Profit Persistence in the Food Industry: Evidence from five European Countries

Size: px
Start display at page:

Download "Profit Persistence in the Food Industry: Evidence from five European Countries"

Transcription

1 Proft Persstence n the Food Industry: Evdence from fve European Countres Adelna Gschwandtner and Stefan Hrsch* *Correspondng author. E-mal: stefan.hrsch@lr.un-bonn.de Paper prepared for presentaton at the EAAE 2011 Congress Change and Uncertanty Challenges for Agrculture, Food and Natural Resources August 30 to September 2, 2011 ETH Zurch, Zurch, Swtzerland Copyrght 2011 by Adelna Gschwandtner and Stefan Hrsch. All rghts reserved. Readers may make verbatm copes of ths document for non-commercal purposes by any means, provded that ths copyrght notce appears on all such copes.

2 1. Introducton Under condtons of perfect competton frm profts that devate from the average, can only be a transtory phenomena. However, t s commonly observed that actual proft rates dffer heavly across frms contradctng the proposton of the compettve envronment hypothess. In ndustral economcs Ban s (1968) Structure-Conduct-Performance paradgm and the more dynamc market-based vew manly focus on ndustry characterstcs as a drver of such abnormal profts. The so called new learnng on the other hand attrbutes abnormal profts to nternal frm specfc resources. Accordng to the resource-based vew, valuable, rare and nmtable resources are the drvers of heterogenety n frm proftablty. Startng wth Mueller (1986) a seres of contrbutons amng to analyze the persstence of abnormal frm proftablty has emerged. Whle most of these studes concentrate on entre manufacturng sectors, studes whch only focus on the food ndustry have been sparse wth the excepton of Schumacher and Boland s (2005) study of proft persstence n the US food ndustry. In order to fll ths gap, ths study ams to quantfy proft persstence and ts determnants for food processors n fve European Unon (EU) member countres: Belgum, France, Italy, Span and the UK, for the perod of The food ndustry 1 consttutes one of the largest economc sectors wthn total EU manufacturng 2 and s coverng a wde range of dfferent economcal actvtes rangng from the processng of bakery products to the producton of a very heterogeneous range of beverages. Food markets are characterzed by strong competton, hgh market saturaton and are confronted wth hgh concentraton n the retal sector. Due to these specal characterstcs a dfferentated examnaton of the food ndustry appears useful and necessary. Besdes beng the frst analyss of proft persstence n the European food ndustry, ths study contrbutes to the lterature n three ways. Frst, contrary to most prevous studes, whch are ether restrcted to publcly quoted frms or to a mnmum frm sze crteron ths study s based on data whch has nearly no frm sze restrctons makng a more precse representaton of the ndustry possble. Ths s an mportant advantage partcularly for a study of the food ndustry snce 96% of all EU food producers are small frms. 3 Furthermore small frms play an mportant role regardng competton wthn ndustres and therefore also for the examnaton of proft persstence. Second, by takng nto account fve EU member countres, whch n 2007 account for 59% of the enterprses and 51% of turnover of the EU-27 food ndustry (Eurostat 2010) a large fracton of the sector s analyzed. Thrd, whle the autoregressve model of order one AR(1) has become the econometrc workhorse of the persstence of profts lterature, ths study estmates autoregressve models up to order 4 and then chooses the best lag model for further analyss. Ths approach s crucal, snce the dynamcs of frm profts can be more complex than the smple AR(1) process could capture. The paper s organzed as follows. Secton 2 descrbes the methodology whle Secton 3 gves a descrpton of the Data. Secton 4 presents the emprcal results and secton 5 concludes. 1 Manufacture of food products and beverages excludng tobacco. Regardng to NACE Rev. 1.1 dvson DA15 2 Contrbutng 11 % of total value added the food ndustry occupes the second place wthn EU-27 manufacturng followng the engne buldng ndustry n Accordng to the SME defnton of the European Commsson small frms are defned as havng employed less than 50 persons and total assets of less than 10Mo.

