1 Introduction
A large prior literature has investigated how government regulations, policies, and actions relate to entrepreneurship (see Blackburn and Schaper (
2016), for a summary). Within this literature, many studies have investigated the role of taxes and start-up costs (Gentry and Hubbard
2000; Djankov et al.
2002; Klapper et al.
2006; Cullen and Gordon
2007; Braunerhjelm and Eklund
2014; Block
2016). Both taxes and start-up costs can be directly influenced by government policy and are shown to have strong relationships with entrepreneurship rates. So far, no study has investigated the role of taxes and start-up costs with regard to innovative entrepreneurship. This is an important oversight, as we believe that the distinction between innovative and non-innovative entrepreneurship is an important one. While many policy-makers and scholars use the terms innovation and entrepreneurship interchangeably (Dreher and Gassebner
2013; Autio et al.
2014) and even try to stimulate one in the hope of getting more of the other, only very few entrepreneurs innovate (Reynolds et al.
2005; Block et al.
2017). Yet, prior research shows that, in particular, such innovative entrepreneurs are responsible for economic growth and development (Autio et al.
2014; Block et al.
2017).
We shall argue that start-up costs and taxes are associated in different ways with innovative entrepreneurship. Start-up costs such as notary charges or registration costs are
one-off costs that increase the barriers of entry into entrepreneurship. High start-up costs are usually associated with low entrepreneurship rates. Yet, high start-up costs may not only influence the quantity of entrepreneurship but also the quality and type of entrepreneurship. In fact, high start-up costs may lead to a positive selection of those individuals who are highly motivated and expect high incomes from entrepreneurship. Monteiro and Assunção (
2012), Branstetter et al. (
2014), and Rostam-Afschar (
2014) find that low start-up regulations attract low quality entrepreneurs with low expected returns. We argue that innovative entrepreneurs expecting high returns (Schumpeter
1934) are more willing, compared to other entrepreneurs, to incur high start-up costs. In addition, they are usually well positioned to attract external funding (Desai et al.
2003). Taxes, in turn, represent
recurring costs reducing the gains from innovation and entrepreneurial profit. They have a deterrent and discouraging effect particularly for risk-taking entrepreneurs with innovative ideas (Hansson
2012). Taxes reduce the expected return on innovation and, thus, discourage innovative entrepreneurship. High tax rates partially remove the “prize” of introducing a new product to the market, while entrepreneurs remain responsible and liable when their ideas fail (Gentry and Hubbard
2000; Cullen and Gordon
2007). In our investigations, we distinguish between income and corporate taxes and how they are associated with innovative entrepreneurship. While income taxes concern the income from unincorporated firms, corporate taxes refer to the income from incorporated firms (Cullen and Gordon
2007). Most innovative entrepreneurs register their businesses as corporations in order to grow faster and attract external funding. Thus, there is reason to believe that corporate taxes should have a stronger relationship with innovative entrepreneurship than income taxes have.
To investigate how start-up costs and taxes relate to innovative entrepreneurship, we use the Global Entrepreneurship Monitor (GEM) dataset, which is composed of 632,116 individuals, including 43,223 entrepreneurs from 53 countries from 2004 to 2011. Conducting both cross-sectional and longitudinal regressions, we can partially confirm our predictions. Corporate taxes show a negative relationship with innovative entrepreneurship, whereas income taxes seem to show no relationship. High start-up costs seem to have a positive relationship with innovative entrepreneurship, although this finding seems to only hold true in cross-sectional and not in longitudinal investigations.
The rest of the paper is structured as follows: the next section develops hypotheses on how start-up costs and taxes relate to innovative entrepreneurship. Then, we describe our data sources, variables, and methods. Subsequently, we present our main results, together with a number of robustness checks. In the final section, we present the main conclusions, implications, and limitations of the study.
3 Data and method
3.1 Data sources
We use both individual- and country-level data. Our individual-level data are from entrepreneurs who have participated in the Adult Population Survey (APS) of the Global Entrepreneurship Monitor (GEM). The data covers 53 countries from 2004 to 2011. GEM is the largest cross-country study of entrepreneurial activity, aspirations, and attitudes (Reynolds et al.
