The newsvendor model

Definition

The newsvendor model handles the decision you only make once: ordering a single quantity, before demand is known, with no possibility of replenishing afterwards. It proves a counter-intuitive and decisive result: the optimal quantity is almost never mean demand. It is a quantile of the distribution, determined by the ratio between what a lost sale costs and what an unsold unit costs.

Page signature
The quantile that maximises margin
mean Q* low demand high demand
Cost ratio
·
Critical ratio
·
Optimal quantity
·
Slide: the economics of the error, not the mean, sets the quantity to order.

Why it matters

When there is no second chance.

Most inventory decisions can be corrected along the way: if you got it wrong, you reorder. Some decisions cannot. The seasonal collection that will not be reordered, the production run launched for a dated release, the volumes committed for an event, the campaign whose product expires: in every one of these cases, you order once, blind, and live with the result. This is precisely the ground of the newsvendor model.

Its contribution is to overturn a widespread reflex. Facing uncertainty, intuition orders the most likely demand, that is, the forecast mean. The model proves that this choice is optimal only in one very particular case: when erring on the high side costs exactly as much as erring on the low side. As soon as the error is asymmetric, and it almost always is, the optimal quantity departs from the mean.

The asymmetry is easy to grasp concretely. On a high-margin product whose leftovers are discounted at little loss, missing a sale costs far more than carrying one unit too many: it is rational to order well above the mean. Conversely, on a low-margin perishable whose leftovers are destroyed, every excess unit is a dead loss, and the optimum drops below the mean.

This reasoning reaches far beyond single-shot decisions. It provides the economic foundation of the service level: it states which service rate it is rational to target, instead of leaving it at an inherited value. That is what makes it one of the rare inventory models that speaks the language of the finance function directly.

Business impact

The central result fits in one sentence: the optimal quantity is the demand quantile corresponding to the critical ratio, that is, the share the cost of a lost sale represents in the sum of the two error costs. That ratio is, mathematically, the service level you ought to target.

The managerial reach is considerable. It means the service rate is not an objective to decree but a consequence to compute. Two references in the same family, with different margins and salvage values, deserve different service levels, and the model says which. This is the rigorous foundation of service differentiation.

The expert lesson concerns sensitivity. The result does not hinge on the precision of the mean forecast but on the quality of two figures, the margin lost on a missed sale and the real loss on an unsold unit. Many companies refine their forecast to a tenth of a point and have never seriously priced what a markdown costs them. The effort is misplaced.

The mechanism

Two error costs, one quantile.

The reasoning starts from the marginal unit. Adding one unit to the order creates two possibilities: either it sells, and it earns back the margin that would have been lost, or it does not, and it costs the gap between its cost and its salvage value. You keep adding units as long as the expected gain exceeds the expected loss, and stop when the two balance.

That balance defines exactly the critical ratio: the cost of a lost sale divided by the sum of the two error costs. A ratio of 0.80 means you should order so as to cover 80% of demand scenarios. The optimal quantity is therefore the 80% quantile of the distribution, not its mean.

The signature above makes this shift visible. The light line marks mean demand, the golden line the optimal quantity. When missing a sale costs little, the optimum stays to the left of the mean; when a lost sale becomes expensive, it moves decisively to the right, and the gap can reach tens of percent. Ordering the mean in that case means leaving margin on the table every season.

Two requirements follow. You need a distribution of demand, not a single point, which points back to probabilistic forecasting: without a measure of dispersion, no quantile can be computed. And you need honest error costs, including on the leftover side the markdown, the liquidation cost, the space occupied, and on the lost-sale side the immediate margin but also, where material, the effect on the customer relationship.

Q* = F−1 ( Cu / ( Cu + Co ) )
Cu the cost of a lost sale, essentially the forgone margin; Co the cost of an unsold unit, its cost less its salvage value; F−1 the inverse of the demand distribution, converting a percentage into a quantity. The ratio Cu / (Cu + Co) is the critical ratio, and it is also the optimal service level. Normal case: Q* = μ + z·σ, with z read at the critical ratio. Marker: if the two costs are equal, the ratio is 0.5 and the optimum meets the mean, the only situation in which common intuition is right.

The traps

Three single-shot decision errors.

01

Ordering the mean

This is the error the model exists to correct. Taking the most likely demand assumes that erring high and erring low cost the same. As soon as margins are strong or leftovers cheap to clear, this reflex mechanically leaves margin untapped, season after season, with nobody measuring it.

02

Underpricing the leftover

The cost of one unit too many is not its purchase price: it is that price less what you genuinely recover, plus liquidation costs, the cost of space and sometimes the effect of markdowns on brand image. An optimistic estimate inflates the critical ratio and drives systematic over-ordering.

03

Applying it to replenishable items

The model assumes no second order is possible. On a reference you replenish mid-season, it oversizes the first order when in-season reordering fills that role at lower risk. The right prior question is always: is this decision truly a single shot?

The rollout

Five steps to arbitrate a single decision.

Isolate the true single-shot decisions

List the cases where no reorder is possible: collections, limited runs, dated products, capacity commitments. These are the only ones where the model fully applies, and often those carrying the highest unit stake.

Price the two error costs

Establish the margin genuinely lost on a missed sale and the net loss on a leftover, salvage value and liquidation costs included. These two figures weigh more on the outcome than any refinement of the forecast.

Produce a distribution, not a point

Obtain the dispersion of expected demand, through probabilistic forecasting or from the history of comparable launches. Without dispersion the quantile cannot be computed and the model stays theoretical.

Set the quantile and own it

Translate the critical ratio into a quantity and document the gap with the mean. That gap must be explained to and accepted by sales and finance, otherwise it will be overridden by hand at the last moment.

Review after the fact

After the season, compare observed stockouts and leftovers with what the model predicted. A systematic excess of leftovers signals a critical ratio set too high, hence mis-estimated error costs; the reverse holds for stockouts.

Neighboring concepts

Read next.

From knowledge to action

Are your season commitments computed or estimated?

On decisions without reordering, the gap between the mean and the optimal quantile is worth margin points, every season. Our Inventory & distribution file prices your error costs and resets your commitments.