Probabilistic and Special-Context Inventory Models

Introduction

Deterministic inventory models assume that demand, replenishment time, and supply conditions are known or sufficiently stable. Most real supply chains violate at least one of these assumptions. Demand fluctuates, lead times vary, products become obsolete, customers may switch suppliers after a stockout, and some items lose value rapidly with age. Inventory control therefore requires models that explicitly represent uncertainty and the economic consequences of being either short or overstocked.

This chapter develops two broad families of models. The single-period, or newsvendor, model is appropriate when a replenishment decision is made before uncertain demand is observed and unsold inventory has limited future value. Stochastic multi-period models are used when inventory can be replenished repeatedly and decisions must account for uncertain demand, lead time, service requirements, and future opportunities. The chapter then adapts these models to perishables, fashion products, spare parts, slow movers, and obsolescence.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain the economic structure of probabilistic inventory decisions.
2. Formulate and solve a single-period newsvendor problem.
3. Distinguish lost sales from backorders and incorporate each into inventory policy.
4. Apply stochastic multi-period reorder-point and order-up-to models.
5. Adapt inventory logic to perishables, fashion goods, spare parts, slow movers, and obsolete stock.
6. Evaluate assumptions, data requirements, governance risks, and managerial trade-offs.

1. The Economics of Uncertainty

An inventory decision balances two asymmetric risks. Ordering too little produces a shortage, which may cause lost margin, customer dissatisfaction, expediting, or contractual penalties. Ordering too much creates holding, markdown, disposal, financing, and obsolescence costs. The best decision is not necessarily the one that minimizes the expected number of stockouts; it minimizes expected total economic cost subject to strategic constraints.

Important terminology includes:

• Demand uncertainty: variation in quantity requested during a specified period.
• Lead-time uncertainty: variation in the time between placing and receiving an order.
• Lost sales: demand that disappears because it cannot be supplied immediately.
• Backorder: unfilled demand recorded for later delivery.
• Overage: inventory remaining after the relevant selling or service horizon.
• Underage: demand that could have been served if additional inventory had been available.
• Service level: a defined probability or proportion of demand served, such as cycle service level or fill rate.

A critical modeling decision is whether shortage demand remains relevant. In a backorder system, unmet demand is carried forward. In a lost-sales system, some or all of it vanishes. The distinction changes both the state transition of the inventory system and the appropriate cost function.

2. The Single-Period Newsvendor Model

2.1 Decision setting

The classical newsvendor problem applies when a product is ordered once before uncertain demand is known and has little or no useful value after the selling season. Examples include newspapers, concert merchandise, fresh bakery products, seasonal promotional goods, and some fashion items.

Let:

• Q = quantity ordered, in units;
• D = random demand during the period, in units;
• p = selling price per unit;
• c = purchase cost per unit;
• v = recovery or salvage value per unsold unit;
• g = shortage or underage cost per unit;
• h = overage cost per unit.

In the basic profit formulation, g = p - c is the contribution margin lost when one unit of demand cannot be served, and h = c - v is the net loss from one unit left over. These expressions assume no separate penalty, goodwill loss, or emergency procurement cost. If such effects exist, they should be added explicitly rather than hidden in the parameters.

The expected-profit objective is:

where (x)^+ = max(x,0). The term min(D,Q) represents units sold, and (Q-D)^+ represents ending excess inventory.

Under the standard assumptions of continuous demand, a known demand distribution, constant unit economics, and a single decision, the optimal quantity satisfies the critical-fractile rule:

where F is the cumulative distribution function of demand, C_u is the underage cost per unit, and C_o is the overage cost per unit. Thus:

For the basic price-cost-salvage model:

The ratio is dimensionless because both numerator and denominator are measured in currency per unit. It is also called the critical ratio. A higher shortage cost increases the target quantity; a higher overage cost reduces it.

2.2 Derivation by marginal analysis

Suppose the order quantity increases by one unit. The additional unit is useful only when demand is at least the expanded order quantity. Its expected marginal benefit is therefore related to the probability that the unit is needed, while its expected marginal loss occurs when demand is below the expanded quantity. At the optimum, the expected benefit and expected loss are balanced:

Rearranging gives the critical-fractile rule. This derivation explains why the model is a quantile rule rather than an average-demand rule.

