Sven Boekestijn Student Number: S2373602 Supervisor: prof. dr. L.H. (Lex) Hoogduin
MASTER
THESIS
(M.Sc. Economics)
2 Abstract
This paper explores the relationship between loose monetary policy and zombification in the euro periphery. With firm level ORBIS data I analyze 6 euro countries that were in financial distress during the global financial and euro crisis for the period 2000-2017. I find that there is a negative correlation between the policy rate of the ECB and the probability of firm becoming a zombie. Problems of reverse causality arise. In addition, I find that lower policy rates increases zombie shares in sectors where there is a high external finance dependence. Here there appears to be a causal relationship.
3
1. Introduction
After the global crisis of 2008, central banks in advanced countries, as a response, have lowered policy rates close to zero. This in order to maintain price stability and in some cases (United States) restore output levels. However, when policy rates have reached the so-called ‘’effective zero lower bound’’, conventional monetary policy is not effective as an instrument anymore. These central banks then had to move to unconventional monetary policies. These include large-scale asset purchases (also known as quantitative easing), long-maturity lending to banks, unlimited access to ECB credit and a new way of communicating intended future policy to affect interest rates in the medium/long term, also known as ‘’forward guidance’’. This blend of ‘’loose’’ conventional monetary policy and unconventional monetary policy is commonly referred to as ‘’ultra-loose’’ monetary policy (Claeys & Zsolt Darvas, 2015) (White, 2013). In the United States the central banks have been tightening monetary policy as the economy in the US has been stabilized. Also, the UK has experienced higher interest rates since 2018, although the rise is not as steep compared to the US. This is not the case for the Eurozone. Up until today, the ECB has kept the policy rate extremely low since the 2008 crisis and will continue to do so in the foreseeable future.1
With persistently low interest rates, one of the ‘’unintended consequences’’, as White (2013) describes it, is that banks feel less pressure to restructure their balance sheets. He argues that very low ‘’risk free’’ rates will encourage banks to offer advantageous borrowing conditions to their clients, even though the banks might suspect that the client is insolvent. This ‘’evergreening’’ behavior of banks favors the weak in the economy, the so-called ‘’zombie firms’’. These companies, which under normal competitive circumstances would exit the market, are now staying alive and trap scarce resources that otherwise could have been used by more productive firms (Adalet McGowan, Andrews, & Millot, 2018). This creative-destruction process might be slowing down as is observed in the periphery countries in the Euro area (Gopinath, Kalemli-Özcan, Karabarbounis, & Villegas-Sanchez, 2017).
Theory suggests that one of the main reasons of the slow recovery post-crisis in the Eurozone is the zombie lending behavior of banks like the episode we have seen in Japan in the 90s (Peek & Rosengren, 2005). Zombie firms have an impact on aggregate productivity in two ways. First, an increase in zombie lending crowds out investment for healthier firms and more
4 productive firms (Andrews & Petroulakis, 2019) (Adalet McGowan, Andrews, & Millot, 2018). Credit from banks that normally would be available for new or existing productive firms are now trapped in unproductive non-viable firms. Second, in a well-functioning economy, capital flows from the least productive firms to the most productive firms through the mechanism of productivity-enhancing capital reallocation. However, through zombie congestion this mechanism is being impaired as empirical literature has shown (Andrews & Petroulakis, 2019) (Gopinath, Kalemli-Özcan, Karabarbounis, & Villegas-Sanchez, 2017). For Europe, these are significant findings as Gopinath et al. (2017) describe that capital misallocation is one of the major factors for the productivity slowdown in in the southern European countries. In particular Spain and Italy, as I will show later on in the analysis also have the highest zombie capital shares in the Eurozone.
In this paper, I will explore the relationship between monetary policy and zombification in the periphery countries of the Eurozone. With ORBIS firm level financial data for the Eurozone, in particular SMEs, I determine to which extent monetary policy had an effect on the rise of zombie firms. My final data set contains firm level data for 6 European countries for the period 2000-2017. I use the MRO rate of the ECB as a proxy for monetary policy. I construct multiple identification strategies for zombie firms and discuss the typical characteristics of these firms. For identifying zombies in my baseline regression, I closely follow Storz et al. (2017) where a zombie firm is defined as a firm that has a negative return on assets and has a relatively low ratio of EBIT over total debt. I avoid identification strategies which involve analyzing a firm’s capabilities to meet interest payments, since these are likely to be highly correlated with the policy rate of the ECB.
Since the dependent variable zombie in my regression is a binary variable, I use a conditional fixed effects logit regression. After controlling for cyclical influences by adding year fixed effects, the odds ratio of the policy interest rate variable is around 0.4. Assuming a causal relationship this signifies that a one percentage point increase in the policy rate would decrease the likelihood of a firm being a zombie that year by 60%. These results are robust to different identification strategies for zombie firms and for different subsamples like the pre-crisis period and by excluding several countries. However, I find that my regressions might suffer from endogeneity issues because of reverse causality.
5 main refinancing rate. The idea behind this is that sectors that are more dependent on external finance, might be more sensitive to changes in the policy rate set by the ECB. I find that indeed that there seems to be a negative relationship between the interaction term and the zombie capital share. A one percentage point increase in the interest rate decreases the capital zombie share by nine percentage points. Here, there does not appear to be a reverse causality problem. The results are robust even with a conservative fixed effects structure on a sector level and controlling for country*year cyclical influences.
My research adds to the literature in a couple of ways. First, I expand the existing literature with more recent data for the years 2015-2017. This allows me to inspect the effects of the lower policy rates in the post crisis period. Most of the previous literature focused on the pre-crisis and the euro crisis period, where zombification was at its peak. In addition, the ORBIS dataset allows me to analyze a relatively large sample of SMEs, where for example Banerjee & Hofmann (2018) only focused on listed firms and zombie prevalence. Lastly, where my sample is similar in size and source to previous literature (Storz , Koetter, Setzer, & Westphal, 2017) (Andrews & Petroulakis, 2019), the focus of this paper is on the influence of the policy interest rate, whilst their focus was on weak banks and zombification.
6
2. Literature Review
Since the late 80’s the prevalence of zombie firms has increased. In the work of Banerjee & Hofmann (2018), they gathered firm-level data in 14 OECD countries from 1987-2016. The analysis showed that the average share of zombie firms of all listed non-financial firms has increased from around 3% in 1987 to around 12% in 2016. Similarly, Adalet McGowan et al. (2018) observe, in their analysis of eight OECD countries, an increase in the share of zombie firms in the years 2003-2013. In the early 2000s in Japan, around 30% of Japanese firms were on life support from banks (Caballero, Hoshi, & Kashyap, 2008). This percentage was still around 5% in the 80s.2
One might ask whether the increase in the incidence of zombie companies is just episodic, generated by financial disruptions, or if it follows a more secular trend. It takes a longer-term perspective to analyze these trends. The literature suggests that zombification has trended up over time in upward shifts during crises, but this does not seem to reverse entirely in the subsequent recovery period (Banerjee & Hofmann, 2018). This is an interesting observation given the fact that recessions can be a good breeding ground for productivity enhancing reallocation. In a normal functioning market, low productive firms during a recession are being forced to either exit the market or restructure (Caballero & Hammour, 1994).
