7.2 Stochastic Inventory Management: Safety Stock, Continuous / Periodic Review & Newsvendor

Stochastic Inventory Management: Safety Stock, Continuous / Periodic Review & Newsvendor

While deterministic inventory formulations establish foundational lot-sizing trade-offs, real-world industrial supply chains operate under relentless uncertainty. Customer demand fluctuates dynamically across time horizons, and supplier lead times vary due to transport bottlenecks, production backlogs, and customs inspections.

Stochastic inventory management integrates statistical probability distributions into inventory control logic, determining appropriate buffer stocks (safety stocks) and reorder thresholds to achieve targeted customer service levels while safeguarding working capital.

---

1. Inventory Control Policies: Continuous vs. Periodic Review

Industrial organizations manage inventory through two primary operating paradigms:

Continuous Review (s, Q) Policy:                Periodic Review (R, S) Policy:
Inventory Level                                  Inventory Level
     ^
   S |       Order Placed (Q)                         S |       Order (Q1)     Order (Q2)
     |       |                                          |       |              |
 ROP +-------*--------------                     ROP +-------*--------------*---
     |      / \                                         |      / \            / \
     |     /   \        Order Placed (Q)                |     /   \          /   \
     |    /     \       |                               |    /     \        /     \
   0 +---/-------*------*----------> Time             0 +---*-------*------*-------*----> Time
         |<-- L ->|     |<-- L ->|                          |<--R-->|<--R-->|
                                                            |<-- R + L ---->|

Continuous Review Policy $(s, Q)$ or $(r, Q)$

In a continuous review system, the inventory state is continuously tracked in real time (facilitated by barcode scanners, RFID readers, and automated ERP transaction logs):

• Inventory Position ($IP$): Decisions are governed not merely by physical on-hand stock, but by the inventory position: $IP = OH + SR - BO$ Where:
  • $OH$ = On-Hand physical stock physically in the warehouse.
  • $SR$ = Scheduled Receipts (open purchase orders placed with suppliers but not yet received).
  • $BO$ = Backorders (unfilled customer commitments awaiting delivery).
• Control Rule: Whenever the inventory position drops to or below the Reorder Point ($ROP$ or $s$), a fixed replenishment order of size $Q$ is immediately initiated ($IP \le s \implies \text{Order } Q$).
• Vulnerability Interval: The system is vulnerable to stockouts only during the replenishment lead time $L$. Once the lead time transpires, the order $Q$ arrives, raising inventory safely above $s$.

Periodic Review Policy $(R, S)$

In a periodic review system, inventory is inspected only at fixed, discrete time intervals of length $R$ (e.g., every Monday morning or on the first of each month):

• Control Rule: At each review epoch $R$, the inventory position $IP$ is counted, and a variable order quantity $q$ is placed to restore the inventory position back up to the target Order-Up-To Level ($S$): $q = S - IP$
• Vulnerability Interval: Because inventory is not monitored between review dates, the system is exposed to stockouts across the entire protection interval of $R + L$ (the review period $R$ plus the delivery lead time $L$). An unexpected surge in demand occurring immediately after a review epoch cannot be rectified until the next review interval plus lead time.

Comparison of Inventory Control Policies

2. Safety Stock Formulations Under Stochastic Regimes

Safety Stock ($SS$) is buffer inventory held to mitigate stockouts caused by stochastic demand surges, supplier transit delays, or both. The general Reorder Point formula is:

$ROP = \text{Expected Demand During Lead Time} + \text{Safety Stock} = \mu_{DL} + SS$

Depending on which parameters exhibit uncertainty, industrial engineers apply three distinct analytical models based on the Law of Total Variance.

