央行独立性: 来自历史和机器学习的视角

We assemble an almost complete set of central bank statutes since 1800 to assess the legal independence of central banking institutions. We use these to extend existing indices of legal independence backward and forward in time. We document the trend toward increased independence post 1980 as well as an earlier, more limited movement in the direction of enhanced independence in the 1920s. We apply natural language processing to current statutes to corroborate our human-reader assessment. Using machine-learning methods, we quantify the extent to which topics in those statutes contribute to the independence measure based on our reading of the statutes. The topic with the largest positive contribution to explaining the cross-country variation in central bank independence encompasses disclosure, transparency, and reporting obligations. The topic with the largest negative contribution covers regulatory powers over inter alia securities markets that complicate the central bank's mandate, make accountability more complex, and render independence problematic.