Forecast Accuracy Metrics - Business Analytics

1. Which model and justification best align with the manager's objective?
   Answer: Choose Model A because its MAE is lower, even though its MAPE is higher. Whenever you see a forecasting question, your first move should be to identify what the manager is actually trying to minimize — then match that goal to the right metric. Here, the manager wants to minimize misallocated units, which is a raw, unit-based concern. That points directly to MAE (Mean Absolute Error), not MAPE. Let's run the numbers. Model A has absolute errors of $40$ and $40$, giving $\text{MAE}_A = \frac{40+40}{2} = 40$. Model B has errors of $10$ and $100$, giving $\text{MAE}_B = \frac{10+100}{2} = 55$. So Model A misallocates fewer units on average — exactly what the manager cares about. For MAPE, Model A produces $\frac{40\%+4\%}{2} = 22\%$, while Model B produces $\frac{10\%+10\%}{2} = 10\%$. Model B wins on MAPE, but MAPE measures relative error, not unit volume. Answer A is correct: Model A's MAE is lower ($40$ vs. $55$), which directly addresses the manager's unit-based objective, even though its MAPE is higher. Answer B is wrong because Model B does not have a lower MAE — $55 > 40$ — so the claim that both metrics favor Model B is factually false. Answer C gets the metric backwards: a lower MAPE is irrelevant when the goal is minimizing misallocated units, and it also picks the wrong model for a unit-based objective. Answer D is an absolute overstatement. Neither metric universally replaces the other — context determines which is appropriate. Study tip: Always anchor your metric choice to the manager's stated goal. Unit goals → MAE; proportional/relative goals → MAPE.
2. Which conclusion is supported by MAE and MAPE for these two models?
   Answer: The models tie on both metrics, although Model X shows consistent underforecasting that the metrics do not reveal. When comparing forecasting models, MAE (Mean Absolute Error) and MAPE (Mean Absolute Percentage Error) both average the absolute values of errors, meaning positive and negative errors do not cancel — they accumulate equally. Run the numbers for both models. Model X has errors of $-10$ and $-10$, giving $\text{MAE} = \frac{10+10}{2} = 10$ and $\text{MAPE} = \frac{10\%+10\%}{2} = 10\%$. Model Y has errors of $-10$ and $+10$, giving $\text{MAE} = \frac{10+10}{2} = 10$ and $\text{MAPE} = \frac{10\%+10\%}{2} = 10\%$. Both metrics are identical across models, confirming that B is correct: the models tie on both MAE and MAPE. However, Model X consistently underforecasts by 10 units each week — a bias that neither metric exposes, since they measure magnitude, not direction. Choice A describes how algebraic (signed) errors cancel, which applies to ME or bias calculations — not MAE or MAPE, which use absolute values. Choice C is simply false; MAPE weights errors as a percentage of actual demand, and since actual demand is the same (100 units) in both weeks for both models, no asymmetric penalty exists. Choice D claims Model Y has lower MAE because it is "unbiased overall," but being unbiased (errors summing to zero) is irrelevant to MAE — what matters is the size of each absolute error, which is identical here. The key study tip: never confuse bias metrics (which use signed errors) with accuracy metrics like MAE and MAPE (which use absolute errors). A model can be perfectly unbiased yet have the same or worse MAE than a biased one.
