Lesson 31: The EOQ Model with No Shortages¶
31.1 Learning Outcomes¶
By the end of this hour a student will be able to:
- State the assumptions of the classical EOQ model and draw its sawtooth inventory profile.
- Derive \(Q^{*}=\sqrt{2C_oD/C_h}\) by calculus and by the equal-costs argument.
- Compute \(Q^*\), \(TC^*\), cycle time, order frequency and reorder point.
- Derive and use the sensitivity ratio \(\dfrac{TC(Q)}{TC^*}=\dfrac12\left(\dfrac{Q}{Q^*}+\dfrac{Q^*}{Q}\right)\).
- Explain why the EOQ is remarkably robust to errors in its inputs.
31.2 The Model and Its Assumptions¶
The Economic Order Quantity model, published by Ford W. Harris in 1913, is the oldest formula in operations research and still the most used.
Assumptions:
- Demand is known, constant and continuous at \(D\) units per unit time.
- No shortages are permitted; stock never runs out.
- Lead time is constant and known (often taken as zero).
- Replenishment is instantaneous — the whole order \(Q\) arrives at once.
- Costs \(C_o\), \(C_h\) and \(C\) are constant; there are no quantity discounts.
- A single item, analysed independently of all others.
The sawtooth diagram¶
stock
Q |\ \ \
| \ \ \
| \ \ \
| \ \ \ slope = -D
0 |____\___________\___________\_______ time
|<-t->|
Stock rises instantaneously to \(Q\) on delivery, falls linearly at rate \(D\), hits zero exactly as the next order arrives, and the cycle repeats.
Consequences read straight off the diagram:
The average of \(Q/2\) follows because the profile is a triangle: the area under one cycle is \(\tfrac12 Qt\), and dividing by \(t\) gives \(Q/2\).
31.3 Deriving the Optimum¶
The total cost function¶
Per unit time:
The purchase term \(DC\) does not involve \(Q\) and therefore cannot affect the optimum. It is customary to omit it and work with the relevant cost
Ordering cost falls like \(1/Q\); holding cost rises linearly in \(Q\). Their sum is a convex U-shape.
By calculus¶
Result 1 (Harris–Wilson EOQ formula). $\(\boxed{\ Q^{*}=\sqrt{\frac{2C_oD}{C_h}}\ }\tag{2.3}\)$ and since \(\dfrac{d^2TRC}{dQ^2}=\dfrac{2C_oD}{Q^3}>0\) for \(Q>0\), this is a global minimum.
By the equal-costs argument (no calculus needed)¶
At \(Q^*\), substitute (2.3) into the two cost terms:
Result 2. At the optimum, annual ordering cost exactly equals annual holding cost. This is the fastest possible check on any EOQ answer: if your two cost figures are not equal, something is wrong.
Setting \(\dfrac{C_oD}{Q}=\dfrac{C_hQ}{2}\) and solving gives (2.3) directly.
The minimum cost¶
Note the elegant forms: \(TRC^* = C_hQ^*\), and also \(TRC^* = 2C_oD/Q^*\).
The complete formula set¶
| Quantity | Formula |
|---|---|
| Optimal order quantity | \(Q^{*}=\sqrt{2C_oD/C_h}\) |
| Minimum relevant cost | \(TRC^{*}=\sqrt{2C_oDC_h}\) |
| Optimal cycle time | \(t^{*}=Q^{*}/D=\sqrt{2C_o/(DC_h)}\) |
| Optimal order frequency | \(n^{*}=D/Q^{*}=\sqrt{DC_h/(2C_o)}\) |
| Average inventory | \(Q^{*}/2\) |
| Reorder point | \(ROP=DL\) (modulo whole cycles) |
How the optimum responds to its inputs¶
Every quantity moves as a square root, which is the model's most important structural fact:
- Quadruple the demand and the order size only doubles — so inventory turnover improves with scale. Large firms hold proportionally less stock than small ones; this formula is why.
- Cutting the ordering cost by a factor of \(4\) (the aim of every JIT programme) cuts the batch size by half, not by a quarter. JIT is EOQ with \(C_o\) driven down, not a rejection of EOQ.
