Patient migration as a response to the regulation of subnational healthcare budgets

ABSTRACT Since 2007, Italian regions with healthcare budget deficits above a specific total funding threshold are obliged to commit to a financial recovery plan (FRP). We employ a quasi-experimental strategy to assess the consequences of FRPs on access to healthcare by identifying their impact on patient migration. We find a 24–31% increase in patient migration as a result of the strictest implementation of FRPs. Our empirical analysis shows that migration is affected by the regional availability of both capital and labour inputs.


INTRODUCTION
The need to contain healthcare expenditure in order to maintain realistic national and subnational fiscal positions has put fiscal rules at the centre of policy debate in many countries.Where the provision of health services is decentralized, fiscal rules are particularly important to prevent lower level government from reckless borrowing in the expectation of a bailout.Despite several studies examining the effects of fiscal rules, discussion of their potential sideeffects has tended to be limited (e.g., Kotia & Duarte Lledó, 2016;Organisation for Economic Co-operation and Development (OECD), 2016;Schakel et al., 2018).One exception concerns the recent debate about the potential adverse consequences of disciplining health expenditure (e.g., Arcà et al., 2020;Atella et al., 2019;Bordignon et al., 2020;Depalo, 2019;Schakel et al., 2017), which focuses on whether fiscal discipline is acceptable in terms of the resulting reduced healthcare opportunities.
We contribute to this debate by studying whether and how the enforcement of fiscal rules aimed at keeping subnational healthcare budgets in check produces an imbalance in the healthcare opportunities available to citizens in different areas and promotes patient migration from fiscally constrained regions.Although we study the Italian context, our analysis and findings can be generalized more widely.
Patient migrationwhich can be considered an example of Tiebout's (1956) voting-with-one's-feet modelis a reliable indicator of perceived poor-quality healthcare services.Therefore, the impact of fiscal discipline on migration brings out indirect but reliable evidence of its effects on the quality of healthcare provided locally.
Italy is an interesting context to investigate the consequences of fiscal rules on citizens' access to care.In 2004, the Italian Parliament passed new legislation, introducing financial recovery plans (FRPs) to control healthcare budget deficits.The law states that regions with fiscal deficits over a specific funding threshold must reach agreement with central government (CG) on how they propose to reduce spending to within budget.Despite wide consensus on the efficacy of FRPs for enforcing fiscal discipline (Arcà et al., 2020;Atella et al., 2019;Bordignon et al., 2020;Depalo, 2019), the consequences for health outcomes and access to care are less clear.
The findings of Bordignon et al. (2020) suggest that FRPs have neither rationed use of healthcare services nor implied significant deterioration in citizens' health.However, Depalo (2019) suggests that regions subject to FRPs managed to contain costs, but that health indicators, such as hospitalization and mortality rates, provide evidence of negative consequences.In all regions subject to FRPs, hospitalization rates decreased, and in some of those regions mortality rates increased.Bordignon et al. (2020) interpret the fall in hospitalization rates as due to a reduction in inappropriate treatments and find no effect of FRPs on mortality rates.They use data on the share of people declaring themselves very satisfied with the quality of care during their most recent hospital stay to assess the impact of fiscal discipline on the quality of regional health services.However, these data reflect the assessment only of those who received hospital healthcare.
In this paper we re-examine the side-effects of fiscal disciplining by focusing on the impact of FRPs on patients' interregional mobility.Since this refers to flows of people moving from their region of residence to satisfy their healthcare needs, it provides evidence of patients' revealed preferences as opposed to their stated preferences gleaned from surveys.
Our analysis is based on a panel of 20 Italian regions observed over the period 2000-19.Our difference-indifferences (DiD) methodology produces findings that are robust to several econometric specifications.They point to increased patient mobility of 24−31% under the strictest form of fiscal discipline, which involves assignment of an ad-acta commissioner.In the case of a standard FRP that does not involve a commissioner, patient mobility increased by around 19%. 1 Overall, our evidence suggests that improvements to budgetary rigour increase the quality gap in the regional health services and increase disparities in access to healthcare (Abatemarco et al., 2020).
Our findings are reinforced by the results of our investigation of the channels through which FRPs can promote migration.Since fiscal discipline generally affects the level of inputs in the healthcare production function, it can be assumed that patient migration is affected by the availability of both capital and labour.Indeed, we find that the more severe the cuts to hospital capacity and/or healthcare employees following enforcement of the fiscal discipline, the higher the outflow of patients.We show that a reduction of four beds (respectively, nine employees) per 10,000 inhabitants increases outflows of patients by about 5% (respectively, 9%).We are not suggesting that fiscal discipline should be abandoned; rather, we want to highlight that awareness of its potentially undesirable consequences is important and that ways should be devised to contain subnational healthcare spending without jeopardizing people's health and access to care.
The paper is organized as follows.Section 2 describes the Italian National Health Service (NHS) and provides a brief review of the literature on patient migration.Section 3 presents some preliminary evidence.Section 4 examines the effects of FRPs on patient migration and the channel through which austerity affects migration.Section 5 examines the consequences of fiscal discipline using a methodology that allows for variations in treatment timings (Callaway & Sant'Anna, 2021).Section 6 concludes.

