Thousands of papers, one dataset: multiple testing in research on Chinese listed firms
"Chinese listed companies" is the third most frequent keyword across the tracked journals over the past year, and most of these papers draw on the same few databases. Yet none of the 1,392 recent papers with such a keyword carries a replication, multiple-testing or publication-bias keyword. How many of their findings would survive a correction for everyone testing on the same data?
What prompted this#
On the trends page, "Chinese listed companies" is the third most frequent keyword over the last 12 complete months, after "China" and "large language models". Since October 2025, 1,392 papers carry a Chinese-listed-company keyword (928 in finance journals, 216 in economics, 172 in management, 54 in accounting). None of them carries a multiple-testing, p-hacking, replication, publication-bias, data-mining or false-discovery keyword.
Questions#
- If every published test on the same panel of firms counts as one of many, how many reported effects clear a stricter threshold, such as a t-statistic of 3?
- Do the results hold out of sample, for example in later years or in firms listed after a paper's sample ends?
- Which outcomes, such as green innovation, ESG scores or digital transformation, attract the most tests?
Possible approach#
Collect the main reported t-statistics and sample periods from a large set of papers that use CSMAR or Wind data, and estimate the false discovery rate with the methods developed for the "factor zoo" in asset pricing.
Next step#
Extract the main coefficient and t-statistic from 200 recent papers on Chinese listed firms and plot their distribution around 1.96.