3 2. Methodology The AR(1) persstence of profts model s a smple regresson of the level of frm profts at a gven pont n tme on the mmedately prevous level: π = + (1), t α + λπ, t 1 ε, t where ε, t s a whte nose error term wth zero mean and constant varance. As n the prevous lterature, proftablty of frm at tme t ( π,t ) s measured as the devaton of the frms return on assets (ROA) from the compettve norm whch usually s approxmated by the mean across the sample of frms. Ths normalzaton serves two ends. Frst, t removes the mpact of macroeconomc cycles. Second, by takng the sample mean as a proxy for normal proft, we can nterpret frm proftablty as devatons from the compettve norm or as abnormal proftablty, wth attendant welfare mplcatons. The coeffcent on lagged proft ( ˆ λ ) reflects the stckness of profts from perod to perod, and can be nterpreted as short-run persstence, and as a measure for the speed of adjustment to the long run level. The long run average of the autoregressve process, on the other hand, yelds the so-called long run projected proft rate (LRPP). The LRPP, defned as ˆ ρ ˆ /(1 ˆ = α λ ) s the steadystate equlbrum value to whch, accordng to the model, the seres s ultmately headng. ρˆ s a measure of permanent rents, whch are not eroded by compettve forces and can therefore be nterpreted as long run persstence. Snce ˆ ρ = 0 mples a long run projected return on assets equal to the compettve norm, the percentage of ρˆ s sgnfcantly dfferent from zero n a gven sample s an ndcator of the degree of proft persstence wthn t. The present study extends the classcal methodology by usng the best lag model. Autoregressve models up to order four have been estmated for each company and Schwarz Bayesan Informaton Crteron (SBC) has been employed n order to decde, whch model best descrbes the adjustment path. After choosng the best lag model, the long-run projected proft rate becomes ˆ ρ = ˆ /(1 ˆ α λj ), where j = { 1...L} s the number of lags of the AR process and ˆ λ = ˆ λj s the speed of adjustment parameter. Ths extenson s mportant snce the adjustment path of proftablty mght be more complex than a smple AR(1). The best lag model allows for more general dynamcs than the smple AR(1) and at the same tme enables comparson wth most of the prevous lterature. Stablty, and convergence upon a fnte steady state, requres that the estmate of λ les n the range between plus and mnus one. Furthermore, the procedure s approprate only for statonary AR processes, snce αˆ /(1 ˆ λ ), the measure for long-run persstence, s not defned for unt root processes, where ˆ λ = 1. If ˆ λ 1, the long-run persstence measure s poorly defned because the denomnator s close to zero. The practce of testng for non-statonarty has ts roots n macroeconomcs, where some macro economc varables can drft off endlessly. But non-statonarty seems far less lkely to be a problem for profts tme seres because a contnuous downward trend n profts must ether drve a frm out of the market or has to cause correcton polces to stop ths downward trend. At the same tme competton s expected to stop any contnuous upward trend n profts. In order to test for statonarty a panel KPSS test was done for each country wth the results presented n secton 4. Persstence analyses based on autoregressve models can serve as a practcal, and potentally powerful, tool both n assessng the status of competton n qute large populatons of frms, and also as a screenng devce for dentfyng partcular cases for polcy revew.

4 3. Data The frm data used for the present analyss stems from AMADEUS, a commercal pan- European balance sheet database compled by Bureau van Djk Electronc Publshng. The analyss s based on the perod 1996 through Followng prevous lterature frm proftablty n year t ( π t ) s measured as the frms ROA n year t normalzed by mean ROA of that year. ROA s calculated as a frm s proft/loss before taxaton and nterests dvded by total assets. 5 The Data was screened by elmnatng frms for whch less than 13 years of ROA data were avalable. Addtonally all frms not assgned to a 4-dgt NACE ndustry, and frms actve n the mscellaneous category (NACE 1589: manufacture of other food products not elsewhere classfed) were removed from the database. Ths was necessary snce the second step of the estmaton whch ams at explanng proft persstence requres a clear allocaton of frms to 4-dgt NACE ndustres. Observatons whch le n the top and bottom 5% of the dstrbuton n each year were removed from the sample n order to prevent the results from beng excessvely nfluenced by outlers. 6 Regardng frm sze some prevous studes have ether used a mnmum frm sze crteron (see e.g. McGahan and Porter 1999) or are restrcted to publcly quoted frms (see e.g. Schumacher and Boland 2005). 7 For the European food ndustry a mnmum sze crteron would lead to a tremendous loss of nformaton, snce small enterprses represent 96% of all food ndustry corporatons. Even though the economc relevance of these frms s rather small snce n 2007 they only accounted for 21% of ndustry turnover (Eurostat 2010) the large number of small frms s a characterstc of the ndustry that should not be neglected. 8 The countres consdered n ths study were chosen by means of remanng sample szes after screenng. Ths resulted n a group of fve countres: Belgum, France, Italy, Span and the Unted Kngdom whch make up 51% of total EU-27 food ndustry turnover n Germany as the European leader regardng food ndustry turnover s not comprsed n the study snce the remanng sample was too small. Ths s caused by the fact that durng most of the perod analyzed n ths study the majorty of German frms were not subject to a legal oblgaton to publsh fnancal statements. Nonetheless the fnal sample comprses four of the top fve countres regardng food ndustry turnover 9 wth a total of 5494 frms actve n dgt NACE Industres. 4. Emprcal Results Ths secton frst examnes the exstence of proft persstence per se by presentng results on the two persstence parameters λˆ and ρˆ whch were estmated for each frm n the sample usng equaton 1. In order to assure convergence frms where ˆ λ > 1 were dropped from the 4 Most proft persstence studes are based on longer tme spans of about years. These studes however manly analyze entre manufacturng sectors whch makes data avalablty over long tme spans easer. In the case of the present study the 13 year tme span s the longest avalable for the European food ndustry. 5 It s mportant to note, that nterests are ncluded n the numerator n order to make the proft measure ndependent of the source of funds used to create total assets. 6 The elmnaton of a sngle proft rate observaton means that there s less than 13 years of ROA data avalable for that frm whch leads to ts overall elmnaton from the sample. 7 McGahan and Porter use 10 mllon US$ n total assets as a cut off value. 8 Nonetheless a mnmum sze crteron could be justfed by the fact that the estmaton consders all frms n an equal way regardless of frm sze, whch causes the results beng largely dependent on the huge number of small enterprses, whose economc relevance however s rather small. 9 Accordng to Eurostat (2010) Germany, France, UK, Span and Italy are the European leaders regardng food ndustry turnover whle Belgum takes eghths place. Data for 2007 except for Italy (2002) and France (2003).

