Croston Intermittent Demand Forecasting

Croston's method forecasts intermittent demand by separately smoothing non-zero demand sizes and the intervals between them, with the Syntetos-Boylan approximation correcting its upward bias.

Croston 방법은 간헐적 수요를 예측할 때 0이 아닌 수요량과 수요 사이 간격을 따로 평활하며, Syntetos-Boylan 근사로 위쪽 편향을 보정한다.

How should I forecast spare-part demand that is zero in most months?

대부분의 달에 수요가 0인 부품의 수요를 어떻게 예측해야 하나요?

Croston's method updates the demand size and inter-demand interval only when a demand occurs, and forecasts their ratio as the demand rate per period. The Syntetos-Boylan approximation multiplies that ratio by one minus half the smoothing constant to remove bias.

When to use Croston Intermittent Demand Forecasting

• Many periods have zero demand.
• You need inventory or reorder planning for slow-moving items.
• The average inter-demand interval exceeds about 1.32 periods.

When to use something else

• Demand is regular with few zeros; use Holt-Winters Exponential Smoothing or the Theta method.
• Demand is driven by known covariates or promotions; use a count regression such as Poisson regression.

Assumptions and requirements

• Demand sizes and intervals are independent.
• Non-negative demand.
• A stationary demand process.

What the analysis reports

• Croston and SBA demand rate per period
• Smoothed demand size and interval
• ADI and CV² demand classification
• In-sample mean absolute error

How to report it

Croston's method (α = [value]) gave a demand rate of [value] per period; the Syntetos-Boylan corrected rate was [value].

Related methods

• Theta Method Forecasting
• Exponential Smoothing (Holt-Winters)

Run Croston Intermittent Demand Forecasting on Quorant