| Course | MT438 Analytics in the Digital Supply Chain |
|---|---|
| Unit | Unit 5 |
| Paper type | Forecast accuracy analysis |
| Length | About 960 words, 6 pages double-spaced |
| Format | APA 7 student paper |
| School | Purdue University Global |
| Program | BS in Business Administration |
| Updated | October 2026 |
Free sample paper for MT438 Unit 5
Convincing Is Not Accurate: An MT438 Unit 5 Forecast Accuracy Analysis of 26 Weeks of Grass Seed
[Student Name]
Purdue University Global
MT438: Analytics in the Digital Supply Chain
Unit 5 Forecast Accuracy Analysis
[Instructor Name]
[Date]
Oakhurst Lawn and Garden, its product and every weekly figure are composite; the sources cited are real.
Error Convention
Throughout this analysis, error means actual shipments minus forecast. When that difference is above zero the forecast ran low; below zero, it ran high. Bias and tracking signals carry the same sign, so a negative bias always means over-forecasting.
The Product and the Season
Oakhurst's 20-pound sun-and-shade grass seed is its largest spring product. Over 26 weeks, January 5 through June 28, 2026, actual shipments totaled 119,841 bags. The season is sharply peaked: about 650 bags a week in early January, rising through March to 14,455 bags in week 13, the week of March 30, and falling back to about 635 by late June. About 71 percent of the half-year's volume ships in the eight weeks from week 9 through week 16.
Three Methods
Method A repeats the same week of 2025, scaled to this year's plan. Last spring came early, so A expects the peak in week 11, two weeks before it arrived. Method B scales a three-year average seasonal profile, the share of half-year volume each week has carried, to the annual plan of about 120,000 bags. Method C is the sales team's consensus: B raised 12 percent across the season, plus another 900 bags a week through the peak weeks 9 to 16, reflecting the team's belief that a new store program would lift spring sales. All three forecasts were frozen on December 15, 2025, so none had the benefit of seeing actual sales.
Four Measures, Defined Before Use
Mean absolute deviation (MAD) averages the weekly misses in bags with their signs dropped. Mean absolute percentage error (MAPE) averages each week's error as a percentage of that week's actual shipments, so every week counts equally. Weighted absolute percentage error divides total absolute error by total actual shipments, so busy weeks count in proportion to their volume. Bias is total error divided by total actual shipments, keeping the sign, and shows direction. Hyndman and Koehler (2006) caution that no single measure suits every purpose, which is why all four are reported.
Results
Method A: MAD 1,840 bags a week, MAPE 45.7 percent, weighted error 39.9 percent and bias plus 6.1 percent. Method B: MAD 362 bags, MAPE 7.8 percent, weighted error 7.9 percent and bias minus 0.2 percent. Method C: MAD 850 bags, MAPE 13.6 percent, weighted error 18.4 percent and bias minus 18.2 percent. Across the season, C forecast 141,690 bags against 119,841 shipped, about 21,850 bags too many.
The consensus forecast about 21,850 more bags than shoppers bought, and almost all of that excess fell in the busiest eight weeks.
Why MAPE Flatters the Consensus
For Method C, weighted error, 18.4 percent, is well above MAPE, 13.6 percent. MAPE gives a quiet January week of 650 bags the same vote as a peak week of 14,455. The consensus was close in the quiet weeks, where it simply copied B, and badly high in the peak weeks, where the 900-bag addition and the 12 percent lift compounded. Because most of its error sits in high-volume weeks, MAPE understates how much stock C would have sent to the wrong place. For a product whose cost is driven by peak-week production, weighted error is the more relevant figure.
Twenty-Two Weeks High
The consensus exceeded actual shipments in 22 of 26 weeks. In the four weeks it ran low, weeks 2, 21, 23 and 25, it missed by just 31 bags each. A forecast that is wrong in one direction nearly every week is not unlucky; it is biased.
