Centralization of public procurement can lower prices for the government's direct purchase of goods and services. This paper focuses on indirect savings. Public administrations that do not procure directly through a central procurement agency might benefit from the availability of centrally procured goods. We exploit the introduction of a central purchasing agency in Italy and find that prices came down by 22% among administrations that bought autonomously. These indirect effects appear to be driven by informational externalities, especially for less competent public buyers purchasing technologically more complex goods. Accounting for indirect savings increases the estimate of direct ones.

I examine the extent to which news of a local politician's corruption leads to electoral spillovers to the political party in neighboring jurisdictions and presidential candidates. Using random audits in Brazilian local governments along with novel data on antenna location and coverage, I identify relevant neighboring jurisdictions. The findings reveal that electoral spillovers extend beyond the area of origin, affecting neighboring local politicians and presidential candidates. Furthermore, by analyzing a decade's worth of national newspaper articles, I show that local corruption gets national media attention, particularly in cases of high corruption. The sentiment of news coverage during national elections is more likely to be negative for parties with national candidates, highlighting the significant role national media plays in amplifying local events and shaping the sentiment.

This paper develops an automated measure of municipal corruption using Brazil’s randomized municipal audit reports from 2003 to 2015. I classify each reported irregularity as severe or non-severe using a dictionary-based approach, then combine the resulting severity counts with structural report features, number of pages, lines, images, and total irregularities, into a single corruption index via principal component analysis. The dictionary’s classification is robust to alternative methods: a supervised classifier trained on hand-coded labels produces nearly identical counts and a nearly identical index (r > 0.98), and GPT-5.3 agrees with the dictionary on all 15,000 randomly sampled irregularities (approximately 12% of the corpus). The dictionary classifier requires neither labelled training data nor a large language model; both are used here only on subsamples, as external benchmarks for validation. Conditioning on state fixed effects, the corruption index correlates strongly with independently hand-coded measures (R2 ≈ 0.70–0.90) and covers twice as many reports as existing hand-coded measures. Using the index, I document a pronounced electoral cycle: corruption declines by 0.15–0.22 standard deviations in the election year and the year preceding it, but is indistinguishable from baseline farther from elections. The cycle does not vary systematically with mayoral party affiliation and is present among both re-electable and term-limited mayors, providing no evidence that partisan enforcement or individual re-election incentives are primary drivers.

We study how legal threats to social media platforms shape online political communication. We exploit the sharp implementation of a 2020 Turkish law that required social media companies to establish local offices and provide user data to the government on request. Assembling a panel of over 700 Twitter users and 7 million posts, we use a differences-in-differences design to compare dissenters, individuals publicly critical of the reform when announced, to other groups unlikely to be affected, such as celebrities. We find that dissenters and political exiles post 15-20 percent fewer tweets following implementation, and engagement with their content falls sharply. These short-run chilling effects emerge despite limited platform compliance, indicating that anticipated enforcement risk, rather than actual enforcement, drives the effect. In the long run, the chilling effects dissipate. Our findings suggest that states can govern online speech indirectly, by regulating platforms rather than users, but that such effects may be self-limiting in the absence of visible enforcement.

We investigate the role of new communication technologies in the fall of autocratic regimes. For this, we use a unique setting: the distribution of photocopy machines, the Xerox program, in communist Hungary. The photocopy machines were seen as a transformative technology, similar to the impact of the Gutenberg press. We use newly digitalized data on machine allocation between 1985 and 1989 and show that areas with machines are more likely to support democratic values, participate in elections, and establish more entrepreneurial activity in the short and long run. Moreover, we show that adjacent areas are also affected. Our results suggest that new communication technologies help overthrow autocracies by promoting democratic values even when political competition is limited and traditional media is censored.