Volume 27 Issue 4 Article 2 December 2025 Shock Transmission in Granular Economies: Impact of Pass- Shock Transmission in Granular Economies: Impact of Pass- Through Effects of Idiosyncratic Microshocks to the Aggregate Through Effects of Idiosyncratic Microshocks to the Aggregate Anamarija Cijan University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia, anamarija.cijan@ef.uni- lj.si Jož e P . Damijan University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia and University of Leuven, Belgium Č rt Kostevc University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia Follow this and additional works at: https://www.ebrjournal.net/home Part of the Economics Commons Recommended Citation Recommended Citation Cijan, A., Damijan, J. P ., & Kostevc, Č . (2025). Shock Transmission in Granular Economies: Impact of Pass- Through Effects of Idiosyncratic Microshocks to the Aggregate. Economic and Business Review, 27(4), 199-220. https://doi.org/10.15458/2335-4216.1361 This Original Article is brought to you for free and open access by Economic and Business Review. It has been accepted for inclusion in Economic and Business Review by an authorized editor of Economic and Business Review. ORIGINAL ARTICLE Shock Transmission in Granular Economies: Impact of Pass-Through Effects of Idiosyncratic Microshocks to the Aggregate Anamarija Cijan a, * , Jože P . Damijan a,b , ˇ Crt Kostevc a a University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia b University of Leuven, Belgium Abstract This paper studies the importance of shocks to the largest rms on the aggregate output. Using rm-level data on eight European countries (2006–2019), we nd that shocks to the largest rms explain an important part of aggregate uctuations. Our paper brings several novelties. Firstly, in addition to the aggregate level, we extend the analysis of the transmission of rm-level shocks to study the shocks at the sectoral level. Secondly, we provide a novel measurement for demand-side shocks within granularity. We show that idiosyncratic shocks affecting the largest 20 rms can explain almost half of the output volatility, which is consistent with Gabaix (2011). Moreover, demand-side shocks contribute a greater share to this volatility compared to supply-side shocks. Finally, we show that the smaller the sample of the largest rms, the larger the propagation effect of the shocks to GDP . This suggests that a few large rms drive a large part of the aggregate volatility, while volatility of other larger rms balances out on average. Keywords: Business cycle, Supply-side shocks, Demand-side shocks, Aggregate uctuations, Granular residual JEL classication: E32, O47, C23, L25, F23 1 Introduction A ggregate uctuations have traditionally been at- tributed to factors such as monetary and scal policies, government spending, aggregate demand shifts, and technological changes (Magerman et al., 2016). These are key mechanisms in real business cy- cle and New Keynesian models (e.g., Christiano et al., 2005; Kydland & Prescott, 1982). However, macroe- conomic shocks alone do not fully explain aggregate volatility (Cochrane, 1994). Several crises, including the global nancial crisis as well as the COVID-19 pandemic, highlighted that rm-level shocks, not just large common shocks, can spread across the economy, leading to substantial aggregate movements (Mager- man et al., 2016; Stumpner et al., 2022). This paper’s motivation is thus to analyse whether rm-level shocks are able to explain aggregate uc- tuations by examining the importance of shocks to the largest rms on the aggregate output. Given the dominance of large rms in modern economies, id- iosyncratic shocks affecting these rms can result in signicant aggregate shocks. As an example, for our sample, during 2007 to 2019, the average proportion of total sales of the largest 20 and 50 rms relative to GDP across all countries stood at 12.5% and 17.9%, respectively. Hence, one can argue than large rms represent a substantial share of the macroeconomic activity, which means analysing their actions is valu- able for gaining insights into the overall economy. Several studies highlight the impact of rm-specic shocks on aggregate uctuations. Gabaix (2011) nds that idiosyncratic shocks to the top 100 U.S. rms explain up to a third of GDP uctuations. Blank et al. (2009) nd that positive shocks at large banks decrease the probability of distress of small banks. Received 24 December 2024; accepted 15 July 2025. Available online 1 December 2025 * Corresponding author. E-mail address: anamarija.cijan@ef.uni-lj.si (A. Cijan). https://doi.org/10.15458/2335-4216.1361 2335-4216/© 2025 School of Economics and Business University of Ljubljana. This is an open access article under the CC BY license (http://creativecommons.org/ licenses/by/4.0/). 200 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Freund and Pierola (2015) demonstrate that the top ve rms contribute up to 30% of nonoil exports in 32 countries, indicating that a single rm can shape the entire comparative advantage of a country. Moreover, this paper also intersects with the lit- erature that focuses on the signicance of sectoral shocks in driving aggregate uctuations, a concept pioneered by Long and Plosser (1983). The central idea is that idiosyncratic shocks impacting a sin- gle sector can yield notable aggregate effects if that sector is deeply interconnected with others through input–output linkages (see Acemoglu et al., 2012; Horvath, 1998). Theoretically, Acemoglu et al. (2012), Acemoglu et al. (2017), and Jones (2011) show that the propagation of idiosyncratic shocks and distortions through input–output linkages can lead to implica- tions for both macroeconomic volatility and economic growth. Empirically, Acemoglu et al. (2015) analyse the impact of large shocks affecting certain rms or industries by examining the input–output network. Also, Carvalho and Gabaix (2013) dene fundamental volatility as the volatility that would emerge from an economy solely composed of idiosyncratic sectoral or rm-level shocks. They show that fundamental volatility is responsible for the uctuations in macroe- conomic volatility across the major world economies. As evidence continues to indicate that the production networks in developed market economies are often controlled by a handful of “superstar” rms (Bernard et al., 2019), there remains a lack of understanding of the differences in the granular effects across countries, sectors, as well as different types of shocks. Also, our paper makes a distinction between sup- ply and demand shocks. Supply shocks impact rms’ production capacity, affecting prices, quantities of fac- tor inputs, or production technology. Such shocks cause price levels and real output to move in oppo- site directions; for example, an adverse shock raises prices and lowers output (Blinder & Rudd, 2013). Temporary negative supply shocks reduce output and employment. Though severe, recessions caused by these shocks are partially an efcient response to decreased economic capacity for goods and services (Guerrieri et al., 2022). Thus, decreased rm sales indicate a supply-side shock. Conversely, Benguria and Taylor (2020) conceptualise a demand-side shock as a reduced borrowing limit for households, lead- ing to reduced consumption, repayment of previous debts, and a consequent decline in overall demand for goods. This reduction in aggregate demand prompts rms to scale back production and subsequently de- crease their demand for intermediate inputs. In other words, lower material costs signal a demand-side shock, as also highlighted by Damijan et al. (2018). Similarly, Carvalho et al. (2016) observe that a drop in demand for a specic product compels rms to curtail their input usage, implying that material costs fall in response to reduced demand. Contrarily, when demand increases—such as through rising foreign demand—rms expand their procurement of materi- als, thereby driving up material costs (Dhyne et al., 2022). Our contributions are threefold. Firstly, as most of the literature on granularity (Blanco-Arroyo et al., 2018; Fornaro & Luomaranta, 2018; Gabaix, 2011; Konings et al., 2022; Miranda-Pinto & Shen, 2019) focuses only on one country, we contribute to the literature by examining granular aspects and cross- country differences in rm size distributions. These differences in rm size distributions lead to differ- ences in granularity and signicant disparities in how rm shocks affect aggregate uctuations across coun- tries. One can argue that countries exhibit varied rm size distributions (Poschke, 2018), and our sample re- ects these differences. Our ndings reveal that when GDP growth is weaker (Italy, Spain), shocks to the largest 20 rms affect GDP growth positively and more strongly. Conversely, in countries with stronger GDP growth (Poland, Hungary), shocks to the largest 20 rms affect GDP growth less strongly but ad- versely. Secondly, the majority of studies examine only a supply-side shock, as they use measures such as labour productivity, TFP , and sales, which depict a supply-side shock (Ebeke & Eklou, 2017). We nd that the economy experiences varying effects depending on the type of shock affecting the largest rms. Key- nesian theory suggests changes in aggregate demand signicantly impact GDP , while classical economists argue changes in aggregate demand have no effect on output (Samuelson & Nordhaus, 2010, p. 593). We address a gap in the literature by introducing a novel measurement for demand-side shocks. Our results show that granular effects are present not only in supply-side shocks but also in demand-side shocks. Also, estimates from xed-effects models in- dicate that demand-side shocks have larger effects than supply-side shocks, conrming the Keynesian view rather than the classical macroeconomic theory. Moreover, demand-side shocks play a more signif- icant role in driving output volatility compared to supply-side shocks, as evidenced by their greater ex- planatory power. Thirdly, by extending our analysis to study rm-level shocks at the sectoral level, we also provide evidence on the differences between sectors in the extent of granularity as we focus on sector-level volatility. This sectoral analysis shows that the three most granular sectors are wholesale, retail, and repair of motor vehicles, manufacturing, and construction. This paper is organised as follows. Section 2 pro- vides a literature review, followed by a theoretical ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 201 model and a calibration which displays that these granular effects matter when analysing macroeco- nomic uctuations. Section 4 presents the empirical approach as well as data used in this paper. Section 5 presents the main empirical results as well as several robustness checks. Section 6 concludes. 