Lesson 32: EOQ with Shortages Permitted (the Backorder Model)¶
32.1 Learning Outcomes¶
By the end of this hour a student will be able to:
- Draw the inventory profile when planned backorders are allowed and identify \(S\), \(B\) and \(Q\).
- Derive the average holding and shortage levels from the geometry of that profile.
- Derive \(Q^{*}\) and \(S^{*}\) for the backorder model and interpret the backorder factor.
- Show that the model reduces to classical EOQ as \(C_s\to\infty\).
- Decide whether a planned-shortage policy is appropriate for a given business.
32.2 Why Plan to Run Out?¶
Deliberately allowing stock-outs sounds like bad management. It is not, provided customers are willing to wait — that is, the demand is backordered rather than lost.
The logic: if a customer is content to wait a few days, the firm can carry less stock and save holding cost, paying instead a smaller shortage cost. Whenever \(C_s\) is genuinely finite and comparable to \(C_h\), the optimal policy is to run out for part of each cycle.
The essential distinction.
- Backorders — unfilled demand waits and is satisfied from the next delivery. The sale is kept; only a penalty is paid. This lesson's model.
- Lost sales — the customer goes elsewhere. The sale, and possibly the customer, is gone. Shortage cost is then per unit short, not per unit per unit time, and the model differs.
Backordering is realistic for made-to-order machinery, spare parts, books and furniture. It is unrealistic for groceries, petrol and medicines.
32.3 The Inventory Profile¶
stock
S |\ |\ |\
| \ | \ | \
| \ | \ | \
0 |---\----------|---\----------|---\------- time
| \ /| \ /|
-B | \______/ | \______/ |
|<-t1->|<-t2->|
|<------ t ---->|
Notation:
| Symbol | Meaning |
|---|---|
| \(Q\) | order quantity (the whole delivery) |
| \(S\) | maximum inventory level — what remains after clearing backorders |
| \(B=Q-S\) | maximum shortage (backorder) level |
| \(t_1=S/D\) | time with stock on hand |
| \(t_2=B/D\) | time in shortage |
| \(t=Q/D\) | full cycle |
When a delivery of \(Q\) arrives, \(B\) units go immediately to waiting customers and \(S=Q-B\) enter stock.
Average inventory and average shortage¶
Both come from triangle areas divided by the full cycle \(t\) — this is the step students most often get wrong, because each average is taken over the whole cycle, not over the sub-interval in which it is non-zero.
Setting \(B=0\) recovers \(S=Q\) and average stock \(=Q/2\), the classical result — a useful check.
32.4 Deriving the Optimum¶
The cost function¶
This is a function of two variables, so we need both partial derivatives.
Optimising over S¶
Result 1. $\(\boxed{\ S^{*}=\frac{C_s}{C_h+C_s}\,Q\ },\qquad B^{*}=Q-S^{*}=\frac{C_h}{C_h+C_s}\,Q \tag{3.2}\)$
The fraction \(\dfrac{C_s}{C_h+C_s}\) is the critical ratio or service level: the proportion of the cycle spent in stock. It has an immediate reading — if shortage is ten times as expensive as holding, \(S^*=\tfrac{10}{11}Q\) and the system is in stock \(91\,\%\) of the time.
Optimising over Q¶
Substituting (3.2) into (3.1) and simplifying, the cost becomes
which is an ordinary EOQ with \(C_h\) replaced by the harmonic-type combination \(\dfrac{C_hC_s}{C_h+C_s}\). Hence
Result 2 (EOQ with backorders). $\(\boxed{\ Q^{*}=\sqrt{\frac{2C_oD}{C_h}}\ \sqrt{\frac{C_h+C_s}{C_s}}\ =\ Q^{*}_{\text{EOQ}}\cdot k, \qquad k=\sqrt{\frac{C_h+C_s}{C_s}}\ }\tag{3.3}\)$ $\(\boxed{\ S^{*}=\sqrt{\frac{2C_oD}{C_h}}\ \sqrt{\frac{C_s}{C_h+C_s}}=\frac{Q^{*}_{\text{EOQ}}}{k}\ }\tag{3.4}\)$ $\(\boxed{\ TRC^{*}=\sqrt{2C_oDC_h}\ \sqrt{\frac{C_s}{C_h+C_s}}=\frac{TRC^{*}_{\text{EOQ}}}{k}\ }\tag{3.5}\)$
Reading the results¶
The single factor \(k=\sqrt{\dfrac{C_h+C_s}{C_s}}>1\) governs everything:
| Quantity | Effect of allowing backorders |
|---|---|
| Order quantity \(Q^*\) | multiplied by \(k\) — order more, less often |
| Maximum stock \(S^*\) | divided by \(k\) — hold less |
| Total cost \(TRC^*\) | divided by \(k\) — always cheaper |
Backordering is never worse than the classical policy, because the classical policy (\(B=0\)) is available and was not chosen. The saving is the factor \(1/k\).
