New evidence on accelerator performance based on funding and location

2.1 From business incubators to seed accelerators

BIs first appeared in the 1980s and underwent rapid growth until the late 1990s. During this period in Europe, most BIs were integrated into the European BIC network. However, this growth slowed in the years following the burst of the internet bubble in 2000/2001. Simultaneously, a new form of BI, the SA, emerged as an important springboard for local entrepreneurs. SAs support the generation and growth of innovative technology-based firms, specialising in software- and Internet-related businesses.

New BI models providing investment and assistance in pre-seed stages have emerged and blossomed in recent years, first in the United States. These models then spread to Europe before rapidly expanding to other parts of the world. This new generation of BIs aims to help and accelerate the creation of innovative companies, from the conception of the initial idea to its initial stages in the market. To do so, these new BIs began providing important business assistance, resources, funding and networking opportunities, and they soon came to be known as SAs or SAPs. These SAPs are described as fast-track processes for new venture development, and they are offered in return for a percentage of equity in the newly established company. The return on investment and profits are made when the SA sells its shares to other investors through exit operations.

The definition of an SA amongst practitioners remains inconsistent. Some BIs refer to themselves as SAs, capitalising on the current hype surrounding SAs. In contrast, others that meet the formal definition of an SA still refer to themselves as BIs (Hochberg, 2016). Although SAs were conceived with the same business structure and philosophy as BIs, some significant differences have emerged. Thus, an SA does indeed follow a specific organisational model in its own right.

The majority of SAs provide an initial seed investment in exchange for accommodation and services (Bliemel et al., 2016; Pauwels et al., 2016). Dempwolf et al. (2014) describe four subtypes of accelerators: innovation, social, university and corporate. All of these accelerators are consistent with Cohen and Hochberg's (2014) definition. Innovation accelerators are the best-known form of SAs. Examples include Techstars and Y-Combinator. Innovation accelerators are still the most widespread kinds of accelerators. Social accelerators have been gaining increasing acceptance since the launch of social entrepreneurship programmes such as the Global Social Venture Competition. Some universities back entrepreneurship programmes linked to hosting entrepreneurs at their own accelerator facilities (Shah and Pahnke, 2014). Finally, corporate accelerators have emerged since 2014 to provide corporations with their own innovation ecosystems in pursuit of the goal of acquiring client start-ups (Page and Garbuio, 2016).

SAs can be described as a more advanced version of BIs (Pauwels et al., 2016). They usually launch an open application process where anyone with a business idea can apply. The best projects are then chosen and enrolled in an SAP. The programme culminates with the presentation of the most successful projects to investors in a public pitch event known as “demo day” (Figure 1).

2.2 Accelerator performance indicators

There has been limited research on accelerators, primarily because of the newness of the phenomenon and limited data availability (Stayton and Mangematin, 2019). Challenges in finding data are considerable and affect researchers' ability to conduct rigorous empirical analyses and performance evaluations. Accelerators have quickly proliferated, but there is a general absence of large-scale representative public databases covering accelerator programmes. This lack of such databases prevents researchers from evaluating the impact of these programmes (Hochberg, 2016).

As Cohen and Hochberg (2014) noted, the scarcity of studies on the performance of accelerators makes it unclear how effective they are. Indeed, little research has explored, even at a descriptive level, the effectiveness of SAPs or the reasons for better or worse results. The measures that should be used to quantify the effectiveness and success of these initiatives are not yet clear.

Much of the limited research on accelerators to date falls into one of the following four categories: (1) conceptual descriptions of the accelerator model (Cohen and Hochberg, 2014; Dempwolf et al., 2014; Hochberg (2016); (2) qualitative assessment of how accelerators may serve to accelerate start-ups (Radojevich-Kelley and Hoffman, 2012; Cohen, 2013; Pauwels et al., 2016; Cohen et al., 2018); (3) empirical studies to assess whether accelerators positively affect the outcomes of the companies that participate in their programmes (Smith and Hannigan, 2015; Cohen et al., 2019; Fehder and Hochberg, 2019; Hallen et al., 2019); and (4) empirical studies to assess whether accelerators have a negative or inconclusive effect on the outcomes of accelerated start-ups (Smith et al., 2013; Gonzalez-Uribe and Leatherbee, 2017; Yu, 2019). Table 1 summarises accelerator studies in terms of the perspective, focus of the study and main findings.

