Safety stock, service level and newsvendor: how much to hold and how much to order
The mathematics of stock for replenished and seasonal goods: safety stock from demand and lead-time variability, reorder point, service level (cycle and fill-rate), the newsvendor problem for a one-off seasonal order, EOQ with caveats and the size pooling effect. The decision: how much to hold above expected demand and how much to order for the season.
Why this document
The replenishment planner and the buyer read this when deciding how much to hold and how much to order. Safety stock costs money and running short costs sales, so the service level is chosen deliberately: each next point of service costs more than the previous, the curve is convex. A seasonal item with no reorder is calculated differently, by the newsvendor problem, because surplus and shortage cost different amounts there. Without the calculation, stock is set by feel and you get both frozen cash and an empty shelf.
At a glance
- What is this document and why is it needed?
- The mathematics of stock for replenished and seasonal goods: safety stock from demand and lead-time variability, reorder point, service level (cycle and fill-rate), the newsvendor problem for a one-off seasonal order, EOQ with caveats and the size pooling effect. The decision: how much to hold above expected demand and how much to order for the season.
- Who owns the document and how often is it rebuilt?
- Inventory planner or buyer; parameters are reviewed quarterly (replenishment) and before each seasonal order (newsvendor); the result feeds replenishment rules, OTB and the working capital budget.
- What mistakes are the most common?
- The contractual lead time is used instead of the actual one → safety stock understated → compute from receipt history and include lead-time spread. Cycle service level and fill rate are mixed up → contractual commitments do not match the plan → state which measure is meant. The same 99% service for the whole range → stock triples without return → service by item role and margin.
For whom
- Brand
- A brand runs a finished-goods warehouse and production orders with long lead times and minimum lots: safety stock is computed from lead time including its variability, and the seasonal volume by the newsvendor problem where the cost of a shortage (lost margin) and of surplus (liquidation) differ.
- Retail
- A retailer holds stock in the warehouse and in stores: pooling sizes and locations reduces required safety stock while the display minimum sets a floor regardless of the formula. Seasonal purchasing is sized by the critical ratio, the replenished base by the reorder point.
What it consists of
Safety stock from demand and supply variability
- If weekly demand has mean d = 120 units and SD σ_d = 38 units, and lead time is L = 3 weeks with SD 0.5, the SD of demand over the lead time is σ_LT = √(L·σ_d² + d²·σ_L²) = 89.1 units; the second term (lead time variability) matters here almost as much as the first.
- Safety stock SS = z·σ_LT, where z is the normal quantile for the target no-stockout probability per cycle; reorder point ROP = d·L + SS.
- The normal approximation works poorly for rare items (demand in units per week) and for short life cycles; then use an empirical quantile of history or a Poisson distribution.
Service level: two different numbers
- Cycle service level is the probability of not hitting zero during the lead time; fill rate is the share of demand served from stock. With Q = 480 units and z = 1.64 (95%) the cycle level is 95% while fill rate is 99.6%: the numbers differ, and contracts and reports must say which is meant.
- Raising service level gets steeper in cost: from 90% to 99% safety stock grows from 114 to 207 units (1.8 times); the last percentage points of service are almost always unprofitable for fashion items.
Newsvendor: a single seasonal order
- A single order without repeat: under-ordering costs lost margin C_u = price − cost = 3,000 RUB/unit, over-ordering costs C_o = cost − liquidation price = 800 RUB/unit. The optimal probability of covering demand is the critical ratio CR = C_u/(C_u + C_o) = 78.9%.
- For demand with mean 900 units and SD 260 units the optimal order is Q* = μ + z·σ = 1109 units (z = 0.8), 23% above expected demand; expected profit is RUB 2,414,838 against RUB 2,305,845 when ordering exactly the mean.
- Rule: a high margin and cheap liquidation push the order above the mean; a low margin and costly liquidation below. For fashion the second happens on trend items.
EOQ and its limits
- The classic EOQ = √(2·D·S/H) at annual demand D = 6,240 units, order cost S = RUB 3,500 and holding cost H = 500 RUB/unit per year gives 296 units; total cost is flat near the minimum (an order of 600 units is only 26% above the minimum), so rounding to a pack multiple or MOQ costs almost nothing.
- EOQ ignores seasonality, obsolescence risk and volume discounts; it is not applicable to seasonal goods and serves as a guide for basics, while real limits come from MOQ, freight and warehouse space.
Sizes: the pooling effect
- If demand across five sizes is independent and the total variance is preserved, safety stock computed size by size is 237 units while for pooled demand it is only 108 units: pooling saves 54%.
- Brand: pools sizes through a central warehouse and top-up deliveries; Retailer: through transfers between stores and through the online stock. The pooling condition: the ability to reallocate fast, otherwise the saving stays on paper.
Brand and retailer: stock is computed differently
- Brand: lead time: from placing the production order to the warehouse; it is long and unstable (fabric, factory capacity, customs), so lead-time spread is often more important than demand spread.
- Retailer: lead time: from a replenishment order to the store shelf; it is short but composed of warehouse, transport, receiving and display, and the spread across stores is set by logistics.
