Predictive Modeling of Environmental Fines Using Historical Accounting Data

Abstract

This paper introduces a novel methodological framework for predicting corporate
environmental fines by leveraging historical accounting data, a previously underexplored data source in environmental compliance forecasting. Traditional approaches
to environmental risk assessment have relied heavily on direct environmental metrics,
regulatory history, and industry-specific factors, often overlooking the rich predictive
signals embedded in financial statements. We propose that patterns in accounting
data—including expense allocations, capital expenditure trends, depreciation methods,
and footnote disclosures—contain latent indicators of environmental management priorities and potential compliance vulnerabilities. Our research formulates the prediction
of environmental fines as a time-series classification problem, employing a hybrid model
that combines autoregressive integrated moving average (ARIMA) components for capturing temporal financial trends with a multilayer perceptron for capturing non-linear
interactions between accounting variables. We construct a unique dataset spanning
1998 to 2004, linking the financial statements of manufacturing firms from Compustat
with environmental penalty records from the EPA’s Enforcement and Compliance History Online (ECHO) database. The model identifies several key accounting predictors,
most notably the ratio of repair and maintenance expenses to capital expenditures, the
volatility of cost of goods sold, and specific linguistic cues in management discussion
and analysis (MDA) sections related to environmental contingencies. Results demonstrate a predictive accuracy of 82.7% in classifying firms that will incur a significant
environmental fine within the next fiscal year, a substantial improvement over baseline
models using only environmental performance indicators. This work establishes a new,
cross-disciplinary research direction at the intersection of environmental informatics,
forensic accounting, and predictive analytics, offering regulators and investors a proactive tool for risk assessment derived from routinely published financial information.