Fear the walking dead: the epidemic of zombie firms in Portugal

i João Afonso Pereira Pita Fear the walking dead: the epidemic of zombie firms in Portugal Masters in Economics Supervisors: Professor Doutor Fernando Manuel Almeida Alexandre Professor Doutor Miguel Ângelo Reis Portela October 2019 ii DIREITOS DE AUTOR E CONDIÇÕES DE UTILIZAÇÃO DO TRABALHO POR TERCEIROS Este é um trabalho académico que pode ser utilizado por terceiros desde que respeitadas as regras e boas práticas internacionalmente aceites, no que concerne aos direitos de autor e direitos conexos. Assim, o presente trabalho pode ser utilizado nos termos previstos na licença abaixo indicada. Caso o utilizador necessite de permissão para poder fazer um uso do trabalho em condições não previstas no licenciamento indicado, deverá contactar o autor, através do RepositóriUM da Universidade do Minho. Licença concedida aos utilizadores deste trabalho Atribuição-NãoComercial-SemDerivações CC BY-NC-ND https://creativecommons.org/licenses/by-nc-nd/4.0/ iii Acknowledgments The accomplishment of this master's thesis had important support and incentives without which it would not be possible and for which I will be forever grateful. Firstly, I want to thank the teachers Fernando Alexandre and Miguel Portela for their guidance and total availability, as well as for the incentives, opinions and criticisms in solving doubts and problems that came from the realization of this work. I also leave here a word of thanks to all the teachers of the School of Economics and Management for the knowledge transmitted and for all the words of encouragement. To my colleagues and friends who have always supported me in accomplishing this work with words and wisdom needed to overcome the problems that arose. Finally, a special thanks to my parents, sister and other members of my family, especially my uncles who have always been present and gave unconditional support, for the financial and emotional support, encouragement and patience in helping with all the difficulties that arose. To them I dedicate this work. Thank you so much, João Afonso Pereira Pita iv STATEMENT OF INTEGRITY I hereby declare having conducted this academic work with integrity. I confirm that I have not used plagiarism or any form of undue use of information or falsification of results along the process leading to its elaboration. I further declare that I have fully acknowledged the Code of Ethical Conduct of the University of Minho. v Medo dos zombies : a epidemia de empresas zombies em Portugal Resumo Este estudo explora as empresas denominadas “zombies” na economia portuguesa. Estas empresas são definidas sendo empresas antigas e que tem problemas persistentes em pagar os juros de empréstimos bancários e desviam a produtividade do trabalho de empresas saudáveis. O tema das empresas zombies já foi analisado desde o caso do Japão na década de 90 até aos mais recentes sobre países da OCDE (nomeadamente Portugal). Vários autores estudaram este tema sendo o mais conhecido Caballero et al. (2008). Primeiro foi usado a ideia de Caballero et al. (2008) sendo aplicada a fórmula simplificada de McGowan et al. (2018). O presente estudo mostra que, entre 2010 e 2018, na economia portuguesa temos empresas zombies seguindo uma análise menos restrita e outra mais restrita. Na primeira, em média, durante os anos de analise temos cerca de 12.8% enquanto que na segunda este número diminui para 7.4%. Na análise menos restrita foi somente tomado em consideração o facto de a empresa ter, pelo menos, 10 anos de idade para remover da análise as novas empresas ( start-ups , por exemplo) e a taxa de cobertura dos juros ser inferior a 1 por, pelo menos, 3 anos consecutivos. Na mais restrita, foi implementado a anteriores rácios financeiros tais como: o retorno sobre o Ativo e o rácio total da divida. Foi também confirmado recorrendo aos modelos Mínimos Quadrados Ordinários, Efeitos fixos e Efeitos aleatórios e, ainda, um modelo Probit, que as empresas zombies tendem a estar relacionados com setores específicos da economia, com certas zonas do país e o seu tamanho (número de empregados). Estas empresas zombies tem implicações significativas nas empresas saudáveis a operar no mesmo setor, reduzindo o emprego e a margem de lucro. Palavras chave: capital, emprego, empresas zombie , endividamento, financiamento, rácios financeiros Classificação JEL: E22; E24; G32; G33; O16. vi Fear the walking dead: the epidemic of zombie firms in Portugal Abstract This study explores firms called “zombies” in the Portuguese economy. These firms are defined as old firms and have persistent problems in paying interest on bank loans and sidetrack labor productivity from healthy companies. The theme of zombie firms has been analyzed from the case of Japan in the 1990s to the most recent on OECD countries (i.e. Portugal). Several authors have studied this theme, the most well-known being Caballero et al. (2008). First the idea of Caballero et. al (2008) being applied, then the simplified formula of McGowan et al. (2018) was used to compute the regression. The present study shows that, between 2010 and 2018, in the Portuguese economy we had zombie firms following a less restricted and a more restricted analysis. In the first, on average, during the years of analysis we have about 12.8% while in the second this number decreases to 7.4%. In the less restrictive analysis, firms need to be, at least, 10 years old to remove bias using newest firms (for example start-ups) and the interest coverage rate is less than 1 per cent. for at least 3 consecutive years. In the more restrictive, it has been implemented to previous financial ratios such as: Return on Assets and the total debt ratio. It was also confirmed by using Ordinary Least Squares, Fixed Effects and Random Effects models, as well as a Probit model, that zombie companies tend to be related to specific sectors of the economy, certain areas of the country and their size (number of employees). These zombie companies have significant implications for healthy companies operating in the same industry, reducing employment and profit margin. Keywords: capital, employment, financial ratios, financing, indebtness, zombie firms JEL Classification: E22; E24; G32; G33; O16. vii Index Acknowledgments ..................................................................................................................... iii Resumo ........................................................................................................................................ v Abstract ....................................................................................................................................... vi List of Tables .............................................................................................................................. ix List of Figures .............................................................................................................................. x List of abbreviations and acronyms ............................................................................................ xi 1.Introduction .............................................................................................................................. 1 2. Macroeconomic Situation of Portugal ..................................................................................... 