In portfolio risk management a lot of focus and attention is (rightly) put on models and methodologies used to compute ex-ante risk measures. And in the context of a multi-asset class universe which is vast by nature, perfect data (market data, terms and conditions provided by the user) and bug-free algorithms are not always possible.
Therefore, one of the key challenges within the portfolio risk management space is to ensure that any risk analytic produced is sound and reliable
In this whitepaper, we:
- take a deeper dive into two accepted tree-based techniques to detect outliers: isolation forest (IF) and random cut forest (RRCF)
- show how efficient these methods can be to detect PL vectors that may need further investigation
- explain how this process can be automated for risk managers to flag impacted portfolios