Msc. Entrepreneurship joint degree Master UvA and VU Melanie Höhle
Stud.ID. UvA 11764104 Stud. ID VU 2625499
Supervisor: Rafael Perez Ribas Application: 1st April 2018 Submission: 27th June 2018
The impact of capital structure on start-up success: A
cross-country study
1
Statement of originality
This document is written by Melanie Höhle who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgement
I would first like to thank my thesis advisor Rafael Perez Ribas of the Faculty of Economics and Business at University of Amsterdam. I received valuable advice whenever I needed support for my research or writing. He consistently provided me feedback which gave me the opportunity to increase the impact of my study and push myself to learn new methods.
Also, I like to express my profound gratitude to my family and my friends for providing me with unfailing support and continuous encouragement throughout my study and through the process of researching and writing this thesis.
2
Abstract
My research investigates how the capital structure of a start-up is related to its growth, productivity and survivorship. This relationship varies across the countries because of external factors such as economic conditions, access to debt and different laws. I first use a cluster method to assess the sensitivity of the start-ups, regarding leverage and liquidity according to their origin country. This enables me matching start-ups from countries with similarities in leverage and liquidity in a cluster. A linear regression model evaluates the relationship of size, liquidity, short-term leverage, total leverage and tangibility with start-up success. I use the data from Orbis on start-ups in the information sector. Based on this a further analysis with the Oaxaca-Blinder Decomposition on the recentered influence function examines the differences on the 10th, 50th and 90th quantile distribution. This method has the advantage to provide detailed
decomposition results with a simple regression-based model. Therefore, the decomposition enables me to evaluate how debt is rewarded in different countries. I assess the expected change of certain countries, assuming other countries predictor levels. My results indicate that short-term leverage is mainly negatively related to growth and survival but mostly positive towards productivity. The total leverage is negatively associated with productivity and mostly with survival but positively towards growth. Hence, short-term debt helps start-ups to bridge cash demanded productivity. Whereas too much debt, represented as total debt, can harm the start-up productivity and survival. The decomposition shows that success does not depend on capital structure as much as it does on the economic condition. While in some countries the economic condition fosters less growing and less productive start-ups, other countries provide a better fit for ‘high flyers’.
Keywords:
Capital structure, Start-ups, Productivity, Firm growth, Oaxaca-Blinder Decomposition, Quantile Decomposition
3
Contents
Statement of originality ... 1 Acknowledgement ... 1 Abstract ... 2 1 Introduction ... 4 2 Literature review ... 7 3 Data ... 123.1 Source and sampling strategy ... 12
3.2 Variables ... 13
4 Empirical model ... 14
4.1 Linear regression model ... 14
4.2 Choice of cluster... 15 4.3 Decomposition ... 15 5 Descriptive statistics ... 18 6 Empirical findings ... 22 6.1 Regression... 22 6.1.1 Growth ... 24 6.1.2 Productivity ... 24 6.1.3 Survivorship ... 25 6.2 Decomposition ... 26 6.2.1 Growth ... 28 6.2.2 Productivity ... 29 6.2.3 Survivorship ... 30 7 Discussion ... 31 8 Conclusions ... 33 9 References ... 35 10 Appendices ... 38 10.1 Cluster method ... 38
4
1 Introduction
The choice of capital is an interesting topic for start-ups because of its many advantages and disadvantages. To fulfil an idea as a start-up cash is needed and therefore it decides either for debt or equity or a mixture of both (Smith, Smith and Bliss, 2011). Start-ups aim to grow fast which raises the question if equity is preferable to do so. In productivity perspective debt can be useful to relieve cash concerns. If once solved, the debt must be paid back. On the one hand this can be harmful in long-term perspective because its pressures the cash-flow. On the other hand, it can even foster the business to evolve quickly and become even more productive. Further, debt solves issues such as the shareholder-manager conflict. Whereas it also increases the risk for the shareholder. The debate about the choice of capital structure of a company is already a long-term topic but still currently relevant.
My study focuses on growth, productivity and survival as success factors as those contribute to overall economic benefits, such as GDP, rather than shareholder’s profits. The countries differ in external factors such as economic conditions, access to debt and laws. Therefore, I evaluate the initial choice of capital on the success of the start-up. My research answers the question what kind of capital structure the best choice for a start-up regarding its future success is. My hypothesis is that a lower leverage ratio leads to higher start-up success.
To test this hypothesis, I use a linear regression model to analyse the impact of the initial leverage, tangibility, liquidity and size on growth, productivity and survival of the start-up. For doing so, the countries, the start-ups are in, are clustered according to their sensitivity in liquidity and leverage. In addition to this analysis, I examine the differences in growth, productivity and survivorship gap. Therefore, I use the recentered influence function by Firpo, Fortin, & Lemieux (2009) in combination with the Oaxaca-Blinder decomposition by Oaxaca (1973). This decomposition provides the advantage to assess the impact on defined unconditional quantiles and allows me to evaluate how debt is rewarded in different countries. I assess the defined unconditional quantiles of 10th, 50th and 90th which describe the differences
for high, medium and low growth and productive start-ups. As the decomposition is threefold, it enables me to assess the endowment, coefficients and interaction. Hence, I evaluate the expected change of certain countries, when assuming other countries predictor levels. Coefficients and interaction of the decomposition provide further detailed results for the assessment. My research design is deductive as the topic is also addressed in previous research
5 and needs clarification and validation. The quantitative approach is used due to the mature research topic.
Over the time several factors have been evaluated in the choice of capital perspective and target leverage ratios have been discussed (Hovakimian, Opler, & Titman, 2001). The pecking order theory and the trade-off theory influences the leverage ratio companies aim to. Also, the macroeconomic environment and business cycle impacts the choice of capital (Hackbarth, Miao, & Morellec, 2006). Firm specific fundamentals such as tangibility, profits, firm size, market-to-book asset ratio are further reasons (Frank & Goyal, 2009). But also personal characteristics of the founder affect the firm’s success (Cooper, Gimeno-Gascon, & Woo, 1994).
My approach differs from previous studies because I use of the Oaxaca-Blinder Decomposition in combination with the recentered influence function to examine the impact of capital choice on defined unconditional quantiles. I use new data generated from Obis database with a cross-country approach for European start-ups. In contrast to many other studies I focus on the revenue, survival and productivity instead of the generated income. Furthermore, I use separate measurements of leverage such as short-term and total liabilities to differentiate the financial needs.
I found that short-term leverage is negatively related to growth in Spain, France, Italy, Portugal, Sweden, Estonia Czech Republic, Germany, Finland, United Kingdom, Russia and Norway, whereas it is positively related to Poland and Ukraine. But the total leverage impacts Ukrainian and Polish start-up growth negatively and in all other countries positively. Furthermore, short-term debt supports productivity positively in Czech Republic, Germany, Finland, United Kingdom, Russia, Norway, Poland and Ukraine. Whereas the total leverage has a negative influence on productivity for all countries. Regarding start-up survival short-term debt helps Ukrainian and Polish start-ups but does not help start-ups in all other countries. The total leverage ratio impacts Ukraine, Poland, Spain, France, Italy, Portugal, Sweden and Estonia negatively regarding survival but Czech Republic, Germany, Finland, United Kingdom, Russia and Norway positively. According to the decomposition results the differences in growth, productivity and survivorship are driven by unexplained factors. Thus, the economic environment seems to be the best explanations for the differences in the countries and not their capital decisions.