3 3. Literature Review .................................................................................................................... 7 4. Theoretical framework and hypotheses ................................................................................. 10 5. Data ........................................................................................................................................ 13 6 – Methods ............................................................................................................................... 15 6.1. – Ordinary Least Squares ............................................................................................... 16 6.2. – Fixed Effects Model .................................................................................................... 17 6.3. – Random Effects Model ................................................................................................ 17 6.4. – Hausman Test .............................................................................................................. 18 6.5. - Probit Model ................................................................................................................. 18 7. Empirical findings ................................................................................................................. 19 7.1. Profile of Zombie firms .................................................................................................. 19 7.2. Regressions used ............................................................................................................ 20 7.3. Less restrictive analysis .................................................................................................. 21 7.4. More restrictive analysis ................................................................................................. 25 7.5. Discussion of results ....................................................................................................... 29 8. Conclusion ............................................................................................................................. 31 viii References ................................................................................................................................. 33 Appendix ................................................................................................................................... 36 ix List of Tables Table 1 - Number of zombies in Portugal (less restrictive) ................................................................... 21 Table 2 - Zombie firms and non-zombies’ performance ....................................................................... 23 Table 3 - Number of zombies in Portugal (restrictive) .......................................................................... 26 Table 4 - Zombies firms and non-zombies’ performance ..................................................................... 26 Table 5 - Firms in the Portuguese economy from 2010 until 2018 ...................................................... 36 Table 6 – NUTS 2 distribution in the Portuguese economy ................................................................. 36 Table 7 – NUTS 3 distribution in the Portuguese economy ................................................................. 36 Table 8 - NACE code and description ................................................................................................. 37 Table 9 - NACE distribution in the Portuguese economy ...................................................................... 38 Table 10 - Number of employees in each firm .................................................................................... 38 Table 11 - Probit model ..................................................................................................................... 39 Table 12 - Margins effects .................................................................................................................. 40 5 Figure 3 - Births, Deaths and Survivors firms in Portugal Notes: The blue line is firms’ births; the orange line is firms’ deaths; the grey line is firms that survived 1 year after they enter the market; and the yellow line is firms that survived 2 years after they enter the market. Source: pordata.pt 2018, Bank of Portugal, PORDATA In the third figure we can see that, in terms of firms’ births, Portugal has had, after the 2008 crisis, the number of births decrease around 60.000 until 2012, in which these births increased again. This can be a sign of the lower interest rates and the rise of loans as shown before. The firms’ deaths reflect what is verified in the births of companies, in the sense that, if the number of births increases, the number of deaths increases. However, between 2008 and 2012 there was a greater number of deaths than births. Currently the numbers are very close. Also, in this figure, we can observe the firms that survived at least 1 and 2 years after entering the market. Unsurprisingly, firms that enter the market tend not to survive for long, and it is curious that about half of the firms created in the previous two years are not in the market in the following years. This is an important point related to zombie firms because since zombie firms may, sometimes, exit the market, the number of firms’ deaths should be sometimes higher than it is shown. Even though is not shown, conclusions about insolvent and in recovering firms can be given. According to a study from a credit and risk management consultant (“InfoTrust”), Portugal had in 2018 almost twice the closing of firms’ comparative to the previous year, having the insolvency and recovering firms a decrease. This may be a sign that firms are surviving thanks to subsidized bank credit. 0 30 000 60 000 90 000 120 000 150 000 180 000 210 000 240 000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Births Deads Survivors for 1 years Survivors for 2 years 6 Figure 4 - Number of employees: Total and by firm size Notes: The blue line is total firms in Portugal; the orange line is SME; and the grey line is big firms. Source: pordata.pt 2018, Bank of Portugal, PORDATA Finally, this last figure shows us that, in Portugal, the number of employees in total firms in Portugal decreased since 2008 until 2013, raising again to number close to the 2008 period. Between SME and big firms, we can see that SME had the biggest impact on the decrease of number of employees and big firms maintain the same along the period of analysis. Also, it is interesting to see that SME firms dominate the market in relation to big firms, which is a characteristic known of the Portuguese economy. As said before and according to the literature revision, SMEs had a big impact on the number of zombies harming the labor market, so this is also a reason for the Portuguese economy having so much zombie firms in relation to other countries in the European Union. 