Causes of Zombification
There is an extensive amount of research on zombification and its aggregate effects, especially since there has been a renewed interest in the topic since the Great Recession. The main cause according to Banerjee & Hofmann (2018) for a rise in zombie firms, according to their data, is that unhealthy firms are more likely to survive for a longer period. The probability that a firm remains in a zombie state the following year has increased from 60% in the late 1980s to 85%. However, the literature on potential causes of this phenomenon is relatively slim.
First, one of the potential causes is the so-called ‘’evergreening’’ behavior of banks. While in the United States during the capital crunch experience in the 1990s banks drastically shrank their loan portfolio in order to increase capital ratios, this is not what was observed in
2 It is to be noted, as the authors mentioned themselves, that this is a unweighted percentage. The asset-weighted
7 Japan. Peek & Rosengren (2005) describe that the reason for this, in part, is that bank regulation and supervision policies provided little incentive for heavily undercapitalized banks with significant nonperforming loans to clean up their balance sheets. Weak banks are facing mutually destruction assurance by their insolvent borrowers. If a borrower cannot make its interest payments, the bank is indirectly being forced to provide additional loans. Otherwise, the bank has to report their increase in nonperforming loans or the firm has to file for bankruptcy. In either case, this would weaken their own reported capital, which could be catastrophic with already heavily undercapitalized banks. Thus, it is in the own interest of the bank to engage in polices of forbearance with their unhealthy borrowers. This causes the bank to ‘’evergreen’’ their loans. Doing this, the balance sheets of banks look better, since the bank is not required to report these ‘’bad’’ loans among their nonperforming loans.
Some empirical findings in the literature indeed observe this ‘’evergreening’’ behavior of banks, for example in the case of Japan. In the paper of Peek & Rosengren (2005) they analyzed detailed data in Japan from individual lenders to individual firms. The authors show that indeed there is an incentive for banks to ‘’evergreen’’ loans to their weakest borrowers. This allows zombie firms to remain in the market. In addition, these troubled borrowers seem to perform poorly even after receiving the additional credit. Similarly, Caballero et al. (2008) also observe an increase of evergreen lending of Japanese banks in the early 2000s using firm-level data. Using not only listed firms, as the previous papers did, but also non-listed firms, Kwon et al. (2015) showed that especially the SMEs were more reliant on bank financing than the larger firms.3 Not surprisingly therefore, the authors observed a higher share (36.2%) of
zombie firms in the unlisted category than the listed firms (25.8%) in 1999.4
Evidence from Japan could provide us with some proximate causes of zombification in the Eurozone. Most of the recent empirical literature on EMU countries linked the increase of zombification to a weak bank environment (Andrews & Petroulakis, 2019) (Acharya, Eisert, Eufinger, & Hirsch, 2019) (Storz , Koetter, Setzer, & Westphal, 2017). For the Eurozone this is especially interesting since the financial crisis severely damaged the health of banks in EMU
3 The authors describe that bank loans accounted for 20% of total debt for large firms but for SMEs, this was
around 40%.
4 Andrews & Petroulakis (2019) describe that further work on the Japanese episode provided some nuance to the
8 countries. This is mainly the case for banks in the periphery countries.5 Acharya et al. (2019)
argue that, even though financial stability was regained in the banking sector, there was not a robust recovery of the real economy. The authors provide systematic evidence that the slow recovery is partially due to zombie lending. This is similar to the previous case observed in Japan. Undercapitalized banks extend advantageous loaning conditions to their impaired borrowers so that they in turn have the liquidity to meet their payments. This way the banks do not have to incur immediate losses and hope these insolvent firms become solvent again in better times. An extensive analysis on the consequences of zombie lending in Italy by Schivardi et al. (2017) find the same results. Storz et al. (2017) also found this positive relationship between stressed banks and zombie lending. However, they did not find the positive link between increase in bank stress and an increase in the indebtedness of SMEs, as in the paper of Acharya et al. (2019).
Next to weak banks and ‘’evergreening’’ behavior as a result, interest rates might have an impact on the increase in zombie lending as well. On the one hand, lower interest rates could reduce the amount of zombie firms since the interest coverage ratio6 improves as the interest expenses decrease. On the other hand, low interest rates reduces the incentive to address debt overhang and resource misallocation. The main channel responsible for this is the banking sector. Low rates decrease the pressure to clean up their balance sheets. According to Borio & Hofmann (2017) this can happen in multiple ways. First, if the interest rate decreases then the discount factor decreases as well. This in turn, increases the expected recovery for non-performing loans making it more likely for banks to roll over then to charge-off these NPLs. Secondly, with low rates the opportunity cost of carrying NPLs on a bank’s balance sheet is low, since the return of alternative investments and the costs of funding these NPLs are low.
Empirically, the evidence connecting low interest rates to zombification is still rather slim. Acharya et al. (2019) use the OMT announcement to link to possible misallocations. They find that the banks that benefited the most from this announcement, through the reevaluation of their sovereign bond holdings, increased their loan supply. However, this was mainly targeted at unhealthy firms and thus increased credit misallocation. Lepetit et al. (2012), in their bank-level regressions in Europe, showed an increase in loan charge-offs with higher short-term nominal interest rates. Similarly, Borio et al. (2015), in their empirical analysis of 109 large
5 Archarya et al. (2019) group them as GIIPS countries (Greece, Ireland, Italy, Portugal and Spain). They observed
that zombie lending was most pronounced in Italy, Spain and Portugal. Storz et al. (2017) used a slightly different sample for the pheriphy countries (they excluded Italy and added Slovenia).
6 The interest coverage ratio is frequently used to identify zombie firms. It indicates how well a firm is able to
9 international banks, found a positive relationship between the short-term nominal interest rate and loan loss provisions of banks. In addition, they found that the sensitivity increases when interest rates are really low. So the impact of changes in short-term nominal interest rate increases when policy rates are close to zero. A long-term perspective analysis of interest rates and zombification is done by Banerjee & Hofman (2018). With firm-level data of listed companies from Datastream Worldscope in 1986-2016, the authors observe a negative relationship between the policy rate and the share of zombie firms per country. In their analysis, they use the measure for a zombie firm as defined by Adalet Mcgowan et al. (2018) as a firm with an interest coverage of less than one for three consecutive years and older than 10 years. Additionally, in their narrow definition of zombie firms, Tobin’s q has to be below the median firm in the sector in a given year. They regressed the country zombie shares on their own lags along with five lags of the nominal short-term interest rate and five lags of price-to-book ratio as a measure of bank health. In contrast to the other literature, they did not find a clear-cut relationship between bank health and zombification. However, the negative relationship between zombie share and the level of interest rate of the previous five years was statistically significant.
Loose monetary policy
The literature linking monetary policy and zombification is relatively slim. Most of the empirical literature has an indirect link through the interest rate channel of monetary policy as described in the previous part. There are some papers, however, studying the direct effects of some unconventional monetary policies on the lending behavior and zombification.
