Income taxation and regional growth: DID evidence from an income tax treatment in Italy

doi:10.21203/rs.3.rs-3962177/v1

Download PDF

Research Article

Income taxation and regional growth: DID evidence from an income tax treatment in Italy

https://doi.org/10.21203/rs.3.rs-3962177/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

This paper offers some fresh empirical evidence about the relation between income taxation and regional growth-related variables. Using the Italian regional income tax treatment of 2007 as a quasi-natural tax experiment, we present various panel data DID estimates of how income per capita and other variables related to economic growth, including labour supply, consumption, and VAT certificates, respond to exogenous income tax changes. Results show that the extensive negative adjustments of response variables measuring supply of labour services offered by taxpayers affected by the treatment is statistically significant, rapid, and strong but not long-time lasting and that treated families reduce in a similar manner their consumption with respect to families in the control (untreated) regions. Analogous adjustment responses characterise the growth of per-capita regional GDP response to tax changes. The estimated aggregate effects are further compared with the spatial-temporal patterns observed for every response variable. Policy issues are also discussed.

JEL codes: C10, C18, C21 H2, E2, E32, E62, C23, C26

Panel Data DID

Convergence tests

Income Taxation

Extensive labour supply change

Taxation and regional growth

Regional/local, as well as central, authorities carry out taxation and expenditure activities. Their policy measures affect local income, employment, investment, and other residents’ decisions. Allocative efficiency and equity arguments in favour of distributing taxation and expenditure power between central and regional/local governments are generally grounded on the hypothesis that those functions are better attended by a combination of local and central government actions, rather than by exclusive central government activities. Similar arguments can be reiterated for other public policies measures such as the control of externalities and some aspects of the redistribution of personal income. A vast economic literature analyses these issues within the framework of fiscal federalism theories. On the contrary, less is known (and investigated) about the effects that regional/local taxes (particularly, income taxes) and expenditure activities can have on regional/local economic variables crucially related to regional/local growth, such as investment, consumption/saving decisions, and, above all, employment choices (e.g., salaried employment, self-employment, etc.). International literature on these issues is not ample and some fresh analysis may be useful to parallel the already abundant literature covering the correlated field concerning the relation between taxation and Beta or Sigma convergence of per-capita regional/local GDPs (Haaf & Kool, 2017; R. Harris, 2011; Higgins, Levy, & Young, 2006).

As for the empirical literature studying the relationships between income taxation and variables conducive to growth at the aggregate level of the national economy, Zidar (2019, p. 1440) notes that published empirical results are generally consistent with the possibility of heterogeneous aggregate effects of income tax changes. His estimations show the importance of the distribution of tax changes across income groups for their overall impact on employment, investment, and consumption. His and other studies provide evidence that lower income households tend to have higher marginal propensities to consume (Dynan, Skinner, & Zeldes, 2004; Jappelli & Pistaferri, 2010; Johnson, Parker, & Souleles, 2006; McCarthy, 1995; Parker, 1999; Parker, Souleles, Johnson, & McClelland, 2013) and so they respond to tax policy by increasing consumption more than high-income groups. Labour supply, consumption, investment, and other growth-related variables of different income groups may respond to income tax policy in an opposite way. On the extensive margin for lower-income groups, Eissa and Liebman (1996), Meyer and Rosenbaum (2001), for instance, show that in the USA the Earned Income Tax Credit (a refundable tax credit for low-to-moderate-income working individuals and couples, particularly those with children, enacted in 1975) has increased labour force participation. Other authors (e.g., Romer & Romer, 2010; Saez, Matsaganis, & Tsakloglou, 2012) show that for high-income earners there is some evidence that the efficiency costs of raising taxes on top-income taxpayers expressed in terms of labour supply and other margins may be limited or can be offset by shifting in the timing or form/source of income acquisition (Auerbach & Siegel, 2000; Goolsbee, 2000). Others analyse the potential adverse base effects of tax changes such as tax avoidance, outright tax evasion, and a general reduction in economic activity (Piketty, Saez, & Stantcheva, 2014). When territorial aspects are accounted for, the analysis considers the tax-induced mobility of inventors and discusses whether income tax rates can be employed by local authorities to attract highly qualified foreigners. For example, Akcigit, Baslandze, and Stantcheva (2016) find that prolific inventors migrate between countries in response to changes in personal top income tax rates and obtain an estimate of the elasticity of the number of foreign inventors (in a country) with respect to the top net-of-tax rate close to one. Moretti and Wilson (2017) examine the mobility responses of inventors to changes in personal top income tax rates across US states and find a corresponding elasticity with respect to the top net-of-tax rate of 1.8.

Yet, when the analysis of the responsiveness of growth-related variables to income tax aims at estimating the individual responses of income, investment and consumption to regionally differentiated income tax changes, the specific issue of the tax effects on local growth (or on local growth-related variables) and local fiscal multipliers (Ramey, 2011a, 2011b) becomes important. This is so because local income taxes may have effects on local growth-related variables and may or may not offset each other in terms of aggregate national growth or overlap to national tax changes relevant for the entire economy. Therefore, it is of considerable importance to understand how local income taxes may affect local decisions (such as regional/local labour supply or residential location) and evaluate the possible consequences of these decisions on the level and geographical distribution of productive activities and on local technological spill over (Widmann, 2023). In addition, since the market value of material goods, and its geographical distribution, may be affected by income taxation, local income tax rates can generate tax externalities of various nature across local jurisdictions (Esteller-Moré & Solé-Ollé, 2002). Baskaran (2021), for instance, presents results from a set of municipalities who, in North Rhine-Westphalia, increased their local property and business tax rates. He studies the revenue and base effects of local property and business tax hikes in a generalized difference-in-differences (DID from now on) design. His results suggest that the property tax hikes had a revenue elasticity of unity even in the long run. He concludes that tax changes had no adverse effects on property tax bases. For the business tax, he finds no significant effects on revenues and tax bases.

The above literature has the great merit of showing that regional/local income taxes are relevant ingredients of regional/local economic decisions. Yet, since it is not directly focused of the relation between income taxes and growth-related regional/local variables, it does not allow drawing some inference on the contribution that regional/local income taxation can have on regional/local economic growth. The search of that inference motivates our paper, which presents some fresh empirical evidence about the relation between income taxation and regional/local growth-related variables using Italy as a case study. Despite Italy is not a federal country, she may provide a good case for studying the reaction to changes in income taxation of growth-related variables. This is so for two main reasons. Since 1970 (the year of the creation of the Italian Ordinary Regions as an application in slow motion of the 1948 democratic Constitution of the Republic) and particularly in the last years, Italian regions have accumulated large tax and expenditure powers, which must have affected somehow the economies of their territories and their growth trajectories. On the other hand, Italy has been (and still is) plagued by regional dualism (an ever-richer North coexisting with a comparatively poorer South) with large per-capita GDP growth differentials across regional territories. Since regional income taxation is not uniform countrywide (regional authorities use different local rates and adopt different tax expenditure measures), income tax heterogeneity may have affected regional growth paths and the realization of growth-related regional variables, such as per capita GDP, labour supply, consumption, etc. We study how exogenous regional income tax shocks may have affected the above variables in Italy.

To contribute to the literature, we adopt the following empirically strategy. We employ a DID regression model to analyse the effects of regional/local income tax changes on variables related to regional economic growth in Italy to evaluate if regional differences in income taxation triggered different effects on local growth-related variables. Implementing this approach, Jakobsen, Jakobsen, Kleven, and Zucman (2019) provide causal evidence on the impact of net-of-tax rate on wealth accumulation in Denmark whereas Kleven, Landais, and Saez (2013) study the causal relationship between tax rates and a particular subset of the population of migrants (football-players super stars). Our DID estimates of the ATET permits to study the timing of the transactions involved in income tax responses (reducing or delaying over time labour supply or consumption) or to identify real responses such as per capita regional income growth. We conduct the empirical study using a large and disaggregated Italian data set containing regional tax and economic data and, to identify the DID empirical model, we exploit a quasi-natural tax experiment that occurred in Italy in 2007. It affected a non-randomly selected subset of Italian regions (treatment group) for a certain number of years and consisted in an exogenously imposed change of their income tax rates with respect to the rates of the other regions (control group). Hence, the 2007 central government policy defines the framework of a quasi-natural tax experiment involving the taxpayers of five (not-randomly-chosen) treated regions vs the taxpayers of the remaining 15 untreated control regions. Accordingly, we may adopt the identification hypothesis of a DID method and to conduct the causal-effect study of the impact of the central government exogenous tax policy on some regional growth-related variables under (testable) parallel trend conditions. Response variables related to local growth (various types of employment, family consumption, etc.) are used to evaluate the short and long run impact of income tax changes taking both time and individual effects into account. The main taxes comprised in the analysis are the tax on regional productive activities (IRAP, a local business income tax in disguise) and the Regional Income Surcharge tax applied by regions on top of the Italian national personal income tax (IRPEF). Results show that differences in regional income tax rates (tax treatment) affect the behaviour of regional economic variables related to growth. The clearest effect is the negative and statistically significant impact of the income tax increase on self-employment, per-capita regional growth, and family consumption, as well as the general absence of anticipatory effects. The same result emerges using other labour response variables (e.g. the annual stock of New VAT Certificates, necessary for self-employment activities) but not investment expenditures. In general, the exogenous tax effect is time lasting but not long lasting. Adjustments of the labour supply to a long-run trend are observable, on average, 3 years after the termination of the treatment.

