly f ths changes the crculaton and transmsson of nformaton whn the frm and eventually the structure of power whn the frm. Yet age as such may not be the crucal or the only element to consder n ths respect. As emphaszed by Daver and Malranta 007 for a large sample of Fnnsh manufacturng companes and workers, senory may be the really damagng feature, partcularly n hgh-tech companes where a certan attude towards change s more lkely related to short tenure and lack of entrenchment n a gven company. Both explanatons may have somethng to say on Italy s productvy dece and perhaps more generally. We smply construct our thought experment to let the data speak on ths. 3. Manageral effort, busness schoog and frm productvy Before delvng nto our emprcal analyss, we make our workng hypotheses more precse whn a logcally coherent two-perod producton functon framework. The manager sde A manager lves for two perods. Today she can eher exert manageral effort work or go to school - a busness school, where she learns novel manageral technques to be employed at work tomorrow. Tomorrow she cannot but work for, n a two-perod framework, there s no pont to go a busness school any more. If workng e>0 she earns a compensaton v n each perod. Hence her lfetme ncome equals v. Ths ncome fnances lfetme consumpton. Wh zero dscount and nterest rates and standard preferences, the workng manager wll smoothly consume c e *= v ε/ n each perod. If nstead she decdes to go to the busness school today, she postpones workng to tomorrow and e=0. In the meanwhle, she learns new busness technques, so that she accumulates manageral
capal b>0. Tomorrow, endowed wh the new busness technques b, she wll earn a compensaton w>v. Then, the lfetme ncome of an educated manager s w. Agan, wh zero dscount and nterest rates and perfect access to cred markets, consumpton smoothng wll preval under standard preferences, so that c b *= w/ n each perod. Note that the educated managers are young n the labor market, for they have no work experence, whle the uneducated managers are old.e. experenced but unaware of new busness technques. The frm sde Frms are born all alke. Then R&D nvestment or abundance of cash flow or other crcumstances enhancng the chance that a frm becomes nnovatve may materalze or not. If any of such crcumstances takes place, the frm becomes an nnovatve frm an I-frm. If R&D does not take place, the frm stays non-nnovatve and becomes an N-frm. The dfference between the two types of frms stems from ther producton functon. The producton functon of each N frm s as follows: Y N = A N K N α N β where α>0, β>0, 0<αβ<, K N and N are respectvely the stock of capal and the number of workers employed n each frm N, A N s the scale parameter n the producton. The parameter A N depends on past manageral effort as follows: A N =E N,- -α-β wh E N =Σe N. In words, the effcent level of producton n frm N s a functon of capal, labor and manageral capal. In turn, manageral capal s accumulated through manageral effort under dmnshng returns. Wh ths producton functon, labor productvy y N = Y N / N s: a decreasng n N and b ncreasng n E N,- The nnovatve frms are dfferent from the non-nnovatve ones for the qualy of manageral effort beng requred. They employ capal and labor as well as tradonal and new manageral technques. Tradonal manageral technques requre experence, whle new manageral technques are those accumulated through busness schoog. Effcent producton occurs through the followng producton functon: Y I = A I K I α I β In the schoog case, access to cred market certanly makes a dfference for the consumpton path. If cred market constrants prevent or restrct borrowng to fund her schoog of today, c * > c *,.e. the consumpton path of the manager wll be upward slopng. We do not study emprcally the consumpton path mplcaton of the manageral decson.
where α>0, β>0, 0<αβ<, K I and I are respectvely the stock of capal and the number of workers employed n each frm I, whle A I s the effcency parameter n the producton. A I s a functon of past manageral effort and busness schoog as follows: A I =E I,- -α-β B wh E I =Σe I and B=Σb, where B s manageral capal accumulated at the busness school n the prevous perod by those who are employed as managers n I-frms. In words, the effcent level of producton n frm I s a functon of capal, labor and manageral capal. In turn, manageral capal s accumulated through manageral effort under dmnshng returns and through busness schoog under constant returns. Ths s dfferent from the N-frms: for an I-frm, beng endowed wh a young Wh ths producton functon, labor productvy y I = Y I / I s: a decreasng n I and b ncreasng or decreasng n E I,-, dependng on whether the posve effect of work on A I more than offsets the negatve effects from the foregone accumulaton of busness school capal entaled by the decson to work nstead of gong to the busness school Equlbrum allocaton of managers Each frm hres managers wh the goal of maxmzng s profs. For each frm N and I ths entals equalzaton of margnal productvy of experenced managers to ther compensaton v. In turn, for frm I, prof maxmzaton also entals equalzaton of the margnal productvy of experenced managers to the margnal productvy of educated managers, whch has to be equal to w. The followng lst of equlbrum condons follow: -α-βe N,- -α-β K N α N β = v -α-βe I,- -α-β K I α I β = v K I α I β = w M= B ne N,- E I,- The latter of whch s a condon of full employment of managers, wh M beng the total number of managers, the total number of I frms and n the total number of N frms. The four condons above determne, up to a gven level of K and, the hrng of uneducated and educated managers n the two groups of frms and the busness schoog premum w/v. Emprcal mplcatons Summng up, the model s mplcatons that we wll explo n the emprcal part of our paper are the followng:
a Frms select themselves nto nnovatve and non-nnovatve ones dependng on some exogenous crcumstances such as the occurrence of R&D spendng, the presence of enough cash flow and others to be specfed. b Experence s good for labor productvy n the non-nnovatve frms and eher good or bad for productvy n the I-frms. We wll use manageral senory as a measure of manageral experence whn the frm. Manageral age wll be a measure of overall manageral experence over and above the experence ganed whn a gven frm. c abor market lberalzaton and other measures that tend to favor the entry of unsklled workers s equally bad for labor productvy n both types of frms. 4. Econometrc framework Based on the model n secton 3, our specfcaton to estmate the effect of board members mean age, average years of senory and the fracton of temporary workers on labor productvy growth s as follows: Y K Temporary = α β γage λsenory µ ε The dependent varable s the long-growth rate of labor productvy for frm at tme t = 003, wh respect to 00. Age s calculated as the average age of the board members when they were nomnated. Senory s calculated as the average wh respect to all the frm board members number of years n the board as of 00 the nal year of our sample. Temporary/ s the share of workers n the frm operatng on a temporary contract full tme part tme as of 00. In the regressons we control also for sectors of producton based on the Ateco007 code, geographcal areas, sze dummes and frm membershp n a group. We run a Chow test of parameter nstably on ths specfcaton equaton because we expect the parameters β,γ,λ, µ to dffer between nnovatve and non nnovatve frms. Innovatve frms ntroduced a product or process nnovaton or both n the three-year perod consdered 00-003. Non nnovatve frms have never ntroduced nnovatons n the perod. The test suggests that the two groups of frms react dfferently to age, senory and temporary workers share gven all the other controls. The p-value of the F-test s n fact equal to zero when the nnovaton dummes refer to product nnovaton only or process nnovaton only, or both see Table. For ths reason, we allow the parameters of nterest to vary across groups, accordng to equaton : Y = α Temporary µ K βd β D γ Age * D γ Age * D Temporary * D µ * D ε λsenor * D λsenor * D The dummes D and D dentfy the two groups of frms D = f nnovatve and D = f non nnovatve. D wll be omted n the regresson because of coleary. We perform Wald tests of parameter nstably for the H 0 hypotheses:
γ = γ H : 0 λ = λ µ = µ and report p-values n the results. Furthermore, we calculate the sem-elastcy of labor productvy wh respect to age, senory and the share of temporary workers, for the whole sample. The second step forward consders the formaton of the nnovaton groups as endogenous. The dea s that frms ntroduce nnovatons because they are more productve, young or ntensve at nvestng nto R&D actvy or nnovatve capal, or maybe because they have more cash flows. The new specfcaton for labor productvy growth s thus a typcal case of swchng regresson model wh endogenous swchng as explaned frstly n Maddala 983. We want to consstently estmate the parameters n two regmes: whether frms are nnovatve regme or non nnovatve regme over the perod of observaton. Ε Ε Y Y = α = α ε nnovatve 0 ε non nnovatve K K 0 β γ Age β γ Age λ Senor λ Senor Temporary µ Temporary µ ε ε f nnovatve f non nnovatve 3 The dstrbuton of the error term ε j j=, can be assumed normal wh zero mean and constant varance σ j. We shall modfy ths strong assumpton n the robustness estmatons. The creron functon to determne whether a frm belongs to regme s the followng: D = f ' ω > 0 D = 0 otherwse 4 To estmate the parameters observe that ED =PD ==P ω >0 s the probably of beng an nnovatve frm. If the error term ω s such that Eω =0 and Vω = the frst stage estmaton method s the prob M. The varables are nstruments correlated wh the decson of ntroducng nnovatons, such as R&D expendure at the begnnng of the sample perod, the number of R&D workers, age of the frm and cash flow at the begnnng of the perod plus the usual controls of equaton. Usually equaton 3 s estmated separately for the two regmes, whose dosyncratc errors are correlated wh ω. In ths case we need to ntroduce correctons for the error condonal mean as n 5 and 5 : φ ' Ε εω = σ εω 5 Φ ' ε D = = Ε σ ω ' ω > 0 D 0 = = D.
