2026 RELATIONSHIPS TO WATCH IN 2027

Sample: 43 completed weekly signals, 2026-01-02 to 2026-10-23; outcomes through 2026-10-30.
These are in-sample rankings from the frozen paper data, not predictions of which rules will work in 2027.
The full 2026 year is not yet available. Both directions of all 27 existing rules were examined.
The five shown passed the disclosed activity/state filters; thresholds were not retuned.

Interpretation: these are descriptive associations, not instructions to add or reduce duration.
The retained position labels are historical scoring conventions. The frozen model does not change.
The timing score subtracts the hindsight benefit of each rule's average directional tilt.
Scores below are gross yield-change points over this sample, not portfolio returns or tradeable profits.

Rank | Indicator | Gross bp | Timing bp | Positive timing years, 2023-25 | State changes
1 | Natural gas | +131.0 | +124.0 | 1/3 | 15
    Condition: Natural gas 4-week return > 0 (one-week-delayed observations).
    Mean next-week yield change when true: -1.37 bp (19 weeks); opposite state: +4.57 bp (23 weeks).
    Timing points by year: 2023 -85.7, 2024 -148.9, 2025 +37.4.
    2026 Jan-Jun +108.5; Jul-Oct +16.4. After dropping the largest helpful timing week: +101.9.
2 | 2s10s curve | +132.0 | +104.1 | 1/3 | 11
    Condition: 2s10s spread 4-week change > 0.
    Mean next-week yield change when true: -2.15 bp (13 weeks); opposite state: +3.59 bp (29 weeks).
    Timing points by year: 2023 -145.8, 2024 +21.8, 2025 -1.7.
    2026 Jan-Jun +47.4; Jul-Oct +67.6. After dropping the largest helpful timing week: +76.4.
3 | WTI oil | +57.0 | +76.2 | 1/3 | 9
    Condition: WTI oil 4-week return > 0 (one-week-delayed observations).
    Mean next-week yield change when true: +0.33 bp (27 weeks); opposite state: +4.13 bp (16 weeks).
    Timing points by year: 2023 -31.8, 2024 -61.3, 2025 +15.2.
    2026 Jan-Jun +40.9; Jul-Oct +36.2. After dropping the largest helpful timing week: +51.4.
4 | Dollar basket | +69.0 | +70.7 | 2/3 | 7
    Condition: Six-currency dollar index 4-week change > 0.
    Mean next-week yield change when true: +0.14 bp (22 weeks); opposite state: +3.43 bp (21 weeks).
    Timing points by year: 2023 +34.2, 2024 -104.7, 2025 +0.0.
    2026 Jan-Jun +20.3; Jul-Oct +48.4. After dropping the largest helpful timing week: +50.6.
5 | German yield | +53.0 | +68.7 | 3/3 | 4
    Condition: German 10-year yield, latest monthly change (as published) > 0.
    Mean next-week yield change when true: +0.42 bp (26 weeks); opposite state: +3.76 bp (17 weeks).
    Timing points by year: 2023 +112.9, 2024 +60.4, 2025 +2.7.
    2026 Jan-Jun +17.5; Jul-Oct +44.4. After dropping the largest helpful timing week: +48.6.

Natural gas, oil and the dollar use observations delayed seven days.
The curve is the four-week change in the 10-year minus 2-year yield.
The German reading is the latest archived MONTHLY change, repeated until a new release.
It is not a contemporaneous weekly Bund move. Four state changes are a limited sample.
Oil and German-yield directions in this watchlist reverse the main checklist directions.
Natural gas and oil have selected signs opposite to a simple inflation-continuation hypothesis.
An already-priced or overshot response could explain a reversal, but these data do not establish that mechanism.
Lower conditional yield changes can still be positive. A positive timing score does not establish an add-duration case.
Natural gas is absent from the frozen 13-rule model. Its selected direction lost timing points in 2023 and 2024.
The main model continues to use its existing definitions; this list is a separate research panel.
Same-week market correlations are in a separate file and never enter the forecast.

In 2026 the main model scored +58.0 bp; reducing duration every week scored +75.0 bp.
The model's constant-bias component was +12.2 bp, leaving +45.8 bp of descriptive timing contribution.
Core inflation stayed on one side of its cutoff, so its timing score is zero.
Published weekly forecasts and actual portfolio execution are needed to assess live usefulness.
All input-vintage, availability and selection limitations in the paper still apply.