5 analyss reducng the ntal sample sze to 4676 frms. 10 The remanng seres were tested for statonarty usng the KPSS Test whch ndcated that n each country around 90% of the seres were statonary. 11 The second step of the analyss ams at explanng proft persstence by estmatng the effect of several frm and ndustry characterstcs upon the two persstence parameters. 4.1 Proft Persstence The short run persstence parameter λˆ reflects the speed at whch a frm s proft rate π t converges to ts long run level ρˆ. Small values of λˆ whch are close to zero ndcate a low degree of persstence of past profts and a quck eroson of short-run rents. Small λˆ values can therefore also be seen as a sgn of hgh competton. Table 1: An overvew of the persstence parameters Belgum France Italy Span UK # obs. ˆ λ < Mean λˆ % of ρˆ s sgnfcantly dfferent from 0 a % of ρˆ s sgnfcantly >0 a % of ρˆ s sgnfcantly <0 a % of λˆ s sgnfcantly dfferent from 0 a % of equatons wth R 2 > a sgnfcant at the 5% level or less Rows one and two of table 1 show the number of frms wth λˆ < 1 and the mean λˆ values for each country. The hghest mean value can be found for the UK (0.232) followed by Span (0.201) and France (0.188) whch ndcates that the average speed of adjustment to the long run level s slower n these countres mplyng weaker competton. Italy (0.143) and Belgum (0.057) on the other hand exhbt lower average λˆ values ndcatng stronger competton n these countres. Overall mean λˆ s for the food ndustry turn out to be rather small compared to other studes based on entre manufacturng sectors. Most prevous studes of entre manufacturng sectors yeld average λˆ values above Competton among food producers therefore seems to be rather strong. These results are not too surprsng snce EU food markets are consdered as beng strongly saturated whch n conjuncton wth a hgh level of prce competton leads to strong competton among producers and therefore to relatvely low average λˆ values. Furthermore whle the mportance of drect sales by producers has dmnshed the retal sector as the man lnk between the food ndustry and the consumers s characterzed by a hgh and stll ncreasng degree of concentraton whch leads 10 Furthermore ρˆ values whch le n the top and bottom 2.5% of the dstrbuton of each country sample were removed. Ths was necessary snce for λˆ values close to one ρˆ s poorly defned leadng to extremely hgh values of the latter. 11 Exact fractons of statonary seres are: Belgum: 95%, France: 88%, Italy: 96%, Span: 87%, UK: 86% 12 Goddard and Wlson (1999) gve an overvew of prevous results for entre manufacturng sectors of seven countres durng varous tme perods. All mean λˆ values except for the US exceed 0.4.