3. Which treatment is most defensible when reporting conventional MAE and MAPE for this evaluation set?
   Answer: Retain the day's absolute error in MAE, but do not report conventional MAPE without a stated zero-value rule. When evaluating forecast accuracy metrics, you need to understand what each metric can and cannot mathematically handle. MAE averages absolute differences — it has no division, so it works for any actual value including zero. MAPE, however, divides by the actual value: $\text{MAPE} = \frac{1}{n}\sum \left|\frac{A_t - F_t}{A_t}\right| \times 100$. When $A_t = 0$, that division is undefined, making conventional MAPE mathematically impossible to compute for that observation. This is exactly why C is the most defensible choice. The absolute error for the zero-actual day is $|0 - 5| = 5$, which is perfectly valid and should be retained in MAE. But because MAPE requires dividing by the actual value, you simply cannot include that day in a conventional MAPE calculation without making an arbitrary methodological decision — and any such decision must be explicitly stated and justified. A is wrong because assigning $0\%$ error implies the forecast was perfect, which is false — the model missed by 5 cancellations. This actively misrepresents accuracy. B is wrong for the opposite reason: $100\%$ is an arbitrary convention with no mathematical grounding; it would distort the MAPE by injecting a number that wasn't computed, just assumed. D is wrong because it overcorrects — excluding the day from MAE throws away valid, computable information. The problem is MAPE-specific, not universal to all metrics. Your study tip: always ask yourself whether a metric involves division by the actual value. If it does, zero actuals are a red flag requiring either exclusion with disclosure or an alternative metric like sMAPE.
4. How does the dashboard result compare with MAPE calculated directly across all four monthly observations?
   Answer: The dashboard produces $20$, while the observation-level calculation produces $15$. Whenever you see a question involving averages of averages, pause and ask yourself: are all underlying observations getting equal weight? That's exactly the trap being tested here. Start with the observation-level MAPE, calculated directly across all four months. You have three errors of $10\%$ and one error of $30\%$, so: $\frac{10 + 10 + 10 + 30}{4} = \frac{60}{4} = 15\%$. Each individual observation contributes equally, and the three Region A observations naturally dominate because there are more of them. Now calculate the dashboard result. Region A's MAPE is $\frac{10+10+10}{3} = 10\%$. Region B's MAPE is $\frac{30}{1} = 30\%$. The dashboard then averages these two regional MAPEs equally: $\frac{10 + 30}{2} = 20\%$. By averaging at the region level first, you've given Region B — which has only one observation — the same influence as Region A, which has three. That inflates the result. So the dashboard produces $20\%$ while the direct calculation produces $15\%$, confirming answer C. Answer A is wrong because equal weighting of regions is not the same as equal weighting of observations — that's the whole point of the question. Answer B has the two values correct but reverses which method produces which number. Answer D is wrong because the dashboard explicitly does not give all percentage errors equal weight; it over-weights Region B's single observation. Study tip: When you see "average of averages," always check whether group sizes differ. Unequal group sizes make the two approaches diverge — a classic trap in business analytics questions on MAPE and similar metrics.
5. Assuming no observations are added or removed, how should the two metrics appear on the new dashboard?
   Answer: MAE should become $24$, while MAPE should remain $8$. When a question asks how metrics behave under a unit change, the key is distinguishing between absolute metrics (which scale with units) and relative/ratio metrics (which are unit-free). MAE (Mean Absolute Error) is an absolute metric — it's the average of raw errors measured in the same units as your data. When you convert values from dollars to thousands of dollars, every actual and forecast value is divided by 1,000, so every error term is also divided by 1,000. An MAE of $24{,}000$ dollars becomes $\frac{24{,}000}{1{,}000} = 24$ (thousands of dollars). MAPE (Mean Absolute Percentage Error), by contrast, expresses each error as a percentage of the actual value — the units cancel out in that ratio. Whether you measure in dollars or thousands of dollars, the ratio $\frac{|\text{error}|}{\text{actual}}$ is identical. So MAPE stays at $8\%$. That makes D correct. Answer A is wrong on both counts: it incorrectly leaves MAE unchanged (ignoring the unit rescaling) and incorrectly scales MAPE (which needs no adjustment). Answer B gets MAE right but still wrongly adjusts MAPE — dividing $8\%$ by 1,000 to get $0.008\%$ confuses a percentage ratio with an absolute number. Answer C correctly leaves MAPE alone but fails to rescale MAE, as if the unit change never happened. A useful rule of thumb: any metric with "%" in its name is typically scale-invariant, while any metric expressed in the same units as your data will rescale proportionally with those units.