0 500 000 1 000 000 1 500 000 2 000 000 2 500 000 3 000 000 3 500 000 4 000 000 4 500 000 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 Dimension Total Dimension PME Dimension Large 7 3. Literature Review The impact of zombie firms is a big deal to the economy of a country, so the study and the comprehension of the way that healthy companies evolved into zombies is an important topic to study in Portugal, since we have a large percentage of zombies, and it is contained in the OECD research program. The phenomenon of zombie firms began in the early 90’s with the collapse of the Japanese asset price bubble, while there was a period of stagnation of Japanese firms known as the “lost decade”. These firms distorted the market and caused significant interruptions in the economy’s recovery. Curiously, few were the firms that declared bankruptcy, and many of them recovered at the beginning of the 21st century. According to Fukuda and Nakamura (2001), reducing the employee strength and selling fixed assets were beneficial for reviving zombie firms. Also, external supports including debt relief and capital reduction were the other important factors for the recovery of these firms. The study from Hoshi and Kashyap (2004) was one of the first to alert to this type of firms, saying that the size of bad loans in Japan was around 7% of GDP. That said, Caballero et al. (2008), one of the most important paper regarding zombie firms, tried to identify these firms based on whether they are receiving subsidized credit, not by looking at their productivity or profitability. Also, they struggle to obtain the data necessary to study this theme mainly because banks and the borrowers aren’t available to provide the raw data that allows the them to determine if they are or not a zombie firm. That said, the study from Caballero et al. (2008) concluded that if banks let zombie firms alive, they hinder efficient human resource allocation and channel the investment from viable firms to unviable ones. When these unviable firms go through a low performance phase, banks tend to provide more credit to them mostly when they work together forming a type of alliance smoothing (in Japan we can see the keiretsu alliance) the access to credit. Lam, Schipke, Tan and Tan (2017) is a study accomplished by observing firms in China, identifying the zombies, and explains the central role that these firms have with state-owned enterprises aligned to low productivity and higher debt. The government had tried various reforms to reduce the 8 deleveraging but with no success. The empirical results of this papers suggest that, to accelerate the restructuring process, it requires a more holistic and coordinated strategy, recognizing losses, reducing the implicit support and liquidating zombie firms. Recent studies have studied zombie firms in the context of the European Union, such as McGowan et al. (2018), defining these firms as those that have, at least, ten years old, excluding startups, and those that ca not cover the interest payments for, at least, three consecutive years. Also, this paper restricts the sample to the pre-crisis period (2003-2007) finding that an increase in zombie share at the industry level is associated with lower investment and employment growth for the average nonzombie. So, the resources trapped in zombie firms was a policy issue already before the crisis that is, today, not solved. In the case of Portugal, those studies told us that, on average, zombies are older, larger (both in terms of number of employees and turnover) and are much less productive than their non-zombie counterparts (Gouveia and Coelho, 2018). McGowan et al. (2018) also shows the share of capital sunk in zombie firms in 2013, and in terms of international comparison, Portugal has around 14%, much more than France (around 6%) or the United Kingdom (around 7,5%). These numbers are decreasing, and it is good for the economy as shown in Osterhold and Gouveia (2018), in which a reduction in the capital sunk in zombie firms fetches positive externalities to non-zombies. Previous papers studied the factor of indebtedness of some Portuguese firms from 1990 to 1995 and found out that the growth of a firm has a positive relation with the indebtedness, which means that the firms with higher asset growth rates are the ones that are more indebted (Jorge and Armada 2001). Storz, Koetter and Setzer (2017) studied, between the years of 2010 and 2014, (post crisis and when Portugal had the (EFAP), the impact of bank stress on the deleveraging process of non-financial small and medium-sized enterprises denominated SMEs, focusing their work in the euro area with two groups: the first one includes Portugal, Greece, Ireland, Spain and Slovenia called by the author “the periphery euro area”, and the second denominated “euro area core” with France and Germany. They concluded that the interaction between the week SMEs and weak banks is a possible cause of misrepresentation in the deleveraging efforts of euro area periphery economies. For the same year, 9 France and Germany had no significant impact of bank stress on zombie firm leverage suggesting that non-financing corporations deleveraging is indeed associated with weak banks, possibly because these banks have an incentive to keep these firms alive (evergreen loans) to avoid having to declare nonperforming loans. Another recent research conducted by OECD, the main financier of the zombie firms are weaker banks, suggesting that the zombie firm problem may partly stem from bank forbearance (McGowan et al. 2017). Also, Oliveira (2008) said that “the principal form of external financing is through the use of the banking system (…)”, but some authors concluded that the owners themselves also invest in their business. This type of bank forbearance happens when banks seek to avoid recognition of the loss of credit granted to suck companies (Alexandre et al., 2017). The economic and financial analysis of a firm is a toll that allows, according to Cardoso Moreira (2001), to know the situation of a company through its past and anticipate future situations, so policy makers should be aware of this phenomenon because it can condition the country’s economic development, and some economist such as Langfield and Pagano (2015), concluded that an increase in the size of the banking system is associated to more risk and less economic growth. 10 4. Theoretical framework and hypotheses As mentioned before, the literature characterizes a zombie firm as those with more than a decade of existence, that do not generate enough revenues in their regular business, persistently dependent on bank credit and those who usually pay high salaries considering the productivity of the sector in which they work. Firstly, to define a zombie firm, it was firstly tried to use the initial idea in the Caballero et al. (2008) where they define zombie firms as those potentially receiving subsidized bank credit and not by looking at their productivity or profitability so that we can evaluate the effect of zombies on the economy. To do this they observed interest payments made by the firm comparing them to an estimated benchmark R* based on the firm debt structure and market interest rate. Those firms with negative interest rates are receiving subsidized credit and are zombies. But the study from Caballero et al. (2008) is very data demanding with a dataset that covers the period from 1981 to 2002, reaching, some years, to 2500 firms. Also, we cannot precisely distinguish between different forms of debt held by companies in our database, such as bank loans and debt securities issued. Therefore, we also cannot observe actual interest payments on different