| City | Nominal Fare (€) | Cost of Living Index | PPP Fare (€) | Num. Taxis | Population | Licenses / 100k inhab. |
|---|---|---|---|---|---|---|
| Bari | 5.00 | 57.4 | 8.71 | 150 | 315473 | 47.55 |
| Bergamo | 9.20 | 58.1 | 15.83 | 41 | 120580 | 34.00 |
| Bologna | 4.10 | 68.8 | 5.96 | 722 | 390734 | 184.78 |
| Brescia | 8.50 | 71.4 | 11.90 | 103 | 199949 | 51.51 |
| Cagliari | 10.00 | 63.2 | 15.82 | 105 | 146627 | 71.61 |
| Catania | 6.50 | 55.4 | 11.73 | 188 | 297517 | 63.19 |
| Florence | 3.30 | 72.3 | 4.56 | 724 | 362353 | 199.81 |
| Genoa | 6.00 | 63.4 | 9.46 | 868 | 563947 | 153.92 |
| Milan | 6.90 | 75.8 | 9.10 | 4855 | 1366155 | 355.38 |
| Modena | 5.90 | 66.2 | 8.91 | 96 | 184739 | 51.97 |
| Naples | 4.00 | 60.2 | 6.64 | 2364 | 908082 | 260.33 |
| Padova | 10.00 | 65.9 | 15.17 | 150 | 207694 | 72.22 |
| Palermo | 3.00 | 56.0 | 5.36 | 319 | 625956 | 50.96 |
| Pisa | 5.00 | 71.8 | 6.96 | 80 | 89450 | 89.44 |
| Reggio Nell’emilia | 7.50 | 61.4 | 12.21 | 60 | 172518 | 34.78 |
| Rimini | 4.80 | 59.5 | 8.07 | 71 | 150630 | 47.14 |
| Rome | 6.00 | 61.2 | 9.80 | 7701 | 2746984 | 280.34 |
| Trento | 7.00 | 65.6 | 10.67 | 40 | 118911 | 33.64 |
| Treviso | 8.00 | 65.2 | 12.27 | 34 | 85770 | 39.64 |
| Trieste | 7.75 | 65.7 | 11.80 | 249 | 198668 | 125.33 |
| Turin | 3.50 | 63.7 | 5.49 | 1501 | 856745 | 175.20 |
| Verona | 7.50 | 67.4 | 11.13 | 176 | 255133 | 68.98 |
Analysis of the Relationship Between the Number of Taxi Licenses Issued per 100k and Fares at Parity of Citizen Purchasing Power in Italy
Abstract
This paper analyzes the relationship between the institutional quota system of taxi licenses and the consumer fares applied in 22 Italian cities. Using official data from the Transport Regulation Authority (ART) and cost-of-living indices from Numbeo, a log-log regression model (OLS) is estimated. The results show a statistically significant negative logarithmic relationship (\(p\text{-value} = 0.028\)) between the density of licenses per 100,000 inhabitants and purchasing power parried (PPP) fares. The empirical evidence rejects the null hypothesis, demonstrating that the administrative block of licenses acts as a barrier to entry that distorts the market to the detriment of consumers, offering strong arguments in favor of the liberalization of the sector.
Introduction
The non-scheduled public transport market in Italy, particularly the taxi sector, is historically at the center of heated political debates, corporate strikes, and hardships for urban users. The current regulation (hinged on Law 21/1992) delegates to individual Municipalities the power to determine the number of active licenses in the territory. This institutional structure has favored dynamics of “Regulatory Capture”, in which local administrations tend to protect the position rents of the existing taxi drivers’ corporation, blocking the issuance of new permits for decades in the face of constantly growing tourist and urban demand.
From the perspective of Public Choice economic theory and the liberal school, such barriers to entry artificially limit supply, generating allocative inefficiencies and upward pressure on prices. The objective of this analysis is to empirically verify whether, at parity of local purchasing power, a higher density of taxi supply translates into more contained fares for citizens.
Methodological Note
In order to conduct the analysis, data extracted from the Numbeo platform were used regarding taxi fares by city and the cost-of-living index, while data from the Transport Regulation Authority platform were used regarding the number of taxi licenses granted in each city. To conduct the analysis in the most correct way possible, it proved necessary to level taxi fares to the cost of living of each individual city, as well as to structure a composition ratio to express the number of taxi licenses, which are counted per 100,000 inhabitants.
Below is the final dataset obtained following the process of cleaning, merging, and normalizing data from different institutional sources.
Final normalized dataset for the 22 Italian cities analyzed
As highlighted by the table, the final dataset allows an equitable comparison of very different urban realities. The standardization of the fare (PPP Fare) and the parametrization of licenses on a demographic basis (Licenses / 100k inhab.) constitute the structural foundations for the subsequent estimation of the model.
Empirical Evidence
Before estimating the continuous functional relationship through regression models, it is useful to preliminarily verify whether a systematic and statistically significant difference exists in real fares between cities characterized by a low density of supply and those with a structurally higher supply.
To this end, the sample of 22 Italian cities was divided into two independent groups using 70 licenses per 100,000 inhabitants as a threshold value (a value close to the sample median):
Group X (Low density): Cities with fewer than 70 licenses per 100,000 inhabitants.
