A Costly Commitment: Populism, Economic Performance, and the Quality of Bureaucracy ∗

We study the consequences of populism for economic performance and the quality of bureaucracy. When voters lose trust in representative democracy, populists strategically supply unconditional policy commitments that are easier to monitor for voters. When in power, populists try to implement their policy commitments regardless of financial constraints and expert assessment of the feasibility of their policies, worsening government economic performance and dismantling resistance from expert bureaucrats. With novel data on more than 8,000 Italian municipalities covering more than 20 years, we estimate the effect of electing a populist mayor with a close-election regression discontinuity design. We find that the election of a populist mayor leads to smaller repayments of debts, a larger share of procurement contracts with cost overruns, higher turnover among top bureaucrats – driven by forced rather than voluntary departures – and a sharp decrease in the percentage of graduate bureaucrats

by neglecting constraints and inflationary consequences, populism leads to economic failure, social unrest, and violence. In their words, because of specific policy preferences, populism is "self destructive." However, recent scholarship questioned the association between populism and specific types of policies, characterizing populism as a thin ideology with a clear view of society but a vague view of the ideal economic or political system (Mudde and Kaltwasser 2017). When departing from a fixed association with policies, the mechanism behind the negative consequences of populism escapes and the idea that populists are doomed to bad performance seems undistinguishable from a mere prejudice. In fact, it has been shown that in some cases populists can be successful at governing (Albertazzi and McDonnell 2015;Biard 2019). The second challenge is empirical. Estimating the effect of populism is a daunting task.
Random occurrences of populist governments are nearly inexistent, and multiple economic and political factors that are correlated with the election of a populist leader also correlate with policy outcomes, limiting the space for causal identification. With this paper we address both challenges.
Theoretically, we build on recent work that looks at the determinants of populists' incentives and policy platforms. When economic, technological, or cultural shocks erode citizens' trust in political elites and the traditional tools of representative democracy, voters demand simple protection policies (see e.g. Guiso et al. 2017). Populist politicians intercept this demand shift towards tighter control over policy-making, and run for office i) committing to a set of policies easy to monitor (e.g., America First or citizenship income), and ii) fuelling voters' distrust in the non-populist opponents, accused of being supported by corrupted elites (Moffitt 2016;Morelli, Nicolò, and Roberti 2021). Trust erosion therefore triggers populism, for it creates incentives for politicians to commit to policies that are easier to monitor, while at the same time accusing traditional parties' candidates of being captured by the elites. Because of such commitments, when populists come to power they implement their agenda with no regard for constraints and expert assessment of the feasibility and consequences of their policies. When facing the resistance of expert bureaucrats', populists will replace them with non-experts in order to ensure the smooth implementation of their policy agenda (Eichengreen 2018;Sasso and Morelli 2021).
Building on these micro foundations of populists' incentives and behaviour, we derive testable expectations about the consequences of populism for policy-making. No matter what the circumstances require, populists carry on with their policy commitments, with detrimental consequences for government economic performance. In order to minimize sabotage from a competent bureaucracy, populists replace experts with non-expert bureaucrats, which results in increased turnover and lower quality of bureaucracy.
Empirically, we test these expectations with novel data on economic performance, public procurement, and bureaucratic composition of municipal governments in Italy over a 20-year period, from 1998 to 2020. We use a close-election regression discontinuity design and compare municipalities in which a populist mayor barely won the elections to municipalities where a populist barely lost in order to isolate the effect of electing a populist candidate. Italian local government is a good case to study the effects of electing a populist government. First, Italy has more than 8,000 municipalities and several populist parties populate the Italian party system at both national and local level, yielding large sample size and large variation in "treatment status." Second, populist parties exert a tight discipline over their members, hence there is likely to be a precise match between the populist nature of the party and the populist attribute of affiliated mayors. Third, the concentration of executive power in mayors and mayors' discretion over the appointments and removals and top bureaucrats create the necessary space to detect sizeable effects on economic performance and the quality of bureaucracy.
The results support the expectations. We find that the (close) election of a populist mayor leads to a 5.3 percentage point increase in the share of public contracts with cost overruns, smaller debt repayments, and -although the estimates are less robust -larger accumulation of debts. Turnover among top bureaucrats increases by 50% compared to the average turnover in the data, and the percentage of bureaucrats with a university degree drops by -13.1 percentage points. By analysing the stated reasons of bureaucrats' departures, we are also able to show that forced rather than voluntary departures drive the increased turnover, suggesting that bureaucrats are forced to leave and do not choose to leave when populists win.
Our findings complement early attempts at investigating the relationship between populism and the economy. Funke, Schularick, and Trebesch (2020) assemble a dataset on populist leaders and find large long-term economic costs of populism, with GDP per capita being more than 10% below compared to the most plausible non-populist counterfactual. Similarly, Magud and Spilimbergo (2021) find negative macroeconomic consequences of populist regimes in Latin American countries. On the bureaucracy side, Peters and Pierre (2020) suggest that populism is likely to translate into lower expertise in government. Bauer and Becker (2020) discuss the public administration goals and strategies of populist governments, showing how purges of personnel and top bureaucrats occurred in many historical cases of populist governments.
Similar conclusions are reached in the theory proposed by Sasso and Morelli (2021). Finally, our work builds on recent scholarship showing that populists can have a significant influence on policy, both when in power (Albertazzi and McDonnell 2015;Paxton 2019), and when outside government (Biard, Bernhard, and Beta 2019). We show that the influence of populist commitments has negative consequences for the performance of municipalities and the retention of competent bureaucrats.