10 had a high ‘’windfall gain’’. However, the major part of this increase in lending volume was targeted to low quality borrowers (i.e. zombie lending). These findings are in line with the existing theory of lower interest rates, increased lending capacity in unhealthy banks and zombie lending.
As is the case for the OMT announcement, other unconventional monetary policies also could induce zombie lending through the same mechanism. Krishnamurthy & Vissing-Jorgensen (2011) found a ge
n
eral reduction in overall yields when studying the quantitative easing programs in the US. Similar evidence is found in the paper of Joyce et al. (2011), which analysis the effect of QE on yield spreads and asset prices in the UK. They found a significant decrease of 100 basis points in medium to long-term gilt yields due to QE. As the case for the EMU, Afonso & Tovar Jalles (2019) found some evidence that unconventional monetary (QE) policy in the Euro area has contributed to the decrease in sovereign yields, especially after the 2009 crisis. In specific, they find a significant reduction in the sovereign yield spread after the implementation of the LTROs and the CBPP1 programs of the ECB. In a similar paper of Afonso et al. (2019) they found that key interest rate announcements of the ECB negatively affected the spreads on sovereign yields in their sample countries. In addition, as in the previous paper, they found that the announcements of nonstandard measures of the ECB decreased sovereign yields.3. Data
In my paper, I use firm-level data from the ORBIS and AMADEUS database of Bureau Van Dijk (BvD)7. Their database contains a large amount of financial, balance sheet and output data of around 250 million companies worldwide. The main advantage of this dataset in comparison to others is that it has a large amount of available data on non-listed SME’s. These SME’s are more likely to depend on external finance from banks than large listed firms as explained by Storz (2017). It is therefore better suitable for my regression, since I am interested in the effect of monetary policy on zombification, and this transmission works through banks.8
7 The ‘’flat-file’’ download contains the combined disks of AMADEUS and ORBIS, where AMADEUS is the
European data edition. In total, this file has financial data of over 50 million European firms in the period 1986-2015. The two additional years (2016 and 2017) were extracted from the ORBIS webserver and added to the datafile.
8 Acharya et al. (2019) explained that the OMT announcement had a positive effect on banks’ balance sheets
11 In previous empirical literature, most of the zombification is observed in the periphery countries (Storz , Koetter, Setzer, & Westphal, 2017) (Acharya, Eisert, Eufinger, & Hirsch, 2019). Therefore, I restrict the data to EMU periphery countries to observe the influence of monetary policy in these regions.9 I closely follow the procedure of Kalemli-Özcan et al. (2015) to transform the raw data into harmonized data for my empirical analysis.10 I limit my data to countries and years where financial data is reported in the euro currency. This way there is no need for harmonization across different currencies. Data is harmonized across consolidation codes and reported units11The closing dates are given each a yearly value.12
Furthermore, I exclude certain observations from my dataset. First, I exclude certain industries that in general show very different firm characteristics, such as the primary sector and the financial sector. For banks for example, one cannot use the same zombie identification strategy as for a regular firm. Because of the NACE code variable in the database, I am able to distinguish between sectors and drop these observations accordingly. The dataset in the analysis is left with NACE codes 10-82 (excluding 64-66, financial companies).
Second, observations with missing key variables are also excluded. These include observations without a consolidation code, number of employees, date of incorporation or firm id number. Also, firms with missing key financials are excluded from the sample. These include missing turnover, provisions, total assets, loans and fixed assets. Also firms with a negative firm age are excluded from the sample.13
Finally, observations that show clear balance sheets inconsistencies are excluded. Specifically, these are zero or negative total assets, non-matching asset and liabilities observations, negative debt, negative fixed assets, negative employment or employment above 2 million (higher than Walmart) .
aggregate lending by the banking sector, but most of this was due to an increase in zombie lending, especially in the periphery countries.
9 In total, I will use 6 countries. Spain, Portugal, Greece, Slovenia, Ireland and Italy. Some might argue that
Slovenia is not actually a periphery country, but since in the previous mentioned literature it has been shown that during the global and euro crisis Slovenia had some severe economic problems, I also include them. Due to data restrictions countries like Cyprus are excluded from my sample. Even though it is a periphery country.
10 The main issue with the ORBIS and AMADEUS datasets is that the data is collected for the private sector with
the aim of financial benchmarking. Thus, some steps need to be taken to use it for economic analysis. Kalemli-Özcan et al. (2015) created an extensive guide where they explain step-by-step how to use and combine different ORBIS and AMADEUS raw data vintages.
11 Observations are reported in regular units, thousands, millions or billions. Transformation into the same units
was therefore required. Observations that showed irregularities after the transformation were dropped.
12 Many firms had multiple reports per ID-YEAR. This was mainly due to a change in report date (for example
from December to March) or because of monthly reports. I filtered out the monthly reports and kept the report with the account date that was the closest to the end of the year.
13 Some firms have a date of incorporation that is later than the financial account year of their financials. This is
12 Identifying zombies
In previous literature, the interest coverage ratio (IC) is one of the most common ways of defining a zombie firm. Adalet McGowan et al. (2018) define zombies, in their base analysis, as a firm that has an interest coverage ratio of smaller than one, for three consecutive years. They argue that the interest coverage ratio is well suited for comparison across countries and that it can capture other channels through which zombie firms may be kept alive, like insolvency regimes or non-performing loans.
This in contrast with the definition of Caballero et al. (2008), who defined zombies as firms that received subsidized credit. Adalet McGowan et al. (2018) created a simplified version of the subsidized credit definition that is also implementable with data from ORBIS. They describe the following formula:
𝑅𝑖,𝑡∗ = 𝑟𝑠𝑡−1𝐵𝑆𝑖,𝑡−1 + (1
5∑ 𝑟𝑙𝑡−𝑗
5
𝑗=1
) 𝐵𝐿𝑖,𝑡−𝑗
This formula is predicated on the idea that a firm should pay a benchmark interest rate (R*) depending of the firm’s debt structure. In this formula rst is the short term interest rate paid on
short term loans (BSi,t) for firm i at the end of year t and rlt is the long term interest rate paid on
the firm’s long term debt (BLi,t).14 When a firm is paying interest that is below this benchmark
interest rate, it is said to be receiving subsidized credit. In order for this definition to be accurate we need very detailed information about a firm’s debt structure. ORBIS does not allow for this, and thus this will not be my main identifier for my analysis. I will use this definition as a robustness test later in my analysis.