Group Y (High density): Cities with 70 or more licenses per 100,000 inhabitants.
The null hypothesis (\(H_0\)) of equality of means of the fares corrected for purchasing power in the two groups is tested against the alternative hypothesis (\(H_1\)) of difference between the means. The R code for extracting the groups and executing Welch’s two-tailed t-test (which does not assume homoscedasticity between groups) is as follows:
x = ita_finale$tariffa_ppp[ita_finale$licenze_100k < 70]
y = ita_finale$tariffa_ppp[ita_finale$licenze_100k >= 70]
test_t = t.test(x, y, conf.level = 0.95)
test_t
Welch Two Sample t-test
data: x and y
t = 1.0285, df = 18.28, p-value = 0.3171
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-1.514617 4.426050
sample estimates:
mean of x mean of y
10.618472 9.162755 With a p-value of \(\sim 0.32\), it is impossible to reject the null hypothesis. From a purely statistical point of view, no significant difference appears between the means of the two groups. However, it should be noted that the test fails mainly for two reasons:
Loss of information following the dichotomization of the analyzed variable.
Reduced sample size and likely presence of strong non-linear relationship components.
Given, therefore, the “failure” of the first approach, we can verify whether a negative relationship exists between the number of licenses per 100k inhabitants and fares at parity of citizen purchasing power through the simple regression model::
modello <- lm(tariffa_ppp ~ licenze_100k, data = ita_finale)
summary(modello)
Call:
lm(formula = tariffa_ppp ~ licenze_100k, data = ita_finale)
Residuals:
Min 1Q Median 3Q Max
-5.4746 -2.6599 0.3777 1.8827 5.3030
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 11.601656 1.057539 10.970 6.52e-10 ***
licenze_100k -0.015107 0.007302 -2.069 0.0517 .
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 3.091 on 20 degrees of freedom
Multiple R-squared: 0.1763, Adjusted R-squared: 0.1351
F-statistic: 4.28 on 1 and 20 DF, p-value: 0.05174To visualize the spatial distribution of the data and evaluate the fit of the estimated regression line against actual observations, the scatterplot is plotted:
As can be noticed from the summary output, the slope coefficient is negative, but the p-value associated with the licenses variable stands around 0.0517. Although it is extremely close to the critical threshold of 5% (\(\alpha = 0.05\)), it does not allow us to declare conventional statistical significance by a hair.
To investigate the nature of this inefficiency, we examine the residual diagnostic plots:
The analysis of the Residuals vs Fitted plot highlights a clear violation of the Gauss-Markov assumptions: the red error interpolation line is not flat and symmetrical around zero, but shows a marked curvature. Forcing a straight line on intrinsically curvilinear data generates systematic prediction errors, explaining why our p-value is on the “weak” threshold of \(0.0517\).
This structural limitation forces us to abandon the linear model in favor of a log-log model:
modello_log <- lm(log(tariffa_ppp) ~ log(licenze_100k), data = ita_finale)
summary(modello_log)
Call:
lm(formula = log(tariffa_ppp) ~ log(licenze_100k), data = ita_finale)
Residuals:
Min 1Q Median 3Q Max
-0.67060 -0.25701 0.08507 0.26077 0.48849
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.2281 0.4255 7.586 2.62e-07 ***
log(licenze_100k) -0.2236 0.0944 -2.369 0.028 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.3242 on 20 degrees of freedom
Multiple R-squared: 0.2191, Adjusted R-squared: 0.1801
F-statistic: 5.612 on 1 and 20 DF, p-value: 0.02802The correction of the functional form has an immediate and radical impact on the robustness of the analysis: we can observe how the p-value has dropped to a value of \(\sim 0.028\), allowing us to reject the null hypothesis with a \(95\%\) confidence level, establishing that there is a statistically significant relationship. We also note that the \(R^2\) index (purely descriptive, but useful for the purpose of the analysis) increases markedly compared to the simple linear model.
To verify that the violation of linearity has been cured, we examine the diagnostic plot of the residuals on the new model:
As can be seen from the graph, the red interpolation line has flattened drastically: the distortion turns out to be smaller compared to the previous analysis. This demonstrates that the relationship between taxi fares adjusted to the urban cost of living and the number of licenses issued per 100k inhabitants follows a negative logarithmic curve, with diminishing marginal returns.
Conclusions
The analysis demonstrates how the artificial barrier to entry created by Italian Municipalities generates enormous distortions in the market, especially in cities with a very low number of licenses relative to the population. When the constraints loosen, fares tend to lower.
Moving past the linear model in favor of a Log-Log one shows the costs of immobility and the benefits of openings: when markets are heavily closed, high taxi fares cause harm to final users. Conversely, openings allow a significant lowering of fares, which slows down as the number of issued licenses increases. This constitutes a refutation of the argument according to which a liberalization, even partial, of the sector would cause an uncontrollable collapse of the price, damaging the supply side, namely the taxi drivers.
In light of these results, the need emerges to overcome the rigid quota system of licenses, moving, possibly, towards the liberalization of platforms such as Uber and Cabify.