6 to strong barganng power on part of the retalers puttng producers under pressure. In most EU countres the top 5 supermarket chans have a market share of around 70 %. As regards the fve countres analyzed n the present study Span shows the hghest 5-Frm concentraton rato (0.79) followed by Belgum (0.77), France (0.69), UK (0.54) and Italy (0.41). 13 For Belgum (lowest average λˆ and hgh concentraton of the retal sector) and for the UK (hghest average λˆ and relatvely low concentraton of the retal sector) results are consstent. The barganng power of retalers s renforced by the ncreasng mportance of ther prvate labels whch n 2006 already acheved a market share of 27%. 14 Establshng prvate labels enables retalers to offer products that are of smlar qualty but usually 10 to 20% cheaper than brands meanng that n cases of low product dfferentaton, prvate labels can acheve hgh market shares. Producers therefore do not have the power to take decsons n the downstream of the value chan. They can only counterval by offerng lower prces or better servces and qualty. Table 1 also gves an overvew of the fractons of sgnfcant proft-persstence parameters for all fve countres. The percentage of ρˆ s sgnfcantly dfferent from zero n row three reflects the fracton of frms wthn each country that do not converge to the average n the long run. It can therefore be seen as an ndcator for the persstence wthn the food ndustres of the fve countres. Ths percentage s around 40% for all countres. Rows four and fve of the table reveal the fact that for Belgum, France, Span and the UK the percentage of frms wth a sgnfcant postve ρˆ value s smlar to the percentage of frms wth a sgnfcant negatve ρˆ value. For Italy n contrast the percentage of frms showng sgnfcantly negatve ρˆ s s much hgher than the percentage of frms wth sgnfcantly postve values. Ths suggests that wthn the Italan food ndustry competton forces are operatng better for frms wth profts above the norm than for frms wth profts below the norm. Overall these results show that a sgnfcant fracton of frms tend to earn profts both above and below the norm that persst n the long run ndcatng that the process of convergence s far from completon. For frms wth sgnfcantly postve ρˆ s some specfc advantages or ndustry wde entry barrers must exst whle for frms wth sgnfcantly negatve ρˆ s mpedments to leave the market e.g. sunk costs or other forms of ext barrers must exst. The percentage of λˆ s sgnfcantly dfferent from zero whch s found n row sx can also be seen as an ndcator for the magntude of the compettve process. The hgher ths percentage, the hgher the number of frms for whch the forces of competton were not strong enough to erode profts wthn one to four years mplyng proft persstence. The percentage dffers between the countres beng the hghest for the UK (29.8) followed by Span (23.5) and France (23.4). Belgum (12.4) and especally Italy (6.0) show a much smaller percentage of sgnfcant λˆ values mplyng that competton n UK, France and Span s weaker as n Italy and Belgum. These results concde wth the mean λˆ values. The last row of Table 1 shows the percentage of equatons where more than 10% of the varaton n proftablty s explaned by the autoregressve process. For each country except Italy ths percentage seems to be larger as n prevous studes, ndcatng that the best lag structure has hgher explanatory power compared to the autoregressve processes of order one on whch most prevous studes are based values accordng to Wjnands et al. (2006), p.45 ff. 14 Datamontor (2006): Tomorrow's Prvate Label Consumers. 15 For four of the seven countres analyzed n Mueller s (1990) study ths percentage s smaller than 50%.

7 4.2 Explanng Proft Persstence For both anttrust polcy and frm stakeholders the determnants of proft persstence mght be more nterestng than the patterns of proft persstence per se. Ths secton therefore analyzes the mpact of frm and ndustry characterstcs on the two persstence parameters λˆ and ρˆ. Data on the followng frm level characterstcs could be deduced from AMADEUS: Market Share (MS) measured as the rato of frm s sales to ndustry sales s expected to be an mportant determnant of frm proftablty. Several studes found a postve relatonshp between market share and proftablty (see e.g. Szymansk et al. 1993). Nonetheless the mpact of market share has not always been unambguous. Gale (1972) one of the earlest efforts fnds that the effect of market share on proftablty s greater for larger frms n ndustres wth hgh concentraton and moderate growth. Gschwandtner (2010) argues that f market share s a measure for dversfcaton a negatve relatonshp can appear snce most studes fnd a negatve mpact of dversfcaton on proftablty (see e.g. Berger and Ofek 1995). 16 Katchova (2005) fnds a smlar dversfcaton dscount for US farmng operatons. Frm age (Age) calculated by means of ncorporaton dates can account for lfecycle effects. One mght expect that agng decreases costs due to learnng effects wthn the frm and learnng spllovers from other frms n the same or n other ndustres. Loderer and Waelchl (2010) however fnd a negatve relatonshp between frm age and proftablty. They argue that corporate agng s attended by organzatonal rgdtes, slower growth and assets whch become obsolete wth tme. Beyond that, as predcted by the rent seekng hypothess corporate governance s pershng and CEO pay s ncreasng as frms grow older. The relatonshp between frm sze (Ln TA) measured as the logarthm of total assets and frm proftablty has not always been unambguous. Due to the fact that prce competton s the domnant competton strategy among food processors, achevng economes of scale through suffcent frm sze s expected to be a very mportant matter. Ollnger et al. (2000) e.g. show, that US chcken slaughterng plants that were two tmes larger than the average-szed plant have 8% lower per unt costs. Furthermore t can be assumed that the complex set of EU legslatons regardng food safety, anmal welfare, addtves and resdues, packagng and labelng put relatvely hgher admnstratve burdens on smaller frms as on frms of larger scale. Especally pre-market approval for new addtves, novel foods, genetcally modfed organsms (GMO s) and health clams are beyond the reach for the vast majorty of small food processors n the EU. 17 In addton, beng of larger sze mght certanly ncrease the ablty to counter the barganng power of retalers. However t has to be noted that there s as well evdence for the neffcency of large frms n the case of dseconomes of scale. In a recent study e.g., Goddard et. al. (2005) fnd a negatve mpact of frm sze on proftablty. The effect of frm growth (Gr. TA) measured as the growth rate of a company s assets s n general supposed to be postve. Yurtoglu (2004) e.g. fnds a sgnfcantly postve mpact of frm growth on long run persstence. However, Gschwandtner (2010) fnds evdence for a negatve relatonshp between frm growth and short run persstence durng the perod whle the effect becomes nsgnfcant n later perods. Two rsk proxes, one for short run rsk and one for long run rsk could be derved. Short run rsk (1/Curr) s measured by the rato of current labltes to current assets whch s the recprocal of a frm s current rato. As a proxy for long run rsk the frm s gearng rato (Gear) has been used, whch s defned as the rato of non-current labltes plus loans to shareholders funds. Frms wth a hgh gearng rato have a hgher rsk of beng unable to fulfll nterest and debt repayment oblgatons. Accordng to rsk theory frms wth hgher rsk should on average have a hgher proft level than companes wth lower rsk. Yurtoglu (2004) 16 In the present study market share could very well be a proxy for dversfcaton snce AMADEUS data to some extent s based on consoldated fnancal statements. 17 Wjnands et al. (2006)