INSTITUTIONAL BACKGROUND
The Italian NHS was established by Law 833/1978 to guarantee appropriate care for all Italian citizens, supporting the constitutional ideal of universal access to healthcare.
The management of the NHS in Italy initially was characterized by ambiguity and confusion in terms of the sharing of power and responsibility among different government tiers (Pavolini & Vicarelli, 2012).Since the 1990s and in line with trends in many industrialized countries, governance of the NHS has been progressively decentralized.A reform approved in 1992-93 meant that regions could choose their organizational model.This resulted in regional differences in the size of local healthcare authorities, their level of integration with hospital facilities and the involvement of private providers (e.g., Mapelli, 2012;Toth, 2014).Regionalization of the NHS was advanced further by legislative Decree No. 56 which came into force in 2000 and made financing of the regional healthcare no longer dependent only on CG transfers, but also on revenues from regional sources.The reform of the Italian Constitution approved in 2001 (Constitutional Law 3/2001) was aimed at achieving a rebalancing of power and competencies between CG and local government, with the CG responsible for aligning regional health policies to CG goals (aimed mainly at enabling access to a basic level of healthcare for all citizens).This constitutional reform enabled the transition to a regionally organized NHS, which included financial responsibility.However, for some years, the CG continued to provide funding due to the absence of any incentives for cost containment (Aimone Gigio et al., 2018).As a result, many regions continued to record large budgetary shortfalls.
Then, in 2004, the Italian Parliament passed Law 311/ 2004, introducing FRPs for regions with large deficits.The law established stricter access to national healthcare funds and required indebted regions to identify the causes of the deficit and to establish plans to put their healthcare systems on a sound financial path.
The requirement for indebted regions to implement FRPs was included, also, in the State-Regions Agreement (Intesa Stato-Regioni), signed in March 2005, and was at the centre of negotiations between the national government and the regions (Health Pact/Patto per la Salute) held in September 2006.The rules governing these plans are set out in Law 296/2006.FRP provisions remain in force for three years and apply any region with a healthcare budget deficit above a certain threshold.This threshold was set, initially, at 7% of the available resources and was later reduced to 5%.These provisions affect both revenues (i.e., automatic regional tax increases) and costs (i.e., recruitment limits).If, after three years, the FRP has not achieved objectives, the provisions are renewed for an additional three years.
In the case of non-compliance with the FRP conditions, the CG can appoint an ad-acta commissioner responsible for implementing a stricter plan.In the event of appointment of a commissioner, the compliance conditions become much more severe (Presidenza del Consiglio dei Ministri, 2009).They entail an automatic increasebeyond the maximum level commonly allowed by state lawof both the regional tax on productive activities (Irap, +0.15 points) and the regional surcharge on personal income tax (Irpef, +0.30 points).These tax increases are combined with cuts in the transfers from the CG and the removal from office of general managers responsible for key areas of the regional health system.In most cases, the initial ad-acta commissioner is the president of the region; if that individual is deemed not to be adequate for the task, then an external (non-elected) person will be appointed.The development of the accompanying legislation has been, however, rather chaotic.Indeed, starting with Legislative Decree 149/2011, the rules establishing whether the president of the region can serve as a commissioner have changed several times.The 2015 budget law disallowed regional presidents from appointment as ad-acta commissioner.This was rescinded by the 2017 budget law, then reinstated by Law 136/2018 and again rescinded by a Constitutional Court decision (247/2019).This resulted in abrupt changes to the person responsible for implementing regions' FRPs.
Table A1 in the supplemental data online provides information on the presence of an ad-acta commissioner in regions required to comply with FRPs and indicates whether this individual was the president of the region.