Tracking Signals
A tracking signal takes the cumulative error to date and scales it by the MAD to date. Trigg (1964) proposed monitoring forecasts this way so that a persistent bias could be detected before it accumulated, and limits of plus or minus 4 are a common rule of thumb. Method C's signal passes minus 4 in week 5, at minus 4.3, and never recovers, ending the season at minus 25.7. Had Oakhurst monitored it, the consensus would have been flagged in early February, well before the peak.
Method A's signal behaves differently. It falls steadily to minus 12.0 by week 12, as A's early peak arrives and actual shipments lag, then climbs to plus 4.0 by week 26 as the late season outruns last year. That swing marks a timing miss rather than a steady bias: A had roughly the right volume in the wrong weeks. Method B's signal stays between minus 2.2 and plus 3.8 all season.
Interpretation
The three methods fail in different ways. A fails on timing: copying one year's season imports that year's weather. C fails on bias: a judgment layered on a sound forecast pushed every peak week high. B fails on neither. Fildes et al. (2009) found in their study of supply chain forecasting that large judgmental adjustments tend to improve accuracy when they rest on specific information, while optimistic upward adjustments frequently make forecasts worse; the consensus here fits the second pattern, since the store program it anticipated added fewer than 40 stores.
Recommendation
Oakhurst should keep Method B as the base forecast for this product. Overrides should remain possible but should name their reason, such as a confirmed promotion or a counted change in stores, and should be recorded so that each override's own accuracy can be tracked. A tracking signal with limits of plus or minus 4 should run weekly on every forecast and trigger a review when it is breached.
References
Fildes, R., Goodwin, P., Lawrence, M., & Nikolopoulos, K. (2009). Effective forecasting and judgmental adjustments: An empirical evaluation and strategies for improvement in supply-chain planning. International Journal of Forecasting, 25(1), 3-23. https://doi.org/10.1016/j.ijforecast.2008.11.010
Hyndman, R. J., & Koehler, A. B. (2006). Another look at measures of forecast accuracy. International Journal of Forecasting, 22(4), 679-688. https://doi.org/10.1016/j.ijforecast.2006.03.001
Trigg, D. W. (1964). Monitoring a forecasting system. Journal of the Operational Research Society, 15(3), 271-274. https://doi.org/10.1057/jors.1964.48
MT438 Unit 5 instructions, in plain terms
The Unit 5 assignment in MT438 compares forecasting methods on real or supplied demand and calls for a judgment about which to trust. State one error convention at the start, usually actual minus forecast, and keep it throughout so every sign reads the same. Describe each method well enough to reproduce, including when it was frozen. Define each accuracy measure before using it: an absolute measure in units, a percentage measure, a volume-weighted measure and bias. Report results for every method on every measure, then explain where measures disagree, because that disagreement is usually the most instructive part. Tracking signals over time show whether a method's error is steady bias or a timing miss. Close by naming the method you would keep and how its error will be watched each week. A table and a chart are common, though the reasoning around them carries the marks.
Inside the MT438 Unit 5 example
The Oakhurst sample fixes its sign convention in a two-sentence opening section. The product and its peaked season come next, with 71 percent of volume in eight weeks. The three methods are described with their freeze date. Four measures are defined before any result appears, citing Hyndman and Koehler. Results follow method by method, and a highlighted sentence turns the consensus bias into about 21,850 surplus bags. A section explains why weighted error exceeds MAPE for the consensus. Another notes that its four low weeks missed by only 31 bags. Tracking signals come next with Trigg: the consensus breaches minus 4 in week 5, while last year's pattern swings from minus 12.0 to plus 4.0, a timing miss. Interpretation separates timing from bias using Fildes and colleagues, and the recommendation keeps the profile with recorded, reasoned overrides.