' ' 0 ' D φ σ ω ω σ ε ω ε ω ε Φ = = Ε = Ε 5 where φ and Φ are, respectvely, the standard normal densy and cumulatve dstrbuton functons. Notce that the condonal dstrbuton of ε j j=, n equaton 3 s normal wh mean Eε j ω = σ εjωω and varance Vε j ω = σ j -σ εjω. Φ = = Φ = otherwse u ' ' D f u ' ' Temporary Senor Age K Y Temporary Senor Age K Y φ σ µ λ γ β α φ σ µ λ γ β α ω ε ω ε 6 Ths system of equatons can be estmated consstently through the OS method second stage after substutng the estmate of n the correcton terms derved from the frst stage. Followng Maddala 983 we can smplfy equaton 3 smultaneously and addng a correcton term,.e. the sum of 5 5, for the whole sample. [ ] ' ' ' ' ' ' 0 * 0 * Temporary Senor Age Temporary Senor Age K D P D Y D P D Y Y φ σ µ λ γ β φ σ µ λ γ β α ω ε ω ε Φ Φ Φ Φ = = = Ε = = = Ε Ε 7 Rearrangng terms n 7, we obtan an estmable specfcaton 8 whch allows us to perform Wald tests of parameters nstably. When the coeffcents of the nteractons are equal to zero, ths procedure s a convenent way to mpose cross-equaton restrctons on the two-regme specfcaton: Temporary Senory Age Temporary Senory Age K Y ξ σ σ φ µ µ λ λ γ γ β β µ λ γ β α ω ε ω ε Φ Φ Φ Φ = ˆ' ˆ' ˆ' ˆ' ˆ' 8 where ξ has a standard normal dstrbuton. When none of the parameters vary across groups, the equaton reduces to the smple orgnal form :
Y K Temporary = α β γage λsenory µ ε whch s the constraned equaton constrant beng equal parameters n all groups to be used n the Chow test. WAD OR CHOW? Apart from the endogeney of capal-labor rato, the problem of endogenous group formaton n the latter case dsappears. As far as the endogeney of the capal stock s concerned, capal accumulaton depends bascally on past nvestment ntenses and nal levels of capal/labor rato, condonal on the sze, sector and group of the frm, whether or not has ntroduced nnovatons. For ths reason we wll also apply an IV method to the system of equatons 6, where the growth rate of the capal/labor rato s nstrumented. 5. The dataset We collected balance sheet data for Italan manufacturng frms and ther board characterstcs n the perod 00-003 from two sources of data. Informaton about employment characterstcs, nnovaton actvy and R&D nvestment at the frm level come from the IX th Survey on Manufacturng Frms by Capala/Uncred. Ths survey has been run n 004 through questonnares dstrbuted to 477 frms. The questonnares nqure about locaton, legal form, group, sales, nvestments, R&D nvestments, nnovaton actvy, exports, labor force characterstcs, fnancal status and ncentves, balance sheets. Most of the quantatve nformaton relates to the prevous three years snce the tme of the survey, separately. Some qualatve answers, nstead, are related to the whole three-year perod,.e. nnovaton actvy. Informaton about balance sheets, age and senory of the board members come nstead from the AIDA database. 3 AIDA s updated every week but mantans balance sheet data for the prevous years as well. Thus we extracted balance sheet ems over the 00-003 spannng to check and correct for nconsstences between the two sources. Incdentally, our chosen sub-perod - the years between 00 and 003 - happens to be a perod durng whch Italy s productvy shortfall has been partcularly severe. Whle we can extract balance sheet from AIDA n the years of nterest to match the two sources, the database regsters just the latest board composon, whch means that we access nformaton about board members of exstng frms on December 3, 007. We know the year at whch the person was nomnated but the duraton of the servce s practcally unknown, and her/hs functon s scarcely avalable. Thus we take the board composon as f was present n 00-003 as well. For each member wh avalable data, we calculate her/hs age at the tme of the nomnaton whn the board, as well as the age n the years 00, 00, 003, the duraton of the servce.e. senory n 00, 00, 003 and for each frm we may calculate an average age and senory of the board. We excluded from the dataset those frms whose board name appeared to be another company, not a physcal person. 3 AIDA s managed by Bureau Van Djk. It collects balance sheets, propretary shares, frm characterstcs and board characterstcs on about 50000 Italan frms. We accessed the data on frms wh at least 800.000 gross sales as of 3 December 007.
Table 3 - Descrptve statstcs of the man varables of nterest Frm level Observatons Mean St.Dev. Mn Max Product 8 45.7 49.83 0 Process 8 48.5 49.99 0 Product & Process 8 9.0 45.39 0 R&D spendng yes/no 78 59.4 49. 0 Incorporaton year 5 946.764 8.9793 900 999 Group 8 49.76 50.0 0 Producton value Th. 5 63667.5 86003 56.909 434989 Producton per worker 5 35.9886 80.5385 6.53767 336.9 Producton-pw 5 5.559637.64484.80564 8.0348 Δ logproducton-pw 06.095536.309769 -.948643.9373 Capal Stock Th. 8 4506.49 486.9 0 8605.9 Capal Stock per worker 8 63.443 9.987 0 68.33 ncapal Stock-pw 8 3.63957.0985 0 7.76796 Δ logcapal Stock-pw 47.67867.48666-3.6583 3.485679 R&D nvestment 576 99998.7 577039 33.33333.8e08 R&D ntensy /Y 576 7.4089 39.999.0038407 634.0 R&D ntensy /Kª 54 38.8543 66.09 0 577.93 Investment ntensy /K 98 855.778 8990.44 0 594760.7 Temporary Workers Rateª 4.045394.355 0 R&D Workersª 49 6.57354 3.05995 0 755 Total Workersª 7 0.9647 53.54 0 99 Indvdual level Age at the nomnaton 348 47.039 0.5808 8 93 Average Board age 348 47.039 4.7583 0 77 Average Senory 4 36.74 7.59477 0 68 Average Board Senory 650 35.084 9.977367 0 57 Note: Dummy varables statstcs are expressed n percentages. a measured n 00 only. measured n 00-003 perod.