Populists' Political Agency and the Strategic Supply of Commitments
Social scientists have embraced different definitions of populism, focusing on ideology, political strategy, or policies (Mudde and Rovira Kaltwasser 2018). First, the "ideational" approach defines populism as a thin-centred ideology which portrays society as divided into the "pure people" and the "corrupt elite," arguing that politics should be the expression of the will of the people (Mudde 2004). Second, the political-strategic approach considers populism as a political strategy employed by politicians in the attempt at "winning and exercising power" (Weyland 2001, 12) while creating unmediated ties between the leader and voters (Weyland 2017). Consistently with these two definitions, the people-vs-elite rhetoric is a straightforward manifestation of populism, which is also the most common measure used in empirical work (e.g., Pauwels 2011;Hawkins et al. 2019;Gennaro, Lecce, and Morelli 2021;Di Cocco and Monechi 2021). A third approach, originating from economics, defines populism as the implementation of redistributive policies that claim to promote the interest of common citizens without regard to the consequences of such policies (Dornbusch and Edwards 1991;Acemoglu, Egorov, and Sonin 2013).
Our theoretical argument endorses a strategic definition of populism, which we believe is the most suitable for studying the consequences of populism. In fact, while the ideational ap-proach does not point to any specific direction when studying the consequences of populism, the economic approach is too narrow in associating populism with a set of redistributive policies . Conversely, the strategic approach allows us to link causes, strategies, and consequences of populism in a single account. As we propose in this section, trust erosion creates demand for simple commitments that are easier to monitor for voters (causes) and triggers a strategic response from politicians: unconditional commitments to policies and anti-elite rhetoric (strategies). Because unconditional commitments are by definition unfit for changing circumstances and hostile to expert assessment of alternatives, populism has detrimental effects on policy-making (consequences).
A common thread in the literature on the causes of populism is the erosion of voters' trust in the traditional tools of representative democracy (Inglehart and Norris 2019;Berman 2021).
When trust is high, politicians are entrusted to adjust policy-making to changing circumstances, in line with a trustee model of representation (Fox and Shotts 2009). Political selection prioritizes competence and when running for office, politicians will stress their level of expertise in delivering their proposed policies. Conversely, when trust is eroded, voters consider politicians -even the competent ones -undeserving of the autonomy characterising the trustee model of representation, perhaps because they are considered selfish rent-seekers captured by interest groups or elites. Voters want to take back control over policy, demanding simple policy commitments that are easy to monitor. Representation shifts to a committed delegate model. Populist politicians intercept this demand and run for office with simple commitments and radical solutions (e.g., zero-immigration types of policies, protectionism, citizenship income). Once a party or politician shifts to the committed delegate model, the best complementary strategy to maximize electoral support is to fuel distrust for the corrupted candidates supported by the corrupted elites, thus sustaining the crisis of trust (see e.g., Mudde 2004;Moffitt 2016). This is why the committed delegate model triggers populism, for the committed delegate rationally chooses simple policy commitments and all the complementary strategies commonly associated with populist behaviour, namely anti-elite rhetoric, anti-media, and anti-experts denigration (Morelli, Nicolò, and Roberti 2021).
Once in office, populists try to stick to their policy commitments. Paxton (2019), for instance, shows that even at local government-level populists are highly incisive with respect to the policy issues they prioritize, especially when they enjoy a large degree of discretion, as evidenced by the author's comparative analysis of municipal governments in Italy and Austria.
Also cross-country evidence points to the significant influence that populist parties have on policies (Biard, Bernhard, and Beta 2019) and how they manage to deliver on their electoral promises, ultimately surviving the electoral cost of governing (Albertazzi and McDonnell 2015).
However, unconditional commitments might lead to inefficiencies and suboptimal outcomes if populist policies are implemented without a careful assessment of their correspondence to the changing economic landscape. As a result of this form of imperative policy-making, we expect populism to worsen government economic performance.
Furthermore, because commitment-type policies are by definition incompatible with adjustments or alternatives, populist politicians are not willing to have their proposed policies undergo expert assessments of their feasibility, for experts may raise concerns about the desirability of populists' policies given the economic and political context (Peters and Pierre 2020).
This might be one of the reasons why populist governments were slower at passing health measures and mobility restrictions during the outbreak of the COVID-19 pandemic in February and March 2020 (Kavakli 2020), and in general performed more poorly (Bayerlein et al. 2021). For populists, expert bureaucrats' recommendations implying policy solutions different from their commitment-type agenda are unwelcome (Mudde and Kaltwasser 2017;Eichengreen 2018). Populists therefore have a strict preference for non-expert bureaucrats who obsequiously implement the populist agenda. As a result, populists will control the bureaucracy by staffing the administration with non-experts who are more likely to implement populist policies. 5 Based on this theoretical framework, we can derive testable expectations about the consequences of populists' commitments. As a result of this form of imperative policy-making, we expect populism to worsen government economic performance. Furthermore, when facing resistance from expert bureaucrats, populists will replace them with non-experts, increasing turnover and decreasing the quality of bureaucracy. We formalize these expectations with three testable hypotheses. Populist governments lead to (i) lower economic performance, (ii) higher bureaucratic turnover, and (iii) lower quality of bureaucracy.
The willingness to implement their commitment-type policies head first, rather than the recurrence of some specific "bad policies," is the real cause of populists' bad economic performance.

