Resources Articles of Interests
The Horse or the Jockey? Evidence from Nascent Firms where a Founder Dies
Posted in M&A
By Hans K. Hvide, May 14, 2008
1 Introduction
Entrepreneurship has for long been viewed as central to economic performance and long- run economic development (e.g., Schumpeter, 1943). We know very little, however, about what are the most critical factors behind the growth and survival of nascent firms, whether it is market conditions, ideas, financial capital, or people. This is unfortunate because better knowledge would be valuable to understand economy-wide creative destruction and growth.
The purpose of this paper is to empirically analyze the importance of founders to nascent firms. How much does a start-up, after its inception, rely upon its founder? One reason to be concerned about this question is that theory suggests that firms that are highly dependent on the founder would encounter constraints to growth such as being less likely to obtain external finance (Hart & Moore, 1994), and scale limitations due to limited attention span of the founder (Gifford, 1998). Thus if the nascent firm rely to a large extent on the founder then this could, at a higher level, be an important detriment to economic growth.
That the founder is an important asset for nascent firms is indirectly supported by findings from Bhide (2000). Bhide studies 100 fast-growing companies from the Inc 500 list in 1989 and finds that 71% of the entrepreneurs in his sample, ”[...] replicated or modified an idea encountered through previous employment”. The founders typically did little formal planning before setting up a business, and the initial business idea was frequently adjusted and modified along the way. This evidence is most reasonably interpreted as the founder being important as a "jockey" in the in the formative phase of the start-up’s life.
Theoretical support for the founder being crucial for nascent firms comes from critical resource theory (Wernerfelt, 1984, and Rajan and Zingales, 2001), where a firm is "a web of specific investments built around a critical resource or resources... At some point the critical resource becomes the web of specific investments itself." (Zingales, 2000). As pointed out by Kaplan et al. (2006), for nascent firms it is reasonable to think of the founder as the critical resource, while after some time non-human assets, or other human assets than the founder, become more important.
The question I ask is how important a founder is to the performance of a nascent firm. Answering this question is hard because we usually do not observe a founder’s level of engagement in developing the nascent firm, and if we do the level of engagement is likely to be highly non-random. For example, the founder being unactive could indicate that the start-up is less promising or that there are serious conflicts between founders.
To learn more about the role of founders we would, as a thought experiment, like to randomly take entrepreneurs out of nascent firms, deport them to a desolate island without means of communication, and study how well their firms fared relative to a control group of firms where the founder stays engaged. The empirical strategy I use attempts to mimic this thought experiment. Using a unique dataset on a large representative sample of new firms from Norway, I exploit the random variation in founder presence through the occurrence of founder death. I analyze the performance of firms where the founder dies and compare it with the performance of firms where the founder stays alive. Differences in performance, after controlling for observable founder characteristics such as age and prior wealth and observable firm characteristics such as size and industry, will be interpreted as due to the importance of the founder. For example, if nascent firms where the founder dies have a much lower survival rate than nascent firms where the founder stays alive, I would interpret this as support for the notion that founder has an important, and non- substitutable, role in the nascent firm.
There are two main issues with interpreting death effects as causal. The first is reverse causality; bad performance of the firm could cause "bad" stress and increase the founder’s death probability (Kiwimäi et al., 2002). Such an effect, if present, would tend to bias our estimates on the importance of the founder downwards. The second issue with interpreting death effects as causal is that founder death could be related to unobservable founder characteristics that affect performance, in particular founder ill health. This possibility is perhaps less of a concern than reverse causality, as poor performance for companies with ill founders would also be evidence supporting the notion that founder engagement is important to firm performance.
The basic results of the paper can easily be summarized. I study the effect of a founder dying on the performance of nascent firms. I find that across a variety of performance measures, such as sales, asset growth, survival, and profitability, the performance of the firms where the founder dies seem only slightly inferior to the firms where the founder stays alive. For example, after controlling for founder and firm characteristics, firms where the founder dies has 5 percentage points lower probability of surviving the first 4 years of operations than firms where the founder stays alive. The effect on sales and asset growth of founder death seems negligible.
If the founder were crucial in jockeying the nascent firm, I would expect strong negative e?ects from the death of founders with a majority ownership share in the firm. While the basic regression analysis is on all the founders, I redo the analysis confining attention to majority founders. It turns out that the results are very similar, in that the effect of founder death seems very slight. For example, if the founder dies the estimate of the
4-year survival probability of the firm is still only 5 percentage points lower than for alive founders. For 6-year survival, there is no difference between the two groups of founders.
