Introduction

Operational risk management involves the assessment and mitigation of potential losses arising from internal processes, people, systems, or external events. Accurate estimation of loss frequency and severity distributions is paramount for effective risk management. However, various data issues can introduce inaccuracies and biases, impacting the reliability of these estimations.

In operational risk analysis, data serves as the foundation for risk assessment and decision-making. Yet, the quality of data can significantly influence the outcomes. In this chapter, we explore the common data issues that can lead to inaccuracies and biases in the estimation of loss frequency and severity distributions.

Data Issues and Their Impacts

1) Underreporting and Reporting Bias: Underreporting occurs when incidents are not recorded due to lack of awareness or reluctance to report. This skews the loss frequency distribution, as only a subset of incidents is captured. Reporting bias arises when certain types of incidents are more likely to be reported, distorting the distribution’s shape.

• Example: An employee might avoid reporting a minor operational error, leading to an underestimation of the true loss frequency.

2) Lack of Historical Data: Insufficient historical data hampers the estimation of rare events. Without a robust dataset, it’s challenging to accurately predict the occurrence of infrequent but high-impact events.

• Example: A new financial product might have limited historical data, making it difficult to gauge potential losses accurately.

3) Changes in Reporting Criteria: Alterations in reporting criteria or categorization can introduce inconsistencies in the data over time. This affects comparability and trend analysis.

• Example: If the criteria for classifying incidents change, historical data might not be directly comparable to new data.

Mitigation Strategies

• Normalization Techniques: Employ techniques like Bayesian estimation or statistical models to account for underreporting and reporting bias. These methods adjust the data to reflect a more accurate representation of incidents.

• Scenario Analysis: When historical data is lacking, scenario analysis becomes crucial. By simulating potential scenarios and their impacts, you can estimate potential losses even without extensive historical data.