Two limits worth checking:
- \(C_s\to\infty\) (shortages intolerable): \(k\to1\), so \(Q^*\to Q^*_{\text{EOQ}}\), \(S^*\to Q^*\), \(B^*\to0\). The model collapses to Lesson 31, exactly as it must.
- \(C_s\to0\) (shortages free): \(k\to\infty\), \(Q^*\to\infty\), \(S^*\to0\). Order enormous quantities and hold nothing — degenerate, and a warning that a low \(C_s\) estimate produces absurd advice.
32.5 Solved Problems¶
Problem 1 — The standard backorder calculation¶
A firm uses \(10{,}000\) units a year. Ordering costs \(₹150\); holding costs \(₹8\) per unit per year; backordering costs \(₹25\) per unit per year. Find \(Q^*\), \(S^*\), \(B^*\), the cost breakdown and the saving over the no-shortage policy.
Solution.
Step 1 — the backorder factor.
Step 2 — the classical EOQ as a base. From Lesson 31, \(Q^*_{\text{EOQ}}=612.37\), \(TRC^*_{\text{EOQ}}=₹4898.98\).
Step 3 — the backorder optimum.
Step 4 — cost breakdown (an essential check).
Note that here ordering cost equals holding + shortage cost (\(2132.01 = 1615.16+516.85\)) — the generalisation of the equal-costs check from Lesson 31.
Step 5 — the saving.
Interpretation. The optimal policy runs out of stock for \(t_2/t=B^*/Q^*=170.56/703.56=24.2\,\%\) of every cycle. That is a deliberate, costed decision, not a failure. The saving of \(₹635\) a year is genuine — provided customers really do wait. If even a quarter of the backordered demand were actually lost at a \(₹40\) margin, the lost contribution would be \(0.25\times170.56\times14.21\text{ cycles}\times ₹40 \approx ₹24{,}000\) a year, dwarfing the saving. Verify the backorder assumption before trusting this model.
Problem 2 — A second calculation with different economics¶
\(D=1200\) units/yr, \(C_o=₹100\), \(C_h=₹5\)/unit/yr, \(C_s=₹15\)/unit/yr. Find the full policy.
Solution.
Breakdown: ordering \(₹474.34\), holding \(₹355.76\), shortage \(₹118.59\); total \(₹948.69\) ✓.
Cycle structure. \(t=Q/D=252.98/1200=0.2108\) yr \(=77\) days, of which \(t_1=189.74/1200=57.7\) days in stock and \(t_2=63.25/1200=19.2\) days short. The firm is out of stock for 19 days in every 77 — one quarter of the time — by design.
Problem 3 — How the policy responds to C_s¶
For Problem 2's data, tabulate the policy as \(C_s\) varies, and comment.
Solution. \(Q^{*}_{\text{EOQ}}=219.09\), \(TRC^{*}_{\text{EOQ}}=₹1095.45\) throughout.
| \(C_s\) | \(k\) | \(Q^{*}\) | \(S^{*}\) | \(B^{*}\) | Fraction of cycle short | \(TRC^{*}\) | Saving |
|---|---|---|---|---|---|---|---|
| 1 | 2.4495 | 536.66 | 89.44 | 447.21 | 83.3 % | ₹447.21 | 59.2 % |
| 5 | 1.4142 | 309.84 | 154.92 | 154.92 | 50.0 % | ₹774.60 | 29.3 % |
| 15 | 1.1547 | 252.98 | 189.74 | 63.25 | 25.0 % | ₹948.68 | 13.4 % |
| 45 | 1.0541 | 230.94 | 207.85 | 23.09 | 10.0 % | ₹1039.23 | 5.1 % |
| 495 | 1.0050 | 220.19 | 217.99 | 2.20 | 1.0 % | ₹1089.95 | 0.5 % |
| \(\infty\) | 1.0000 | 219.09 | 219.09 | 0 | 0 % | ₹1095.45 | 0 |
Interpretation.
- The fraction of the cycle spent short is exactly \(\dfrac{C_h}{C_h+C_s}\) — read straight from (3.2). At \(C_s=C_h=5\) the system is short half the time.
- The saving is substantial only when \(C_s\) is comparable to or smaller than \(C_h\). Once \(C_s>10C_h\) the model is barely distinguishable from classical EOQ.
- This is a strong practical guide. If a stock-out costs more than about ten times the holding cost — which it usually does in retail — do not bother with the backorder model. Its complexity buys under \(5\,\%\).
Problem 4 — Inferring C_s from a chosen service level¶
A firm cannot estimate \(C_s\) but insists on being in stock \(95\,\%\) of the time. With \(C_h=₹8\), \(C_o=₹150\) and \(D=10{,}000\), what \(C_s\) does that imply, and what policy follows?
