Baseline OLS results

In Table 2, we present the effects of economic freedom on income levels and growth rates for the full sample of districts. In Panel A, the effects of economic freedom on income levels are presented while Panel B presents short-run effects of economic freedom on cross-district growth rates. In column (1), the full sample estimate indicates that a 1 basis point increase in synthetic index of economic freedom is associated with an 8.5% increase in income level. The point estimate is economically large and statistically significant at 5%. The size of the long-run effect of economic freedom indicates a strong and persistent relationship with respect to income levels. In particular, a 1 basis point differences in the synthetic index of economic freedom roughly corresponds to the observed quantitative difference between the Neunkirchen (Saarland) and Günzburg (Bayern) districts, which is approximately equivalent to the distance between the median and 80^th percentile of the distribution. Our evidence readily suggests that the synthetic district-level index of economic freedom alone explains up to 17% of the per capita income variance across districts, while the full covariates explain up to 88% of cross-district income differences.

Table 2.

Economic freedom, growth and income levels across German districts

Notes: Table 2 presents the effects of economic freedom on real GDP per capita and its growth rate. The dependent variables are the log per capita GDP and the GDP growth rate. Each regression specification contains the set of unobserved state-fixed effects and administrative-fixed effects. The set of control variables in each underlying specification consists of the log initial per capita GDP, log state-level investment/GDP ratio and the fraction of population with successfully completed university entrance exams. The standard errors are adjusted for arbitrary and non-random disturbances arising from heteroscedastic distribution of error variance and serially correlated residuals simultaneously at district level, administrative cluster level and state level using Cameron et al. (Reference Cameron, Gelbach and Miller2011) multiway clustering scheme and the appropriate moment-based empirical distribution function. Asterisks denote statsitically significant regression coefficients at 10% (*), 5% (**), and 1%(***), respectively.

The long-run effect of economic freedom on income level is stable across multiple subsamples. In columns (2) and (3), the districts from Bayern and Baden-Württemberg are split off from the full sample. In spite of the coefficient drop to 0.084 and 0.041 as a result of excluding high-freedom districts, the evidence confirms the beneficial effect of economic freedom on income level, since the magnitudes are significant at 5%. In columns (4) through (6), the effect of economic freedom on income level jumps back to its full sample counterpart once districts from the states of Brandenburg, Hessen and Mecklenburg-Vorpommern are excluded from the full sample. In Panel B, we reject the notion that the beneficial effect of economic freedom is the result of the convergence to the mean. In the full set of specifications, the beneficial effect of economic freedom is conditional on the initial per capita income. The coefficient on initial per capita income is both negative and statistically significant at 5%. Greater economic freedom tends to foster growth after taking into account that districts with lower initial per capita income have grown faster than the districts with high initial per capita income. Excluding the districts with below-average per capita income, the coefficient on initial per capita income is substantially weaker, which is expected since these districts comprise the high-growth cluster in the post-unification period.

The fraction of the cross-district variation in income levels explained by economic freedom alone is stable in the range between 13 and 20%. Our evidence suggests that the main source of disparity in the magnitude of the effect arises from the composition of the sample. Excluding former West German states from the full samples tends to underscore the full-sample effect of economic freedom on growth, while excluding former East German districts tends to raise the magnitude of the effect considerably. This is further indicated in columns (11), (12) and (14), where the districts from three former East German states are excluded from the core sample.

In Panel B, the short-run effects of economic freedom on growth are presented. In column (1), a 1 percentage point increase in the synthetic index of economic freedom is associated with a 12.3% improvement in the rate of economic growth, holding everything else constant, and conditional on initial per capita income. The coefficient on initial per capita income (-0.511 with cluster-robust SE = 0.034), highlights a rapid degree of convergence between former East and West German districts from 1995 onwards. The effect appears to be stable across subsamples. Excluding districts from the high-freedom states such as Bayern and Baden-Württemberg tends to raise the underlying magnitude of the effect considerably to 0.155 (cluster-robust SE = 0.033), and 0.126 (cluster-robust SE = 0.034) compared to column (1). The beneficial effect of economic freedom on growth does not disappear across columns (9) to (14), where similar exclusion checks are performed on the remaining states. Our evidence indicates large growth gains from greater economic freedom. Taking our preferred specification in column (1), expanding the level of economic freedom from the 25^th percentile (Steinfurt, Nordrhein-Westfalen) to the 75^th percentile (Cham, Bayern) is associated with a 19% increase in growth rate suggesting that the growth dividend from greater economic freedom is considerable.