forms of debt. Observable overall interest expenses may not necessarily show the actual payments during a certain year. Knowing the previously present difficulty, McGowan et al. (2018) define zombie firms based on a simplified formula from Caballero et al. (2008) adding, to the definition of the interest coverage ratio, a criterion to the age of the firm. In Amadeus, we have the date of incorporation for each firm so doing basic computations we got only firms that have only ten years or more in the market. That said, a zombie must have an interest coverage ratio less than one for three consecutive years. Initially, the level of EBIT is observed and divided by the interest paid by firms. This will give us the interest coverage ratio and if it is lower than one for three consecutive years we can say, in practice, this means that a firm must take on additional debt to cover its interest payments. It is typical to identify zombies based on a single weak performance metric, but this easily includes firms who are growing and whose weak profitability is often temporary. For example, investment involves a necessary trade-off between short term costs and future productivity and profitability growth, which may cause performance metrics to supposedly decline before revenue 11 catches up. Start-up companies are particularly vulnerable to this. So, they used the age of the firms to differentiate a real zombie from an innovative start-up who may still have comparatively high operating costs and low revenue but since this database doesn’t have how long firms had been on the market, we can’t differentiate zombies’ firms from start-ups. After that, the empirical framework uses pooled cross-section micro data to explore the distortionary effects of zombie firms on the performance of non-zombie firms. That said, on the regressions non-zombies should have also ten years but the interest coverage ratio higher than one for three consecutive years and the following financial ratios. Following Schivardi et al. (2017) a criterion of total debt to total assets ratio, also known as, total debt ratio will be implemented and if a firm has a value lower than 40% it would be considered a non-zombie (being tested with other ratios as well to see the impact in the number of zombies). Thirdly, it will be used the Return of Assets (ROA) to analyze the profitability (or “quality”) of the firm to replace Tobin’s q, as mentioned on Miyajima and Yafeh (2007). Tobin’s q is used to express the relationship between market valuation and intrinsic value, in other words, it is a means of estimating whether a given firm or market is overvalued or undervalued. The use of ROA is similar to Tobin’s q because it tells investors an idea of how effective the firm is in converting the money it invests into net income, so the higher the ROA, the better, because more money is made with less investment. Summing up, a firm is considered a non-zombie, whenever they have the interest coverage ratio higher than one for, at least, three consecutive years; have more than ten years in business, the total debt ratio is lower than 40% and the ROA is higher than 0. Our non-zombie dummy is thus equal to 0, whenever the firm fulfills criteria all these criteria for the current period, and 1 otherwise. Following the specification in Caballero et al. (2008) and McGowan et al. (2018), we test whether zombies entail negative spillover effects on viable firms. We depend on panel data from 2010 to 2018 to estimate a reduced-form equation to see the impact of zombie congestion as: 𝑌𝑖𝑠𝑡 =𝛽0+𝛽1𝑛𝑜𝑛𝑧𝑜𝑚𝑏𝑖𝑒𝑖𝑠𝑡 +𝛽2𝑛𝑜𝑛𝑧𝑜𝑚𝑏𝑖𝑒𝑖𝑠𝑡 ∗𝑍𝑠𝑡 +𝛽3𝑟𝑒𝑣𝑒𝑛𝑢𝑒𝑖𝑠𝑡 +𝛽4𝑠𝑖𝑧𝑒𝑖𝑠𝑡 + 𝜀𝑖𝑠𝑡 (1) 12 Where Y denotes the profit margin and employment growth of firm i , in industry s , at year t. The dummy nonzombie takes the value of 1 for non-zombie firms and 0 otherwise. This variable is defined as shown before whenever the firm fulfills all four criteria. Z is the share of industry capital sunk in zombie firms. With the variable of 𝛽3 it is possible to see the impact of sales in the regression. The variable β4 is a firm control variable, so it is a dummy variable that defines the size of the firm (1 to 10, 11 to 19, 20 to 49, 50 to 99, 100 to 249 and 250+) in terms of employment. It is expected that 𝛽2will be negative, implying that more resources are sunken in zombie firms, for the profit margin and employment growth since zombie firms reduce the ability and capital for nonzombie firms to grow. As well, 𝛽1 may be negative if zombie firms receive large amounts of subsidized credit but, it’s shown in the literature that it could be positive due to zombie firms not being able to spend as much as healthy firms. For the dummy variable of size, a raise in the number of employees the higher is expected to be the number of zombies. Operating Revenue may also influence the capital and employment growth and should tell us if the firm is doing good in term of sales or if they just receive the subsidized credit. 13 5. Data In the literature (see Caballero et al 2008; McGowan et al. 2018; Alexandre et al. 2017), it has been observed that the survival of zombie firms may distort competition and weaken market efficiency. Healthy markets are characterized by a process of creative destruction, where insolvent or unprofitable firms reduce their share of labor and successful firms invest and create new jobs. When zombies participate in the market, they raise demand for labor and intensify competition for market share. This has the consequence of lowering product prices and increasing wages, effectively congesting growth conditions for more promising firms. To define a zombie firms, several criteria are used, from least restricted to the most restricted. As said before, this study will primarily use the Caballero et al. (2008) formula and the simplest form used by McGowan et al. (2018) with the addition of others financial variables, using available information to determine which firms are receiving subsidized credit using firm-level data from Bureau van Dijk’s (BvD) Amadeus database. This is a database of comparable financial information for public and private companies among countries in Europe. The purpose of using this database is that it encompasses data from Portuguese companies with significant relevance to the business and from financial ratios that are routinely used in the financial analysis of companies. Also, this database is in panel data and has the period which will be analyzed (2010-2018). It would be better if we could use the Sistema de Contas Integradas das Empresas (SCIE) from the Instituto Nacional de Estatística (INE) but due to some restrictions it was not possible to use. The only problem in this period is that it coincides in a substantial part with the Economic and Financial Assistance Program (EFAP) that occurred in Portugal between 2011 and 2014, after the worldwide crisis of 2008. Also, not analyzing the period before 2008 is a big shrinkage because we don’t observe what happened in a good economic period. Thus, in addition to a short period of time (8 years of analysis), it is an atypical period, in which economic activity is extraordinarily retracted. Nevertheless, as said before, it would be interesting to analyze the impact of the EFAP in the context of zombie firms, but a longer period of dataset would be necessary. Another important point of