4.13.6 Advanced Extensions

Basic Prerequisites: 04 · Aggregating Risk Types and Stress Testing & Scenario Analysis.

---

1. Intuition & Practical Objective

The fixes to the failure modes of 05 are structural, not parametric, and they live at the macroprudential level - the level of the system, not the firm. This page is the launchpad for two ideas:

1. Countercyclicality in policy. Because per-firm VaR/leverage rules are procyclical (§05), regulators add a macroprudential layer that builds buffers in the boom so they are available in the bust, and that is calibrated on credit/leverage growth (the Credit-to-GDP gap) rather than current volatility. This is the CCyB and the broader “leaning against the wind” toolkit.
2. Stress integration as the aggregation solution (Bellini). Since cross-type aggregation by copula is model-fragile (§04), Bellini’s practical alternative is to integrate risk types through a shared macro factor: one scenario-driver $Z$ (e.g. GDP, rates, spreads) simultaneously shocks the market book, the credit PDs, and liquidity, then all losses land in a single P&L. Integration by scenario sidesteps the untestable copula - you don’t have to assume dependence, you generate it from a common macro path.

The objective: see that the right aggregation is a model of a shared economy, not a statistical joiner of marginal distributions.

---

2. Mathematical Ground Truth & Derivations

Countercyclical capital buffer (BCBS 2010; the credit-gap rule). Let $g_t$ be the credit-to-GDP gap (credit ratio minus its trend). The CCyB add-on is

$${\text{buffer}}_t=0.3125\cdot \max (g_t-2,0)\text{(i.e. }0\%\text{ at a 2pp gap, }2.5\%\text{ at }10\text{pp, capped at }2.5\%\text{),}$$

so the buffer rises as credit outgrows the trend - before the bust - and releases as the gap shrinks, feeding capital back into the system during the downcycle. This is the anti-procyclical mirror of the VaR-target loop in §05: it removes the need to delever in the bust because capital was pre-positioned.

Macro-factor credit link (Vasicek / Bellini CLE). Let a single macro factor $Z\sim N(0,1)$ drive every borrower’s default probability through a logistic link

$$\text{PD}(Z)=\frac{1}{1+e^{-(\alpha +\beta Z)}}.$$

Borrowers are independent given $Z$, so the portfolio default count for $N$ loans is, conditional on $Z$,

$$L^{\text{credit}}=\frac{1}{N}\sum _{i=1}^N{1}_{U_i\le \text{PD}(Z)}.$$

Now add the bank’s market (trading) book as a function of the same $Z$, $L^{\text{market}}=c+bZ+\xi$. The integrated loss is

$$L^{\text{tot}}(Z)=w_m{L}^{\text{market}}(Z)+w_c{L}^{\text{credit}}(Z),$$

and the bank’s integrated expected shortfall is ${\text{ES}}_{\alpha }(L^{\text{tot}})$ over scenarios $Z\sim N(0,1)$ - a single number that commits the market and credit books to the same macro scenario. This is Bellini’s core move: aggregate by joint scenario, not by adding separately-estimated marginal capitals. The resulting integrated ES is different from (and in §3 below, below) the naive sum of the two standalone ES, because a single scenario can’t simultaneously trigger each book’s own idiosyncratic tail - and that “un-achievable” sum is exactly the subadditivity slack of §04.

---

3. Computational Implementation - Bellini-style integrated stress

We run a bank with a trading book and a loan book under a shared macro factor $Z$. Standalone = each book’s own ES estimated separately; integrated = ES of the sum under the same $Z$-scenario. Plus the CCyB credit-gap rule. Stdlib only.

Read the output. The macro factor does the coupling: lending PDs climb from 1.80% at the mean, through 0.17% in a boom, to 16.8% in a $+2$ macro stress. The two standalone ES stack to 14.37, but the integrated figure - market and credit forced onto the same macro scenario - is 14.13: the naive sum overstates by ~0.24 because a single bad scenario can’t simultaneously strain each book’s idiosyncratic versus its macro component. More important than the (modest) saving is what the method commits to: it replaces an untestable copula with a tested shared-economy model - you audit the macro driver $Z$ and the PD and market-loading coefficients ($\beta$, $b$), which are identifiable from data, instead of trusting an unobservable ${\lambda }_u$. The CCyB rows make the policy explicit: the buffer is zero while the gap is negative (bust) and builds to 1.9% at a +3% gap - capital accumulate in the boom, released in the bust, precisely anti-procyclical to the §05 loop.

---

4. Failure Modes & First-Principles Breakdowns

1. The macro factor is still a model. Integrated stress is only as good as the assumed $Z$ and loadings; a wrong factor (or a tail that $N(0,1)$ can’t reach) silently understates the joint tail. Scenario selection must itself be stress-tested (bridges Stress Testing).
2. CCyB is procyclical in its own right if mis-calibrated. A gap computed with a bad “trend” can release buffers at the wrong time; the 2.5% cap is a policy judgement, not a calibration.
3. Integrated ≠ correct. Bellini integration plausibly beats naive summation, but it does not resolve tail-dependence ambiguity - it pins, rather than estimates, the dependence via the macroeconomic channel, so model risk moves but does not disappear.
4. Regulatory gaming. Any published systemic measure (MES/SRISK/CoVaR) invites arbitrage once banks optimise against it - a Goodhart failure inherent in all three-measure toolkits.

---

5. References

• Bellini, Stress Testing and Risk Integration in Banks (2016)
• BCBS, Basel III: A Global Regulatory Framework for More Resilient Banks (2010), and the 2017 finalisation
• Schuermann, Stress Testing Banks, IJCB 10(2) (2014)
• Quagliariello (ed.), Stress-testing the Banking System (2009)
• McNeil, Frey & Embrechts, QRM (2015)

---

6. Connected Graph Bridges

• Back: 05 · Failure Modes · Index Hub
• Forward topic pages: Stress Testing & Scenario Analysis · Capital & RWA · Basel & Regulation
• Sibling mechanism: Margin & Funding Spirals (what “release the buffer” prevents) · Credit Risk & the Merton Model (the PD/$Z$ link)