In Table 3, we examine the robustness of the key economic freedom parameters across subsamples. The robustness of the key parameters is assessed by splitting district types off the full sample specification. The effect of economic freedom on growth rate and income level is replicated on the core sample, large- and small-state samples to unravel the sensitivity of economic freedom in influencing per capita incomes and growth. In Panel A, the evidence confirms the positive effect of freedom on income level, which does not appear to be driven either by the institutional type of district or by the geographic composition of the sample. In Panel B, the effect of freedom on income appears to be stable in large-state subsamples while the growth effect of economic freedom appears to be slightly weaker. Lastly, in Panel C, the beneficial effects of economic freedom on per capita incomes are confirmed on a small-state sample whereas the short-run effects of freedom on growth rate tend to be slightly weaker and marginally significant at borderline levels. The effects of economic freedom are not driven by the institutional type of districts either. The underlying coefficient on economic freedom is in the 0.073 to 0.143 coefficient range across (long-run) income level regression specifications, and in the 0.103 to 0.174 coefficient range for cross-district growth regressions. The effects of economic freedom on both economic outcomes tend to be particularly stronger in larger states and slightly weaker in smaller states.

Table 3.

Effects of economic freedom on income level and growth of German districts across various subsamples

Notes: Table 3 presents the effects of economic freedom on income level and its growth rate over time across various sub-samples. The standard errors are robust against heteroscedasticity and serially correlated stochastic disturbances using the non-nested empirical distribution function based on the moment conditions and Cameron et al. (Reference Cameron, Gelbach and Miller2011) multiway clustering scheme. Asterisks denote statistically significant regression coefficients at 10% (*), 5% (**) and 1% (***), respectively.

Effect heterogeneity

In Table 4, the quantile regression estimates of the underlying cross-district growth and income model specifications are presented. Columns (1) to (5) present the effects of economic freedom on income levels. In column (1), the full sample specification suggests that a 1 basis point improvement in the index of economic freedom is associated with a 6.6% increase in income level. In column (2), the effect of economic freedom on income level at the 25^th percentile corresponds to the full sample OLS estimate with similar standard errors. In column (3), the effect at the median is slightly lower than at the 25^th percentile. In column (4), the long-run effect of the economic freedom tends to rebound and recover the effect magnitude from the 25^th percentile of the distribution. In column (5), the point estimate at the 90^th percentile of the outcome distribution tends to increase by 10 basis points relative to the effect at the 75^th percentile, and suggests that improving the level of economic freedom in the 90^th percentile of district-level income distribution tends to boost per capita incomes in the respective percentile by 6.4%. Lower-income districts tend to gain disproportionately less from higher economic freedom than higher-income districts: the effect tends to weaken at the median, recover at the 75^th percentile of the distribution, while it gains strength in the upper tail of the district-level per capita income distribution. Columns (6) to (10) present the short-run effect of economic freedom on district-level growth rates. In column (6), the full sample freedom growth estimate suggests that a 1 basis point improvement in the synthetic index of economic freedom is associated with a 12% improvement in the growth rate. Compared with its long-run counterpart, the short-run effect of economic freedom unveils substantial heterogeneity across various tails of the growth distribution. Column (7) reveals that improving economic freedom by 1 basis point in the 25^th percentile of growth distribution is associated with a 13.7% expansion in short-run growth rate. The short-run effect tends to weaken substantially at the median of the growth distribution, as indicated by column (8). In column (9), the effect at the 75^th percentile of the distribution is similar to the effect at the 25^th percentile while the effect gains notable strength in column (10), which highlights considerable gains from greater economic freedom in the upper tail of growth distribution.

Table 4.

Effects of economic freedom on income level and growth rate across different quantiles of the outcomes

Notes: Table 4 presents the effects of economic freedom on real GDP per capita and its growth rate over time across different quantiles of each outcome. The standard errors are cluster-robust against the multiple spatial dimensions of heteroscedasticity and serially correlated stochastic disturbances using Machado and Santos Silva (2015) quantile regression function. Asterisks denote statistically significant regression coefficients at 10% (*), 5% (**) and 1% (***), respectively.

Counterfactual scenario

The ultimate question pertaining to the robustness of our estimates concerns the counterfactual scenario. Would German districts with below-average economic freedom significantly improve growth rates and income if they moved up the ladder of economic freedom? We construct a counterfactual scenario by focusing on the districts with a level of economic freedom on the synthetic index below the 90^th percentile, to separate high-freedom districts from low-freedom ones. The level of economic freedom at the 90^th percentile roughly corresponds to the level of economic freedom observed in the Landshut district in Lower Bavaria. Specifying the counterfactual threshold at the highest level of economic freedom observed in our sample is implausible, since an array of observable and idiosyncratic characteristics can be the driving force behind top level of economic freedom. Our counterfactual scenario invokes growth and income levels computed by using the point estimates from equations (3.1) and (3.2) in the following setup:

(5.1) $$\begin{eqnarray} \ln y_{i \in \left\{ {1\left[ {EF < Q_{.90}^{EF}} \right]} \right\},j,k,l}^{\it Counterfactual} &=& \ln A + {{\hat{\gamma }}_1} \cdot EF_{j,k,l}^{Q\left( {.90} \right)} + {{\bf X^{\prime}}}\beta + \sum\limits_{j = 1}^J {{\hat{\phi }}_j} \cdot 1\left[ {i \in j} \right]\nonumber\\ && +\, \sum\limits_{k = 1}^K {{{\hat{\alpha }}_k} \cdot 1\left[ {i \in j \in k} \right]} + \sum\limits_{l = 1}^L {{{\hat{\Theta }}_l} \cdot 1\left[ {i \in j \in k \in l} \right]}\quad \end{eqnarray}$$