8 usng the standard devaton of profts as a measure for frm rsk fnds that frms wth hgher varablty n profts also tend to have hgher long run profts. However, the bgger part of prevous studes fnds a negatve relatonshp between varous rsk measures and proftablty (see e.g. Mueller (1986) and Gschwandtner (2005)). Industry characterstcs were obtaned from Eurostat s annual detaled enterprse statstcs on manufacturng subsectons DA-DE and total manufacturng. 18 Concentraton measured by the four-frm concentraton rato (CR4) s the ndustry characterstc one would most lkely expect to affect the level of proft persstence. Frms n ndustres characterzed by collusve behavor as a result of hgh concentraton mght have the ablty to prevent entry leadng to a hgher degree of proftablty. Many studes report a postve relatonshp between concentraton and proft persstence (see e.g. Yurtoglu 2004). However there s also the possblty that hgh concentraton leads to very strong rvalry between frms wthn an ndustry resultng n a negatve mpact on proftablty. As regards ndustry sze (NF) whch s measured by the number of frms n an ndustry, one mght expect that the hgher the number of establshments n the ndustry the hgher the volatlty of profts, the stronger the competton and therefore a lower degree of proft persstence s to be found. The growth rate of an ndustry measured by the growth rate of the number of frms n an ndustry (Gr.NF) may also be an mportant factor explanng proft persstence. Its net effect on proft persstence however s ambguous at a theoretcal level. On the one hand n ndustres wth rapd growth the ablty of ncumbents to mantan ther market shares mght decrease leadng to a reducton of olgopolstc dscplne and subsequently to a decrease n proftablty and ts persstence. On the other hand n ndustres characterzed by fast growng output frms are not under pressure to reduce prces n order to ncrease sales and therefore abnormal profts mght be mantaned over tme. Some of the dfferences n proft persstence may be explaned by the degree of research and development. Permanent changes n consumer preferences related to ssues of health, convenence, varety, ethcs and safety as well as changes n technology make R&D an mportant ssue n the food ndustry. It has to be recognzed though that R&D n the food ndustry has a dfferent character as e.g. n the electronc ndustry. Conventonal foods and beverages have been n ths world for a long tme and the nventon of completely new ones s rather unusual. New food products are therefore manly varatons of older ones, e.g. product extensons by the use of new addtves, varatons n taste and packages desgned for dfferent consumpton moments. 19 In general one mght expect that the ntensty of R&D at the ndustry level s a bass for product dfferentaton and for the creaton of entry barrers for new frms whch enables ncumbents to earn hgh profts that persst over tme. Warng (1996) e.g. fnds a postve mpact of the ndustry-level R&D ntensty on short run persstence. In the present study the share of expendture for research and development n total ndustry sales (R&D) serves as a measure for R&D. The results of the regresson analyss are presented n Table 2. The frst equaton always explores the mpact of the ndustry and frm characterstcs on the short run persstence rate ( ˆ λ ) whle the second equaton always explans the mpact on the long run persstence parameter ( ˆ ρ ). Snce both persstence measures are estmated parameters, the equatons were weghted by the nverse of ther standard errors accordng to Saxonhouse (1976). All explanatory varables except age are mean values over the tme span analyzed. As regards frm characterstcs the mpact of market share (MS) seems to be ambguous as predcted. Whle the mpact s sgnfcantly postve on short run persstence n Italy 18 Avalable onlne at: 19 Accordng to Stewart-Knox and Mtchell (2003) only 7-25% of the food products launched can be consdered as beng truly novel.