2 Literature review Traditional theory, specically the diversication argument based on the law of large numbers, dis- misses the possibility that substantial aggregate uc- tuations arise from individual shocks to rms or specic sectors. As Lucas (1977) argued, these shocks average out and only exert negligible effects on the overall economy. Consequently, aggregate output sta- bilises very rapidly around its mean. In an economy comprising n sectors that are subject to indepen- dent shocks, the magnitude of aggregate uctuations would be proportional to 1= p n. Hence, at highly disaggregated levels, only negligible effects are no- ticeable. This argument dismisses the existence of linkages between rms and sectors, even though these connections serve as a propagation mechanism for idiosyncratic shocks throughout the economy. In- terconnections among sectors can lead to a slower stabilisation of aggregate output around its mean compared to what is proposed by the diversication argument. This implies that sectoral shocks play a more substantial role in driving aggregate uctua- tions; a concept referred to as granularity (Acemoglu et al., 2012). Gabaix (2011) argues that when the rm size distribution has a very fat tail, aggregate volatility decreases in line with 1= ln N. Many aggregate uctu- ations can be traced back to the “grains” of economic activity, particularly to large rms; an idea known as the granular hypothesis. This hypothesis argues that idiosyncratic shocks to large rms can create nontrivial aggregate shocks impacting GDP and, via general equilibrium, all rms. An economy is deemed granular if shocks to the largest rms can induce ag- gregate uctuations (di Giovanni & Levchenko, 2012). Firm size distribution in Gabaix (2011) serves a simi- lar function to the intersectoral network in Acemoglu et al. (2012). The granular approach is used in analyses that ex- plain the uctuations of several macroeconomic indi- cators. Various studies indicate that a few large rms have a disproportionate effect on GDP uctuations. As an example, Gabaix (2011) examines U.S. data from 1951 to 2008 and nds that idiosyncratic shocks impacting the top 100 rms contributed to roughly one third of GDP uctuations. Likewise, Ebeke and Eklou (2017) investigate idiosyncratic shocks among the 100 largest rms across eight European countries from 2000 to 2013, concluding that 40 percent of GDP variation could be attributed to idiosyncratic shocks affecting these rms. Similarly, comparable effects are also found for Spain (Blanco-Arroyo et al., 2018), Fin- land (Fornaro & Luomaranta, 2018), and Australia (Miranda-Pinto & Shen, 2019). On the other hand, Wagner and Weche (2020) conclude that the Ger- man economy is not a granular economy. They nd that idiosyncratic shocks to the largest 100 rms did not seem to have a substantial impact on explaining aggregate volatility, thus contradicting a consider- able portion of the empirical evidence supporting the granular hypothesis. Furthermore, studies also explore the granular hypothesis in terms of its in- uence on aggregate sales, in addition to its effects on GDP . These studies conrm that rm-level shocks impact aggregate sales in several countries, includ- ing France (di Giovanni et al., 2014), Sweden (Friberg & Sanctuary, 2016), and Chile (Grigoli et al., 2023). Other studies conrm that other measures, such as total factor productivity (TFP), exhibit indications of granularity, including in the U.S. (Baqaee & Farhi, 2019), Ireland (Papa, 2019), and Kazakhstan (Konings et al., 2022). Lastly, the granular hypothesis is also ex- amined by analysing data from nancial institutions. Amiti and Weinstein (2018) use Japanese data from 1990 to 2010 and nd that idiosyncratic bank shocks explain 30 to 40 percent of aggregate loan and invest- ment uctuations. The literature on granularity is also closely related to the literature on the role of rm heterogeneity in explaining aggregate uctuations in outcomes such as unemployment and trade. For instance, Moscarini and Postel-Vinay (2012) examine the role of large and small employers in job creation across different business cycles. They identify a negative correlation between the net job creation rate of large employers and the level of aggregate unemployment. More- over, di Giovanni and Levchenko (2012) examine a model with heterogeneous rms that are subjected to idiosyncratic rm-specic shocks, calibrated us- ing data for the 50 largest economies globally. They discover that smaller countries are more prone to higher volatility arising from idiosyncratic shocks to large rms as they have fewer rms. Further- more, macroeconomic volatility increases with trade opening as large rms gain even more importance. Consequently, trade can trigger a 15–20% increase in aggregate volatility in some small open economies. Relatedly, Wagner (2013) uses panel data for German manufacturing exporting rms during the 2008–2009 crisis. The author shows that idiosyncratic shocks to very large rms play an important role in shaping the export collapse as the top 10 rms in an industry accounted for around one third of export uctuations. 202 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Lastly, it is crucial to differentiate the pass-through effects between small and large rms. Amiti et al. (2014) nd that large import-intensive exporters in Belgium have a 50% exchange rate pass-through, while small nonimporting rms have nearly complete pass-through. Amiti et al. (2019) also show that small rms exhibit no strategic complementarities and a complete pass-through of marginal cost shocks into their domestic prices, while large rms show strong strategic complementarities and an incomplete pass- through to their domestic prices. 3 Theoretical model and calibration This section briey presents the model proposed by Gabaix (2011) and provides a calibration of GDP volatility using Orbis data. Gabaix uses an islands economy with N rms. Production is assumed to be exogenous, and initially there are no linkages be- tween rms. The main equation of the model depicts the below expression for standard deviation of GDP growth, denoted ass GDP : s GDP D N X iD1 s 2 i S it Y t 2 ! 1 2 (1) Here, total GDP is labelled as Y t ,s i represents rm i’s volatility, and S it is the quantity of a homogenous consumption good produced by rm i in time t with- out any factor input. Hence, the variance of GDP represents the weighted sum of the variance s 2 i of idiosyncratic shocks where weights are equal to the squared share of output produced by rm i (Gabaix, 2011). Next, the examination of the 1= p N argument for the (ir)relevance of idiosyncratic shocks is presented. If one assumes there is a large number of rms, N, idiosyncratic uctuations disappear in the aggregate. Also, assuming that rms have initially identical size that is equal to 1=N of GDP and identical standard deviation (s i Ds), the standard deviation of GDP growth, denoted as s GDP , is then equal to s GDP D s p N . According to Gabaix (2011), the estimate of rm volatilitys is equal to 12%, and an economy has ND 10 6 rms. The GDP volatility per year is then equal to 0.012%. Arguably, that is far too distant from the empirically measured size of macroeconomic uctu- ations of approximately 1%, thus economists often resort to aggregate shocks. More general modelling assumptions anticipate a 1= p N scaling (Gabaix, 2011). The above proposition assumes that rm size distri- bution is thin-tailed. According to Axtell (2001), the size distribution of U.S. rms is instead well approxi- mated by the power law; more specically, it exhibits the Zipf distribution. This result applies globally, and there is an improving understanding of the origins be- hind this distribution. Accordingly, it has been shown that if the rm size distribution has fat tails, s GDP declines more slowly than 1= p N. More specically, if one knows the GDP of several countries, but not the size of their respective rms, except that, for example, they exhibit Zipf’s law, the volatility of a country of size N is proportional to 1= ln N (Gabaix, 2011). We then follow Gabaix (2011) to account for input– output linkages and for the endogenous response in inputs to initial disturbances. Our calibration shows that the effects are of the right order of magnitude to account for macroeconomic uctuations. First, as- sume an economy with N competitive rms that buy intermediary inputs from one another. Firm i exhibits Hicks-neutral productivity growth. It can be observed that TFP shocks can be calculated without knowing the input–output matrix, as the sufcient datum for the impact of rm i is its size, measured by its sales. Moreover, h is the sales Herndahl index: hD N X iD1 sales it GDP t 2 ! 