2005). It collects data on individuals about different aspects of their entrepreneurial activity, such as the innovativeness of their ventures, as well as their personal start-up motivations, entrepreneurial ambitions, and human capital characteristics, which make the GEM data suitable for use in our research.
At the country level, we use the World Bank Doing Business (WBDB) database and the World Competitiveness Yearbook (WCY) for information on start-up costs and tax rates. The WBDB database contains several measures of business regulations and their enforcement for 155 countries from 2004 to the present. These measures demonstrate the regulatory expenses and procedures of undertaking business and have been used in prior research to analyze regulatory influences on the productivity and growth of entrepreneurs (e.g., Dreher and Gassebner
2013; Braunerhjelm and Eklund
2014). We use the WCY for information about corporate and personal income tax rates, as well as for some control variables (e.g., GDP growth, GDP per capita). WCY includes annual data for 18 years for more than 50 countries that participate in the executive survey conducted by the IMD World Competitiveness Center. Several previous studies have used WCY measures to study the impact of country-level factors on entrepreneurship (e.g., van Stel et al.
2007; Hessels et al.
2008).
3.2 Sample
The total GEM sample for 2004 to 2011 is composed of 689,399 individuals aged 18 to 64, including (early-stage and established) entrepreneurs, employees, unemployed individuals, students, and retirees. Of these, 57,796 persons are early-stage entrepreneurs (8.4%) (i.e., individuals who are setting up businesses), as well as entrepreneurs who have started their own business in the last 42 months. For the purpose of this study, we focus on whether such early-stage entrepreneurs (that we henceforth call “entrepreneurs”) are innovative (see also the description of variables as follows).
Table
A.1 in the electronic supplementary material shows the number of individuals and entrepreneurs per country and distinguishes between innovative and non-innovative entrepreneurs.
3.3 Variables (individual-level regressions)
Our dependent variable is innovative entrepreneur. This variable is measured at the individual level, based on a question in the GEM survey asking entrepreneurs whether they provide a new product or service to the market. The variable is a dummy variable that takes the value one when the product or service offered is perceived by the entrepreneur to be new to customers and takes the value zero otherwise.
Our main independent variables are
start-up costs and
corporate and
income tax, which are measured at the country level. Start-up costs reflect the expenses required by law to register a new venture in a country. Tax refers to the (logarithm of) corporate and personal income tax rates in a country. Table
1 provides a more detailed overview and description of our independent variables. Tables
A.1 and
A.2 in the electronic supplementary material provide more insights into the values of the dependent and independent variables per country and per year, respectively.
1Table 1
Data sources for the main country-level variables
Start-up costs | The average costs of obtaining legal status to operate a firm, measured as a percentage of per capita income. It contains all recognizable official expenses such as fees, costs of forms and procedures, photocopies, fiscal stamps, and legal and notary charges | WBDB |
Corporate tax rate | Maximum corporate tax rate, calculated on profit before tax | WCY |
Income tax rate | Maximum personal income tax rate as a percent of the individual’s income | WCY |
One might ask why the share of innovative entrepreneurship is 41.8% in Chile or 33.9% in Jordan while it is, for example, 11.6% in the UK. The explanation is that advancement of an economy does not necessarily mean that its entrepreneurs are innovative. Entrepreneurs can still be mainly imitative or adopters in an advanced economy so that the ratio of innovative entrepreneurs would be low. In developing countries, there is more room to innovate in the market as many ideas are not yet tested and can be learned/copied from more developed countries. Hence, ideas in developing countries might not be globally innovative, but they are at least new in their local market.
In addition, we add to the regression model a number of individual and country-level control variables that are common determinants of innovative entrepreneurship (Acs and Audretsch
1987; Koellinger
2008; Anokhin and Schulze
2009; Autio et al.