Frms n the Capala dataset wh nfo n 00-003 exstng also n 007 to match AIDA nformaton are 356 that s 85.3% of the sample. Frm-ndvdual observatons are 08. We frst test for potental sample selecton of these frms, n terms of age, sze, locaton and sector of producton younger, bgger or partcular sectors could have a hgher survval rate, hgher productvy or nnovaton capacy. The dscusson of the potental selecton bas s placed n the Appendx secton. The data then need a cleanng procedure because of nconsstences between brth dates and nomnaton dates of the ndvdual board members. Only 348 58.6% of total observatons dstrbuted n 8 frms contan sensble nformaton on brth and servce dates, whch fnally becomes our longudnal or quas -panel dataset wh frms as uns and board members as the longudnal dmenson, n the years 00-003. Table 4 - Frm sze and area dstrbuton n the fnal sample Small Medum arge North West North East Centre South Freq 55 730 43 43 407 9 99 % 0.77 59.45 9.79 35. 33.4 5.55 6.
6. Estmaton results and comments Table 5 shows the OS coeffcents and robust standard errors of the estmates of labor productvy growth rates -year rates on our varables of nterest equaton, nteracted wh the group dummes. Column and refer to nnovatve frms n general D = group of frms ntroducng a product or process nnovaton n the 00-003 perod, D = group of non nnovatve frms. Column 3 and 4 refer to product nnovatve frms D = group of frms ntroducng a product nnovaton and column 5 and 6 refer to process nnovatve frms D = group of frms ntroducng a process nnovaton. Table 5 - OS Estmates of abor Productvy long growth rates Dependent varable Y all nnovatons Product Process Coeffcent Robust Coeffcent Robust Coeffcent Robust Std. Err. Std. Err. Std. Err. K.06***.095.07***.09.04***.094 Innovatve frms.350***.0693.59***.0503.357***.057 Age*Innov -.003***.0004 -.003***.0006 -.0043***.0005 Age*non_Innov.006**.003 -.0004.0008.007**.0009 Senor*Innov.0009.006.00039.007.0009.006 Senor*non_Innov.007*.006.0059.006.007*.006 Temporary*Innov -.94***.03 -.04***.0386 -.979***.0367 Temporary*non_Innov -.6.097 -.40**.0700 -.95*.0734 Small -.090***.0098 -.0879***.0098 -.0896***.0098 Medum -.099***.0087 -.0986***.0087 -.0996***.0086 Group.043***.0063.049***.0063.0445***.0063 Constant -.48.097.046.0869 -.56.0964 Sem-elastcy of: Age -.0007.004 -.0036 ***.00 -.006**.00 Senory.0036.003.000.003.0035.003 Temporary share -.30***.0978 -.346***.0808 -.37***.087 Chow test for nstably F3,348 6.98 [0.000] 7.63 [0.000] 9.75 [0.000] Wald test for: Age [0.000] [0.005] [0.000] Senory [0.000] [0.004] [0.000] Temporary share [0.484] [0.439] [0.400] R 0.7 0.56 0.7 N 4 4 4 Note: * 0%, ** 5%, *** % level of sgnfcance, p-values n brackets. Sze, areas, ndustry and group dummes are ncluded n all regressons.
The nnovaton dummy reports a coeffcent of 0.35 wh hgh statstcal sgnfcance. Ths means that there exsts a sgnfcant drft for productvy growth for the Italan nnovatve frms, even n the years consdered. Ths s partcularly true for those frms ntroducng process nnovaton, whle the drft s nearly half that sze for those frms ntroducng product nnovaton. For the group of nnovatve frms column the estmate of the age coeffcent s negatve and statstcally sgnfcant, whle takes oppose sgn for the non nnovatve group. Havng a younger board of governor on average spurs productvy for currently nnovatve frms. It appears that an older board s assocated to hgher productvy n currently non nnovatve frms. The semelastcy for the whole sample s nstead not sgnfcantly dfferent from zero. If we defne nnovatveness by product nnovaton, then we obtan the results of columns 3-4. The dfference here s that average board age seems not to be of any mportance for non nnovatve frms. The sem-elastcy for the whole sample s equal to -0.36 and s statstcally sgnfcant. When process nnovaton s taken nto account see column 5-6, the mpact of the average board age on productvy s -0.43% for nnovatve frms, and 0.7% for non-nnovatve frms. The semelastcy for the whole sample s equal to -0.6% and s statstcally sgnfcant. The coeffcent of senory s zero for the nnovatve and slghtly posve for the non nnovatve group of frms, both wh general nnovaton and process nnovaton. The elastcy for the whole sample s not sgnfcantly dfferent from zero as well n all columns. The coeffcent estmate for the temporary workers share s statstcally sgnfcant and negatve - 0.94 for the nnovatve group but zero for the non nnovatve one s -0.04 for product nnovatve frms and -0.4 for frms whch have not ntroduced any product nnovaton. The elastcy for the whole sample s negatve and sgnfcant -0.3. The negatve value for age coeffcent turns out to be mostly drven by frms undertakng process nnovaton. Frms whch do not ntroduce process nnovaton show nstead sgnfcantly posve coeffcents for age and senory. The presence of older and more senor board members appears to be assocated wh hgher productvy n frms whch do not nnovate. Young members are assocated wh hgher productvy n nnovatve frms, whle senory does not seem to matter. The dstncton between the group of nnovatve and non nnovatve frms s nstead mmateral for the share of temporary workers. Havng a hgher rate of temporary workers s always assocated wh lower productvy. To confrm our results, Table 5 reports p-values of the Wald tests of parameter nstably for the H 0 hypotheses: γ = γ H : 0 λ = λ µ = µ The test for Age coeffcent γ rejects the null of equal parameters p-value of the test s equal to zero. The test for Senory λ rejects the null as well, even for product nnovaton whose p-value s < 0.0. Ths means that the two groups of frms behave dfferently on these aspects. The test for the share of temporary workers µ cannot reject the null of equaly: ths means that the estmated coeffcents do not vary across groups. In general the three Chow tests reveal that the constraned model has to be rejected.