The main issue regarding the interest coverage ratio for my analysis is that monetary policy might have an impact on the average amount of interest paid by companies. Expansionary monetary policy decreases policy rates of the central bank and therefore can have a downward effect on the lending rate of financial institutions. All firms, including unproductive ones, will therefore see an improvement in their interest coverage ratio. Even if
14 I collected data for bank lending rates to corporates from the ECB Statistical Data Warehouse. Only complete
13 these firms have not become more profitable or realize an increase in operating income. As a result, a lot of zombie firms will not be captured by the IC ratio as a definition for zombies. If we look at Figure 1, I observe that indeed over time, the average amount of interest paid by firms has declined. In the entire periphery one observes that the median interest rate paid by firms on loans and long-term debt has decreased from around 9.5% in 2000 to around 3.7% in 2017. The biggest difference is observed in Portugal where the median has decreased from 9.8% in 2000 down to 2.8% in 2017. Keeping everything else equal, this would improve the IC ratios and reducing the amount of zombie firms.
For the baseline regression of my analysis, I therefore closely follow the definition set up by Storz et al. (2017). A firm is considered a zombie when it has either (i) a negative return on assets or (ii) negative net investments15, and in addition (iii) a low debt service capacity16
for at least (iv) three consecutive years. So, these requirements have to be fulfilled for the current and previous two years in order to be viewed as a zombie. In addition, we only consider a firm a zombie if it is older than 10 years in our regression, as we do not want to confuse young
15 Total fixed assets change in comparison to the previous year. 16 EBIT over financial debt (loans + long-term debt) lower than 6%
Figure 1.
Percentage of interest paid over total financial debt
The graph shows the evolution of the amount of interest expenses a firm had to pay over its total financial debt. Where total financial debt is defined as long-term debt + loans. The percentages are median values of all firms in the sample per country year. The graph includes Spain (ES), Italy (IT), Portugal (PT) and the median value of all 6 countries in the sample (Spain, Portugal, Greece, Ireland, Slovenia and Italy). Showing a separate line for each country provides similar results in terms of the downward trend. The total sample size was 12,096,870 observations over a period of 17 years. Source: ORBIS data and author’s calculations. 0% 2% 4% 6% 8% 10% 12% 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Interest paid over financial debt
14 expanding firms with zombie companies. Requirement (iii) ensures that firms receiving subsidized credit are not captured as healthy firms, which often can be the case with the interest coverage ratio. Especially in my analysis, where interest rates17 have been decreasing over time,
this measure is more suitable for identifying zombie firms. Companies that have a debt servicing capacity of 6% means that they have an interest coverage ratio of 1. considering that the median percentage of interest paid of the whole sample is around 6%. Through this method performance effects are captured more accurately and identifying zombies becomes less blurry as interest rates go down. In addition, this definition of zombie firms also captures the increase in the leverage of a firm, as this would lower the debt servicing capacity ratio.
Characteristics of zombie firms
Before continuing to the analysis, I first identify some of the characteristics of zombie firms. I analyzed the data from 2015-2017 of all 6 countries and used the debt servicing capacity mentioned above for identifying zombie firms.
In Figure 2 a quick overview of zombie characteristics is listed. Panel A shows the zombie share differences for firm size. In my sample the larger the firm gets, the higher the probability of the firm being a zombie. The share size of zombie firms is around 6.8% for the small countries and around 8% for large countries. Adalet McGowan et al. (2018) argue that larger firms are more likely to receive government subsidies because they are less willing to incur the losses of relatively large firms, especially during times of crisis. The authors add to this that bank forbearance and bank relationship is observed more regularly if the bank views the company as a relatively “big” client. This also adds to the increasing probability of large firms turning into zombies. Very large companies, however, see a slight decrease in the zombie share compared to the large firms. This is in line with Hoshi (2006), who describes that with very large firms the idea, that being a large firm increases the probability of becoming a zombie firm, does not hold anymore. In panel B, a similar result is observed. Now, as a measure of size, the number of employees is used. As the number of employees increases, so does the likelihood of being a zombie. Companies on the low end, however, have a higher probability of being a zombie.
15 The likelihood of being a zombie firm is higher for firms as the firm’s age increases. Panel C shows that share of zombie firms increases by a substantial amount once the firm has reached the ten-year-threshold. The zombie share increases from 7% to about 9%. The idea behind this is as firms get older, they tend to increase in size or number of employees. This would in turn increase the likelihood of bank forbearance or government subsidies. Interesting to note however, is that in this sample firms between 1-5 years old have the lowest likelihood
Figure 2.
Different characteristics of zombie firms in 2017
Panel A: Share of zombie firms by ORBIS size qualifications
Panel B: Share of zombie firms by number of employees
Panel C: Share of zombie firms by firm age Panel D: Zombie share in terms of firms and capital stock
Note: Zombie firms are calculated as firms with a debt servicing capacity lower than 6% with either negative investments or negative net income for the years 2015-2017. Countries included in the sample are Spain, Greece, Ireland, Italy, Portugal and Slovenia. Number of observations is 3,073,168. ORBIS provides firms with a firm size variable. Very large firms are either listed or have a total asset value of over 200 million euros. Large firms have total asset value of at least 20 million euros. For medium sized firms the required minimum of total assets is 2 million euros. Panel A, B and C are expressed in percentages of the total within the specific group. Panel D shows the contrast between the percentage of zombie firms per country and the share of capital sunk in zombie firms per country. Source: ORBIS data and author’s calculations
16 of being a zombie. This is contrary to the idea that start-ups may be struggling in the beginning to have positive economic returns.18
In panel D, both the share of zombie firms as the share of capital sunk in zombie firms is illustrated. Here, the share of capital sunk is a weighted measure for zombification and is commonly used in previous literature (Adalet McGowan, Andrews, & Millot, 2018) (Banerjee & Hofmann, 2018) (Andrews & Petroulakis, 2019). This weighted average is better suited to derive an order of magnitude and economics consequences of these zombie firms.19 In some countries, for example Portugal or Slovenia, the differences between these shares are not as substantial, respectively 4% and 5% for Spain and 4% and 6% for Slovenia. Large differences are observed in Greece for example where the zombie share is around 1%, but the share of capital sunk lies around 4%. Even stronger contrasts are seen in Italy, where the share of capital sunk is almost fivefold compared to the firm share. Exactly the opposite is the case for Ireland and Spain, where firm shares are higher than the weighted capital shares.
Progression of zombification
There has been a prevalence of zombie firms in the euro periphery during the global financial crisis and the following euro crisis. Panel A of Figure 3 illustrates this increase over time from 2003-2017. In 2003 the share of zombie firms was relatively low with a median of around 1%. Then around 2010-2011, there is a significant rise in zombie firms which remains persistent and peaks around 2013-2015. From there, the share of zombie firms has decreased, while some shares still remain higher than pre-crisis levels. Spain, Ireland and Portugal have experienced relatively the largest differences in shares, respectively 13% and both 8% at peak level. Note that Ireland experiences the rise earlier than Spain and Portugal. The global financial crisis affected Ireland in a relatively severe manner while Spain, and the other periphery countries, experienced the biggest shock during the euro crisis that followed. Greece, Slovenia and Italy have similar progressions in terms of zombie shares, with a peak around 6%, but have not experienced zombification as severe as Spain. Interestingly, there is a slight decrease in
18 Adalet McGowan et al. (2018) saw in their data that there were two peaks in zombie firms. One was in the group
were the firm’s age was relatively low and the other peak was observed in relatively old firms. They used a sample that was from 2013, in the midst of the euro crisis. Therefore, one could argue that startups, especially in crisis times would have a harder time generating positive economic returns in comparison to a startup in 2017, when the euro crisis was relatively stabilized.
