Annual financial
statements are one of the information sources used to assess the risk that a
firm is involved in money laundering, tax fraud or organised crime
infiltration. This paper compares the detection tools based on the financial
statements of Italian firms, developed along two distinct strands: academic
research and the studies of the Financial Intelligence Unit for Italy (UIF).
The five methodological approaches examined – a logistic regression model, two
composite indicators, and two machine learning models – serve as tools for
identifying firms linked to organised crime, companies issuing invoices for
non-existent transactions, and firms interposed in complex tax fraud schemes.
The comparison considers four dimensions: detection purpose and target,
reference firms, information set, and analytical method. It shows that risk
assessment depends on three main choices: which financial statement items to
use, which firms to take as reference, and how the data are processed. As these
choices vary across tools, financial statements do not provide a uniform
information base, and differences in reference firms, validation conditions,
and performance measures make reported results not directly comparable. The
paper’s originality lies in proposing a criterion for variable selection: when
variables have equal predictive power, preference should be given to those that
firms targeted for detection can alter only at a high cost. This provides a
descriptive and conceptual contribution by introducing the cost of manipulating
observable characteristics as an additional criterion for selecting predictors.