All the regressons nclude control varables lke sze, geographcal areas, ndustry and group dummes. Table 6.a -Prob Maxmum kelhood for Innovaton dummes frst stage Prob M Innovaton Product Process Coeff Robust Std. Err. Coeff Robust Std. Err. Coeff Robust Std. Err. R&D: yes.7469***.0354.8407***.035.457***.0333 R&D workers t-.87***.067.305***.044.09***.038 Cash flow t.007***.0004.0008**.00035.006***.0003 Small.99***.0463.0934**.0464.476***.044 Medum.307***.036.356***.0373.87***.0356 Constant -.387***.0359 -.995***.0379 -.64***.0357 Pseudo-R 0.59 0.965 0.0636 N 0640 0640 0640 Table 6.b Second stage results for system 6, OS and SS Dependent OS SS OS SS Y Coeff Robust Std. Err. Coeff Robust Std. Err. Coeff Robust Std. Err. Coeff Robust Std. Err. Innovatve frms Non Innovatve frms α 0.05***.035.098***.089 α.037.074 0.0467** 0.0 β.34***.0965.87**.00 β.449***.37 0.3 0.33 γ -.0038***.0004 -.0030***.0004 γ.000.000 0.003** 0.00 λ.00.00.004.00 λ -.03***.009-0.0069** 0.003 µ -.838***.038 -.03***.0339 µ -.35**.056-0.37** 0.4 -σ εω -.0876***.09 -.033***.09 σ εω.056.07 0.063*** 0.04 R 0.58 0. R 0.0768 0.066 Sargan 5.6 [0.076] Sargan 0.648 [0.73] N 706 6437 N 949 070 Note. All the regressons nclude sze, sector and area dummes. The frst stage refers to column and of Table 6a. The SS method nstruments the growth rate of the capal stock per worker.
Table 6.a and Table 6.b show the -stage method of estmaton of the parameters n 4 and 6, whch take nto account the endogenous formaton of the two groups. Table 6.a shows the estmates of the frst stage decson to nnovate. The prob estmatons of Table 6.a are used as a frst stage n the swchng regresson wh endogenous swchng eq. 4. We add the prob for product nnovaton n columns 3 and 4 or process nnovaton n columns 5 and 6. These last regressons are useful to understand the mportance of the nstruments n determnng the decson of nnovatng. For example, engagng nto R&D actvy and hrng R&D workers have a bgger mpact on product nnovaton than process nnovaton. Per worker current Cash flow s an mportant dmenson for ntroducng process nnovatons. Table 6.b shows the estmates of the second stage swchng regresson: the Maddala method wh OS on eq. 6 and a SS method to take nto account the endogeney of the growth rate of capal stock per worker. 4 The nstruments used for the growth rate of the capal stock predcton are the nal level of the capal/labor rato, age of the frm at the begnnng of the sample perod, nvestment ntensy at the begnnng of the sample perod, sze, area and sector dummes. The frst comment over Table 6.b regards the fact that close coeffcents are obtaned for the three man varables of nterest, when usng OS and SS methods on the sample of nnovatve frms. The OS has a slghtly hgher R though 0.58, and the Sargan test for over-dentfyng restrctons cannot reject the hypothess of valdy of nstruments at the 5% level [p-value of the Sargan test = 0.076]. The mpact of mean age of the board members on productvy growth s negatve and statstcally sgnfcant γ,os = -0.0038, γ,ss = -0.003. The mpact of mean senory of the board members s essentally zero. The mpact of the share of temporary workers s negatve and sgnfcant µ,os = -0.838, µ,ss = -.03 for nnovatve frms. These latter estmates confrm the results of Table 5. The correcton for swchng regresson s negatve and sgnfcant -σ εω, OS = -0.0876. The OS estmate of the capal share appears overestmated α,os = 0.05 whle s equal to 0.098 when the capal stock s nstrumented. If we compare the OS and the SS results for the sample of non nnovatve frms, the results dffer n some aspects. The SS method gves a slghtly hgher R 0.066 and the Sargan test cannot reject the valdy of the nstruments p-value = 0.73. The OS coeffcent for the capal stock growth s zero, whle s halved n SS compared to the result for nnovatve frms α = 0.0467. The mpact of the mean Age s null accordng to OS, and posve and sgnfcant wh SS γ = 0.003. Ths latter result s coherent wh that of Table 5. 4 We run smlar regressons of equatons 7 and 8 whch allows to test for parameters equaly between the two regmes. The results confrm the presence of sgnfcant dfferences n the estmates of the coeffcents, n partcular as far as Senory and the rate of Temporary workers are concerned, between the two regmes. Table wh these other regressons are avalable upon requests.