Case 1: Stochastic Demand, Constant Lead Time

Assume daily demand is independent and identically distributed with mean $\bar{d}$ and standard deviation $\sigma_d$, while lead time $L$ is constant and deterministic:

• Expected lead time demand: $\mu_{DL} = \bar{d} \cdot L$
• Variance of lead time demand: $\sigma_{DL}^2 = L \cdot \sigma_d^2$
• Standard deviation of lead time demand: $\sigma_{DL} = \sigma_d \sqrt{L}$
• Safety Stock and ROP: $SS = z \cdot \sigma_d \sqrt{L}$ $ROP = \bar{d} \cdot L + z \cdot \sigma_d \sqrt{L}$ Where $z$ is the standard normal deviate associated with the desired cycle service level ($CSL$).

Case 2: Constant Demand, Stochastic Lead Time

Assume demand is strictly constant at rate $d$, but lead time varies with mean $\bar{L}$ and standard deviation $\sigma_L$:

• Expected lead time demand: $\mu_{DL} = d \cdot \bar{L}$
• Variance of lead time demand: $\sigma_{DL}^2 = d^2 \cdot \sigma_L^2$
• Standard deviation of lead time demand: $\sigma_{DL} = d \cdot \sigma_L$
• Safety Stock and ROP: $SS = z \cdot d \cdot \sigma_L$ $ROP = d \cdot \bar{L} + z \cdot d \cdot \sigma_L$

Case 3: Both Demand and Lead Time Are Independent Stochastic Variables

When both daily demand ($D \sim (\bar{d}, \sigma_d)$) and lead time ($L \sim (\bar{L}, \sigma_L)$) fluctuate independently, lead time demand $D_L$ is a compound random variable. By the Law of Total Variance:

$\sigma_{DL}^2 = E[L] \cdot Var(D) + (E[D])^2 \cdot Var(L) = \bar{L} \sigma_d^2 + \bar{d}^2 \sigma_L^2$

Taking the square root yields the standard deviation of lead time demand:

$\sigma_{DL} = \sqrt{\bar{L} \sigma_d^2 + \bar{d}^2 \sigma_L^2}$

• Safety Stock and ROP: $SS = z \sqrt{\bar{L} \sigma_d^2 + \bar{d}^2 \sigma_L^2}$ $ROP = \bar{d} \cdot \bar{L} + z \sqrt{\bar{L} \sigma_d^2 + \bar{d}^2 \sigma_L^2}$

3. Service Level Metrics: Cycle Service Level vs. Fill Rate

Industrial engineers must distinguish between two fundamentally distinct definitions of customer service:

             Cycle Service Level (CSL, Type 1)          Fill Rate (Beta, Type 2)
Definition:  Probability of NO stockouts per cycle      Fraction of DEMAND satisfied from stock
Formula:     CSL = P(D_L <= ROP) = Phi(z)               Beta = 1 - (Expected Shortage / Q)
Focus:       Frequency of stockout events               Magnitude / Volume of shortages
Sensitivity: Independent of lot size Q                  Directly improves as Q increases

Type 1 Service Level: Cycle Service Level ($CSL$)

Cycle Service Level ($CSL$) is the probability that demand during lead time does not exceed the reorder point:

$CSL = P(D_L \le ROP) = \Phi(z) = 1 - \alpha$

Where $\alpha$ is the cycle stockout probability.

Limitation: CSL considers only whether a stockout occurs, completely ignoring how many units are short. Whether the warehouse is short by 1 unit or 10,000 units, the cycle is scored identically as a failure. Furthermore, CSL does not reflect annual customer experience: if a firm orders twice a year ($Q$ is large), a 90% CSL implies a stockout once every 5 years ($2 \times 0.10 = 0.2$ stockouts/year); if it orders 50 times a year ($Q$ is small), the same 90% CSL implies 5 stockouts every year!

Type 2 Service Level: Fill Rate ($\beta$)

Fill Rate ($\beta$) represents the proportion of total customer demand satisfied immediately from available inventory:

$\beta = 1 - \frac{\text{Expected Shortage per Cycle } (ESC)}{Q} = 1 - \frac{E[n(ROP)]}{Q}$

Where $Q$ is the replenishment order batch size, and $ESC$ is the expected number of units short per replenishment cycle.

