Case Study 14.2: Cornerstone's Market Risk Transformation — From VaR to FRTB

The Situation

Organization: Cornerstone Financial Group (fictional composite multi-jurisdictional institution) Challenge: End-to-end FRTB implementation across Cornerstone's EU, UK, and US trading operations Timeline: 2022–2025 Regulatory backdrop: CRR3 (EU), PRA FRTB rules (UK), pending US implementation

Background

Cornerstone Financial Group operates trading desks across three jurisdictions: Frankfurt (EU regulated, CRR3 scope), London (UK regulated, PRA FRTB scope), and New York (US regulated, pending FRTB implementation). The trading book contains:

• Fixed income: G10 sovereign bonds, investment grade and high yield corporate bonds, structured credit
• FX: G10 and EM currency trading for clients and proprietary
• Rates derivatives: interest rate swaps, swaptions, inflation-linked
• Equity: index futures and delta-hedging for structured products
• Commodity: energy derivatives (Brent crude, European gas)

Total trading book market value: approximately €24 billion. Current market risk capital: €580 million.

The FRTB program at Cornerstone was a 3-year, €45 million investment — one of the most significant technology programs in the risk function's history.

The Data Infrastructure Challenge

The first FRTB challenge Cornerstone encountered was not about risk models — it was about data.

FRTB's Non-Modellable Risk Factor (NMRF) rules require institutions to demonstrate that each risk factor has at least 24 price observations per year, with no 90-day gap in observations, using data from a "real" transaction or committed quote.

Cornerstone's risk technology team conducted an NMRF assessment for all 1,847 risk factors in its risk model. The findings:

• 1,204 risk factors: modellable (sufficient observations, no gaps)
• 412 risk factors: marginally modellable (sufficient overall count, but occasional gaps requiring careful documentation)
• 231 risk factors: definitively non-modellable — either insufficient observations or significant gaps

The 231 non-modellable factors included: - Corporate bond CDS spreads for 45 smaller issuers (insufficient CDS trading) - EM local currency yield curve points for maturities beyond 10 years in less liquid markets - Volatility surface points for out-of-the-money options in less liquid currencies - Gas basis spreads (difference between UK NBP and Netherlands TTF gas prices) at specific forward maturities

The business decision: For the definitively non-modellable factors, Cornerstone had two options: 1. Accept NMRF treatment and pay the stress scenario capital charge 2. Reduce or eliminate positions in these risk factors to reduce NMRF capital

Cornerstone's trading management reviewed the NMRF list with the business. Decision: reduce positions in the 45 smaller corporate issuers (the capital cost exceeded the revenue potential); maintain EM and energy positions and absorb NMRF capital charges (these positions were core to client service mandates).

NMRF capital charge: €42 million annually — a new permanent capital cost.

The PLA Testing Program

Cornerstone ran PLA tests across 14 trading desks for 12 months prior to FRTB go-live. Results:

Analysis of failures:

High Yield Credit: Similar to the IG credit issue — issuer-specific spread dynamics not captured by sector-level indices. But the HY market is more heterogeneous than IG; fixing this requires more issuer-specific factors and data.

Structured Credit (CLO): CLO tranche valuations depend on complex waterfall mechanics, prepayment speeds, and collateral pool dynamics. The risk model used simplified credit spread risk factors — a fundamental model adequacy issue. CLOs remain on SA.

European Gas: The gas market experienced extraordinary volatility in 2021–2022 following the Russia-Ukraine conflict. The risk model's gas risk factors (TTF spot, TTF forward curve) did not capture the full idiosyncrasy of the gas basis — regional price differences driven by physical supply/demand dynamics that cannot be modeled from financial price data alone.

Commodity Physical: Physical commodity positions (physical gas storage contracts) are difficult to hedge mathematically — their risk depends on physical delivery logistics that don't map to financial risk factors.