1 2 (2) Moreover, the volatility of the TFP growth, denoted as s TFP , is equal to s TFP D hs p , where s p is the stan- dard deviation for growth rates of total sales and h is the sales Herndahl index. One can then examine the empirical magnitude of the key variables of the volatility of the TFP growth, namely the volatility of rm size of the largest rms. As an example, the volatility of one of the measures of growth rates in- cludes 1ln sales it . For each year one then calculates the cross-sectional variance among the largest rms of the previous year and takes the average. The volatility of GDP can be calculated as s GDP Dms p h, where m reects factor usage, s p is the standard deviation for growth rates of total sales, and h is the sales Hernd- ahl index. The following three benchmarks can be used for factor usage. Firstly, a short-term model with xed capital in the short run and the Frisch elasticity of labour supply equal to 2 yieldsmD 1:8. Secondly, assuming exible supply of capital, the value of m amounts to 4.5. Lastly, under the neoclassical growth model where TFP is assumed to follow a geometrical random walk where only capital can be accumulated in the long run, m is equal to 1.5. One can take the average of the three above values and obtainmD 2:6 (Gabaix, 2011). Fig. 1 summarises the selected measures for our sample, with the sales Herndahl index and GDP volatility displayed on the left y axis, and the stan- dard deviation of sales growth rates for the 20 largest rms shown on the right y axis. Here, the average ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 203 Fig. 1. Sales Herndahl, standard deviations for growth rates of sales, and GDP volatility (in %), by country (average for 2006–2019). Source: Orbis database. sales Herndahl is quite large for the selected Eu- ropean countries during the period 2006–2019, as it amounts to around 8.2%. Hungary (13.4%) and Slovenia (12.7%) exhibited the highest values (the number of large rms is relatively large, compared to GDP), while Italy (5.1%), Spain (5.2%), France (5.9%), Poland (7.7%), Sweden (7.8%), and Portugal (8.1%) all had values below the average of the sample (the number of small rms is relatively large, compared to GDP). Next, in order to calculate GDP volatil- ity, standard deviations for growth rates of sales for the 20 largest rms in each country were calcu- lated. For each year t, the cross-sectional variance of growth rate was calculated, s 2 t D K 1 P K iD1 g 2 it (K 1 P K iD1 g it ) 2 , with KD 20. The average standard de- viation is (T 1 P T tD1 s 2 t ) 1=2 . On average, the standard deviation for growth rates of sales for all countries is around 19.5%. The largest rms were the most volatile in Hungary, France, and Italy. In contrast, the largest rms were less volatile in Sweden, Slove- nia, Portugal, Spain, and Poland as these countries all had values below the sample average. Finally, the average GDP volatility for all countries in the sample stands at around 4.3%. Arguably, this suggests that idiosyncratic volatility is signicant enough at the macrolevel. For comparison, in Gabaix (2011) the standard de- viation for growth rates of total sales is 12%. The sales Herndahl index is quite low, amounting to 5.3%, whereas it accounts for 22% in an average over all countries. In other words, the U.S. is a country with relatively small rms compared to GDP . Thus, the granular hypothesis would likely be harder to prove. GDP volatility (s GDP ) for the U.S. amounts to 1.7%, whereas for a typical country GDP volatility amounts to 6.8%. Arguably, this is on the order of magnitude of GDP uctuations; thus, at the macrolevel idiosyn- cratic volatility appears to be quantitatively large enough to matter (Gabaix, 2011). 4 Empirical approach We test whether granular effects are observed in propagation of shocks on the set of eight Euro- pean countries. Firstly, we follow the methodology proposed by Gabaix (2011), while employing indica- tors for both supply- and demand-side shocks. The granular residual, serving as an indicator of these shocks, is derived as the weighted sum of large rms’ sales or material costs growth rate subtracted from the corresponding average growth rate across rms. Specically, changes in the sales growth rate reect a supply-side shock, whereas changes in the growth rate of material costs signify a demand-side shock. Secondly, the paper extends the analysis by investi- gating the impact of shocks in the largest rms at the 204 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 sectoral level. These shocks are dened as the annual differences in the shares of either sales (supply side) or material costs (demand side) of the largest rms in the sectoral value of both variables. This section rstly outlines the econometric ap- proach for both the aggregate granular residual and for sectoral level analysis. Secondly, descrip- tive statistics are presented. The baseline model tests whether GDP is affected by either supply- or demand- side shocks in the largest rms; shocks transmitting through the largest 20 rms in case of aggregate gran- ular residual and the largest ve rms for sectoral analysis. Robustness checks for the granular residual include additional estimations of the growth rates of sales and material costs (Eqs. (5) and (6) below). We also vary the denition of “large rms” to include 3, 5, 10, 20, 50, or 100 largest rms for the granular residual and 3, 5, and 10 largest rms for sectoral analysis. 4.1 Econometric approach This section rstly presents a parsimonious measure—granular residual—of the shocks to the 20 largest rms, ranked by their lagged sales. The main challenge lies in identifying idiosyncratic shocks. Aggregate shocks may affect large rms, rather than shocks in the largest rms driving aggregate uctuations. This “reection problem” does not have a general solution (Manski, 1993). We use various tools to measure the share of idiosyncratic shocks. Moreover, granular residuals are constructed for each country separately. Focusing on the country average growths enables the removal of the inuences stemming from the structural disparities in sectoral productivity growth across different countries (Ebeke & Eklou, 2017). While we address potential simultaneity concerns via mean differencing and xed effects, we acknowl- edge that this approach may not fully isolate causal effects. Ideally, an instrumental variable strategy or dynamic panel framework would strengthen identi- cation. Here, we follow the approach of Gabaix (2011) and Ebeke and Eklou (2017), who also rely on OLS re- gressions and a xed-effects model in similar granular residual setups. Firstly, to account for supply-side shocks, one can use the change in sales growth rate of rm i in country c, at time t, denoted as g ict (aligned with several stud- ies that examine supply-side shock as sales growth, see, e.g., Guerrieri et al., 2022): g ict D sales i;t sales i;t 1 sales i;t 1 (3) Creating the granular residual by using sales growth indicates that the source of aggregate uc- tuations lies in supply-side shocks. Nonetheless, to account for demand-side shocks one can examine a change in the material costs growth rate of rm i in country c, at time t, denoted as z ict (see Damijan et al., 2018): z ict D MC i;t MC i;t 1 MC i;t 1 (4) Rather than relying solely on the previously de- ned expressions for changes, we estimate rm-level growth rates for the largest K rms to capture both sales and material cost dynamics more effectively. Specically, we regress changes in sales and material cost growth rates on two specications, summarised in X ict : rstly, on the mean growth rates of the top K rms and their interaction with log sales (rm size), and secondly, on their interaction with the square of log sales as a robustness check, as shown below: g ict Db 0 X ict C+ ict (5) z ict Db 0 X ict C+ ict (6) These regressions allow us to extract residuals fol- lowing the methodology of Gabaix (2011), and these residuals serve as granular shocks representing the idiosyncratic component of rm-level growth. Ulti- mately, our objective is to assess whether this idiosyn- cratic component, captured in error term—denoted as + ict —can explain uctuations in GDP growth. Next, the ideal granular residual, denoted as0 ct , can be em- pirically approximated by the equation of the growth in GDP: 0 ct :D K X iD1 S ic;t 1 S c;t 1 + ict (7) The residual represents the sum of idiosyncratic rm shocks, weighted by rm relative size. Put dif- ferently, the weight is the ratio of the lagged sales of the rm (S ic;t 1 ) over the total lagged sales (S c;t 1 ), similarly as in Konings et al. (2022). Then, one needs to investigate what share of the total variance of GDP growth originates from the granular residual, as the theory states that GDP growth is g Yt Dm0 t . Importantly, + ict needs to be extracted by estimat- ing the evolution of sales growth for the largest K rms of the previous year on a vector of observables specied above. The estimate of idiosyncratic rm- level supply-side shocks can be formed asc + ict D g ict b b 0 X ict orc + ict D z ict b b 0 X ict for demand-side shocks. The granular residual is thus dened as (Gabaix, 2011; ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 205 Konings et al., 2022): 0 ct :D K X iD1 S ic;t 1 S c;t 1 c + ict (8) If the measured granular residual0 ct is close to the ideal granular residual0 ct , identication is achieved. Gabaix (2011) then presents the particularisation that is transparent and does not demand much data, while turning out to do as well as the more complicated measures. The simplest procedure is to control for the mean growth rate in the sample. Put differently, X ict D ¯ g ct or X ict D ¯ z ct , where ¯ g ct D K 1 P K iD1 g ict and ¯ z ct D K 1 P K iD1 z ict . As indicated by the equations, we compute the average across the largest K rms. In our baseline analysis, K is set to 20, corresponding to the 20 largest rms, while we also vary K in robustness checks. Next, the granular residual is the weighted sum of a rm’s growth rate difference from the av- erage growth rate (Gabaix, 2011): 0 s ct D K X iD1 S ic;t 1 S c;t 1 (g ict ¯ g ct ) (9) 0 d ct D K X iD1 S ic;t 1 S c;t 1 (z ict ¯ z ct ) (10) This adjustment helps to remove the impacts of common shocks affecting all rms and sectors in each country every year. These encompass, among other factors, policy shocks related to aggregate