Replication Data and Code for “Trade Policy Has Memory: Path-Dependent Adjustment under Repeated Tariff Shocks”
Description
This dataset is the replication package for the paper “Trade Policy Has Memory: Path-Dependent Adjustment under Repeated Tariff Shocks” by Chen Kenan and He Qingren. The paper studies the dynamic effects of repeated U.S. tariff shocks on Chinese exports, combining the 2018 Section 301 episode, the evolution of product-level U.S. exposure between shocks, the renewed tariff shock in 2025, and the partial policy reversal in 2026. The package contains analysis-ready datasets, tariff-treatment and policy-timing inputs, HS concordances, estimation code, figure and table code, and verification routines. It reproduces the paper’s results on short-run trade destruction, the evolution of destination exposure, state-dependent incidence under the 2025 shock, the approximately USD 50.2 billion exposure counterfactual, and the asymmetric restoration dynamics following the 2026 policy reset. It also includes robustness analyses using alternative seasonal benchmarks and flexible functional forms for the exposure counterfactual. Large external provider databases are not redistributed where redistribution is restricted or unnecessary. Instead, the package includes compact analysis-ready panels together with source documentation and data-preparation programs that reconstruct the analysis layer for researchers with lawful access to the underlying source files. The main source families are UN Comtrade, U.S. Census merchandise trade data, China Customs trade statistics, CEPII BACI, official tariff and policy records, and HS concordance files. The default replication begins from the bundled analysis-ready data. Running the master script regenerates the estimation results, main figures, online-appendix figures and tables, and a numerical verification report. The frozen package passes 62 automated verification checks with zero failures.
Files
Steps to reproduce
Software requirements Python 3.11 or later is recommended. Required Python packages are listed in requirements.txt. Replication from the bundled analysis-ready data Download and extract the replication package. Open a terminal in the root directory of the package. Create and activate a Python virtual environment if desired. Install the required packages: pip install -r requirements.txt Run the master replication script: python run_all.py The script creates output/generated/, re-estimates the paper’s results, regenerates the figures and table inputs, and runs numerical verification. A successful replication ends with: Verification complete: 62 checks; 0 failures. Upstream reconstruction from original source files The default replication starts from the bundled analysis-ready panels because some large provider source files are not redistributed. Researchers with lawful access to the original source files can reconstruct the analysis layer using the programs in code/preparation/ and the exact source-to-panel instructions in DATA_PREPARATION.md. The principal preparation modules reconstruct the first-shock panels, U.S. importer panels, long-run BACI panels, second-shock and exposure panels, and the 2026 court-reset panels. The reconstructed files are written to data/analysis/, after which python run_all.py reproduces the estimation and exhibit pipeline. Canonical reference-result files are stored only for validation. The estimation routines do not read these files; they are used exclusively by code/verify.py to confirm numerical parity.