Forecasting products, tested before they are offered

Budgets and contracts need a defensible range, not a point guess. In testing, no model we tried beat a simple benchmark for these series, so Northline does not pretend to forecast direction; it shows the benchmark path and a range built from how far that benchmark has actually missed in the past, with the measured hit rate beside it.

"Held in testing" is the share of later outcomes inside 80% ranges built the same way, out of sample (target 80%; across all series tested it ranged from 55% to 100%). The track record counts overlapping recent ranges, so it is not twelve independent tests. The ends of a range are not limits on what can happen.

• This is not a forecast of direction. The central path is the stronger of "no change" and "recent trend".
• A range is built from past errors; a new kind of shock can exceed it. Its ends are not limits on what can happen.
• In testing, 80% ranges covered between about 68% and 91% of later outcomes depending on the series; the figure for each series is shown.
• National figures; your own costs may follow them differently. That response is your assumption.

The prime rate that prices most floating-rate business credit follows the Bank of Canada target rate closely and almost immediately in the data. Where a relationship is that direct, honest value is a clear, checkable calculation of what a rate change would mean for the interest bill, not a forecast of what the Bank will do (in testing, no model beat "no change" for rates).

Observed 2010-01 to 2026-10: the prime rate stayed 1.96 to 2.37 points above the Bank of Canada target, and moved 0.97 points per point of policy change over 67 non-overlapping three-month windows (standard error 0.01). Enter your floating-rate borrowing and the share that reprices, and Studio shows the arithmetic.

• This is a conditional scenario, not a prediction that rates will move.
• Pass-through is the observed history of the prime rate; lenders' margins and your own contract terms are yours to enter.
• Mortgage rates follow bond yields as well as the policy rate; they are not quantified here.

The new housing price index is smooth and persistent, so a simple momentum model beat its benchmark out of sample at one month in testing. That is the only forecasting claim in this study that survived correction for testing many series. It is shown as a research preview with the benchmark path and range, not as a released forecast.

• Unemployment and employment forecasts. No model beat the benchmark at any tested horizon (error ratio 0.86 to 3.23, none significant).
• Oil, natural gas and exchange-rate price forecasts. No model beat the benchmark. Range and sensitivity tools are offered instead; USD/CAD has too little published history (about 115 months) to test.
• CPI, wage and earnings direction forecasts. After comparing with a trend benchmark, no model was reliably better. Planning ranges are offered instead.
• Housing starts, mortgage balances, manufacturing sales, Alberta unemployment. No reliable skill beyond the benchmark (starts: one month only, before correction). Not offered as forecasts.
• Interest-rate forecasts (policy, prime, mortgage rates). No model beat "no change" and the error ranges covered only 55% to 79% of outcomes. Conditional sensitivity is offered instead.

One click turns a range or a rate scenario into scenarios in your business analysis. You state how strongly your costs, sales and borrowing respond; Northline never supplies that. Results show what each number rests on, and the ranges are labelled as ranges, not forecasts.