Municipal Government and Populism in Italy
Italian municipalities represent a good case to study the consequences of populism for government performance and bureaucratic quality. First, the presence of several populist parties in Italy makes it less arbitrary for researchers to measure the populist attribute of candidates.
Second, executive power is highly concentrated and populist mayors and executive committees have significant leeway to implement their commitment-type policies. Third, municipal bureaucrats play a central role in the administration of public policies, and mayors have large discretion over the appointment and removal of top municipal bureaucrats.
Consistently with various definitions of populism (Mudde and Rovira Kaltwasser 2018), empirical work generally identifies populist parties based on the presence of people-vs-elite rhetorical elements in electoral manifestos and other party sources (e.g., Pauwels 2011; Di Cocco and Monechi 2021). Measurement becomes more challenging when moving the focus from parties to individuals. Two strategies are available, based on individual behaviour or party identification. The suitability of these strategies rests on data availability for the former (e.g., individual speeches or campaign messages), and the presence of clearly populist parties for the latter. Gennaro, Lecce, and Morelli (2021) and Hawkins et al. (2019), for instance, measure populism for US congressional candidates and chief executives based on the analysis of speeches.
However, when populism is a characteristic of political parties and there is no textual data to produce individual-level estimates, researchers can rely on candidates' party identification and code populist politicians based on whether they identify with a populist party or not. While using party-level characteristics to infer individual-level characteristics clearly implies a loss of construct validity, we believe it is a suitable strategy for the Italian context, where several parties are clearly considered populist and deviation from party lines is punished by populist parties (Fasone 2020). 6 We obtain a list of populist parties in Italy from the PopuList Database, a collective enterprise that brings together researchers in the attempt to provide a comprehensive list of populist parties in Europe (Rooduijn et al. 2019). There are six populist parties in Italy, which ranks fourth in Europe for number of populist parties: Lega (Nord), Movimento 5 Stelle (Five-Star Movement, 5SM), Fratelli d'Italia, Forza Italia -Il Popolo della Libertà (only until 2018), and two minor parties, Liga Veneta and Lega d'Azione Meridionale. In Section A of the SI we provide qualitative information about these parties.
Italian municipal government features a directly-elected mayor, a directly-elected local council, and an executive committee appointed by the mayor, with elections held every five years.
Mayoral candidates are linked to one or more party-lists of candidates to the local council. The lists mirror the political parties at the national level, although it is common for candidates to be linked to civic lists (i.e., liste civiche) that do not match any of the parties at the national level. Voters can cast a ballot for the mayoral candidate directly, for one of the lists linked to the candidate, or both. Votes cast for a list automatically count towards the number of votes of the mayoral candidate linked to the list. The candidate who wins a plurality is elected mayor, except for municipalities with more than 15,000 inhabitants, where the two candidates with the largest vote share run in a second round if none of the candidates obtain 50% of the votes in the first round.
Municipalities are responsible for a wide set of services, from primary schooling to local police, waste management, public roads and infrastructure, social services, and security. As a result, municipal governments enjoy large degrees of financial autonomy and they are responsible for the procurement of goods and services. Municipalities also have large bureaucratic apparatuses, accounting in 2017 for 12% of the 3.5 million employees working in public organisations in Italy. 7 The bureaucratic organisation of Italian municipal governments consists of two types of employees, those with managerial rank (qualifica dirigenziale) and those without managerial rank (qualifica non dirigenziale). For simplicity, we shall consider bureaucrats without managerial rank as rank-and-file employees and bureaucrats with managerial rank public managers. While rank-and-file employees have limited discretion on policy, public managers espulsioni-parlamentari-governo-draghi-11478409/.
7 See the 2017 Census of Public Organisations, National Institute of Statistics http://istat.it/it/censimentipermanenti/istituzioni-pubbliche. are responsible for the implementation of directives adopted by the executive committee, the financial and personnel management of the municipality, public service delivery and monitoring.
Rank-and-file employees are generally hired through public competitions and with permanent contracts agreed at the national level through collective bargaining, whereas the mayor and the executive committee have larger discretion over the hiring and departures of public managers. There are three ways to hire public managers: public competitions published by the municipality, mobility across organizations, and direct appointment with temporary contracts.
Municipal governments have increasingly opted for fixed-term contracts that allow them to exert a tighter control over bureaucrats, with many scholars arguing for the emergence of a "spoil system" in Italian local government (Borgonovi and Ongaro 2011). The average share of temporary contracts among all bureaucrats with managerial rank is 20%, increasing from 16% in 2003 to 25% in 2019.
While mayors have full discretion over temporary contracts, permanent contracts can be terminated only for serious failures. However, mayors have discretion over the allocation of managerial tasks and demotion within the government, and are therefore able to incentivize public managers to leave even when the contract is permanent. 8 Mayors can therefore affect turnover in municipal government through the termination of temporary contracts and (indirectly) through mobility across organizations for permanent contracts. 15% of municipalities have public managers, and the median number of managers in those municipalities is 4.

Data
We assemble a rich dataset on municipal governments in Italy covering more than 20 years, from 1998 to 2020. We collect three sets of data: i) municipal elections, mayoral candidates, and partisan affiliation of politicians, ii) economic performance and public procurement data, and iii) bureaucratic composition.
We combine several sources of data. We obtained data on all municipal elections, mayoral candidates, and their party affiliations from the Historical Electoral Archive of the Ministry of the Interior (1989Interior ( -2020. We used the Database on Local Administrators for other information on mayors (e.g., job, gender, and level of education), local councillors, and members of the executive committee . For our measures of performance we collected novel data on public procurement (2012-2020) from the archives of the National Anticorruption Authority, which we combined with data on economic performance from the National Institute of Statistics (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019). Finally, data on the number of bureaucrats, their rank, education, type of contract, and data on hirings and departures is obtained from the Annual Account of the Italian General Accounting Office (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019). This is at the same time an extremely rich and complex source of data which allows us to capture variation in bureaucratic composition of municipal governments over 20 years and across 8,405 unique municipal governments. To the best of our knowledge, it is the first time it is used in scholarly work. 9

Populist Candidates
We code populist mayoral candidates based on the political party-lists linked to the candidate. We delegate the identification of populist parties to the PopuList database (Rooduijn et al. 2019), commonly used in scholarly work and regularly peer-reviewed by more than eighty academics from different countries. PopuList codes parties as populist if they interpret the people and the elites as two antagonist entities; embrace the idea of popular sovereignty; and uphold an ideal vision of the people and a denigratory vision of elites.

Economic Performance
We present three measures of economic performance. First, we calculate the percentage of procurement contracts with cost overruns. Public procurement data is often used in the literature to study performance (Spenkuch, Teso, and Xu 2021;Decarolis et al. 2021). We follow this literature and compute the yearly percentage of all public contracts awarded by the municipality where payments are greater than the adjudicated costs. Since 2012, every municipality has to publish on its website a large set of information about each contract such as object, cost, duration, beneficiary, and awarding procedure (Article 1(32), Law 190/2012).
This information is also sent to the Anticorruption Authority for verification and quality control, before being published in the authority's open-access archives.
9 Detailed description of the original sources is reported in Section B of the SI.
We web-scraped the archives and assembled a dataset with 1.25 million contracts worth at least 1,000 euros awarded by 5,527 municipalities from 2012 to 2020. 10 For each contract we create a variable equal to 1 if payments exceed the initial adjudicated cost, and 0 otherwise.
We then compute the average of this metric for every municipality in any given year, obtaining a dataset of 25,628 municipality-year observations. 11 The other two measures of performance capture the fiscal quality and sustainability of the administration, proxied by the accumulation and repayment of residual liabilities. Let us consider this example. In year , municipality i's balance sheets report residual liabilities for 2,000 euros, consisting of the difference between committed and actual payments in the previous year − 1. In year , municipality i commits to payments for 13,000 euros and eventually pays 10,000 euros, thus accumulating other 3,000 euros of liabilities. Debt accumulation is computed as the ratio between current and initial liabilities (3, 000/2, 000 = 1.5). Debt repayment is the ratio between disposed and accumulated liabilities, and measures the ability of municipality i to repay more debts than what it accumulates. A good fiscal performance is associated with low levels of debt accumulation and high levels of debt repayment.