this database is that it analyzes births and deaths of companies as well as demographic indicators. For the current year, the population is constituted by all the companies 14 that carry out an activity of production of goods and/or services, in Portugal. In addition, the database has information to be able to perform a dimensional analysis of each firm, along with a sectorial analysis in which firms will be divided by seven regions according to NUTS 2 and twenty-five regions according to NUTS 3. To study the dimension of the firm, this study will have five size classes that are: firms with less than ten employers, firms with less than fifty employers but over than ten, firms with less than one hundred but over fifty employers, firms with lower than two hundred and fifty employers but over than one hundred and big enterprises with over two hundred and fifty employers. This size of the firms was made accordingly to Alexandre et al. (2017) and according to a variable available in Amadeus (It should be noted that this variable is according to the European Union classification based on total assets and operating revenue). For the sectorial analysis it will be divided by letters according to the sector in question, i.e. each letter will represent a sector of activity with a total of seventeen sectors. Using the data available in Amadeus, a set of 378,887 firms was obtained for Portugal but after some restrictions and adjustments the number was reduced to 245,015 mainly due to missing values. To better explain this decrease in the number of observations, it’s computed in Stata that, if important financial variables are missing, such as, total assets, capital, profit or EBIT, the firms are removed. This command, for example, to the year of 2010 decreases the number of firms from the 378,887 to 194,237. Another point is, if the firm has more than 7 variables missing (the maximum is 11) it gets removed (in this point, for the year of 2010 we have 191,151 firms). Other restriction was if the firm has years missing, this means if the firm have reports for 2010 and 2012 but misses the year of 2011, it will get 1 (equals to 1 year missing) until the maximum that is 9. At this stage we can notice more firms dropped in the earlier data such as 2018, 2017 and 2016. (for example, in 2017, previously we had 314,198 firms and now only 259,654 firms). This can be explained because, in Portugal, firms tend to report the year n in the n+1 year or, in some cases, they send wrong data and the regulators ask to correct this. Then firms that do not have recent year data (2018) are deleted. Finally, we have our database ready to the analysis with an increase in the number of firms since 2010, from 153,795 to 219,749 in 2018. This can cause a problem in the analysis since it was not possible to verify the real reason for this decrease (see Table 5 on appendix). 21 𝐿𝑛(𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡)𝑖𝑠𝑡 =β0+β1nonzombieist +β2nonzombieist ∗Zst +β3revenueist + β4size+𝜂𝑖+εist (8) It is expected that 𝛽2 will be negative since zombie congestion reduce the ability or incentives for non-zombie firms to grow. The coefficient of the non-zombie 𝛽1 will be hard to understand because, as said before, it can be either negative for some cases or positive, depending on the subsidized credit that firms receive. 7.3. Less restrictive analysis As said before, a less restrictive criteria will be used to see how many zombies we have in the Portuguese economy. In this part, it was only used two criteria: the firms’ age (higher than 10 years) and the interest coverage ratio (less than one for 3 consecutive years). Using the Stata to compute these criteria, we got, for Portugal, an average of zombie firms, between 2010 and 2018, of 12.8% (see Table 1). Also, we can observe a decrease since 2010, from 13% to 10.8% in 2018. Also, from 2011 until 2013, the number of zombies raises in 2.7 percentual points. This can be explained mainly because of two reasons. Either these firms exited the market or had a restructuring. According to the literature, when these restrictions were used, the value of zombie firms was around 10%. As we can see, the number of zombie firms in Portugal is slightly higher than the results given by the literature. At of this is crucial to deeply analyze the zombies with some more criteria and to see whether and where zombies have more impact on the Portuguese economy. Table 1 - Number of zombies in Portugal (less restrictive) Year Zombies 2010 13.0% 2011 12.9% 2012 15.0% 2013 15.6% 2014 14.1% 2015 12.6% 2016 11.4% 22 2017 11.1% 2018 10.8% Source: Author’s own calculations. In the regressions (see Table 2) we can notice a difference in the number of observations, from 899,937 for the profit margin variable and 929,326 for the employment growth. This was a constant problem with this database because, for example, the profit margin variable has a lot of missing values in Amadeus. That said, observing the R squared, which is 21.7% for the profit margin, 21.7% of the variation can be explained by the independent variables in the model and for the Log of the employment we have a higher R squared in which this regression can explain 65.6% of the employment growth. For the profit margin regression, the p-value for almost all variables is, approximately, 0 so since it is a low p-value (below 0.05 for 95% confidence level) this indicates that we can reject the null hypothesis. In other words, a change in the independent variable is associated with changes in the response at the population level and all variables are statistically significant. Only the size when the number of employees is between 50 and 100, we have 0.165 and for the significance level of 0.05 it is not statistically significant. For the employment growth, we only have one p-value different from 0 but it is below 0.05, so we reject the null hypothesis. The estimated coefficients are positive for 𝛽1 and for the 𝛽2 we have a positive sign in the profit margin regression but a (low) negative results in the employment growth. This means that we have an increase in the profit margin and employment growth when the firms are not zombie. Also, the interaction between the nonzombie firms and the share of industry capital sunk in zombies’ firms, reduces the profit margin but raise the employment growth slightly. We can also see the impact of the dimension of the firm. According to the computations, the firms with high number of employees, ceteris paribus , translates into higher profit margin and higher employment growth. To notice that firms with fewer than 20 employees have negative signs and therefore have a negative impact on profit. The remaining firms have positive values, notably large companies (with more than 250 employees) are those with a higher coefficient value. After that it is important to pay attention to the homoscedasticity. It is a term for designating constant variance of the errors terms for distinct observations. If the homoscedasticity assumption is not valid, we can list some effects on the model: 1The standard errors of the estimators obtained by OLS are incorrect and therefore the 23 statistical inference is not valid; 2We can no longer say that OLS is the best estimators for β, although they may still be nonbias. Doing the White test to detect heteroscedasticity (absence of