The question is deceptively simple: do countries produce more startups because their corporate law gives founders and investors greater contractual freedom? This visualisation explores a more cautious version of that question — testing whether broad legal and regulatory proxies explain new-business creation once entry costs, regulatory quality, and innovation conditions enter the picture. The exercise uses available cross-country proxies, not hand-coded measures of statutory flexibility. The findings are associations, not causal proof. But the patterns are consistent enough to prompt a harder question about where the weight of explanation actually lies.

Scatter plot of 24 advanced economies. Entry friction (x-axis, higher = harder to start a business) against new business density — new limited-liability registrations per 1,000 working-age people (y-axis). Estonia is the clear outlier with very low entry friction and by far the highest startup density at 24.7. Japan sits at the opposite extreme with high entry friction and very low startup density at 0.41. New Zealand, UK, and Canada cluster in the low-friction, higher-density area. Germany, Austria, and Poland cluster in the high-friction, low-density area.

Easy In, More Out

Entry friction vs. new business density — 24 advanced economies, c. 2019. New business density = new limited-liability registrations per 1,000 working-age people.

24Countries in sample
24.7Estonia — highest startup density
0.41Japan — lowest startup density
−0.69Correlation: friction × density

New business density vs. entry friction (standardised), 24 countries

Entry friction proxy: inverse of the World Bank Doing Business Starting a Business Score (2019), standardised. Higher values on the x-axis = harder to register a business.
Note on Japan: Japan's entry friction score (86.1) is similar to Italy (86.8) and Spain (86.9) yet its startup density (0.41) is less than one seventh of theirs — illustrating why a simple bivariate relationship does not tell the full story. The multivariate regression in Tab 3 controls for this.
Source: World Bank Entrepreneurship Database (IC.BUS.NDNS.ZS); World Bank Doing Business (IC.REG.STRT.BUS.DFRN). Preferred year 2019; fallback to nearest available year.

The pattern. Across 24 advanced economies, countries where it is administratively harder to register a company tend to create fewer new businesses per head of working-age population. The relationship is not mechanical — other forces are clearly at work — but it is visible and consistent. Estonia sits at one extreme: very low registration friction and by far the highest startup density in the sample. Japan, Germany, and Austria sit near the other end: higher administrative barriers, very low new-business creation per capita. The pattern holds across both common law and civil law countries.

What this chart cannot show: whether any particular policy would reduce this friction in any specific context, or whether the companies being registered are innovative startups or routine service firms. Those questions require a sharper research design.
Scope and limitations

Sample. Advanced economies in Europe plus comparable OECD-style jurisdictions where the relevant public indicators are available. The United States is included in the underlying data file where available, but the World Bank startup-density dependent variable is missing for the US and it is excluded from the regression sample. Working sample in the regressions: 22–24 countries depending on missing values.

Period. 2019 where possible, with fallback to the nearest available year (2020, 2018, then 2017). This keeps the exercise pre-pandemic and avoids mixing the startup shock of 2020 with structural country differences.

Entry friction proxy. Built from the inverse of the World Bank Doing Business Starting a Business Score. Because the original score is higher when starting a business is easier, it is reversed and standardised; in the regression, higher entry friction therefore means starting a business is harder. This proxy reflects domestic business-registration conditions in each country. It is not a measure of cross-border incorporation ease and cannot speak to the likely effects of any specific policy proposal.

Investor-protection / governance proxy. Built by averaging standardised World Bank Doing Business measures on minority-investor protection, shareholder rights, ownership and control, and corporate transparency. This is a broad governance proxy — close to the substantive content of company law — not a pure index of freedom of contract in corporate charters.

This visualisation is a first-pass empirical exercise, not a causal identification strategy. Legal proxies are derived from available indicators, not hand-coded indices of mandatory versus default rules. Findings should be read as a prompt for further research — specifically, a hand-coded corporate-law flexibility index tested against entry costs and institutional controls — rather than as definitive evidence about the value of corporate-law design.

Data sources. World Bank Entrepreneurship Database (IC.BUS.NDNS.ZS); World Bank Doing Business (IC.REG.STRT.BUS.DFRN and legacy minority-investor/governance indicators); World Bank Worldwide Governance Indicators; World Development Indicators. Analysis: Prof. Marco Ventoruzzo, supported by OpenAI Codex for data extraction and calculation.