6 Furthermore, I test other related variables such as size, liquidity and tangibility on start-up success. I found evidence that the larger the start-up, the less it grows and less productive but the mainly more survival. Increased liquidity supports start-up growth. More cash decreases survival and productivity in Ukraine and Poland. Whereas in all other countries this fosters productivity and survival. Tangible assets help ups grow but negatively impact the start-up productivity as those tie start-up cash. In the case of survival tangibles are beneficial for start-start-ups in Spain, France, Italy, Portugal, Sweden, Estonia Czech Republic, Germany, Finland, United Kingdom, Russia and Norway but a disadvantage for start-ups in Ukraine and Poland. When focusing on the different quantiles start-ups from Czech Republic, Germany, Finland, United Kingdom, Russia and Norway perform better in low growth quantile and Poland and Ukraine start-ups perform better in median and high growth quantile. In terms of productivity start-ups from Spain, France, Italy, Portugal, Sweden and Estonia have an advantage. Further, Polish and Ukrainian start-ups are the most likely ones to survive.
According to Barclay, Morellec, & Smith (2001) taking on debt has a negative impact on growth if there is no debt capacity given through assets. Also, Lang, Ofek, & Stulz (1995) conclude that leverage has a negative influence on growth. Robb & Robinson (2014) state that liquidity matters for the company’s success. My study confirms this as liquidity is mainly positively related to growth, productivity and survival. Furthermore, my decomposition results verify that other factors such as the environment and company specifics drive the success (Cooper et al., 1994).
The following thesis first provides a literature overview regarding capital structure choices. Afterwards, the data, empirical model and the descriptive statistics follows. Results and discussions present the takeaways from the study. Finally, the conclusion reports the key message of the paper as well as its limitations.
7
2 Literature review
The discussion about equity and debt evolved in 1958 when Modigliani & Miller (1958) state that under market efficiency the market value of a firm and the cost of capital is independent from the capital structure. Hence, the leverage is not influencing the firm’s value or the costs. This was revised in 1963, when Modigliani & Miller (1963) claim that under real world conditions the leverage affects the return and market value.
Start-ups often face the decision which means of capital to use - debt or equity to finance the business case on a given condition of the available capital and credit supply. Equity has the advantage that the risk is shared between the shareholders, whereas also the earnings are split between all of them (Smith, Smith and Bliss, 2011). Debt has the benefit that the earnings generated through the start-up will belong to the owner but also the whole entrepreneurial risk associated with the start-up must be borne by himself. The venture capitalist diversifies the risk through several investments (Smith, Smith and Bliss, 2011). Also, a venture capitalist or a business angel provides the founder with advice and guidance and might interfere in business decision. If the founder uses debt, he controls his own future decisions (Smith, Smith and Bliss, 2011).
There are several further reasons to consider for the capital decision. Taking on debt depends on the stage the start-up is in. For the seed stage, equity might be favoured while if the start-up is in the late stage, the debt finance can be favoured (Smith, Smith and Bliss, 2011). The reason for this is that the start-up has already build up track records of credibility to convince debt providers. But debt can be still a good choice in the early stage, if there are enough assets to secure the debt. Furthermore, it depends how urgent the financial means are needed as some debt forms take more time such as governmental subsidised debt (Smith, Smith and Bliss, 2011). Harris & Raviv (1991) argue that leverage increases the owner’s share by two ways: the founder’s return becomes more sensitive to firm performance and as debt is non-voting for management decisions, it concentrates the power to the founder.
Also, the assessment of banks can be harmful towards start-ups when the business is screened for a debt application (Bustamante & Acunto, 2018). This may result in a potential shock towards the fundamental business model as potential lenders impact business decisions. Bustamante & Acunto (2018) argue that the initial capital structure influences future
8 performance. Their panel study of US ups from the Kauffmann survey suggests that start-ups with a lower initial leverage perform better in the future.
In general, firms decide for debt, when expecting a positive future cash flow from an idea or project. But the trade-off between taking the debt and the cost of debt needs to be faced. For the long-term perspective the initial capital structure of a company is important because it remains stable over a long time period (Lemmon & Zender, 2008). Hence, the initial leverage of a company is often constant. Firms tend to adjust their leverage infrequently and not according to their possibilities and dynamics in the economy. (Strebulaev, 2006).
According to Graham & Leary (2011) taking on debt has the advantage of tax benefits and fixed interest payment. But taking debt has the disadvantage of bankruptcy cost, which arise with higher probability of facing bankruptcy when using debt instead of equity (Graham & Leary, 2011). Hence, the price of the debt realising a certain project is crucial for this decision. Hovakimian et al. (2001) stress the trade-off theory between the benefits of debt and its cost, as they claim there is a target leverage ratio, which the company aims to. So, on the one hand if the company has the choice between debt and equity, it will move towards the target leverage ratio. On the other hand, Hennessy & Whited (2005) claim there exist no target leverage ratio and Hovakimian et al. (2001) soften the argument through the fact the trade-off theory is more static. Due to that firms tend to follow also the pecking order theory.
The pecking order theory introduced by Myers & Majluf (1984) states that debt is a preferred choice of capital to equity due to information asymmetry. The argumentation is that the founder has an information advantage over potential investors and thus overvalues the company. Therefore, the investor offers a lower value for a higher stake in the company. So, for this case debt can be used to mitigate the risk of asymmetric information as it has the advantage to provide a fixed interest rate and repayment. This can raise the lenders trust (Harris & Raviv, 1991). Frank & Goyal (2003) oppose the pecking order theory claiming that equity becomes more and more important especially for small companies. Their findings show that the pecking order theory is only valid for large corporates. Lemmon & Zender (2010) oppose as they argue that debt is preferred over equity especially for small high growth firms. In their point of view the debt capacity has to be taken into consideration to present the preferred choice of capital.
9 Since the perfect market environment does not exist, start-ups face agency costs which can also be mitigated by the use of debt (Miglo, 2016). Agency cost arise due to a shareholder versus manager conflict. If the manager earns only a small part of the generated profit, the manager’s intention can be less focused on value-maximization. Taking debt can also signal the founder’s commitment and firm’s quality to the lender (Miglo, 2016). Also because the founder has to provide a personal guarantee. Frank & Goyal (2009) argue that more diversified businesses have a lower default risk and mature firms with an established reputation face less agency cost. This implies that start-ups need especially profitable ideas and projects to offset those expenses.
Additionally, the access to debt market of the country is important for the capital structure decision (Frank & Goyal, 2009). Having a rating increases the access to market debt (Faulkender & Petersen, 2006). Unfortunately, for start-ups this mainly not the case, as market debt is mainly for mature companies. But this might change in the future as the start-up, ‘Early metrics’, is providing ratings for start-ups and scale ups (Early metrics, 2018).
Furthermore, if there is a credit rationing, which means that the debt lenders limit their lending volume, start-ups favour equity (Miglo, 2016). Also, volatile cash flows of the start-up influence the access to debt (Frank & Goyal, 2009). This indicates that very volatile cash flows mean less access to debt. Volatile cash flows can have macroeconomic reasons or are due to the business cycle phase the company is in (Frank & Goyal, 2009).
According to Hackbarth, et. al. (2006) firms should adjust their policy according to macroeconomic environment as well as the business cycle phase they are in. So macroeconomic conditions in the country matter because of the expected inflation (Frank & Goyal, 2009). If the inflation increases, more companies take on debt because they need to pay back less value with high inflation over time. Thus,. Korajczyk & Levy (2003) especially stress the argument that macroeconomic conditions affect the choice of capital. They present differences between financially constrained and unconstrained firms. Unconstrained firms choose a preferred macroeconomic timing to issue debt capital whereas constrained ones do not as they are taking on any money (Korajczyk & Levy, 2003). But not only the choice of capital is influenced, also the firm performance is subject to macroeconomic conditions (Cooper, 1993).
