After the abrupt end of China’s Zero COVID policy at the end of 2022, the debt held by local government financing vehicles (LGFVs) on behalf of their local governments had soared to at least US$8 trillion. Unable to resolve the problem, some local governments now are cutting public services due to a lack of funds. COVID spending itself is not a sufficient explanation of the crisis. Using an array of quantitative data, we show that changing central government policies at different points during the pandemic played a determining and sometimes catastrophic role in creating debt and undermining the operation of LGFVs. The three red lines against the real estate sector and the call for LGFVs to take up the slack and buy land no longer wanted by a collapsed sector eroded and eventually pulled the rug out from under the inherently vulnerable LGFVs. The mountains of LGFV debt facing local governments across China today are more a result of policy decisions than COVID expenditure shocks. A perfect storm of policy instability in a pandemic exposed the vulnerability of longstanding weak institutions that have troubled localities since the mid-1990s, underscoring the need for systemic fiscal reform.

Does the adoption of artificial intelligence (AI) solve the principal-agent problem inherent in bureaucratic control in autocratic regimes? In this paper, I argue that the localized procurement and deployment of government administrative AI technologies exacerbate pre-existing information asymmetries between the center and its localities, and increase the likelihood of local agents strategically circumventing central oversight. I draw on an original dataset of over 31 million central and local government procurement documents from 2002-2022. Leveraging a 2015 policy change that instituted a monetary size reporting threshold and a robust design including DiD and event study, I construct the first-of-its-kind measures of local strategic actions aimed at evading central oversight. I find that among a range of local political and economic factors, investments in government administrative AI technologies stand out most, significantly associated with less central control. This paper challenges the conventional wisdom, showing that the adoption of AI in authoritarian regimes can paradoxically weaken bureaucratic control, particularly when there is a misalignment of interests between the center and its localities, and when the technology is in the hands of state agents.

What determines political promotion in authoritarian regimes? Do autocrats sort on the competence or political loyalty of their subordinates? A long-running scholarly debate in both economics and political science concerns whether the selection of leaders in the Chinese bureaucracy is meritocratic or personalistic. Using the latest demographic, career experience, and economic performance data for more than four thousand leaders starting from the prefecture- level in China from 1995-2015, I find that the career outcomes of political leaders are largely determined by patronage networks formed in early periods of their career. Early patronage relationships remain the best predictor of a leader’s later career advancement even after controlling for economic performance. Adopting a novel network-based measurement strategy, this study goes beyond existing one-to-one, binary coding schemes of political connection to capture neighborhood views of leaders in the network. To the best of my knowledge, this paper is one of the first to use advanced machine learning and network analysis methods to quantify the network impact of political patronage on promotion and to compare it with performance among nationwide politicians in large bureaucracies.

This paper examines political censorship on WeChat, the most popular social media platform in China, as an instance of information control in authoritarian regimes. The study investigates censorship behaviors on ten high-profile public accounts that discuss politics to answer the question: does censorship in China permit political criticism? The answer may differ across various media platforms. While previous empirical evidence suggests that criticizing the Chinese government on Weibo and BBS forums does not increase the chance of being censored, this analysis of WeChat shows the opposite. Drawing on 3,015 human-coded long articles from WeChat, I find that criticizing the government substantively increases the possibility of removal – massively higher for explicit criticisms and substantially for implicit ones. This suggests that China’s censorship program is platform-dependent instead of uniform across all media platforms, and the target of censorship depends on the platform’s capacity to mobilize collective action and to host long political criticism. Interviews with former censors and employees from internet service providers further reveal a classic principal-agent dilemma in the censorship implementation process. These findings have potentially significant implications for reconsidering censorship theories, bringing private companies into censorship studies, and examining strategies that authoritarian regimes employ to perpetuate their rule in the digital age.