Bureaucratic Turnover and Quality
To measure the consequences of populism for bureaucratic quality and turnover, we focus on public managers. We do this for two reasons. First, public managers have large levels of discretion in the administration of policies and populist politicians are more likely to replace bureaucrats in strategic decision-making positions. Second, as discussed above, it is easier for politicians to fire and hire managers compared to rank-and-file employees.
We measure bureaucratic turnover in municipality and year as the sum of managers who leave (departures) and join (hirings) the government divided by the total number of managers in the same year. The precise metric is given by the following formula: 10 Several contracts have typos in the reported information: 6,629 contracts have a termination date that is earlier than the starting date and 348 contracts have payments that are more than 100 times as large as the adjudicated costs. To clean the dataset, we removed these contracts. 11 In Section E.4 in the SI we show our results are robust when computing the percentage of contracts where the difference between payments and costs exceeds 5% and 10% of the awarded costs.  Perfect measures of quality of bureaucrats are hard to produce, for quality is a latent, multifaceted concept. We follow a long tradition of work in political economy that uses education as a proxy for the quality of politicians and interpret bureaucrats' quality as the percentage of managers with a university degree (in the Italian context, see e.g. Galasso and Nannicini 2011;Baltrunaite et al. 2014). By jointly looking at turnover and education we provide a comprehensive picture of the costs of populism for the bureaucracy. In fact, while turnover alone can produce leadership vacuums and undermine institutional continuity in planning and implementation (Lewis 2011), these negative consequences are likely to be enhanced when coupled with a non-expert-replace-expert dynamic. Table 1 reports descriptive statistics for the main variables in the full dataset and in the analysis dataset, namely the subset of municipalities where a populist candidate ran for office.
3,244 out of the 8,569 municipalities in our dataset had a populist candidate in any of the elections they held. In the third and fourth columns, we report statistics for municipalities where the populist candidate won and lost the elections.

Close-election Regression Discontinuity Design
Identifying the effect of populist governments on our outcomes of interest is a challenging task. Municipalities governed by a populist may differ from municipalities governed by a non-populist mayor due to many unobservable characteristics. However, municipalities where populist candidates win the elections by very thin margins can be, in expectation, comparable to municipalities where the populist candidate barely lost. Close-election sharp RDD is a common method for estimating the effect of winning candidates' characteristics on downstream outcomes. This approach has been used in the context of Italian municipalities by Gagliarducci and Paserman (2012) and Casarico, Lattanzio, and Profeta (2021), to estimate the effect of electing a female mayor on government termination and fiscal policy, by Romarri (2020) We intentionally define as the LATE of electing a populist candidate, instead of the LATE of populism alone. In fact, because close elections do not randomly assign candidate characteristics, the RD estimator will recover the effect of the populist attribute of mayors plus all other individual-and municipality-level characteristics that distinguish populist from non-populist and that allow the former to remain in close-election (Marshall 2022). 13 In the following section we describe how balance tests for mayor-level characteristics can help us characterize the compound nature of the estimand, as recommended by recent methodological literature (Marshall 2022).
We estimate with a continuity-based approach that uses nonparametric local polynomial methods for estimation and inference. We fit local WLS models where weights are determined by the triangular kernel function based on the ratio between the distance of unit from the cutoff and the mean-squared-error minimising bandwidth ℎ. The closer the units are to the cutoff, the larger the weight. Units outside the optimal bandwidth receive a weight equal to zero, therefore estimation is performed on a restricted sample of units so that ∈ [−ℎ, +ℎ].
We use the automatic bandwidth selector proposed by Calonico, Cattaneo, and Titiunik (2014), which aims to minimize the mean-squared-error of the local polynomial RD point estimator.
Inference adjusts for the variability introduced in the bias-estimation step and uses a variance estimator that yield robust bias-corrected confidence intervals and p-values (Cattaneo, Idrobo, and Titiunik 2019). Therefore, point estimates will not be centred in the confidence interval.
We estimate the following full treatment-interaction model: We use subscript for election year and for calendar year. Recall that is a dummy for treated units above the cutoff, is the margin of victory (i.e., the running variable), and a robust error term clustered by municipality. We include a set of pre-treatment covariates in the specification as well as year, municipality, and election-year dummies ( , , and ) to boost efficiency (Calonico et al. 2019). The coefficient is the RDD estimator and identifies the average outcome jump at the cutoff after partialling out the effect of the covariates and the fixed effects. Mayors stay in office for five years, while the outcome variables are at the municipality-calendar year level. is therefore the average yearly effect of electing a populist mayor within the government term.