2.3 Worked example: seasonal product

A retailer purchases a promotional product for $20 and sells it for $50. Unsold units can be liquidated for $8. Demand is normally distributed with mean 1,000 units and standard deviation 250 units. Assume demand is continuous and the normal approximation is acceptable.

The underage and overage costs are:

The critical ratio is:

The standard normal quantile associated with 0.7143 is approximately z = 0.57. Therefore:

The practical order quantity is approximately 1,143 units, subject to packaging and minimum-order constraints. The result is above mean demand because the margin lost from a stockout is greater than the net loss on liquidation.

The calculation should not be interpreted as a guarantee that 71.43% of actual demand will be met. It is a model-based target quantile under the stated economics and distributional assumptions.

2.4 Discrete demand and operational constraints

When demand is discrete, the exact optimum is the smallest feasible quantity satisfying:

If products are ordered in case packs, the candidate quantities must be rounded and compared. Similarly, capacity limits, budget constraints, supplier minimums, multiple products competing for shelf space, and demand substitution can invalidate the one-product unconstrained solution.

The newsvendor model also becomes unreliable when unsold stock can be carried forward at substantial value. In that situation, leftover inventory is not simply an overage loss; it affects future decisions and belongs in a multi-period model.

3. Stochastic Multi-Period Inventory Models

3.1 Inventory position and inventory status

A multi-period policy must distinguish physical stock from stock that is already committed. Inventory position is commonly defined as:

where OH_t is on-hand inventory at time t, OO_t is on-order inventory, and BO_t is backordered demand. All terms use units of product. In a lost-sales system, backorders are normally zero, but lost demand may still influence performance measurement.

A reorder point policy orders when inventory position reaches a specified threshold. An order-up-to policy raises inventory position to a target level at review times.

3.2 Continuous-review reorder-point model

A common policy is an (s,Q) policy: order quantity Q when inventory position is at or below reorder point s. If lead-time demand is uncertain, a practical reorder point is:

where μ_L is expected demand during lead time, σ_L is the standard deviation of lead-time demand, and z is the standard-normal factor associated with the desired cycle service level. The units of both μ_L and σ_L are units of product.

If demand per period is independent with mean μ_d and standard deviation σ_d, and lead time is a fixed integer L periods, then:

If both demand and lead time are random, with independent demand across periods and independent lead time, a widely used approximation is:

and:

This variance expression includes demand variability accumulated over lead time and variability caused by lead-time length. It is an approximation and should be replaced by empirical lead-time-demand distributions when data are sufficient or when dependence is material.

3.3 Periodic review and the order-up-to level

Under a periodic-review policy, inventory is reviewed every R periods and replenishment arrives after lead time L. A typical order-up-to level is based on demand over the protection interval L + R:

where S is the order-up-to level. With independent demand and fixed protection interval:

At a review epoch, the order quantity is:

This policy is often called an (R,S) policy. It is especially useful where frequent continuous monitoring is costly or where supplier and customer processes operate on scheduled cycles.

3.4 Service measures and shortage treatment

Cycle service level is the probability of no stockout during a replenishment cycle. Fill rate is the fraction of demand supplied immediately. These measures are not equivalent. A system can experience occasional stockouts but satisfy most demand, producing a high fill rate and a lower cycle service level.

For backorders, unmet demand remains in the system. A backorder cost may include administrative handling, delayed revenue, contractual penalties, and customer dissatisfaction. For lost sales, the demand is removed from future outstanding demand, but the firm may lose margin and future customer value. Consequently, the same reorder point may produce different economics under the two shortage regimes.

A simple expected-cost objective for a period can be expressed as:

where c_h, c_b, c_l, and c_o are unit costs for holding, backorders, lost sales, and other excess or obsolescence costs; and H, B, L_s, and O are the corresponding quantities. The terms must be defined so that no cost is counted twice.

3.5 Dynamic programming perspective

When decisions affect future demand, capacity, or product value, a state-based model is more appropriate. Let x_t denote the inventory state at period t, q_t the order decision, and D_t random demand. A simplified inventory transition is:

with suitable modifications for lead times, backorders, perishability, and lost sales. A finite-horizon objective may be written:

where V_t is the minimum expected future cost from state x_t, and C is the current-period cost. Dynamic programming is conceptually powerful but may become computationally demanding when there are many products, locations, age classes, or correlated uncertainties.