The mpact of mean senory s negatve and que sgnfcant wh both methods λ,os = -0.03, λ,ss = -0.0069. Ths result s oppose wh respect to that of Table 5. The mpact of the share of temporary workers do not dffer sgnfcantly across the methods µ,os = -0.35, µ,ss = -0.368 and s negatve and sgnfcant. Ths result dffers from that of Table 5 where the correspondent coeffcent s zero. The correcton for swchng regresson s not statstcally sgnfcant for OS whle s posve and sgnfcant for SS σ εω,ss = 0.063. Notce that the number of observatons used wh the SS s less than the number of observatons n OS regressons, due to the fact that nstrument varables for capal stock contan few more mssng values. As a robustness check we replcate the estmates of equaton for the sub-sample of managers who are supposed to take decsons whn the board. We select those frms whose board contans competence-specfc managers, who are supposed to nfluence the decson to ntroduce nnovatons n the frm, eventually. The sgn and sze of the man coeffcents are confrmed. 5 We fnally mplement the Maxmum kelhood Endogenous Swchng model whch allows us to obtan consstent and effcent estmates. Followng the method suggested by okshn and Sajaa 004 we estmate the equaton system 3 frst by ncludng the growth rate of the capal stock as s analogously to the OS method n Table 6.b, second by substutng the varable wh s predcted value obtaned through a frst-stage regresson wh nstruments analogously to the SS method used n Table 6.b. The results of the M Endogenous Swchng Model are shown n Table 6.c, below. 5 Results are avalable upon request.
Table 6.c. Second stage results for system 6, Maxmum kelhood effcent method Dependent M M predct kˆ M M predct kˆ Y Coeff Robust Std. Err. Coeff Robust Std. Err. Coeff Robust Std. Err. Coeff Robust Std. Err. Innovatve frms Non Innovatve frms α.99***.095.79***.048 α.033.004.365**.030 β.33*** 0.099.354***.8 β.338***.95.39**.338 γ -.0035***.0005 -.009***.0006 γ.006.004.00.003 λ.0006.00 -.000.00 λ -.0097***.008 -.005***.008 µ -.58***.0388 -.877***.047 µ -.75*.0988 -.77*.097 σ.7***.0378.85***.045 σ.34***.044.34***.0444 ρ -.345.36 -.396.868 ρ -.084.095 -.07.49 Mlls rato.487.339.486.3309.3***.540.33***.509 Condon Exogenous K Endogenous K Wald χ.3 [0.9].97 [0.608] PInno.7.9.7.07 N 9865 9865 Note. All regressons nclude sze, area and sector dummes. Columns 3 and 4 show M wh predcted value of the growth rate of capal stock for nnovatve frms. Columns 7 and 8 replcate M wh predcted value for the non nnovatve cluster. Wald Ch-square tests the ndependence of the resduals n system 3. PInno s the predcted probably of beng n the nnovatve regme. The results for nnovatve frms confrm those of Table 6.b. The estmated share of the capal stock s equal to 0.79 when capal stock s predcted. The parameters for mean age and senory have the same values as n Table 6.b The temporary worker share coeffcent s slghtly lower both n M and M predcted. The value of ρ equal to zero means that the correlaton between the resduals of the second-stage equaton and the selecton equaton s null. Ths means that an nnovatve frm does no better or worse n ncreasng productvy than a frm whch randomly chooses to be n a regme. The capal share for non nnovatve frms s zero wh M and equal to 0.365 posve and sgnfcant for M predcted. Ths value s n between the OS and SS estmates of Table 6.b. Our estmates also ndcate that the mpact of average board age for non nnovatve frms s zero. Ths result s dfferent from that of Table 6.b and Table 5, where the SS coeffcent s posve
and statstcally sgnfcant. Senory s negatve and sgnfcant for productvy growth of non nnovatve frms, equvalently to Table 6.b. The mpact of the share of temporary workers s negatve and sgnfcant even f the sze of the mpact s lower wh M than SS µ,m µ,mpred = -0.75. Fnally, the correlaton between selectng the regme and the second-stage equaton s stll zero. Ths s confrmed by the Wald test of ndependence across the equatons of the system 3. To gan a better understandng of the numercal mplcatons of our results, we take the M predcted as the best coeffcents for calculatng some exercses of comparatve statcs. If we evaluate the absolute value of elastcy of labor productvy to the mean age of the board members for nnovatve frms 49.5 years, ths s equal to 0.43. Ths means that a hgher age by, say, 0 percent around 5 years of age lowers labor productvy by -.43%, whle there would be no productvy change n non-nnovatve frms. If we evaluate the elastcy of labor productvy wh respect to senory, there would be no mpact of a change n senory spells on productvy for nnovatve frms. The elastcy for non-nnovatve frms would be equal to 0.373 at the mean senory 35.5 years. Ths means that a hgher experence whn the frm by 0 percent around 3 years and a half would ental a varaton of labor productvy by -3.73%, for the non-nnovatve frms only. Fnally the elastcy of labor productvy wh respect to the share of temporary workers s equal to 0.006 for nnovatve frms, and 0.0057 for non nnovatve frms measured at the mean rate values. If the share of temporary workers ncreases by 0%, then productvy wll decrease by 0.6% for nnovatve frms on average and by 0.57% for the non nnovatve ones.
