Runxi Niu

This paper studies how credit assessments shape firms’ financial vulnerability before downturns. Using Norwegian firm data, I find that financial information predicts bankruptcy within agency rating groups. Better ratings are linked to lower borrowing costs among firms with similar measured risk. I build a model in which lenders’ assessments change loan prices and firms’ financing decisions. Assessment-based pricing raises credit by 0.9%, but increases losses from a later 5% fall in aggregate productivity by 4.5%, when both economies face the same loan pricing and aggregate prices. About two-thirds of the gap comes from changes in credit allocation across risk groups. The extra risk remains after the rating advantage fades but adds little to the fall in total investment. In general equilibrium, differences in real activity are larger after bank-capital shocks than productivity shocks, partly because of loan pricing during the shock.

We develop a new approach to evaluating environmental policies in settings where pollution spills across administrative boundaries. Our approach integrates atmospheric particle-trajectory data with natural-experiment designs to construct a measure of policy exposure at affected locations, regardless of whether they lie inside or outside regulated jurisdictions. This addresses a key limitation of conventional jurisdiction-based designs, which may misclassify treated locations when pollution travels across borders. Applying the method to China’s Two Control Zones policy, we show that conventional estimates become increasingly attenuated when cross-boundary spillovers are important, while our approach better captures the policy effect. When spillovers are limited, the two approaches perform similarly.

We study which texts help forecast U.S. CPI inflation at horizons of one to twelve months. Our retrieval-augmented generation (RAG) system combines historical CPI data with passages from central bank communications and financial news. In an out-of-sample evaluation, the best RAG configuration outperforms conventional econometric benchmarks, with larger gains at longer horizons. Central bank communications help more at short horizons, while news contributes more at longer horizons. Text also improves density forecast calibration, bringing prediction interval coverage closer to its nominal level. Demand and energy text help forecasts, while monetary policy content is consistently associated with worse accuracy.