Validity
An important falsification test for the RDD consists of ensuring that, near the cutoff, treated units are similar to control units in terms of observable characteristics. Balance tests are generally employed to provide evidence in support of the continuity of potential outcomes assump-tion, where several pre-treatment covariates are used as placebo outcomes. However, as we are interested in the effect of a candidate characteristic, the RDD estimand should be interpreted as a compound treatment effect. In most cases, this interpretation is sensible, for causal claims about fixed characteristics like populism, gender, and race should be operationalized as a "bundle of sticks" (Hall 2015;Sen and Wasow 2016). However, detecting meaningful discontinuities in candidate-level characteristics that are conceptually different from populism can help characterize the compound nature of treatment (Marshall 2022). Figure D.2 in the SI shows balance tests for 18 pre-treatment covariates. Municipalities above and below the cutoff are very similar with respect to demographic, geographic, and political characteristics of the municipality. Importantly, we find no discontinuity for the value of the margin of victory from the previous election. Populists and non-populists mayors are also similar with respect to several individual characteristics (e.g., education, gender, and age).
However, populist candidates are more likely to be new entrants compared to their non-populist counterparts. We find that the probability of populist mayors to be incumbents is 12 percentage points lower. While the absence of discontinuities in most of the covariates strengthens our confidence that municipalities where a populist mayor won and lost by thin margins are similar in expectation, the discontinuities in the incumbency status of non-populist mayors might suggest that populist candidates benefit electorally from being challengers, and their lack of government experience allows them to remain in close elections. This difference might confound the effect of the populist attribute of mayors. In fact, bureaucratic turnover might be the result of change in political leadership of the municipality and poorer performance might be the product of inexperience. In Section Challenger Status and (Right-Wing) Populism we address this source of confounding.
We perform several additional tests and we report the results in the SI. We document the absence of sorting at the cutoff with density tests aimed at detecting whether there is a proportional number of elections where populist candidates barely won or lost ( Figure D.1).
In Figure D.3 we report estimates from alternative placebo margins of victory ranging from −25% to +25%. When adjusting the estimation for multiple testing we find no discontinuities at 95% level in 83% of the tests with placebo cutoffs -when pooling all the p-values -and in 78% of the tests -when performing the multiple-test adjustment separately for each outcome.
While the number of discontinuities at placebo cutoffs is not negligible, it is important to note that we never find a simultaneous discontinuity for all our five (nor even four) outcomes in any of the 46 placebo cutoffs used. Next, we perform a set of falsification tests with lagged values of the outcomes, using the margin of victory in election to estimate jumps at the cutoff in the outcomes during the years between election and − 1 and find a discontinuity only for one outcome (see Table D.4). We also show that the estimates are overall robust to alternative bandwidth selections and bandwidth selectors ( Figure E.4 and Figure E.5). In Section E.2 we perform power analysis to ensure we have enough statistical power to detect the treatment effects we actually estimate (Stommes, Aronow, and Sävje 2021). Finally, as suggested by Cattaneo, Keele, and Titiunik (2021), we estimate the effects without using covariates (see Table E.6 in the SI) and only debt accumulation displays instability in the point estimate and generally low precision.
The outcome which performs worst in these tests is debt accumulation. The power analysis shows we are not well-powered to detect even small effects for this outcome, the point estimate after removing the covariates from the estimation as well as selecting larger bandwidths is pushed towards zero 14 Debt accumulation is also the outcome with the lowest share of nonsignificant estimates at placebo cutoff (67%). 15 In the section below we present the results for all the five outcomes, although those for debt accumulation should be treated with the necessary caution warranted by the results of these tests.

Results
Merging the database on government performance and bureaucratic composition with election data where one populist candidate was running against at least one non-populist candidate, we obtain a sample of 7,897 municipality-elections pairs, for a total of 31,545 municipality-year observations. 2,164 elections were won by a populist (27% of the total). In most cases, populist candidates were supported by one populist party (i.e., 6,153 elections), 1,575 candidates by two, and only 169 candidates were supported by three populist parties (i.e., the right-wing populist  Table 2 shows the RD results. We report the estimated effect of electing a populist mayor in close electoral races on the five outcomes as well as 95% cluster-robust confidence intervals. 17 Consistently with the RD plots, there are economically and statistically significant effects on government economic performance. Municipalities with a populist mayor do a worse job of repaying debts accumulated in the current year, with debt repayment decreasing by -0.05 points (-5% compared to the mean in the data), and the share of procurement contracts with cost overruns increases by 5.3 percentage points. Despite its weak performance in the validity tests, there is a positive relationship between electing a populist mayor and debt accumulation, with the ratio of new over old liabilities increasing by 0.02 points, equal to 4% the average value in the dataset (i.e., 0.59). While we believe the effects estimated for debt accumulation should be interpreted with caution, the direction of the point estimate is consistent with the other two performance outcomes, which jointly lend support to our argument for which populist commitments are bad for government economic performance.
Electing a populist mayor also leads to an increase in turnover among public managers 16 In Table A.1 in the SI we show the distribution of populist candidates across each pair of supporting populist parties.
17 Results without covariates are reported in Table E.6 in the SI. Only for debt accumulation the 95% CI includes 0 (robust p.value = 0.346).  In Section F.3 in the SI we perform additional tests to detect heterogeneity in treatment effects based on the first or second part of government term, geographical location, and population size of municipalities and we do not find marked differences, except for the effects on performance outcomes in the sample of municipalities with population size below the median value, which generally display larger effects. This is a surprising finding, which clearly deserves further investigation. Even though this difference might suggest that the consequences of populist commitments are particularly relevant for small polities, it is worth noticing that in the context of Italian municipalities "smaller" also means less fragmentation of executive power.

Quality of Bureaucrats
The size of executive committees and local councils is set by the law as a function of population thresholds and the number of bureaucrats is highly correlated with population size (correlation equal to 0.98), hence mayors in smaller municipalities -while enjoying the same level of discretion in the management of the municipality -face the resistance of fewer bureaucrats, fewer members of the legislative body, and coordinate policy-making with smaller executive cabinets. As a result, the election of a populist mayor in municipalities with similar tasks but higher concentration of power might translate into larger costs for the economic performance of government, without the need to alter the organization of the bureaucracy. The relationship 18 We find positive effects on both the number of hirings and departures (see SI Table F.8). Moreover, as a falsification test, we find much smaller effects for rank-and-file employees, with the effect on turnover even being of opposite sign (see Table F  between this finding and the scope conditions of our argument is discussed in more depth in the conclusions.