7. Conclusons Italy s prolonged productvy slowdown s not a cyclcal contngency and deserves careful nvestgaton. In ths paper, we gve our contrbuton evaluatng two of the man vews proposed to explan such slowdown. We fnd that the abor Supply vew s consstent wh our data, whle as to the abor Demand vew, our results ndcate that manageral experence s not unambguously correlated wh productvy performance n the entre sample. Defne patterns of correlaton are present though, once the whole sample s spl nto nnovatve and non-nnovatve clusters. Age, n partcular a measure of overall experence n the labor market appears to be weakly posvely correlated wh productvy n non-nnovatve frms, whle s negatvely correlated wh productvy for nnovatve frms. The pattern of correlaton for senory s nstead less robust and requres further nvestgaton. The cross-sectonal statstcal analyss of long-dfferences based on frm-averaged data s also not problem-free. A bg ssue s potental reverse causaton. The statstcal relatons we ntend to analyze pos that the temporary share of workers, age or senory varables are the ndependent varables and productvy the dependent varable. But cross-secton data as such be they observed at a gven pont n tme or averaged over tme may only ndcate correlaton, not causaton. Therefore, f the estmated coeffcent kng senory and productvy s negatve, ths may not ndcate that the frms where aged or senor managers are employed are less productve. Rather, the negatve correlaton may smply sgnal that older or senor managers tend to stay longer n less productve and older frms, featurng outdated machnes and methods of producton, probably because they managed to put n place successful relatons, whle new, nnovatve and hghproductvy plants may be more often matched to young and brllant managers. If ths s the case, we would be wrongly nterpretng what causes what, attrbutng to senory or age a causal nfluence on plant productvy, whch may go the other way around. Ths s why we mplement our SS specfcaton. Our expectaton s that by choosng predetermned nstruments we may lessen the smultaney problems. Surely, a lot of unobserved heterogeney n plant productvy s stll there n the data even once we have augmented the lst of productvy determnants wh dummes and other control varables. Yet the problem of nterpretng the statstcal results from cross-sectonal estmates arses f and only f the unobserved therefore unmeasured frm varables are correlated wh the ncluded explanatory varables. For example, f manageral ably a typcally unobserved frm varable were unrelated to the tenure persstence of managers, then leavng out of the emprcal analyss would not be a major problem. Ths may or may not be the case though. If manageral ably s not observed and therefore omted from the analyss but turns out to be correlated wh some ncluded varable such as the tenure persstence of managers, s effect may be pcked up by the negatve estmated relaton between senor managers and productvy. We would be mspercevng the effect of manageral ably on productvy as f were the causal effect of manageral senory on productvy. To tackle ths problem, we control for a few dummy varables that capture some, though presumably not all, of the unobserved determnants of frm productvy.