demand, such as scal and/or monetary policies, signicant structural reforms (Ebeke & Eklou, 2017). Lastly, the largest 20 rms are sorted according to their year- over-year lagged sales. Utilising lagged sales ensures that even if large companies face a negative shock, they remain included in the sample of large rms (Gabaix, 2011; Konings et al., 2022). Then, GDP growth (using pooled OLS estimation) is regressed on the measure of granularity (0 s ct and0 d ct , respectively) interacted with countries and on a crisis dummy, where, ct presents the error term: Y ct Db 1 Cb 2 0 s ct C b 3 0 s ct Country t Cb 4 Country t C b 5 Crisis t C, ct (11) Y ct Db 1 Cb 2 0 d ct C b 3 0 d ct Country t Cb 4 Country t Cb 5 Crisis t C, ct (12) More specically, the model tests whether the mea- sure of granularity (also interacted with country xed effects) is able to explain GDP growth. The global nancial crisis (GFC) dummy accounts for the global nancial crisis and includes the period from 2009 to 2012. We explore two types of shocks. To ensure the robustness of our analysis, we preform various tests by varying the denition of “largest” rms when computing the granular residual. Moreover, an- other robustness check involves expanding the set of explanatory variables in estimating both sales and material costs growth rates as alongside g ict and z ict and their interactions with the logarithm of rm size, their interactions with the square of the logarithm of rm size are also included. Additionally, we use a xed-effects model to estimate the average impact of both supply and demand shocks to GDP growth. Next, the paper extends the analysis by investigat- ing demand and supply shocks within the samples of largest rms at the sectoral level. Supply and demand shocks are dened as the annual differences in the shares of sales or material costs of the largest rms in the sectoral value of the corresponding variable, respectively. 1 Here, we acknowledge that sectoral dynamics may be more comprehensively captured through the use of dynamic panel models. In this study, we partially address these dynamics by in- cluding country xed effects and estimating separate regressions for each sector, thereby accounting for sector-specic trends and unobserved heterogeneity. While this approach helps mitigate concerns related to omitted dynamics, we recognise that future re- search could benet from employing dynamic panel estimators to capture cumulative and lagged effects across sectors more explicitly. The supply-side shock is dened as: t s ct D 1 K K X iD1 sales i;t sales j;t sales i;t 1 sales j;t 1 (13) Similarly, the demand-side shock is dened as: t d ct D 1 K K X iD1 MC i;t MC j;t MC i;t 1 MC j;t 1 (14) Here, i depicts the rm, j indicates the sector, and t is the year. The paper identies the KD 5 largest rms in each sector in the Orbis dataset, using the previous year’s sales and, as before, excluding rms in the oil, energy, and nance sectors. GDP growth is then regressed on this novel measure of granularity with country interactions terms and the GFC dummy for each sector individually. Y ct Db 1 Cb 2 t s ct C b 3 t s ct Country t Cb 4 Country t Cb 5 Crisis t C, ct (15) 1 Note that the values are not deated. 206 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Y ct Db 1 Cb 2 t d ct C b 3 t d ct Country t C b 4 Country t Cb 5 Crisis t C, ct (16) The model investigates the impact of two types of shocks on GDP at the sectoral level. Firstly, the model examines a supply-side shock in the largest ve rms, where the granular residual (t s ct ) is the difference in the shares of sales of the largest rms in the sectoral sales. Secondly, the model investigates a demand-side shock in the largest ve rms, where the granular residual (t d ct ) is the difference in the shares of material costs of the largest rms in the sectoral material costs. The crisis dummy represents the same period as in the previous model. To ensure the robustness of our analysis, we explore different denitions of “largest rms” in specifying measures of granularity. 4.2 Data We use rm-level data from the Orbis database (Bu- reau van Dijk) by KU Leuven. The Orbis database includes information on both listed and unlisted rms. The nancial and balance sheet data origi- nate from national business registries, which adhere to country-specic legal and administrative ling mandates. While most countries require that limited liability rms register upon formation, the criteria in terms of the rm size for reporting balance sheet details vary among countries (Kalemli-Ozcan et al., 2022). 2 With millions of rms, Orbis is a valuable re- search resource. The data include annual observations from 2006 to 2019 of the following variables: rm ID, sales, and material costs. Following Bajgar et al. (2020) we rely on unconsolidated accounts to avoid dupli- cating accounts. Additionally, many large rms do not report consolidated accounts, and the majority of rms in Orbis provide only unconsolidated accounts. Consequently, we base our analysis on unconsoli- dated data. 3 While Orbis tends to be biased towards larger rms, this bias varies across sectors. By includ- ing all sectors (except nance and mining) we aim to mitigate this bias to some extent. To address Orbis limitations, we focus on eight European countries se- lected based on specic criteria. According to Bajgar et al. (2020), these countries demonstrate relatively high coverage of aggregate employment, output, and value added, do not have rounding issues, have a low prevalence of limited nancial (LF) accounts, and feature a high percentage of rms that le accounts that are then reported to Orbis (Kalemli-Ozcan et al., 2022). Therefore, we believe that for these selected eight countries the analysis is robust. Next, we adopt a balanced sample approach, requiring rms included in the sample to be present throughout the entire pe- riod. 4 Additionally, missing, negative, or zero values for sales or material costs are replaced with linear interpolations. If after linear interpolation, there is still at least one missing value for either sales or ma- terial costs for a certain rm, that rm is excluded. Moreover, when the growth rate of sales or material costs surpasses 20% (the same threshold as Gabaix, 2011), these values are replaced with interpolated val- ues. 5 Industries are categorised using 2-digit NACE (Nomenclature of Economic Activities) Rev. 2 codes. Like other studies on granularity, we exclude rms engaged in mining (due to uctuations of worldwide commodity prices) and nancial institutions (as sales are a poor proxy for their output). Excluding these rms has minimal impact, while it is conceptually more appropriate. The ten sectors covered are agri- culture, forestry and shing; manufacturing; water supply, sewerage, waste management; construction; wholesale, retail, and repair of motor vehicles; trans- portation and storage; accommodation and food ser- vices; information and communication; real estate; and scientic technical and other business activities. Lastly, data for GDP growth rates are obtained from the IMF World Economic Outlook Database. Table 1 provides data on average sales, material costs, as well as number of rms in the sample during the 2006–2019 period. For the largest 20 rms across analysed countries, the rankings are based on lagged sales, mirroring the approach used in the empirical analysis. The full data consist of approximately 5.6 million rms on an annual basis. One can see that there are notable differences in sales and material costs between average rms and the largest rms. Descriptive data for the largest 50 rms can be found in the appendix (Table A1). Table 2 shows the skewness and kurtosis when all rms in the sample are plotted according to their size (based on log sales) in year 2019 (the last year in the sample). 6 Results indicate that skewness for selected countries lies around 0.076, varying from 0.392 for Italy to 0.14 for Slovenia, indicating the distributions are slightly skewed. All countries 2 For detailed information on which rms are excluded in particular countries, please refer to Kalemli-Ozcan et al. (2022), Table A.6.1. 3 Should duplicate accounts with the same ID still exist and one of them has the consolidation code U2, the other accounts are removed. 4 As some rms do not report their nancials every year, excluding them only due to inavaliblity of data can disort the results. 5 Several studies using Orbis data apply interpolation and imputation to improve data quality (Fujimoto et al., 2022; Gal, 2013; Kalemli-Ozcan et al., 2022). These approaches help ensure more reliable rm-level analysis from Orbis data and make the results more robust. For possible biases please refer to Kalemli- Ozcan et al. (2022). 6 A normal distribution is characterised by a skewness of 0 and a kurtosis of 3. ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 207 Table 1. Descriptive statistics for all and for largest 20 rms, by country. All rms Largest 20 rms Material costs (in million Material costs (in million Sales (in million EUR) EUR) No. of rms Sales (in million EUR) EUR) All countries 3.183 2.398 5,579,619 3845 2360 Spain 2.664 (69.906) 1.868 (59.191) 764,040 6427 (3653) 4457 (2964) France 3.575 (132.933) 2.650 (113.248) 788,813 13,212 (12,913) 7927 (10,651) Hungary 0.950 (29.837) 6.929 (74.298) 353,316 758 (546) 510 (513) Italy 2.762 (86.567) 1.959 (71.363) 1,637,572 5159 (4953) 3019 (3819) Poland 11.006 (430.934) 3.154 (43.874) 1,097,413 2441 (3568) 1198 (1444) Portugal 1.133 (26.984) 0.874 (26.141) 279,831 1184 (866) 709 (812) Sweden 2.476 (59.057) 1.195 (17.067) 565,874 1072 (579) 713 (506) Slovenia 0.900 (17.286) 0.558 (14.822) 92,761 503 (355) 343 (302) Note. Mean values (standard deviation). Source: Orbis database. Table 2. Skewness and kurtosis based on log sales for 2019, by country. Skewness Kurtosis p value for Shapiro–Wilk test p value for Shapiro–Francia test Spain 0.058 4.42 .00000 .00001 France 0.113 3.70 .00000 .00001 Hungary 0.166 3.78 .00000 .00001 Italy 0.392 5.47 .00000 .00001 Poland 0.140 2.90 .00000 .00001 Portugal 0.008 4.40 .00000 .00001 Sweden 0.113 3.88 .00000 .00001 Slovenia 0.140 4.23 .00000 .00001 Source: Orbis database. (except Poland) exhibit excess kurtosis (> 3), indicat- ing a nonnormal, leptokurtic, distribution. Leptokur- tic distributions are characterised by a higher peak, thinner “shoulders,” and fatter tails. This can also be seen from Fig. A1 in the appendix, which shows rm size distributions (based on log sales) for year 2019 for each country. The presence of these fat tails, as can be seen from our data, causes the central limit the- orem to break down, allowing idiosyncratic shocks to large rms to inuence aggregate outcomes. In turn, this implies that aggregate outcomes can be dis- proportionately driven by a few large, idiosyncratic shocks, including those to large rms, sectors, or busi- ness groups, which do not necessarily average out at the macrolevel (Gabaix, 2011). Moreover, to mitigate heteroskedasticity, we applied two standard correc- tions. First, we used robust standard errors clustered at country level to ensure valid inference. Second, we log-transformed rm size to reduce skewness and stabilise variance, minimising the inuence of large rms and improving model t. Next, the largest rms are segmented into ten distinct sectors based on their area of activity, as indi- cated by their respective NACE codes. These largest 20 rms predominantly operate in manufacturing and wholesale, retail, and repair of motor vehicles as these rms comprise just over 74% of all rms across all countries. The gure of each country’s sectoral de- composition over the examined period is provided in the appendix (Fig. A2). This gure displays the distri- bution of the 20 largest rms by sector in each country over time and illustrates how the dominant indus- tries among leading rms have shifted over time, reecting broader economic and structural changes within each country. Lastly, Fig. A3 (in the appendix) shows GDP growth alongside two measures of gran- ular residuals—the traditional supply-side version based on Gabaix (2011) and our novel demand- side measure—across our eight European countries from 2007 to 2019. The residuals are calculated us- ing data from the largest 20 rms in each country. In our selected sample, most prominently in Spain, 208 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 France, Italy, Poland, and Sweden, both residuals track GDP growth closely, with visible comovement during downturns and recoveries. Overall, the g- ure supports the idea that rm-level shocks can play a meaningful role in macroeconomic dynamics, and that both shock measures may be relevant in explain- ing some country-specic business cycle patterns. 5 Empirical results This section presents econometric results derived from estimating the granular residual models (11) and (12) at the aggregate and sectoral levels, xed-effects model, as well as robustness checks. Firstly, to calcu- late the granular residual, we identify the 20 largest rms by country and year based on the previous year’s sales. To handle outliers in the database, we follow Gabaix (2011) and winsorise the extreme de- meaned growth rates at 20%. 7 Secondly, the analysis is extended by investigating shocks within the largest rms at the sectoral level. 5.1 Granular residual This section presents results derived from estimat- ing the granular residual model (xed-effects model and Eqs. (11) and (12)). We test whether GDP is affected by shocks in the largest 20 rms. Table 3 presents estimates of the effects of the granular resid- ual on GDP growth (coefcients of granular residual from the xed-effects model, coefcient b 3 for in- teractions and coefcient b 4 for country dummies in Eqs. (11) and (12)). These regressions support the granular hypothesis. The model’s explanatory power is reasonably high, at 45.6% for supply-side shocks and 48.4% for demand-side shocks. Impor- tantly, demand-side shocks play a more prominent role in driving output volatility as compared to supply-side shocks. As shown by the estimates of the xed-effects model, the impact of the average demand-side shock across all countries is signicant as well as larger than that of the supply-side shock. Next, pooled OLS re- gression indicates that Spain (our controlling country) exhibits positive coefcients, suggesting that shocks in the largest rms have procyclical effects on GDP growth. Relative to Spanish granular effects France, Italy, Sweden, and Slovenia also show procyclical ef- fects on GDP growth. Conversely, for Portugal we nd negative coefcients, indicating countercyclical effects of the largest rms on GDP growth. This might be due to severe effects encountered during the GFC or due to a specic sectoral structure of large rms (composition effect). Supply-side shocks in the largest 20 rms have countercyclical effects in Hungary and Poland, while the opposite holds for demand-side shocks. During the observed period, 2007–2019, the aver- age GDP growth rates were lower in Italy, Portugal, Spain, and France, as compared to Poland, Sweden, Hungary, and Slovenia. Granular effects of countries with lower average GDP growth rates (Italy, Portu- gal, Spain) are the strongest, while granular effects of countries with higher average GDP growth rates (Poland, Hungary, and Sweden) are the weakest. Fig. 1 shows the relative importance of large rms (sales Herndahl) as well as volatility of large rms. For Italy, Spain, and France (countries with a more equal size distribution), empirical evidence suggests that granular effects are even stronger compared to other countries. On the other hand, even though Hun- gary and Slovenia are countries where, relative to GDP , large rms tend to be more dominant, the gran- ular effects are relatively weak compared to other countries. This could reect institutional or structural differences beyond the scope of this paper’s analysis. 5.2 Firm-level shocks at the sectoral level This section provides results obtained from sectoral analysis by estimating models (15) and (16). Note that here we test whether GDP is affected by shocks in the largest ve rms (instead of 20 rms) at the sectoral level. The full results are reported in Tables A2 and A3 in the appendix. Here, we rst describe all results and then proceed with sectors where the strongest gran- ular effects are observed. These regressions mostly support the granular hypothesis. Firstly, the extent of granularity differs across coun- tries, with Spain continuing to be the controlling country. The largest granular effects (the highest aver- age coefcients for both shocks relative to Spanish ef- fects) are observed in Italy and Sweden, and the weak- est in Poland and France. The extent of granularity also differs across sectors; the largest granular effects are observed in wholesale, retail, and repair of motor vehicles, followed by manufacturing and construc- tion. On the other hand, the weakest effects are esti- mated in real estate; water supply, sewerage, waste management; and scientic technical and other busi- ness activities. Cross-sectoral heterogeneity arises from differences in market concentration, interrm 7 When estimating c + ict , we winsorised c + ict at MD 20%. That is, by replacing it with T(c + ict ), if T(x)D x ifjxj M and T(x)D sign(x)M ifjxj> M. We also performed the regression without data winsorisation. Overall, the results are similar in signicance, magnitude, and direction. Nonetheless, the winsorised data approach provides greater explanatory power. ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 209 Table 3. Coefcients of the granular residual for 20 largest rms, by country. Supply-side shocks in largest 20 rms Demand-side shocks in largest 20 rms FE model Granular residual 0.603 0.926 (0.780) (0.356) Pooled OLS model Interactions Spain 7.406 7.336 (0.760) (0.424) France 6.211 6.840 (0.536) (0.412) Hungary 7.920 6.081 (0.944) (0.427) Italy 4.083 7.016 (0.112) (0.254) Poland 7.873 7.246 (0.669) (0.450) Portugal 10.047 8.279 (1.146) (0.796) Sweden 7.169 6.679 (0.535) (0.409) Slovenia 6.871 5.588 (0.008) (0.171) Country dummies France 0.118 0.592 (0.025) (0.017) Hungary 1.069 1.350 (0.040) (0.012) Italy 0.829 0.774 (0.035) (0.026) Poland 3.285 3.289 (0.014) (0.014) Portugal 0.116 0.127 (0.011) (0.038) Sweden 1.118 1.304 (0.043) (0.015) Slovenia 1.035 0.929 (0.124) (0.009) Global nanical crisis dummy 2.380 2.480 (0.779) (0.572) Constant 1.505 1.408 (0.249) (0.189) Observations 104 104 R 2 .456 .484 Note. Dependent variable: GDP growth. FE model, pooled OLS estimations. Reported coefcients for interactions are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. Standard errors in the parentheses. p < .01. p < .05. * p < .1. linkages, and sector-specic dynamics. Highly con- centrated and interconnected sectors (manufacturing and motor vehicle trade) are more exposed to rm- level shocks (di Giovanni et al., 2014; Eurostat, 2021; Gabaix, 2011). Di Giovanni et al. (2014) show that input–output linkages amplify shock transmission within sectors. The construction sector’s sensitivity stems from its cyclical nature, driven by interest rates, housing demand, and public investment (van Sante, 2023). Shocks to large construction rms can be par- ticularly impactful, with employment over twice as volatile as the cyclical component of GDP (Sun et al., 2013). Next, the nature of different shock types can be examined. Across all countries, just above 60% of all coefcients have positive values, indicating shocks at the largest rms have a procyclical effect on GDP growth, which is mostly true for manufacturing, while the accommodation and food services sector ex- hibits countercyclical effects on GDP growth. Shocks at the largest rms in Portugal mostly have coun- tercyclical effects on GDP growth across all sectors, 210 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Table 4. Supply- and demand-side shocks in the largest ve rms at the sectoral level for selected sectors, by country. Supply-side shock Demand-side shock Manufacturing Construction Motor vehicles Manufacturing Construction Motor vehicles ES 0.416 0.752 0.352 0.675 0.143 1.358 (0.134) (0.143) (0.594) (0.093) (0.058) (0.314) FR 0.034 0.112 0.354 0.274 0.258 0.778 (0.066) (0.084) (0.517) (0.050) (0.176) (0.241) HU 0.813 0.762 1.296 0.949 0.353 1.787 (0.221) (0.110) (0.520) (0.165) (0.049) (0.262) IT 0.293 0.465 6.318 0.036 0.048 4.814 (0.051) (0.016) (0.861) (0.061) (0.132) (0.453) PL 0.640 1.053 0.743 0.330 0.241* 2.033 (0.089) (0.087) (0.353) (0.101) (0.124) (0.168) PT 0.759 0.668 1.332 0.037 0.543 2.410 (0.095) (0.021) (0.264) (0.049) (0.051) (0.130) SE 2.629 3.607 1.406 0.368 3.278 0.631 (0.504) (0.207) (0.593) (0.104) (0.176) (0.293) SI 0.286 1.113 2.364 0.150 0.577 2.976 (0.062) (0.185) (0.440) (0.037) (0.072) (0.190) GFC 2.392 2.772 2.193 2.313 2.809 2.310 (0.615) (0.742) (0.639) (0.622) (0.677) (0.617) Const. 