Patient mobility
If patients have a free choice of healthcare provider, they will tend to migrate towards centres of excellence, which might be distant from their place of residence (Balia et al., 2020;Brenna & Spandonaro, 2015).This applies to both Italy and other countries.For example, Moscelli et al. (2016) show that following the relaxation of restrictions on hospital choices in 2006, the proportion of UK patients who chose to bypass their nearest provider almost doubled from 25% to almost 50%.
If certain demand factors (e.g., wealth, age; Fattore et al., 2014) are excluded, Italians are generally free to choose their healthcare provider.Fabbri and Robone (2010) found that patient flows tend to be from areas where patients are enrolled in health plans provided by less wealthy local health authorities to areas with better endowed health authorities.This suggests that the presence of more up-to-date technology and better services have a strong effect on migration for healthcare.Similarly, Balia et al. (2014) show that the strongest pull factors are hospital capacity, advanced technology and access to regional health services.Instead Levaggi and Zanola (2004) and Balia et al. (2018) suggest that patient mobility is conditioned by hospital capacity, measured as numbers of hospital beds.
However, the number of hospital beds is a somewhat rough indicator of healthcare service provision.In the Italian NHS case, a degree of caution is advised in particular when assessing the productive capacity of Italy's northern regions, where a progressive 'de-hospitalization' strategy has been implemented and many services are being provided in outpatient clinics (Toth, 2014).In the south of Italy, these re-organization processes have been much slower and less effective, so that public health expenditure is still unbalanced on hospital care.Therefore, the number of hospital beds is likely a more reliable indicator of the overall productive capacity of the health systems in these regions.

