Integrating Linear Regression and Multi-Criteria Decision Making for Assessing Financial Statement Risks in Manufacturing Firms

Technical Summary: Integrating Linear Regression and Multi-Criteria Decision Making for Assessing Financial Statement Risks in Manufacturing Firms

Problem Statement
The evaluation of financial performance in manufacturing firms faces a methodological fragmentation. Existing literature typically separates two critical approaches:

1. Theoretical Time-Discounted Models: These rigorously incorporate the time value of money (TVM) but often rely on oversimplified assumptions that fail to capture the multidimensional and interdependent nature of economic, operational, and managerial decision criteria.
2. Expert-Based Weighting Methods (e.g., AHP): These provide descriptive rankings of criteria based on expert judgment but lack empirical validation regarding their actual influence on discounted performance outcomes.

There is a critical gap in the literature regarding an integrated framework that quantitatively bridges time-discounted economic theory with data-driven analysis of multiple evaluation criteria. Traditional linear models and historical ratios often fail to capture complex relationships, while pure AI approaches may lack interpretability. This study addresses the need for a transparent, interpretable method that combines theoretical discounting with empirical regression to assess control system efficiency and financial risks.

Methodology
The proposed framework employs a hybrid approach combining deterministic discounting models with linear regression analysis.

1. Theoretical Discounting Model:
   The study first develops a model to convert costs and benefits occurring at different times into present-value terms using compound interest functions. The model accounts for three types of financial flows:

   • One-time expenditures: Discounted using the benefit function $\theta(t) = (1 + \alpha)^{-t}$.
   • Time-proportional costs (e.g., operating savings): Modeled as continuous flows integrated over time, yielding a total discounted value $r' = I\tau(1 - e^{-t/\tau})$, where $\tau = 1/\ln(1+\alpha)$.
   • Complex cost structures (e.g., system development): Costs are attributed to a "center of gravity" of the project duration ($t_y$). The discounted cost at startup is calculated as $C'_y = C_y(1 + \alpha)^{0.5t_y}$.

2. Linear Regression Integration:
   To empirically assess how these discounted economic performances are influenced by multiple criteria, the study utilizes linear regression as an approximation method.

   • Variables: The dependent variable is the discounted economic effect ($\bar{R}_i$). The independent variables are Average Weight Scores (AWS) derived from expert assessments. These include weights for primary criteria (e.g., Cost Structure, Investment Risk, Reliability, Cash Flow) and secondary criteria.
   • Model Specification: The relationship is modeled as:
     $$\bar{R}_i = \beta_0 + \beta_1 CSR_i + \beta_2 LR_i + \beta_3 IR_i + \beta_4 CFR_i + \varepsilon_i$$
     Where $\beta_j$ represent the regression coefficients quantifying the marginal contribution of each criterion to the overall discounted performance.
   • Estimation: The coefficients are estimated using Ordinary Least Squares (OLS). The study also references Ridge regression as a potential extension to handle multicollinearity, though the primary results presented rely on the standard OLS formulation.

Key Results

• Quantification of Criteria Influence: The regression analysis successfully quantifies the contribution of specific criteria (such as cost structure and investment risk) to the overall discounted economic performance. The coefficients provide a transparent basis for ranking criteria according to their actual economic impact rather than just theoretical importance.
• Model Fit and Linearity: In a preliminary analysis of criterion weights, the study observed that while the relationship between rank and weight is not perfectly linear for the top positions (where top-ranked criteria carry disproportionately large weights), the relationship becomes closer to linear from rank 5 onward. The overall coefficient of determination ($R^2 \approx 0.55$) indicates a moderate linear fit, suggesting that while linear regression serves as a useful baseline, it captures only part of the variance.
• Interpretability: The model successfully bridges the gap between abstract discounting principles and empirical data. It demonstrates that expert-derived weights can serve as effective explanatory variables for predicting discounted economic outcomes.

Significance and Claims
The paper claims that this integrated framework offers a practical advancement in financial performance evaluation for manufacturing firms by:

• Bridging Theory and Data: It connects theoretical time-discounting principles with empirical data, overcoming the limitations of purely theoretical models (which oversimplify interactions) and standalone expert-weighting methods (which lack empirical validation).
• Enhancing Decision-Making: By providing a transparent, data-driven ranking of evaluation criteria based on their marginal economic contributions, the approach supports more robust, evidence-based decision-making regarding control system efficiency and resource allocation.
• Improving Interpretability: Unlike "black box" machine learning models, this linear regression approach maintains high interpretability, allowing managers to understand exactly how factors like cost structure or cash flow characteristics influence financial performance.

The authors maintain a modest stance, acknowledging that the linear model is an approximation. They note that the framework assumes linear relationships and that future research could improve predictive accuracy by incorporating nonlinear relationships, dynamic panel data models, or industry-specific calibrations. The study does not claim to replace advanced AI but rather to provide an interpretable baseline that effectively utilizes expert data within a rigorous economic framework.