1.321 0.973 1.381 1.321 1.442 1.532 (0.241) (0.359) (0.238) (0.219) (0.173) (0.235) Obs. 96 96 96 96 96 96 R 2 .471 .473 .531 .473 .496 .545 Note. Dependent variable: GDP growth. Pooled OLS estimations. Standard errors in the parentheses. GFCD global nancial crisis dummy. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. p < .01. p < .05. p < .1. while the opposite holds for Sweden, which is con- sistent with the ndings on the granular residual, presented in Section 5.1. Secondly, based on the results of this analysis, the strongest signicant granular effects are observed in the following three sectors: wholesale, retail, and re- pair of motor vehicles; manufacturing; and construc- tion (Table 4). Therefore, these sectors are analysed further, although other sectors also exhibit signicant, yet less strong values. Results show that PIGS (Portugal, Italy, Spain) countries are most adversely affected by shocks in the largest ve construction rms. Hungary, Poland, and Slovenia (as well as Portugal) are most adversely af- fected by shocks in the largest ve rms in wholesale, retail, and repair of motor vehicles, while GDP growth in Hungary is also affected by shocks in large manu- facturing rms. In France, supply-side shocks in the largest rms do not show signicant effects on GDP growth, while demand-side shocks show procyclical effects in manufacturing and wholesale, retail, and re- pair of motor vehicles. In Sweden, all granular effects exhibit procyclical effects. Based on these results, we can divide these eight countries into three groups. The rst group consists of France and Sweden; the second group includes Portugal, Italy, and Spain (PIGS countries), and the third group is comprised of Poland, Slovenia, and Hungary. Looking at the data, in manufacturing, the granular effects are negative only in Hungary. These negative granular effects are, for example, observed in 2009, when Hungarian GDP fell by 6.6%, whereas the share of the largest ve rms in sectoral sales and material costs increased by 3.9% and 3.8%, re- spectively. Even though GDP decreased, the largest rms still performed well. This indicates that the largest manufacturing rms have on average a coun- tercyclical effect on GDP growth. However, the latter was more affected by sluggish performance of other parts of the economy. In construction, the granular effects are negative mostly in Spain, Italy, and Por- tugal. In Spain, these negative granular effects are observed especially in 2015, where Spanish GDP in- creased by 3.8%, whereas the share of the largest ve rms in sectoral sales and material costs de- creased by 2.8% and 2.2%, respectively. This indicates that even though GDP growth already started to ac- celerate, negative shocks in the largest construction rms still dragged down the rate of GDP growth. In Portugal, these negative granular effects are ob- served especially in 2016, when Portuguese GDP increased by 2%, whereas the share of the largest ve rms in sectoral sales and material costs de- creased by 2.9% and 2.2%, respectively. In Italy, these ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 211 Table 5. Coefcients of supply-side granular residuals for 3, 5, 10, 20, 50, and 100 largest rms, by country. Largest 3 rms Largest 5 rms Largest 10 rms Largest 20 rms Largest 50 rms Largest 100 rms Spain 1.750 2.278 2.725 7.406 5.735 1.000 France 9.053 5.220 0.091 6.211 4.348 0.235 Hungary 1.746 2.414 4.005 7.920 6.880 1.594 Italy 9.608 5.620 0.033 4.083 2.929 5.241 Poland 0.457 1.004 3.444 7.873 6.447 1.667 Portugal 7.579 1.862 6.202 10.047 6.194 1.700 Sweden 29.679 4.996 9.211 7.169 6.922 3.314 Slovenia 1.171 3.093 2.545 6.871 7.902 5.550 Observations 104 104 104 104 104 104 R 2 .473 .455 .491 .456 .471 .513 Note. Dependent variable: GDP growth. Pooled OLS estimations. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. p < .05. Table 6. Coefcients of demand-side granular residuals for 3, 5, 10, 20, 50, and 100 largest rms, by country. Largest 3 rms Largest 5 rms Largest 10 rms Largest 20 rms Largest 50 rms Largest 100 rms Spain 6.513 2.135 4.691 7.336 2.477 1.050 France 2.949 1.08 4.133 6.840 1.143 0.901 Hungary 5.714 1.514 5.028 6.081 1.325 0.457 Italy 0.571 1.664 4.733 7.016 2.641 1.289 Poland 6.336 3.969 4.826 7.246 3.142 1.655 Portugal 7.403 2.124 5.788 8.279 2.750 1.364 Sweden 20.606 9.083 6.731 6.679 1.885 0.933 Slovenia 5.991 0.207 0.361 5.588 3.205 1.971 Observations 104 104 104 104 104 104 R 2 .451 .434 .480 .484 .508 .513 Note. Dependent variable: GDP growth. Pooled OLS estimations. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. p < .05. negative granular effects are most prominent in 2013, when GDP fell by 3%. Nevertheless, the share of the largest ve rms increased by 3.2%. In wholesale, retail, and repair of motor vehicles, the granular ef- fects are negative in Hungary, Poland, Portugal, and Slovenia. 5.3 Robustness checks Additional tests for the robustness of granular residual analysis involve further estimations of sales and material cost growth rates, considering their in- teractions with rm size and its squared value. Also, sample sizes vary, encompassing the largest 3, 5, 10, 20, 50, and 100 rms for the granular residual, and the largest 3, 5, and 10 rms for sectoral analysis. Firstly, robustness checks are performed for the granular residual. For instance, when estimating the growth rates of sales and material costs (denoted as g it and z it , respectively) the model incorporates not only g ict and z ict and their interaction with rm size but also their interaction with rm size and its squared value. These estimations produce consistent results. More- over, using a larger sample of top rms with KD 50 and KD 100 of rms provides similar results as well. These results strongly support the granular hypothe- sis. On the other hand, using a smaller sample of top rms, KD 3, KD 5, or KD 10, yields lower explana- tory powers yet comparable results. Also note that one needs to be cautious in interpreting results from highly selective rm samples, for example, K 5, as in these cases country coefcients exhibit somewhat lower levels of signicance, especially for demand- side shocks. Tables 5 and 6 present these results. On average, countries exhibit stronger granular effects (higher coefcients) when a smaller sample of top rms is included. This indicates that a few largest rms affect GDP the most. Accordingly, as the sam- ple size of the largest rms decreases, the impact of shocks on GDP increases, meaning that the volatility of a few of the largest rms signicantly contributes to aggregate volatility, whereas the uctuation of other large rms balances out on average. Nonetheless, there is a drop in the number of statistically signicant coefcients when analysing the effect of 3 and 5 rms (especially for demand-side shocks). This implies that it might be suitable to analyse at least the 10 largest rms in an economy. Secondly, dening largest rms by sector with KD 3 and KD 10 of rms produces consistent 212 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 results (see Tables A4 and A5 in the appendix). Sim- ilar as before, we analyse the following three sectors: wholesale, retail, and repair of motor vehicles, man- ufacturing, and construction. On average, countries exhibit the strongest granular effects (the highest co- efcients) when the 3 largest rms are analysed. This indicates that the largest three rms affect GDP the most. For both shock types, these largest rms have mostly procyclical effects on GDP growth. These re- sults are aligned with the ndings on the granular residual. 