Corporate Financial Distress: From theoretical framework and prediction models to restructuring strategies, and an empirical framework for knowledge-intensive firms

The theme of this work is corporate financial distress. The thesis develops across five interconnected chapters. The first establishes the theoretical framework of corporate financial distress, examining its causes, lifecycle dynamics, and spillover effects. The second surveys the landscape of crisis prediction models, from the pioneering ratio-based analyses of Beaver to the multivariate models of Altman and Ohlson, through to the Italian contribution and the frontier of machine learning. The third shifts the focus from diagnosis to response, examining the tools of financial and operational restructuring, the turnaround frameworks of Bibeault and Slatter and Lovett, and the role of private equity in the governance transitions of Italian family SMEs. The fourth applies these frameworks empirically, through an ex-post Z-Score analysis of Amazon.com (1997–2007) and a creative accounting detection study of Bending Spoons S.p.A. (2019–2024) using the Beneish M-Score and the Sloan Accruals Ratio. Both cases demonstrate that quantitative models identify anomalies accurately, but cannot distinguish their cause without qualitative judgment. The fifth chapter develops and tests the KI-Score, an original distress prediction model incorporating intangible asset intensity and earnings quality variables, calibrated on technology SMEs through three sequential empirical studies on Italian and European samples drawn from AIDA and ORBIS. Across all three studies, the KI-Score consistently outperforms the Altman Z-Score, with AUC gains ranging from 9 to 16 percentage points. The unifying conclusion is both simple and fundamental: accounting data can tell us whether a company is heading toward crisis, but only if the models are used with an awareness of their assumptions, their limits, and the context in which they are applied. Quantitative models are powerful screening tools, not self-interpreting instruments. The real analytical work begins after the model has spoken.