homoscedasticity), we can see that the null hypothesis of constant variance can be rejected at 5% level of significance. The implication of the above finding is that there is heteroscedasticity in the residuals. This can be due to measurement error, model misspecifications or subpopulation differences. Consequences of the heteroscedasticity are that the OLS estimates are no longer BLUE (Best Linear Unbiased Estimator). Standard errors will be unreliable, which will further cause bias in test results and confidence intervals. To solve this issue the regression was corrected by using the robust command. Table 2 - Zombie firms and non-zombies’ performance OLS FE RE OLS FE RE VARIABLES Profit Margin Profit Margin Profit Margin Ln(emp) Ln(emp) Ln(emp) Non-zombie 30.30*** 20.01*** 23.33*** 0.199*** -0.018*** 0.002*** (0.065) (0.081) (0.072) (0.002) (0.001) (0.001) Non-zombie * Zombie Share 0.0246*** -0.105*** -0.050*** -0.002*** 0.0001** -0.0001** (0.002) (0.003) (0.003) (7.40e-05) (5.55e-05) (5.57e-05) Operating Revenue 0.01*** 0.151*** 0.02*** 0.004*** 0.009*** 0.007*** (0.002) (0.02) (0.006) (0.0008) (0.0004) (0.0002) Employees 10<X<=20 -0.623*** 0.274** -0.0269 1.607*** 0.594*** 0.787*** (0.065) (0.111) (0.086) (0.002) (0.002) (0.002) Employees 20<X<=50 -0.267*** 0.701*** 0.264** 2.340*** 1.127*** 1.483*** (0.075) (0.172) (0.113) (0.002) (0.003) (0.003) Employees 50<X<=100 0.108 1.003*** 0.529*** 3.177*** 1.671*** 2.198*** (0.134) (0.286) (0.191) (0.005) (0.005) (0.005) 24 Employees 100<X<=250 0.987*** 1.822*** 1.364*** 3.948*** 2.241*** 2.942*** (0.181) (0.433) (0.273) (0.007) (0.008) (0.007) Employees 250<X 1.372*** 1.559** 1.351*** 5.276*** 2.913*** 3.846*** (0.267) (0.716) (0.437) (0.009) (0.013) (0.012) Observations 899,937 899,937 899,937 929,326 929,326 929,326 R-squared 0.217 0.081 0.692 0.194 Number of Firms 138,923 138,923 140,321 140,321 Notes: Non-zombie denotes firms in Portugal that are not zombies, with more than 10 years and an interest coverage ratio higher than 1 for 3 consecutive years. Zombie share denotes the capital sunk in zombie firms. Ln(emp) means the Log of the employment. Operating Revenue is represented in millions. This panel is from data collected from 2010 until 2018. Standard errors in parentheses. *** denotes statistical significance at the 1% level, ** significance at the 5% level, * significance at the 10% level. Source: Author’s own calculations. As stated before, two models will be used to deeply analyze this study. Firstly, it was applied in Stata the Fixed effects model to the profit margin variable to check the impact of the independent variables that can vary over time. Secondly, a Random effects model was conducted using random values. The FE told us that, by observing the p-value which is the probability that the null hypothesis for the full model is true, and since is lower than 0.05 (in this case is approximately 0) we have at least some of the parameters being nonzero. The t-values test the hypothesis that each coefficient is different from 0 and in this case, we reject the t-value for every variable because it is higher than 1.96 (for a 95% of confidence), so the variables have a significant influence on profit margin. Also, the "rho" value (0.53685) tell us that 53.7% of the variance is due to differences across panels. The rho is also known as the interclass correlation and tells how strongly the observations within each resemble each other. Another relevant value is the coefficient of the regressors regarding the size of the firm. For example, profit margin raises 2 percentual points when the firms with more than 250 employees increases by one unit. In both, profit margin and employment growth regressions we have a f-statistic in where the null hypothesis is that all the coefficients in the model except for the intercept are zero, so: 25 𝐻0:𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 𝑖𝑛 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡𝑠 𝑛𝑜𝑡 𝑠𝑦𝑠𝑡𝑒𝑚𝑎𝑡𝑖𝑐 vs. 𝐻1:𝐻𝑜 𝑖𝑠 𝑛𝑜𝑡 𝑡𝑟𝑢𝑒 (9) Since the value is 0, we reject the null hypothesis (reject 𝐻0), so this means that we have, at least, one 𝛽 different from 0 and all independent variables (together) are statistically significant Meanwhile the RE says that the p-value is, again, approximately 0. The two-tail p-values are all 0 and this tests the hypothesis that each coefficient is different from 0, so it can be said that all the variables have a significant influence on profit margin. The "rho" in this case is 0.429, 10.8 percentage points lower than the FE. Using the Hausman test, we can decide therefore which model is most suitable to this analysis. That said, we have: 𝐻0: Random effects model is preferred 𝐻1: Fixed effects model is preferred That said and observing our results we have a p-value of, approximately, 0, so we reject the 𝐻0 hypothesis. With these results, our regressions indicate that the present of these zombies’ firms, in the Portuguese economy, may have amplified the negative consequences of the 2008 crisis but may also slow down the economy recovery by distorting the money that could went to healthy firms. 7.4. More restrictive analysis After studying the number of zombies in Portugal and doing a restrictive regression, it is important to see the impact of the financial ratios when we consider a firm to be zombie. Using the criteria described earlier in the section 4, a firm needs to fill all the criteria to become a zombie so, in this point we will have a smaller number of zombies, in average, 6.6% of total firms in Portugal. Also, it is important to notice (see Table 3) that, as the less restrictive analysis, the number decreased from 2010, with 7.4% of the Portuguese being zombies to 5.5% in 2018, but it raises from 2011 to 2013. This cannot be explained properly but again, the fact that we have a lot of missing data, mainly in the oldest years, can influence the data. 26 Table 3 - Number of zombies in Portugal (restrictive) Year Zombies 2010 7.4% 2011 6.8% 2012 7.8% 2013 8.2% 2014 7.3% 2015 6.5% 2016 5.8% 2017 5.6% 2018 5.5% Source: Author’s own calculations. In the regressions (see Table 4) we can also notice almost the same difference in the number of observations, from 899,937 for the profit margin variable and 929,326 for the employment growth. Hereupon, observing the R squared, 11.5% of the profit margin can be explained by the independent variable in this model and for the employment growth we have a higher R squared just like before, which is better, because this regression can explain 65.7% on the employment growth. The R2 is almost the same using the more restrictive analysis. Table 4 - Zombies firms and non-zombies’ performance OLS FE RE OLS FE RE VARIABLES Ln(emp) Ln(emp) Ln(emp) Profit Margin Profit Margin Profit Margin Non-zombie 0.217*** -0.011*** 0.005*** 27.15*** 13.28*** 16.89*** 27 (0.003) (0.002) (0.002) (0.091) (0.098) (0.091) Non-zombie * Zombie Share -0.001*** -7.78e-05 -0.0002*** 0.193*** 0.112*** 0.149*** (7.16e-05) (5.17e-05) (5.21e-05) (0.002) (0.003) (0.003) Operating Revenue 0.004*** 0.009*** 0.007*** 0.015*** 0.191*** 0.026*** (0.0008) (0.0004) (0.0002) (0.003) (0.020) (0.006) Employees 10<X<=20 1.612*** 0.594*** 0.787*** 0.201*** 0.328*** 0.364*** (0.002) (0.002) (0.002) (0.069) (0.114) (0.090) Employees 20<X<=50 2.345*** 1.127*** 1.482*** 0.624*** 0.940*** 0.901*** (0.003) (0.003) (0.003) (0.079) (0.176) (0.120) Employees 50<X<=100 3.181*** 1.670*** 2.198*** 0.759*** 1.320*** 1.096*** (0.005) (0.005) (0.005) (0.142) (0.294) (0.201) Employees 100<X<=250 3.950*** 2.241*** 2.941*** 1.481*** 2.471*** 1.998*** (0.007) (0.008) (0.007) (0.192) (0.444) (0.289) Employees 250<X 5.277*** 2.912*** 3.845*** 1.570*** 2.609*** 1.989*** (0.010) (0.013) (0.012) (0.284) (0.735) (0.464) Observations 929,326 929,326 929,326 899,937 899,937 899,937 R-squared 0.691 0.193 0.115 0.031 Number of Firms 140,321 140,321 138,923 138,923 Notes: Non-zombie denotes firms in Portugal that are not zombies, with more than 10 years, an interest coverage ratio higher than 1 for 3 consecutive years, ROA below 0 and total debt ratio below 40%. Zombie share denotes the capital sunk in zombie firms. Ln(emp) means the Log of the employment. Operating Revenue is represented in millions. This panel is from data collected from 2010 until 2018. Standard errors in parentheses. *** denotes statistical significance at the 1% level, ** significance at the 5% level, * significance at the 10% level. Source: Author’s own calculations. 