10 So, leverage needs to get adjusted for the company’s pursued strategy (Frank & Goyal, 2009). This implies for the initial capital structure, that the leverage should be chosen according to the expected growth in the strategy (Hennessy & Whited, 2005). The capital is closely linked to the start-up strategy because it supports indirectly through better training and planning as well as it shows credibility towards venture capitalist and venture lenders (Cooper et al., 1994). Furthermore, capital impacts directly by financing ambitious strategies, possible changes of direction, rapid growth and provides time if needed (Cooper et al., 1994). For example, the initial capital is used to buffer shocks in capital such as production delays and to finance intensive growth strategies. Therefore, the initial financial capital influences the performance measures failure, marginal survival and growth (Cooper et al., 1994). Lang et al. (1995) emphasize the negative relation between leverage and growth. Leverage does not reduce growth opportunities if firms have recognized growth opportunities. Further, they argue that leverage can even proxy growth opportunities. This shows that high leverage signals high growth opportunities and low leverage signals low growth opportunities (Lang et al., 1995). They further state that leverage is even more relevant to proxy growth than liquidity through operating cash flow. Barclay et al. (2001) confirm as well that debt is negatively associated with growth. They argue that this relationship is valid under the condition of having growth opportunities and an available level of debt capacity. Hovakimian, Hovakimian, & Tehranian (2004) empirically test the relationship of debt and growth opportunities. They found that high market-to-book companies have better growth opportunities. Thus, those firms aim for low debt ratios and hence for equity finance.
Moreover, Frank & Goyal, 2009 argue that the main factors determine the capital structure are industry median leverage, tangibility, profits, firm size and market-to-book asset ratio. They stress the industry median leverage as some industries take on more debt than others. This is because some industries are more capital intensive such as the mining industry and thus a higher funding amount is needed whereas other areas demand less capital such as the information sector. Another point of view is provided by MacKay & Phillips (2005) who argue that the industry fixed effects influences the financial structure of a firm less than the firm fixed effects. But still they also state that there are industry related factors which explain the intra-industry variation. Thus, it can be concluded that industry is a driver, but firm’s individual characteristics impact the financial structure even more. According to Frank & Goyal (2009) tangibility is a driver for leverage, as companies with more fixed assets, can use them as collateral. They argue
11 besides that the firm size is a reason for the choice of capital, as larger companies, regarding total assets, can afford a higher leverage ratio.
Furthermore, there are unobservable factors influencing debt decision such as transaction costs (Titman & Wessels, 1988). Also, Rajan & Zingales (1995) found many unexplained factors when investigate capital structure decisions of the G7 countries. They found the different choices in capital structure are due to different laws in bankruptcy filings. The reason for this is that the riskiness of taking debt versus equity differentiates in the G7 countries (Rajan & Zingales, 1995).
According to Miglo (2016) taking too much debt in the need of financial flexibility can lead to financial constraints. He argues that developing start-ups have a high demand for financial flexibility to pursue their business idea. Hence, a debt overhang can occur by taking on too much debt which can turn the start-up into a financially difficult situation. Also, the debt reduces the free available cash, because the debt needs to be paid back soon, which worsens the situation (Harris & Raviv, 1991). According to Robb & Robinson (2014) up to 80-90% of the entrepreneurs rely on owner equity and bank debt, as there is a need for liquidity. Even personal credit cards are taken to reach a level of approximately 40% of leverage according to US start-ups from the Kauffmann survey (Robb & Robinson, 2014).
Furthermore, Cooper, et al. (1994) explain that also human capital influence the performance measures failure, marginal survival and growth. If the founder has a clear understanding of his resources and its usability, it will add to survive and grow. But also, individual factors such as the level of education and entrepreneurial parents are impacting the survival and growth. Further, Cooper et al. (1991) found that being male and having a partner is positive related to growth. Besides, Cooper (1993) argues that the personal character of the entrepreneur, the founding process and non-economic goals of the founder impacts the future success. Thus, if the founder has a risk averse character who himself also does not want to use debt in general, then the likelihood that he would use debt to finance his company decreases.
12
3 Data
3.1 Source and sampling strategy
The data contains start-ups founded between 2008 and 2011 in the information sector. This information is collected from Orbis Database. Each start-up is observed for five years. Year zero is the year of incorporation. Since the financial reporting is established for most companies in year one, rather than year zero, the first year is used to express the initial financial year.
The data is limited to the information sector according to north American industry classification (NAICS) 51. According to United States Department of Labor (2018) NAICS 51 includes the publishing industries, broadcasting, web search portals, data processing, motion pictures, sound recording and telecommunications. For the further analysis I only split into two subsectors; telecommunication and non-telecommunications.
The benefit of selecting the information sector is that it is a less cost intensive sector and hence provides an easier founding field than for example businesses with high entry barriers such as the mining industry. Another filter is used to exclude branches and subsidiaries that are part of corporate hierarchies. Therefore, I define the start-ups as ultimate owner (>50,01%). I also filter for start-ups with filled liabilities in year one, which provides the advantage of results with available data.
The data was collected for European countries where more than 500 start-ups are available to inspect, such as Portugal, Spain, Russia, United Kingdom, Sweden, France, Germany, Estonia, Italy, Poland, Czech Republic, Ukraine, Finland and Norway. The number of total start-ups observed is 20,545. The data was collected in March and April 2018. Table 1 provides an overview of the sample data.
13 Table 1: Number of start-ups per country
This table shows the number of start-ups for each researched country. Missing indicates the worst-case scenario. Meaning the less available variable.
Country Number of start-ups Missing in (%)
Portugal 970 396 41% Spain 1,500 504 34% Russia 1,141 759 67% United Kingdom 2,513 1987 79% Sweden 1,496 619 41% France 3,381 1,380 41% Germany 630 546 87% Estonia 605 300 50% Italy 5,165 1,181 23% Poland 716 378 53% Czech Republic 474 249 53% Ukraine 523 226 43% Finland 646 334 52% Norway 785 328 42% Total 20,545 100% 3.2 Variables
The success of a start-up can be explained as growth, productivity and survival.
Growth is measured as the difference between the logarithm of operating revenue in year five and in year one.
Productivity is measured by the logarithm operating revenue in year five divided by the fixed assets of year four. The fourth year is used as the company generates its revenue with the capital from the previous year. Productivity values below zero are excluded from the sample.
Survivorship is a dummy variable that is one if the start-up survived and zero otherwise. The start-up is indicated as closed, if it is bankrupt or dissolved. But if the company was taken over or merged it is flagged as survived because this is not considered as failure.
The explanatory variables in my model are total leverage, short-term leverage, tangibility, liquidity and size.
Total leverage is measured by total liabilities (current and non-current) divided by total assets both in year one.
Short-term leverage is expressed by current liabilities divided by total assets in year one.
Tangibility is indicated by the ratio of fixed assets divided by total assets in year one. Liquidity measures cash & cash equivalent divided by total assets in year one.
14 Size is expressed by the logarithm of total assets in year one. Size values below zero are excluded. The number of employees was not an appropriate measure for productivity as well as for size since there are 57% missing values in the described dataset.
Sector is a dummy that defines if the start-up is operating in the telecommunication subsector or not
For the further analysis I winsorize the variables, which means ordering the non-missing values and generating a new variable, which replaces the highest value. The measure used is 1%.
4 Empirical model
4.1 Linear regression modelMy research consists of three linear regression models that aim to evaluate the impact of the initial short-term leverage, total leverage, tangibility, liquidity and size on growth, productivity and survival. The regressions are run for each cluster, which are described in Section 4.2.
Model assumes that Growth, Productivity and Survivorship is:
Y, = α + 𝛽 ∗ 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒, + 𝛽 ∗ 𝑆ℎ𝑜𝑟𝑡 − 𝑡𝑒𝑟𝑚 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒, +𝛽 ∗ 𝑇𝑎𝑛𝑔𝑖𝑏𝑖𝑙𝑖𝑡𝑦, + 𝛽 ∗ 𝐿𝑖𝑞𝑢𝑖𝑑𝑖𝑡𝑦, + 𝛽 ∗ 𝑆𝑖𝑧𝑒, + 𝜇 + 𝜇 + 𝜖,
(1)
where Y represents growth, productivity and survivorship. The i indicates the start-up, µt the
specific effect of the year of incorporation, µs the specific effect of the telecommunications
sector and 𝜖 the error term. The year of incorporation is included to control for differences that can occur due to the time span of founding date between 2008 and 2011. Further, the sector variable is included to control if the start-up is in the telecommunications sector. Start-ups in this sector differentiate regarding their business model in comparison to the other sectors motion pictures, broadcasting and data process.