A first look at the data
The literature proposes several measures for patient mobility.In this paper we consider a variable we call escape rate.Specifically, for any region i and year t, we compute patient mobility as the ratio of the number of individuals from region i hospitalized in region j (with j ≠ i) to the total number of individuals resident in region i hospitalized at time t.The escape rate indicates the share of the region's unmet healthcare needs.
Figure A1 in the supplemental data online shows how the escape rate has changed over the period 2000-19 in the five Italian geographical macro-areas.Tables A2 and A3 online provide the variable definitions and descriptive statistics.It can be seen that the escape rate since 2007 has increased in the southern regions.
Figure 1 points to the possible impact of fiscal discipline on escape rates.Figure 1(a) depicts the escape rate evolution, based on partitioning Italian regions into two groups: treated regions, subject to an FRP (either standard or with an ad-acta commissioner) at some point in time after 2007 (black line); and 'control' regions never subject to an FRP (grey line).
The evolution of the escape rate in the two groups is quite similar up to 2006 (in section 4.1 we test parametrically that the standard common trend hypothesis holds).However, after 2006, we observe divergence.In 2009, the curves capturing the evolution of the escape rate for the two groups intersect and then move apart.
In Figure 1(b), the grouping is based on regions with healthcare services overseen by a commissioner appointed by CG at some point in time (black line) and from all other regions (including those working according to a standard FRP or no FRP) (grey line).Figure 1(b) includes the assumed counterfactuals, that is, the expected escape rate trend in the two groups had a FRP not been imposed.
Figure 1 provides preliminary evidence that escape rates exhibit similar increasing pre-treatment trends in both the treated and control regions.After 2007, in the control regions the escape rate shows the opposite trend.
Figure 2 depicts migration (captured by the escape rate) towards treated and control regions, respectively, distinguishing flows by origin (treated or control region).We observe a strong decline in migration from the control to the treated regions, which explains the decline in the escape rate in the control regions after 2007 (Figure 1a).As a potential explanation notice that by depleting the stock of resources available at the regional level, financial austerity may discourage inflows from more fiscally virtuous regions.
Table 1 reports the number of regional health system employees and hospital beds.
Given the need to reduce health expenditure, the number of beds (per 10,000 inhabitants) has decreased, regardless of the presence or not of an FRP, and shows a slightly higher reduction in the treated regionsfrom 42.23 to Patient migration as a response to the regulation of subnational healthcare budgets 2209 REGIONAL STUDIES 32.65 (Welch t-test: t ¼ −10.88, p ¼ 0.00)compared with the control regions: from 40.60 to 34.15 (Welch t-test: t ¼ −9.20, p ¼ 0.00).Treated regions also experienced a strong reduction in the number of employees per 10,000 inhabitants, from 112.89 to 102.41 (Welch t-test t ¼ −3.89, p ¼ 0.00), whereas in the control regions there was no appreciable reduction.
The data, which are available from the authors on request, also show that in both groups regional factors that could potentially affect mobility (e.g., share of the population aged 65 or over, per capita income, level of healthcare expenditure) present similar trends in the period under consideration.
Figure 3 shows that, before 2007, most of the treated regions experienced lower escape rates than the average in the control regions.The exceptions are Calabria, a region traditionally characterized by extremely poor healthcare provision, and Molise, whose small size encourages cross-border shopping for care.
Overall, after 2007, the escape rate increased in the treated regions from 9.48% to 11.49%, but remained stable in the control regions (Table 1).In general, escape rates increase with the presence of a non-standard FRP, that is, with the appointment of a commissioner in charge of overseeing healthcare spending (Figure 3).
Figure A2a in the supplemental data online compares the numbers of regional health system employees in the treated (black lines) and control regions (grey lines).Before 2007, with the exception of Liguria, all treated regions had a smaller number of healthcare workers per inhabitant than the average observed in the control group.The striking reduction in the number of workers following enforcement of fiscal discipline worked to increase the gap in labour input between the groups of regions (Figures A2b and A2c online compare the respective evolution in the numbers of nurses and doctors).Figure A2d online shows that, before 2007, the number of hospital beds per inhabitant was higher in four out of 10 treated regions than in the control group.However, numbers decreased with the imposition of an FRP.
Note that the reduction in the number of beds was a consequence, mainly, of cuts brought about in public hospitals.Before 2007, the number of (public) hospital beds in the treated regions was lower than in the control regions; after 2007 it reduced more than in the control regions (Aimone Gigio et al., 2018).After 2007, in the treated regions, the number of public hospital beds per inhabitant was 25.43 compared with 29.12 in the control regions.In the treated regions, the reduction in bed numbers in public hospitals was partially offset by an increase in the number of beds in private hospitals (Table 1).