9 Table 2: Regressors explanng the estmated parameters of Equaton 1 (heteroskedastcty consstent t-values) Cons. CR4 NF Gr. NF R&D MS Age Ln TA Gr. TA Gear 1/Curr R² p Belgum (1) (2) France (3) (4) Italy (5) (6) Span (7) (8) UK (9) (10) (-1.353) (0.756) (-2.006) (0.588) (1.400) (1.222) (0.600) (-0.061) (-1.421) (-2.170) (1.003) (0.477) 0.625* (1.957) (-0.439) (-1.123) (-0.314) (1.518) (1.258) (-0.344) (1.632) (0.978) * (1.700) ** (2.552) (-1.061) (-1.383) (-0.028) (-0.704) (0.788) (0.049) (0.400) (0.463) * (1.676) 5.191*** (3.581) (-0.746) (-1.059) 6.834** (2.412) (-0.433) * (-1.755) (-0.328) (0.322) (-0.878) (-0.912) (-0.337) (1.245) (1.254) *** (-5.794) (-0.761) (-0.775) (0.166) * (-1.942) (0.234) (0.103) (-0.504) (-0.386) ** (2.536) *** (-2.785) (-0.220) 0.030*** (2.860) (-1.629) ** (-2.285) (-0.013) * (-1.714) (-0.557) (-0.983) (0.469) *** (-4.480) (-1.295) ** (-2.093) (0.723) (0.223) 0.088* (1.958) (-0.303) 0.152*** (5.265) (-0.246) (-0.418) (0.386) 0.122*** (2.586) 0.255* (1.659) 0.352** (2.411) 1.399* (1.820) 0.185*** (15.883) 0.098*** (2.826) 0.096*** (3.253) (0.969) (-0.721) ** (-2.254) 0.983*** (3.807) 1.966** (2.046) ** (-2.113) (0.835) (0.263) (-0.407) (-0.305) ** (-2.450) (-0.795) (-0.098) (-1.493) *** (-3.259) (1.615) (-0.905) (0.139) (-1.506) (0.963) (1.077) (0.114) (-1.164) (-1.638) ** (-2.525) (-0.429) (0.988) Dependent varables: Equatons 1,3,5,7,9: λˆ ; Equatons 2,4,6,8,10: ρˆ. Snce the dependent varable s an estmated parameter, values are dvded by ther standard errors. Industry varables: CR4 = 4 frm concentraton rato; NF= # of frms n Industry; Gr.NF = Growth rate of the number of frms n the ndustry; R&D = Share of R&D expendture n ndustry value added. Frm varables: MS = frm sales/ndustry sales; Age = frm age; Ln TA = natural logarthm of total assets; Gr.TA= growth rate of total assets; Gear = gearng rato; 1/Curr = 1/current rato. All varables except age are averages over the sample perod Numbers n parentheses are whte heteroskedastcty-consstent t-values. p refers to the p-value of the F-test.