References Bandera, Orana, ug Guso, Andrea Prat and Raffaella Sadun 008, Italan managers: fdely or performance?, Report presented at the Annual Rodolfo De Benedett Foundaton Conference The rug class, Revsed: September Crépon Bruno, Duguet E. and Jacques Maresse 998, Research, Innovaton and Productvy: an Econometrc Analyss at the Frm evel, NBER Workng Paper 6696, Aug. 998 Daver, Francesco 009, Italy, before and after ehman Brothers, VoxEu.org, June 5 Daver Francesco and Cecla Jona asno 005, Italy s dece: gettng the facts rght, Gornale degl Economst ed Annal d Economa, December, 365-40 Daver Francesco and Mka Malranta 007, Age, senory and labour costs: lessons from the Fnnsh IT revoluton, Economc Polcy, 8-75 Gordon Robert J. and Ian Dew-Becker 008, The role of labor market changes n the slowdown of European productvy growth, CEPR Dscusson Paper, February 008 Grlches v 979, Issues n assessng the contrbuton of research and development to productvy growth, The Bell Journal of Economcs, 9-6 Hall Bronwyn, Francesca ott and Jacques Maresse Employment, nnovaton and productvy: evdence from Italan mcrodata, Industral and Corporate Change, 008 Hall Bronwyn, Jacques Maresse, ous Branstetter and Bruno Crépon, Does cash flow cause Investment and R&D: An Exploraton Usng Panel Data for French, Japanese and Uned States Scentfc Frms, Dscusson Paper Nuffeld College Oxford n 4, 998 okshn Mchael and urab Sajaa 004: Maxmum-lkelhood estmaton of endogenous swchng regresson models, Stata-journal, vol. 4, n.3, pp. 8-89 Maddala, G., 983 med-dependent and Qualatve Varables n Econometrc, Econometrc Socety Monographs No. 3, Cambrdge Unversy Press, New York Pars Mara aura, Fabo Schantarell and Alessandro Sembenell, Productvy, Innovaton and R&D: Mcro Evdence for Italy, European Economc Revew, 006, 037 06
Appendx We control for sample selecton that could actually come up when Capala IX survey data are matched wh AIDA balance sheet of frms present n 007. Not all Capala frms exst n AIDA regster. Nonetheless, we manage to retan almost 86% of the Capala sample. Therefore, we check n what type of characterstcs do frms n-sample and out-of-sample dffer. Fgure A. In and Out Sample dstrbuton of Capala frms by sze Frm dstrbuton by number of workers n and out-sample Fracton of frms 0...3.4-0 -50 5-50 5-499 500 e oltre Sze ntervals out-of-sample n-sample Fgure A shows the dstrbuton by class of workers of the frms falg n and out of our fnal panel. The panel tends to mantan medum sze frms manly 87%, whle keepng around 79% of the medum-large and large frms. As far as the very small frms, our panel keeps 8% of them. Formally, the test for ndependence hypothess rejects the null Pearson ch-square4 = 5.7455, p- value = 0.000 meanng that beng n or out of sample depends n a certan way on frm sze. We lose 5.6% of frms located n North-West part of Italy ombarda, Pemonte, gura, Valle d Aosta, 3.9% of the frms located n the North-East Trentno A.A., Veneto, Frul V.G., Emla Romagna, 3.5% of the frms located n the Centre Toscana, Umbra, Marche, azo and 5.8% of the frms located n the South. The Pearson ch-square4 = 3.450 wh p-value = 0.49 says that there s statstcal ndependence between the regonal dstrbuton and beng n or out of sample. Tradonal sectors wh lower Ateco 99 code,.e. Food and Beverages, Textles, Clothes, Tobacco, tend to be underrepresented wh respect to the orgnal Capala sample, as we can see from Fgure A. In any case f we consder Hgh-Tech versus the others, there s an ndependent dstrbuton of frequences n and out of sample Pearson ch-square test = 0.395 wh p- value=0.530.
Fgure A. Dstrbuton of frms by Ateco 99 classfcaton, n and out-sample Fracton 0.05..5 Dstrbuton of frms by Sectors n and out of sample 5 6 7 8 9 0 3 4 5 6 7 8 9 30 3 3 33 34 35 36 37 Ateco 99 out of sample n-sample We then run a two-sample t test wh equal varances to test for equaly of average frm age between the two groups n-sample, out-sample. The results hghlght that frms outsde the sample are on average 3 years older, and the dfference n means s statstcally sgnfcant. Group Obs Mean Std. Err. Std. Dev. [95% Conf. Interval] Out sample 570 3.87.938.4 30.03 33.7 In sample 3469 8.87.35 9.6 8.4 9.5 Combned 4039 9.9.309 9.67 8.69 9.90 dff 3.00.887.59 4.74 Degrees of freedom: 4037 Ho: meanout - meann = dff = 0 Ha: dff < 0 Ha: dff 0 Ha: dff > 0 t = 3.3795 t = 3.3795 t = 3.3795 P < t = 0.9996 P > t = 0.0007 P > t = 0.0004 Fnally, we run an assocaton tests to check for ndependence between beng an nnovatve frm and beng n or out of sample, to evaluate whether less nnovatve frms are those kcked out of the fnal panel. The Pearson ch-square tests are lsted for dfferent types of nnovaton actvy:
R&D expendures n 00-003 yes/no Pearson ch-square = 3.5 p-value = 0.06 Introducng product nnovatons yes/no Pearson ch-square = 7.94 p-value = 0.007 Introducng process nnovatons yes/no Pearson ch-square =.89 p-value = 0.39 Introducng both process and product nnovatons yes/no Pearson ch-square =.49 p-value = 0.34 We reject the hypothess of ndependence for R&D expendure and product nnovaton only. That means that frms nvestng nto R&D and ntroducng product nnovatons have a slghtly hgher probably to survve. We cannot reject the null for process nnovatons or both knds of nnovatons, nstead. Introducng process nnovatons or not provde a frm equal probably to reman n our sample.