CFOs: Surviving in a New Era - AGA

16
Predictive and statistical analytics
Quantitative analytics are hardwired into the
operations cultures of many government entities;
indeed, some offices and even whole agencies like the
U.S. Census Bureau exist simply to apply predictive
and statistical analytics in ways that produce useful
information for planners, marketers, citizens and
decision makers. Yet, predictive and statistical
analytics have yet to permeate the administrative
decision making of many government entities,
including their financial functions. This is changing,
but very gradually.
Predictive or statistical analytics or modeling uses
data mining, statistical analysis, game theory and
geospatial analysis to extract information from
data, and then applies it to predicting trends and
patterns and to identifying emerging phenomena.
This is useful for risk management, helping to
prevent bad things from happening and to ensure
that good things happen as intended. The core
of predictive analytics relies on capturing relationships
among explanatory variables and the
predicted variables from past occurrences, then
exploiting knowledge of the relationships to predict
future outcomes. Government entities in our
survey use these tools and methods for:
• Collecting revenues or fees
• Credit scoring for loans
• Detecting erroneous and improper payments
• Detecting fraud
• Forecasting environmental trends
and behaviors
• Identifying risk profiles
• Optimizing resource allocations
• Predicting program portfolio or
economy levels
• Setting priorities for resource allocations
• Underwriting
• Validating budget processes and assumptions.
Analytic tools in
financial management
We asked federal executives what types of predictive
and statistical analytic tools they use as part
of financial management for their entities. The
most frequent responses were:
Data
mining
Pattern
Trend
Pattern
Data Data Data
Statistical
analysis
Game
theory
Geospatial
an
• Sampling techniques used for testing internal
controls, audits and other related purposes
• Dashboards and balanced scorecards, often
manual but sometimes linked to sophisticated
data-mining systems
• Predictive modeling, based primarily on
historical internal data but sometimes incorporating
outside factors and data for “what-if”-
type analysis
• Trend analysis using historical data
• “Homegrown” applications, typically based on
spreadsheets or PC database software, often
searching for anomalies