10 and long run persstence n Span t s sgnfcantly negatve for long run persstence n Italy and the UK. Whle nterpretng these results one has to keep n mnd that market share mght be a proxy for dversfcaton. Dorsey and Boland (2009) e.g. fnd that dversfcaton of US food producers nto actvtes wthn the food economy 20 leads to premums whle dversfcaton to unrelated actvtes outsde the food economy does not lead to premums whch mght n part be an explanaton for the ambguous results. As suggested by Gale (1972) the mpact of market share on proftablty can depend on the envronment n whch frms operate. The ambguous results could therefore also be a consequence of the wde range of economcal actvtes allocated on dfferent countres wth dfferent envronmental and ndustry structures that can be found wthn the European food ndustry. Frm age has a sgnfcant negatve mpact on long run persstence n Belgum, Italy and Span renforcng for these countres the fndngs of Loderer and Waelchl (2010) that corporate agng s attended by organzatonal rgdtes, slower growth and assets whch become obsolete wth tme. Frm sze (Ln TA) has a sgnfcant postve mpact on short run persstence n all countres except Italy. The mpact on long run persstence n the UK and Span s also sgnfcantly postve. These results emphasze the fact that beng of suffcent scale s a very mportant matter n the food ndustry. Larger frms seem to be able to counterval the superorty of retalers, to offer lower prces and seem to be less affected by admnstratve burdens lke premarket approval or the handlng of legslaton. The mpact of frm growth (Gr. TA) s manly postve. Belgum and Span show sgnfcantly postve coeffcents for both persstence measures. For France the mpact on short run persstence s sgnfcantly postve. A negatve coeffcent can only be found for long run persstence n Italy and for short run persstence n the UK. If frms seek growth through dversfcaton because they have exhausted growth n ther prmary feld of acton a negatve relatonshp between growth and proftablty may result as a consequence of a negatve relatonshp between dversfcaton and proftablty n a smlar way as t s suggested for market share. In contradcton wth rsk theory the mpact of frm rsk appears to be negatve. The gearng rato has a sgnfcantly negatve mpact on long run persstence n France and Span. The recprocal of the current rato has a sgnfcantly negatve mpact on the long run persstence measure n Span. The negatve relatonshp between rsk and proftablty whch was already found n prevous lterature also seems to apply to the food ndustres of these countres: less rsky frms are the ones that acheve hgher profts that last over tme. The mportance of Industry concentraton (CR4) n order to explan proft persstence seems to be neglgble. Only for France the coeffcent has a postve mpact on short run persstence merely sgnfcant at the 10% level ndcatng that some degree of colluson took place wth attendant welfare mplcatons for the frms. Frms operatng n large and fast growng ndustres n Belgum and France seem to obtan a hgher degree of proft persstence. Both measures have a sgnfcantly postve mpact on long run persstence n Belgum and on short run persstence n France. In addton ndustry growth has a postve mpact on the long run persstence measure n Italy and a negatve one n Span whch, however, s only slghtly sgnfcant at the 10% level. The relatonshp between R&D and proft persstence has to be nterpreted on the bass of the specal character of R&D n the food ndustry. The bulk of the coeffcents are negatve, however the nfluence s only sgnfcant for long run persstence n Italy and the UK. These results emphasze the fact that the vast majorty of new food products (72-88%) fal. 21 Stewart-Knox and Mtchell (2003) state that orgnal and truly new products are more lkely 20 Dorsey and Boland consder dversfcaton wthn the food economy as ntegraton whle dversfcaton refers to actvtes outsde of the food economy. 21 Stewart-Knox and Mtchell (2003)

11 to be successful than product extensons. Ths s further confrmed by the fndng of Hoban (1998) who states that the falure rate of truly new products s only 25%.The fact that only 7-25% of launched food products can be consdered as beng truly novel mght therefore explan to some degree the negatve mpact of R&D. Wess and Wttkopp (2005) analyzng the mpact of retaler concentraton on product nnovaton n German food manufacturng fnd that as a result of retalers upstream market power food processors acheve lower profts whch reduces ther ncentve for cost-ntensve product nnovaton. Hgh retaler concentraton mght therefore ntensfy the fact that only a small amount of launched products s truly novel and based on cost-ntensve R&D whle the bulk of nnovatons are only varatons or product extensons wth lesser success. Overall the results ndcate that the methodology for new food product development s n the urgent need of mprovement. Even though the bgger part of equatons s overall statstcally sgnfcant as shown by the p- 2 values of the F-Test t has to be noted that the adjusted R s n general relatvely small. The man reason s that several other varables lke mport, export, advertsng or merger actvty have prevously been found to be related to proft persstence. Unfortunately these varables could not be taken nto consderaton n the present analyss due to data lmtaton. 5. Concluson The precedng analyss of proft persstence n the European food ndustry ndcates that the process of converge towards a compettve norm s far from completon snce n each of the fve countres analyzed a sgnfcant fracton of around 40% of the frms tends to earn profts above or below the compettve norm whch persst even n the long run. As ndcated by the mean λˆ values proft persstence seems to be hghest n the UK and lowest n Belgum and Italy. Furthermore for the UK the percentage of λˆ s sgnfcantly dfferent from zero whch represents the fracton of frms for whch the forces of competton were not strong enough to erode profts wthn one to four years s hghest n the UK. These fractons are lowest n Belgum and Italy renforcng the fndng that compettve forces work slower for UK frms and faster n Belgum and Italy. However compettve forces do not seem to affect all frms equally. As the second step of the estmaton shows some of the observed dfferences n the degree of proft persstence can be explaned by specfc frm and ndustry characterstcs. As regards frm characterstcs t was shown that especally young, large frms that are also characterzed by fast growth are the ones earnng hgh profts that persst. Regardng ndustry characterstcs especally the sze and the growth rate of the ndustry n whch a frm operates seem to have a postve mpact on ts proft persstence. Addtonally frm rsk and the level of ndustry R&D expendture have a negatve mpact on proft persstence n some countres. Comparng the results wth other studes analyzng manly entre manufacturng sectors t has to be noted that the degree of persstence s lower n the food ndustry manly due to a hgh degree of market saturaton, strong prce competton and a hghly concentrated retalng sector. Another strkng dfference s the mportance of frm sze. Whle many prevous studes fnd evdence for the neffcency of large frms, beng of suffcent scale seems to be a very mportant matter n the food ndustry. A fnal crucal dfference s the specal characterstc of R&D and ts negatve nfluence on proft persstence. Contrary to other sectors nnovaton for the most part seems to be unsuccessful. From the researchers pont of vew an extenson of the analyss on the retalng- and wholesalng sector and an analyss of ndvdual 3-dgt NACE ndustres wthn food manufacturng could be a startng pont for further research. However, the purpose of the present study was to gve a frst mpresson of the proft persstence phenomena n the European food ndustry.

