CTA Treasury Reversal

Signal as of April 2026: Cheap / Undervalued (+0.20pp vs. fair value) | Z-Score: +0.58σ | R² = 0.877 | Updated June 05, 2026

Model Methodology

This model estimates the 10-year Treasury yield as a linear function of two Fed policy stance variables, calibrated by OLS regression over the full historical sample.

DGS10 = α + β₁·DGS2 + β₂·PolicySpread + β₃·TaylorGap

Regressor	Definition	Coeff.	Interpretation
α	Intercept	+1.969	Structural level offset
DGS2	2-Year Treasury yield	+0.656	Market pricing of short-rate path (level anchor)
PolicySpread	DGS2 − FEDFUNDS	+0.536	Hike expectations (+) / cut expectations (−)
TaylorGap	FEDFUNDS − Taylor Rule rate	+0.148	Hawkish vs. rule (+) / dovish vs. rule (−)

Taylor Rule: r = 2.5 + π + 0.5·(π − 2.0) − 1.0·(UNRATE − NROU), where π = Core PCE YoY and NROU = CBO natural rate of unemployment.

Model fit: R² = 0.877 | In-sample RMSE = 0.53pp | Sample: Dec 1995 – Apr 2026

Residual = Actual − Fair Value. Mean-zero by OLS construction.
Positive → yield above model → market cheap/undervalued | Negative → yield below model → market rich/overvalued

Reversal Signal Methodology (Z-Score Panel)

The raw residual identifies direction of mispricing but not timing. The Z-Score panel normalizes the residual against its own trailing 36-month history to measure how statistically extreme the current deviation is, and flags when an extreme episode ends.

Z = (Residual − 36m rolling mean) / 36m rolling std

Z > +1σ: yield is unusually cheap vs. policy fundamentals (green shaded zone)
Z < −1σ: yield is unusually rich (red shaded zone)
▼ Sell reversal: z crosses back below +1σ from above — cheap episode confirmed over, yield likely reverting down
▲ Buy reversal: z crosses back above −1σ from below — rich episode confirmed over, yield likely reverting up

Background shading in the top panel (Row 1) highlights the same extreme-z periods so reversal markers can be visually matched against actual yield movements. The ±1.5σ dotted lines flag historically rare extremes (<15% of observations). Over the 1990–present sample, this approach generated approximately 13 sell and 15 buy reversal confirmations.

Sources: Federal Reserve H.15 (DGS10, DGS2, FEDFUNDS), Bureau of Economic Analysis (PCEPILFE), Bureau of Labor Statistics (UNRATE), Congressional Budget Office via FRED (NROU). All data via FRED.