This is a detailed research piece. If you find value in institutional-quality hedge fund analysis, support this work on Patreon.
Citadel Wellington returned 15.1% in 2024 while consensus steepener trades imploded. The edge: replacing principal component analysis with explicit tenor-level risk reporting before policy-driven curve dislocations broke historical covariance structures.
The Trade Setup
Early 2024: hedge funds priced six Fed rate cuts by year-end. Consensus position: 2s30s bull steepeners betting the short end would rally harder than the long end.
Trade mechanics:
Long 2-year Treasury futures/swaps
Short 30-year Treasury futures/swaps
DV01-neutral hedge ratio (equal dollar duration)
P&L driver: 2s30s spread widening when short-end yields fall faster than long-end
The 2s30s spread traded around -50bp mid-January. By mid-February: -83bp — a 33bp adverse move in 30 days. The long 2Y position lost value as 2-year yields rose instead of falling (rate cut expectations collapsed), while the short 30Y position also bled as long-end yields remained elevated, creating losses on both legs of the trade.
What broke: Sticky inflation (January CPI 3.1% vs. 2% target) and strong employment delayed cuts. ECB and Bank of Canada didn’t cut until early June; the Fed waited until September 18, 2024. The consensus trade burned.
Why PCA Failed
Principal component analysis decomposes yield curves into orthogonal factors:
PC1 (level): 80–85% variance, parallel shifts
PC2 (slope): 10–12% variance, steepening/flattening
PC3 (curvature): 3–5% variance, butterfly moves
PCA optimizes for historical variance explanation. First principal component typically loads heaviest on intermediate maturities (5–10Y), lighter on wings (2Y, 30Y). This assumes historical factor relationships persist.
Q1 2024 regime break: 2-year yields repriced +40–50bp as rate cut expectations collapsed, devastating the long 2Y leg. 30-year yields stayed anchored by term premium dynamics, preventing profits on the short 30Y leg. This wasn’t a PC1 parallel shift — it was idiosyncratic short-end repricing that historical covariance couldn’t capture.
Failure mechanism:
Historical PC1 loadings assumed 5Y yields would move ~1.5x more than 30Y
Actual dynamics: 2Y moved aggressively, 30Y range-bound
The DV01-neutral hedge embedded an implicit directional exposure to whichever tenor decoupled from historical patterns
When central bank expectations shift violently, PCA’s historical eigenvectors become invalid. The model explains past variance, not structural breaks in policy expectations.
Market Observables Framework
Chief Risk Officer Joanna Welsh implemented a switch from PCA to direct tenor-level risk reporting (2Y, 5Y, 10Y, 30Y yields).
Operational distinction:
PCA: “Portfolio has +$2mm exposure to PC1, -$800k to PC2”
Market observables: “+$1.5mm DV01 at 2Y, -$600k at 10Y, flat at 30Y”
Welsh emphasized avoiding getting “too academic” in favor of “a more pragmatic way of thinking about curve risk.”
Framework advantages:
Tenor-specific transparency: Portfolio managers see exactly which maturities drive P&L. No abstract statistical factors requiring interpretation.
Regime-adaptive hedging: Zero assumptions about historical eigenvector loadings. When 2Y yields decouple, exposures are immediately visible without recalculating factor sensitivities.
Operational clarity: Risk limits become directly actionable. “Reduce 2Y exposure by $500mm” executes immediately versus “decrease PC2 loading by 15%” requiring eigenvalue recalculation and PM interpretation.
Welsh noted this helped navigate 2024’s “erratic rates environment” by isolating “a very specific set of risk factors, rather than rely on generic statements like ‘curve risk’ when some people were in the short end and some people were in the long end.”
The approach requires monitoring more data points — 8–10 tenors versus 3 principal components. But when policy expectations drive non-parallel moves, precision beats parsimony.
Performance Validation
Wellington gained ~1% in August 2024 when the Bank of Japan’s surprise rate hike triggered yen carry unwinds and VIX spiked. Full-year 2024: 15.1% versus industry average 7.4% — a 770bp outperformance with all five strategies positive (as Risk.net reported).
The tenor-based framework allowed PMs to:
Isolate specific curve exposures before macro surprises materialized
Adjust positions in real-time as Fed expectations shifted
Avoid factor hedges that broke under non-parallel repricing
Welsh attributed outperformance to what Citadel “didn’t do” — steering clear of trades that became “a banana skin for people.”
The Quantitative Lesson
Dimensional reduction fails when covariance structures destabilize. PCA’s elegance — collapsing 10+ tenors into 3 factors — optimizes for parsimony and historical variance explanation. But 2024 demonstrated the cost when policy expectations shift violently.
The critical failure mode: PCA assumes PC1 (parallel shifts) dominates returns. When 2Y yields reprice independently due to Fed expectations while 30Y yields remain anchored, historical factor loadings become worthless. The “neutral” position carries hidden directional exposure to whichever tenor deviates from historical patterns.
Market observables impose operational overhead — tracking each tenor’s DV01 versus a single PC1 exposure. But in policy-driven volatility where correlations break down, explicit tenor exposures allow real-time risk adjustment without waiting for covariance recalibration.
Citadel’s 770bp outperformance validated Welsh’s pragmatic approach: sophisticated statistical frameworks must subordinate to operational clarity when regime changes render historical relationships invalid. The hedge fund’s success wasn’t predicting Fed policy — it was maintaining transparent tenor-level visibility when consensus PCA models broke.
For quantitative risk managers: PCA works until it doesn’t. When macro policy uncertainty dominates, reverting to first principles — tracking actual market observables — beats optimizing for statistical elegance derived from stable regimes.
Sources
Performance & Strategy:
Citadel Wellington 15.1% full-year 2024: CNBC (Jan 2, 2025)
Wellington 11.2% through November, industry 7.4%: Risk.net — Hedge Fund of the Year: Citadel (Nov 26, 2024)
August 2024 +1% performance: Reuters (Sep 4, 2024)
Trade Mechanics & Market Data:
4. Steepener trade details: Risk.net — How Steepener Trades Burned Hedge Funds (Jun 26, 2024)
5. January 2024 CPI: Bureau of Labor Statistics (Feb 13, 2024)
6. Fed rate cut timeline: Federal Reserve FOMC Statement (Sep 18, 2024)
7. ECB/Canada rate cuts: Reuters — ECB, Canada Cut Rates (Jun 6, 2024)
Technical Framework:
8. Joanna Welsh quotes (all PCA vs. market observables commentary): Risk.net — Hedge Fund of the Year: Citadel (Nov 26, 2024)
9. Litterman & Scheinkman PCA framework: Common Factors Affecting Bond Returns, Journal of Fixed Income (1991)
10. PCA variance decomposition: MDPI — Principal Component Analysis in Yield Curve Scenarios (May 2022)
11. DV01 hedging methodology: CME Group — Treasury Futures Hedging
August 2024 Volatility:
12. Bank of Japan rate hike: BOJ Policy Statement (Jul 31, 2024)
13. VIX spike and carry unwind: BIS — The August 2024 Volatility Spike (Aug 2024)
All claims verified against primary sources. Direct quotes from Joanna Welsh sourced from Risk.net interviews. Performance figures confirmed across multiple outlets. Technical PCA framework based on Litterman-Scheinkman (1991) and contemporary academic literature.
📊 Support this research: https://www.patreon.com/c/NavnoorBawa
Cover photograph: Ken Lund from Reno, Nevada, USA, CC BY-SA 2.0, via Wikimedia Commons.