The paper is structured as follows. Tax factors considered in this study are described in section 2 together with the narrative of the Italian 2007 tax experiment. Details on the latter are in Appendix 1. Section 3 contains the discussion of the DID identification and a descriptive statistic survey of the data set with plots of the response variables used in the DID analysis. In section 4 the empirical results are commented in comparison with the results of the previous international literature. Policy implications are also discussed. Section 5 includes some additional robustness analysis and some new test procedures for detecting the presence of pre-treatment parallel trends in our panel data set (we call it a falsification test). Section 6 briefly concludes.

The measures adopted by the Italian government at the end of 2006 are a good example of a possible exogenous income tax exogenous treatment. In 2006, the Italian central government imposed to 5 out of 20 Italian regions (Abruzzo, Campania, Lazio, Molise, and Sicily) an increase - automatic and compulsory - of both their IRAP tax rates and their regional Income Tax Surcharge. The measure was adopted to force those five regions to collect additional local resources and use them exclusively to reduce the accumulated deficits of their regional health care systems (when this was equal to or greater than 7% of the previous year value). The content of the 2006 policy (described in details by Caruso and Dirindin (2019) is summarized below (a more detailed description of the taxes involved in the treatment is in Appendix 1):

IRAP tax rates. The ongoing standard IRAP tax rate existing in 2006 (4.5%) and employed by the five regions mentioned above was increased from the tax year 2007 onwards by almost a 1%. This implied that the actual tax rate for those regional taxpayers became equal to 5.25%. IRAP tax rates for taxpayers resident in other regions were not modified.
Regional Income tax surcharge. From the tax year 2007 onwards, the five “treated” regions had to increase by 0.5% the ongoing rate of 1.23% they applied as a surtax on the personal income tax base (determined by the national Law) of their residents. Yet, since the five regions were under “controlled administration” by the central government, the surcharge rate was further increased by an additional 0.3%. The net total increase was equal to 0.8% leading to an income Tax Surcharge rate of 2.03%.
A third element of the treatment is the following. Since the institution of the regional income surcharge, the Italian regions were permitted to allow deductions from the income surcharge payments instead of providing subsidies, vouchers, service vouchers and other social support measures determined by their regional legislation in favour of some taxpayers. These powers were taken away from the five treated regions. As a result, they could not compensate the surcharge increase with tax deductions.

For both taxations, the tax rates increase was compulsory, quantitatively relevant and finalized to collect an additional tax yield hopefully sufficient to cover the prolonged budget unbalances of their regional health care systems. No other uses were permitted. The law introducing the new regime for the five regions indicated that the measures had to be adopted for a 3-year period. The other 15 regions were not compelled to increase their rates and they decided not to modify their ongoing rates.

The exogenous and unexpected increases of IRAP Tax and Personal Income surcharge mainly fell on Individual Business and Self-Employed Professionals (the main group of IRAP taxpayers within a region). Yet also the generality of taxpayers paying the national Income Tax was affected by the measure although less severely (they had to face only the increase of the regional personal income surcharge). Hence, based on the realization of a state of the word (the deficit of their regional health care systems) determined when the measure was not in effect, taxpayers, in the three-year period starting in 2007, were exposed to an exogenously determined tax increase of the main regional taxes. Then, the 2006 tax law determined the condition for the estimation of the causal-effect relationship between tax treatment and response variables as average treatment within a fully identified panel data DID model.

Our approach parallels the one conducted using municipal budget data by Baskaran (2021). Due to a reform of the local equalization scheme in 2003, a set of municipalities in North Rhine-Westphalia (NRW) increased their local property and business tax rates by one to two percentage points, while the remaining municipalities kept their rates constant. Baskaran (2021) uses this variation across municipalities and over time to study the revenue and base effects of local property and business tax hikes in a generalized DID design. In the present case, the exogenous treatment was designed by the central government to modify asymmetrically regional and not municipal taxation. Therefore, the Italian and German tax treatments are not entirely equivalent. An increase in local tax rates by local governments may not increase revenues proportionally because of adverse base effects. First, mobile firms may respond to municipal tax rise by moving to jurisdictions where the tax burden is lower (Wilson, 1986; Zodrow & Mieszkowski, 1986). In the case of property taxation, high property taxes may lead to an outmigration of residents and to lower property values and a declining property tax base. Further reasons for adverse base effects are increased tax avoidance, outright tax evasion, and a general reduction in economic activity (Piketty et al., 2014). On the contrary, if we consider the regional scale of the exogenous tax hike of 2007, the first of the above concern may not be entirely justified. Tax minimizing migration should be across regions, which may imply higher moving costs¹. Finally, a reason to refer to the regional level is that the analysis of regional income taxation is less vulnerable to the observation that, in a decentralised tax system, the effects of tax policies enacted by one government are not confined to its own jurisdiction. If the tax base is mobile among jurisdictions, tax rate increases by one local government may affect the amount of taxes collected by other local governments as well as the behaviour of its residents. These sources of fiscal interdependence are called in the literature vertical and horizontal tax externalities and have been empirically investigated with respect to some federal state such a Canada (Esteller-Moré & Solé-Ollé, 2002). Fiscal interdependence may reduce the robustness of the empirical analysis of the effects of local taxation.

In what follows we conduct a DID estimation of the effect of the income tax treatment using a panel data set of growth-related variables summarised in Table 1 below. To illustrate the estimated model, we introduce the following notations:

${y}_{it}=$ Observed response variable in region i at time t

$$D1= \left\{\begin{array}{c} 0 \text{w}\text{h}\text{e}\text{n} \text{t}\text{h}\text{e} \text{r}\text{e}\text{g}\text{i}\text{o}\text{n} \text{i}\text{s} \text{u}\text{n}\text{t}\text{r}\text{e}\text{a}\text{t}\text{e}\text{d} \left( \text{i}\text{r}\text{r}\text{e}\text{s}\text{p}\text{e}\text{c}\text{t}\text{i}\text{v}\text{e} \text{o}\text{f} \text{t}\right) \\ 1 \text{w}\text{h}\text{e}\text{n} \text{t}\text{h}\text{e} \text{r}\text{e}\text{g}\text{i}\text{o}\text{n} \text{i}\text{s} \text{t}\text{r}\text{e}\text{a}\text{t}\text{e}\text{d} \left( \text{i}\text{r}\text{r}\text{e}\text{s}\text{p}\text{e}\text{c}\text{t}\text{i}\text{v}\text{e} \text{o}\text{f} \text{t}\right)\end{array}\right.$$

$$D2= \left\{\begin{array}{c} 0 \text{f}\text{o}\text{r} \text{y}\text{e}\text{a}\text{r}\text{s} \text{i}\text{n} \text{w}\text{h}\text{i}\text{c}\text{h} \text{t}\text{h}\text{e}\text{r}\text{e} \text{w}\text{a}\text{s} \text{n}\text{o} \text{t}\text{r}\text{e}\text{a}\text{t}\text{m}\text{e}\text{n}\text{t} \left(\text{i}\text{r}\text{r}\text{e}\text{s}\text{p}\text{e}\text{c}\text{t}\text{i}\text{v}\text{e} \text{o}\text{f} \text{i}\right)\\ 1 \text{f}\text{o}\text{r} \text{y}\text{e}\text{a}\text{r}\text{s} \text{i}\text{n} \text{w}\text{h}\text{i}\text{c}\text{h} \text{t}\text{h}\text{e}\text{r}\text{e} \text{w}\text{a}\text{s} \text{t}\text{r}\text{e}\text{a}\text{t}\text{m}\text{e}\text{n}\text{t} \left(\text{i}\text{r}\text{r}\text{e}\text{s}\text{p}\text{e}\text{c}\text{t}\text{i}\text{v}\text{e} \text{o}\text{f} \text{i}\right)\end{array}\right.$$

The total sample period is T = (1995… 2021) with t = (1995 … 2006) = No-Treatment period (dummy variable D2 = 1) and t = (2007 … 2021) = Treatment period (dummy variable D2 = 1)

${\varvec{X}}_{it}$ = vector of cofactors/controls (to be specified in each equation)

As noted above, we define the target or response variable $y$ in terms of potential outcomes and estimate the average effect of treatment on the treated units. This compares the potential outcomes with treatment to the potential outcomes with no treatment, in the treated group. Written mathematically, the estimated effect of the treatment is the β coefficient of panel data OLS equation

$${y}_{it}=Constant+\alpha \times D1+\delta \times D2+ \beta \times \left(D1\times D2\right)+ {{\varvec{X}}_{it}\gamma +\epsilon }_{it}$$

where ${\epsilon }_{it}$ is a White Noise error term.

