Forecasting and Trend Analysis
Every forecasting tradition eventually meets the same objection: you only remember the calls that landed. The remedy is not a better framework, it is a record written before the outcome is known.
We ran two on the same ledger. The first was a traditional reading framework, applied by practitioners who had used it for years. The second was a trend model over the same underlying series — rates, volumes, and dates. Both had to commit to a direction and a time window before the period closed.
The interesting result was not the hit rate, which was close enough on our sample that neither could be called ahead. It was where the two disagreed. Those cases split cleanly in two: periods where the series had broken a long-standing pattern, which the reading framework flagged early and the model absorbed slowly, and periods of ordinary noise, where the model was right and the reading found a structure that was not there.
The use of that split is not to crown a winner. It is to know which kind of surprise you are looking at — a judgement no single framework can make on its own.