What Associates with Startup Creation?

Bivariate correlations with log new business density — 24 countries. Variables ranked by absolute correlation strength.

Legal / regulatory proxy Economic / institutional control

Correlation with log new business density (n = 24)

Bivariate (Pearson) correlations. A negative correlation means higher values of that variable associate with fewer new businesses; a positive correlation means the opposite. Simple correlations do not control for anything else — the regression in Tab 3 does. Orange bars = legal/regulatory proxies; blue bars = economic and institutional controls.

Entry friction leads. The ease of registering a business has the strongest simple association with startup density in this sample (r = −0.693): where it is harder to start a business, countries tend to see fewer new registrations per head. Regulatory quality — the perceived competence and predictability of a country's regulatory environment — also associates positively (r = 0.478).

The investor-protection and governance proxy sits near the bottom (r = 0.118). This variable captures elements of company law closest to its substantive governance content — minority-investor rights, shareholder protections, ownership and transparency rules. In the raw data, it associates with startup creation only weakly. A simple correlation controls for nothing else; the regression in the next tab asks whether the pattern survives with multiple factors considered together.
Correlation is not causation. These are associations in a small cross-country snapshot. Variables that move together with startup density in the raw data may do so because of a third factor. The regression in Tab 3 is a partial control for this, but it cannot eliminate the possibility of omitted variables.

The Full Picture

Standardised OLS coefficients — parsimonious multivariate model, n = 24. Each bar shows the estimated change in log startup density associated with a one-standard-deviation shift in that variable, holding all others constant.

Significance: *** p < 0.01 ** p < 0.05 * p < 0.10 n.s. not significant

Standardised OLS coefficients: what explains startup density once everything is in the model?

OLS multivariate regression. Dependent variable: ln(1 + new business density). Explanatory variables standardised (mean 0, s.d. 1). Bars coloured by statistical significance level. Whiskers show ±1 standard error.
R² = 0.616  |  Adjusted R² = 0.509  |  F-statistic = 5.776 (p = 0.002)  |  n = 24.
* p < 0.10   ** p < 0.05   *** p < 0.01

Two variables show statistically meaningful associations. Entry friction (β = −0.376, p < 0.01) remains negatively associated with startup density after controlling for innovation, productivity, and institutional quality. Regulatory quality (β = +0.297, p < 0.05) is also positive and meaningful — countries where regulation is perceived as competent and predictable tend to have higher startup creation. This is a measure of how the regulatory environment functions in practice, not what company law statutes prescribe.

The investor-protection and governance proxy contributes essentially nothing in this model (β = 0.006, p = 0.96). That variable is the one closest to the substantive governance content of corporate law — minority-investor rights, shareholder protections, ownership transparency. Once entry costs and institutional quality are already in the model, it adds no explanatory power.

That does not settle any debate about corporate law's importance. It does shift the burden of proof: in this cross-country snapshot, the measurable governance content of company law does not appear to explain startup density once the operating environment around firms is accounted for. A hand-coded index of statutory flexibility — distinguishing mandatory from default rules across jurisdictions — would be the right next step.
What this does not prove. This is not a causal design. It does not identify a legal reform shock, and it does not distinguish innovative startups from ordinary new limited-liability registrations. It also uses legal proxies available cross-country, not hand-coded measures of mandatory versus default corporate-law rules. The sample is small (n = 24) and the results should be read as a disciplined prompt for debate and a basis for a sharper legal coding project.
The headline finding

In this cross-country sample, new-business creation tracks the ease of registration and the general quality of the regulatory environment more closely than the available proxy for corporate-law governance content. The result is an association, not a verdict on corporate law. But it does suggest that the strength of the connection between detailed company-law design and startup creation should not be assumed — it needs to be tested, with sharper legal data than are currently available cross-country.

Full regression table
Variable Coefficient Std. Error t-stat p-value VIF
OLS, n = 24. Explanatory variables standardised. Dependent variable: ln(1 + new business density).
R² = 0.616  |  Adjusted R² = 0.509  |  F = 5.776 (p = 0.002).
* p < 0.10   ** p < 0.05   *** p < 0.01