- Brand: the seasonal purchase is sized by the critical ratio with the shortage cost as lost wholesale margin and the surplus cost as a return, partner compensation and liquidation.
- Retailer: the seasonal purchase is sized by the critical ratio with the shortage cost as lost retail margin and the surplus cost as a markdown with lost shelf space.
- Brand: safety stock is held at the central finished-goods warehouse for partner top-ups; the stock unit is model-color-size and size pooling is possible through top-up deliveries.
- Retailer: safety stock is split between the warehouse and stores: the display minimum sets a floor in the store regardless of the formula, and pooling is possible through transfers and the online stock.
- Brand: the factory minimum lot (MOQ) and pack multiple often exceed the optimal order: the decision rests on the cost of surplus and the possibility of combining models into one order.
- Retailer: pack multiple and the supplier minimum order set the order step: for the replenished basics it pays to combine orders by supplier into one so stock is not inflated.
- Brand: a brand fixes the service level in the partner contract (OTIF, fill rate); penalties for short delivery are part of the shortage cost, so it is computed from the contract terms.
- Retailer: a retailer sets the service level by category role (core, fashion, image) and by the share of empty shelf; lost sales are estimated from demand on neighboring days and from the size curve.
Where the data comes from
| Data source | Fields |
|---|---|
| Weekly sales and stock | Demand by SKU or model for at least 26-52 weeks, with demand reconstruction on out-of-stock days. |
| Deliveries | Actual time from order to receipt, lead-time spread, share of short deliveries, MOQ and pack multiple. |
| Item economics | Price excluding VAT, cost, residual liquidation price, holding and capital cost, size curve. |
How to build it step by step
- Step 1. Split the range into replenished basics and seasonal items: reorder points for the first, newsvendor for the second.
- Brand: Step 2. Take the actual time from production order to warehouse over past lots and its spread, not the contractual lead time.
- Retailer: Step 3. Take the time from a replenishment order to the store shelf (warehouse, transport, receipt, display) and the spread across stores.
- Step 4. Estimate mean and SD of weekly demand on reconstructed demand; check the distribution shape (tail, zeros).
- Step 5. Set the service target by item role (core, fashion, image) considering the price of surplus and shortage; compute z.
- Step 6. Compute σ_LT, safety stock and reorder point; check against the display minimum and pack multiple and take the larger.
- Step 7. For seasonal items compute CR, the optimal order Q* and the profit at neighboring volumes; choose a volume considering MOQ and risk.
- Step 8. For basics use EOQ as a guide but confirm the lot by MOQ, multiple and freight cost.
- Step 9. Estimate the cost of stock (SS × cost × capital rate) and show it next to the service gain.
- Step 10. Revisit parameters when demand or lead time shifts: tracking signal and actual service level trigger a recalculation.
Formulas
| Measure | Formulas | Note |
|---|---|---|
| SD of demand over lead time | σ_LT = √( L·σ_d² + d²·σ_L² ) | First term: demand spread, second: lead-time spread; with a constant lead time σ_L = 0. |
| Safety stock and reorder point | SS = z·σ_LT; ROP = d·L + SS | For periodic review the protection interval is L + R (review period). |
| Fill rate | FR = 1 − σ_LT·G(z) / Q, G(z) = φ(z) − z·(1 − Φ(z)) | G(z) is the normal loss function; Q is the order size. |
| Critical ratio | CR = C_u / (C_u + C_o); Q* = μ + z(CR)·σ | C_u = price − cost; C_o = cost − liquidation price; normal demand. |
| Newsvendor expected profit | E[π(Q)] = (p − s)·E[min(D, Q)] − (c − s)·Q; E[min(D, Q)] = μ − σ·G((Q − μ)/σ) | p price, c cost, s liquidation price; holding and capital costs are added to c. |
| EOQ and total cost | EOQ = √(2·D·S / H); TC(Q) = D·S / Q + H·Q / 2 | A flat minimum: at Q = 0.8·EOQ cost is 2.5% above the minimum, at Q = 1.2·EOQ 1.7%; the general deviation formula for Q = (1 + x)·EOQ is ½·(1 + x + 1/(1 + x)) − 1. |
| Pooling effect | SS_separate = z·√L·Σσ_i; SS_pooled = z·√L·√(Σσ_i²) | Valid for independent sizes; with positive correlation the gain is smaller. |
Formulas in notation
- Loss function and fill rate
G(z) = φ(z) − z·[1 − Φ(z)]; FR = 1 − σ_LT·G(z)/Q
Expected shortage per cycle σ_LT·G(z) divided by the order.- Newsvendor profit
E[π(Q)] = (p − s)·( μ − σ·G(z_Q) ) − (c − s)·Q, z_Q = (Q − μ)/σ
Expected profit when ordering Q.- EOQ
Q_EOQ = √( 2·D·S / H ); TC(Q) = D·S/Q + H·Q/2
Minimum of total ordering and holding cost.- Size pooling
SS_pool / SS_sep = √(Σ σ_i²) / Σ σ_i
Ratio of pooled to separate safety stock.