15 4.2 Choice of cluster
To evaluate the countries efficiently, clustering is a method to describe differences and similarities in the data (Everitt, Landau, Leese, & Stahl, 2011) . Hence, start-ups from countries that have similarities are grouped together. For the process of clustering the 14 countries, the standardized regression coefficients of leverage and liquidity for all three-models growth, productivity and survivorship are used per country (see Appendix Table 9). Those coefficients indicate the dissimilarity/similarity matrix.
I use the k-median approach which minimizes the error terms and reduces the distance while part n observations into k cluster using a dissimilarity matrix (see Appendix cluster method) (Everitt et al., 2011). The stopping rule of Calinski-Harabasz pseudo f index indicates which cluster choice is the best (see Appendix Table 7). For this dissimilarity matrix three cluster is the best choice.
Thereafter, I tested this suggestion with the hierarchical approach wardslinkage, which uses a similarity matrix to minimize the loss of each cluster while forming possible clusters (Ward, 1963). The stopping rule Duda/Hart indicates how many clusters are the best choice (Appendix Table 8). I ordered the countries according to the wardslinkage in each cluster, which is presented in Table 2 (see Appendix Table 9).
Table 2: Cluster
This table provides an overview of each cluster according to the countries and number of start-ups in the cluster.
Cluster Countries Number of start-ups
C1 Spain, France, Italy, Portugal, Sweden, Estonia 13,117
C2 Czech Republic, Germany, Finland, United Kingdom, Russia, Norway
6,189
C3 Poland, Ukraine 1,239
4.3 Decomposition
To distinguish the differences in growth, productivity and survivorship by cluster a regression-based decomposition method is applied. The unconditional recentered influence function regression (RIF method) in combination with the Oaxaca blinder (OB) decomposition is mostly used for defining the wage gap between female and male (Firpo et al., 2009). The RIF regression evaluates the contribution of each variable on growth and productivity at different
16 quantiles while comparing the cluster. Because survivorship is a dummy variable, no different quantiles can be evaluated in this case. Hence, only the OB decomposition is applied on survivorship. After the regression coefficients are generated through the RIF method, the OB decomposition performs the detailed decomposition to evaluate each quantile unconditionally. Unconditional indicates in this context that the decomposition is not impacted by the condition effect such as the countries in the certain cluster.
The RIF method initiated by Firpo, Forting and Lemieux (2009) provides a detailed composition for quantiles. RIF can be described similar to a simple regression, but the dependent variable is replaced by the recentered influence function of the statistic of interest (Fortin, Lemieux, & Firpo, 2010). IF (y; v) denotes the influence function which corresponds to an observed growth or productivity y for the distributional statistic of interest called v(FY).
Hence, RIF can be defined as RIF (y; v) = v (FY )+IF (y; v).
The first step is to calculate the RIF for the quantile q(𝜏) of 10%, 50% and 90%. RIF can be modelled as (Fortin et al., 2010 p.73):
𝑅𝐼𝐹(𝑦; 𝑄 , ) = 𝑄 +τ − 1{y ≤ 𝑄 } 𝑓 (𝑄 )
(2)
whereas Q(𝜏) denotes the population in 𝜏th quantile, 1 is the indicator function which indicates
whether the growth and productivity y is at or under quantile 𝑄 . The kernel density function is defined as fy and estimated for each q(𝜏). 𝑅𝐼𝐹(Yi; 𝑄 ) is used to estimate the RIF for each
observation. Thus, the coefficients of the unconditional quantile regressions for all groups would be (Fortin et al., 2010 p.74):
𝛾 , = (∑∈ 𝑋 ∗ 𝑋 ) ∗ ∑∈ 𝑅𝐼𝐹 𝑌 ; 𝑄 , ∗ 𝑋 , 𝑔 = 𝐴, 𝐵 (3)
The groups A, B describe the clusters. As there are three cluster, three comparing groups are formed. Thus, group 1 compares cluster 1 against cluster 2, group 2 compares cluster 1 to cluster 3 and group 3 compares cluster 2 to cluster 3.
17 In the second step RIF is used to estimate the effect of a change of distribution of a given covariate on the marginal quantile q(𝜏) of Y using the OB decomposition (Fortin et al., 2010 p.75):
∆ = 𝑋 (𝛾 , − 𝛾 , ) + (𝑋 − 𝑋 )𝛾 , (4)
∆ + ∆ ∙ (5)
whereas the second term, which displays the contribution of each covariate is explainable by:
∆ = ( (𝑋 − 𝑋 ) 𝛾 , ∙ (6)
The results of the decomposition are threefold. The endowment measures the expected change in the compared cluster if assuming the other cluster predictor levels which is called structure effect. The coefficient explains the occurred differences regarding the coefficients. The interaction describes the effect of the combination of coefficient and endowments which is called composition. The reweighting allows to calculate the distribution of another cluster given the same distributional characteristics of the evaluated cluster. The advantage is to calculate the decomposition at a specific point of the distribution.
18
5 Descriptive statistics
In the following text I describe first the explained variables and then the explanatory variables. Table 3 summarizes the descriptive statistics.
Table 3: Descriptive statistics
This table provides an overview of each variable per cluster regarding number of observations (N), mean, standard deviation, 5% quantile and 95% quantile.
Variables cluster N mean sd p5 p95
Growth 1 9,348 0.30 1.50 -1.98 2.75 2 2,327 0.24 1.69 -2.37 3.12 3 656 0.42 1.96 -2.99 3.84 Productivity 1 8,080 2.73 4.04 0.73 7.92 2 1,979 2.92 4.42 0.72 9.47 3 581 3.03 4.81 0.73 9.53 Leverage 1 12,194 0.79 0.82 0.04 1.61 2 5,467 1.09 1.54 0.00 3.79 3 945 0.95 1.48 0.00 3.24 Short-term Leverage 1 12,414 0.65 0.67 0.01 1.37 2 5,644 0.82 1.16 0.00 2.85 3 1,046 0.79 1.10 0.00 2.64 Tangibility 1 12,415 0.25 0.27 0.00 0.83 2 5,598 0.19 0.29 0.00 0.88 3 1,049 0.28 0.33 0.00 0.94 Liquidity 1 11,979 0.31 0.29 0.00 0.93 2 5,155 0.42 0.35 0.00 1.00 3 953 0.29 0.32 0.00 1.00 Size 1 12,344 4.23 1.64 1.52 7.07 2 5,206 4.13 2.14 0.92 8.27 3 949 3.53 1.93 0.64 7.11
The start-ups in the third cluster (C3) grow the most, with the highest variability, followed by start-ups in the first (C1) and the second cluster (C2). This is because the average Growth is 0.24 for C2, 0.30 for C1 and 0.42 for C3. The variability in Growth is the highest for C3 with 1.96 which is also displayed through the lowest 5% quantile of a negative growth of - 2.99 and the highest 5% quantile of 3.84. Hence, C3 grows the most but has the risk of decreasing revenue in comparison to C1 and C2. C1 has the lowest variability of 1.5 and the 5% and 95% quantiles are -1.98 and 2.75 respectively. These values of C1 are lower than those in C2 and C3. Finally, C2 ranges between C3 and C1.
19 C3 is also most productive on average but has the highest variability. This is because C3 has a mean productivity of 3.03 followed by C2 with 2.92 and C1 with 2.73. Hence, C2 has less growth but higher productivity than C1. Also, C2 reaches close to the 95% quantiles of C3 with 9.47 versus 9.53 for C3 respectively showing the high productivity of both. C1 has a clearly lower productivity with 7.92.