19 Empirical literature has frequently linked high capital shares sunk in zombie firms to low productivity, rising
17 Figure 3
Percentage of zombie firms per country
The graph in Panel A illustrates the progression of zombie firms in the euro periphery countries for the period 2003-2017. Panel B provides a quick overview of the zombie share at the beginning of the sample and at the end. A firm is identified as a zombie if it has a debt service capacity lower than 5% and either a negative return on assets or a negative net investment. All requirements have to be met for two consecutive years. Only firms older than 10 years can be considered a zombie. Percentage are calculated as the amount of zombie firms over total firms in sample per country. Countries included are Spain (ES), Greece (GR), Ireland (IE), Italy (IT), Portugal (PT) and Slovenia (SI). Data for Slovenia starts in 2007, when the euro was adopted. For this reason Slovenia is excluded from Panel B. Source: ORBIS data and author’s calculations
Panel A: Zombie share per year 2003-2017
Panel B: Differences in zombie capital shares per country for year 2004, 2014 and 2017
18 zombification after 2015. Previous literature about zombification mainly
saw a constant increase in zombie share as well as the amount of resources sunk in zombie companies. This could be because existing literature analyzed zombification in the Eurozone up until 2015.20 On the other hand, zombification is still relatively high in comparison with 2004. Panel B of Figure 3 demonstrates the differences in zombie capital shares of the beginning of the sample and the end, with the peak value in between. Spain had a capital share of less than 2%, while in 2017 this share was still around 6%. While in Panel A it seemed that Italy did not have a relatively high share of zombie firms, looking at the capital shares shows a different picture. Here, one can see that Italy in 2017 has the highest share of capital sunk in zombie firms of around 9%. This capital share in Italy in 2004 was still around 2%. So out of the five countries shown in Panel B, three show the same pattern of higher shares of zombie firms post crisis than pre-crisis. This is line with empirical findings of Banerjee & Hofmann (2018), where they have found that zombification peaks during crisis times, but that during the recovery period not all of these firms exit the market. Only in Ireland and Portugal the reversal of zombification seems to be happening in the years after the crisis. Ireland had a zombie capital share of around 1% in 2004 and decreased to almost 0% in 2017. Portugal went from a 9% capital share in 2004 to around 5% in 2017
Loose monetary policy and control variables
To measure the policies of the ECB I use the main refinancing rate of the ECB as a proxy. I obtained this data from the ECB Statistical Data Warehouse for the years 2000-2017. In addition, I include country controls in my analysis, GDP and the 10-year Government Bond yield. The GDP data is retrieved from AMECO (European Commission database) and is defined as the contribution of change (in percentages) in GDP at constant prices of total consumption.
For a complete overview of the variables see Table A1.
20 Acharya et al. (2019) observed an increase in zombie lending after the OMT announcement with observations
19
4. Empirical Framework
For establishing the relationship between zombie firms and monetary policy, I follow Andrews & Petroulakis (2019) and use following fixed effects conditional maximum likelihood panel estimation as a baseline model for 6 countries21, over the period 2000-2017:
𝑍𝑜𝑚𝑏𝑖𝑒𝑖𝑠𝑐,𝑡 = 𝛼1𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡𝑟𝑎𝑡𝑒𝑐,𝑡−2+ 𝛼2𝑍𝑜𝑚𝑏𝑖𝑒𝑖𝑠𝑐,𝑡−2 + 𝛼3𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑠𝑐,𝑡−2+ 𝛾𝑡+ 𝛼𝑖 + 𝜀𝑖𝑠𝑐,𝑡
(1)
In their regression they use the variable “bank health” to establish the relationship between bank health and zombification. I choose here for a conditional logit regression whereas they use OLS. Using a fixed effect structure for allows me to observe for unobserved characteristics of individual firms that do not change over time. With a conditional logit model this means that only observations that change over time will be incorporated in the model. This means that if a firm is a zombie during all periods in the dataset, it is eliminated from the regression. The same applies to firms that are non-zombies during the whole period.
My focus on monetary policy and its impact on zombification. Therefore, I use the policy interest rate as a proxy. The dependent variable Zombie will take a value of 1 when the firm is classified as a zombie in a given year and 0 otherwise. To capture the effects of loose monetary policy on zombification, I use different measures of the interest rate in the analysis. In the baseline regression, however, the variable Interestrate is the main refinancing rate of the ECB. The interest rates enter the regression with a two-year lag to capture the effect of zombification at the beginning stage. This, because a firm is classified as a zombie if it shows persistent financial weakness for at least three consecutive years.22 I also add a lagged value of the dependent variable. This is because, presumably, if a firm was a zombie firm the previous year, the chances of it being a zombie in the current year is higher than if it were not a zombie in the previous year. The firm level control variable firm size as measured as number of employees is added to the regression.23 In addition country level economic variables like the
21 These countries are: Spain, Portugal, Greece, Slovenia, Ireland and Italy. For comparability purposes, only
financials for the years after 2007 were included for Slovenia, as this is the moment they adopted the euro.
22 In further robustness tests, more years are added. For example 4 consecutive years of financial weakness means
also that the policy rate variable enters with a three year lag instead of two.
23 In the paper of Andrews & Petroulakis (2019), they show that firm size (number of employees) is indeed a
20 GDP growth rate and the ten year government bond yields are also added. These control variables are captured in the explanatory variable Controls. The regression uses a firm and year fixed effects, 𝛾𝑡+ 𝛼𝑖 , which captures time-varying shocks as well as time-invariant firm
characteristics. Since my hypothesis is that loose monetary policy increases zombification (so a decrease in the interest rate causes an increase in zombie firms), I expect α1 < 0 (and an odd
ratio < 1) in the conditional logit regression. An odd ratio can in my case being interpreted as the likelihood of a firm being a zombie as a consequence of a unit change of one of the explanatory variables, all else equal, compared to not changing the explanatory variable by one unit. So if the main refinancing rate would change by one percentage point and the odds ratio is 1, then changing the main refinancing rate would not impact the likelihood of a firm being a zombie. If it is smaller than one, for example 0.9, then increasing the main refinancing rate by one percentage point would decrease the likelihood of a firm being a zombie by 10%, all else equal.
Data diagnostics
Before I start analyzing the results of my regressions, it is important to do basic data diagnostics to tackle some potential issues regarding the data. First, I start with potential multicollinearity. Table A2 in the appendix shows the different VIFs for my explanatory variables in my regression. The mean VIF is smaller than 10, which tends to be the lower limit for further inspection (Henseler, Ringle, & Sarstedt, 2015). In the paper of Ringle et al. (2015) they argue for a lower limit VIF of 5. In my case all the values are around 2, so no strong sign of multicollinearity. Table A3 shows the correlation between the explanatory variables. Government bonds seem to be weakly correlated with the main refinancing rate (policy rate) and GDP. The correlation is not above .5, so the relationship is not very strong.