Reading the MT438 Unit 5 grading rubric
In forecast comparisons the MT438 graders look first for definitions and a stated sign convention, since a bias figure without its convention cannot be interpreted. One accuracy measure is rarely enough; marks go to papers that report several and explain why they rank methods differently. Explaining why MAPE and weighted error diverge earns analysis credit, especially when the explanation is tied to where errors fall in the season. Tracking signals are rewarded when they are read over time, separating a timing miss from a persistent bias. Graders also credit papers that distinguish a method's failure mode instead of declaring a single winner without comment. A recommendation that includes how the forecast will be monitored shows practical sense. Tables and charts should be labeled and referenced in the text. Sources on forecast accuracy and judgment should support specific claims, cited in APA format at Purdue Global.
MT438 Unit 5 help: mistakes that cost points
The most frequent Unit 5 problem is reporting error without a sign convention, which leaves every bias figure ambiguous. Close behind is relying on MAPE alone; it treats quiet weeks and peak weeks equally and can make a method biased in busy weeks look acceptable. Students sometimes compute a tracking signal once at the end instead of week by week, losing the story of when a method went wrong. Methods are often described too loosely to reproduce. Some papers declare the most sophisticated-sounding method the winner. Another common gap is treating judgment as good or bad in general rather than asking whether each override had a reason. Before submitting, check that every sign in your paper means the same thing. Then look for weeks where your measures disagree; explaining one such week usually shows more understanding than another table.
Get an MT438 Unit 5 example written to your instructions
Send the demand history and forecasts in your Unit 5 case, the prompt and your rubric. A composite analysis comes back with its sign convention stated, each error measure defined before use, a week-by-week tracking signal and a recommendation saying what would trigger a review. That opening custom sample costs you nothing and needs 24-48h. The paper above is an original model document written by our desk, not a submitted student paper and not an official Purdue University Global document.
More MT438 papers and BS in Business Administration samples
- MT438 Unit 1: Supply Chain Data Inventory Post
- MT438 Unit 2: Order Data Quality Audit
- MT438 Unit 3: Supply Chain Metrics Exercise
- MT438 Unit 4: Operations Dashboard Critique
- MT438 Unit 6: Seminar Reflection
- MT438 Unit 7: Visibility Technology Brief
- MT438 Unit 8: Root Cause Analysis
- MT438 Unit 9: Predictive Model Evaluation
- MT438 Unit 10: Analytics Recommendation Memo
- MT437 Unit 5: Pick Method Comparison
- MT436 Unit 5: Request for Proposal
- MT435 Unit 5: Job Scheduling Exercise
- MT434 Unit 5: Shipment Consolidation Study
MT438 Unit 5 questions, answered
How do MAPE and weighted absolute percentage error differ?
MAPE averages each period's percentage error, so every period counts equally whatever its volume. Weighted error, often called WAPE, divides total absolute error by total actual volume, so high-volume periods count more. In seasonal data they can differ sharply. Reporting both, and explaining any gap, shows the grader you understand what each measure rewards.
How is a tracking signal calculated?
Divide the running sum of forecast errors by the mean absolute deviation to date. A forecast without bias keeps the signal near zero; persistent over- or under-forecasting pushes it away. Limits of plus or minus 4 are a common rule of thumb for flagging review. Read it period by period, because a signal that crosses a limit and returns tells a different story from one that keeps climbing.
Should I use MAPE when some periods have very low demand?
Use it with caution. When actual demand is small, even a modest miss becomes a huge percentage and can dominate the average. Many analysts add a weighted measure or exclude near-zero periods with a stated rule. Whatever you decide, say so explicitly, since graders tend to notice when a low-volume stretch drives an accuracy result.
Where can I find a free MT438 Unit 5 sample paper?
The whole MT438 Unit 5 analysis is printed here, scoring three grass seed forecasts over 26 weeks on four error measures and a weekly tracking signal.
When does the consensus forecast get flagged in the MT438 Unit 5 sample?
Its tracking signal passes minus 4 in week 5, at minus 4.3, and keeps falling to minus 25.7, so monitoring would have flagged it in early February.