Identification strategy and results
As already mentioned, the strictest FRPs, imposed for non-compliance with fiscal rules, involve oversight from a commissioner.In some cases, the regional president was appointed as ad-acta commissioner.In the case of continued fiscal problems, an external (non-elected) individual is appointed.
To obtain more precise results, our empirical exercise focuses, first, on FRP with an ad-acta commissioner.
Table A4 in the supplemental data online reports summary statistics for some key variables for regional healthcare supply, sorting regions by standard/non-standard FRP.Note that we consider the autonomous provinces of Trento and Bolzano as constituting a homogeneous area, that is, the single region Trentino-Alto Adige, so that our sample includes 20 regional health systems.This increases comparability with previous studies (e.g., Arcà et al., 2020;Cavalieri & Ferrante, 2016;Depalo, 2019).However, we replicated the main analysis on the full sample of 21 healthcare systems (see Tables A8-A10 online).All our results are fully confirmed.
The figures suggest that following the imposition of the strict FRP (i.e., with a commissioner) regions experienced a deeper retrenchment and, thus, a larger reduction in the resources available for healthcare.For example, in those regions required to adhere to the standard FRP, the number of employees per 10,000 inhabitants fell from 116 to 109 compared with from 110 to 96 in regions with a commissioner.Similarly, in non-standard FRP regions, the number of public beds per 10,000 inhabitants fell from 32.79 to 23.72 compared with from 34.65 to 27.14 in standard FRP regions.
Before conducting the estimates, we ran non-parametric tests to assess the hypothesis of a parallel trend between the treated and control regions In Table A5 in the supplemental data online, columns 1 and 2 refer to the full sample of Italian regions; columns 3 and 4 refer to the subsample of regions working under an FRP.We check the statistical significance of the interaction term - EC i × Yearwhere EC i is a dummy indicating whether the region had a commissioner, and Year is the time trend before its appointment.Columns 1 and 3 show that, in the case of both samples, the estimated coefficient of the interaction term is small and not statistically significant; the p-value of the joint test indicates that the parallel trend assumption is not rejected.
We test next for possible anticipated effects of the treatment by estimating a model in which the EC i dummy is interacted with all the year dummies.Columns 2 and 4 report the estimated coefficients of the leads and lags, relative to the first year of FRP implementation (a necessary condition for imposition of the strictest fiscal discipline).The estimates rule out possible anticipated effects.The coefficients of the years 2000-06 lack statistical significance, which is consistent with the parallel trend assumption.In 2008, the sign on the coefficients switches from negative to positive and, starting from 2009, the first year when commissioners were appointed, they become statistically significant.
The lack of an anticipation effect suggests that any other changes to the institutional framework that had different effects on regions with and without a commissioner, are not influencing escape rate patterns.Overall, the evidence suggests that the following DiD model does not deliver biased estimations of the causal effect: where EscapeRate it is the escape rate in region i at time t; EC i is a time-invariant binary variable that takes the value 1 if the region experienced appointment of a commissioner at some point in time; Post it is a dummy variable that takes the value 1 in any year t in which a commissioner was present in region i; EC i × Post it is the interaction term: the coefficient of this term is the DiD causal effect of interest; X it is a set of variables controlling for demand and supply factors; and λ i and μ t are, respectively, regional and year fixed effects.
Table 2 shows the effect of the FRP with a commissioner, on escape rates.Our regressions include region and year fixed effects; all the specifications include robust standard errors clustered at the regional level.Columns 1-4 compare the treated and control regions.A region is considered treated in a given year if, in that year, it experienced the presence of a commissioner, otherwise it is considered a control region.When considering the whole sample, the results suggest an increase in escape rates, due to the presence of a commissioner, in the range 24-31% (Table 2, last row).The coefficient of the interaction term is significant and stable and has the expected positive sign.Column 4 shows that it is robust to the introduction of controls for demand and supply: GDP per capita, which is a measure of the patient's ability to bear the ancillary costs of mobility; percentage of the population aged 65 or over, which proxies for healthcare needs; Education, measured as the share of population that did not complete primary education; Obesity rate, which can be interpreted as a measure of the length of the time horizon (e.g., Beraldo et al., 2013) and the importance that the individual places on his or her future health status (Grossman, 1972); and Cesarean rate, which measures inappropriate care (e.g., Francese et al., 2014) and, hence, overall quality of regional healthcare.
As a robustness check, we examine the reliability of our selection strategy by redefining the treatment and control groups.Table 2, column 5, reports the results of a regression that excludes from the control group regions subject to the standard FRP, that is, we compare regions with a commissioner with those never exposed to an FRP.The coefficient of interest is consistent in both size and statistical significance with the previous results.In column 6 we restrict the analysis to the sample of regions subject to an FRP at some point in time and identify the treatment with the presence of a commissioner (the control group is made up by all the regions subjected to standard FRP).In this specification, the effect related with the presence of a commissioner is smaller in size and statistically significant only at 10%.Although the model is estimated on only half the observations, which suggests some caution in interpreting the results, it provides preliminary evidence that the standard FRP has an impact on mobility and that the stricter FRP exacerbates this effect.
We ran an additional regression, distinguishing regions that experienced an external non-elected commissioner (in at least one month of the year analysed), from regions where the commissioner was the regional president.Table A11 in the supplemental data online confirms that imposition of the stricter FRP increases the escape rate, regardless of whether the commissioner is president of the region or an external, non-elected individual.It is interesting that, in line with our expectations, the magnitude of the effect is larger if the commissioner is an external non-elected individual.