Those results are derived by comparing the performance of firms where the founder dies with firms where the founder stays alive. The panel structure of the data allows us to, as an alternative regression strategy, consider within-firm effects. I analyze survival, growth and profitability in a window around the founder death event. Consistent with the cross-sectional analysis, the estimates suggest that the effect of founder death is slight. For example, the one-year survival rate going into the year of death is almost identical to the one-year survival rate going out of the year of death. Moreover, the firms seem to experience healthy sales and asset growth in the two-year window around founder death.
Overall, the results suggest that the entrepreneur’s importance as a jockey to the nascent firm is quite limited; once firms are set up, the founder seems to be substitutable. The results should not, however, be interpreted as the founder being unimportant but rather that, as emphasized long ago by Hayek and Schumpeter, the main function of the entrepreneur could be to discover new opportunities or to, so to speak, create the horse.
As economists we know surprisingly little about where firms come from and what constitutes a firm in its infancy. One related strand of work is the Stanford Project on Emerging Companies. As we do, they study a panel of young firms but ask a different set of questions. Baron and Hannan (2002) summarize the findings of the project as showing that initial models of employment tend to persist. When the employment models are changed, employee turnover increases and performance declines. Kaplan et al. (2006) study strategy and management changes in a sample of 156 fast-growing companies that eventually go public. They find that between being venture capital financed and the initial public offering, almost none of these companies change their line of business, while the management team changes quite frequently. Thus for this sample of companies, the initial idea seemed more stable than the initial management team. Since all the companies in this study were very successful, it was hard to evaluate whether the idea or the personnel were more important to performance.
Worrel et al. (1986) study the stock market response of the death of 127 key executives of listed
The horse/jockey terminology comes from the venture capital community, where a long-standing debate concerns whether the key determinant of the success of a young company is a strong business/market (the horse) or strong management/founders (the jockey). While venture capitalists would prefer to invest in young companies with both a strong business and a strong founder team, different venture capitalists tend to resolve trade-offs differently. For example, Gompers and Lerner (2001) describe how Tom Perkins of Kleiner Perkins mainly considered a company’s technological position, and Don Valen- tine of Sequoia assessed the size and growth of the market. On the other hand, Arthur Rock, an early investor in Apple, put more weight in the quality of the founding team. In support of the latter view, Gladstone and Gladstone (2002, p.91-92) quotes an old saying as “You can have a good idea and poor management and lose every time. You can have a poor idea and good management and win every time.”
The paper is structured as follows. Section 2 presents the data and some summary statistics. Section 3 discusses the empirical strategy, and Section 4 contains the results. Section 5 concludes.
2 Data and empirical strategy
2.1 Data
I use a dataset consisting of 6,800 incorporated limited liability firms started up between 1996 and 2003 in
1. Accounting information from Dun & Bradstreet’s database of accounting figures based on the annual financial statements. This data includes variables such as sales, assets, and profits for the years 1992-2006, as well as 5-digit industry codes.
2. Data on individuals from 1986 to 2005 prepared by Statistics Norway. These records are based on register data and tax statements, and include the anonymized personal identification number and yearly sociodemographic variables such as gender, age, education in years, taxable wealth, and income. The records also identifies the year of death (if applicable).
3. Founding documents submitted by new firms to the government agency ‘Bronnoy- sundregisteret’. These data includes the start-up year, total capitalization, and the personal identification number and ownership share of all founders with at least 10 percent ownership share.
For each new firm identified in 1) we compile a list of founders identified through 2) and match their associated sociodemographic information from 3). Due to alterations in the reporting requirement in 1997 we were able to match around 80% of the founders in companies founded after 1997 and around 20% before. We are then left with about 12,500 unique founders. For some firms, the first year of financial reporting, defined through 1), is later than the year of incorporation defined by 2). For these firms, I define the first year as the first year of reporting.
An adverse consequence of the low barriers to starting up an incorporated company and its favorable tax treatment, is that many start-ups, particularly within real estate, are tax-shelters or have minimal activity. This problem was dealt with in two ways. First, by over-sampling manufacturing and IT since tax shelters are less likely to occur in these industries (also these industries include companies with high growth potential and exhibit high variation in capital-intensity). We selected all start-ups within the high tech sectors NACE 23-35 and 72 from 1994-1998, and all start-ups within manufacturing and IT, NACE 15-37 and 72 from 1999-2002. We added a random 25 percent sample of other non-financial private sector start-ups from 1999-2002. We expanded the sample after 1998 because the cost of collecting data for the more recent period is lower. Second, to further reduce the share of “empty shells” firms are included only if they have at least NOK 500 000 in sales and at least two persons employed during one of the first two years of operation. Avoiding sampling empty companies is important as the incorporation documents had to be hand-collected by research assistants at considerable cost.