6. Which statement correctly compares the models?
   Answer: Model A has lower MAE, while Model B has lower MAPE because relative errors receive equal weight. When comparing forecast models, you need to know that MAE (Mean Absolute Error) and MAPE (Mean Absolute Percentage Error) measure accuracy differently — MAE uses raw units, while MAPE uses percentages, giving equal weight to relative errors regardless of volume. Start by computing the absolute errors for each model: - Model A: $|2-0| = 2$ (specialty) and $|200-180| = 20$ (standard) → $\text{MAE} = \frac{2+20}{2} = 11$ - Model B: $|2-1| = 1$ (specialty) and $|200-160| = 40$ (standard) → $\text{MAE} = \frac{1+40}{2} = 20.5$ So Model A has the lower MAE. Now check MAPE: - Model A: $\frac{2/2 + 20/200}{2} = \frac{1.00 + 0.10}{2} = 55\%$ - Model B: $\frac{1/2 + 40/200}{2} = \frac{0.50 + 0.20}{2} = 35\%$ Model B has the lower MAPE. This confirms D — Model A wins on MAE (dominated by the high-volume part's raw error), while Model B wins on MAPE (penalizes the 100% miss on the specialty part equally with the standard part). Choice A is wrong because while Model A has lower MAE, its MAPE is worse. Choice B is wrong because a smaller specialty-part error doesn't determine overall model superiority on either metric. Choice C reverses the conclusion — it claims Model B has lower MAE, which the calculations disprove. Your study tip: whenever a question mixes high-volume and low-volume products, expect MAE and MAPE to tell different stories — that divergence is almost always the point being tested.
7. What is the most appropriate interpretation for choosing between the models?
   Answer: MAE and MAPE tie, so the retailer should supplement them with an asymmetric business-cost measure. Whenever you see a question involving forecast accuracy metrics alongside asymmetric business costs, your first instinct should be to calculate the metrics carefully — then ask whether those metrics actually capture the cost structure the business faces. Here, both models produce an average absolute error of $\frac{10+10}{2} = 10$ units, and both produce a MAPE of $\frac{10\%+10\%}{2} = 10\%$. Model X misses by $-10$ in both periods (always underforecasts). Model Y misses by $-10$ then $+10$ (one under, one over). Since MAE and MAPE treat all errors by their absolute magnitude — ignoring direction — both models score identically on both metrics. That tie means neither metric alone is sufficient for the retailer's decision. Because stockouts (underforecasts) are substantially more costly than overforecasts, the retailer needs an asymmetric cost measure — one that penalizes negative errors more heavily — to distinguish the models. That logic confirms A as correct. B is wrong because MAE does not allow overforecasts to offset underforecasts — it uses absolute values, so cancellation is impossible by definition. C is wrong because MAPE also uses absolute percentage errors; it contains no mechanism to incorporate stockout costs or directional bias. D is wrong on two counts: neither metric actually selects Model X (they tie), and "less variable" errors are not what MAE or MAPE measure — variability of errors is a separate concept entirely. Your study tip: when a question mentions asymmetric costs, that's a signal that standard symmetric metrics (MAE, MSE, MAPE) are insufficient. Always check whether the business context demands a directional or cost-weighted error measure.