Administrative autonomy and sample selection: a first check
Our previous estimates considered the universe of Italian regions.This would seem an appropriate choice, as all the regions are in fact involved in the 'exchanging patients' game, regardless of their degree of legislative and administrative autonomy.However, Bordignon et al. (2020) point out that special statute regions enjoy greater independence and suggest that comparison between special statute and ordinary statute regions might introduce bias.
To ensure that our findings do not depend on ad hoc sample selection, we replicated the analysis in the previous section on the sample of ordinary statute regions only.Table A6 in the supplemental data online shows that all the results hold.The coefficient of interest is highly significant and stable, suggesting an impact on escape rates due to the presence of a commissioner, ranging between 21% and 30%.

Exploring the channel through which fiscal discipline affects migration
In this section we investigate the channel through which a FRP conditions migration, focusing on the stricter (nonstandard) FRP.Since the more severe constraints enforced by a commissioner generally affect the variables that enter as inputs in the healthcare production function, then a natural starting pointconsistent with the evidence Patient migration as a response to the regulation of subnational healthcare budgets 2213 REGIONAL STUDIES discussed in section 3is to examine the effect of an FRP on migration via its impact on both capital and labour inputs.
We measure capital inputs as variations in the number of (public and overall) hospital beds.This measures the hospital's capacity to satisfy local demand for healthcare.Note: In this and the following tables: robust standard errors are shown in parentheses; ***, **, *statistical significance at 1%, 5% and 10%, respectively.First four columns: all the regions.A region is considered treated in a given year if, in that year, it experienced the presence of a commissioner.Column 5 excludes from the sample all the regions subjected to a standard financial recovery plan (FRP).Column 6 excludes from the sample all the regions never subjected to an FRP.

REGIONAL STUDIES
For labour inputs we use number of regional health service employees and numbers of physicians and nurses.
Our empirical strategy relies on identifying a coefficient associated with a triple interaction term.We estimate: where EC i , Post it , X it , λ i and μ t are defined as above, and: ΔY i is the variation in input availability observed after 2007; and EC i × Post it × ΔY i is a triple interaction term whose coefficient is the DiD causal effect of interest.
In estimating (2), the treatment variables are both exposure to non-standard FRP and degree of variation in input availability (Table 3).
The coefficient of the baseline interaction term disappears with the introduction of the input variables, suggesting that exposure to the non-standard FRP affects patient migration only indirectly.Note that the coefficient of the baseline interaction (EC × Post) in ( 2) is an indication of the effect of the appointment of a commissioner in a context where there is no change in the input variables.This means that it is not directly comparable with the coefficient of the same explanatory interaction (EC × Post) in Table 2. Also, the coefficient of the triple interaction term, computed including the intensity of the cuts in (public and overall) hospital beds, has a negative effect on mobility.This suggests that as the number of hospital beds per inhabitant decreases in the treated regions, in the post-2007 period migration increases.Specifically, a reduction of four beds per 10,000 inhabitants increases patient outflows by about 5%.Similarly, a reduction of nine regional health service employees per 10,000 inhabitants increases the outflow of patients by about 9%.If we focus on particular worker categories, such as doctors and nurses, we find no effect on patient outflows.
Table 1 shows that the number of public health system employees has been stable in the no FRP regions over time (129 employees per 10,000 inhabitants), was much lower in regions subject to an FRP before 2007 and has decreased sharply since then (from 112 to 102 per 10,000 inhabitants).
Overall, the reduced availability of inputs following enforced fiscal discipline can be interpreted as a contraction of the regional health system's capacity to satisfy citizens' healthcare needs, which, potentially, promotes migration.
Table 3 shows that when variations in either medical doctors or nurses are considered, the coefficients of the baseline interaction term (EC × Post) become statistically significant.This suggests, also, that patient migration is better explained by a decline in the overall level of provision of healthcare assistance than by reduced availability of a particular productive input.