Worked example
Safety stock and service for an item as the service level rises
| Service level, % | z | Safety stock, units | Reorder point, units | Fill rate, % |
|---|---|---|---|---|
| 80 | 0.84 | 75 | 435 | 97.9 |
| 90 | 1.28 | 114 | 474 | 99.1 |
| 95 | 1.64 | 146 | 506 | 99.6 |
| 97.5 | 1.96 | 175 | 535 | 99.8 |
| 99 | 2.33 | 207 | 567 | 99.9 |
Demand 120 units per week, SD 38, lead time 3 weeks with SD 0.5, order 480 units. Safety stock grows non-linearly: from 80% to 99% it grows 2.8 times while fill rate rises only from 97.9 to 99.9%. Seasonal part: the optimal order is 1109 units at a critical ratio of 78.9%. The data is illustrative.
Figures
Data as a table
| Measure | Safety stock, units |
|---|---|
| 50 | 0 |
| 60 | 23 |
| 70 | 47 |
| 80 | 75 |
| 85 | 92 |
| 90 | 114 |
| 95 | 146 |
| 97.5 | 175 |
| 99 | 207 |
| 99.5 | 229 |
Safety stock grows non-linearly with service level. Each next point of service costs more than the previous: the curve is convex. Axis: cycle service level, %.
Data as a table
| Measure | Expected profit, RUB thousand |
|---|---|
| 500 | 1,473 |
| 600 | 1,739 |
| 700 | 1,975 |
| 800 | 2,167 |
| 900 | 2,306 |
| 1000 | 2,387 |
| 1100 | 2,415 |
| 1200 | 2,399 |
| 1300 | 2,353 |
| 1400 | 2,290 |
Newsvendor expected profit and order volume. The maximum lies above mean demand (900 units) near 1,100 units; right of the peak the curve falls slower than on the left.
Data as a table
| Measure | Order and holding cost, RUB thousand per year |
|---|---|
| 100 | 243 |
| 150 | 183 |
| 200 | 159 |
| 250 | 150 |
| 296 | 148 |
| 350 | 150 |
| 450 | 161 |
| 600 | 186 |
| 800 | 227 |
EOQ total cost: a flat minimum. Near EOQ cost barely changes: rounding to a multiple and MOQ are acceptable.
Data as a table
| Measure | Safety stock, units |
|---|---|
| XS | 31 |
| S | 51 |
| M | 59 |
| L | 54 |
| XL | 42 |
| Sum | 237 |
| Pooled | 108 |
Safety stock by size and the pooling effect. The sum of per-size stocks is clearly larger than the stock for pooled demand.
Specifics by category
- Apparel
- Basics are sized by reorder point, seasonal items by newsvendor; the liquidation price falls toward season end, so C_o grows and the optimal order falls for late reorders.
- Footwear
- Safety stock is computed by model, not size; edge sizes (the smallest and largest shares) need a minimum cell; a high unit price and a high return share raise the cost of surplus.
- Accessories
- A low unit price and slow turnover: surplus is cheap, shortage is costly at gift peaks; basic accessories get a higher service level, seasonal gift sets use a critical ratio computed at the peak.
- Kids
- Sizes go obsolete fast (the child grows), so surplus is costly and shelf life limited; school wear is a peak with a hard deadline and a high shortage cost.
- Men
- Demand is steadier, σ_d lower, so safety stock on basics is small; a high repeat share allows a higher service level on the core without large stock.
- Women
- More fashion items with a short cycle: newsvendor is used more often and forecast accuracy determines σ; a wide size grid and colors fragment demand, so pooling matters.
Common mistakes
- The contractual lead time is used instead of the actual one → safety stock understated → compute from receipt history and include lead-time spread.
- Cycle service level and fill rate are mixed up → contractual commitments do not match the plan → state which measure is meant.
- The same 99% service for the whole range → stock triples without return → service by item role and margin.
- The seasonal order is set at the forecast mean → sales are lost at a high margin → compute the critical ratio and the optimal quantile.
- EOQ is applied to seasonal goods → the order ignores season end → EOQ only for basics, season by newsvendor.
- Safety stock is computed per size and summed → excess stock → compute at the pooled level where reallocation is possible.
Russian specifics
- Stock costs in Russia include a high cost of capital (loan rates above the return on stock): C_o and H are computed with the company real rate rather than an illustrative one; the rate is taken from the financial model and revisited when the key rate changes.
- Inventory is measured for reporting at cost or net realizable value if lower (FSBU 5/2019): the liquidation price in newsvendor must be aligned with the provision for inventory write-down. Rule: Order 180n of 15.11.2019: Ministry of Finance Order of 15.11.2019 No. 180n (FSBU 5/2019 Inventories) (rule check not completed).
- Labeling of light-industry goods and receiving times affect the actual lead time: include code processing in L, otherwise the reorder point will be lower than needed. Rule: Decree 1956 of 31.12.2019: Government Decree of 31.12.2019 No. 1956 (marking rules for light-industry goods) (rule check not completed).