FORMATION OF THE MDA MODEL FOR FORECASTING FINANCIAL DISTRESS AS THE BASIS OF ANTI-CRISIS FINANCIAL MANAGEMENT OF AGRICULTURAL ENTERPRISES

Abstract

The article aims to develop an empirically verified financial distress forecasting model for agricultural enterprises and to justify its use as a formal early-warning instrument within anti-crisis financial management. The author’s contribution is twofold: a reproducible process architecture that links problem formulation, data preparation, model estimation, validation, threshold calibration and managerial integration; and an applied discriminant risk index constructed on the financial statements of agricultural enterprises of Bolhrad district. Financial distress is operationalized through the YouControl financial scoring class and specified for a one-year forecasting horizon, where explanatory indicators are taken for the previous year and the target label is assigned for the subsequent year, which prevents information leakage. The initial population comprises 88 enterprises; after applying inclusion and exclusion criteria the sample contains 84 enterprises, and 80 complete observations are retained for modelling, with an imbalanced class structure (13 distress and 67 non-distress cases). To enhance robustness, extreme values of candidate indicators are limited by winsorization at the 1st and 99th percentiles, followed by standardization to comparable scale; multicollinearity is controlled via correlation screening and pre-filtering across key financial blocks. Variable selection is conducted with a stepwise discriminant procedure guided by Wilks’ lambda and classification metrics, emphasizing sensitivity to the distress class and balanced accuracy rather than overall accuracy alone. The recommended specification includes indicators of leverage, equity manoeuvrability, absolute and intermediate liquidity, the ratio of receivables to payables in commercial operations, and sales profitability. The selected model demonstrates high sensitivity (0,9231) with balanced accuracy of 0,6556; reflecting an intentionally cautionary profile with a low probability of missing distressed enterprises but a notable rate of false alarms among financially stable firms. Scientific novelty lies in the combined decision rule for specification choice (structural separation plus balanced classification quality) and in transforming the discriminant index into a four-zone risk scale. Thresholds are calibrated using the receiver operating characteristic curve to obtain a low-risk boundary, a screening boundary and a critical action boundary, enabling gradual escalation of managerial response. Practical value is provided by a ready-to-implement zoning framework that supports routine monitoring of liquidity, capital structure and profitability and translates model signals into standardized anti-crisis actions with feedback for re-calibration. The conclusions substantiate that the zonal scale increases managerial interpretability and supports preventive decisions before legal insolvency procedures occur.