6 Conclusion This paper demonstrates that idiosyncratic shocks to large rms can generate signicant aggregate volatility, consistent with the breakdown of the cen- tral limit theorem in the presence of a fat-tailed rm size distribution. Using theory, calibration, and em- pirical evidence, we show that individual shocks affecting large rms are a key driver of business-cycle uctuations as they explain a signicant portion of aggregate volatility. Our analysis makes several contributions. First, we highlight cross-country differences in rm size distri- butions, extending the largely single-country focus of existing literature. Second, we introduce a novel mea- sure of demand-side shocks—complementing the literature’s focus on supply-side shocks—and nd that demand shocks have stronger effects on output volatility. Third, we examine sectoral-level granular- ity and show that sectors such as manufacturing, construction, and wholesale—where large rms are dominant and interlinked—exhibit the most pro- nounced granular effects. Finally, robustness tests conrm that the smaller the subset of top rms con- sidered, the greater their shock propagation to GDP . We conclude that shocks to the largest 20 rms ac- count for nearly half of aggregate output volatility and additionally, we show that demand-side shocks account for a larger portion of output volatility than supply-side shocks. These ndings have important implications for both macroeconomic modeling and policy. Tradi- tional models that rely on representative agents or homogenous sectors may underestimate the impact of rm-specic dynamics. Incorporating granular resid- uals into forecasting models could improve their abil- ity to predict uctuations. From a policy perspective, understanding which rms and sectors dispropor- tionately drive volatility can improve the design of targeted interventions. For instance, monetary pol- icy may have outsized effects in sectors such as construction, while scal policies such as targeted in- frastructure spending could be more efcient than broad-based stimulus. Moreover, antitrust and indus- trial policy should account for systemic risks posed by dominant rms whose shocks can ripple through the economy. Firm-level data should therefore play a central role in macroeconomic surveillance. While our results are robust, several limitations remain. First, cross-country differences could be fur- ther explored by accounting for institutional and structural factors, including market rigidities and reg- ulatory environments, which may inuence the trans- mission of rm-level shocks. Second, the identica- tion strategy for demand-side shocks—using material costs as a proxy—may not fully capture the complex- ity of rm-specic demand uctuations; alternative proxies would help validate and strengthen our nd- ings and enrich the analysis. Thirdly, our analysis acknowledges cross-sectoral differences in the impact of idiosyncratic shocks. However, fully accounting for sector-specic characteristics, including varying pro- duction technologies and market structures, remains unexplored. These heterogeneities could inuence how shocks propagate within and across sectors. Fu- ture research could therefore pursue more detailed sectoral and country-specic modeling alongside the use of additional demand-side shock proxies. Declarations The authors did not receive support from any organisation for this study. The authors have no com- peting interests to declare that are relevant to the content of this article. References Acemoglu, D., Carvalho, V . M., Ozdaglar, A., & Tahbaz-Salehi, A. (2012). The network origins of aggregate uctuations. Economet- rica, 80(5), 1977–2016. https://doi.org/10.3982/ECTA9623 Acemoglu, D., Ozdaglar, A., & Tahbaz-Salehi, A. (2015). Networks, shocks, and systemic risk (Working Paper 20931). National Bureau of Economic Research. https://doi.org/10.3386/w20931 Acemoglu, D., Ozdaglar, A., & Tahbaz-Salehi, A. (2017). Microeco- nomic origins of macroeconomic tail risks. American Economic Review, 107(1), 54–108. https://doi.org/10.1257/aer.20151086 Amiti, M., Itskhoki, O., & Konings, J. (2014). Importers, exporters and exchange rate disconnect. American Economic Review, 104(7), 1942–1978. https://doi.org/10.1257/aer.104.7.1942 Amiti, M., Itskhoki, O., & Konings, J. (2019). International shocks, variable markups and domestic prices. Review of Economic Stud- ies, 86(6), 2356–2402. https://doi.org/10.1093/restud/rdz005 Amiti, M., & Weinstein, D. E. (2018). How much do idiosyncratic bank shocks affect investment? Evidence from matched bank- rm loan data. Journal of Political Economy, 162(2), 525–587. https://doi.org/10.1086/696272 Axtell, R. (2001). Zipf distribution of U.S. rm sizes. Science, 293(5536), 1818–1820. https://doi.org/10.1126/science.1062081 Bajgar, M., Berlingieri, G., Calligaris, S., Criscuolo, C., & Timmis, J. (2020). Coverage and representativeness of Orbis data (OECD Science, Technology and Industry Working Papers 2020/06). Organisation for Economic Co-operation and Development. https://doi.org/10.1787/c7bdaa03-en ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 213 Baqaee, D. R., & Farhi, E. (2019). The macroeconomic impact of microeconomic shocks: Beyond Hulten’s theorem. Econometrica, 87(4), 1155–1203. https://doi.org/10.3982/ECTA15202 Benguria, F., & Taylor, A. M. (2020). After the panic: Are nancial crises demand or supply shocks? Evidence from international trade. American Economic Review: Insights, 2(4), 509–526. https:// doi.org/10.1257/aeri.20190533 Bernard, A., Dhyne, E., Magerman, G., Manova, K., & Moxnes, A. (2019). The origins of rm heterogeneity: A production network approach (Working Paper 25441). National Bureau of Economic Research. https://doi.org/10.3386/w25441 Blanco-Arroyo, O., Ruiz-Buforn, A., Vidal-Tomás, D., & Alfarano, S. (2018). On the determination of the granular size of the econ- omy. Economics Letters, 173, 35–38. https://doi.org/10.1016/j .econlet.2018.08.020 Blank, S., Buch, C. M., & Neugebauer, K. (2009). Shocks at large banks and banking sector distress: The banking granular resid- ual. Journal of Financial Stability, 5(4), 353–373. https://doi.org/ 10.1016/j.jfs.2008.12.002 Blinder, A. S., & Rudd, J. B. (2013). The supply-shock explanation of the Great Stagation revisited. In M. D. Bordo & A. Orphanides (Eds.), The Great Ination: The rebirth of modern central banking (pp. 119–175). National Bureau of Economic Research. Carvalho, V ., & Gabaix, X. (2013). The great diversication and its undoing. American Economic Review, 103(5), 1697–1727. https:// doi.org/10.1257/aer.103.5.1697 Carvalho, V . M., Nirei, M., Saito, Y. U., & Tahbaz-Salehi, A. (2016). Supply chain disruptions: Evidence from the Great East Japan Earthquake. Becker-Friedman Institute working paper 2017-01. http://dx.doi.org/10.2139/ssrn.2893221 Christiano, L., Eichenbaum, M., & Evans, C. (2005). Nominal rigidi- ties and the dynamic effects of a shock to monetary policy. Journal of Political Economy, 113(1), 1–45. https://doi.org/10 .1086/426038 Cochrane, J. (1994). Shocks. Carnegie-Rochester Conference Series on Public Policy, 41, 295–364. https://doi.org/10.1016/0167 -2231(94)00024-7 Damijan, J., Kostevc, ˇ C., & Polanec, S. (2018). The role of granularity in propagation of a micro demand shocks in Slovenia [Mimeo]. School of Economics and Business, University of Ljubljana. Dhyne, E., Kikkawa, A.K., Komatsu, T., Mogstad, M., & Tintelnot, F. (2022). Firm responses and wage effects of foreign demand shocks with xed labor costs and monopsony (Working Paper 30447). Na- tional Bureau of Economic Research. https://doi.org/10.3386/ w30447 di Giovanni, J., & Levchenko, A. A. (2012). Country size, interna- tional trade, and aggregate uctuations in granular economies. Journal of Political Economy, 120(6), 1083–1132. https://doi.org/ 10.1086/669161 di Giovanni, J., Levchenko, A. A., & Mejean, I. (2014). Firms, destinations, and aggregate uctuations. Econometrica, 82(4), 1303–1340. https://doi.org/10.3982/ECTA11041 Ebeke, C., & Eklou, K. M. (2017). The granular origins of macroe- conomic uctuations in Europe (Working paper WP/17/229). International Monetary Fund. Eurostat (2021). Key gures on European business, industry. https://ec.europa.eu/eurostat/cache/htmlpub/key_gures _on_european_business_2021/industry.html Fornaro, P ., & Luomaranta, H. (2018). Aggregate uctuations and the effect of large corporations: Evidence from Finnish monthly data. Economic Modelling, 70, 245–258. https://doi.org/10.1016/ j.econmod.2017.11.012 Freund, C., & Pierola, M. D. (2015). Export superstars. Review of Eco- nomics and Statistics, 97(5), 1023–1032. https://doi.org/10.1162/ REST_a_00511 Friberg, R., & Sanctuary, M. (2016). The contribution of rm-level shocks to aggregate uctuations: The case of Sweden. Economics Letters, 147, 8–11. https://doi.org/10.1016/j.econlet.2016.08.009 Fujimoto, S., Takayuki, M., & Ishikawa, A. (2022). Interpola- tion of non-random missing values in nancial statements’ big data using CatBoost. Journal of Computational Social Science, 5(2), 1281–1301. https://doi.org/10.1007/s42001-022 -00165-9 Gabaix, X. (2011). The granular origins of aggregate uctua- tions. Econometrica, 79(3), 733–772. https://doi.org/10.3982/ ECTA8769 Gal, P . N. (2013). Measuring total factor productivity at the rm level using OECD-ORBIS (OECD Economics Department Working Papers No. 1049). Organisation for Economic Co-operation and Development. https://doi.org/10.1787/5k46dsb25ls6-en Grigoli, F., Luttini, E., & Sandri, D. (2023). Idiosyncratic shocks and aggregate uctuations in an emerging market. Journal of Development Economics, 160, Article 102949. https://doi.org/10 .1016/j.jdeveco.2022.102949 Guerrieri, V ., Lorenzoni, G., Straub, L., & Werning, I. (2022). Macroeconomic implications of COVID-19: Can negative sup- ply shocks cause demand shortages? American Economic Review, 112(5), 1437–1474. https://doi.org/10.1257/aer.20201063 Horvath, M. (1998). Cyclicality and sectoral linkages: Aggregate uctuations from sectoral shocks. Review of Economic Dynamics, 1(4), 781–808. https://doi.org/10.1006/redy.1998.0028 Jones, C. I. (2011). Misallocation, economic growth, and input-output economics (Working Paper 16742). National Bureau of Economic Research. https://doi.org/10.3386/w16742 Kalemli-Ozcan, S., Sorensen, B., Villegas-Sanchez, C., Volosovych, V ., & Yesiltas, S. (2022). How to construct nationally representa- tive rm level data from the Orbis global database: New facts and aggregate implications (Working Paper 21558). National Bureau of Economic Research. https://doi.org/10.3386/w21558 Konings, J., Sagyndykova, G., Subramanian, V ., & Volckaert, A. (2022). The granular nature of emerging market economies: The case of Kazakhstan. The Economics of Transition and Institutional Change, 31(2), 429–464. https://doi.org/10.1111/ecot.12346 Kydland, F., & Prescott, E. (1982). Time to build and aggregate uctuations. Econometrica, 50(6), 1345–1370. https://doi.org/10 .2307/1913386 Long, J. B., Jr., & Plosser, C. I. (1983). Real business cycles. The Jour- nal of Political Economy, 91(1), 39–69. https://doi.org/10.1086/ 261128 Lucas, R. (1977). Understanding business cycles. Carnegie-Rochester Conference Series on Public Policy, 5, 7–29. https://doi.org/10 .1016/0167-2231(77)90002-1 Magerman, G., De Bruyne, K., Dhyne, E., & Van Hove, J. (2016). Heterogeneous rms and the micro origins of aggregate uctuations (Working Paper No 312). National Bank of Belgium. Manski, C. F. (1993). Identication of endogenous social effects: The reection problem. Review of Economic Studies, 60, 531–542. https://doi.org/10.2307/2298123 Miranda-Pinto, J., & Shen, Y. (2019). A granular view of the Aus- tralian business cycle. The Economic Record, 95(311), 407–424. https://doi.org/10.1111/1475-4932.12495 Moscarini, G., & Postel-Vinay, F. (2012). The contribution of large and small employers to job creation in times of high and low unemployment. American Economic Review, 102(6), 2509–2539. https://doi.org/10.1257/aer.102.6.2509 Papa, J. (2019). What is behind aggregate productivity growth in Ireland? A granular approach [Research paper]. Department of Business Enterprise and Innovation. Poschke, M. (2018). The rm size distribution across countries and skill-biased change in entrepreneurial technology. American Eco- nomic Journal: Macroeconomics, 10(3), 1–41. https://doi.org/10 .1257/mac.20140181 Samuelson, P . A., & Nordhaus, W.D. (2010). Economics (19th edi- tion). Douglas Reiner. Stumpner, S., Gaulier, G., Carluccio, J., Fontagné, L., & Bricongne, J.