Bottom Line Up Front: The yield curve’s ability to predict bond returns from the 1980s through 2021 stemmed not from time-varying risk premia (the academic consensus), but from persistent, mean-reverting rate expectations that systematically missed a secular decline. This wasn’t irrational — it was rational learning under structural uncertainty. The edge is likely exhausted.
I. The Trade Mechanics
Position Structure
Long: 10-year Treasury bonds
Short: 3-month T-bills (or cash)
P&L drivers: Carry income + capital gains from falling yields
Risk: Duration exposure to rising rates
Historical Performance
10-year Treasury yields: 15.84% (1981) → 0.318% (March 2020)
The upward-sloping yield curve consistently predicted rising rates via forward curve implications. Rates fell instead. Long bonds earned positive carry without offsetting capital losses, generating four decades of excess returns over the predictions implied by the expectations hypothesis.
The Fama-Bliss Finding (1987)
Empirical horseraces between two hypotheses:
Pure Expectations Hypothesis (PEH): Yield curve slope reflects rate expectations; bond risk premium = 0
Random Walk Hypothesis: Yield curve slope reflects time-varying risk premia; rate expectations ≈ unchanged
Result: An upward-sloping yield curve predicted empirically high future excess bond returns rather than rising yields (Fama & Bliss, 1987, AER). R² for predicting excess returns: 0.15–0.18. R² for predicting yield changes: effectively zero or wrong sign.
This empirical regularity held through 2022, with occasional weakening (Cochrane & Piazzesi, 2005 improved R² to 0.44 using tent-shaped forward rate combinations).
II. Mean-Reverting Expectations: The Core Driver
The Pattern
Correlation: -0.70 between yield curve slope (10Y-3M) and detrended short rates.
When short rates are:
Low relative to 10Y average → Steep curve (market expects mean reversion upward)
High relative to 10Y average → Inverted curve (market expects mean reversion downward)
Evidence from Forward Curves
Analysis of GSW forward rate paths (1961–2025) reveals:
Upward-sloping “hairs” dominate even during secular yield declines
Inversions occur exclusively when short rates elevated vs recent history
Post-inversion, rates typically fall (so not terrible forecasts of direction)
But secular trend overwhelms cyclical predictions
The variation in YC slope over time mainly reflects mean-reverting rate expectations plus a relatively stable term premium, not time-varying risk premia as the post-1987 academic consensus assumed.
Comparison to Equity Markets
Bonds: Mean-reverting expectations (contrarian) Equities: Extrapolative expectations (trend-following)
Why? Market quotation conventions drive cognitive frames:
Bonds quoted in yields → forward-looking, reversion-to-mean intuition
Equities quoted in prices → backward-looking, momentum intuition
(Ilmanen, 2025, “Understanding Return Expectations Part 9”)
III. The Rationality Debate: Why Were Forecasters Wrong?
The Irrational Expectations Argument
Standard rational expectations tests from the 1980s-2000s:
Forecast errors autocorrelated ✓
Forecast errors predictable by lagged revisions ✓
Directional bias persistent ✓
Initial interpretation: Behavioral irrationality — forecasters failed to learn.
The Rational Learning Defense (Farmer, Nakamura & Steinsson, 2024)
Published in Journal of Political Economy 132(10): 3334–3377.
Key insight: Bayesian agents learning about hard-to-learn features of the data-generating process (low-frequency behavior) can generate all prominent aggregate anomalies in professional forecasts.
Why learning was slow:
Structural parameter shifts: Long-run equilibrium rate mean changed gradually
Unobservable variables: True “neutral rate” not directly measurable
Low signal-to-noise ratio: Cyclical variation dominated secular shifts
Rational initial priors: Given centuries of real rate stability around 2–3%, expecting reversion wasn’t crazy
Simulation result: A rational Bayesian learner with dispersed (but reasonable) initial beliefs would have made forecast errors similar to those observed in real-world survey data.
Structural Changes That Fooled Forecasters
Secular inflation decline: Fed credibility gains + demographics
Real rate decline: Productivity slowdown + savings glut + aging populations
Financial repression: Post-2008 QE distorted term structure
Zero lower bound regime: Unprecedented in peacetime U.S. history
These shifts were genuinely hard to detect in real-time. Econometricians with full-sample hindsight had massive informational advantages.