The crucial parameter to estimate is β. If we call R a binary income tax treatment indicator, we have $\widehat{\beta }\equiv CATET=\mathbb{E}\left({y}_{i,t}^{1}-{y}_{i,t}^{0}|{X}_{it};{R}_{i,t}\right)$ or $\widehat{\beta }\equiv ATET=\mathbb{E}\left({y}_{i,t}^{1}-{y}_{i,t}^{0}|{R}_{i,t}\right)$ if cofactors are included or excluded. CATET (Conditional Average Effect upon Treated) or ATET (Average Effect upon Treated) will be interpreted as the mean effect (conditional or unconditional) of the “tax treatment” for those taxpayers who were compelled to participate in the central government program of income tax changes (i.e. the residents in the treated areas). Unless it is not explicitly specified, in all estimates the average treatment effect is estimated by adjusting for both cross sectional and time effects. Note that outcomes variables are independent on treatment conditional upon X. Then the unconfoundedness hypothesis holds. For (1) to identify the parameters of interest, tax shocks need to be exogenous conditional on fixed effects and controls. Intuitively, this identifying assumption is that the national tax shock of 2007 is not favouring regions that are doing poorly relative to how fast they normally perform in terms of dependent variables. Then, the validity of comparing outcomes of regions having different distributions of the response variables relies on three key assumptions: (1) the national tax shock is exogenous, (2) targeted tax shocks are unrelated to any possible targeted level of the response variables, and (3) outcomes from untreated regions provide a reasonable counterfactual since the 2007 tax shock was absent. Since we control for region and year fixed effects in Eq. (1), the first assumption maintains that Italian policy makers of the time were not systematically setting income tax policy to respond to idiosyncratic regional shocks other than the budget unbalance of their health care systems.

3.1 Data description and parallel trends

We collected regional ($N=20$) macroeconomic indicators on a yearly basis from 1995 to 2021 (i.e., $T=27$ years). Data were provided by the Italian National Statistical Institute (ISTAT) and the Italian Treasury (Department of Finance). The set of available indicators includes: regional GDP (total and per capita values); total regional value added; employment (number of employed persons and employment rate) and self-employment (measured as the share of self-employed people out of the total annual employment in the region); family consumption; the number of new VAT certificates²; the Internal Fixed Gross Investments³ for non-financial sectors; and, the indirect taxes collected within the region. Descriptive statistics of the response variables used in this paper with unit root test and cross-correlation tests are shown in Table 1.

Table 1

Descriptive statistics of response variables and cofactors (monetary variables are in 2015-chained values)
Variables	Obs.	Mean	SD	Min	Max	Panel unit roots tests H₀: Panels contain unit roots			Panel Stationarity	Pesaran CD test for cross-sectional dependence
20 Units /Regions T = 1995, …, 2021						Levin-Lin-Chu (LLC)	Harris-Tzavalis (HT)			Test statistics	P-value	Average joint T		Mean abs($\varvec{\rho }$)
Regional GDP (Euros)	540	8.35e + 09	8.24e + 09	4.25e + 08	3.86e + 10	t* = -2.429 PV = 0.0076		ρ = 0.6488 PV = 0.0057	Yes, both tests	43.348	0.000	27.00	0.64
Per-capita regional GDP (Euros)	540	27699.96	7453.601	15313	43103	t* = -1.928 PV = 0.0269		ρ = 0.6669 PV = 0.0210	Yes, both tests	53.49	0.000	27.00	0.75
Regional Value Added per regional worker (Euros)	540	49832.05	13346.88	4253	76483	t* = -2.314 PV = 0.0103		ρ = 0.6615 PV = 0.0146	Yes, both tests	53.331	0.000	27.00	0.83
Internal Fixed Gross Investment (Euros)	540	1.62e + 10	1.56e + 10	8.45e + 08	8.07e + 10	t* = -2.594 PV = 0.0047		ρ = 0.7903 PV = 0.9093	Yes LLC No HT	48.668	0.000	27.00	0.68
Total regional Employment (Units)	540	1210098	1063136	58600	4884300	t* = -2.239 PV = 0.0126		ρ = 0.8682 PV = 0.9997	Yes LLC No HT	45.583	0.000	27.00	0.65
Self-employment (% of total employment)	355	13.55634	2.32783	6.9	20.4	t* = -5.298 PV = 0.000		t* = 0.995 PV = 1	Yes LLC No HT	19.578	0.000	17.51	0.37
New VAT Certificates (Units) (2001–2021)	421	25853.42	22523.57	879	90191	t* = -4.879 PV = 0.0000		ρ = 0.6287 PV = 0.1595	Yes LLC No HT	24.256	0.000	20.20	0.53
Regional Family Consumption (Euros)	540	50667.72	43999.08	2593	204864	t* = -3.0287 PV = 0.0012		ρ = 0.6649 PV = 0.0185	Yes, both tests	43.009	0.000	27.00	0.60
Indirect taxes collected within the region	520	9176.237	9888.198	358	53105	t* = -0.6909 PV = 0.2448		ρ = 0.6645 PV = 0.0370	No LLC Yes HT	65.586	0.000	26.00	0.93
Notes. Sources: ISTAT and Italian Treasury (Department of Finance). Mean, standard deviation (SD), min and max are computed using the overall sample. All monetary values are expressed as values linked to the 2015 reference year: the linking provides a measure of the economic aggregate of interest in terms of volume, i.e. net of the underlying price dynamics. Note that the sum of the chained values of the components of an aggregate is not equal to the chained value of the aggregate itself.

Under the null hypothesis of cross-section independence, the Pesaran CD test statistics is normally distributed, i.e., CD ~ N(0,1), and p-values close to zero indicate that data are correlated across panel groups. In the Levin-Lin-Chu (LLC) unit root test, a common autoregressive parameter for all panels is assumed, so it does not allow for the possibility that some regional variables contain unit roots while others do not. To mitigate the effects of possible cross-sectional correlation among time series, we removed the cross-sectional means from the series (demeaning). Also, we included for all the tests a linear trend term. The LLC test requires that the ratio of the number of panels to time periods tend to zero asymptotically. Hence, it is not well suited with relatively few periods. Hence, we also present the Harris–Tzavalis (HT) unit root test, which assumes that the number of panels tends to infinity while the number of periods is fixed. For both tests, the null hypothesis is H₀: Panels contain unit roots.

Note that the reported statistics are computed on the overall sample and they do provide information about the overall scale and the variability of the macroeconomic indicators under consideration. However, they do not provide any insight about the territorial heterogeneity typically affecting the Italian economic measurements (see, for example, De Philippis, Locatelli, Papini, and Torrini (2022); Federici, Ferrante, and Parisi (2023); Lo Cascio, Mazzola, and Epifanio (2019)). In the Appendix (Figures A2, A3 and A4) we show the evolution of the considered regional indicators for selected years along with the 1995–2021 average. For almost all indicators included in the sample, the charts show a pronounced regional heterogeneity, highlighting in many cases either the North-South gap (e.g. GDP per capita, value added and employment), which persists over the decades, or clear patterns linked to urbanisation degree and population density (e.g. regional household consumption).

Unit root and cross-correlation are tested due to the panel macroeconomic nature of the dataset. Adding the cross-section dimension to the time series dimension offers an advantage in testing for nonstationary and cointegration since cross-section increases the data set used in those tests, thus improving their power. However, the cross-sectional dimension also brings some new problems into question, namely the existence of cross-sectional dependency that can bias usual panel data unit root test results in small samples. The Pesaran (2007) cross-sectional dependence test checks for this issue. Our findings show, as expected, the presence of high correlation among regional variables across panels/regions. The importance of a priori check of the existence of unit roots in the panel data comes from the already known effect that the presence of unit roots in time series may cause a misinterpretation of estimated results. The unit root hypothesis is tested through the Levin, Lin, and Chu (2002) and R. D. F. Harris and Tzavalis (1999) unit root tests (i.e., for both, the null hypothesis is that the panels contain unit roots) accounting for linear trend and cross-correlation adjustments. With the only exception of the variable “Indirect taxes collected within the region Regional Value Added per regional worker”, we reject the null that the panel contains unit roots and conclude that the series are trend stationary. Since in the OLS estimation of (1), panels are considered to be homogeneous, the opposite would be a drawback. The absence of unit roots is important for our DID analysis. In the most intuitive sense, stationarity means that the statistical properties of a process generating a time series do not change over time because the statistical properties of the data generation process do not change over time. It does not mean that the series does not change over time, just that the way it changes does not itself change over time. If an exogeneous event occurs (such as an exogenous tax change) changes in the response variables can be, loosely speaking, attributed to the influence of the event and not to some long run property of the series.