28 In this model we have got the same p-values (approximately zero for all variables) as before so our sample data provide enough evidence to reject the null hypothesis for the entire set. The data favor the hypothesis that there is a non-zero correlation. The beta (𝛽1) in this model has the same signal as the restrictive model, being positive for both dependent variables. So even with the additional restrictions, we have a positive impact of the non-zombie firms on the profit margin and the employment growth. The share of industrial capital sunk in zombie firms with the interaction with the non-zombies is almost the same as the previous one, being negative for the employment and positive for the profit margin. This means that, having more capital sunk in zombie firms can contribute slightly to a higher profit margin to non-zombie firms. This was not expected but it can be explained because if zombie firms can get subsidized credit. In both, profit margin and employment growth regressions we have a f-statistic in where the null hypothesis is that all the coefficients in the model except for the intercept are zero, so: 𝐻0:𝛽1=0,𝛽2=0,𝛽3=0,𝛽4=0 vs. 𝐻1:𝐻𝑜 𝑖𝑠 𝑛𝑜𝑡 𝑡𝑟𝑢𝑒 (10) Since the value is 0, we reject the null hypothesis (reject 𝐻0), so this means that we have, at least, one 𝛽 different from 0 and all independent variables are statistically significant. The number of employees, in this regression, tell us the same conclusion as the previous one. So, a raise in the number of employees that each firm has, contributes positively to a raise in either the profit margin or the employment growth To these regressions it is, as well, applied the Fixed effects model and the Random effects model. To the profit margin, the FE model provides almost the same results as the less restrictive regressions provided with a p-value being approximately 0. To notice that the two-tail p-values vary in the size of the firm, mainly when firms have more than 250 or between 10 and 20 employees, but still less than 0.05 so we can say that it is a significant influence on profit margin. The "rho" is 0.557 and another point is that our non-zombie variable with restrictive financial ratios has a t-value of 139 (the higher the t-value, the higher the relevance of the variable). Again, using the RE model the p-value is approximately 0 as well as all two-tail p-values on all variables. Observing the value on the Hausman test, we again should reject the 𝐻0. To the employment growth, FE model show us the same results as the previous models with 29 approximately 0 for the p-value and the two-tail p-values but a "rho" of 0.902. The difference are the values for the firm size regarding the t-values with around 300 in each category of the number of the employees (the highest is within the group of more than 20 but less than 50), so it can be said that this variable has a significant influence on the employment growth. The RE model on the employment, the important thing to highlight is the fact that our nonzombie variable and the interaction between the non-zombie and the capital sunk in the zombie firms have different values for the two-tail p-values with 0.01 and 0.03 respectively. Regarding the employment growth these two variables have a significant impact. Using the Hausman test, previous shown, for the employment growth, we should reject the 𝐻0 hypothesis since the p-value for this test is, approximately, 0. 7.5. Discussion of results Looking at the regressions and the results obtained, the regressions, either the less restrictive or the more restrictive, have almost the same impact in the profit margin and employment growth Portuguese firms. Even with a smaller R square, the fixed effects model should be considered as the most suitable for this regression. Looking at this model we can conclude that, for the restrictive analysis, for example, profit increases by 0.2 percentage points when a firm is non-zombie instead of a zombie, that means that, non-zombie firms have higher profits than zombies (as expected). In this regression, the log of the employment is negative and suggests that employment growth tends to slow if the firm is non-zombie. Also, the operating revenue is positive for both dependent variables which means that, an increase in these variables will increase the operating revenue of the firm. For the more restrictive analyzes, our variable for non-zombie decreases our values for the estimator. So, the profit margin increases by 0.27 percentage points when it is a non-zombie firm, this means that a non-zombie should report higher profit margin that a zombie firm. Again, the log of the employment suggests that the employment growth tends to slow if the firm is non-zombie (negative sign). The operating revenue in this regression is also positive for both variables. 30 Our regressions showed that, if the firm has, for example, more employees, the profit margin and the employment growth tend to increase, ceteris paribus . Also, our variable for the non-zombie is always positive indicating that they have a positive impact on the two dependent variables and the interception between these and the capital sunk in zombie firms provide vital information that zombie firms can corrupt the market. The literature told us that zombie firms, in average, are older and less productive and with this database we can obtain almost the same results (Gouveia and Coelho, 2017). Other authors such as Osterhold and Gouveia (2018), presented a study where they show that zombies, in Portugal, are decreasing as almost as the results presented, and the reduction in the capital sunk in zombie firms fetches positive externalities to non-zombies. Other papers like McGowan et al. (2018), that do not talk directly about the Portuguese economy, find that an increase in zombie share at the industry level is associated with lower investment and employment growth for the average non-zombie and according to both regressions this is true for Portugal. Also, Barnett et al. (2014) provide a relevant study about the employment behavior that allows us to see that zombie firms tend to be more size dependent. Barros et al. (2017) also analyzed the zombies using restrictions on the sectors. They had the same results presented on the point 7.1., where zombies tend to be from real estate, accommodation and food services and construction. Regarding Portugal we have other papers that talk about where the zombies tend to concentrate in terms of zones and, as we could saw the North of Portugal have more firms but less zombies regarding, for example, the Algarve or the autonomous regions of Madeira and Azores. Since the database used was not the best but the results were almost the same as previous and relevant papers (with some differences), these regressions can be accepted and may help in order to analyze more deeply zombie firms. Other important aspect to notice is that the financial ratios may have a huge role when trying to define and describe zombie firms. 