Siqi Chen.

This project systematizes more than 200 legal documents, distinguishing localization mandates, conditional transfer mechanisms, and discretionary state authority while accounting for statutory exceptions and overlapping instruments.

A six-model LLM-as-judge pipeline in Python uses anchored scoring rubrics and structured JSON outputs to construct the Cross-border Data Restriction Index. Validation includes cross-model reliability (ICC = 0.791), human benchmarking (r = 0.83 across 53 texts), and jurisdiction anonymization tests.

Methods: LLM-based measurement · Text as data · Measurement validation

Firms weigh the compliance costs of data protectionism against their outside options, while governments selectively accommodate firms aligned with industrial policy priorities. The framework draws on firm heterogeneity, New New Trade Theory, and state–multinational corporation bargaining.

The empirical design defines exposure by industry data intensity and estimates heterogeneous exit patterns using firm, year, and multinational firms’ home-country fixed effects, company-clustered standard errors, pre-trend tests, and robustness checks.

Methods: Event-study difference-in-differences · Firm-level panel analysis

The theoretical framework maps firms’ compliance choices—including building data centers, purchasing foreign affiliates, and contracting local providers—to predictions about host-country service exports and imports.

The analysis combines two-way fixed-effects panel models with estimators robust to heterogeneous treatment effects (FEct, IFEct, Matrix Completion, and CSDID). Sectoral analyses use UN Comtrade and UN ServiceTrade (BPM6) data, with lagged country-level controls and country-clustered standard errors.

Methods: Panel regression · Heterogeneous treatment effects · Sectoral analysis