2014; Fritsch and Wyrwich
2018). At the individual level, the following variables are included: formal education (a dummy variable that indicates whether entrepreneurs have a university education), entrepreneurial networks (a dummy variable indicating whether the entrepreneur knows someone personally who started a new business in the last 2 years), perception of entrepreneurial skills (a dummy variable indicating whether the entrepreneur perceives him- or herself to have relevant skills, knowledge, and experience for setting up a business), recent prior entrepreneurship experience (a dummy variable that indicates whether someone has quit as an entrepreneur in the past 12 months), established business ownership (a dummy variable that equals one if the respondent owns a business older than 42 months), and gender (a dummy variable that equals one for males). Age and age squared are also included. We further added “year” and “industry” as dummy variables to the regression model. The following industries are distinguished: business services (financial intermediation, real estate, renting, and business activities), consumer-oriented services (hotels and restaurants, other services), extractive industries (agriculture, fishing, mining, and quarrying), and transforming (manufacturing, electricity, gas, water, construction, trade, repairs, transportation, storage, and communication). At the country level, we include GDP growth and the (logarithm of) GDP per capita, which are both taken from the WCY database. After removing observations with missing values, we retained a sample of 632,116 individuals including 43,223 entrepreneurs.
3.4 Regression methods
We analyze our data at both individual and country levels. At the individual level, we run Heckman probit regressions; at the country level, we employ fixed effect panel regressions and (Bayesian) first-difference regressions.
3.4.1 Individual-level cross-sectional regression: Heckman probit model
Our dependent variable
innovative entrepreneur is binary, and we use various probit regressions. We cluster the individual-level data by countries to avoid underestimating standard errors (Huber and Stanig
2011). Furthermore, we employ a Heckman probit model to reduce a potential selection bias when assessing the influence of start-up costs and taxes on the likelihood for entrepreneurs to be innovative. This is mainly because start-up costs and taxes could affect the entry of individuals into entrepreneurship (Gentry and Hubbard
2000; Djankov et al.
2002; Wen and Gordon
2014), in addition to their effect on innovative entrepreneurship. As such, trying to estimate the influence of start-up costs and taxes on an entrepreneur’s likelihood to innovate may lead to biased estimators when such potential selection bias is not taken into account. Heckman correction (probit) models are used to address this methodological issue. Additionally, we test for the presence of a selection bias through likelihood ratio tests: The likelihood ratio test of rho (which compares the log likelihoods of the selection plus outcome models with the log likelihood of the probit model with sample selection) confirms that a Heckman model is indeed necessary (Table
2).
Table 2
Heckman probit regression (dependent variable: innovative entrepreneur)
Predicted probabilities | 0.11 | | 0.10 | | 0.10 | | 0.10 | |
| Marginal effect | t statistics | t statistics | Marginal effect | t statistics | t statistics | Marginal effect | t statistics | t statistics | Marginal effect | t statistics | t statistics |
Country-level variables |
Start-up costs | 0.001 | 2.26** | − 0.60 | | | | | | | 0.001 | 2.05** | − 0.90 |
Corporate tax rate (log) | | | | − 0.010 | − 5.37*** | − 0.73 | | | | − 0.009 | − 2.29** | 0.77 |
Income tax rate (log) | | | | | | | − 0.017 | − 2.01** | − 1.23 | − 0.001 | − 0.13 | − 1.29 |
GDP per capita (log) | − 0.002 | − 0.20 | − 4.80*** | − 0.014 | − 1.53 | − 4.11*** | − 0.007 | − 0.82 | − 3.95*** | − 0.002 | − 0.21 | − 5.15*** |
GDP growth rate | 0.003 | 1.66* | 0.86 | 0.002 | 1.17 | 0.92 | 0.002 | 1.47 | 0.85 | 0.003 | 1.83* | 0.79 |
Individual-level control variables |