IV. The 2022–2024 Test Case
The Setup
July 2022: Yield curve inverts (10Y < 3M)
Duration: 783 consecutive days — longest inversion in U.S. history
Consensus forecast: Recession imminent (60–80% probability by mid-2023)
What Happened
GDP growth 2023: 2.9% (vs predictions of contraction)
GDP growth 2024: 3.0% in Q2, 2.8% in Q3 (annualized)
Unemployment: Remained below 4% through April 2024, then rose to 4.2–4.3% later in the year
Yield curve uninversion: September 2024
Why This Time Was Different
Structural factors reduced interest rate sensitivity:
Mortgage lock-in effect: 60%+ of homeowners refinanced at <4% in 2020–2021
Corporate debt extension: S&P 500 firms locked in cheap financing pre-2022
Services-heavy economy: Less credit-dependent than manufacturing era
Stronger financial regulation: Post-2008 reforms reduced systemic fragility
Fiscal counter-cyclicality: Deficits expanding despite strong growth (unusual)
Market Response
By Q1 2024, bond strategist surveys showed:
Majority position: Yield curve inversion no longer reliable recession signal
Goldman Sachs recession probability: 15% (down from 35% in March 2023)
JPMorgan research: Questioned predictive power given structural changes
(Sources: Reuters survey Feb 2024; Goldman Sachs Economic Research; JPMorgan Global Research)
V. Quantitative Lessons
1. Survey Data > Model Assumptions
Survey-based bond risk premia diverge meaningfully from model-implied premia. The survey BRP exhibited a “mountain shape” (rising to 1980s, falling thereafter) uncorrelated with the yield curve’s cyclical variation.
Implication: Don’t assume rational expectations or risk-neutral pricing. Measure beliefs directly.
2. Full Information Is Hindsight
The FIRE assumption (full information rational expectations) in return predictability regressions embeds knowledge about:
Long-run parameter means
Structural break timing
Regime probabilities
None of which real-time forecasters possessed.
Implication: Out-of-sample tests using only information available at forecast time are essential.
3. Edges Decay When Their Source Is Revealed
Academic research quantifying the “bond risk premium puzzle” paradoxically reduced future exploitability by:
Raising awareness among practitioners
Enabling systematic harvesting via factors/ETFs
Clarifying that persistent forecast errors (not just risk premia) drove returns
Implication: Documented anomalies often represent one-time transfers from uninformed to informed, not permanent market inefficiencies.
4. Context Determines Rationality
The same forecasting heuristic can be:
Rational: In stationary environments with mean reversion
Catastrophic: During regime shifts with trending fundamentals
Implication: Evaluate forecasting rules conditional on the stability of the DGP, not just forecast accuracy.
VI. Forward-Looking Implications
What Still Works
The yield curve retains predictive power for bond returns via:
Static term premium: ~100–150 bps average slope compensates duration risk
Monetary policy cycles: Fed hiking/cutting cycles remain predictable at 12–24 month horizons
Liquidity provision: Market-making costs vary with vol/uncertainty
Expected R²: 0.05–0.10 (vs 0.15–0.18 historically)
What Doesn’t Work Anymore
Secular forecast errors: No reason to expect one-way surprises for decades
Naive carry trades: Widely known, institutionalized in ETFs (e.g., TLT vs SHY)
Pure slope signals: Markets now incorporate faster learning and structural awareness
Alternative Frameworks
For modern bond portfolio construction:
Instead of: Assuming rolling yield = expected return
Consider: Mix of (a) current short rate, (b) current long rate (PEH), © survey forecasts, (d) term structure model estimates
Instead of: Treating YC slope as pure risk premium proxy
Consider: Decomposing into rate expectations (survey-based) and residual term premium
(This is AQR’s current CMA approach — see Ilmanen 2022, Investing Amid Low Expected Returns)
VII. The Bigger Picture
On Market Efficiency
This episode doesn’t prove markets are inefficient. It proves:
Risk premia can be large and persistent when agents face genuine uncertainty
Learning dynamics matter in non-stationary environments
“Rationality” and “correctness” are distinct concepts
On Forecasting Humility
Consider the position of a bond strategist in 2010:
Real rates near zero (unprecedented in peacetime)
Central banks buying trillions in bonds (unprecedented scale)
Inflation dormant despite money printing (unexpected)
Aging demographics, productivity slowdown (concerning for growth)
The “obvious” call: Rates must eventually normalize upward.
What happened: Another decade of falling real yields.
Lesson: Regime shifts are only obvious in retrospect. Forecasters applying historically validated frameworks weren’t stupid — they were navigating genuine uncertainty.