The average of total leverage is the highest for C2 followed by C3 and finally C1. This is due to the fact that C2 has a leverage mean of 1.09 indicating that the start-ups have the same amount of liabilities as in total assets. C2 is also more variable then the other two and has the highest 95% quantile of 3.79. Hence, C2 has the highest total leverage ratios and thus the most debt finance. The highest 5% of C3 is also quite high with about 55% less than the second with 3.24 but much less has C1 with 1.61. The short-term leverage variable is similar to the leverage.
C3 has the highest tangibility on average. This is because the mean of tangibility is 0.28 for C3 followed by C1 with 0.25 and then C2 with 0.19. Also, C3 has the highest variability and the highest 5% quantile of 0.94. This high ratio indicates that nearly every asset is tangible.
The liquidity position is the best for C2 with 0.42 mean whereas C3 has the lowest liquidity position with 0.29. The variance is around 0.3 for all three clusters. The 95th quantile
is close to one which would indicate a 100% asset in liquidity.
The largest start-ups are in C1, then C2 and then C3. However, the variance in size is the highest for C2 with also the highest 5% of 8.27. C1 and C3 have smaller start-ups in terms of size in the best 5% of 7.11 and 7.07, respectively.
I can conclude that C3 achieves high growth and high productivity with less cash and partly less debt than the other two clusters but has also high risk to fail in terms of productivity and growth. Also, its average size is much lower in comparison to C1 and C2. C1 consists of larger start-ups with a stable approach, using less debt, less risky and thus also less successful. C2 start-ups use debt financing and have therefore more cash and are productive with it. Table 4 reports the descriptive of the dummy variables.
20 Table 4: Descriptive statistics for dummy variables
This table presents the proportion of survived, not survived and the number of start-ups in the telecommunications sector per cluster. Also, the split of the year of incorporation is presented.
Cluster 1 in % 2 in % 3 in % Total Survived 11,339 86% 5,386 87% 1,181 95% 17,906 Not survived 1,778 14% 803 13% 58 5% 2,639 Total 13,117 100% 6,189 100% 1,239 100% 20,545 Telecomm. 2,049 16% 1,222 20% 235 19% 3,506 No Telecomm. 11,068 84% 4,967 80% 1,004 81% 17,039 Total 13,117 100% 6,189 100% 1,239 100% 20,545 Incorporation year 2008 3,452 26% 1,727 28% 309 25% 5,488 Incorporation year 2009 3,326 25% 1,791 29% 348 28% 5,465 Incorporation year 2010 3,725 28% 1,784 29% 368 30% 5,877 Incorporation year 2011 2,614 20% 887 14% 214 17% 3,715 Total 13,117 100% 6,189 100% 1,239 100% 20,545
The most survived start-ups are in C3 with a survival rate of 95% followed by C2 and C1 with 87% and 86%, respectively. Around 16% to 20% of the start-ups are in the telecommunications sector. The split of year of incorporation is quite evenly divided despite the last year 2011 which is lower for C2 and C3.
Quantile-Quantile plots are a graphical tool to compare two different distributions. For my data set I compare C1 versus C2, C1 versus C3 and C2 versus C3 for both explained variables Growth and Productivity. Survivorship is not applicable for this tool because it is a dummy variable. Graphic 1 shows the quantile-quantile plots.
21 Graphic 1: Quantile-quantile plots
The quantile-quantile plots show the different distribution forms of the clusters for the variable growth and productivity.
Growth: Cluster 1 vs. 2 Growth Cluster 1 vs. 3
Growth Cluster 2 vs. 3
22 Productivity: Cluster 2 vs. 3
From growth perspective, it can be concluded that, C1 performs best for low growth level and C3 performs best for the high growth level. C2 is in between and the ‘normal’ case. The graphic for the growth of C1 vs. C2 shows that start-ups in C2 grow less than in C1 for the low growth level. But for an increase of growth level the shape is normal. When comparing the growth of C1 vs. C3, I can conclude that C1 grows more than C3 for the low growth level. But on a high growth level C3 performs better than C1 and C2. From productivity perspective, the most productive start-ups are even more productive in C3 than in C2 and C1. For the low productivity level, the shape is normal.
6 Empirical findings
First, I am going to present the research findings of the linear regression models for growth, productivity and survivorship. Next, I analyse the results from the decomposition as well for growth, productivity and survivorship.
6.1 Regression
In this section I present the results from the regression analysis. Table 5 summarizes the regression results.
23 Table 5: Regression results
The table presents the standard beta coefficients for each regression of Growth, Productivity and Survivorship. The regressions are run for each cluster and in ‘full’ with all data. The robust standard errors are in parentheses and significance is indicated by *** p<0.01, ** p<0.05, * p<0.1.
Cluster 1 2 3 Full 1 2 3 Full 1 2 3 Full
Explained variable
Growth Productivity Survivorship
Size -0.129*** (0.0123) -0.095*** (0.0205) -0.098* (0.0467) -0.120*** (0.0103) -0.157*** (0.0305) -0.145*** (0.0549) -0.185*** (0.104) -0.154*** (0.0259) 0.022* (0.00208) 0.063*** (0.00254) -0.028 (0.00340) 0.027*** (0.00151) ST. Leverage -0.019 (0.0796) -0.047 (0.0803) 0.237** (0.198) -0.024 (0.0596) -0.007 (0.163) 0.087* (0.258) 0.071 (0.284) 0.021 (0.126) -0.007 (0.0122) -0.052* (0.00869) 0.155 (0.0235) -0.023 (0.00699) Leverage 0.007 (0.0682) 0.154*** (0.0583) -0.184* (0.162) 0.045* (0.0491) -0.006 (0.144) -0.012 (0.174) -0.106** (0.147) -0.009 (0.0976) -0.026 (0.0103) 0.005 (0.00659) -0.214 (0.0187) -0.016 (0.00553) Liquidity 0.018 (0.0715) 0.081** (0.138) 0.034 (0.415) 0.034** (0.0631) 0.022 (0.239) 0.028 (0.568) -0.067 (1.321) 0.029* (0.221) 0.066*** (0.0130) 0.031 (0.0180) -0.030 (0.0256) 0.049*** (0.0100) Tangibility 0.101*** (0.0743) 0.146*** (0.151) 0.080 (0.281) 0.113*** (0.0651) -0.182*** (0.179) -0.196*** (0.396) -0.181** (0.912) -0.178*** (0.161) 0.007 (0.0137) 0.036* (0.0200) -0.008 (0.0234) 0.015 (0.0107) N 8894 2022 522 11438 7367 1624 405 9396 11699 4678 771 17148
24 6.1.1 Growth
Growth can be explained by size, leverage, short-term leverage, liquidity and tangibility. Size impacts growth negatively and there is the tendency the larger the start-up the stronger is the negative impact. The strongest impact is on C1 with - 0.129, then C3 with - 0.098 and followed with - 0.095 for C2.
Taking more short-term debt does not help start-ups to grow in C1 and C2 but those in C3. This is due to the fact that short-term leverage influences growth negatively for C1 and C2 but C3 positively. The negative relation is - 0.019 for C1, - 0.047 for C2 and positive at 0.237 for C3. Further, more debt is beneficial for start-ups in C2 and slightly beneficial for start-ups in C1 but not beneficial for those in C3. This can be viewed as Total Leverage differs to short-term leverage as the coefficient is positive for C1 with 0.007 and C2 with 0.154 but negative for C3 with - 0.184.
More cash does not truly imply higher growth. This is because liquidity has a low positive impact, the most for C2 with 0.081, then C3 with 0.034 and then C1 with 0.018. Further, having more tangible assets is supportive for start-up growth. This can be viewed in the relationship between tangibility and growth in C2 with 0.146, followed with 0.101 for C1 and 0.080 for C3.