Eventually, the presence of parallel trend in pre-treatment periods for a group of response variables is of great importance for DID. For some of our response variables it is shown in the following Fig. 1, which shows how the outcome variables for treated and control groups evolve over time, before and after the treatment. Parallel trends require that, in the absence of treatment, the difference between the treatment and control group is constant over time. Violation of parallel trend assumption will lead to biased estimation of the causal effect. It is frequently assumed that the smaller the time period tested (for example a 2-year period), the more likely the assumption is to hold, but with a period of 27 years the possibility that treated and control groups have different outcome trends (which may generate time convergence) cannot be excluded.

Visual inspection of Fig. 1 shows that parallel variations prior to 2007 is a reasonable hypothesis for the response variables and it is suggestive that the identification proposition of the DID are valid. Plots reproduced in Appendix 2 (Fig. A6) show the time series of index numbers with baseline 2006 = 100 for all available response variables. Abruzzo, Campania, Lazio, Molise and Sicilia are the five treated regions, while the non-treated regions are averaged across year (solid black lines). The plots indicate that the time behaviours of the response variable accord with the above parallel trend interpretation. Yet, the parallel trend hypothesis will be subjected to statistical test based on an augmented version of Eq. (1) and is discussed at the beginning of the next section. The presence of anticipation effects will be also tested.

DID results are presented in Table 2.

Table 2

DID estimations of the Italian income tax treatment
Response Variables	Total Employment	Self-Employment	New VAT Certificates	Regional GDP Growth per capita	Regional GDP Growth per capita (HP filter)	Regional GDP per capita	Log Family Consumption
ATET or CATET
Treated vs untreated regions	-29401.111	-1.1650436**	-4592.5248**	− .00492499*	− .03186035***	-629.18586***	− .049**
1996	6690**				-0.0045002	-259.30395	.0064782*
1997	10230**			0.00385154	-0.00703847	-332.96859	.0391751***
1998	21495***			0.00627141	-0.00637149	-115.83798	.0688513***
1999	33675***			0.00675496	-0.00703858	93.668916	.0924181***
2000	55815***			.01882805***	-0.01289044	601.68919	.118144***
2001	78415***		(base)	0.00002573	-0.01804747	656.04639	.1233866***
2002	98220***		362.62472*	− .0215791***	-0.01728926	382.62437	.1187407***
2003	115750***		1009.5801**	− .0232905**	-0.01858051	73.910492	.1240574***
2004	123135***	-1.3032172***	2759.9581***	− .01468758**	-0.02405906	-107.40569	.1329757***
2005	129990***	-0.79481413	3532.96***	− .01878086**	-0.02915511	-312.69061	.1399013***
2006	154155***	-1.1050093**	4066.7757***	-0.00507515	-0.03699559	-116.92608	.1539774***
2007	176985.28***	-1.7357198***	6032.483***	-0.01214461	-0.03874279	-2.1277481	.156496***
2008	179790.28***	-1.2009668**	6118.9498***	− .04189799***	-0.04368337	-687.60307	.1438508***
2009	158895.28***	-1.4413136***	6257.9583**	− .07688143***	-0.03531405	-1955.6365	.1276458***
2010	150955.28***	-1.6152737***	6245.3666**	-0.01254494	-0.04403754	-1832.6498	.143082***
2011	154300.28***	-0.57854018	5098.0129*	− .01639645*	-0.05142562	-1823.484	.1476519**
2012	150960.28***	-2.1962287***	5637.865*	− .04757515***	-0.04597545	-2314.2495	.1080306***
2013	128780.28***	-1.896113**	4588.6282	− .04166368***	-0.04600308	-2823.6999*	.0820907***
2014	129685.28***	-2.5526536***	6938.0525**	− .02099425**	-0.04919986	-2855.4027*	.0806337***
2015	137625.28***	-2.0213505**	4246.2576	-0.00320001	-0.0591872	-2712.2537*	.101242***7
2016	154245.28***	-1.8697245**	3578.8324	-0.00947221	− .06638269*	-2543.8082	.1122534***
2017	168730.28***	-1.2597971	4181.602	-0.00137235	− .07464948*	-2204.5669	.1276559***
2018	180345.28***	-2.1817539**	3998.3382	-0.00729039	− .08428351**	-2043.9405	.1382362***
2019	186980.28***	-1.5534404*	3778.6483	-0.01148007	− .09223028**	-1955.0961	.1414657**
2020	159470.28***	-3.1028488***	2645.0976	− .10811968***	− .07597091**	-3679.2373**	.0235429
2021	166375.28***		3013.8352	.04887077***	− .09266339**	-2413.492	.0738984**
Controls
log_K		-0.79632567	-453.96235	0.00989756	.08774462***	2241.2117***
log_Val		-11.716371***	-13460.289	0.03106403	.23681953**	9867.1645**
_cons	1095230***	160.11159***	178491.2	− .54361653*	5.672762***	-128581.29***	10.29218***
N	540	355	420	520	540	540	540
Ptrend Test: H₀ Linear parallel trend	F(1,19) = 0.04	F(1, 19) = 2.30	F(1, 19) = 2.02	F(1, 19) = 0.02	F(1, 19) = 1.18	F(1, 19) = 0.34 Prob > F = 0.5674 H₀ cannot be rejected F(11, 19) = 7.29 Prob > F = 0.0001 Reject H₀	F(1, 19) = 0.62 Prob > F = 0.4398 H0 cannot be rejected F(11, 19) = 7.12 Prob > F = 0.0001 Reject H₀
	Prob > F = 0.8459	Prob > F = 0.1455	Prob > F = 0.1715	Prob > F = 0.9030	Prob > F = 0.2903
	H₀ cannot be rejected	H₀: cannot be rejected	H₀: cannot be rejected	H₀: cannot be rejected	H₀ cannot be rejected
Granger Type Test: H₀ No Anticipation effects	F(11, 19) = 4.00	F(3, 19) = 1.50	F(5, 19) = 1.34	F(10, 19) = 9.99	F(11, 19) = 11.29
	Prob > F = 0.0040	Prob > F = 0.2466	Prob > F = 0.2893	Prob > F = 0.0000	Prob > F = 0.0000
	Reject H₀	H₀ cannot be rejected	H₀ cannot be rejected	Reject H₀	Reject H₀
Notes. Results from DID estimation of Eq. (1) with robust SE. H₀ in the Ptrend test is that pre-treatment linear trends are parallel. Granger-type test is for anticipatory effects. H₀ is that there are no effects in anticipation of treatment. In the fourth and fifth columns, the dependent variable is the logged per-capita income of the regional residents. We used the original logged series and the Hodrick–Prescott high-pass filter to separate the time series into trend and cyclical components. The trend component may contain a deterministic or a stochastic trend. The smoothing parameter determines the periods of the stochastic cycles that drive the stationary cyclical component. With the exception of the first, columns report CATET estimates obtained using Controls in addition to Time. ATET estimates (not reported) are very similar. Data source: see previous Table.

Before commenting each specific results, we discuss the test for the presence of the parallel trend and anticipation effects. The last two rows of Table 2 report the estimated F-tests for parallel trend hypothesis and the test for absence of anticipation effects. The former is based on the results of an augmented linear DID model based on our Eq. (1) with additional terms that capture the differences in slopes between treated and controls. As for anticipation effects, Table 2 reports the results of a Granger type test obtained from a linear DID model based once again on our Eq. (1) augment in this case by interacting the dummy variable that marks treated observations with dummy variables for time periods prior to the treatment to capture any potential anticipatory treatment effects. Anticipation effects are import, particularly for policy interpretation of the results.

The results of the above tests allow us to conclude (at any significance level) that data are consistent with the hypothesis of parallel trend but not always with the absence of anticipation effects hypothesis. Notwithstanding the importance of these statistical tests, the visual inspection of the plots of the series (Fig. 1) remains of paramount importance. Yet, one should recall that the Granger-type test is less robust than the parallel trend F-test since it consumes more degrees of freedom because it estimates a higher number of parameters. Hence, the parallel trends test has higher statistical power. Test indications will be commented with reference to the results obtained for each response variables.

4.1 Total Employment, Self-Employment, and New VAT Certificates

If the determinants of employment decisions are based on the hypothesis that the individual responds to the risk adjusted relative earnings opportunities, then income tax changes should affect their behaviour at both intensive and extensive level. (Saez, 2002, p. 1043) rightly emphasises that the shorter the time period upon which the income tax is assessed the more relevant is the extensive effect of the tax changes. The results reported in the first three columns of Table 2 are obtained for response variables representing possible different ways of measuring labour supply at the extensive margin. The first two are the annual levels of Total Employment and Self-Employment (the latter measured as the share of self-employed people out of the total annual employment in the region). We use Total Employment rather than Labour Force Participation because the latter includes unemployed people who are actively searching for work or only temporarily laid off. While labour force participation might have the advantage of not changing with transitions between employment and unemployment, Total Employment is more relevant for welfare calculations and tax incentive analysis. Then, a third variable (New VAT certificates) is studied to include estimates for a response variable that, as it will be discussed below, is indirectly related to the binary participation decision based on after-tax income.