Forced or Voluntary Departures
Our expectation for which populist mayors get rid of expert managers with detrimental consequences for the quality of the bureaucracy hinges on expert bureaucrats being forced to leave the administration. However, an alternative, observationally equivalent mechanism could be at play, whereby managers voluntarily decide to leave the administration following the election of a populist mayor. For instance, Bolton, Figueiredo, and Lewis (2021) argue that top bureaucrats might decide to leave the organisation when facing a newly-elected principal with diverging policy positions, either because they are marginalized by the new government or because the value they obtain from public office decreases.
To adjudicate between these two mechanisms, we dig into the data on bureaucrats' departures, which allows us to code whether the departure was a mayor's or bureaucrat's decision.
The original data features eight categories of departures, which are displayed in Table 3 and described in more detail in Section F.2 in the SI. After reviewing the legislation with the support of a public manager from an Italian municipality, we are able to classify categories based on whether they represent a decision of the mayor (forced) or of the manager (voluntary).
Some categories are either too ambiguous to be considered forced or voluntary departures (e.g., "Other Reasons") or they are forms of natural termination of the contract (e.g., "Termination of Contract (40 years of contributions)").
We then estimate the effect of electing a populist mayor on four separate outcomes: total  These results corroborate the expectations we derived from characterizing populism as a form of unconditional policy commitment. Populist governments deteriorate the financial sustainability of the government and incur in inefficiencies in the management of procurement contracts. In the attempt to get rid of internal bureaucratic resistance, populists are also more likely than non-populist to replace expert bureaucrats with less educated ones, thus decreasing the overall quality of the bureaucracy. We therefore interpret lower performance and weakened bureaucracy as two sides of the same coin: populists' costly commitments.

Challenger Status and (Righ-Wing) Populism
We present two robustness tests to further strengthen our results. First, we address the possibility of the estimated effects being confounded by the challenger status of populist mayors.
The lower probability of populist mayors to be incumbent might confound the effect of the  Figure 3: RD estimates of the effects of electing a populist mayor excluding incumbent mayors (we keep the first term of all mayors who have been elected more than once) and excluding incumbent mayors and non-incumbent mayors supported by at least one incumbent party. Same estimation, covariates, and period of analysis reported in Table 2. populist attribute of mayors. Changes in performance and bureaucratic organisation might be the result of inexperience and reasonable change in administrative leadership resulting from the election of a new mayor (see e.g., Bolton, Figueiredo, and Lewis 2021). To rule out this possibility, we perform two additional tests removing i) incumbent mayors (i.e., mayors serving the second term), as well as ii) mayors who are serving the first term (i.e., challengers) but who are supported by an incumbent party. Specifically, we remove incumbent mayors and first-term mayor if at least one of the parties supporting the mayor was also supporting the previous, incumbent mayor. With these two tests we therefore address both candidate-and party-level incumbency status. Figure 3 shows the estimated effects across the total sample of mayors, the sample where we compare populist and non-populist challengers, and a sample where we compare populist and non-populist challengers who are not supported by incumbent parties. The estimates remain in the expected direction and distinguishable from 0 across all the two additional samples, suggesting that the populist attribute of the mayors -rather than their challenger status -is driving the results.
Second, except for the 5SM, the other populist parties are also right-wing, hence there may be a concern about what is driving the results, whether being populist or being right-wing.
While it is not possible to conclusively isolate the effects of the two components, we show that  Figure 4: RD estimates with 95% cluster-robust CI of the effect of electing a populist mayor under alternative coding strategies. Panel labels report the party omitted when coding the candidates as populist. Same estimation, covariates, and period of analysis reported in Table 2. the results are not exclusively driven by one single right-wing populist party. In Figure 4 we display the main results reported in Table 2

Conclusions
In this paper we explore how populism affects government economic performance and bureaucratic quality. We argue that when citizens lose trust in traditional political parties and elites, populist politicians have an incentive to propose a commitment-type policy platform that can be easily monitored by voters. We derive testable implications of populists' commitmenttype policies, and we test them in the context of Italian municipalities. We find that when a populist mayor barely wins the election debt repayments decrease and cost overruns on procurement contracts increase, suggesting lower financial sustainability and efficiency in government economic performance. We also find that populist mayors lead to higher turnover among top bureaucrats, who in turn are on average less educated.
Two notes on the generalizability of our results are in order, concerning the scope conditions of the theoretical set-up and the outcomes we selected for the analysis.
Our study of municipal government in Italy raises questions about the generalizability of our findings to other contexts, both vertically -to higher levels of government -and horizontallyto local governments of other countries. In particular, it is worth discussing two important scope conditions that, if matched, could possibly allow our argument and findings to travel beyond Italian municipalities: strong executive power, and a significant set of policy responsibilities granted to the mayor.
The concentration of executive power can be the product of multiple institutional set-ups, chiefly the electoral system -favouring single-party rather than coalition governments -and by the stock of policy responsibilities bestowed upon the executive by the constitution. Coalition governments might constrain the populist agenda, and inter-party bargaining can marginalize the demands of populist coalition partners. Similarly, political systems which do not grant considerable discretion to chief executives might pose institutional constraints on the commitmenttype agenda of populists. Therefore, we expect our argument to be informative about the effects of populism for central governments too, which clearly enjoy high levels of discretion and authority. This is in line with the evidence in Funke, Schularick, and Trebesch (2020) and Biard, Bernhard, and Beta (2019), who document the sizeable influence of populist politicians and parties on policies and the economy at national level. On the bureaucracy side, our framework is compatible with bureaucracies which can at least partially resist the will of elected politicians. Bureaucratic resistance ultimately requires a clear allocation of tasks and organisational structure and significant stocks of delegated discretion in managing policies. In political systems where crony or corrupted practices prevail it is harder for expertise to affirm as a distinctive feature of bureaucratic policy-making, and populist politicians might not be concerned about replacing experts with non-experts.
Power concentration and policy responsibilities are important factors highlighted in comparative accounts of local government and populism. For instance, Paxton (2019) finds that the populist mayor of the city of Padua in Italy managed to implement restrictive anti-immigrants policy as a result of the large formal power that mayors enjoy, whereas a feeble influence was achieved by the populist mayor of the city of Wels, in Austria, where municipal governments enjoy a lower degree of autonomy. These features are present in the local government of other countries too. German municipalities directly elect a strong mayor and enjoy a considerable degree of legal and fiscal autonomy (Höhmann 2017). Outside Europe, Brazilian (Akhtari, Moreira, and Trucco 2022) municipal governments display similar features to Italian municipalities, with large bureaucracies and significant autonomy for key public services. Conversely, the autonomy of the executive leader is arguably weaker in forms of government where the executive leader ultimately responds to the will of the local council. This is the case, for instance, for Spanish (Sanz 2019), Finnish, Swedish (Pettersson-Lidbom 2012, and French municipalities (Tricaud 2021), which -despite the marked decentralization in the management of public services -do not directly elect the head of the executive.
Finally, it should be noted that the effects we find on economic performance might be only one fraction of the factors affected by populist commitments. Longer-term effects might hamper the economic attractiveness of the municipality with detrimental consequences on investments and local economic growth (in a similar vein to the findings in Funke, Schularick, and Trebesch 2020). On the bureaucracy side, while the data allows us to prioritize the demand-side mechanism of turnover over a supply-side one based on bureaucrats' self-selection and voluntary departures, we cannot rule out alternative ways through which populism undermines bureaucracy. An important one, highlighted in theoretical work, is that those expert bureaucrats who remain in the administration can "pause" their commitment to good-quality policies and feign to be non-expert while waiting out the incumbent government (Sasso and Morelli 2021;Cameron and Figueiredo 2020). Future research could study other facets of performance and examine the conditions under which bureaucrats are willing to compromise on policy today to remain in their post tomorrow.
While it is important to highlight these specificities of the Italian context and more general scope conditions of the theoretical framework, we believe that the main gist of the paper is rather general. Populism is on the rise across the world 19 and it is likely to have sizeable consequences for the performance of government and interfere with the appointment and removal decisions that characterize the relationship between political principals and bureaucratic agents. 19 The aggregate 2020 vote share of populist parties in Europe doubled with respect to the total share in 2010 (from approximately 15% to more than 30%) (Rooduijn et al. 2019).