High level of education | 0.019 | 4.42*** | 0.48 | 0.017 | 3.99*** | 0.43 | 0.018 | 4.22*** | 0.33 | 0.018 | 4.35*** | 0.36 |
Entrepreneurial networks | 0.013 | 1.85* | 13.50*** | 0.014 | 2.09** | 13.32*** | 0.013 | 1.94* | 13.30*** | 0.012 | 1.82* | 13.68*** |
Perceived entrepreneurial sills | 0.052 | 3.10*** | 23.01*** | 0.054 | 3.46*** | 22.67*** | 0.054 | 3.36*** | 22.43*** | 0.050 | 3.10*** | 22.91*** |
Gender (male = 1) | 0.004 | 0.47 | 4.36*** | 0.004 | 0.49 | 4.14*** | 0.005 | 0.50 | 4.01*** | 0.003 | 0.33 | 4.45*** |
Age | − 0.001 | − 0.65 | 2.35** | − 0.001 | − 0.49 | 2.34** | − 0.001 | − 0.53 | 2.39** | − 0.001 | − 0.80 | 2.39** |
Age square | 0.00001 | 0.47 | − 3.52*** | 0.000003 | 0.31 | − 3.48*** | 0.000006 | 0.33 | − 3.56*** | 0.00001 | 0.64 | − 3.59*** |
Established business ownership | 0.022 | 1.21 | − 6.80*** | 0.021 | 1.19 | − 6.79*** | 0.023 | 1.23 | − 6.79*** | 0.021 | 1.19 | − 6.83*** |
Prior entrepreneurship experience | 0.022 | 2.96*** | 11.63*** | 0.021 | 2.96*** | 11.72*** | 0.021 | 2.89*** | 11.98*** | 0.022 | 3.22*** | 12.42*** |
Industry dummies | | Yes | | | Yes | | | Yes | | | Yes | |
Employment status dummies | | | Yes | | | Yes | | | Yes | | | Yes |
Year dummies | | Yes | Yes | | Yes | Yes | | Yes | Yes | | Yes | Yes |
Constant | | − 3.56*** | − 0.24 | | − 1.99** | − 0.57 | | − 2.51** | − 0.59 | | − 2.98*** | − 0.10 |
Sample size | | 43,223 | 632,116 | | 43,223 | 632,116 | | 43,223 | 632,116 | | 43,223 | 632,116 |
Number of countries | | 53 | | 53 | | 53 | | 53 |
Likelihood ratio test (rho = 0) (prob > chi2) | | *** | | *** | | *** | | *** |
The Heckman model has one selection and one outcome equation. The selection equation (the first stage) estimates entry into entrepreneurship, including all the abovementioned individual- and country-level predictors. We also add the individuals’ employment status (dummy variables indicating whether someone is employed, unemployed, a student, or a retiree) to the selection equation. The outcome equation (i.e., the second stage) estimates whether or not an entrepreneur is innovative. The Heckman probit model is similar to other Heckman correction models (Heckman
1976,
1979; Puhani
2000) in how it corrects for selection bias, except that the outcome-dependent variable is a dummy variable and not a metric variable. The main control variables correspond to Braunerhjelm and Eklund (
2014).
3.4.2 Country-level longitudinal regressions
Next to the individual level, we analyze our data in an aggregated form at the country level and thereby employ a longitudinal perspective. To understand the effect of our predictors on the outcome variable
within each country, we estimate panel data regressions using data from 2004 to 2011. The dependent variable is the
share of innovative entrepreneurship in a particular country in a particular year. The variable is calculated from the individual-level variable innovative entrepreneur and measures the aggregate number of innovative entrepreneurs as a percentage of all entrepreneurs per country. We included country-level main and control variables in line with Braunerhjelm and Eklund (
2014). Such panel data investigations are only possible at the country and not at the individual level, because the individual-level GEM dataset is not a panel dataset; every year, it uses a different or new sample of individuals (Reynolds et al.
2005).
We analyze our country-level dataset through both a fixed effect panel data regression and a first-difference regression. The first-difference regression is estimated in a Bayesian way. The main reason is that first differencing reduces the number of observations per country substantially making classical null hypothesis significance testing difficult. Bayesian analysis, in turn, exploits fully the information provided in small samples. It is able to investigate the relationships between variables using small samples; sample size does not influence its ability to test whether a particular relationship is “true” or not (see Block et al. (
2014), for a deeper discussion of Bayesian analysis).
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.