STOP 28 / 31Make the pieces add up.
Layout
STOP 28 · In practice

Make the pieces add up.

EXPLORE~6 min
Back to route map
  • Northstar forecasts total sales, categories, and regions.
  • Forecasts built independently can disagree: the category totals may not sum to the company forecast.
Ways to make Northstar's forecasts coherent. Illustrative Northstar examples, not measured business results.
ApproachWhat it meansNorthstar example
Independent

Forecast the company and each category separately; their totals need not agree.

Northstar's supplied Home, Electronics, and Apparel forecasts sum to $109k while the company forecast says $115k.

Bottom-up

Forecast the parts and add them to get the total.

Northstar keeps category detail but replaces the company's own forecast; noisy small categories can make the sum noisy.

Top-down

Forecast the total and allocate it using chosen shares.

Northstar's stable historical category shares may miss Apparel's newer growth or a change in shopping mix.

Reconciliation

Adjust forecasts across levels so aggregation constraints hold, using a specified weighting/covariance model.

  • Northstar can combine information from the total and categories, but poor covariance estimates can hurt.
  • The lab uses supplied diagonal weights, not fitted MinT.
FIG 28.1

Do the forecasts add up?

WHAT THIS SHOWS

Home, Electronics, and Apparel have their own forecasts, alongside a separately modeled company total. Compare leaving them independent, adding parts upward, splitting the total downward, and reconciling all levels.

The question it answers: How can Northstar's category forecasts agree with its company forecast?

Approach
Northstar · total$115,000
Home$52,000
Electronics$31,000
Apparel$26,000

Parts add to $109,000 · total $115,000 · gap $6,000

Adds up?No
Apparel share23.9%
Base Apparel share23.9%
Fine print & method

All dollar forecasts are supplied teaching values. Top-down uses fixed historical shares of 50%, 30%, and 20%, not the current category forecasts. Reconciliation is a weighted least-squares projection with supplied diagonal variances [4,1,1,2], not a fitted MinT covariance estimate. Consistency does not guarantee greater accuracy; evaluate that out of sample.

CHECK YOURSELF · CHOOSE AN ANSWER

What does forecast reconciliation enforce?

Your learning note

Private · just for you