Hence, I conclude that debt has diverse effects on the different clusters in terms of growth. Clear results are that a larger size has a negative impact on growth and that more fixed assets support the growth.
6.1.2 Productivity
Productivity can be explained by size, leverage, short-term leverage, liquidity and tangibility. The larger the start-up, the less productive it is. Since size is negatively related to productivity with - 0.185 for C3, - 0.157 for C1 and - 0.145 for C2.
Further, taking on short-term debt has a positive effect on productivity for C2 and C3 and a negative effect on C1. This is because the relation of short-term leverage on productivity is 0.087 for C2, 0.071 for C3 and - 0.007 for C1. Due to the reason that C1 has the largest start-ups on average, for those an increase in short-term debt has nearly no effect. But more debt regarding the total debt the start-up has, impacts the productivity negatively. This can be viewed
25 as the total leverage is negatively related to productivity by -0.106 for the C3, - 0.012 for C2 and - 0.006 for C1. So, the strongest negative impact the leverage has is on C3.
Having more cash supports start-ups in C1 and C2 slightly but decreases productivity for start-ups in C3. This is due to the fact that liquidity is positively related to productivity by 0.028 for C2, 0.022 for C1 and negatively with - 0.067 for C3. Further, more fixed assets make the start-up less productive. This can be viewed as tangibility is negatively associated with productivity by - 0.196 for C2, - 0.182 for C1 and - 0.181 for C3.
Hence, I conclude that a larger size, fixed assets and more debt decreases the productivity of a start-up. Whereas cash mainly fosters productivity.
6.1.3 Survivorship
Survivorship can be explained by size, leverage, short-term leverage, liquidity and tangibility. The larger the start-up the more likely it survives is applicable for C1 and C2. The positive relationship between size and survivorship is 0.063 for C2 and 0.022 for C1. C3 has a negative relation between size and survivorship by - 0.028.
Furthermore, taking on short-term debt is negative for start-ups in C1 and C2 but saves start-ups in C3. This negative effect of short-term leverage is - 0.052 for C2, - 0.007 for C1 and positive for C3 with 0.155. Also, debt decreases the chance of survival. This can be viewed as the negative relation of total leverage is the highest with - 0.214 for C3 and - 0.026 for C1 and slightly positive with 0.005 for C2.
Additionally, more cash helps C1 and C2 start-ups to survive. This is because liquidity is positively related to survival by 0.066 for C1, 0.031 for C2 and - 0.030 negatively for C3. Also, having tangible assets makes survival more likely for C1 and C2 start-ups. Since the relation of tangibility to survival is 0.036 for C2, 0.007 for C1 and negative for C3 by - 0.008.
Therefore, I can conclude that nothing saves C3 start-ups despite short-term debt. But a larger size, cash and tangible assets saves C1 and C2 start-ups. Whereas debt is mainly negatively associated with survival.
26 6.2 Decomposition
In this section I present the results from the RIF in combination with the OB decomposition. Table 6 reports the decomposition results. The difference describes the growth, productivity and survivorship gap between the two compared cluster.
27 Table 6: Decomposition results
The table gives an overview of the decomposition results for comparing C1vs.C2, C1 vs. C3 and C2 vs. C3 for the quantiles 10%, 50% and 90%. The difference effect is threefold into endowment x, coefficient ß and interaction xß. For Survivorship only the 50% quantile is provided because it is a dummy.
Cluster 1 vs. 2 1 vs. 3 2 vs. 3 1 vs. 2 1 vs. 3 2 vs. 3 1 vs. 2 1 vs. 3 2 vs. 3
Dependent variable Quantile Q10 Q50 Q90
Growth difference -0.2142 0.3491 0.5633 0.0772 -0.0258 -0.1030 0.5199 -0.4032 -0.9230 x -0.0361 0.0054 0.1388 -0.0123 0.0064 0.0468 -0.0238 -0.0737 0.0090 ß -0.2657 0.4489 0.5847 0.0800 0.0126 -0.1071 0.5340 -0.3168 -0.9274 xß 0.0876 -0.1052 -0.1601 0.0095 -0.0448 -0.0427 0.0097 -0.0126 -0.0046 Productivity difference 0.2342 0.1010 -0.1332 0.4560 0.1960 -0.2600 1.0792 0.5298 -0.5494 x 0.0082 -0.0012 -0.0077 0.0078 -0.0231 -0.0461 -0.0112 -0.1525 -0.3478 ß 0.2373 0.0714 -0.1478 0.4560 0.1587 -0.2892 1.1645 0.4732 -0.6793 xß -0.0112 0.0307 0.0222 -0.0078 0.0605 0.0753 -0.0740 0.2091 0.4777 Survivorship difference -0.0134 -0.1045 -0.0911 x 0.0015 -0.0029 -0.0091 ß -0.0104 -0.1132 -0.0970 xß -0.0045 0.0117 0.0150
28 6.2.1 Growth
Growth is driven by unexplained factors like the economic condition in this case and not by the capital decision. The difference is mainly impacted by the coefficient and less by the endowment or interaction.
On the 10th quantile the main coefficients impacting the difference are liquidity
negatively by - 0.2365, which shows a better liquidity position for C2 start-ups in comparison to C3 ones. The year of incorporation has a strong positive impact by 0.2003 on the difference (see Appendix Table 10 for detailed decomposition results). Due to the fact year of incorporation is a dummy, it provides no explanation of the difference.
Furthermore, the best environment for low growth start-ups is C2. As the ranking on the 10th quantile is as follows C2 grows the most, then C1 and then C3. This means that the
discrimination is the highest in C3. The highest growth difference is between C2 vs. C3. If C3 had the same characteristics as C2, C3 would grow 0.5633 more. The best environment for median growing and high growth start-ups is C3. As the ranking shifts to C3, C1 and then C2 on the 50th and 90th quantile showing that C3 benefits in growth in comparison to C1 and C2.
This interpretation is in line with the quantile-quantile graphics from Section 5.
Further, high growth start-ups in C3 have an advantage in terms of short-term debt to C2. The coefficient for short-term leverage is -0.4837 on the 90th quantile. Whereas the total
debt coefficient shows that C2 has an advantage over C3 by 0.2810 (see Appendix Table 10 for detailed decomposition results). Additionally, a larger size is a disadvantage for high growth start-ups. When comparing C1 vs. C2 the coefficient is - 0.4882 at the 90th quantile. As the
largest start-ups are in C1, C2 start-ups would grow less if they had C1 characteristics.
Hence, I can conclude that growth success is not driven by the finance decision. C2 start-ups perform best on the low growth quantile and C3 performs best in median and high growth quantile even though it has disadvantages in liquidity and debt. Graphic 2 shows the growth differences in the clusters.
29 Graphic 2: Growth difference
The graphic shows growth differences for the 10th, 50th, and 90th quantile.
The graphic 2 illustrates that growth differences between C1 vs. C2 are increasing whereas differences between C1 vs. C3 and C2 vs. C3 declines negatively. Therefore, C3 mitigates characteristics of C1 and C2 and has the highest ability to grow, after reaching the 50th quantile.
6.2.2 Productivity
The productivity is as well as the growth not driven by the choice of capital but by unexplained factors such as the economic condition in the certain cluster countries. The difference in comparing the clusters is mainly driven by the coefficients. Especially tangibility negatively impacts productivity (see Appendix table 11 for detailed decomposition results). Since productivity is expressed as revenue to fixed assets.
The best environment for productivity is C1. When ranking the cluster according to their differences. C1 has the highest productivity, followed by C3 and C2. This is valid in all quantiles. Also, start-ups in C2 would be much more productive if they would be in C1. The highest difference is on all quantiles is between C1 vs. C2. On the 10th quantile the difference
30 A larger size influences high productive start-ups negatively. When comparing C2 vs. C3 the coefficient is 0.2046 on the 50th quantile and increases to 0.9729 on the 90th quantile.