31.4 Sensitivity: Why EOQ Is Forgiving¶
Suppose you order \(Q\) rather than the optimum \(Q^*\). Write \(k=Q/Q^*\). Then
and substituting \(Q^*=\sqrt{2C_oD/C_h}\) gives
Result 3 (sensitivity ratio). $\(\boxed{\ \frac{TRC(Q)}{TRC^{*}}=\frac12\left(\frac{Q^{*}}{Q}+\frac{Q}{Q^{*}}\right)=\frac12\left(k+\frac1k\right)\ }\tag{3.1}\)$
| \(k=Q/Q^*\) | 0.5 | 0.8 | 0.9 | 1.0 | 1.1 | 1.25 | 1.5 | 2.0 |
|---|---|---|---|---|---|---|---|---|
| Cost penalty | +25 % | +2.5 % | +0.6 % | 0 | +0.5 % | +2.5 % | +8.3 % | +25 % |
The EOQ is extraordinarily flat near its optimum. Ordering \(20\,\%\) too much or too little costs only about \(2\,\%\) extra. The formula is symmetric in \(k\) and \(1/k\): doubling the order and halving it are equally bad.
Why this matters practically. The inputs \(D\), \(C_o\) and \(C_h\) are estimates, often poor ones. But an error of \(50\,\%\) in \(C_o\) moves \(Q^*\) by only \(\sqrt{1.5}=22\,\%\), which by (3.1) costs about \(2\,\%\). So the EOQ answer can be safely rounded to a convenient pack, pallet or lorry-load. This robustness — not the formula itself — is why EOQ has survived a century of criticism.
31.5 Solved Problems¶
Problem 1 — The standard calculation¶
A firm uses \(10{,}000\) units of a component a year. Placing an order costs \(₹150\); holding one unit for a year costs \(₹8\). The item costs \(₹40\) and the lead time is \(12\) days. The firm works \(250\) days a year. Find \(Q^*\), \(TRC^*\), the cycle time, the number of orders and the reorder point.
Solution.
Check by components:
Total annual cost including purchases: \(TC = 10000(40)+4898.98 = ₹404{,}898.98\).
Interpretation. Note that the purchase cost of \(₹4\) lakh dwarfs the \(₹4899\) of relevant cost. This is typical, and it is why management attention often sits in the wrong place: a \(1\,\%\) better purchase price saves \(₹4000\) a year, almost as much as perfecting the order quantity. Both are worth pursuing; only one is usually pursued.
Problem 2 — Holding cost given as a percentage¶
A dealer sells \(2400\) units a year of an item costing \(₹50\). The ordering cost is \(₹350\) per order and the carrying charge is \(20\,\%\) of the item's value per year. Find \(Q^*\), the number of orders, the cycle time and \(TRC^*\).
Solution.
Practical rounding. \(5.86\) orders a year is not implementable. Consider \(6\) orders of \(400\) units:
a penalty of \(₹1.22\) per year — \(0.03\,\%\). Round freely. Check with (3.1): \(k=400/409.88=0.9759\), and \(\tfrac12(0.9759+1.0247)=1.0003\) ✓.
Problem 3 — Sensitivity in action¶
For Problem 1, evaluate the cost of ordering \(306\), \(766\) and \(1225\) units instead of \(612\).
Solution. With \(Q^*=612.37\) and \(TRC^*=₹4898.98\):
| \(Q\) | \(k=Q/Q^*\) | \(TRC(Q)\) | Ratio | Penalty |
|---|---|---|---|---|
| 306.19 | 0.50 | ₹6123.72 | 1.2500 | +25.0 % |
| 489.90 | 0.80 | ₹5021.45 | 1.0250 | +2.5 % |
| 612.37 | 1.00 | ₹4898.98 | 1.0000 | 0 |
| 765.47 | 1.25 | ₹5021.45 | 1.0250 | +2.5 % |
| 918.56 | 1.50 | ₹5307.23 | 1.0833 | +8.3 % |
| 1224.74 | 2.00 | ₹6123.72 | 1.2500 | +25.0 % |
Verify one row by hand. At \(Q=918.56\):
and by (3.1): \(\tfrac12(1.5+0.6667)=1.0833\) ✓.
Interpretation. The two extreme rows are identical in penalty — \(k=0.5\) and \(k=2\) both cost \(25\,\%\) more. This symmetry in \(k \leftrightarrow 1/k\) is exact, not approximate, and it means a manager who is unsure whether demand is \(5000\) or \(20{,}000\) can order for \(10{,}000\) and be no more than \(25\,\%\) wrong in cost while being \(100\,\%\) wrong in the forecast.
Problem 4 — Working backwards from a policy¶
A firm currently orders \(1200\) units at a time. Annual demand is \(18{,}000\) units and the holding cost is \(₹6\) per unit per year. Management believes the current policy is optimal. What ordering cost does that imply, and if the true ordering cost is \(₹150\), how much is the firm losing?
Solution.
Step 1 — the implied \(C_o\). If \(1200\) were optimal,
Step 2 — the true optimum. With \(C_o=150\):
Step 3 — the loss.
Check with (3.1): \(k=1200/948.68=1.2649\), \(\tfrac12(1.2649+0.7906)=1.0277\), and \(5692.10\times1.0277=₹5849.99\) ✓.