A Populist Parties in Italy
In this section we provide qualitative information about the five main populis parties in Italy.
The Lega (Nord) was established in 1991, with a strong regionalist and secessionist focus (Meardi and Guardiancich 2021). Together with Giorgia Meloni's Fratelli d'Italia, established in 2012, the new leadership of Salvini in 2013 moved the party to a more right-wing, conservative, nationalist platform, with a strong focus on anti-immigration and security. The Five-Star Movement, instead, was established as a more participatory and deliberative platform, advocate of direct democracy, and with a strong anti-elite identity (Mosca and Tronconi 2019). Despite some stark differences, they all share the aversion to austerity measures and see the European Union as a technocratic organisation which puts at risk the national interest (Caiani and Graziano 2019). Silvio Berlusconi's Forza Italia and, then, Il Popolo della Libertà, has been portrayed as the first instance of populist party in Italy, known for its anti-establishment and people-centric rhetoric against corrupt elites, intellectuals and members of the judiciary (Van Kessel 2015). However, the populist nature of Forza Italia, which replaced Il Popolo della Libertà after its dissolution in 2013, has been recently questioned in the literature. PopuList, for instance, stops coding Forza Italia as populist after 2018, for it is considered a borderline case (Rooduijn et al. 2019). In Figure 4 we show that our results are robust to omitting these two parties from the list of populist parties.
The features that can be found in Italian populism are far from being a distinct characteristic of Italian politics. The combination of anti-establishment sentiment with right-wing conservatism, holistic-representation ambitions, and economic nationalism can be found in many countries and regions of the world, from Latin to North America, Europe, Asian, and Africa alike [Moffitt2016].

B Datasets
In this section we provide detailed information on the source of data used to assemble the dataset.
We obtained election data directly from the Ministry of Interior. Similar data -with less detailed information on the denomination of the party-lists supporting each candidate -is reported in the Historical Electoral Archive accessible at https://elezioni.interno.gov.it/opendata. The dataset includes information about every mayoral candidate in the total population of municipal elections from 1989 to 2020. We focus on the elections where one populist candidate ran against at least one non-populist candidate. Municipalities with more than 15,000 inhabitants have a two-round electoral system, where the two most voted candidates compete in a second round when no one obtains more than 50% of votes in the first round. When a second-round occurred, we focused on the two candidates running in the second round. As a result, if a populist was running in the first round but did not qualify to the second round, the election is excluded from the sample.
Data on bureaucratic composition of municipal governments is obtained from the Annual Account of the Italian General Accounting Office, which is a department within the Ministry of the Economy and Finance (available at https://contoannuale.rgs.mef.gov.it/). The richness of this data allows us build fine-grained measures of bureaucratic turnover and level of education from 2001 to 2019 across all Italian municipalities. Importantly, we are able to focus on key bureaucrats within municipal governments, namely those with managerial rank. To do this, we subset each datasets of the Annual Account to macro-categories of contracts which contain the word dirigente (manager).
Database on Local Administrators with information on the number of local councillors and members of the executive committee, as well as the job, gender, date and place of birth of the mayor, is available at http://dait.interno.gov.it/elezioni/open-data. Data on debt repayment for the period 2008-2019 available at http://dati.statistiche-pa.it. Procurement data available at https://dati.anticorruzione.it. Demographic and territorial data were downloaded from the National Institute of Statistics database at http://dati.istat.it/.
Because not every dataset resorts to unique code identifiers, we alternated merging strategies using strings that combined both the municipality and region name, the unique identifiers assigned by the National Institute of Statistics, or the unique code attached to each municipality's budget data.

C Treatment and Control Group
In the table below we report a summary description of the treatment and control groups for the RD analysis.

Treatment Group
1) one populist candidate runs against one or more non-populist candidates in election year 2) the populist candidate barely wins 3) we analyse outcomes for that municipality during every calendar year until the next election in + 1 Control Group 1) one populist candidate runs against one or more non-populist candidates in election year 2) the populist candidate barely loses 3) we analyse outcomes for that municipality during every calendar year until the next election in + 1

D Validity of RDD
In this section we report falsification tests for the regression discontinuity design.