Also, on the 90th quantile the coefficient between C1 and C2 is very high at – 1.0578 indicating
the negative impact of size on productivity if C2 ups would have the same size as C1 start-ups. Furthermore, for high productive start-ups C1 and C2 have an advantage in terms of cash in comparison to C3 start-ups. This is because the difference coefficient of liquidity of C2 vs. C3 is 0.4549 and 0.3343 for C1 vs. C3 on the 90th quantile (see Appendix table 11 for detailed
decomposition results).
I conclude that, the environment of C1 is the most supportive in terms of productivity, whereas C3 faces deficits regarding cash availability to be more productive. Graphic 3 shows the productivity differences in the cluster.
Graphic 3: Productivity difference
The graphic shows productivity differences for the 10th, 50th, and 90th quantile.
The graphic illustrates that productivity differences in C1 vs. C2 and C1 vs. C3 are increasing with rising quantiles. Whereas C2 vs. C3 productivity difference increases negatively. So, the gap widens on high productivity level.
6.2.3 Survivorship
The survival is driven by the unexplained factors in the decomposition. So, the differences between the clusters are impacted by the coefficient. Start-ups in C3 survive more likely, whereas C1 fail more likely (see Appendix table 12 for detailed decomposition results),
31 which is also expressed in the Section 5, the descriptive statistics. The clusters are ranked as C3, C2 and C1. Graphic 4 shows the survivorship differences.
Graphic 4: Survivorship difference
The graphic shows survivorship differences for the 50th quantile.
7 Discussion
Despite the mainly negative relationship of leverage with productivity and survivorship, the choice of capital structure is not the main factor explaining differences across countries. These differences remain unexplained according to my defined factors and therefore can be reasoned through different economic conditions.
On the one hand growth can even be fostered through debt in the following countries: Spain, France, Italy, Portugal, Sweden, Estonia, Czech Republic, Germany, Finland, United Kingdom, Russia and Norway. On the other hand, debt decreases growth in Ukraine and Poland. Reasons could be for Ukraine the political situation as the war began in Ukraine in 2014. Hence, entrepreneurs are likely not to take on debt or only for short-term. This is because the relation of short-term debt to growth is positive in Ukraine and Poland. Another reason could be that Poland and Ukrainian start-ups face less access to debt.
I also find that the larger the start-up, the less growth and the less productive is the start-up and the more likely to survive which is in line with Frank & Goyal (2009).
32 Furthermore, tangibility fosters growth and survival as for example patents give start-ups opportunities to build up a certain software without interference of competitors. For the case of survival, they can still sell the patent or liquidate fixed assets.
More cash supports ambitious growth strategies which confirms Robb & Robinson (2014). So, start-ups need cash to expand into a new market. Also, cash fosters productivity and survival in Spain, France, Italy, Portugal, Sweden, Estonia, Czech Republic, Germany, Finland, United Kingdom, Russia, Norway. Only in Ukraine and Poland, more cash does not make them more productive or support the survival.
The Global Entrepreneurship Index which represents the health of the entrepreneurship ecosystem in a certain country, does not indicate disadvantages for Poland (Global Entrepreneurship Index (2018). Whereas Ukraine and Russia are the countries with obstacles for entrepreneurs. I cannot completely confirm this as according to my results there are disadvantages for Poland such as less debt access.
The decomposition results show that the choice of capital is not impacted by start-up decisions but by economic conditions. According to Frank & Goyal (2009) the expected inflation can be a reason for the choice of capital. During my observation time especially, the inflation in Russia and Ukraine was volatile and peaked at about 17% in Russia and nearly 60% for Ukraine in 2015 (Trading economics, 2018). This can have a strong impact on the success of the start-up and choice of capital because start-ups are less mature and not yet build up capital buffers. For the rest of the sample countries the inflation rate remained less volatile and lower. Other reasons for differences in the cluster, which I examined in the decomposition, can be personal character, strategy and non-economic goals that are not considered in my study.
Furthermore, Czech Republic, Germany, Finland, United Kingdom, Russia and Norway offer advantages for the low growing start-ups. As these start-ups have a general high level of leverage, which means more availability of cash, enables them to finance their growth strategies. Spain, France, Italy, Portugal, Sweden, and Estonia have larger start-ups and have less access to debt, which makes them less successful regarding growth but makes them focus on productivity. Ukraine and Poland represent the ‘winners’ for median and high growth start-ups. Even though both countries face difficulties regarding debt access, and by that to cash, start-ups take the risk for high growth which enables them to become ‘winners’ or underperform
33 worse than others. As Ukrainian and Polish start-ups have more tangible assets than the other countries, this can be one reason why they are more likely to survive. Since tangible assets help start-ups to survive, because they can liquidate them.
Moreover, my results show that the total leverage increases the likelihood of failure in Spain, France, Italy, Portugal, Sweden, Estonia, Poland and Ukraine. This supports the statement of Miglo (2016) who argues that debt can lead to a debt overhang, which can result in bankruptcy. Even though Czech Republic, Germany, Finland, United Kingdom, Russia and Norway have the highest leverage, their survival is more likely than Spain, France, Italy, Portugal, Sweden and Estonia. The definition of bankruptcy varies in the different countries as well as the accounting standards. Therefore, this can be a reason that some countries have a higher leverage and still are more likely to survive. Also, it needs to be considered that going bankrupt can be reasoned to the fact that the start-up idea is not valuable. Thus, bankruptcy provides a natural selection. Hence, start-ups in Poland and Ukraine are not willing or able to take this risk, even though the idea would have been valuable.
8 Conclusions
Finally, debt can negatively impact the productivity and survival. So, equity is the preferred choice. But every start-up has to decide carefully on their own which means of capital for them is the preferred choice as individual factors need to be considered. The start-up success depends on financial fundamentals as well as on external factors such as the economic condition in the certain country or personal reasons such as family situation of the founder.
My study contributed in theoretical perspective as I used a new approach of Oaxaca-Blinder decomposition in combination with the recentered influence function on growth, productivity and survival differences. I covered a wide range of European countries and tested the hypothesis with empirical data on the information sector over five years. But my study has limitations. The clustering is done according to country of origin instead of the specifications of each start-up. As each start-up itself has different characteristics it may have more similarities with a start-up in another country than in the same country.
Furthermore, the external validity is limited as I am researching only few countries, the research can be different in other countries, especially Asia, Africa or America. Also, my focus
34 is on the information sector and therefore diverse sectors may response differently to the hypothesis. Hence, a future goal can be to rearrange the cluster, broaden the research with more countries and include more sectors. Also, for timing it can be different in another period as my focus is on 2008 and 2016. Moreover, the internal validity is limited because of other existing definitions of start-ups success instead of growth, productivity and survivorship.