The first two response variables react to the income tax treatment differently: Total Employment does not appear to be affected by the tax increase whereas Self-Employment does. Self-Employed people are the taxpayers mostly affected by the 2007 tax treatment since they will pay both taxes whose rates have being increased in the five treated regions (see Appendix). Then, our DID estimations show that the share of Self-Employed workers is reduced by the increase of income taxation. The estimated negative effect on Self Employment and the non-significant effect on Total Employment are consistent with the observed temporal trends reported in Appendix 2. In Figure A5, we report the absolute variation between 2006 and 2009 of the variables examined, while in Figure A6 we show the time series of index numbers with base year 2006 (the time stamp prior to the policy introduction). We can clearly notice that for self-employment, the strongest reductions were registered within the five treated regions (the colours are more markedly green) compared to the rest of the country; furthermore, the index numbers show how all treated regions suffered a reduction between 20% and 30% in the four years following the reform, with a partial but homogenous recovery from 2011 onwards, while the average of the untreated regions actually increased. The same charts also reveal that the total regional employment has changed heterogeneously in the treated sample, with Lazio increasing (+ 3%) and Campania and Sicily experiencing a marked reduction (-6% and − 3% in 2009, respectively), while Abruzzo remained substantially unchanged. Such territorial heterogeneity may have internally balanced the policy effect, leading to estimates that are not statistically significant. These results accord with previous literature (e.g., Heim, 2010), including those models which consider explicitly how stochastic shocks in demand or cost functions affect the decision to become self-employed (Appelbaum & Katz, 1986; Kihlstrom & Laffont, 1979). In this type of models, it is obvious that an asymmetric tax reduction, favouring the self-employed relative to other workers, increases self-employment. The converse result represented by the tax treatment studied here (tax increase) is not a surprise⁴. Other authors put emphasis on other tax opportunities open to self-employed. Robson and Wren (1999) state that higher tax rates lead individuals to take up self-employment to take advantage of the greater opportunity for tax avoidance and evasion that self-employment offer relative to paid employment (i.e., pure wage earners). Given the high level of tax evasion in Italy, we take this argument very seriously.

Anticipatory effects on Self-Employment should be excluded as well as absence of parallel trends (see the last two rows of Table 2). The plot of Fig. 3 shows that the impact of the tax treatment on Self-Employment is statistically significant in the first year of the treatment and during the next 3 years but coefficients are no longer significant after the initial period. Then, the tax effect loses strength and significance as we move away from the initial year of the treatment. Still, results indicate that income tax changes affect the level of self-employment in the treated areas although the effect is short time-lasting. More recently, Heim (2010) estimates the extent to which self-employment income responds to changes in the net-of-tax share using a panel of tax returns that spans 1987–1996. The results suggest that the elasticity of reported self-employment income to the net-of-tax share is approximately 0.9, implying a real elasticity (net of any reporting response) of around one. Reported estimated elasticities tend to be larger for higher income taxpayers, married males, and females. In addition, the elasticity of self-employment income is considerably larger than the elasticity wage and salary income estimated using the same methodology.

The third response variable is the New VAT Certificates. Operators who want to start legally a (new) business activity must obtain a VAT certificate from tax authorities. Hence, this flow variable measures the net increase of economic operators requiring and obtaining a formal VAT Certificate from the tax authorities, net of the number of certificates returned back to tax authorities or extinguished during each year. Clearly, employees are excluded from this set of taxpayers. In a DID model a CATET/ATET parameter will correspond to the estimate of the effect of the income tax treatment on the (official) net flow of new labour supply provided in each region by professional business operators. Then, observing the post treatment behaviour of this response variable and the difference between treated and untreated (future) taxpayers may provide indication on the extensive effect of income taxation on that specific set of taxpayers. The variable clearly indicates the intention of some economic operators in some segments of the potential labour force of starting a professional business activity (plumbers, mechanics, shop owners, engineers, architects, etc.). Those who apply for a VAT certificate/permission are subjected to the payment of both IRAP and IRPEF. Therefore, they represent the labourer category most affected by the tax treatment of 2007. How this variable reacts to the tax treatment permits us to interpret the tax treatment results in terms of extensive margin reaction to taxation or participation decisions for we interpreted the realization of the variable as the result of a binary choice: either apply or not apply for the certificate as a result of the tax change. In this simplification (which excludes options such as running the business in the black market), the only decision is a participation decision which, in turn, only depends on the relative after-tax income when he works or when he is unemployed. As stressed Saez (2002, p. 1048) this view is obviously a crude simplification of reality but captures the extensive margin labour supply decision. If one assumes away income effects, the decision to participate depends only on the difference between (after tax) income with participation and unemployment transfers without participation.

The estimation of the tax treatment indicates that an average decrease of the new applications for the VAT Certificates in the range of 4600 units (about the 6% of the average 2006 value), which is the estimate of the average annual reduction of new (official) operators in the treated regions with respect to the untreated ones. This means that the tax treatment has discouraged new professionals to enter the relevant markets in treated regions or has led some of the already active operators to leave the (official) markets.

Once again, the test reported in the last rows of Table 2 suggest that we cannot reject the null that trends are parallel and that anticipatory effects are absent even for this variable. Recall that operators (old and new) requiring a VAT Certificate to carry a business are the main IRAP taxpayers, and the result seems to indicate that the tax base of IRAP is rigid to anticipatory adjustments.

As for the effect of the tax treatment over time, the plots of the Betas coefficients shown in Fig. 3 indicate that the effect of the tax treatment on the supply of professional services is strong, statistically significant (slightly at the 95% all over the sample), and not restricted to the very short run. The latter remark is strongly justified by the observed graphical tendencies. The index numbers, in particular, clearly show that the pattern of the treated group and of the control group are widely divergent and that the regions affected by the tax policy have experienced large reductions (around 20% in 2009) in the number of VAT certificates compared to the rest of Italy.

Another way to interpret the results starts from the acknowledgement that VAT Certificates are necessary to start company activities and not only individual professional business. Then, we may evaluate the empirical results in terms of the effects of income tax changes on creation of new companies by entrepreneurs who plan to exploit some business opportunities (a fundamental force for economic growth first identified by Schumpeter and formalized by Aghion and Howitt (1990). Economic policies aimed at fostering the entry of new companies may be potentially benefits for innovation, competition, employment, and growth (see Aghion & Howitt, 2006). In the case studied here we should conclude that the above DID results parallel those obtained by some recent studies which include taxation as a determinant of the incorporation decision (see Barrios, Huizinga, Laeven, & Nicodème, 2009; Djankov, Ganser, McLiesh, Ramalho, & Shleifer, 2010) or some more recent contributions in which taxation is not seen as an exogenous policy instrument (as it is done in my DID study), but that it is seen as an instrument that rather reacts to current (or past) business conditions (Da Rin, Di Giacomo, & Sembenelli, 2011). Altogether, our results show that self-employment might be reduced by income tax increases. As in Wen and Gordon (2014) where in a probit model the tax convexity variable and the net-of-tax income difference between self- and paid employment have the predicted (negative) signs and there is a high level of statistical significance for the probability of self-employment, in our estimation self-employment significatively reacts to income tax increase. The result obtained for the variable New VAT Certificates confirm that income tax increases are perceived as disincentives to entry the formal self-employment professional markets and produce negative effects at the extensive margin.

4.2 Income tax effects on per-capita income

Three income response variables are considered: Regional GDP Growth per capita, Regional GDP Growth per capita (Hodrick-Prescott filtered), and Regional GDP per capita in levels. The graphical trends (Figures A5 and A6) of the three indicators show a marked generalised reduction following the introduction of the policy. While the average of the untreated regions shows a pro-cyclical pattern with quick alternations of growth and stagnation, Sicily, Campania and Molise follow a smooth process of GDP loss that is not reabsorbed in the following years. The models support these tendencies, as the DID estimates confirm that the three response variables negatively react to the income tax treatment. For the growth variables we use both the original logged series and the Hodrick–Prescott high-pass filter to separate the time series into trend and cyclical components. Conditioning factors are the log of Gross Capital and the log of the Value Added per Worker (average labour productivity). We find that personal income growth is negatively affected by income tax increase and this result accords with the previous extensive labour supply contraction shown in the first three columns of Table 2. As for the values of per-capita income in levels, results show that regional per capita income is reduced by the treatment by (approximately) a value corresponding to a 2.3% of the national mean value reported in Table 1 for the entire data set and to a 2.7% of the per capita income measured in treated regions only. We have also explored if per capita income is parallel for the control and treated regions prior to 2007 by checking for the parallel trend assumption. Test results show that anticipated adjustments cannot be excluded. Recall that the per capita income proxies the income tax base. So, some taxpayers in treated regions may have anticipated the tax treatment (particularly the tax change hitting income from business activity via IRAP) and taken advantage of any existing opportunities to plan in advance tax reducing strategies, possibly coordinated by well-informed professional advisers.

