Syntha Lab RU

Safety stock from forecast error

Counts safety stock from the forecast accuracy rather than the demand scatter: the weekly error from WAPE, the accumulation of the error over the lead time and review with the correlation of weeks, the service level. Shows the bounds for independent and fully linked errors.

Replenishment and stock cover For whom: Buyers and planners who count safety stock at a long lead time.

Calculate on your data →

At a glance

What question does it answer?
What safety stock is needed to cover the forecast error over the lead time and the review period at the chosen service level.
What data is needed?
The average demand and WAPE for the item class.
How to read the result?
Compare the stock with the bounds at independent and linked errors and with the stock from the demand scatter. Example on the default values: Safety stock, units: 279; Safety stock rounded to the pack, units: 279; Safety stock in weeks of demand: 2.8; Weekly forecast error σ, units: 43.8. The figures are illustrative: enter your own to see your result.

Why and how to reason

Safety stock is needed against the forecast error over the lead time and the review, not against the demand scatter. In the example demand is 100 units a week, WAPE 35%, horizon 6 weeks, correlation 0.3, service 95%: the weekly error is 43.75, over the term 169.4, safety stock 279 units (2.8 weeks); the bounds are 176 units for independent and 432 units for fully linked errors.

When to use it

Apply at long lead times (imports) and for seasonal goods; the classic stock from the demand scatter is counted by reorder and safetyfill.

How to read the result

Compare the stock with the bounds by correlation and with the stock from the demand scatter, if given.

Common mistakes

The demand scatter is taken instead of the forecast error, weekly errors are added as independent, someone else’s WAPE is taken.

What data is needed?

  • The average demand and WAPE for the item class.
  • The lead time and the review period.
  • The correlation of weekly errors and the service level.

Inputs

FieldUnitValue in the exampleMeaning
Average demand, units per weekpcs100The average weekly demand by the forecast for the lead time (cleaned, baserate). The example value is illustrative.
Forecast error WAPE%35Σ|actual − forecast| ÷ Σ actual for your item class on the weekly horizon. Take your own history, not someone else’s. The example value is illustrative.
Lead time, weekswk4From order to receipt. From China six to ten weeks make the error the main risk item. The example value is illustrative.
Order review period, weekswk2How often you order: safety stock covers the lead time plus the review period. The example value is illustrative.
Correlation of errors of adjacent weeks×0.30 means independent errors, 1 fully linked (the forecast is understated every week in a row). Estimate from history, 0.3 by default. The example value is illustrative.
Target service level%95The probability of not running out within an order cycle; 50% gives zero stock. The example value is illustrative.
Pack multiple, unitspcs1Safety stock is rounded up to a whole number of packs. 1 means no rounding. The example value is illustrative.
Demand scatter, units per week (optional)pcs—If given, the classic stock from the demand scatter is compared with the stock from the forecast error. Empty: no comparison. The example value is illustrative.
Weeks to the end of the season (optional)wk—If given and below the lead time, the horizon is cut to it: toward the end of the season the error accumulates less. Empty: no limit. The example value is illustrative.

How it is calculated

StepFormula
Weekly errorσ = 1.25 × WAPE × average demand = 1.25 × 35% × 100 = 43.75
HorizonT = lead time + review period = 6 weeks
Error over the termσ(T) = σ × √(T + T(T − 1)ρ) = 169.4
Multiplierz = inverse normal function of 95% = 1.645
Safety stockz × σ(T) = 278.7 units

Worked numbers

Default values from the app; in the interactive version you enter your own.

Safety stock, units
279z × the weekly error × the root of the accumulation over the term.
Safety stock rounded to the pack, units
279
Safety stock in weeks of demand
2.8
Weekly forecast error σ, units
43.8
Error over the term σ(T), units
169.4
Horizon T, weeks
6
Service level multiplier z
1.64
Bound: independent errors (ρ = 0), units
176
Bound: fully linked errors (ρ = 1), units
432
Classic stock from demand scatter, units
demand scatter not given
Safety stock by error correlation, units
Bar chart: ρ = 0 176, ρ = 0.3 279, ρ = 1 4320100200300400500ρ = 0ρ = 0Safety stock: 176176ρ = 0.3ρ = 0.3Safety stock: 279279ρ = 1ρ = 1Safety stock: 432432
Data as a table
MeasureSafety stock
ρ = 0176
ρ = 0.3279
ρ = 1432

The model: the weekly standard deviation of the error is taken as 1.25 × WAPE × average demand (an approximation for a normal error). This is an assumption, not a norm; check it on your history.

What to do with the result: put the stock into the reorder point (reorder), compare its cost with the cost of a shortage and recount when the lead time or the forecast error changes.

A typical mistake: taking the demand scatter instead of the forecast error and adding weekly errors as independent, understating the stock.

Nuances and limits

  • Taking the demand scatter instead of the forecast error.
  • Adding weekly errors as independent.
  • Taking someone else’s WAPE.

Terms used

Calculate on your data →