2.2 Summary statistics
The following table describes the characteristics of founders that are alive at the end of 2005 (d = 0) versus those that die sometime before the end of 2005 (d = 1).
The table depicts summary statistics of founders and the firms they start up, broken down on whether the founder dies or not. Previous income and previous wealth are calculated as means over five years prior to start-up year. No founders is the number of founders with at least 10 percent ownership share. Leverage ratio is defined as book value of debt divided by assets. IT is defined as fraction of IT start-ups (NACE 72). Krone values are expressed in 2002 kroners.
Panel A contains information about the founders. Of the about 12 500 entrepreneurs, 181 (1.7%) die at some point between starting up the firm and the end of 2005. The firm is on average 3.2 years old when the founder dies. 78 percent of the founders that die do so before the end of the sixth year of operations. Not surprisingly, the entrepreneurs that consecutively die are on average older on the start-up date than the entrepreneurs that stay alive. They are also more wealthy and have a slightly higher ownership share in the firm, and about one year less of education on average. Panel B describes summary statistics for firms, broken down on whether the founder died or not. The firms with subsequently dead founders are somewhat larger both in terms of initial equity, assets, and the number of employees. Since being older is correlated with both higher death probability and a higher wealth, these findings are not surprising.
2.3 Empirical strategy
The question I ask is how important a founder is to the performance of a nascent firm. Answering this question is hard because we usually do not observe a founder’s level of engagement in developing the nascent firm, and if we do the level of engagement is likely to be non-random. To exploit variation between nascent firms with respect to founder death provides an opportunity to overcome these estimation problems. My empirical strategy is to exploit the random presence of founders created through variation in the death event in order to identify the causal effect of founders. The basic regression model explains, for each founder, the firm performance in the following manner,
Start-up performance = β(founder characteristics) + γ(start-up characteristics) + λd + є
(1)
where d is an indicator variable that equals 1 if the founder dies and equal to 0 if the founder stays alive. My main interest is to estimate λ, which I interpret as the economic magnitude of how much start-ups rely on their founder. The performance variables used are defined in Section 2.4.
There are two main issues with interpreting λ in (1) as a causal effect. The first is reverse causality; it is possible that bad performance of the start-up could increase the stress level of the founder and increase his death probability. The health literature suggests that increased death probability is associated with "bad" stress, i.e., a combination of high work demand and limited ability to cope with it, rather than high work demand by itself. This suggests that a possible effect from performance to death probability should primarily be for founders where the firm performs badly. Thus reverse causality should, if present, should bias our estimate of λ downwards.
An indirect way of assessing the importance of reverse causality is to investigate whether entrepreneurs have higher death probability compared to non-entrepreneurs with the same socioeconomic characteristics. My (unreported) analysis of this question suggested that entrepreneurs have a somewhat lower death probability.10
The second issue with interpreting λ as a causal effect is that omitted variables could make the residual in (1) correlated with the death probability. It would then appear as death is driving performance while it would in fact be the omitted variable. An obvious candidate for omitted variable is the health status of the founder. This possibility is perhaps less of a concern than reverse causality, as poor performance due to illness of the founder would also support the notion that founder presence is important. Second, ill founders could adjust the ambitions of the firm when setting it up, for example by having a lower initial capitalization or by recruiting additional founders as back-up. Such adjustments could have a detrimental effect on subsequent firm performance. If such conscious adjustments were the case, we would expect firms where the founder dies to be smaller and to have a higher number of founders at the start-up date than firms where the founder stays alive. This issue is addressed below.
To investigate whether firms where the founder dies are different at the start-up date than firms where the founder stays alive, I regress the firm variables initial equity, number of founders, number of employees, and assets on the founder variables. The founder variables are age, gender, education level, previous income, and previous wealth. The results, reported in the following table, does not suggest the presence of systematic differences between firms where the founder dies and firms where the founder stays alive.
Each row is a separate regression where the dependent variable is given in the first column. Founder variables (age, gender, previous income, and previous wealth), firm variables (log equity, number of founders, two-digit industry), and start-up year are included as controls. The estimation method is ordinary least squares in (1)-(4) and probit in (5). In (5), the marginal effect is reported. Robust standard errors clustered at firm level in parenthesis. *** Significant at the 1 % level ** Significant at the 5 % level * Significant at the 10 % level.