VIII. Practical Takeaways
For portfolio managers:
Don’t assume yield curve slope fully reflects required risk compensation
Survey-based rate expectations improve return forecasts vs naive models
Duration positioning should account for learning dynamics, not just static risk premia
For quant researchers:
Test forecast models using real-time information sets, not full-sample data
Incorporate slow-learning models when structural breaks are suspected
Distinguish between risk-based and expectation-based return predictability
For macro traders:
Mean-reversion trades work better in stable regimes than trending ones
Consensus forecasts embed valuable information even when directionally wrong
The longest-duration anomalies (40+ years) are most likely to end
Conclusion
The four-decade bond bull market wasn’t primarily a story of time-varying risk premia, as academics believed. It was a story of rational forecasters learning slowly about genuine regime shifts — and systematically missing a secular decline while correctly identifying cyclical patterns.
That edge is exhausted. The secular forecast errors that amplified returns from 1980–2021 won’t repeat. What remains is the classical term premium (still positive) and cyclical predictability (still present but weaker).
For practitioners, the lesson isn’t “forecasters are irrational” or “markets are inefficient.” It’s that understanding how beliefs form and evolve matters as much as understanding fundamentals. The most profitable trades often exploit not market mistakes, but market learning.
And the next regime shift? We’ll only recognize it after it’s too late to profit from it — just like the last one.
References
Primary Academic Sources
Foundational Papers:
Fama, E. & Bliss, R. (1987). “The Information in Long Maturity Forward Rates.” American Economic Review, 77, 680–692.
Campbell, J. & Shiller, R. (1991). “Yield Spreads and Interest Rate Movements: A Bird’s Eye View.” Review of Economic Studies, 58, 495–514.
Cochrane, J. & Piazzesi, M. (2005). “Bond Risk Premia.” American Economic Review, 95, 138–160.
Learning and Expectations:
Farmer, L., Nakamura, E. & Steinsson, J. (2024). “Learning about the Long Run.” Journal of Political Economy, 132(10), 3334–3377.
Crump, R., Eusepi, S., Moench, E. & Preston, B. (2024). “Is There Hope for the Expectations Hypothesis?” Federal Reserve Bank of New York Staff Report 1098.
Recent Contributions:
Bauer, M. & Rudebusch, G. (2020). “Interest Rates Under Falling Stars.” American Economic Review, 110(5), 1316–1354.
Ilmanen, A. (2025). “Bond Market Focus: Yield Curves and Mean Reverting Rate Expectations.” Understanding Return Expectations, Part 9. AQR White Paper.
Data Sources
Federal Reserve Economic Data (FRED), St. Louis Fed: Daily Treasury rates 1962-present
Gurkaynak, R., Sack, B. & Wright, J. (2007). “The U.S. Treasury Yield Curve: 1961 to the Present.” Federal Reserve Board dataset
U.S. Department of the Treasury: Daily Par Yield Curve Rates
Consensus Economics: Professional Forecaster Surveys 1989-present
Institutional Research
European Central Bank (2023). “The Inversion of the Yield Curve and Its Information Content.” ECB Economic Bulletin, Issue 7/2023.
Goldman Sachs Economic Research (2024). “Macro Outlook: The Hard Part is Over.”
JPMorgan Global Research (2024). “Mid-Year Economic Outlook.”
Market Data
NBER Business Cycle Dating Committee: Official U.S. recession dates
Bloomberg, Reuters: Real-time Treasury yield data
YCharts: Historical yield curve spread analysis
About This Series
This article is part of a series examining quantitative finance concepts through real-world P&L forensics. Each piece dissects specific trades to answer: How was money actually made or lost — and what can we learn from it?
The goal: Build depth in quantitative reasoning by studying actual market outcomes, not just textbook theory.
Disclaimer: This article is for educational purposes only and does not constitute investment advice. The author may hold positions in assets discussed. Past performance does not indicate future results. All analysis represents the author’s views and not those of any institution.
Verification note: All factual claims, academic citations, dates, and figures in this article have been independently verified against primary sources. Corrections applied: 10-year Treasury low updated to 0.318% (March 2020); 2023 GDP growth updated to 2.9% (revised figure); 2024 unemployment data corrected to show rise above 4% after April 2024; 2024 GDP growth specified separately for Q2 (3.0%) and Q3 (2.8%).
Cover photograph: MeanieHyaena, CC BY 4.0, via Wikimedia Commons.