Identification and results
In this section we investigate whether the standard FRP also affects patient migration.We employ the same methodology as in the previous sections to estimate: where the interaction term of interest, FRP i × Post it , is a binary variable that takes the value 1 for the years when the region i was subject to an FRP (FRP i takes the value 1 if the region was exposed to fiscal discipline; Post it equals 1 for the years when the region was affected by a FRP).Table 4 reports the results.Columns 1-5 compare the treated with the control regions.A region is considered treated in a given year if, in that year, it was subject to an FRP, and otherwise is considered to be in the control group.The results in column 6 are based on a different rationale, according to which any region subject to an FRP is considered as 'persistently' treated, that is, since the year in which the plan was imposed.We conducted this test since it shares the same logic of those suitable to assess causal effects in the case that the units are not treated at the same moment (see section 5.2).The interaction term is statistically significant and has the expected sign, even after including fixed effects (columns 1-3).Column 4 shows that the inclusion of control variables turns the coefficient of interest statistically insignificant.This raises the question of whether the standard FRP triggers patient migration; we test for this by conducting the additional estimations reported in columns 5 and 6.
The sample in column 5 includes only those regions where a commissioner was appointed at some point over the time of analysis; the treated group in this exercise includes region i if that region is subject to standard FRP at time t, then FRP i × Post it equals 1.The same region belongs to the control group at time t -1 when it is not yet subject to standard FRP.The causal effect is strongly significant and points to increased patient mobility of 28%.This suggests that adoption of an FRP affects patient migration only in those regions that, at some point in time, were under control of a commissioner.
To check this, we investigate whether the effects of fiscal discipline are confined only to those years when the FRP is in place.Figures A2a-A2d in the supplemental data online show that the effects on input availability persist over time, until the FRP is removed.
It seems that regions subjected to FRPs do not easily recover from the trend induced by austerity on the stock of available healthcare resources.This is confirmed by the estimations in Table 4, column 6, which considers all regions subject to an FRP as 'persistently' treated, that is, starting from the year in which the plan was implemented (in (3) Post it is valued at 1 for all the years subsequent to first imposition of the FRP in the region i).The coefficient of interest is statistically significant and robust to the introduction of controls and suggests an increase in patient migration of about 19% as the result of fiscal austerity. 2 This approach is consistent with the long-term evolution of input availability in regions that were subject to fiscal regulation for only short periods (Liguria: three years; Sardinia: two years).It is also consistent with the approach in section 5.2.

Exploring the average treatment effect (ATE) over time
To corroborate our results, in this section we exploit the methodology proposed by Callaway and Sant'Anna (2021) to test for causal effects in the case of multiple time periods and variations in treatment timings.This methodology assesses group-time ATEs, a natural generalization of the average treatment effect for the treated subpopulation (ATT).Estimation of the group-time parameters requires construction of several treatment groups.Each group is formed by aggregating regions that are treated for the first time in the same year.It then compares each treatment group with a corresponding control group that includes either 'never-treated' or 'not-yet-treated' regions.
In addition to the well-known identification assumptions for the canonical DiD set-up, Callaway and Sant'Anna (2021) posit that having once been treated, the unit is always considered treated.The rationale for this assumption is that the unit does not 'forget' the treatment experience, which is in line with our interpretation of persistence of FRP effects.
One of the main advantages of this methodology is that it allows for different aggregation schemes, each of which concisely summarizes the group-time ATTs with respect to a particular dimension of interest.In what follows we focus on an aggregation scheme that allows us to identify whether the ATT varies with the length of exposure (persistence) to the treatment.This scheme is based on the clustering of group-time ATTs into ATEs at different lengths of exposure to the treatment.For a specific length of exposure, the aggregated parameter is the ATE across all groups participating in the treatment, over the same time periods.
Figure 4 depicts the results of the event study assessing the effects of an FRP.The horizontal axis is the time (years) since first treatment; if positive, it measures the length of exposure to the treatment.The pre-treatment estimates provide a pre-test for the parallel trend assumption; the post-treatment estimates show the pattern of the treatment effect of interest, depending on the length of exposure.
Overall, the plots support the previous results; the coefficients estimating the impact of fiscal discipline on patient migration increase with length of exposure to the treatment.Note: Columns 1-4: a region is considered treated in a given year if it was effectively subjected to the FRP; column 5: sample restricted to regions where a commissioner was appointed at some point over the time of analysis; and column 6: any region subjected to an FRP is considered as 'persistently' treated.