Interpretation. Overstating \(C_o\) by \(60\,\%\) (\(240\) vs \(150\)) costs only \(2.8\,\%\) — under \(₹160\) a year. Two useful conclusions. First, the flatness again: it would not be worth a consultant's fee to fix this. Second, the diagnostic technique itself is valuable: inverting the EOQ formula to find what a firm's current policy implies about its costs frequently exposes an assumption nobody had noticed they were making.
Problem 5 — Testing the assumptions¶
For each situation say whether classical EOQ applies, and if not, which assumption fails.
| Situation | EOQ valid? | Which assumption fails |
|---|---|---|
| (a) Steel bar, steady use, one supplier, no discounts | Yes | — |
| (b) Supplier offers 5 % off on orders above 500 units | No | Assumption 5 (constant price) → Lesson 35 |
| (c) The firm makes the item itself at 500 units/day | No | Assumption 4 (instant replenishment) → Lesson 33 |
| (d) Backorders are acceptable and cost ₹20/unit/yr | No | Assumption 2 (no shortages) → Lesson 32 |
| (e) Demand is 500 ± 200 units per month | No | Assumption 1 (deterministic demand) → safety stock models |
| (f) Fresh produce spoiling in 4 days | No | Implicitly assumes no perishability → newsvendor |
| (g) Total warehouse capacity is 800 units across 5 items | No | Assumption 6 (single item) → constrained multi-item EOQ |
Interpretation. Every assumption that fails has a corresponding extension, and Lessons 32–35 build four of them. But note something important about (e): even when demand is uncertain, the EOQ is usually still used for the order quantity, with the uncertainty handled separately through safety stock. The reason is exactly Result 3 — the cost curve is so flat that using \(\bar D\) in the formula costs little, so it is not worth a more sophisticated batch-sizing rule. Robustness lets EOQ survive violations of its own assumptions.
31.6 Practice Problems¶
- \(D=1500\)/yr, \(C_o=₹200\), \(C_h=₹15\)/unit/yr. Find \(Q^*\), \(n^*\), \(t^*\), \(TRC^*\). (Answer: \(200\) units, \(7.5\) orders/yr, \(48.7\) days, \(₹3000\)/yr)
- \(D=8000\)/yr, \(C_o=₹500\), item costs \(₹120\), carrying rate \(25\,\%\). Find \(Q^*\) and \(TRC^*\). (Answer: \(C_h=₹30\); \(Q^*=516.40\); \(TRC^*=₹15{,}491.93\))
- Show that \(t^{*}=\sqrt{2C_o/(DC_h)}\) and evaluate it for Problem 1 of §4 in days. (Answer: \(0.06124\) yr \(=15.31\) working days on a 250-day year)
- A firm's EOQ is \(400\) units. What order quantity would raise total relevant cost by exactly \(10\,\%\)? (Answer: solve \(\tfrac12(k+1/k)=1.1\): \(k=1.5583\) or \(0.6417\), i.e. \(Q=623.3\) or \(256.7\))
- Demand is revised upward from \(4000\) to \(9000\) units a year. By what factor should \(Q^*\) change, and by what factor does \(TRC^*\) change? (Answer: both by \(\sqrt{9000/4000}=1.5\))
31.7 Summary¶
- EOQ assumes constant known demand, no shortages, instant replenishment, constant costs, one item.
- Read the sawtooth: \(t=Q/D\), \(n=D/Q\), average inventory \(=Q/2\).
- \(\boxed{Q^{*}=\sqrt{2C_oD/C_h}}\) and \(\boxed{TRC^{*}=\sqrt{2C_oDC_h}}\).
- At the optimum, ordering cost = holding cost. Use this as an instant check.
- The constant purchase cost \(DC\) never affects \(Q^*\) — but it usually dominates the total bill.
- Everything scales as a square root: quadrupling demand only doubles the order size, so inventory turnover improves with scale. JIT is EOQ with \(C_o\) driven down.
- Sensitivity: \(\dfrac{TRC(Q)}{TRC^*}=\dfrac12\left(k+\dfrac1k\right)\). A \(20\,\%\) error costs \(2\,\%\); halving and doubling are equally bad at \(+25\,\%\) each.
- That flatness is the model's real contribution: round the answer to a convenient pack size and do not agonise over the inputs.
Next lesson: L32 — EOQ with planned shortages (the backorder model).
Practise now: unit3-queue-inventory-lab.html → EOQ Explorer: drag \(D\), \(C_o\) and \(C_h\), watch
the sawtooth and the two cost curves cross, and see the flat bottom of the total-cost curve.