Due to the vast amount of panels I have in this dataset, it is quite a challenge to perform a heteroscedasticity test on the model. Table A4 present the results of a heteroscedastic probit model. The Wald test of the sigma squared rejects the null hypothesis of no heteroscedasticity in the model, so it appears that the errors are not homoscedastic. Since my dataset has a very large number of panels (N) and a relatively small T, using a fixed effects model and clustering the errors by each panel (firm) is an appropriate solution to deal with autocorrelation24 and heteroscedasticity in my model.
24 The Wooldridge test for autocorrelation gives an F-value of 190,452 (probability > F = 0.000), so the null
21 Table A5 shows the Levin-Lin-Chu unit-root test for the explanatory variables. The government bond variable shows the highest p-value, but is still relatively low. For all variables the null hypothesis of the panels containing unit roots is rejected against the alternative hypothesis that all panels are stationary.
In addition, I performed a Hausman test to compare the random effects model to the fixed effects model. Both random effects models, logit and probit, were compared to the conditional fixed effects logit model. The χ2 values were respectively 133,309 and 147,635 , so
the null hypothesis of no systematic differences in the coefficients is rejected. Intuitively, it is reasonable to assume that due to a wide range of possible omitted variables25 that causes zombification, individual firm effects are correlated with the regressors. A fixed effect models therefore seems more appropriate.
5. Results
Table 1 presents the results of the baseline estimation (equation 1) of my analysis. In this regression a conditional fixed effect structure is used. This means that only firms that differ across time in terms of the zombie variable are used in the regression. Firms that are zombie for the entire period or firms that are non-zombie for the entire period are excluded from the sample. The dependent variable is zombie, which takes the value of 1 when the firm is identified as a zombie firm. The regression is estimated with firm and year fixed effects. Firm fixed effects enables me to capture time-invariant effects like differences between individual firms, but also at the same time, differences between countries and industries. The year fixed effects capture global economic shocks in the model.26 The explanatory variables enter the regression with a two-year lag, this in order to capture the zombie firm at the beginning stage.
Column 1 shows that the interest rate variable enters the regression with an odds ratio less than 1 as expected. This indicates that an increase in the policy rate of the ECB decreases the likelihood of a firm being a zombie. To be precise, a one percentage point increase in the policy rate would decrease the likelihood of a firm being a zombie by 70%. With the policy rate as the only time-varying variable, it is likely that the regression suffers from endogeneity
25 Weak banks, insolvency regimes and exit margins
26 The majority of the year dummies are highly significant at a 1% level. In the period 2008-2009, during the global
22 Table 1
Baseline panel regressions results for dependent variable zombie
The dependent variable is the binary indicator zombie. This variable takes the value of 1 when a firm has a debt servicing capacity ratio of below 6% and either negative investments or negative return on assets. Both requirements have to be met for three consecutive years. Policy interest rate is the main refinancing rate as set by the ECB. GDP is the Gross Domestic Product growth rate in comparison to the previous year at constant prices of total consumption. Government Bond is the yield of a 10-year government bond of a country in a specific year. The policy interest rate, GDP and Government Bond variables all enter the regression with a 2-year
lag. Zombiet-1 is a zombie dummy describing if the firm was identified as a zombie in the previous period. The Employment variables
are dummies for the size qualification of a firm. Robust standard errors are clustered by firm (parenthesis). Sample of 2000-2017. Countries included are PT, IT, ES, SI, GR and IE. Significance levels are denoted respectively *** for 1%, ** for 5% and * for 10%. Both the employment and year fixed effects are significant at a 1% level, so these are not denoted by asterisks. Year 2017 was omitted due to multicollinearity. Coefficients are given as odds ratios. Source: ORBIS data and author’s calculations.
(1) (2) (3) (4)
Panel A: Conditional Maximum Likelihood Estimation (CMLE) with firm fixed effects
Policy interest rate 0.309*** 0.324*** 0.404*** 0.393***
23 because of omitted variables. These are added to the regression in the following columns. Previous literature has demonstrated that zombification tends to intensify during times of crisis (Banerjee & Hofmann, 2018). At the same time, counter-cyclical monetary policy ensures that interest rates go down during an economic downturn. It is therefore apparent that these two variables show a strong association, while the causal relationship may be less clear-cut.
In order to account for economic cyclicality, column 2 adds the GDP growth rate and the government bond yields. Both enter the regression with a significant odds ratio of less than one. For GDP, intuitively, this is as expected. As GDP goes up, the economy is growing, firms become more profitable and this puts a downward pressure on zombification.
The odds ratio below one for the government bond yields also implicate that as these yields go up, the likelihood of being a zombie goes down. This is interesting, because one could expect this sign to be positive because during the euro crisis, government bond yields in some periphery countries (Portugal, Ireland and Greece) had risen dramatically. The main cause for this was a significant loss of trust of the investors in the solvability of these countries. One can assume therefore that an increase in these yields signifies an economic downturn.27 This would increase the likelihood of a firm turning zombie. An argument for the opposite effect can also be made. As a response to the drastic increase in sovereign yields, the ECB successfully calmed markets down by the OMT-announcement. As explained previously, in the literature section, Acharya et al. (2019) found that these unconventional
policies increased zombie lending by banks. In addition, as discussed before, QE also might have contributed to the decrease in sovereign yields at a later stage and also might have increased zombie lending through the same mechanism as the OMT announcement. The fact that the sign here in my regression is negative therefore shows that the latter explanation is dominant.28
Column 3 adds the lagged value of the dependent variable zombie. The highly significant positive sign indicates that if a firm was a zombie firm in the previous year, it is more likely it will be a zombie this year as well. In this case having an odd ratio of 3.9 signifies that it is almost four times more likely for a firm to be zombie if it was a zombie in the previous year, compared to the firm not being a zombie in the previous year, all else equal. The fixed effect structure in this regression ensures that we are not capturing time-invariant firm
27 This is also what we observe in reality. In 2011 and 2012, Greek 10-year bond yields sat around 15% and 20%.
In this period GDP growth rates in crisis were the lowest in a decade, respectively -8% and -7%. See Figure A1.
28 Most of the variation in sovereign yields is observed during the crisis. So for the period 2009-2015. Before that,
24 characteristics in the lagged zombie variable. So here I am able to disentangle unobserved heterogeneity for the dynamic variable.
In the last column, the firm size dummy variable enters the regression. Firms here are divided into size groups of employees. The signs are significantly positive and they compare to the base rate of a firm that has more than 250 employees, so the largest firms. The odds ratio diminishes in value as the number of employees increases. In contrast to what we observed in the data section, this indicates that as the firm expands, the likelihood of it becoming a zombie goes down. The firm fixed effect structure is mainly responsible for this contrasting effect as it captures the most of the firms characteristics like size. Only when a firm switches from size qualification (i.e. time variant observations), then this effect is captured by the control variable. Only expanding firms are therefore captured by this variable, and these firms are relatively productive firms since they have the capacity to expand. Due to missing values for employees, I lose about 7 million observations. This does not influence the significance of the regression by any substantial amount. I therefore in future regressions do not include employees groups as a size proxy, since most of the effect is captured in the fixed effect structure.