4. Special-Context Models

4.1 Perishables

Perishables have finite usable life. A unit’s value depends not only on quantity but also on age, remaining shelf life, temperature history, and regulatory status. A single aggregate inventory number can therefore be misleading.

The operational foundations are:

• first-expired, first-out issuing rather than merely first-in, first-out;
• age-specific demand and spoilage records;
• explicit markdown, donation, disposal, or substitution pathways;
• shelf-life constraints in procurement and customer promises;
• cold-chain and traceability controls where relevant.

For a product with a one-period life, the newsvendor model may be appropriate. For several periods of remaining life, an age-structured state is required. If x_a,t denotes units with age a at time t, then demand allocation and deterioration must be modeled by age. A unit that remains physically present but cannot legally or commercially be sold is not effective inventory.

4.2 Fashion and seasonal goods

Fashion demand is difficult because it combines demand uncertainty, short selling windows, trend effects, substitution, and markdown competition. Early-season demand observations can be used to update the distribution for later demand, creating a value for postponement and rapid replenishment.

Useful practices include limited initial buys, option contracts, supplier postponement, test markets, responsive replenishment, and planned markdown rules. The relevant overage cost includes not only purchase cost less salvage value but also brand dilution, selling expense, and the opportunity cost of scarce retail space.

4.3 Spare parts and maintenance inventory

Spare parts differ from ordinary finished goods because demand is often intermittent, failures may be correlated with installed equipment, and the cost of downtime can greatly exceed the part’s purchase price. A low-volume item may therefore deserve a high service target.

For critical parts, decisions should consider:

• failure-rate and reliability data;
• installed base and operating hours;
• repair, cannibalization, and refurbishment options;
• emergency procurement and transportation time;
• system-level consequences of part unavailability;
• commonality across equipment models.

Poisson or compound-Poisson demand models can be useful for independent failure events, but they should not be imposed when failures are driven by common shocks, maintenance campaigns, or changing operating conditions. Criticality classification is often more useful than a single service-level rule for all parts.

4.4 Slow movers and intermittent demand

Slow-moving items have many zero-demand periods and occasional large demands. Average demand and standard deviation calculated without recognizing this structure can produce unstable safety stocks. Croston-type methods and related intermittent-demand procedures estimate demand occurrence and demand size separately, although their use should be validated against the organization’s service and cost objectives.

Pooling across locations, substituting technically equivalent parts, repair, supplier-held stock, and make-to-order arrangements can be more effective than simply increasing safety stock. Forecast accuracy should be assessed with measures suitable for intermittent demand; percentage errors can be undefined or misleading when actual demand is zero.

4.5 Obsolescence and product substitution

Obsolescence may result from technology change, regulation, fashion, packaging revision, or the introduction of a successor product. Inventory policy should include the probability and timing of value loss, not merely historical demand.

A practical governance process uses lifecycle gates: introduction, growth, maturity, decline, end-of-life announcement, final buy, and last-time-buy review. At each gate, planners should reassess demand, remaining service obligations, supplier commitments, recoverability, and substitution. Forecasts should be scenario-based where structural change is likely.

5. Applied Case: Illustrative Medical-Device Component

Background

The following is an illustrative case. A manufacturer of diagnostic equipment stocks a replacement sensor. The sensor costs $180, has a supplier lead time averaging four weeks, and is required to keep installed equipment operational. Demand is intermittent, with an average of 12 units per week and a standard deviation of 7 units per week. Lead time is assumed fixed for the initial analysis. Management wants a 95% cycle service level.

Problem

The current reorder point is based only on average lead-time demand. Stockouts have caused field-service delays, while excess units risk design obsolescence when a new sensor version is introduced.

Analysis

Expected lead-time demand is:

The standard deviation of four-week demand, under the simplifying assumption of independent weekly demand, is:

For a 95% cycle service level, z is approximately 1.645. The reorder point is therefore:

The existing policy of 48 units protects against average demand but provides no statistical safety allowance. However, the result must be challenged: demand may be correlated with equipment failures, demand may be censored by prior stockouts, and lead time may not be fixed.