The specification in (1) is cross-sectional. The data includes information on death year, and therefore gives scope for tests that explores changes in firm performance after a founder dies. While the data tells us year of death, it does not give information on the cause of death. Aggregate statistics from the US Department of Health suggest that the most common cause of death for persons in the 45-65 age bracket (containing two-thirds of the dead founders) is cancer (35 percent) and heart decease (20 percent). Cancer in particular is typically associated with a preceding period of illness that could affect the founder’s productivity well before the death event. The changes in firm performance from the death event can therefore be hard to track down, even if there are large cross-sectional effects of founder death. I therefore initially focus on the cross-sectional, between-firm, specification in (1), and then in Section 3.5 analyze within-firm changes in a window around the year of death.
2.4 Performance measures
I measure firm performance by three different measures. The first measure is business survival. A firm is defined to survive t years of operations if it reports a (negative or positive) profit and has at least NOK 50,000 in sales in year t + 1. Survival provides a simple measure of performance but is rather crude; for example lack of survival could mean that the firm has been bought up or that the entrepreneur closes it down due to better opportunities in the labor market. The second measure is profitability, as measured by operating returns on assets (OROA). OROA is the standard performance measure in a large accounting and financial economics literature (see e.g. Bennedsen et al., 2007a, and the references therein) and defined as the ratio of earnings before interest and taxes (EBIT) to the total asset base used to generate them. Unlike returns on equity or returns to capital employed, OROA compares firm profitability relative to total assets. In contrast to net income-based measures such as return on assets, OROA is not affected by capital structure or dividend policy differences across firms. As asset base when calculating OROA, I use the average of the assets at the end of year t − 1 and at the end of year t. Neither survival nor profitability measure start-up size, indications of value potential and job creation. I therefore add sales, number of employees, and total assets as third measures of performance, evaluated at the end of the fourth and at the end of the sixth year of operations.
3 Basic results
I now turn to the main question; how much nascent firm are hit by the founder dying. I start out by comparing the performance of firms where the founders subsequently die with the firms where the founders stay alive. To avoid that outliers drive the results, I winsorize the yearly OROA values at the 5 percent level.
Since I can track all companies through their first four years of operations, I first evaluate firm performance over a four-year horizon. To accommodate possible trends, I also evaluate firm performance over a six-year horizon. A six-year horizon reduces the total sample by 20 percent, to about 10,000 founders, as firms started up later than 2001 are then excluded from the sample.
A firm is defined to survive year t if it reports a profit and at least NOK 50,000 in sales in year t+1. OROA is computed using the average of assets at the end of year t − 1 and at the end of year t. The number of OROA observations in Panel B is 34,767 for d=0 and 501 for d=1. The number of OROA observations in Panel C is 50,944 for d=0 and 735 for d=1. Krone values are expressed in 2002 kroners.
Panel A reports the mean survival rates of the start-ups where the founder subsequently dies (the third column) with start-ups where the founder stays alive (the first column). Panel B reports size and sales after four years, and profitability during the first four years, broken down on founder type. Panel C reports the same as Panel B but on a six-year horizon. This first cut of the data does not suggest that the founder dying has any strong negative effects on the firm. For example, the 6-year firm survival rate is 60 percent for dead founders and 61 percent for alive founders. The start-ups of the dead founders have on average an OROA of 6 percent, while the average OROA of the alive founders is 8 percent. Using a t-test, neither of the difference in survival rates nor in OROA are significant at the 10 percent level.
Founders that die are older, wealthier, and start up larger firms. In the following regressions, I investigate firm performance controlling for founder and firm characteristics at the start-up date. The controls include the founder variables age, gender, previous income, and previous wealth, and the firm variables size (log equity at the start-up date), two-digit industry dummies, and start-up year.
The estimation method is probit in (1) and ordinary least squares in (2)-(5). The reported coefficient in (1) is the marginal effect evaluated at the mean of the other variables. Founder variables (age, gender, previous income, and previous wealth), firm variables (log equity, dummies for two-digit industry and for start-up year) are included but not reported. In the OROA regressions (5) and (10), firm age dummies are also included. Robust standard errors clustered at firm level in parenthesis. NOK 10 000 are added to previous income, previous wealth, assets and sales before taking logs. *** Significant at the 1 % level ** Significant at the 5 % level * Significant at the 10 % level
In the first four rows of Panel A, I regress measures of firm performance after four years on the explanatory variables, and in row (5), I use each individual-year as an independent observation. Across rows (1)-(5), neither the size or the significance level of the estimated coefficient suggest a clear effect of the founder dying. The only coefficient that is close to being significant is on survival. The estimated coefficient on survival suggests that firms where the founder dies has about 5 percentage points lower 4-year survival probability. In row (5), the relation between founder death and yearly profitability is weakly negative and insignificant. The results in Panel B speak to firm performance after six years of operations. The results are very similar to in Panel A, in that the only coefficient that is close to significant is on 6-year survival probability. The magnitude of the coefficient on survival is close to coefficient on 4-year survival; firms where the founder dies has about 6 percentage points lower survival probability. Since the results suggest very weak negative effects of the founder dying it seems unlikely that reverse causality, as discussed in Section 3.2, has any strong effects.
