A proper out-of-sample test
Rolling-origin backtesting tests next 1–3 months without using future observations. Every ranked method faces the same test months.
FIELD TOOL Nº 01 / FORECAST CLUB
Make the numbers
make sense.
Welcome to Forecast Club — the first analytical field tool by El Explorador. Six forecasting methods, one clever workspace. Turn demand history into 12 months of foresight, without rebuilding another Excel model.
LET'S FORECAST ↗NO CODING. NO SIGN-UPS.
JUST YOU & YOUR DATA.
JUST YOU & YOUR DATA.
STAYCURIOUS
✳
01 — THE STARTING POINT
Bring your numbers.
INPUT: HISTORICAL MONTHLY DEMAND ↙
YOUR DATA, PLEASE✳
Drop that spreadsheet!
Drag an .xlsx or .csv right here, or pick one from your device.
✳
THE RECIPE / WHAT TO UPLOAD
Four columns.
Endless possibilities.
S.NOYEARMONTHACTUAL DEMAND
012025JAN1,240
022025FEB1,310
032025MAR1,190
Works with 6–36 months of monthly data (up to 120 supported). Dates should have no missing months. Seasonal models need more history.
✳ No subscriptions. No servers.
Everything happens on your device.
Everything happens on your device.
02 — THE RESULTS ARE IN
Let's see what's next.
THE FUTURE ISN'T GUARANTEED. ↙
THE BIG PICTURE / 12 MONTHS AHEAD
Demand, past & predicted.
ACTUAL FORECAST
✳ Forward projections are estimates, not realized accuracy. Change models above to compare trajectories.
03 — THE MODEL SHOWDOWN
Six ways to look ahead.
Same held-out months. Same 1–3-month horizons.
A fair fight between eligible models.
HISTORICAL BACKTEST / ERROR SCORECARDSCROLL SIDEWAYS FOR MORE METRICS →
04 — THE MODEL PLAYGROUND
Tinker away, analyst.
Change α, β, γ, period T or window lengths.
Forecasts and rankings will recalculate.
05 — THE NUMBERS TO KEEP
A full year on paper.
All six models side by side.
Your spreadsheet is a click away.
THE NEXT 12 MONTHSUNITS OF ACTUAL DEMAND
06 — THE FINE PRINT
Numbers need
good judgement.
Seasonality isn't magic
Holt–Winters uses additive seasonal components. For a 12-month cycle, 24 months are needed to forecast; at least 27 to compare on held-out months.
Accuracy has its caveats
WAPE is the primary ranking metric, with RMSE as a tie-break. MAPE ignores months with zero actual demand; bias means forecast minus actual.
Built for iteration
Methods are independent modules. Expand later with ARIMA, SARIMA, intermittent-demand models, or new evaluation metrics.