(5.2) $$\begin{eqnarray} \Delta \ln y_{i \in \left\{ {1\left[ {EF < Q_{.90}^{EF}} \right]} \right\},j,k,l}^{\it Counterfactual} &=& \ln A + {{\hat{\lambda }}_1} \cdot EF_{j,k,l}^{Q\left( {.90} \right)} + {{\bf X^{\prime}}}\alpha + \sum\limits_{j = 1}^J {\mu _j} \cdot 1\left[ {i \in j} \right]\nonumber\\ && +\, \sum\limits_{k = 1}^K {{\theta _k} \cdot 1\left[ {i \in j \in k} \right]} + \sum\limits_{l = 1}^L {{\Gamma _l} \cdot 1\left[ {i \in j \in k \in l} \right]}\quad \end{eqnarray}$$

where ln y^{Counterfactual} denotes the counterfactual level of income with the level of economic freedom at the 90^th percentile, and Δln y^{Counterfactual} is the equivalent counterfactual growth rate. The sample consists of the district-level observations below the 90^th percentile economic freedom threshold, indicated by i ∈ {1[EF < Q^EF _.90]}, j, k, l The vector X represents structural and sectoral covariates from the baseline regressions. Our benchmark specification used to construct the counterfactual scenario comprises the model specification with covariates, and standard errors clustered at the district, administrative unit, regional government and state levels to prevent serially correlated stochastic disturbances from undermining the inference on the two key parameters of interest, ${\hat{\gamma }_1}$ and ${\hat{\lambda }_1}$. Using cross-district income and growth model specification, we obtain ${\hat{\gamma }_1} = .066$ (cluster-robust SE = 0.021) and ${\hat{\lambda }_1} = .123$ (cluster-robust S E = 0.034). The differences in economic freedom account for up to 17% of differences in income level across districts, and for up to 3% of differences in cross-district growth rates. We use the point estimates ${\hat{\lambda }_1}$ and ${\hat{\gamma }_1}$ to calculate the counterfactual district-level income and growth outcomes, taking into account the outcomes’ variance driven by the covariates.

Figure 2 presents the counterfactual distribution of per capita income across districts. The transition to the 90^th percentile of economic freedom is associated with marked income and growth gains. For districts characterized by low levels of economic freedom, the gains are remarkable. For Aachen district, the counterfactual scenario indicates a 55% increase in per capita income. Similarly, for the low-freedom districts Delmenhorst and Kaiserslautern, our counterfactual estimates indicate 45% and 41% improvements in per capita incomes. The counterfactual scenario suggests that the per capita income level is higher than the actual level in 89% of the below-90^th percentile districts. The remaining 11% of districts are either characterized by exceptionally high income levels (München, Frankfurt am Main, Darmstadt) or display the counterfactual outcome very close to the actual level. The counterfactual Gini coefficient for the cross-district per capita income distributions drops by 4 basis points (from 0.19 to 0.15) suggesting that improving the level of economic freedom to the 90^th percentile in below-cutoff districts tends to alleviate cross-district inequality.

Figure 2.

Counterfactual GDP per capita with 90^th percentile economic freedom.

Figure 3 replicates the counterfactual scenario for the cross-district growth specification. The evidence unveils noticeable growth gains. About 65% of the districts tend to experience a higher rate of economic growth compared to the actual rate. The most substantial improvement is indicated for the districts Herne, Delmenhorst, Erzgebirgkreis and Leverkusen, where the growth rate gains at least 1.5 percentage points. The remaining third of the districts experience either exceptionally high rates of growth or the counterfactual rates as closely matched with the actual rates. The counterfactual outcomes highlight a notable decrease in cross-district growth inequality. The cross-district Gini coefficient on growth rates tends to drop by 20 percentage points, i.e. from 0.42 to 0.22.

Figure 3.

Counterfactual growth rates with 90^th percentile economic freedom.

From the normative perspective, the remaining question is what specific policy recommendation the counterfactual scenario could offer. The policy recommendation critically depends on the loading of eigenvectors in the synthetic index of economic freedom from section 2. The counterfactual scenario suggests that expanding economic freedom tends to reduce growth disparities and boost income levels across districts. Policies aimed at reducing the property tax burden and income tax rates and constraining public sector employment tend to have the most beneficial impact on expanding economic freedom, and are set to foster growth and income levels to the greatest degree. Policies aimed at reducing the business turnover taxes and public debt would tend to have a sizeable but moderate impact on economic freedom and growth, while the reduction of value-added taxes is set to influence economic freedom to the minimal degree. Nevertheless, reforms favouring lower property tax burden and lower individual income tax rates would yield a disproportionately favourable effect on growth and income levels compared to the other four components of economic freedom.