With less than two hundred dead founders, one might ask whether the statistical tests have sufficient power to reject the null hypothesis of no difference between the two groups. To illustrate, say that the true 4-year survival rate for the alive group of founders is 10 percentage points higher than the performance of the dead founders. Using the estimated standard deviations from Table 3, a one-sided 10 percent level t-test on differences in survival rate has 95 percent power, meaning that the null hypothesis will be rejected at the 10 percent level with a 95 percent probability. The corresponding power of a test of a 10 percentage points difference in OROA is 100 percent. Doing the same type of exercise, but now assuming that the true difference in survival rates is 5 percentage points, gives a test power of approximately 60 percent for survival and 95 percent for OROA. The joint power, i.e., the probability of rejecting both null hypothesis given that one of them were false would thus be higher than 95 percent. These calculations suggests that the lack of evidence of strong differences between the groups is not due to low power of the tests.
I have exposed these basic results to a variety of robustness checks, such as running the regressions using one sector industry codes, using sales growth and asset growth as dependent variables rather than sales and assets after 4 (6) years, using median regressions rather than ordinary least squares, including squared terms on the main explanatory variables such as wealth and income of the founder, and size of the firm, and excluded firms where a firm (rather than an individual) is the largest founder. The results were the same. I have also run the regressions on firm level rather than on founder level. In these regressions I kept only one yearly observation per firm and defined a death dummy that equals one if at least one of the founder of the firm dies. The estimated coefficients on the death dummy in these regressions were very similar to in Table 4.
4 Additional analysis
The basic analysis suggested that founder death has only a weak effect on the performance of the nascent firm. The current section considers two extensions of the basic analysis.
A possible reason for the weak results is that the analysis contains founders with a small ownership stake and that die "late" in the early stage of the firm’s life. In Section 4.1, I redo the analysis but now on majority founders that die before six years of operations. A different reason for the weak results could be that omitted variable bias skews the estimated coefficients towards zero. To assess this possibility, Section 4.2 considers within-firm effects of founder death, by analyzing changes in performance in a two-year time window around the founder death event.
4.1 Majority founders
The following table contains summary statistics on majority founders (who own at least 50 percent of the initial equity) that die before the end of the sixth year of operations.
Panel A contains information about the founders and Panel B about the firms they form. Of the 5 400 majority entrepreneurs, 64 (1.2%) die before the end of the sixth year of operations. The majority founders that die are older than the majority founders that stay alive, are more wealthy, and start up larger companies. Turning to performance, we have the following summary statistics.
A firm is defined to survive year t if it reports a profit and at least NOK 50,000 in sales in year t+1. OROA is computed using the average of assets at the end of year t − 1 and at the end of year t. The number of OROA observations in Panel B is 15,092 for d=0 and 170 for d=1. The number of OROA observations in Panel C is 22,163 for d=0 and 252 for d=1. Krone values are expressed in 2002 kroners.
5 Conclusion
While entrepreneurship is often considered central to economic performance and growth, we know little about what the most critical factors are for the performance of new firms. Using unique Norwegian data, I have analyzed the importance of the founder for firms in their nascency by exploiting variation in the occurrence of founder death.
The empirical results are quite striking. Both the cross-sectional and the within-firm analysis suggest that founder death has only a slight negative effect on firm performance. For example, after controlling for founder and firm characteristics at the start-up date, the predicted excess 4-year survival rate for nascent firms where the founder dies become less profitable, but only in a short time span around the founder death, which suggests that founder death incurs an adjustment cost rather than a permanent drop in profitability. For other variables such as sales or asset growth, I failed to find any effect of founder death. Hence for these infant firms, once they are set up the founder seems highly substitutable. These findings have two main implications. First, it seems that nascent firms are much less dependent on their founder than what one could expect. This is good news since firms where the founder is non-essential can be expected to have easier access to credit and be less exposed to bottlenecks due to limited capacity of the founder. Second, the findings have implications for how we view the entrepreneur. Although our results suggest that the founder’s role as a jockey to infant firms seems rather unimportant, the results should not be interpreted as the founder being an unimportant figure for the firm. Rather, the results suggest that the founder’s role as a creator and visionary prior to the start-up date is much more important than the role as a jockey.