We may contrast the above results with Zidar (2019, p. 1423). A clear similarity emerges between our response results and his, although, in our case, estimates indicate a more pronounced decline of the response variables in treated regions. This quantitative difference can be interpreted by recalling that the tax treatment consisted here is the increase of the tax rates of two taxes affecting net income of a large set of taxpayers: IRAP (paid by personal business and professionals, but not by employees) and the higher regional income surcharge levied by treated regions on the generality of the taxpayers resident in those regions (see appendix 1 for the list). The exogenous increase of both taxes (the regional income Surcharge and the increase of the business quasi-income tax IRAP), taken in isolation, apparently reduces the post treatment regional GDP levels and growth over time. Clearly this drastic conclusion should be moderated by recalling that the effects on income and growth, via the local multiplier effects, should be evaluated by weighting the effects of both taxes and expenditures (Ramey, 2011a, 2011b).

As for the effects of the treatment over time, the plots show the time value of the treatment coefficient.

The impact on per capita GDP growth is not strong at the beginning but it has leads effects during the years following the treatment. These effects become stronger and statistically significant as time passes.

4.3 Family consumption

Turning to consumption, we observe that log estimates (last column of Table 2) show a negative value of the overall consumption elasticity in the treated region to be approximately 0.5%. Yet, this finding is consistent with the results commented above about the tax effects on the GDP per capita and with the basic hypotheses about marginal propensities to consume. The 2007 tax treatment included the increase of the regional Income Tax Surcharge and consequently it made overall (national and regional) income taxation more progressive in treated regions. Thus, by changing the disposable income distribution in treated regions, a more redistributive taxation may have changed the relative aggregate demand for goods and services and the estimated elasticity of Table 2 can interpreted as the net aggregate result generated by these changes. This in turn may have lead to changes in the demand for workers in various sectors, particularly those offering more income-elastic goods. With non-homothetic consumption preferences, when we allow consumption to adjust (i.e., when income effects are at work), any change in progressivity differentially affects the labour supply of skilled and unskilled workers because the compositions of agents' consumption baskets change with their disposable income. If this is the case, the marginal utility of income may vary across income groups, distinctively affecting the incentives to work of each group (Oni, 2023, p. 4). Moreover, if we look at the two plots of Fig. 2 and at the Granger-type test results reported at the end of Table 2 (the null of absence of anticipatory effects of tax treatment on consumption must be rejected), we find that in our data set the permanent income hypothesis is not consistent with the data⁵.

The length of exposure of consumption to the tax treatment is shown in Fig. 2 below.

The tax shock has a strong effect during the first years following the treatment and then loses intensity and significance (at the 95%). Similar deductions can be made from the observed trends in household consumption (Figures A5 and A6), which show that the treated regions have comparable behaviour to that of regional GDP. In particular, the graphs suggest that the fall in the treated regions was stronger than in the control group between 2008 and 2011, being partially reabsorbed by some regions (Lazio, Sicily and Campania did not catch up). This may suggest that durable consumption might be affected by the treatment at the beginning of the post treatment period and then, since current consumption may be measured with error, its point estimates may be less reliable than other response variables such as, for instance, self-employment. However, the results are somewhat not clear-cut for no distinction is made for durable goods consumption vs current consumption and, above all, for different income groups. However, we should recall that we are evaluating the effects of two tax increases: one affecting the generality of income taxpayers (with an increase of the marginal income tax rate of 0.8%) and one hitting personal business (with an increase of the marginal income tax rate of 1%) we may hazard the interpretation, absent any other reliable information, that the reduction of consumption is mainly attributable to the taxpayers paying both income taxes, and therefore to those belonging to the set of taxpayers who obtain their income from business activities.

There are two key threats to the validity of the estimates of the tax shock: are tax changes really endogenous and do differential pretreatment trends exist? The nature and the mechanism of the Italian 2007 income tax experiment considered here exclude the first threat but not necessarily the second. As it was discussed earlier, the presence of different trends may jeopardise the parallel trend assumption of a DID. Given the macroeconomic nature of our regional data set, the presence of converging trends of response variables (for instance, mean values of Sicilian and Piemonte’s family consumption reduce their time gaps over time) cannot be excluded a priori. Hence in addition to the tests reported in Table 2 and to the visual inspection of the plots of Fig. 1, we propose a falsification test that excludes that the response variables follow some converging trend. To this end a log(t) convergence club analysis is conducted on the available data. This analysis is important from the parallel trends perspective because parallel trends require that, in the absence of treatment, the difference between the treatment and control group is constant over time. As it was already stressed, violation of parallel trend assumption will lead to biased estimation of the causal effect. Admittedly, it can be assumed that the smaller the time period tested (for example a 2-year period), the more likely the assumption is to hold, but with a period of 27 years the possibility that treated and control groups have different outcome trends (which may generate time convergence among them) cannot be excluded. Time convergence, i.e. converging trends of the response variables for treated and control groups (employment, labour supply, consumption, etc.), would then be at odds with the hypothesis of parallel trends. This is so because the change (or “path”) in outcomes over time that units in the treated group would have experienced if they had not been included in the treatment may not be the same as the path of outcomes that units in the control group actually experienced.

To test for the presence of convergence of each of the response variables analysed by the DID we perform an econometric convergence analysis following Phillips and Sul (2007, 2009) who proposed the novel “log t” regression test approach to test the convergence hypothesis based on a nonlinear time-varying factor model. The proposed approach has the following merits: first, it accommodates heterogeneous agent behaviour and evolution in that behaviour. Second, the proposed test does not impose any particular assumptions concerning trend stationarity or stochastic non-stationarity and consequently it is robust to the stationarity property of the series. In other words, it could be used whatever the unit root results reported in Table 1. Yet, to make the test more adherent to the time span of the tax treatment we conduct it a) using the entire sample period (1995–2021) and for the pre and post treatment periods and for the two sets (treated and control) of data. Results of log(t) tests are reported below.

Table 3

a: Phillips-Sul log(t) tests of convergence. t-stat in parentheses (H0 is that there is convergence)
Response Variables (treated and untreated regions)	1995–2021	1995–2006	2007–2021
Total Employment	-0.7846 (-185.436) Rejected	-0-7846 (-185.436) Rejected	-0.9543 (-44.2028) Rejected
Self-Employment	-1.0471 (-3.9066) Rejected	-1-0471 (-3.9066) Rejected	-1.3775 (-5.4437) Rejected
New VAT Certificates	-0.8194 (-37.4199) Rejected		-0.9361 (-55.5262) Rejected
Regional GDP Growth p-c	2.7964 (1.85) H₀ Not Rejected	-4.6937 (-2.0528) H₀: Rejected	3.3804 (2.1556) H₀ Not rejected
log Regional GDP Growth p-c HP filter)	-0.8112 (-253.81) Rejected	-0.9415 (-40.8417) Rejected	-0.9571 (-53.9166) Rejected
Regional GDP p-c	-0.8233 (-23.3134) Rejected	-0.9610 (-26.2895) Rejected	-0.8512 (-32.0493) Rejected
Family Consumption	-0.8788 (-82.5425) Rejected	-1.0069 (-53.7084) Rejected	-0.9222 (-37.5706) Rejected

Notes. The Table shows the results of a log(t) test for overall convergence of the response variables used in the DID analysis (see Table 1) proposed by Phillips and Sul (2007, 2009) in order to investigate the presence of convergence by adopting the Andrews estimator of long-run variance (fixed or adaptive bandwidth of the kernel) based on Quadratic Spectral kernel. If the t-test of log(t) test (reported in parentheses) is smaller than − 1.65, then the null hypothesis of convergence of the variable is rejected at 5% CI. Data availability of New VAT Certificates permits to run the test for the post treatment period only. Results show that during the pre-treatment period the hypothesis of convergence is not consistent with the data and this finding does not contradict the DID parallel trend hypothesis. In one case (regional GDP growth per capita), post treatment period and the entire period show convergence.

Table 3

b: Phillips-Sul log(t) tests of convergence. t-stat in parentheses (Untreated regions only)
Response Variables (untreated regions)	1995–2021	1995 2006	2007–2021
Total Employment	-0.7667 (-160.2177) Rejected	-1.0048 (-33.1669) Rejected	-0.9503 (-44.9480) Rejected
Self-Employment	-1.2245 (-6.6226) Rejected	-2.6070 (-18.1581) Rejected	-1.0843 (-4.5208) Rejected
New VAT Certificates	-0.7446 (-30.1970) Rejected		-0.9188 (-48.3598) Rejected
Regional GDP Growth p-c	2.2234 (1.4464) H₀ not rejected	-4.6663 (-2.2425) Rejected	2.8482 (1.2841) H₀ not rejected
log Regional GDP Growth p-c HP filter)	-0.7526 (-153.3184) Rejected	-0.9037 (-37.4473) Rejected	-0.9260 (-50.2768) Rejected
Regional GDP p-c	-0.7894 (-18.6610) Rejected	-0.9299 ( -23.6503) Rejected	-0.8154 (-34.8615) Rejected
Family Consumption	-0.8783 (-78.1308) Rejected	-1.0022 (-54.3171) Rejected	-0.9188 (-40.0264) Rejected

Notes. The Table shows the result of log(t) test of convergence for the untreated regions only. Data availability of New VAT Certificates permits to run the test for the post treatment period only. See notes below Table 2 for interpretation.