Accelerators have attracted the attention of researchers because they provide a window into early-stage entrepreneurship, which has historically been difficult to observe (Aldrich and Yang, 2012). However, the existing research is highly fragmented and has yet to form into a robust corpus of knowledge built around a core framework with a shared understanding of questions, methodologies and knowledge gaps (Cohen et al., 2019).

Few studies have used quantitative analyses to measure the impact of a global set of SAs on the performance of their accelerated start-ups. Gonzalez-Uribe and Leatherbee (2017) used a sample of 3,258 applicants to an individual accelerator programme (Start-up Chile) and found that access to certain basic services, such as the coworking space provided by the programme, had a limited impact on the future performance of Start-up Chile graduates. Cohen et al. (2019) used a sample of 146 US accelerators and 100 interviews to confirm a connection between SAP design and the performance of the accelerated start-ups. Fehder and Hochberg (2019) examined a list of 59 accelerators founded between 2005 and 2013. They concluded that the arrival of an accelerator is associated with a significant increase in the volume of seed and early-stage deals, driven by outside investor groups and the emergence of new local early-stage investors. Hallen et al. (2019) used a matched sample from four cohorts of eight top US accelerator programmes to compare treated and untreated start-ups. They found evidence that accelerators substantially aid and accelerate venture development. “Novel learning” was observed to be the key driver of the accelerator effects. Finally, Yu (2019) compared start-ups affiliated with 13 accelerator programmes to (non-accelerated) start-ups backed by venture capitalists (VCs). The findings suggest that new ventures admitted to accelerators are less likely to reach key milestones. In contrast to these recent studies, our subject of analysis is the accelerator itself rather than the hosted companies.

2.3 The accelerator's location

Analysis at the country level has attracted ample attention in the BI literature. Many BI studies have focused on developed countries (Chen, 2009), principally the United States (Mian, 1997; Rothaermel and Thrusby, 2005) and European countries (CSES, 2002; Clarysee et al., 2005). Many BI studies provide comparisons between BI activity in these two markets (Aerts et al., 2007). Other studies have focused on particular countries such as the United Kingdom (Soetanto and Jack, 2013), Finland (Abetti, 2004), Sweden (Lindelof and Lofsten, 2002), Germany (Schwartz and Hornych, 2008, 2010; Schwartz, 2013), Italy (Colombo and Delmastro, 2002), Israel (Rothschild and Darr, 2005), Spain (Peña, 2004) and Portugal (Ratinho and Henriques, 2010). Others have examined developing countries (Akçomak, 2009) such as Brazil (Etzkowitz et al., 2005), the Gulf Cooperation Council countries (Mubaraki and Busler, 2010) and Turkey (Özdemir and Sehitôglu, 2013). Whilst there is abundant coverage of BIs, broad studies based on worldwide surveys of incubators are practically non-existent.

Europe and the United States host a comparable number of start-up programmes per capita. In Europe, the number of SAs has increased dramatically since the start of the financial crisis in 2007. Between 2007 and 2013 the number rose by almost 400% (Salido et al., 2014). SAs have emerged as a plausible way of creating job opportunities and technology-based businesses, revealing innovative ways to offer products that can conquer the international market and grow without the need for huge injections of capital (Christiansen, 2009; Cohen, 2013).

The accelerator phenomenon was born in the United States, and despite extensive globalisation, it is still the undisputed leader in terms of the number of acceleration programmes. Of the top 20 SAPs, 15 are located in the United States. Silicon Valley pioneers new forms of the original SA model. The United States also plays a leading role in the development of university-driven accelerators. Start-x (Stanford) and Skydeck (UC Berkeley) offer notable examples. The same is true of corporate accelerators, which are now flourishing around the world. The purpose of our first hypothesis is to test the extent to which the United States leads in SAs.

2.4 Investment in SAs

A key indicator of the prospects and expectations of most high-tech companies, especially start-ups, is the presence of funding by external investors, primarily VCs. SAPs are expected to make their hosted start-ups more appealing to VCs and business angels. Similarly, firms that succeed in attracting external investors are expected to have more chances of survival and growth. These better chances are because there is generally a positive association between VC finance and growth, although this view is not unanimous (Bottazzi and Da Rin, 2002). As noted by Bertoni et al. (2011), most studies of VCs suffer from a bias because they consider only IPO firms. This approach leaves privately held firms unstudied, the vast majority of which are start-ups.

