Outlier detection is the process of identifying data points that differ drastically from so-called “normal instances” in a given data set. Outlier detection delivers critical information across many different domains in finance, such as financial reporting, fraud detection, and portfolio risk management. Recognizing anomaly patterns not only helps us detect potential errors at early stages, but also enables us to uncover potential insights on the underlying data.
This paper by Milliman’s Hyunsu Kim and Michael Leitschkis considers a technique called the Isolation Forest, which overcomes the shortcomings of classic anomaly detection algorithms. The technique has already been successfully applied across various disciplines, ranging from astrophysics to private wealth management. In their paper, the authors apply the Isolation Forest approach to the life insurance market and discuss an algorithm for an efficient automated detection on outliers in both small and large data sets.