4.10 Operational Risk

Basic Prerequisites: Probability & Measure Theory and Statistics & Inference. (these are the folder-level prerequisites for pages 02–06; page 01 states its own, smaller, entry requirements)

1. Intuition & Practical Objective

Operational risk is the risk of loss from failed internal processes, people, and systems, or from external events - the residual bucket after market risk (prices move) and credit risk (counterparties default). A rogue trader, a failed settlement engine, a hacked network, a mistaken model input, a regulator’s fine for money-laundering controls: all of these are operational losses. They are not explained by any single traded factor, which is why they cannot be hedged the way a stock or a bond can - and why the whole discipline is really about measuring a tail we cannot trade away and pricing the capital needed to survive it.

This folder is the operational-risk topic-folder for the Kwant-Atlas build. It is a hub: it gives you the fast formula lookup below, and routes you to six sub-pages that walk from raw intuition through the loss-event taxonomy, frequency–severity modelling, the aggregate-loss / Loss Distribution Approach (LDA), the failure modes, and the Basel capital + risk-transfer extensions.

The one-sentence essence. “Operational losses arrive as a random number of randomly sized hits per year - model the annual count as Poisson, model the per-event size as a heavy-tailed severity law, convolve them into an aggregate annual-loss distribution, and read off the 99.9%-one-year capital from its tail - the Loss Distribution Approach.”

2. Mathematical Ground Truth & Derivations

Quick-Reference Lookup. All formulas below are cross-checked against the BCBS Basel II AMA section (2006, ¶664–683), the BCBS Revised SMA for operational risk (2016, d305) and the Basel III: Finalising post-crisis reforms (2017, d424), plus McNeil–Frey–Embrechts (2015) and Panjer (2006). The numbers in the check column were re-executed and reproduced exactly from the verified corpus (see §3).

Notation: $N$ = number of loss events in one year; $X_i$ = size (severity) of the $i$-th event; $S={\sum }_{i=1}^N{X}_i$ = aggregate annual loss; $\lambda =\mathbb{E}[N]$ = expected frequency; $F_S$ = CDF of aggregate loss; $\alpha$ = confidence level (99.9%).

Critical caveat - the tail dominates. For equal expected severity, a Pareto tail ($\xi =1.5$) gives ${\text{VaR}}_{99.9}/\text{EL}\approx 13$ versus $\approx 2.2$ for a lognormal tail. Operational-risk capital is a tail number; misspecifying the severity tail swamps every other modelling choice.

3. Computational Implementation - the LDA engine

This runs on the standard library only (random), simulating the compound Poisson process and reading operational VaR from the empirical aggregate distribution. It reproduces the verified numbers above.

4. Failure Modes & First-Principles Breakdowns

Hub signposts - the folder’s failure-mode analysis lives in 05 · Failure Modes & Practice. In one line each:

1. Data scarcity & the unobservable tail - the 99.9% loss is, by definition, rarer than almost any bank’s loss history, so the number you care most about is the one you can least measure; even a 5,000-year sample estimates a heavy-tail VaR with ~26% standard error.
2. Severity misestimation - a one-parameter slip in the tail index changes capital by multiples; fitting the body of the data tells you almost nothing about the tail that drives the answer.
3. Frequency–severity confusion & dependence - Poisson assumes independent events, but real losses cluster and co-move in stress, breaking the clean convolution that makes LDA tractable.
4. The measurement-and-incentive loop - capital computed from loss history rewards not reporting losses, so the data feeding the model is endogenously corrupted.

5. References

• BCBS, Basel II: International Convergence of Capital Measurement and Capital Standards (2006)
• BCBS, Basel III: Finalising post-crisis reforms (2017, d424)
• Panjer, Harry H., Operational Risk: Modeling Analytics (2006, Wiley)
• Shevchenko, Pavel V., Modelling Operational Risk Using Bayesian Inference (2011, Springer)
• McNeil, Frey & Embrechts, Quantitative Risk Management (2015, Princeton)
• Embrechts, Klüppelberg & Mikosch, Modelling Extremal Events for Insurance and Finance (1997, Springer)

6. Connected Graph Bridges

• Foundational base: Probability & Measure Theory · Statistics & Inference
• Sibling topic: Extreme Value Theory & Fat Tails (the tail machinery severity needs)
• Regulatory home: Basel & Regulation
• Sub-pages (in-folder): 01 From Zero · 02 Loss Event Types · 03 Frequency–Severity · 04 Aggregate Loss & LDA · 05 Failure Modes · 06 Advanced Extensions

Beginner: start at 01 · Practitioner: start at 05