Approach and solution

The firm should use a differentiated policy. High-criticality hospitals receive priority allocation; lower-criticality customers may accept planned substitution. The inventory position is calculated using on-hand, on-order, and backordered quantities. A quarterly review estimates demand from the installed base and failure exposure rather than relying only on shipment history. Engineering provides a probability-weighted obsolescence scenario, while procurement negotiates a smaller minimum order and an end-of-life return option.

Outcome

The analytical reorder point rises to approximately 71 units, but the company does not apply it blindly. It combines the target with a lifecycle cap, supplier flexibility, and a review of the successor sensor. This avoids treating service level as the only objective.

Learning points

1. Safety stock is a response to uncertainty, not a substitute for better data.
2. Criticality changes the economic cost of shortage.
3. Inventory position is more informative than on-hand stock alone.
4. Obsolescence risk can make a statistically justified quantity financially unacceptable.
5. Cross-functional governance is essential for spare-parts decisions.

6. Technology, Data, and Governance

Modern inventory systems can combine point-of-sale data, maintenance records, supplier lead-time histories, weather, promotions, and product lifecycle information. Machine-learning models may improve demand classification or detect changing patterns, but they do not remove the need for economic parameters, causal reasoning, or operational constraints.

AI-supported replenishment should be governed through:

• documented training data and forecast horizons;
• monitoring for bias, drift, and stockout-censored demand;
• human approval for end-of-life and critical-service decisions;
• audit trails for overrides and parameter changes;
• scenario testing under disruption;
• access controls for commercially sensitive data.

Ethically, firms should not conceal shortage risk from customers when products affect safety, health, or essential services. Regulatory requirements may govern expiry dates, traceability, recalls, controlled products, and retention of service parts. A low-cost inventory policy is not acceptable if it creates unreasonable safety or compliance risk.

7. Limitations and Best Practices

Probabilistic inventory models are conditional representations, not physical laws. Their quality depends on demand data, distributional assumptions, cost estimates, and correct treatment of stockouts. Common limitations include nonstationary demand, dependence between demand and lead time, substitution, capacity constraints, parameter uncertainty, and strategic customer behavior.

Best practice is to:

1. define the decision horizon and service metric precisely;
2. separate lost sales from backorders in both data and equations;
3. estimate lead-time demand directly when possible;
4. use empirical or scenario distributions when normality is implausible;
5. include salvage, markdown, disposal, and obsolescence economics;
6. segment products by lifecycle, criticality, demand pattern, and perishability;
7. validate policies through simulation or controlled pilots;
8. review parameters after promotions, disruptions, engineering changes, and supplier transitions;
9. combine statistical targets with managerial constraints and governance;
10. measure total cost and customer outcomes, not only inventory turns.

Chapter Summary

Probabilistic inventory control recognizes that demand, lead time, product value, and shortage consequences are uncertain. The single-period newsvendor model selects a demand quantile by balancing underage and overage costs. Its critical ratio is appropriate when a decision is made once and leftover stock has limited future value.

Stochastic multi-period models represent replenishment over time. Reorder points protect against lead-time demand, while order-up-to policies protect against demand during the lead time and review interval. Inventory position incorporates on-hand, on-order, and backordered quantities. Lost sales and backorders require different interpretations because unmet demand either disappears or remains outstanding.

Perishables require age and shelf-life control; fashion goods require lifecycle and markdown logic; spare parts require criticality and failure-based analysis; slow movers require intermittent-demand methods; and obsolescence requires lifecycle governance. Technology can improve forecasting and monitoring, but sound policy still depends on valid data, explicit economics, cross-functional judgment, and ethical accountability.

Key Terms

Newsvendor model: A single-period inventory model for uncertain demand and limited leftover value.

Underage cost: Economic cost of ordering one unit too few.

Overage cost: Economic cost of ordering one unit too many.

Critical ratio: Underage cost divided by the sum of underage and overage costs.

Salvage value: Recoverable value of inventory remaining after the selling horizon.

Lead-time demand: Demand occurring while replenishment is in transit.

Inventory position: On-hand plus on-order inventory minus backorders.

Reorder point: Inventory-position threshold that triggers replenishment.

Order-up-to level: Target inventory position under periodic review.

Cycle service level: Probability of no stockout during a replenishment cycle.

Fill rate: Proportion of demand supplied immediately from available stock.