Solution. From (3.2), the in-stock fraction is \(\dfrac{S^*}{Q^*}=\dfrac{C_s}{C_h+C_s}\). Setting this to \(0.95\):
Interpretation — why this inversion is the professional approach. The firm never had to name a shortage cost. Instead it named a service level, which managers can discuss meaningfully, and the model reports the implied \(C_s=₹152\) per unit-year. The conversation then becomes: "Your 95 % target is equivalent to valuing a stock-out at ₹152 per unit per year — about 19 times the holding cost. Is that right?" That question can be answered; "what is your shortage cost?" usually cannot.
The policy saves only \(₹124\) a year over pure EOQ — consistent with Problem 3's finding that a high implied \(C_s\) leaves little on the table.
Problem 5 — When the model must not be used¶
A pharmacy considers a planned-shortage policy for a life-saving drug, arguing that \(C_h\) is high (refrigeration) and patients will wait for the next delivery. Critique this.
Solution.
The model is inapplicable, for four separate reasons.
- Demand is not backordered. A patient needing the drug today cannot wait a week. The demand is lost (or worse), so the shortage cost is not "per unit per unit time" but a step function of catastrophic size.
- \(C_s\) is not commensurable with money. Any finite \(C_s\) produces a \(B^*>0\); the model structurally cannot recommend zero shortage unless \(C_s=\infty\). A model whose form forbids the right answer must not be used to justify the wrong one.
- Demand is probabilistic, violating Lesson 31's assumption 1. Even a policy of never planning shortages would need safety stock.
- Regulatory and ethical constraints are hard constraints, not costs to be traded off.
What to do instead. Formulate as a service-level problem: choose the stock policy that meets demand with probability \(\ge99.9\,\%\), then minimise cost subject to that constraint. The constraint expresses the ethical requirement; optimisation happens inside it.
The general principle. An optimisation model trades off whatever you put in its objective. If something must not be traded off, it belongs in the constraints, not the objective. Recognising that distinction is the mark of a competent modeller — and misplacing it is how quantitative methods acquire a bad name.
32.6 Practice Problems¶
- \(D=2000\)/yr, \(C_o=₹250\), \(C_h=₹10\), \(C_s=₹40\). Find \(Q^*\), \(S^*\), \(B^*\), \(TRC^*\). (Answer: \(353.55\), \(282.84\), \(70.71\), \(₹2828.43\))
- Show that \(S^{*}/Q^{*}\) is independent of \(D\) and \(C_o\), and explain why that makes sense.
- For Problem 1 of §4, how long is the firm out of stock in each cycle, in days on a 250-day year? (Answer: \(t=17.59\) days, \(t_2=4.26\) days)
- A firm's data give \(Q^{*}=500\) and \(S^{*}=400\). What is \(C_s/C_h\)? (Answer: \(S/Q=0.8=C_s/(C_h+C_s)\), so \(C_s=4C_h\))
- Prove that \(TRC^{*}_{\text{backorder}} \le TRC^{*}_{\text{EOQ}}\) for all \(C_s>0\), with equality only in the limit \(C_s\to\infty\).
32.7 Summary¶
- Planned shortages are rational only when demand is genuinely backordered, not lost.
- Averages are taken over the whole cycle: average stock \(=S^2/2Q\), average shortage \(=(Q-S)^2/2Q\).
- \(TRC(Q,S)=\dfrac{C_oD}{Q}+\dfrac{C_hS^2}{2Q}+\dfrac{C_s(Q-S)^2}{2Q}\), optimised in two variables.
- Critical ratio: \(\dfrac{S^*}{Q^*}=\dfrac{C_s}{C_h+C_s}\) — the fraction of the cycle in stock, and a directly interpretable service level.
- With \(k=\sqrt{\dfrac{C_h+C_s}{C_s}}\): \(\ Q^*=kQ^*_{\text{EOQ}}\), \(\ S^*=Q^*_{\text{EOQ}}/k\), \(\ TRC^*=TRC^*_{\text{EOQ}}/k\). Order more, hold less, pay less.
- As \(C_s\to\infty\) the model reduces exactly to classical EOQ — the required consistency check.
- The saving matters only when \(C_s\) is within about an order of magnitude of \(C_h\). Beyond that, use plain EOQ.
- When shortage cost cannot be priced, invert the model: choose a service level and report the implied \(C_s\).
- Requirements that must never be traded off belong in the constraints, never in the objective.
Next lesson: L33 — The production (finite replenishment) model without shortages.
Practise now: unit3-queue-inventory-lab.html → Backorder Lab: drag \(C_s\) and watch \(S^*\), \(B^*\)
and the shaded shortage region of the sawtooth respond.