Accelerated start-ups have better chances of attracting VC investment and closing investment rounds if they adapt to the VCs' preferences for investing in firms whose founders have management, educational and professional experience (Bertoni et al., 2011; Colombo and Grilli, 2010; Puri and Zarutskie, 2008). These are precisely the areas where most entrepreneurs improve during SAPs.

Firms with VC investment tend to excel over others in most performance indicators (Gompers and Lerner, 2001; Dennis, 2004). In the context of start-ups, closing successive investment rounds is vital and offers the route to a marketable solution and the gateway to customers. Prestigious VC funds provide extra marketplace credibility to participating firms and greater attractiveness to new investors. In addition, these start-ups seem to have easier access to valuable skills and resources (Colombo et al., 2006; Hsu, 2006) and have more chance to grow in employment terms (Bertoni et al., 2011). Davila et al. (2003) performed a broad study of 494 Silicon Valley start-ups, concluding that the quality, reputation and credibility of new ventures is enhanced when an investment round is undertaken by a VC.

In accordance with these findings and conclusions from the literature, we assume that receiving sufficient investment from a VC by closing an investment round above US$ 1 million improves start-ups' expectations and growth prospects. Most entrepreneurs starting ventures in Silicon Valley share the view that closing at least an A round of investment (US$1–5 million) and, ideally, a B round (over US$5 million) is the main success indicator. For practically all such start-ups, this amount is viewed as sufficient to keep pace and fuel their growth. The arguments in this section lead us to formulate our second hypothesis.

3.1 Analysis model

The previous literature review reveals that, despite the vast number of empirical studies assessing the impact of BIs, there is a lack of consensus on BI performance measurement. In addition, the absence of a single standard method makes any analysis of BI efficiency and performance even more difficult (Phan et al., 2005; Bergek and Norrman, 2008; Schwartz and Gothner, 2009; Ratinho and Henriques, 2010). Further, few studies have used a robust quantitative approach to assess the economic effects of incubator organisations. In addition, most results and findings are inconclusive and somewhat contradictory. Table 2 summarises some of the most significant variables that have previously been used in the BI literature.

Given the lack of specific variables for measuring SAPs and in light of the BI literature review in the previous section, we propose a model for measuring SA performance, with variables grouped into three categories.

1. Size. Variables in this category provide quantitative information regarding the actual size of the accelerators: (1) Total funding: total amount of capital invested in the participating companies; (2) Total employees: total number of employees in the participating companies; (3) Total rounds: total number of investment rounds; (4) Total companies: total number of accelerated companies in each accelerator.

2. Location and age. This category comprises two typical control variables: (1) Country: location of the accelerator (United States or elsewhere); (2) Founding year: period in which the accelerator started to operate (1995–2000, 2001–2005, 2006–2010 or 2011–2014).

3. Performance ratios. Indicators and ratios suggested by Crunchbase (2018): (1) Total exits: amount of capital obtained by the accelerator through the exit of participating companies. This variable is only available for accelerators that have exited companies; (2) Average total exits per company (total exits/total companies); (3) ROI (return on investment) factor (total exits/total funding) × 100, which reflects the return on investment by the accelerator through company exits; (4) Average total funding per company (total funding/total companies); (5) Average total investment rounds per company (total rounds/total companies); (6) Average employees per company (total employees/total companies).

3.2 Data

One of the main limitations to increasing knowledge about SAs lies in the absence of large-scale representative databases that include data on programme features and the companies that enter and graduate from the programmes (Cohen and Hochberg, 2014). In accordance with the accelerator definition used in this study and to ensure a certain degree of homogeneity, we limited the type of SAs to those that meet the following selection criteria: (1) located in the United States and elsewhere, (2) at least four years old, (3) take equity in exchange for investment, and (4) are not mostly funded by private investors. An initial set of 191 SAs worldwide met these selection criteria. Of these, 100 were included in Seed-DB [1] an online accelerator database that probably represents the largest public repository of accelerators and graduate data (Hochberg, 2016).

The other 91 were hand-collected from crunchbase, which is an open source database with partnerships with more than 400 venture capital firms, accelerators, incubators and angel groups to ensure the accuracy of the data (Yu, 2019). Crunchbase tends to have more early-stage transactions than similar databases, which makes it ideal for hand-collecting data on the companies in our sample. Then, we used AngelList and LinkedIn profiles for verification purposes.

