20240229-IMF-Republic_of_Slovenia_Technical_Assistance_Report-Data-Driven_Risk_Assessment_Follow-Up_11页_319kb

Technical Assistance Report Summary: Republic of Slovenia – Data-Driven Risk Assessment (Follow-Up)

I. Core Content and Purpose

This report is a follow-up to the November 2022 field-based mission aimed at enhancing the Republic of Slovenia's Revenue Administration (SFA) capabilities in data-driven risk assessment for corporate income tax (CIT). It outlines the progress made by SFA in adopting the tools and methodologies recommended by the IMF's Technical Support Team (STX), as well as the next steps and recommendations for further improvements in compliance risk management.

II. Main Recommendations and Key Points

1. Emphasis on Data Utilization and Integrity

• The SFA has already laid a strong foundation for data analytics, with a competent "Analysis unit" and timely investments in data infrastructure.
• The priority is to systematically use existing data sources (e.g., audits, tax returns, Country-by-Country-Reports, balance sheets) to improve risk assessment models.
• Data integrity is crucial, and the use of unique taxpayer identifiers enables seamless merging of datasets.

2. Implementation of Empirical Risk Assessment Models

• The SFA has developed four pilot empirical models, including:
  • An advanced CIT audit selection model (Random Forest)
  • An advanced VAT audit selection model (Random Forest)
  • A workflow to evaluate existing business risk rules against historical data
  • A Country-by-Country-Reporting (CbCR) anomaly detection tool
• These models aim to provide real-time risk assessment using all available data, with the CIT model increasing predicted strike rates (corrections > €5,000) from 30% to 70%.

3. Focus on Taxpayer Size and Risk Profiling

• The SFA should increase the emphasis on taxpayer size in risk assessment.
• A risk profile for the 30 largest taxpayers should be created, which can be automated using transfer pricing (TP) indicators.
• Client account managers should be assigned to these large taxpayers to ensure compliance.

4. Use of Simple Analytics and Benchmarks

• The STX recommends starting with simple analytics (e.g., deviations from industry averages) to evaluate business risk rules.
• Benchmarks and standard deviations should be used to root risk analysis in systematic data rather than intuition.

5. Gradual Introduction of Advanced Models

• The SFA should prioritize simpler risk analysis methods before moving to advanced predictive models.
• The Random Forest model is a good example of an advanced tool that should be used after the groundwork is laid.

III. SFA's Implementation Strategy

• The SFA has outlined a short-term strategy for implementing data-driven risk assessment in CIT:
  • Focus on large taxpayers and use existing TP business risk indicators.
  • Implement automatic risk analysis in transfer pricing.
  • Use questionnaires, preventive visits, and cooperative compliance to promote voluntary compliance.
• The annual supervision plan of the Special Financial Office (SFO) will be adapted to include comprehensive audits for all taxes, led by specialists in transfer pricing.

IV. Areas for Improvement

• The CIT audit selection model currently uses both comprehensive and partial audit data, which may distort results. Separate models should be developed for each type of audit.
• The model is based on corrected tax returns, which may not reflect real-world non-compliance. It should instead be trained on initial tax returns to better mimic audit selection practices.
• The risk rules should be evaluated against historical data to ensure their effectiveness and accuracy.

V. Next Steps and Priorities

1. Senior Management Prioritization

• Action: Prioritize high-value data analytics projects and allocate protected time for development.
• Deadline: Short term (within 12 months)
• Responsibility: Director General (DG) and department heads

2. Risk Assessment Enhancements

• Action: Develop business risk indicators for CIT, evaluate existing risk rules, and improve data workflows.
• Deadline: Short term (within 12 months)
• Responsibility: Analysis unit and relevant business owners

3. Compliance Risk Management

• Action: Reestablish the Compliance Risk Management Unit (CRMU) and ensure it supports the entire SFA.
• Deadline: Short term (within 12 months)
• Responsibility: DG in consultation with senior management

4. Automation and Anomaly Detection

• Action: Expand the use of anomaly detection in transfer pricing using CbCR data and other sources.
• Deadline: Short term (within 12 months)
• Responsibility: Analysis unit and TP unit

5. Staff Training and Development

• Action: Establish learning programs where data scientists train non-experts in analytics.
• Deadline: Short term (within 12 months)
• Responsibility: Analysis unit and heads of divisions
• Tools: Use KNIME® for low-barrier training and collaboration.

6. Staff Retention and Recruitment

• Action: Improve terms for data analysts to ensure retention and recruitment.
• Deadline: Short term (within 12 months)
• Responsibility: DG and HR

VI. Conclusion

The SFA has made significant progress in implementing the recommendations from the November 2022 mission. The speed of adoption and dedication of staff were highly praised. The next steps involve refining existing models, improving data workflows, and establishing a more structured compliance risk management system. The focus should remain on simple, effective tools before moving to more advanced predictive analytics.