Computing Expected Shortage and the Unit Normal Loss Function

Assuming lead time demand follows a normal distribution $N(\mu_{DL}, \sigma_{DL})$, the expected shortage per cycle is evaluated via the Unit Normal Loss Function $L(z)$:

$ESC = \sigma_{DL} \cdot L(z)$

Where the unit normal loss integral is defined mathematically as:

$L(z) = \int_z^\infty (t - z) \phi(t) dt = \phi(z) - z [1 - \Phi(z)]$

Where $\phi(z) = \frac{1}{\sqrt{2\pi}} e^{-z^2/2}$ is the standard normal probability density function, and $\Phi(z)$ is the standard normal cumulative distribution function.

To find the safety factor $z$ required to achieve a designated target fill rate $\beta$:

$L(z) = \frac{Q (1 - \beta)}{\sigma_{DL}}$

Once $L(z)$ is computed, the corresponding $z$-value is retrieved from standard normal loss tables or software functions. Notice that as batch size $Q$ increases, the required $L(z)$ increases, which decreases the required $z$ and lowers safety stock. Larger replenishment orders naturally bolster fill rate because stockouts can occur only during the brief lead time window following each order.

4. Single-Period Stochastic Inventory: The Newsvendor Model

The Newsvendor (or Newsboy) Model addresses single-period inventory optimization for perishable, highly seasonal, or rapid-obsolescence items where inventory cannot be carried over to subsequent selling periods without severe financial loss (e.g., dated periodicals, high-fashion apparel, perishable chemicals, holiday electronics, or spare parts with one-time lifetime buys).

Cost Architecture

The optimal order quantity balances the marginal cost of ordering too much against the marginal cost of ordering too little:

• Cost of Overage ($C_o$): The loss incurred for each unsold unit remaining at the end of the selling period: $C_o = c - s$ Where $c$ = unit acquisition/production cost, and $s$ = salvage or liquidation value per unit ($s < c$). If disposal costs are incurred, $C_o = c - s + \text{disposal fee}$.
• Cost of Underage ($C_u$): The opportunity loss (unrealized profit) incurred for each unit of unmet customer demand: $C_u = p - c + g$ Where $p$ = retail selling price ($p > c$), and $g$ = customer goodwill penalty / contractual late penalty.

Marginal Economic Analysis Derivation

Let $Q$ represent the stocking quantity, and let demand $D$ be a continuous random variable with cumulative distribution function $F(x) = P(D \le x)$ and density $f(x)$.

If the vendor increments the stocking quantity from $Q$ to $Q + 1$:

• The additional unit will be sold if demand exceeds $Q$ ($D > Q$), yielding a marginal gain of $C_u$. The probability of this event is $P(D > Q) = 1 - F(Q)$.
• The additional unit will remain unsold if demand is less than or equal to $Q$ ($D \le Q$), resulting in a marginal loss of $C_o$. The probability of this event is $P(D \le Q) = F(Q)$.

At the optimal stocking point $Q^*$, the expected marginal gain of adding the $(Q+1)$-th unit must exactly balance its expected marginal loss:

$E[\text{Marginal Gain}] = E[\text{Marginal Loss}]$

$C_u \cdot [1 - F(Q^*)] = C_o \cdot F(Q^*)$

Expanding and rearranging terms:

$C_u - C_u F(Q^*) = C_o F(Q^*)$

$C_u = (C_u + C_o) F(Q^*)$

Solving for $F(Q^*)$ establishes the fundamental Critical Ratio (Critical Fractile):

$F(Q^*) = P(D \le Q^*) = \frac{C_u}{C_u + C_o} \equiv CR$

Probability Density f(D)
     ^
     |                 /\ 
     |                /  \ 
     |               /    \ 
     |              /      \ 
     |    Area =   /        \ 
     |      CR    /|         \ 
     |           / |          \ 
   0 +----------+--+-----------+----------> Demand (D)
                   Q*

Determining Optimal Stocking Quantity ($Q^*$)

1. Uniform Demand $D \sim U[a, b]$: $F(Q) = \frac{Q - a}{b - a} = CR \implies Q^* = a + CR \cdot (b - a)$
2. Normal Demand $D \sim N(\mu, \sigma)$: $Q^* = \mu + z \cdot \sigma$ Where $z$ is the standard normal value satisfying $\Phi(z) = CR$:
   • If $CR > 0.50$, then $C_u > C_o \implies z > 0$ (stock above the mean).
   • If $CR < 0.50$, then $C_u < C_o \implies z < 0$ (stock below the mean).
   • If $CR = 0.50$, then $C_u = C_o \implies z = 0 \implies Q^* = \mu$.