8. How will this translation affect MAE and MAPE?
   Answer: MAE will remain $2$, while MAPE will decrease to approximately $1.74$. Whenever you see a question about forecast error metrics, ask yourself: what does each metric actually measure, and what happens to that measurement when the data shifts? MAE (Mean Absolute Error) measures the average absolute difference between actuals and forecasts. Here, the errors are $|10-12| = 2$ and $|20-18| = 2$, giving $\text{MAE} = 2$. After adding $100$ to both actuals and forecasts, the errors become $|110-112| = 2$ and $|120-118| = 2$. The differences are identical — adding the same constant to both sides cancels out — so **MAE stays at $2$**. MAPE (Mean Absolute Percentage Error) divides each absolute error by its actual value. Originally: $\frac{2}{10} = 20\%$ and $\frac{2}{20} = 10\%$, averaging to $15\%$. After translation, the actuals become $110$ and $120$, so: $\frac{2}{110} \approx 1.82\%$ and $\frac{2}{120} \approx 1.67\%$, averaging to approximately $1.74\%$. Because the denominator grows while the errors stay fixed, MAPE decreases. This confirms answer C. Answer A incorrectly adds $100$ to the MAE itself, confusing the error metric with the values. Answer B correctly fixes MAE but assumes MAPE is scale-invariant — it isn't, because the percentage is anchored to the actual values. Answer D gets MAPE right but incorrectly shrinks MAE, misunderstanding that constant shifts cancel in subtraction. The key study tip: MAE is translation-invariant (constant shifts cancel), but MAPE is not — larger actuals produce smaller percentage errors even when absolute errors are unchanged. Watch for questions that exploit this asymmetry.
9. What should be reported if MAE and MAPE are calculated only from observations for which actual outcomes are available?
   Answer: MAE is $20$ units, and MAPE is $12.5$. When calculating forecast accuracy metrics like MAE and MAPE, you can only use periods where actual demand is known — incomplete observations must be excluded entirely. Here, only Weeks 1 and 2 qualify; Week 3 has no actual demand yet. Start by finding the absolute errors for each available week. Week 1: $|100 - 90| = 10$ units. Week 2: $|200 - 230| = 30$ units. MAE is the average of these: $\frac{10 + 30}{2} = 20$ units, confirming the first part of answer B. For MAPE, divide each absolute error by its actual demand, then average: $\frac{10}{100} = 10\%$ and $\frac{30}{200} = 15\%$. Averaging those: $\frac{10\% + 15\%}{2} = 12.5\%$, confirming B entirely. Answer A incorrectly averages over three periods (dividing by 3 instead of 2), as if Week 3 were a valid observation — a classic "don't include unknowns" trap. Answer C correctly computes MAE as 20 units but reports MAPE as 15%, which is only Week 2's individual percentage error, not the two-week average. Answer D appears to use incorrect error values altogether, possibly confusing the forecast with actual values or applying a wrong formula entirely. The key study tip here: always identify which observations are complete before plugging into any accuracy formula. On exam questions involving rolling forecasts, watch for incomplete periods designed to tempt you into using all rows in the table — only rows with both a forecast and an actual belong in your calculation.
10. What are the forecast's MAE and MAPE across the two weeks?
    Answer: MAE is $15$ units, and MAPE is exactly $20$. When evaluating forecast accuracy, two of the most common metrics are MAE (Mean Absolute Error) and MAPE (Mean Absolute Percentage Error). MAE measures average absolute deviation in units; MAPE expresses that same deviation as a percentage of actual demand. Knowing both formulas cold is essential here. Start by finding each period's absolute error. Week 1: $|50 - 40| = 10$ units. Week 2: $|100 - 120| = 20$ units. The MAE is simply the average: $\frac{10 + 20}{2} = 15$ units. For MAPE, divide each absolute error by its actual demand, then average those percentages. Week 1: $\frac{10}{50} = 20\%$. Week 2: $\frac{20}{100} = 20\%$. Average: $\frac{20\% + 20\%}{2} = 20\%$ exactly. That confirms B is correct — MAE of 15 units and MAPE of exactly 20%. Choice A gets the MAE right but reports MAPE as approximately 20.83%, which would result from incorrectly averaging the raw errors over total demand ($\frac{30}{144} \approx 20.83\%$) rather than averaging the per-period percentage errors. Choice C reports MAE as 30, which is the sum of errors, not the average — a classic mistake. MAPE of 40% compounds that error by also summing the percentages without averaging. Choice D fabricates both values with no clear calculation path. A reliable study tip: always average after computing per-period metrics, not before. MAE = average of absolute errors; MAPE = average of absolute percentage errors. Summing without dividing is the most common trap on these calculations.

Start this quiz