-C. (2022, July 27). Large rms react more strongly to macro shocks, and it matters. VoxEU column. Sun, Y. M., Mitra, P ., & Simone, A. (2013). The driving force be- hind the boom and bust in construction in Europe (Working paper WP/13/181). International Monetary Fund. https://doi.org/ 10.5089/9781484392485.001 van Sante, M. (2023, September 4). Europe’s construction sector set to slow as demand plummets. ING Think. https://think.ing.com/articles/european-construction-sector -slows-as-demand-plummets/ 214 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Wagner, J. (2013). The granular nature of the great export collapse in German manufacturing industries, 2008/2009. Economics, 7(1), Article 20130005. https://doi.org/10.5018/economics-ejournal .ja.2013-5 Wagner, J., & Weche, J. P . (2020). On the granularity of the German economy—First evidence from the top 100 companies panel database. Applied Economics Letters, 27(21), 1768–1771. https:// doi.org/10.1080/13504851.2020.1722790 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 215 Appendix Table A1. Descriptive statistics for largest 50 rms, by country. Mean value (SD) Sales (in million EUR) Material costs (in million EUR) All countries 2212 1387 Spain 3839 (3163) 2658 (2450) France 7207 (9539) 4410 (7358) Hungary 410 (450) 265 (385) Italy 3263 (3505) 1967 (2614) Poland 1289 (2445) 698 (1007) Portugal 707 (676) 447 (565) Sweden 695 (483) 457 (398) Slovenia 290 (285) 195 (228) Source: Orbis database. 216 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Fig. A1. Kernel density plots of log sales for 2019, by country. Source: Orbis database. ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 217 Fig. A2. Sectoral composition of the largest 20 rms over time, by country. Source: Orbis database. Fig. A3. GDP growth and granular residuals over time, by country. Source: Orbis database. 218 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Table A2. Supply-side shocks in largest 5 rms at the sectoral level, by country. 1 2 3 4 5 6 7 8 9 10 ES 0.220 0.416 0.125 0.752 0.352 0.188 0.080 0.320 0.109 0.916 (0.007) (0.134) (0.035) (0.143) (0.594) (0.040) (0.026) (0.032) (0.030) (0.091) FR 0.069 0.034 0.534 0.112 0.354 0.195 1.314 1.201 0.182 0.727 (0.179) (0.066) (0.228) (0.084) (0.517) (0.033) (0.236) (0.073) (0.064) (0.087) HU 0.154 0.813 0.219 0.762 1.296 0.524 0.916 1.329 0.086 0.990 (0.082) (0.221) (0.003) (0.110) (0.520) (0.040) (0.073) (0.034) (0.086) (0.119) IT 0.869 0.293 0.275 0.465 6.318 1.062 0.006 0.296 0.258 0.500 (0.135) (0.051) (0.100) (0.016) (0.861) (0.038) (0.009) (0.051) (0.306) (0.295) PL 0.025 0.640 0.894 1.053 0.743 0.515 0.480 0.334 0.444 1.087 (0.059) (0.089) (0.164) (0.087) (0.353) (0.041) (0.127) (0.001) (0.096) (0.089) PT 0.381 0.759 0.691 0.668 1.332 0.077 2.316 1.244 0.065 0.436 (0.040) (0.095) (0.016) (0.021) (0.264) (0.248) (0.650) (0.077) (0.062) (0.042) SE 0.691 2.629 0.436 3.607 1.406 0.133 0.203 0.018 0.150 1.975 (0.004) (0.504) (0.132) (0.207) (0.593) (0.138) (0.069) (0.013) (0.046) (0.215) SI 0.113 0.286 0.076 1.113 2.364 1.879 0.111 0.313 0.162 0.701 (0.061) (0.062) (0.067) (0.185) (0.440) (0.211) (0.001) (0.091) (0.022) (0.118) GFC 2.710 2.392 2.886 2.772 2.193 2.513 2.514 2.505 2.662 2.705 (0.639) (0.615) (0.611) (0.742) (0.639) (0.688) (0.742) (0.639) (0.660) (0.708) Const. 1.508 1.321 1.545 0.973 1.381 1.370 1.463 1.784 1.522 1.421 (0.212) (0.241) (0.218) (0.359) (0.238) (0.251) (0.251) (0.181) (0.220) (0.248) Obs. 96 96 96 96 96 96 96 96 96 96 R 2 .463 .471 .445 .473 .531 .468 .432 .498 .421 .443 Note. Dependent variable: GDP growth. Pooled OLS estimations. Standard errors in the parentheses. GFCD global nancial crisis dummy. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. Largest rms are identied based on their lagged sales; 1—Agriculture, forestry, and shing, 2—Manufacturing, 3—Water supply, sewerage, waste management, 4—Construction, 5—Wholesale, retail, and repair of motor vehicles, 6—Transportation and storage, 7—Accommodation and food services, 8—Information and communication, 9—Real estate, 10—Scientic technical and other business activities. p < .01. p < .05. p < .1. ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 219 Table A3. Demand-side shocks in largest 5 rms at the sectoral level, by country. 1 2 3 4 5 6 7 8 9 10 ES 0.766 0.675 0.111 0.143 1.358 0.134 1.923 0.602 0.146 0.061 (0.062) (0.093) (0.013) (0.058) (0.314) (0.027) (0.117) (0.062) (0.005) (0.026) FR 0.443 0.274 0.127 0.258 0.778 0.139 1.823 0.626 0.149 0.027 (0.087) (0.050) (0.022) (0.176) (0.241) (0.035) (0.196) (0.051) (0.004) (0.044) HU 1.024 0.949 0.191 0.353 1.787 0.478 2.502 0.300 0.081 0.034 (0.045) (0.165) (0.034) (0.049) (0.262) (0.001) (0.042) (0.049) (0.020) (0.035) IT 0.014 0.036 0.125 0.048 4.814 0.095 1.911 0.604 0.453 0.093 (0.060) (0.061) (0.010) (0.132) (0.453) (0.041) (0.119) (0.054) (0.001) (0.083) PL 0.642 0.330 0.025 0.241 2.033 0.018 1.802 0.641 0.266 0.022 (0.006) (0.101) (0.005) (0.124) (0.168) (0.052) (0.152) (0.076) (0.026) (0.012) PT 1.453 0.037 0.017 0.543 2.410 0.085 1.938 0.535 0.072 0.170 (0.048) (0.049) (0.037) (0.051) (0.130) (0.026) (0.151) (0.095) (0.060) (0.051) SE 0.368 0.368 0.035 3.278 0.631 0.154 1.591 0.601 0.173 0.993 (0.048) (0.104) (0.054) (0.176) (0.293) (0.047) (0.015) (0.048) (0.002) (0.262) SI 0.666 0.150 0.438 0.577 2.976 0.806 1.470 0.518 0.182 0.096 (0.069) (0.037) (0.022) (0.072) (0.190) (0.037) (0.042) (0.059) (0.007) (0.003) GFC 2.794 2.313 2.368 2.809 2.310 2.421 2.495 2.548 2.552 2.449 (0.656) (0.622) (0.564) (0.677) (0.617) (0.596) (0.733) (0.686) (0.661) (0.702) Const. 1.427 1.321 1.488 1.442 1.532 1.300 1.053 1.441 1.493 1.469 (0.207) (0.219) (0.195) (0.173) (0.235) (0.227) (0.270) (0.233) (0.220) (0.241) Obs. 96 96 96 96 96 96 96 96 96 96 R 2 .444 .473 .470 .496 .545 .507 .460 .432 .425 .425 Note. Dependent variable: GDP growth. Pooled OLS estimations. Standard errors in the parentheses. GFCD global nancial crisis dummy. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. Largest rms are identied based on their lagged sales; 1—Agriculture, forestry, and shing, 2—Manufacturing, 3—Water supply, sewerage, waste management, 4—Construction, 5—Wholesale, retail, and repair of motor vehicles, 6—Transportation and storage, 7—Accommodation and food services, 8—Information and communication, 9—Real estate, 10—Scientic technical and other business activities. p < .01. p < .05. p < .1. Table A4. Supply-side shocks in largest 3, 5, and 10 rms at the sectoral level for selected sectors, by country. Largest 3 rms Largest 5 rms Largest 10 rms 1 2 3 1 2 3 1 2 3 ES 0.769 1.044 2.235 0.416 0.752 0.352 0.675 1.021 0.205 FR 0.463 0.308 2.726 0.034 0.112 0.354 0.269 0.361 0.236 HU 1.137 1.061 1.107 0.813 0.762 1.296 0.842 1.075 1.141 IT 0.194 1.185 11.019 0.293 0.465 6.318 0.363 0.421 1.969 PL 0.318 2.227 1.796 0.640 1.053 0.743 1.078 1.857 0.218 PT 0.223 1.235 0.807 0.759 0.668 1.332 0.340 0.709 1.390 SE 0.987 5.164 4.066 2.629 3.607 1.406 1.391 3.445 2.392 SI 0.126 1.830 0.058 0.286 1.113 2.364 0.098 1.400 1.754 GFC 2.334 2.907 2.092 2.392 2.772 2.193 2.541 2.516 2.305 Const. 1.202 0.973 1.428 1.321 0.973 1.381 1.290 0.59 1.424 Obs. 96 96 96 96 96 96 96 96 96 R 2 .452 .494 .540 .471 .473 .531 .480 .491 .492 Note. Dependent variable: GDP growth. Pooled OLS estimations. Standard errors in the parentheses. GFCD global nancial crisis dummy. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. Largest rms are identied based on their lagged sales; 1—Manufacturing, 2—Construction, 3—Wholesale, retail, and repair of motor vehicles. p < .01. p < .05. p < .1. 220 ECONOMIC AND BUSINESS REVIEW 2025;27:199–220 Table A5. Demand-side shocks in largest 3, 5, and 10 rms at the sectoral level for selected sectors, by country. Largest 3 rms Largest 5 rms Largest 10 rms 1 2 3 1 2 3 1 2 3 ES 0.943 0.326 1.584 0.675 0.143 1.358 0.735 0.607 1.399 FR 0.579 0.074 1.909 0.274 0.258 0.778 0.287 0.494 1.344 HU 1.057 0.374 0.839 0.949 0.353 1.787 0.805 0.662 2.088 IT 0.533 1.001 6.539 0.036 0.048 4.814 0.707 0.770 0.850 PL 0.718 0.069 1.208 0.330 0.241 2.033 0.207 1.146 1.055 PT 0.316 0.828 0.121 0.037 0.543 2.410 0.097 0.275 2.300 SE 0.357 4.122 3.633 0.368 3.278 0.631 1.101 2.074 1.157 SI 0.526 0.283 0.171 0.150 0.577 2.976 0.326 0.825 2.662 GFC 2.388 2.700 2.192 2.313 2.809 2.310 2.496 2.809 2.517 Const. 1.355 1.763 1.385 1.321 1.442 1.532 1.343 1.056 1.714 Obs. 96 96 96 96 96 96 96 96 96 R 2 .460 .533 .540 .473 .496 .545 .520 .464 .505 Note. Dependent variable: GDP growth. Pooled OLS estimations. Standard errors in the parentheses. GFCD global nancial crisis dummy. Reported coefcients are taken directly from results; to obtain the coefcient for, e.g., France, one needs to sum up the coefcients for Spain and France. Largest rms are identied based on their lagged sales; 1—Manufacturing, 2—Construction, 3—Wholesale, retail, and repair of motor vehicles. p < .01. p < .05. p < .1.