12 References Ban, J.S. (1968): Industral Organzaton, 2 nd ed., John Wley & Sons, New York. Berger, P.G., and Ofek, E. (1995): Dversfcaton s Effect on Frm Value. Journal of Fnancal Economcs, 37, pp Datamontor (2006): Tomorrow's Prvate Label Consumers. On Dec. 19th 2010 at: Dorsey, S. and Boland, M. (2009): The mpact of ntegraton strateges on food busness frm value. Journal of Agrcultural and Appled Economcs, 41(3), pp Eurostat (2010): Annual detaled enterprse statstcs on manufacturng subsectons DA-DE and total manufacturng, European Commsson, Luxembourg (Belgum). On Nov. 16th Gale, Bradley T. (1972): Market Share and Rate of Return. The Revew of Economcs and Statstcs, 54(4), pp Goddard, J. A. and Wlson, J. O. S. (1999): The persstence of proft: a new emprcal nterpretaton. Internatonal Journal of Industral Organzaton, 17, pp Goddard, J., Tavakol, M and Wlson, J. O. S. (2005): Determnants of proftablty n European manufacturng and servces: evdence from a dynamc panel model. Appled Fnancal Economcs, 15, pp Gschwandtner, A. (2005): Proft persstence n the very long run: evdence from survvors and exters. Appled Economcs, 37, pp Gschwandtner, A. (2010): Evoluton of proft persstence n the US: evdence from three perods. The Manchester School, forthcomng. Hoban, T. J. (1998). Improvng the success of new product development. Food Technology, 52, pp Katchova, A.L. (2005): The Farm Dversfcaton Dscount. Amercan Journal of Agrcultural Economcs, 87(4), pp Loderer, C. and Waelchl, U. (2010): Frm age and performance. Workng Paper Unversty Bern. McGahan, A. M. and Porter, M. E. (1999): The Persstence of Shocks to Proftablty. The Revew of Economcs and Statstcs, 81(1), pp Mueller, D. C. (1986): Profts n the Long Run, Cambrdge Unversty Press, Cambrdge. Mueller, D. C. (ed.) (1990): Dynamcs of Company Profts: An Internatonal Comparson. Cambrdge Unversty Press, Cambrdge. Ollnger, M., MacDonald, J. and Madson, M. (2000): Poultry Plants Lowerng Producton Costs and Increasng Varety. Food Revew, 22(3), pp Saxonhouse, G. R. (1976): Estmated Parameters as Dependent Varables. The Amercan Economc Revew, 66, pp Schumacher, S. K. and Boland, M. A. (2005): The persstence of proftablty among frms n the food ndustry. Amercan Journal of Agrcultural Economcs, 87(1), pp Stewart-Knox, B. and Mtchell, P. (2003): What separates the wnners from the losers n new food product development? Trends n Food Scence & Technology, 14, pp Szymansk, D. M., Bharadwaj, S. G. and Varadarajan, P. R. (1993): An Analyss of the Market Share-Proftablty Relatonshp. The Journal of Marketng, 57(3), pp Warng, G. F. (1996): Industry Dfferences n the Persstence of Frm-Specfc Returns. The Amercan Economc Revew, 86(5), pp Wess, C. R. and Wttkopp, (2005): Retaler concentraton and product nnovaton n food manufacturng. European Revew of Agrcultural Economcs, 32(2), pp Wjnands, J.H.M., van der Meulen, B.M.J. and Poppe, K.J. (eds.) (2006): Compettveness of the European Food Industry: An economc and legal assessment European Commsson, Belgum. Yurtoglu, B. B. (2004): Persstence of frm-level proftablty n Turkey. Appled Economcs, 36(6), pp