35
9 References
Barclay, M. J., Morellec, E., & Smith, C. W. (2001). On the Debt Capacity of Growth Options. SSRN Electronic Journal, 79(1), 37–60. https://doi.org/10.2139/ssrn.271949
Bustamante, M. C., & Acunto, F. D. (2018). Banks’ Screening of Startups: The Role of Initial Leverage, (May 1, 2018). Retrieved from https://ssrn.com/abstract=3073044
Cooper, A. C. (1993). Challenges in predicting new firm performance. Journal of Business Venturing, 8(3), 241–253. https://doi.org/10.1016/0883-9026(93)90030-9
Cooper, A. C., Gimeno-Gascon, F. J., & Woo, C. Y. (1991). A resource-based prediction of new venture survival and growth. Academy of Management Proceedings, Vol. (1991), Retrieved from https://doi.org/10.5465/ambpp.1991.4976561
Cooper, A. C., Gimeno-Gascon, F. J., & Woo, C. Y. (1994). Initial human and financial capital as predictors of new venture performance. Journal of Business Venturing, 9(5), 371– 395. https://doi.org/10.1016/0883-9026(94)90013-2
Earlymetrics. (2018) Earlymetrics – Start-up Rating. Retrieved from http://earlymetrics.com/ Everitt, B. S., Landau, S., Leese, M., & Stahl, D. (2011). Cluster Analysis. Quality and Quantity, 14. https://doi.org/10.1007/BF00154794
Faulkender, M., & Petersen, M. A. (2006). Does the source of capital affect capital structure? Review of Financial Studies, 19(1), 45–79. https://doi.org/10.1093/rfs/hhj003
Firpo, S., Fortin, N. M., & Lemieux, T. (2009). Unconditional Quantile Regressions. Econometrica, 77(3), 953–973. https://doi.org/10.3982/ECTA6822
Fortin, N., Lemieux, T., & Firpo, S. (2010). Decomposition Methods in Economics. Working Paper 16045. Retrieved from http://www.nber.org/papers/w16045
Frank, M. Z., & Goyal, V. K. (2003). Testing the pecking order theory of capital structure. Journal of Financial Economics, 67(217-248). https://doi.org/10.1016/S0304-405X(02)00252-0
Frank, M. Z., & Goyal, V. K. (2009). Capital structure decisions: Which factors are reliably important? Financial Management, 38, 1–37.
https://doi.org/10.1111/j.1755-053X.2009.01026.x
Graham, J. R., & Leary, M. T. (2011). A Review of Empirical Capital Structure Research and Directions for the Future. Annual Review of Financial Economics, 3(1), 309–345.
https://doi.org/10.1146/annurev-financial-102710-144821
Hackbarth, D., Miao, J., & Morellec, E. (2006). Capital structure, credit risk, and macroeconomic conditions. Journal of Financial Economics, 82(3), 519–550. https://doi.org/10.1016/j.jfineco.2005.10.003
Harris, M., & Raviv, A. (1991). The Theory of Capital Structure. The Journal of Finance, 46(1), 297–355. Retrieved from www.jstor.org/stable/2328697
Hennessy, C. A., & Whited, T. M. (2005). Debt dynamics. The Journal of Finance, 60(3), 1129–1165. Retrieved from https://doi.org/10.1111/j.1540-6261.2005.00758.x
36 Hovakimian, A., Hovakimian, G., & Tehranian, H. (2004). Determinants of target capital structure: The case of dual debt and equity issues. Journal of Financial Economics, 71(3), 517–540. https://doi.org/10.1016/S0304-405X(03)00181-8
Hovakimian, A., Opler, T., & Titman, S. (2001). The debt-equity choice. Journal of Financial and Quantitative Analysis, 36(1), 1–24. https://doi.org/10.2307/2676195
Korajczyk, R. A., & Levy, A. (2003). Capital structure choice: Macroeconomic conditions and financial constraints. Journal of Financial Economics, 68(1), 75–109.
https://doi.org/10.1016/S0304-405X(02)00249-0
Lang, L., Ofek, E., & Stulz, R. (1995). Leverage, Investment and Firm Growth. NBER Working Paper 5165. Retrieved from http://www.ssrn.com/abstract=271949
Lemmon, M. L., Roberts, M. R., & Zender, J. F. (2008). Back to the beginning: Persistence and the cross-section of corporate capital structure. Journal of Finance, 63(4), 1575–1608. https://doi.org/10.1111/j.1540-6261.2008.01369.x
Lemmon, M. L., & Zender, J. F. (2010). Debt capacity and tests of capital structure theories. Journal of Financial and Quantitative Analysis, 45(5), 1161–1187.
https://doi.org/10.1017/S0022109010000499
MacKay, P., & Phillips, G. M. (2005). How does industry affect firm financial structure? Review of Financial Studies, 18(4), 1433–1466. https://doi.org/10.1093/rfs/hhi032
Miglo, A. (2016). Capital Structure in the Modern World. https://doi.org/10.1007/978-3-319-30713-8
Milligan, G. W., & Cooper, M. C. (1985). An examination of procedures for determining the number of clusters in a data set. Psychometrika, 50(2), 159–179.
https://doi.org/10.1007/BF02294245
Modigliani, F., & Miller, M. H. (1958). The American Economic Review, 48(3), 261–297. Retrieved from http://www.jstor.org/stable/1809766
Modigliani, F., & Miller, M. H. (1963). Corporate Income Taxes and the Cost of Capital: A Correction. The American Economic Review, 53(3), 433–443.
https://doi.org/10.2307/1809167
Myers, S. C., & Majluf, N. S. (1984). Corporate financing and investment decisions when firms have information that investors do not have. Journal of Financial Economics, 13(2), 187–221. https://doi.org/10.1016/0304-405X(84)90023-0
Oaxaca, R. (1973). Male-Female Wage Differentials in Urban Labor Markets. International Economic Review, 14(3), 693–709. Retrieved from http://www.jstor.org/stable/2525981 Rajan, R. G., & Zingales, L. (1995). What Do We Know about Capital Structure? Some Evidence from International Data. Journal of Finance, 50(5), 1421–1460.
https://doi.org/10.1111/j.1540-6261.1994.tb00086.x
Robb, A. M., & Robinson, D. T. (2014). The capital structure decisions of new firms. Review of Financial Studies, 27(1), 153–179. https://doi.org/10.1093/rfs/hhs072
37 Smith, J. K., Smith, R. L., and Bliss, R. T. (2011). Entrepreneurial Finance: Strategy,
Valuation, and Deal Structure, Stanford University Press. https://www.sup.org/books/title/?id=18738
Strebulaev, I. A. (2006). Do Tests of Capital Structure Theory Mean What They Say? The Journal of Finance, 62(4), 1747–1787. https://doi.org/10.1111/j.1540-6261.2007.01256.x Titman, S., & Wessels, R. (1988). The Determinants of Capital Structure Choice. The Journal of Finance, 43(1), 1–19. https://doi.org/10.1111/j.1540-6261.1988.tb02585.x
Trading Economics (2018) 20 million Indicators from 196 countries. Retrieved from https://tradingeconomics.com /
United States Department of Labor (2018): The Information sector. Retrieved from https://www.bls.gov/iag/tgs/iag51.htm
Ward, J. H. (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58(301), 236–244.
38
10 Appendices
10.1 Cluster method
To cluster the data the non-hierarchical approach k-median and the hierarchical approach wardslinkage was used. The goal of the k-median approach is to minimize the error terms and reduce the distance while part n observations into k cluster using a dissimilarity matrix. For this case n is the number of countries, 14, and k is the number of cluster. The algorithm of k-median relocates the coefficients to the cluster whose exemplar (centre of a cluster) is the nearest regarding the dissimilarity measure (Everitt et al., 2011). Afterwards the cluster get re-evaluated until no move of a single object improves the cluster criterion. The k-median approach is more robust with regard to outliers than the k-mean approach because of using the median instead of the mean (Everitt et al., 2011). The lack of homogeneity can be expressed as (Everitt et al., 2011 p. 112):
ℎ(𝑚) = 𝑚𝑖𝑛 ,…, [ 𝛿 , ]
(7)
where ℎ(𝑚) denotes the lack of homogeneity of the mth group, 𝛿 measures the dissimilarity
between the Ith object in the qth cluster and the vth object in the kth cluster with n
m objects in
the dissimilarity matrix and r=1. This index indicates the minimum sum over the dissimilarities between all objects in cluster m and a single cluster member.
According to Everitt et al. (2011) p. 113 the k-median cluster criteria can be explained as a minimizing problem of:
𝐶(𝑛, 𝑔) = min ,…, [ℎ(𝑚)] (8)
where C indicates the cluster criterion that is aimed to be minimized depending on n, the number of objects and g, the number of cluster.