Technology Implementation

FRTB required significant technology infrastructure upgrades at Cornerstone:

New data vendor agreements: Cornerstone subscribed to Bloomberg's FRTB data service — providing daily market price observations from real transactions across 12,000+ instruments, enabling NMRF modellability assessments. Annual cost: €2.8 million.

Risk engine upgrade: Cornerstone's risk engine (Murex) was upgraded to Murex 3.1, which included FRTB-native support: sensitivity-based SA calculation, ES with liquidity horizon scaling, PLA test computation, NMRF identification and capital calculation.

P&L data infrastructure: PLA testing requires daily HPL (hypothetical P&L) calculation — a clean P&L from risk factor movements only, stripped of intraday trading effects, new trades, and non-risk adjustments. This required new daily feeds from the finance system and a new HPL calculation process in the risk engine.

COREP C 31.00/C 32.00 reporting: The new COREP templates for FRTB SA (C 31.00) and IMA (C 32.00) required significant configuration in Cornerstone's reporting platform. The C 31.00 template alone had 847 data cells across risk classes and instrument types.

Capital Impact

Cornerstone's final FRTB capital impact:

Net capital increase: €114 million (+19.7%). The IMA desks actually showed lower capital under FRTB — the combination of ES (more tail-sensitive but at 97.5% vs 99% VaR) and liquidity horizon weighting produced lower aggregate capital for the highly liquid G10 desks. The SA desks drove the overall increase.

The net €114M capital increase was funded by retained earnings — Cornerstone had anticipated and planned for this in its capital planning.

Lessons Learned

Cornerstone's Head of Market Risk, Dr. Yuki Tanaka, identified three key lessons from the 3-year program:

Lesson 1: Data quality is the foundation. "We spent six months just on NMRF modellability assessment — cataloguing 1,847 risk factors, validating observation counts, identifying gaps. That foundation work was more labor-intensive than anything on the model side."

Lesson 2: PLA test failure reveals model inadequacy that was there all along. "The structured credit desk's PLA failure was not FRTB's fault — it revealed that our CLO risk model had never been adequate. We just couldn't see it because VaR backtesting at the aggregate level masked desk-level model weaknesses. FRTB's desk-level PLA test is a genuine improvement in model governance."

Lesson 3: SA is not a simple fallback. "We thought of SA as a penalty — move to SA if you fail IMA. But FRTB's SA is itself complex. Getting the sensitivity-based SA calculation right for an options book or a structured credit portfolio requires real quantitative effort. 'We'll just use SA' is not as simple as it sounds."

Discussion Questions

1. Cornerstone identified 231 non-modellable risk factors — 12.5% of all risk factors. The NMRF capital charge was €42 million. If Cornerstone could reduce this to 5% through better data sourcing, what would be the approximate capital saving? What investment in data vendor relationships would be justified?

2. The Structured Credit (CLO) desk failed the PLA test decisively (Spearman correlation: 0.44). Dr. Tanaka commented that the PLA failure "revealed model inadequacy that was there all along." What governance mechanism should have caught this model inadequacy before FRTB forced the issue? How does this relate to SR 11-7 model validation requirements?

3. Cornerstone's overall FRTB capital increase was 19.7% — €114 million. The IMA desks actually saw lower capital post-FRTB while SA desks drove higher capital. Is this the expected industry-wide pattern? What types of institutions would see the largest capital increases from FRTB, and which would see relatively modest impacts?

4. The European Gas desk's PLA failure was partly attributable to the Russia-Ukraine conflict causing extraordinary market conditions. If the model had been tested against pre-2022 data, it might have passed. Should regulators allow institutions to exclude extraordinary periods from PLA testing? What are the arguments for and against?

5. Cornerstone's FRTB implementation cost €45 million over three years. The ongoing annual cost includes €2.8M in data vendor fees plus significant internal resources. Is this cost proportionate to the regulatory benefit of FRTB — i.e., does the improved market risk capture justify the compliance cost? Consider both the institutional and systemic perspectives.