5. Worked Engineering Examples

Example 1: Safety Stock and ROP Under Compound Uncertainty

Problem Statement: An aerospace assembly plant sources specialized titanium fasteners from an overseas supplier. Operational parameters are:

• Mean daily demand ($\bar{d}$): 50 units/day
• Standard deviation of daily demand ($\sigma_d$): 8 units/day
• Mean replenishment lead time ($\bar{L}$): 16 days
• Standard deviation of lead time ($\sigma_L$): 3 days
• Target cycle service level ($CSL$): 97.72% ($z = 2.00$)

Calculate the standard deviation of lead time demand, required safety stock, and reorder point.

Solution:

Step 1: Compute variance and standard deviation of lead time demand: $\sigma_{DL}^2 = \bar{L} \sigma_d^2 + \bar{d}^2 \sigma_L^2$ $\sigma_{DL}^2 = 16 \times (8)^2 + (50)^2 \times (3)^2 = (16 \times 64) + (2,500 \times 9)$ $\sigma_{DL}^2 = 1,024 + 22,500 = 23,524$ $\sigma_{DL} = \sqrt{23,524} \approx 153.375\text{ units}$

Notice: The lead time variance component ($22,500$) contributes $\frac{22,500}{23,524} = 95.6\%$ of total system uncertainty!

Step 2: Calculate Safety Stock ($SS$): $SS = z \times \sigma_{DL} = 2.00 \times 153.375 = 306.75 \approx 307\text{ units}$

Step 3: Calculate Reorder Point ($ROP$): $\mu_{DL} = \bar{d} \times \bar{L} = 50 \times 16 = 800\text{ units}$ $ROP = \mu_{DL} + SS = 800 + 307 = 1,107\text{ units}$

When the inventory position drops to 1,107 units, an order is placed. The 307 units of safety stock protect against both fastener demand surges and supplier transit delays.

---

Example 2: Newsvendor Model for High-Tech Product Launch

Problem Statement: An electronics manufacturer is scheduling production for a limited-edition smart gaming device with a single seasonal selling window. Engineering and marketing supply the following financial estimates:

• Selling price ($p$): $280 per unit
• Manufacturing production cost ($c$): $100 per unit
• Post-season liquidation salvage value ($s$): $80 per unit
• Market demand is normally distributed: mean $\mu = 2,000$ units, standard deviation $\sigma = 400$ units.

Determine the cost of underage, cost of overage, critical ratio, and optimal production run size.

Solution:

Step 1: Determine Underage and Overage Costs: $C_u = p - c = \$280 - \$100 = \$180\text{/unit (lost profit per unsold demand)}$ $C_o = c - s = \$100 - \$80 = \$20\text{/unit (net loss on leftover units)}$

Step 2: Compute the Critical Ratio ($CR$): $CR = \frac{C_u}{C_u + C_o} = \frac{180}{180 + 20} = \frac{180}{200} = 0.90$

Step 3: Find Optimal Standard Normal Deviate ($z^*$): From standard normal cumulative distribution tables, find $z$ such that $\Phi(z) = 0.90$: $z = 1.282$

Step 4: Compute Optimal Stocking Quantity ($Q^*$): $Q^* = \mu + z \cdot \sigma = 2,000 + (1.282 \times 400) = 2,000 + 512.8 = 2,512.8 \approx 2,513\text{ units}$

Because the profit margin ($180) is nine times larger than the salvage penalty ($20), the firm aggressively builds 513 units beyond expected demand to ensure a 90% in-stock probability.