Trent TurnerLocal Demand Momentum
Testing whether recent demand improves forecasts beyond annual seasonality
Retained component
- My Role
- I developed and evaluated this component within the demand-forecasting project, including benchmark comparison and interpretation.
- Business / Analytical Question
- Can recent normalized demand capture the rate of change as well as the current level?
- Evidence
- Six completed weeks of seasonally normalized demand. 2026 sensitivity check: rising-season MAE 1.27 vs. 1.47; declining-season 1.92 vs. 1.80, both vs. recent 4-week benchmark.
- Decision / Finding
- Retain momentum; prioritize post-peak adaptation for further testing.
- Outcome / Status
- Retained component: local momentum helps the adaptive level respond to changing demand. Post-peak behavior remains a refinement question. Analytical validation; no operational deployment is claimed.
Business Question
Can recent normalized demand capture the rate of change as well as the current level?
Why It Matters
A flat rolling mean is slow when demand is rising or falling quickly. Seasonal normalization separates the recurring annual slope from current-year momentum.
Data / Method
Estimate a local linear endpoint across the previous six normalized observations. The documented weights are −4, −1, 2, 5, 8, and 11, divided by 21.
Analysis
Six completed weeks, one local endpoint
Relative-demand index · 2026 W15–W20 approximation
Illustrates the documented local-trend operation using workbook observations. The endpoint illustrates the local trend on the same protected index scale; it cannot be used to recover the absolute forecast. This is not a momentum-only validation.
View chart data
| Week | Indexed observations | Six-week local linear fit |
|---|---|---|
| 15 | 88.40 | 89.78 |
| 16 | 112.40 | 87.90 |
| 17 | 82.20 | 86.01 |
| 18 | 62.00 | 84.12 |
| 19 | 46.80 | 82.24 |
| 20 | 118.60 | 80.35 |
Method detail
This is a local trend estimate, not a positive-weight moving average. Recent observations receive progressively greater weight. The final forecast floors the relative estimate at 0.01.
Results
Rising season · W01–W20
2026 calendar-alignment sensitivity check · approximate MAE · weekly New Customers
Post-peak adaptation: the primary failure mode
2026 calendar-alignment sensitivity check · approximate MAE · weekly New Customers
Interpretation
Post-peak adaptation emerged as the primary failure mode in this sensitivity check. In W21–W36, adaptive MAE was 1.92 versus 1.80 for the recent four-week benchmark: the simpler recent-history estimate adjusted faster during decline. Next, test post-peak adaptation using exact daily alignment and chronological validation.
Limitations & Considerations
The phase results evaluate the full architecture under an overlap-weighted calendar approximation. They are directional evidence, not a standalone momentum ablation or exact 2026 validation.
How this fits into the forecasting system
Retained component: local momentum helps the adaptive level respond to changing demand. Post-peak behavior remains a refinement question.