Narrative Disclosure and Private Credit Risk: Text-Based Evidence from BDC Filings Amid Macro-Financial Shocks

Abstract

A persistent difficulty in monitoring private-credit risk is that narrative and quantitative information in periodic filings are produced jointly but evaluated separately. This leaves open the question of whether disclosure language is a useful signal of risk management behaviour or merely an echo of conditions already visible in published data. For business development companies (BDCs), this separation carries a particular cost: the sector sits at the intersection of private credit, fair-value accounting, and floating-rate funding, where filing language about portfolio conditions and the macro environment may reflect the cycle itself rather than add to what published rate and spread data already reveal. This paper asks two questions. First, do aggregate BDC text measures of macro and portfolio-credit language co-move with key macro series over time? Second, does cross-sectional text intensity relate in a stable, linear way to the same BDC’s reported ratios and their volatility? Using dictionary-based filing scores linked to over 590 BDC observations and macro series from 2010 to 2025, we find macro text in filings correlates strongly with variables such as the Federal funds rate and the two-year Treasury yield. Portfolio-credit text lines up with corporate spreads and the unemployment rate. At the firm-year level, associations between text and balance-sheet outcomes are weak. This indicates that BDC narratives are linked with the macro cycle, but there is not a tight mapping to risk metrics in reported financials from year to year, consistent with a degree of insulation in private credit from prevailing macro conditions. For creditors, investors, and supervisors of private-credit vehicles, this asymmetry of macro co-movement without firm-level signal has direct implications for how narrative disclosure should be weighted in risk monitoring and governance frameworks. The aggregate regression results are based on sixteen annual observations and should be interpreted accordingly.