37 PT11B - Alto Tamega 10,142 0.58 29.41 PT11C - Tamega e Sousa 60,301 3.48 32.88 PT11D - Douro 24,773 1.43 34.31 PT11E - Terras de Tras-os-Montes 14,775 0.85 35.16 PT150 - Algarve 80,746 4.65 39.82 PT16B - Oeste 60,870 3.51 43.33 PT16D - Regiao de Aveiro 59,985 3.46 46.78 PT16E - Regiao de Coimbra 68,685 3.96 50.74 PT16F - Regiao de Leiria 65,359 3.77 54.51 PT16G - Viseu Dao Lafoes 38,445 2.22 56.73 PT16H - Beira Baixa 11,661 0.67 57.40 PT16I - Medio Tejo 35,550 2.05 59.45 PT16J - Beiras e Serra da Estrela 30,290 1.75 61.19 PT170 - Area Metropolitana de Lisboa 508,769 29.33 90.52 PT181 - Alentejo Litoral 13,387 0.77 91.29 PT184 - Baixo Alentejo 15,079 0.87 92.16 PT185 - Leziria do Tejo 36,112 2.08 94.24 PT186 - Alto Alentejo 15,045 0.87 95.11 PT187 - Alentejo Central 24,581 1.42 96.53 PT200 - Regiao Autonoma dos Acores 22,555 1.30 97.83 PT300 - Regiao Autonoma da Madeira 37,708 2.17 100.00 Total 1,734,879 100.00 Notes: Firms from database divided in NUTS 3 presented in the Portuguese economy. Source: Author’s own calculations according to Amadeus database. Table 8 - NACE code and description 1 A AGRICULTURE, FORESTRY AND FISHING 2 B MINING AND QUARRYING 3 C MANUFACTURING 4 D ELECTRICITY, GAS, STEAM AND AIR CONDITIONING SUPPLY 5 E WATER SUPPLY; SEWERAGE, WASTE MANAGEMENT AND REMEDIATION ACTIVITIES 6 F CONSTRUCTION 7 G WHOLESALE AND RETAIL TRADE; REPAIR OF MOTOR VEHICLES AND MOTORCYCLES 8 H TRANSPORTATION AND STORAGE 9 I ACCOMMODATION AND FOOD SERVICE ACTIVITIES 10 J INFORMATION AND COMMUNICATION 11 K FINANCIAL AND INSURANCE ACTIVITIES 12 L REAL ESTATE ACTIVITIES 13 M PROFESSIONAL, SCIENTIFIC AND TECHNICAL ACTIVITIES 14 N ADMINISTRATIVE AND SUPPORT SERVICE ACTIVITIES 15 O PUBLIC ADMINISTRATION AND DEFENCE; COMPULSORY SOCIAL SECURITY 38 16 P EDUCATION 17 Q HUMAN HEALTH AND SOCIAL WORK ACTIVITIES 18 R ARTS, ENTERTAINMENT AND RECREATION 19 S OTHER SERVICE ACTIVITIES 20 T ACTIVITIES OF HOUSEHOLDS AS EMPLOYERS; U0NDIFFERENTIATED GOODSAND SERVICES-PRODUCING ACTIVITIES OF HOUSEHOLDS FOR OWN USE 21 U ACTIVITIES OF EXTRATERRITORIAL ORGANISATIONS AND BODIES Notes: NACE distribution using letters (from A to U) to define each sector in the economy Source: NACE according to Eurostat Table 9 - NACE distribution in the Portuguese economy NACE code Number of Firms Percentage Cumulative A 61,901 3.57 3.57 B 3,788 0.22 3.79 C 212,061 12.22 16.01 D 1,900 0.11 16.12 E 4,747 0.27 16.39 F 172,889 9.97 26.36 G 473,301 27.28 53.64 H 96,557 5.57 59.21 I 151,872 8.75 67.96 J 40,006 2.31 70.27 K 26,573 1.53 71.80 L 81,294 4.69 76.48 M 169,302 9.76 86.24 N 53,648 3.09 89.33 O 142 0.01 89.34 P 23,180 1.34 90.68 Q 106,861 6.16 96.84 R 19,465 1.12 97.96 S 35,392 2.04 100.00 Total 1,734,879 100.00 Notes: NACE distribution using the letters previously defined to see how many firms there is on this database Source: Author’s own calculations according to Amadeus database. Table 10 - Number of employees in each firm Number of employees Frequency Percentage Cumulative X<=10 1,477,335 85.15 85.15 10<X<=20 128,174 7.39 92.54 39 20<X<=50 86,841 5.01 97.55 50<X<=100 24,103 1.39 98.94 100<X<=250 12,559 0.72 99.66 X>250 5,867 0.34 100.00 Total 1,734,876 100.00 Notes: Number of employees in each firm divided per categories. Source: Author’s own calculations according to the Amadeus database Table 11 - Probit model Variables Parameters 10<X<=20 -0.4609*** (0.0070) 20<X<=50 -0.5175*** (0.0084) 50<X<=100 -0.4239*** (0.0143) 100<X<=250 -0.3910*** (0.0191) 250<X -0.3621*** (0.0273) Agriculture, forestry and fishing -0.1075*** (0.0106) Mining and quarrying 0.2579*** (0.0285) Manufacturing -0.0559*** (0.0059) Electricity, gas, steam and air conditioning supply -0.3041*** (0.0603) Water supply. sewerage, waste management and remediation activities -0.3081*** (0.0445) Construction -0.0785*** (0.0063) Transportation and storage -0.1598*** (0.0073) Accommodation and food service activities 0.4826*** (0.0056) Information and communication -0.1754*** (0.0150) Financial and insurance activities -0.3356*** (0.0208) Real estate activities 0.1038*** 40 (0.0078) Professional, scientific and technical activities -0.3580*** (0.0074) Administrative and support service activities -0.1899*** (0.0124) Public administration and defense 0.6649*** (0.1751) Education 0.2327*** (0.0143) Human health and social work activities -0.5441*** (0.0097) Arts, entertainment and recreation 0.1704*** (0.0187) Other service activities 0.3100*** (0.0113) North -0.1147*** (0.0043) Algarve -0.0257*** (0.0081) Centre -0.1534*** (0.0048) Alentejo -0.0927*** (0.0078) Azores -0.0111 (0.0156) Madeira 0.2194*** (0.0105) pseudo-R2 0.05 log likelihood -340,804.38 N 935,458 Notes: Probit model using the size of the firm, NACE code and the NUTS 2 distribution. N is the number of observations. : Robust standard errors in parentheses. ***denotes statistical significance at the 1% level, ** significance at the 5% level, * significance at the 10% level. Source: Author’s own calculations according to the Amadeus database Table 12 - Marginal effects for the Probit model (Table 11) Variables Marginal effects X<=10 0.1437*** (0.0004) 10<X<=20 0.0656*** (0.0008) 20<X<=50 0.0589*** (0.0009) 50<X<=100 0.0703*** (0.0018) 41 100<X<=250 0.0747*** (0.0026) 250<X 0.0787*** (0.0039) Agriculture, forestry and fishing 0.1090*** (0.0019) Mining and quarrying 0.1914*** (0.0076) Manufacturing 0.1188*** (0.0010) Electricity, gas, steam and air conditioning supply 0.0770*** (0.0086) Water supply. sewerage, waste management and remediation activities 0.0765*** (0.0063) Construction 0.1144*** (0.0011) Wholesale and retail trade 0.1301*** (0.0006) Transportation and storage 0.0997*** (0.0011) Accommodation and food service activities 0.2572*** (0.0015) Information and communication 0.0970*** (0.0025) Financial and insurance activities 0.0727*** (0.0028) Real estate activities 0.1529*** (0.0017) Professional, scientific and technical activities 0.0697*** (0.0009) Administrative and support service activities 0.0946*** (0.0020) Public administration and defense 0.3181*** (0.0613) Education 0.1847*** (0.0037) Human health and social work activities 0.0483*** (0.0009) Arts, entertainment and recreation 0.1689*** (0.0046) Other service activities 0.2057*** (0.0030) North 0.1191*** (0.0006) Algarve 0.1369*** (0.0016) Centre 0.1119*** (0.0007) AML 0.1424*** 42 (0.0006) Alentejo 0.1234*** (0.0014) Azores 0.1400*** (0.0033) Madeira 0.1951*** (0.0027) N 935,458 Notes: Using the size of the firm, NACE distribution and NUTS 2 to compute the marginal effects. N is the number of observations. Source: Author’s own calculations according to the Amadeus database