Lost sales: Demand that is not served and does not remain as an outstanding order.

Backorder: Demand recorded for later fulfillment.

Perishable inventory: Stock whose value or usability declines with time.

Intermittent demand: Demand pattern containing many zero-demand periods and occasional requirements.

Spare part: Component held to support maintenance or repair.

Obsolescence: Loss of inventory value caused by technological, regulatory, or market change.

Protection interval: Time span covered by a periodic-review inventory target.

Review Questions

Short Answer Questions

1. What economic trade-off does the critical ratio represent?
2. Distinguish lost sales from backorders.
3. Define inventory position and state its components.
4. Why is lead-time demand variability important in a reorder-point model?
5. Why can a high fill rate coexist with a lower cycle service level?

Long Answer Questions

1. Derive the newsvendor critical-fractile rule and explain its assumptions.
2. Compare continuous-review and periodic-review stochastic inventory policies.
3. Explain why spare-parts inventory should be segmented by criticality rather than managed using one service level.
4. Discuss how perishability and obsolescence alter the meaning of “excess inventory.”
5. Evaluate the benefits and limitations of using machine learning in probabilistic replenishment.

Analytical / Application Questions

1. A product costs $30, sells for $75, and has a salvage value of $10. Demand is normally distributed with mean 2,000 and standard deviation 400 units. Calculate the critical ratio and approximate optimal order quantity.
2. Weekly demand has mean 20 units and standard deviation 8 units. For a fixed five-week lead time and a 97.5% cycle service level, calculate the reorder point using the normal approximation.
3. A slow-moving spare part has low annual demand but causes major customer downtime when unavailable. Design a policy that combines statistical analysis, criticality, repair, substitution, and obsolescence controls.

Multiple Choice Questions with Answers

Question 1

In the basic newsvendor model, the critical ratio is calculated as:

A. Overage cost divided by underage cost

B. Underage cost divided by the sum of underage and overage costs

C. Selling price divided by purchase cost

D. Mean demand divided by demand standard deviation

Correct Answer: B. Underage cost divided by the sum of underage and overage costs

Question 2

A product sells for $50, costs $20, and has a salvage value of $8. What is its critical ratio?

A. 0.286

B. 0.400

C. 0.714

D. 0.833

Correct Answer: C. 0.714

Question 3

Which inventory quantity includes on-hand inventory, on-order inventory, and outstanding backorders?

A. Cycle stock

B. Safety stock

C. Inventory position

D. Available capacity

Correct Answer: C. Inventory position

Question 4

With independent demand and a fixed lead time of L periods, lead-time demand standard deviation is commonly estimated as:

A. Lσ_d

B. σ_d ÷ L

C. √L × σ_d

D. μ_d + Lσ_d

Correct Answer: C. √L × σ_d

Question 5

Under a periodic-review order-up-to policy, the protection interval generally includes:

A. Only the review period

B. Only supplier lead time

C. Lead time plus the review period

D. The product’s entire lifecycle

Correct Answer: C. Lead time plus the review period

Question 6

Which statement best distinguishes a backorder from a lost sale?

A. A backorder is always caused by excess inventory

B. A backorder remains an outstanding requirement, whereas a lost sale may disappear

C. A lost sale is recorded for later shipment

D. There is no operational difference

Correct Answer: B. A backorder remains an outstanding requirement, whereas a lost sale may disappear

Question 7

A key reason to use age-specific records for perishables is that:

A. All units have identical remaining value

B. Physical presence always implies saleability

C. Usability and value depend on remaining shelf life

D. Demand is always deterministic

Correct Answer: C. Usability and value depend on remaining shelf life

Question 8

Which characteristic is especially important in spare-parts inventory?

A. All parts have identical shortage consequences

B. Demand is always smooth and seasonal

C. Equipment criticality and downtime consequences

D. Salvage value is always zero

Correct Answer: C. Equipment criticality and downtime consequences

Question 9

If underage cost increases while overage cost remains constant, the newsvendor optimal quantity generally:

A. Decreases

B. Remains exactly unchanged

C. Increases

D. Becomes zero

Correct Answer: C. Increases

Question 10

For a 95% cycle service level using a normal approximation, the safety factor z is approximately:

A. 0.50

B. 1.00

C. 1.645

D. 2.576

Correct Answer: C. 1.645