The presence of missing data for some variables forced us to delete some SAs initially exported from the Seed-DB database. The final sample consisted of 116 SAs, of which 72 were located in the United States and the remaining 44 in other countries.

4.1 Descriptive analysis

Total funding ranged from a minimum of US$ 9,000 to a maximum of US$ 2.2 billion. In addition, 62.93% of accelerators invested more than US$ 1 million in their accelerated companies.

SAs have not yet excelled as employment seedbeds. Only one had generated more than 1,000 jobs in participating companies, and over 80% of SAs had not yet created 100 new jobs. Only 25 accelerators participated in 10 or more rounds. In terms of the number of participating companies, 46 accelerators hosted 20 or more start-ups and 10 hosted more than 50. The largest accelerator supported 585 new ventures.

As expected, younger SAs hosted fewer companies, created fewer jobs, generated less total investment and completed fewer total rounds. When the accelerator had been operating for four years, the differences in terms of size indicators tended to grow exponentially, as shown in Table 3.

The location and age of the accelerators are of interest. The accelerators were grouped into four age intervals based on their founding year: 1995–2000, 2001–2005, 2006–2010 or 2011–2014. Our data confirm the young status of the SA phenomenon, with 91.38% of SAs founded from 2006 onwards (Table 4).

In terms of performance and effectiveness indicators, the most significant profitability ratio (ROI) was only available for 19 SAs.

4.2 Results of bivariate and multivariate analyses

Bivariate and multivariate analyses were used to test the two hypotheses. The bivariate analysis provided statistically significant results regarding the differences between two groups based on a single variable.

The existence or absence of significant differences between groups of SAs in terms of the variables in the model was verified using the non-parametric Kruskal–Wallis* or Mann–Whitney U** test. The required level of significance in the comparisons was 95%.

For testing Hypothesis 1, a bivariate analysis was conducted using country of location of the SAs as the grouping variable. US accelerators (72 in total) were thus distinguished from non-US accelerators (44 in total). Table 5 summarises the results of this bivariate analysis. We observed statistically significant differences based on country of origin for the following variables: total rounds (category 1), average total funding per company (category 3), average employees per company (category 3) and average total rounds per company (category 3). All four variables had higher values for the SAs located in the United States. Three of the variables corresponded to category 3 (performance ratios), which indicates that levels of profitability and efficiency are higher in US accelerators.

Next, to test Hypothesis 1 with greater precision, we ran a binomial model with country as the dependent variable. This variable took a value of 1 if the accelerator was located in the United States and 0 if the accelerator was located elsewhere. The intrinsic features and nature of the data made binomial logistic regression models suitable.

From the initial set of six variables, those used in the regression model were selected using the stepwise regression method. After each variable was added, all candidate variables in the model were checked to observe whether their significance had been reduced below the specified tolerance level. The Akaike information criterion (AIC) was employed to compare the different models. The model with the lowest AIC comprised only two independent variables:

The results are displayed in Table 6, with the estimated coefficient and the standard error. In this model, the only significant variable was average total funding per company. The model fit was satisfactory, with an AIC value of 136.12.

Average total funding was the only variable identified by both the bivariate and multivariate analyses. It is therefore considered the key component characterising SAs located in the United States.

To test Hypothesis 2, we applied a Tweedie distribution for generalised linear models (GLMs; tweedie), with the logarithm of the average total rounds per company as the dependent variable. To check the normality of the continuous variables, a Shapiro–Wilk test was run. All the p-values were greater than the significance level of 0.05, which implied that the variables did not follow a normal distribution.

When running GLMs, several models are typically feasible and valid. Three GLMs models were run with the average total rounds per company as the dependent variable and with the following independent variables:

1. Model 1: founding year, country, average total funding, average employees and total funding

2. Model 2: total companies, total exits, total funding, total employees and country

3. Model 3: total companies, % multiplicator factor, average total funding, average total exits, average employees and country.

After running all the models, the best model – and the one that was selected – was based on Model 1. It comprised three significant independent variables: average funding, average employees and total funding. Consequently, one key finding is that the SAs that close most rounds of investment per company are those that have a higher amount of funding per company, a higher average number of employees per company and a larger amount of funding being raised from investors.

Accordingly, as stated in Table 7, these are the three key factors that SAs should prioritise to outperform others in terms of ability to close more investment rounds for start-ups participating in their programmes.