Table 3

c: Phillips-Sul log(t) tests of convergence. t-stat in parentheses (Treated regions only)
Response Variables (untreated regions)	1995–2021	1995 2006	2007–2021
Total Employment	-0.8333 (-137.0381) Rejected	-1.0310 (-34.5509) Rejected	-0.9679 (-39.6925) Rejected
Self-Employment	-2.1374 (-2.3277) Rejected	-2.0493 (-1.5835) H₀ not rejected	-1.0843 (-4.5208) Rejected
New VAT Certificates	-0.8244 (-62.8409) Rejected		-0.9373 (-35.9240) Rejected
Regional GDP Growth p-c	3.2923 (2.1344) H₀ not rejected	-3.8758 (-1.6369) H₀ not rejected	1.4841 (0.8986) H₀ not rejected
log Regional GDP Growth p-c HP filter)	-0.7645 (-44.4023) Rejected	-1.0604 (-44.7164) Rejected	-0.8736 (-35.9767) Rejected
Regional GDP p-c	-0.6442 (-13.5016) Rejected	-1.2055 (-24.0175) Rejected	-0.7757 (-19.9970) Rejected
Family Consumption	-0.8118 (-82.0494) Rejected	-1.0345 (-51.2356) Rejected	-0.9468 (-33.0851) Rejected

Notes. The table shows the result of log(t) test of convergence for the untreated regions only. Data availability of New VAT Certificates permits to run the test for the post treatment period only. See notes below Table 2 for interpretation. The Table shows that convergence seems consistent with the data for Self-Employment among Treated regions during the pre-treatment period. If the t-test of log(t) test is smaller than − 1.65, then the null of convergence of the variable is rejected at 5% CI. Data availability of New VAT Certificates permits to run the test for the post treatment period only.

Results show that during the pre-treatment period the hypothesis of convergence is not consistent with the data and this finding accords with the DID parallel trend hypothesis. In one case (regional GDP growth per capita), the null of convergence cannot be rejected at usual significance levels for the entire time period and for the post treatment period (this contradicts the P-trend test reported in Table 2 for that variable) but no convergence is found for the pretreatment period.

Obtaining the rejection of the null of convergence of the response variables when pooling treated and control observations just indicate that the series do not converge. However, it does not exclude that trends are not parallel because the series diverge. Yet, a different result (not rejecting the null hypothesis of convergence) would be a contradiction of the parallel trend hypothesis and that is why we propose this test in addition to those reported in Table 2.

DID results illustrated in this paper provide robust evidence that aggregate regional economic variables related to growth are substantially responsive to income tax changes. However, spatial heterogeneity within the treatment (and control) groups may result in the offsets of causal effects and consequently to poorly significant aggregate effects. Large labour supply extensive responses to income tax treatment are the main finding of our DID analysis. They are consistent with the evidence presented on labour force participation by previous literature that the supply of professional self-employed person is adversely affected by the tax increase. The share of self-employed out of total occupation is immediately reduced by the exogenous increase of income taxation but apparently, the effect is not permanent. In our estimates, readjustments of that segment of the labour supply to a long-run trend are observable 3 years after the treatment. Anticipatory effects are not consistent with the data and this may be relevant in terms of policy measures. Moreover, the absence of anticipatory effects somehow implies that the tax base of one of the two taxes analysed in this paper (a quasi-income business tax called IRAP in the Italian tax legislation) may be more rigid to anticipatory adjustments than the tax base of the general personal income tax (called IRPEF in the Italian tax legislation). The specific result about New VAT Certificates indicates that exogenous modification of regional taxation has robust extensive repercussion on the supply of professional activities, possibly delating or cancelling the decisions of potential operators to enter their specific official markets or leading some of them to enter the underground economy. Results on per-capita income and family consumption are coherent with the above findings since similar adjustment responses characterise the reactions to exogeneous income tax changes of the growth of per-capita regional GDP and family consumption.

Author Contribution

The authors contributed equally to the submitted manuscript.

Acknowledgments

This paper is part of a research project on income taxation and growth in Italy. We owe the idea of using the tax experiment of 2007 to Alessandro Santoro from the University of Milano-Bicocca. Details about the Italian budget law of December 2006 (containing the tax rules used in this paper) were kindly provided by Enza Caruso and Giuseppe Pisauro.

Conflict of interest/Competing interests: The authors have no competing interests to declare that are relevant to the content of this article.

Funding: The authors did not receive support from any organization for the submitted work.

Data availability and codes: All results presented in this paper can be reproduced using the R and Stata 18 software. The codes were developed entirely by the authors. Attached to the submission files, we attach the complete set of dataset and scripts. All scripts and the data will be made available on a dedicated GitHub at the end of the peer-review process.

Figures: All images included in the paper were created by the authors and do not require any publication permission.