Including control variables in the regression does not seem to change the sign of the odds ratio of the policy rates, nor does it change the significance. It does however, change the magnitude of the coefficient. In the first two columns, it is likely that the interest rate variable is overestimated because of omitted variables. As GDP and the lagged zombie variable are added, the odds ratio comes closer to one, but still highly significant. This indicates that loose monetary policy might have an adverse effect on zombification.
25 Robustness of results
In table A6 the distribution is given of my data per country. As one can see in the table, Italy and Spain are responsible for the majority of my observations. Out of around 17 million observations in total, Italy is responsible for around 11 million and Spain for around 5 million. To check whether my regressions are not mainly driven by these countries, I divided the sample in different country groups and different time windows. The latter controls for the strong intensification of zombie firms during and after the crisis.29
The first two columns are regressions of the subsample pre-crisis (2000-2009) and post-crisis (2010-2017).30 The coefficients of the policy interest rate does not seem to change, so the relationship remains. The magnitude of these numbers however, vary greatly. In the pre-crisis sample the odds ratio is around 0.0002 while in the post-crisis it is 0.754. A reason for this could be the lower variability of the interest rates during this period. If zombification increases in that period and the policy rate show smaller movements, then a large part of the deviation is attributed to the policy rate. This also observed in the coefficient of the government bonds, where an increase of 1% in the yields would translate to a 6 fold increases in the likelihood of becoming a zombie. Estimators in these samples should therefore be interpreted with caution, since it is difficult to determine the actual magnitude of the effect. It is interesting to note that both government bonds and GDP have now odd ratios larger than 1. This suggests that economic growth and government bond yields are positively related to zombification. The former might be due to an increasing trend of zombie firms that has been occurring since the 80’s (Banerjee & Hofmann, 2018) which in the pre-crisis sample might be attributed to economic growth in my regressions.
The next four columns in Table 2 present the results of the subsamples of different countries. For Spain and Italy, due to the large sample sizes of these countries, I ran separate regressions. The fifth column excluded Spain and Italy. Greece is excluded in the last regression. 31 For these columns the coefficients continue to indicate the same relationship between the policy rate and zombification, since the coefficient does not change signs or becomes greater than one. Also, differences in magnitude seem to be less sensitive to changes in the sample selection of countries relatively to the year selection in the first two columns.
29 Andrews & Petroulakis (2019) realized that circumstances changed dramatically during the global and euro
crisis, so in their analysis they also divided their sample these two categories (pre-crisis, post-crisis).
30 This is the starting year of the intensification of zombification in my sample (see Figure 3 panel A).
31 Since Greece is under heavy surveillance of the European Commission since the euro crisis, they might be an
Table 2
Conditional Logistic Regression of subsamples
The dependent variable is the binary indicator zombie. This variable takes the value of 1 when a firm has a debt servicing capacity ratio of below 6% and either negative investments or negative return on assets. Both requirements have to be met for three consecutive years. Policy interest rate is the main refinancing rate as set by the ECB. GDP is the Gross Domestic Product growth rate in comparison to the previous year at constant prices of total consumption. Government Bond is the yield of a 10-year government bond of a country in a specific year. The policy interest rate, GDP
and Government Bond variables all enter the regression with a 2-year lag. Zombiet-1 is a zombie dummy describing if the firm was identified as a zombie in the previous period. Robust standard
errors are clustered by firm (parenthesis). Sample of 2000-2017. Countries included are PT, IT, ES, SI, GR and IE. Significance levels are denoted respectively *** for 1%, ** for 5% and * for 10%. Coefficients are given as odds ratios.
(1) (2) (3) (4) (5) (6)
Pre-crisis (<2010) Post-crisis(>2009) Spain Italy Without Spain &
Italy Without Greece
Panel A: Conditional Maximum Likelihood Estimation (CMLE) with firm fixed effects
Policy interest rate 0.0002*** 0.754*** 0.827*** 0.654*** 0.314*** 0.407***
(0.000023) (0.0048) (0.0049) (0.002) (0.047) (0.0058) GDP 1.12*** 1.170*** 0.679*** 0.093*** 0.947*** 0.919*** (0.0197) (0.0072) (0.0021) (0.0035) (0.015) (0.0041) Government Bond 6.77*** 1.12*** 0.714*** 1.346*** 1.009 0.972*** (1.368) (0.0079) (0.0040) (0.0065) (0.013) (0.0067) Zombiet-1 1.10*** 3.122*** 4.234*** 5.061*** 2.46*** 3.90*** (0.0118) (0.012) (0.025) (0.023) (0.057) (0.015)
Year fixed effects YES YES NO NO YES YES
R2 0.18 0.12 0.26 0.20 0.21 0.26
Observations 296,185 1,110,814 856,583 1,187,736 73,960 2,139,131
Table A7 presents the results of the regressions using different definitions of zombie firms. As explained in the data section, it is quite difficult to define a zombie and there are multiple way to go about it. In panel A, we use the same definition as in the baseline results. Only I take more conservative measurements in these columns. I use definitions where firms show financial weaknesses for four consecutive years instead of 3. Also, a combination of both negative profits and negative investments instead of either one of the two. In the last column firms are considered a zombie when they have a debt servicing capacity of less than 5 percent, where in the baseline regressions it was 6 percent.
Panel B uses the IC definition as used in Adalet McGowan et al. (2018). The results also seem robust to more conservative versions of this definitions like an IC less than 0.5 for 4 consecutive years. The odds ratio of the policy interest rate is substantially different from Panel A with an average of around 0.75. This means that if the interest rate increases with 1% the probability of a firm becoming a zombie is reduced by 25%. While this percentage was around 45% in Panel A. This demonstrates that issue as discussed in the data section exists when using the IC ratio for identifying zombies. If the policy rate goes down, then the bank lending rate is likely to go down, which in turn increases the IC ratio of a firm. With the debt servicing capacity definition, the amount of interest paid by a firm is out of the equation. This explains the stronger effect of the policy interest rate in Panel A,
In addition, I use the definition for a zombie firm used in Caballero et al. (2008) of a firm that receives subsidized credit. These results are presented in Panel C. This definition exploits specifically the channel of bank forbearance. All coefficients are significant and negative (odds ratio <1), which reinforces the hypothesis that lower policy rates would lead to an increase of bank forbearance. Only Spain and Italy were used in the sample, due to data availability. Important to note is that Panel C does not observe the effect of bank forbearance directly since we do not have bank-firm relationship data. So coefficients are interpreted by implication indirectly.
Causality
28 for omitted variables due to the time and individual effects added in the regression. Multiple identification strategies like the interest coverage ratio and the subsidized credit strategy of Caballero et al (2008) controlled partially for the measurement error that can arise. Aside for the omitted variable bias and the measurement error, endogeneity can also exist when there is a reversed causal relationship between the dependent variable and the explanatory variable. In order to investigate this relationship, I reduce my panel data to the country level and use the capital share of zombie firms per country per year as the dependent variable. All the explanatory variables remain the same since they were already at a country level or equal across all countries in the case of the policy rate.