REGIONAL STUDIES
The results are robust to the introduction of the previous used controls.

DISCUSSION AND CONCLUSIONS
In general, looking at society's fundamental goals, it seems that the introduction of fiscal discipline, rather than working as a Pareto improvement, re-proposes the classic equity-efficiency dilemma.Our findings support the conclusions of Depalo (2019) and Arcà et al. (2020), who emphasize that recovery plans achieve their fiscal objectives, but at the very high price of an increase in preventable deaths and increased inequity in access to healthcare both within and between regions.
The substantial level of patient mobility we document suggests that spending cuts are achieved mainly by reducing the level of inputs required to provide adequate healthcare and that inefficiencies in the management of resources are reduced only marginally.In this sense, the effect of FRPs can be considered part of an ongoing and broader (decentralization-related) process of differentiation among local healthcare systems, which risks increasing the differences between northern and southern regions of Italy in terms of citizens' access to adequate care Patient migration as a response to the regulation of subnational healthcare budgets 2217 REGIONAL STUDIES (e.g., Aimone Gigio et al., 2018;Pavolini & Vicarelli, 2012;Toth, 2014).This is not surprising since, potentially, decentralization increases territorial inequalities.However, it is problematic in countries such as Italy, which traditionally are characterized by stark socio-economic divides.This is not to say that decentralization is in itself bad.As emphasized by Cavalieri and Ferrante (2016), the response to decentralization might be heterogeneous due to differences in the characteristics of local governments and populations.These differences must be taken into account in the governance of the process in order to avoid that the gains in terms of healthcare performance in economically advantaged areas are achieved at the expense of healthcare performance in the economically disadvantaged ones.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.

NOTES
1. Bordignon et al. (2020) conducted instrumental variables estimates to test whether FRPs stimulate patient migration, but found no statistical significance.The differences with our findings may be because we do not limit the analysis to ordinary statute regions, and we also consider a longer time span (which works to almost double the size of our sample).2. Table A7 in the supplemental data online reports the results for the subsample of ordinary statute regions.

Figure 1 .
Figure 1.Evolution of the escape rate, treated and control regions: (a) standard financial recovery plan (FRP); and (b) FRP with commissioner.Note: (a) The black (respectively, grey) line clusters regions which have (respectively, never) undergone the FRP; and (b) the black (respectively, grey) line clusters regions where a commissioner was present at some point in time (respectively, never present).The vertical dotted line depicts the period of FRP discipline rollout.

Figure 2 .
Figure 2. Evolution of the escape rate towards: treated regions (a); and control regions (b).Note: The black line captures the escape rate from regions subjected to financial recovery plan (FRP); and the grey line refers to the escape rate from regions never subjected to an FRP.The vertical dashed line depicts the period of FRP discipline rollout.

Figure 3 .
Figure 3.Comparison of the escape rate in treated regions with the average escape rate in control regions.Note: The vertical dotted lines encompass the years in which the region was subjected to a financial recovery plan (FRP).A unique vertical dotted line indicates that the region was still subjected to an FRP in 2019.The vertical dashed line indicates the year in which the government appointed a commissioner.

Figure 4 .
Figure 4. Dynamic treatment effect of financial recovery plan (FRP) on the escape rate: FRP without commissioner (a); and FRP with commissioner (b).

Table 1 .
Summary statistics by financial recovery plan (FRP) exposure.
Note: First two columns: only regions subjected to the FRP; and last two columns: only regions never subjected to an FRP.2212Sergio Beraldo et al.

Table 2 .
Presence of a commissioner and escape rates.

Table 3 .
Inputs reduction and escape rates.

Table 4 .
Standard financial recovery plans (FRPs) and escape rates.