Aghion, P., & Howitt, P. (1990). A model of growth through creative destruction. In: National Bureau of Economic Research Cambridge, Mass., USA.
Aghion, P., & Howitt, P. (2006). Appropriate growth policy: A unifying framework. Journal of the european Economic Association, 4(2-3), 269-314.
Akcigit, U., Baslandze, S., & Stantcheva, S. (2016). Taxation and the International Mobility of Inventors. American economic review, 106(10), 2930-2981. doi:10.1257/aer.20150237
Appelbaum, E., & Katz, E. (1986). Measures of Risk Aversion and Comparative Statics of Industry Equilibrium. American economic review, 76(3), 524-529. Retrieved from http://www.jstor.org/stable/1813369
Auerbach, A. J., & Siegel, J. M. (2000). Capital-Gains Realizations of the Rich and Sophisticated. American economic review, 90(2), 276-282. doi:10.1257/aer.90.2.276
Barrios, S., Huizinga, H., Laeven, L., & Nicodème, G. (2009). International taxation and multinational firm location decisions. Retrieved from https://EconPapers.repec.org/RePEc:euf:ecopap:0356
Baskaran, T. (2021). The revenue and base effects of local tax hikes: evidence from a quasi-experiment. International Tax and Public Finance, 28(6), 1472-1518. doi:10.1007/s10797-021-09657-2
Blau, D. M. (1987). A Time-Series Analysis of Self-Employment in the United States. Journal of Political Economy, 95(3), 445-467. Retrieved from http://www.jstor.org/stable/1831972
Caruso, E., & Dirindin, N. (2019). Salute ed Economia-Questioni di economia e politica sanitaria: Il mulino.
Da Rin, M., Di Giacomo, M., & Sembenelli, A. (2011). Entrepreneurship, firm entry, and the taxation of corporate income: Evidence from Europe. Journal of Public Economics, 95(9), 1048-1066. doi:https://doi.org/10.1016/j.jpubeco.2010.06.010
De Philippis, M., Locatelli, A., Papini, G., & Torrini, R. (2022). La crescita dell’economia italiana e il divario Nord-Sud: trend storici e prospettive alla luce dei recenti scenari demografici [Italian economic growth and the North-South gap: Historical trends and future projections in light of the recent demographic scenarios]. Bank of Italy Occasional Paper(683).
Djankov, S., Ganser, T., McLiesh, C., Ramalho, R., & Shleifer, A. (2010). The Effect of Corporate Taxes on Investment and Entrepreneurship. American Economic Journal: Macroeconomics, 2(3), 31-64. doi:10.1257/mac.2.3.31
Dynan, K. E., Skinner, J., & Zeldes, S. P. (2004). Do the Rich Save More? Journal of Political Economy, 112(2), 397-444. doi:10.1086/381475
Eissa, N., & Liebman, J. B. (1996). Labor Supply Response to the Earned Income Tax Credit*. The quarterly journal of economics, 111(2), 605-637. doi:10.2307/2946689
Esteller-Moré, A., & Solé-Ollé, A. (2002). Tax Setting in a Federal System: The Case of Personal Income Taxation in Canada. International Tax and Public Finance, 9(3), 235-257. doi:10.1023/A:1016212110137
Evans, D. S., & Leighton, L. S. (1989). The determinants of changes in U.S. self-employment, 1968–1987. Small Business Economics, 1(2), 111-119. doi:10.1007/BF00398629
Federici, D., Ferrante, F., & Parisi, V. (2023). The educational roots of persistent territorial disparities: the case of Italy. GeoJournal, 88(2), 2063-2082. doi:10.1007/s10708-022-10740-2
Goolsbee, A. (2000). What Happens When You Tax the Rich? Evidence from Executive Compensation. Journal of Political Economy, 108(2), 352-378. doi:10.1086/262122
Haaf, K., & Kool, C. (2017). Determinants of regional growth and convergence in Germany. USE Discussion Paper Series, 17(12).
Harris, R. (2011). Models of regional growth: past, present and future. Journal of Economic Surveys, 25(5), 913-951. doi:https://doi.org/10.1111/j.1467-6419.2010.00630.x
Harris, R. D. F., & Tzavalis, E. (1999). Inference for unit roots in dynamic panels where the time dimension is fixed. Journal of Econometrics, 91(2), 201-226. doi:https://doi.org/10.1016/S0304-4076(98)00076-1
Heim, B. T. (2010). The responsiveness of self-employment income to tax rate changes. Labour Economics, 17(6), 940-950. doi:https://doi.org/10.1016/j.labeco.2010.02.010
Higgins, M. J., Levy, D., & Young, A. T. (2006). Growth and Convergence across the United States: Evidence from County-Level Data. The Review of Economics and Statistics, 88(4), 671-681. doi:10.1162/rest.88.4.671
Jakobsen, K., Jakobsen, K., Kleven, H., & Zucman, G. (2019). Wealth Taxation and Wealth Accumulation: Theory and Evidence From Denmark*. The quarterly journal of economics, 135(1), 329-388. doi:10.1093/qje/qjz032
Jappelli, T., & Pistaferri, L. (2010). The Consumption Response to Income Changes. Annual Review of Economics, 2(1), 479-506. doi:10.1146/annurev.economics.050708.142933
Johnson, D. S., Parker, J. A., & Souleles, N. S. (2006). Household Expenditure and the Income Tax Rebates of 2001. American economic review, 96(5), 1589-1610. doi:10.1257/aer.96.5.1589
Kihlstrom, R. E., & Laffont, J.-J. (1979). A General Equilibrium Entrepreneurial Theory of Firm Formation Based on Risk Aversion. Journal of Political Economy, 87(4), 719-748. doi:10.1086/260790
Kleven, H. J., Landais, C., & Saez, E. (2013). Taxation and International Migration of Superstars: Evidence from the European Football Market. American economic review, 103(5), 1892-1924. doi:10.1257/aer.103.5.1892
Levin, A., Lin, C.-F., & Chu, C.-S. J. (2002). Unit root tests in panel data: asymptotic and finite-sample properties. Journal of Econometrics, 108(1), 1-24.
Lo Cascio, I., Mazzola, F., & Epifanio, R. (2019). Territorial determinants and NUTS 3 regional performance: A spatial analysis for Italy across the crisis. Papers in Regional Science, 98(2), 641-678. doi:https://doi.org/10.1111/pirs.12372
McCarthy, J. (1995). Imperfect insurance and differing propensities to consume across households. Journal of Monetary economics, 36(2), 301-327. doi:https://doi.org/10.1016/0304-3932(95)01214-1
Meyer, B. D., & Rosenbaum, D. T. (2001). Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers*. The quarterly journal of economics, 116(3), 1063-1114. doi:10.1162/00335530152466313
Moretti, E., & Wilson, D. J. (2017). The Effect of State Taxes on the Geographical Location of Top Earners: Evidence from Star Scientists. American economic review, 107(7), 1858-1903. doi:10.1257/aer.20150508
Nguyen, A. D. M., Onnis, L., & Rossi, R. (2021). The Macroeconomic Effects of Income and Consumption Tax Changes. American Economic Journal: Economic Policy, 13(2), 439-466. doi:10.1257/pol.20170241
Oni, M. H. (2023). Progressive income taxation and consumption baskets of rich and poor. Journal of Economic Dynamics and Control, 157, 104758. doi:https://doi.org/10.1016/j.jedc.2023.104758
Parker, J. A. (1999). The Reaction of Household Consumption to Predictable Changes in Social Security Taxes. American economic review, 89(4), 959-973. doi:10.1257/aer.89.4.959
Parker, J. A., Souleles, N. S., Johnson, D. S., & McClelland, R. (2013). Consumer Spending and the Economic Stimulus Payments of 2008. American economic review, 103(6), 2530-2553. doi:10.1257/aer.103.6.2530
Pesaran, M. H. (2007). A simple panel unit root test in the presence of cross‐section dependence. Journal of applied econometrics, 22(2), 265-312.
Phillips, P. C. B., & Sul, D. (2007). Transition Modeling and Econometric Convergence Tests. Econometrica, 75(6), 1771-1855. doi:https://doi.org/10.1111/j.1468-0262.2007.00811.x
Phillips, P. C. B., & Sul, D. (2009). Economic transition and growth. Journal of applied econometrics, 24(7), 1153-1185. doi:https://doi.org/10.1002/jae.1080
Piketty, T., Saez, E., & Stantcheva, S. (2014). Optimal Taxation of Top Labor Incomes: A Tale of Three Elasticities. American Economic Journal: Economic Policy, 6(1), 230-271. doi:10.1257/pol.6.1.230
Ramey, V. A. (2011a). Can Government Purchases Stimulate the Economy? Journal of Economic Literature, 49(3), 673-685. doi:10.1257/jel.49.3.673
Ramey, V. A. (2011b). Identifying Government Spending Shocks: It's all in the Timing*. The quarterly journal of economics, 126(1), 1-50. doi:10.1093/qje/qjq008
Robson, M. T., & Wren, C. (1999). Marginal and Average Tax Rates and the Incentive for Self-Employment. Southern Economic Journal, 65(4), 757-773. doi:https://doi.org/10.1002/j.2325-8012.1999.tb00198.x
Romer, C. D., & Romer, D. H. (2010). The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks. American economic review, 100(3), 763-801. doi:10.1257/aer.100.3.763
Rubolino, E., & Giommoni, T. (2023). Taxation and mobility: Evidence from tax decentralization in italy.
Saez, E. (2002). Optimal Income Transfer Programs: Intensive versus Extensive Labor Supply Responses*. The quarterly journal of economics, 117(3), 1039-1073. doi:10.1162/003355302760193959
Saez, E., Matsaganis, M., & Tsakloglou, P. (2012). Earnings Determination and Taxes: Evidence From a Cohort-Based Payroll Tax Reform in Greece *. The quarterly journal of economics, 127(1), 493-533. doi:10.1093/qje/qjr052
Souleles, N. S. (1999). The Response of Household Consumption to Income Tax Refunds. American economic review, 89(4), 947-958. doi:10.1257/aer.89.4.947
Wen, J.-F., & Gordon, D. V. (2014). An Empirical Model of Tax Convexity and Self-Employment. The Review of Economics and Statistics, 96(3), 471-482. doi:10.1162/REST_a_00388
Widmann, R. (2023). Immigrant inventors and local income taxes: Evidence from Swiss municipalities. Journal of Public Economics, 219, 104822. doi:https://doi.org/10.1016/j.jpubeco.2023.104822
Wilson, J. D. (1986). A theory of interregional tax competition. Journal of Urban Economics, 19(3), 296-315. doi:https://doi.org/10.1016/0094-1190(86)90045-8
Zidar, O. (2019). Tax Cuts for Whom? Heterogeneous Effects of Income Tax Changes on Growth and Employment. Journal of Political Economy, 127(3), 1437-1472. doi:10.1086/701424
Zodrow, G. R., & Mieszkowski, P. (1986). Pigou, Tiebout, property taxation, and the underprovision of local public goods. Journal of Urban Economics, 19(3), 356-370. doi:https://doi.org/10.1016/0094-1190(86)90048-3

Rubolino and Giommoni (2023) conducted a recent study of Italian inter-municipal migration possibly motivated by municipal income tax differences.
The new VAT certificates measures the annual new flow of VAT certificates granted by the tax authorities to entities wanting to start some business activity. It approximates the net flow of annual business entities entering the respective professional markets and it is computed as the Change of the annual stock of existing certificates = Existing Certificates at t-1 + New Solicited Certificates at t – (Expired Certificates at t – Cancelled Certificates at t).
The Internal Fixed Gross Investments (non-financial sector), which consist of the acquisitions (net of sales) of fixed capital carried out by resident producers to which the increases in value of non-produced tangible goods is added.
Hoover, in many models of the specification of the relationship between employment and income tax, no distinction is typically made between the marginal average tax rate, and this make the policy evaluation more difficult. For example, Blau (1987) uses measures of the marginal tax rate in his study of the determinants of U.S. self-employment, and Evans and Leighton (1989) use the average tax rate. The impression is that for these authors distinction is largely irrelevant.
This is so because consumption should not respond to predictable fluctuations in income (see for example, Souleles, 1999, p. 947, who finds significant evidence of excess sensitivity in the response of households’ consumption to their income tax refunds). Nguyen, Onnis, and Rossi (2021, p. 456 and Fig. 4) also find a significant sensitivity of consumption to general level of taxation.

No competing interests reported.

Appendices.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Income taxation and regional growth: DID evidence from an income tax treatment in Italy

Status:

Version 1

Abstract

Figures

1. Introduction

2. The Italian income tax treatment of 2007

3 The DID model identification

3.1 Data description and parallel trends

4. Results

4.1 Total Employment, Self-Employment, and New VAT Certificates

4.2 Income tax effects on per-capita income

4.3 Family consumption

5. More on Validity and Robustness of the DID estimations

6. Conclusions

Declarations

Author Contribution

References

Footnotes

Additional Declarations

Supplementary Files

Status:

Version 1