D.2 Continuity of Potential Outcomes
Mayor Municipality Robust Z-Statistic Figure D.2: RD robust Z-statistics of the effect of electing a populist mayor on pre-treatment covariates with vertical line at 95% confidence level. Unit of analysis is municipality-election year for covariates that do not change within the government term, whereas unit of analysis is municipality-calendar year for covariates that do change within the government term (i.e., population, number of rank-and-file employees, and n. of public managers). Estimates constructed using local polynomial estimators with triangular kernel and CER-optimal bandwidth [as suggested by Cattaneo2019, Ch. 5]. Robust p-values using bias-correction with cluster robust standard errors at municipality level. Same covariates as in Table 2 Red coefficients at true cutoff (margin of victory = 0). Blue coefficients when p.value after multiple testing adjustment is smaller than 0.05. Multiple-testing adjustment performed separately for each outcome variable with Bonferroni procedure to control for the false discovery rate. RD estimates constructed separately on control unit when placebo cutoff < 0, and on treated unit when placebo cutoff > 0. Placebo cutoffs very close to 0 (i.e., 1%, 2%) omitted due to small sample size. Estimation performed using local polynomial estimators with triangular kernel and MSE-optimal bandwidth. Confidence interval constructed using bias-correction with cluster robust standard errors at municipality level. Average 78 Table D.3: Share of non-significant estimates at placebo cutoffs ranging from −25 to +25 after multipletesting adjustment separately for each outcome as displayed in Figure D.3.  Figure E.4: RD estimates with 95% cluster-robust CI of the effect of electing a populist mayor with alternative bandwidths. Red coefficients estimated with MSE-optimal bandwidth and grey coefficients statistically significant at 95% level. Alternative bandwidths selected automatically ranging from half to 2 times the MSE-optimal bandwidth at interval of .1 (i.e., ℎ × , where = 0.5, 0.6, 0.7, ..., 2). Estimates constructed using local polynomial estimators with triangular kernel and MSE-optimal bandwidth. Robust p-values and confidence interval constructed using bias-correction with cluster robust standard errors at municipality level. Covariates include: population (log), surface (sq.km), surface at hydro-geological risk (sq.km), number of local councillors, gender, secondary education, degree, and white-collar job of mayor (all dichotomous), year, municipality, and year-election dummies. Estimate Figure E.5: RD estimates with 95% cluster-robust CI of the effect of electing a populist mayor with alternative bandwidth selection procedures (panel labels): mean-squared-error and coverage-error-rate optimal selectors both symmetric and asymmetric. MSE-optimal selector minimizes the asymptotic MSE of the point estimator, whereas the CER-optimal bandwidth minimizes the asymptotic coverage error rate of the robust bias-corrected confidence interval (Cattaneo, Idrobo, and Titiunik 2019). Blue coefficient for MSE-optimal selector used in baseline analysis. Estimates constructed using local polynomial estimators with triangular kernel. Robust p-values and confidence interval constructed using bias-correction with cluster robust standard errors at municipality level.

E.2 Power Analysis
Stommes, Aronow, and Sävje (2021) assess the validity of 36 studies using the RDD published in top political science journals and find that most studies are poorly powered to detect anything but large effects. To strengthen the validity of our results, we conducted power analysis with the method implemented in the rdpower package [Cattaneo2019e]. We estimate the power of a two-tailed test at the 5% significance level. We use the default settings of the package and investigate power with respect to an effect size equal to the one estimated and reported in Table 2. Table E.5 below reports the statistical power to detect such effects. Except for debt accumulation, the probability of detecting true positive is greater than the conventional 0.8 threshold for each of our outcomes.

F Additional Evidence on the Mechanism
In this section we provide additional tests to clarify or corroborate the mechanism underpinning our theoretical argument.

Outcomes
Hirings  Table F.9: RD estimates of the effect of electing a populist mayor on turnover and education of rank-and-file employees. Estimates constructed using local polynomial estimators with triangular kernel. Robust p-values and confidence interval constructed using bias-correction with cluster robust standard errors at municipality level, h is the MSE-optimal bandwidth. Covariates include: population (log), surface (sq.km), surface at hydro-geological risk (sq.km), number of local councillors, gender, secondary education, degree, and white-collar job of mayor (all dichotomous), year, year-election. Period of analysis: 2001-2019.

F.2 Supply and Demand side of Turnover
Here we describe the various categories of managers' departures analysed in Figure 2. For every municipality with public managers, we have data on the number of departures. Every departure is associated with one of eight categories listed in Table 3. These categories can be clustered in three groups: voluntary departures, forced departures, and a third cluster of unknown reasons.
Resignations consists of public managers unilaterally deciding to terminate their contract of employment. Similarly, it is public managers who need to apply and request to be transferred to other administrations. On the other hand, mayors can decide to terminate the contract of public managers (especially with fixed term contracts), they can fire them, and they can force the retirement when there are age limits. Normal retirements automatically occur after 40 years of work, and transfers due to outsourcing materialize when public services are outsourced to a public-owned company. There is also a big black box consisting of "other reasons", which is the largest category of managers' departures.

F.3 Heterogeneity Analysis
In Table F.10 below, we report the estimated differences in the RD coefficients from two samples: second and first half of the government term. There are no differences that are statistically significant at standard confidence levels. The results do not change if we perform the analysis separately for each year of the government term.   One reason why we fail to detect significant differences over the government term is that personnel changes are spread quite evenly over the government term, although with a slight decreasing trend, as evidenced by the descriptive statistics in Table F.11 below.

Turnover Departures Hirings
Year of Government Term Mean SD Mean SD Mean SD  Table F.11: Descriptive statistics of turnover, hired and fired managers over the government term. Departures and Hirings are averages of the number of managers that leave and join the municipality. Turnover is measured as the sum of managers who join and leave divided by the total number of managers in any given year.
As for the geographical differences, in Table F.12 below we report the estimated differences in the RD coefficients from two samples: municipalities located in the Centre/South and North of the country, which is the most intuitive geographical divide given the persisting socio-economic differences between the North and South of the country.
As displayed in the table, there are two differences that are statistically significant at standard confidence levels for two performance outcomes. These differences, however, are in contrasting directions. While populists elected in the Centre-South do a better job of repaying debts, they exhibit larger shares of procurements contracts with cost overruns.   In Table F.13 below we show the difference in the effects estimated from a subset of the dataset used for the analysis with population below and above the median (8,616 inhabitants). Overall, we find smaller effects in larger municipalities for performance outcomes and no difference for turnover and the percentage of graduate managers.