Table A8 shows the same regression as the baseline, but now the zombie capital share per country is the dependent variable. The results seem similar as in the conditional logit model. The policy interest rate has a negative significant sign in both regressions, respectively -1.2 and -0.94. The economic interpretation of this is that a one percent increase in the policy rate leads to a 0.94-1.2% reduction in the amount of capital sunk in zombie firms. It is interesting to note that here GDP and government bonds are not significantly different from zero.
An argument against these results could be that zombie firms in itself might trigger economic instability and therefore the ECB would have to respond by lowering the interest rate. This can be a reverse causal relationship. Performing a Granger causality test helps determining this causal effect. The capital share of zombie firms seems to appear stationary in my sample.32 Performing a Granger causality test on the dependent variable zombie capital share and the
policy rate however, shows that with 1-4 lags I cannot reject the null hypothesis that the policy rate does not Granger-cause the zombie variable (see table A9). In contrast, using a two year
lag of the zombie variable for explaining the policy rate rejects the null-hypothesis at a 1% level. This implicates that the regression suffers from endogeneity due to reversed causality. Because of this potential reverse causality, it is difficult to infer any causal relationship for the policy rate to potential zombification.
A more robust and informative way to look at the relationship between the policy rate and zombification is to look at differences across sectors. I do so by assessing if lower policy rates have a stronger effect in sectors where firms are more dependent on external finance. In the paper of Banerjee & Hofmann (2018) a similar test was performed. If the charge off rates
32 Performing a Granger-causality test requires the variables to stationary, otherwise it will only produce spurious
29 of banks increase as the interest rates go up (Borio et al., 2017), then this effect should be present especially in these sectors. By analyzing on a sectorial level with an interaction term, I am better able to address issues of reverse causality and omitted variables, since I can control for unobserved macroeconomic factors at the country level in any year.
Table A10 present the results of the panel regression of zombie capital shares per sector in each country in year t.33 Panel A is a random effects regression with year fixed effects. The policy rate and external finance dependence interaction term enters the regression with a negative significant sign. This shows that the policy rate pushed up the zombie share in those sectors which have a relatively high dependence on external finance. Adding control variables do not seem to change the sign of the coefficient nor its significance. Interestingly, policy rates seem to push down zombie shares in those sectors where external finance dependence is relatively low due to its positive sign without the interaction term.
Panel B present the alternative fixed effects model, this model captures unobserved time invariant characteristics of each sector in any country. The year*country fixed effects controls for time-variant effects in the year-country cell. Control variables like the GDP growth rate and government bond yields are insignificant in these regressions. Partially this is due because their impact is already captured in the fixed effects structure. Similar to panel A, the coefficient of the interaction term is around -7.2 and significant at the 1% level. This means that an increase of one percentage point in the policy interest rate results in a 7.2 percentage point reduction of the capital zombie share.
Discussion and Policy Implications
From the results as the discussed above, it seems that there is a causal relationship between the main refinancing rate of the ECB and the share of capital sunk in zombie firms in these 6 euro periphery countries. If we assume that the coefficient of -7.2 for the interaction term is accurate, the total change of 4% in the main refinancing rate from 2000 until 2017, would account for 17.5% of the rise in zombie shares in these 6 euro periphery countries if evaluated at the mean sector level of external finance dependence. As explained in the introduction, an increased zombie capital share has shown to suppress total factor productivity and may weaken economic performance. (Adalet McGowan, Andrews, & Millot, 2018) (Banerjee & Hofmann, 2018).
33 Note that if I regress the two year lag policy rate as dependent variable against the interaction term of zombie
30 On the other hand, Figure 3 shows us that zombification has been decreasing since 2015. One could attribute this fall in zombie shares to the rise in earnings before interest and taxes that started around 2014 as Figure A2 shows. Since in my identification strategy zombie firms are partially defined by their EBIT to total debt ratio, this would mean that an economic upturn or increase in GDP would, for the larger part, be responsible for this reversal of zombification.
The next question one could ask then is to which extent monetary policy has an effect on economic performance and would then therefore be indirectly responsible for this fall in zombie shares. According to Banerjee & Hofmann (2018) lower policy rates tend to boost aggregate demand and increase employment and investment in the short run. However, lower rates also can cause misallocation of resources and lower productivity growth due to the prevalence of zombies. If the latter effect is strong enough, this could have detrimental effects on economic growth, pressing down interest rates even further. This eventually could lead to a downward spiral. Where policy rates have to be lowered to compensate for the lack of productivity growth, which in turn even further slows down productivity growth.
Haldane (2017) argues that for monetary policy there seems to be a tradeoff between on the one hand maintaining stability in the economic system by lowering policy rates which eases the interest burden that firms face, and on the other hand the lowered productivity as a result of these low rates due to possible zombie prevalence. In his hypothetical simulation for the United Kingdom, he shows that a potential increase of 4% in the interest rate may boost productivity with 2-3 percent and would come with a significant macro-economic cost, since around 10% of the companies would go bankrupt.
31
6. Conclusion
This paper investigated the potential effect that monetary policy has on zombification in the euro periphery area. First, I find that on a firm level an increase in the policy rate of the ECB leads to a reduction in the likelihood of a firm becoming a zombie. These results were robust to different subsamples for separate countries and years. However, using subsamples sometimes did change the coefficient by a substantial amount. Therefore this coefficients should be interpreted with caution. Also using different identification strategies of zombie firms did not change the results significantly. Even though I controlled for a large part for omitted variables and measurement errors, reverse causality made it difficult to infer a causal relationship.
In order to address reverse causality issues, I constructed a country level panel. Here, the dependent variable was the share of capital sunk in zombie firms for a country in a specific year. Similar results are observed as in the firm level regression. An increase in the policy rate pushes down the zombie share. However, reverse causality seemed to be an issue in this regression. In an attempt to control for this, I constructed a sector level panel. Zombie shares were calculated on a sector level in a country for a specific year. By adding the interaction term of external financial dependence the reversed causation was controlled for and the negative relationship between the policy rate and zombie shares remained.
32 Future research could dig further into the causal relationship of monetary policy and zombification, especially on a firm level, since it is hard to control for reverse causation. Especially with the incredibly rich dataset of ORBIS, it opens many doors for further firm level investigation and identification of zombie firms.
In addition, further research should explore individual bank-firm relationships with respect to monetary policy. Past research has had a main focus on weak banks and zombie firms when investigating this bank-firm relationship. It would be interesting to observe lending behavior over time when policy rates change. This is mainly what Acharya et al., (2019) already did in his paper, but only with listed firms. Also, this would allow to further explore the bank forbearance channel.
In my paper, the effects of unconventional monetary policy has not been studied yet, and previous literature addressing the relationship with zombie firms is relatively slim. So in order to construct policy implications for the ECB, more research in this area is needed as well.
