Accounting Information Quality and Its Influence on Lending Decisions

Abstract

This research introduces a novel, cross-disciplinary framework for evaluating
accounting information quality (AIQ) by integrating principles from computational
linguistics, network theory, and behavioral finance, moving beyond traditional financial ratio analysis. We posit that the structural and semantic properties of
financial disclosures—often overlooked in conventional lending models—contain
predictive signals about borrower creditworthiness that complement quantitative
metrics. Our methodology employs a hybrid approach combining natural language
processing (NLP) to assess narrative coherence and transparency in management
discussion and analysis (MDA), alongside a graph-based model to evaluate the relational integrity and consistency between financial statement line items. We test
this framework on a unique dataset of small and medium enterprise (SME) loan
applications, where traditional collateral and cash flow data are often limited. The
results demonstrate that our composite AIQ score, derived from textual and relational features, significantly improves the predictive accuracy of default models
compared to models using only financial ratios. Notably, we find that semantic
clarity in risk disclosures and the topological robustness of the accounting network are strong, non-linear predictors of loan performance, especially for younger
firms. This study contributes an original, computationally-grounded lens to the
assessment of accounting quality, offering lenders a more nuanced tool for credit
decision-making in informationally opaque environments and challenging the primacy of purely quantitative analysis in lending models.