"Rebel on the Canal Disrupted Trade Access and Social Conflict in China, 1650–1911", conditionally accepted at the American Economic Review https://www.yimingcao.com/uploads/6/5/6/3/65630513/canal.pdf
This paper finds that the closure of China’s Grand Canal, which served as a permanent shock to regional trade access, led to more rebellions along the canal, with disrupted trade access being the underlying causal mechanism. It uses 1826 as the starting date of Canal closure, and describes that "the reform was implemented gradually".
However, @Zhangtaisu writes on twitter:
"This is an interesting paper, and the basic conclusion that Qing grain tribute trade routes had a large effect on regional socioeconomic stability is probably correct. However, there’s one big oddity in the results, which is that its effects begin with the first sea-shipping experiment by the Qing Court in 1826, instead of the more permanent sea shipping reforms in the later 1840s. As anyone who knows this history can tell you, the 1826 experiment lasted for a grand total of 1 year, and canal shipping resumed at normal levels shortly afterwards. It wasn’t until two decades later that more systemic changes were put in place. So why would a 1 year experiment have such a large and durable effect on local stability? You might say that local populations could anticipate the 1840s reforms after 1826, but that seems highly unlikely given the Imperial Court’s rapid backtracking. So what was the casual mechanism in the 20 years in-between? The paper really needs an answer for this."
The paper does *not* mention the full reopening of the canal in 1827. So now the problem becomes: if authors use late 1840s as the starting date, which is the *correct* date of permanent canal closure, there would be severe pre-trends in their data. If authors use 1826 as they do now, they get good results and zero pre-trend but now the causal mechanism is very unlikely to be be "disrupted trade access" as they claim because trade resumed in 1827.
Shuo Chen and Yiming Cao can quickly shut up Taisu Zhang by running another regression with moved DID start year.