| Course | MT296 Supply Chain Management Applications |
|---|---|
| Unit | Unit 2 |
| Paper type | Demand forecast exercise |
| Length | About 900 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 MT296 Unit 2
From 24.9 Percent Error to 6.1: A Unit 2 Demand Forecast Exercise on Two Years of One Pierogi Flavor
[Student Name]
Purdue University Global
MT296: Supply Chain Management Applications
Unit 2 Demand Forecast Exercise
[Instructor Name]
[Date]
The pierogi maker, its shipments and every figure are composite and illustrative; all forecasts and errors were calculated with a short script; the sources cited are real.
The Data
The data are monthly shipments of potato-cheddar twelve-count pierogi, in cases, for two calendar years.
Year one: January 7,389; February 8,555; March 8,944; April 7,622; May 6,378; June 5,833; July 5,444; August 5,755; September 6,611; October 7,778; November 10,189; December 12,832; total 93,330
Year two: January 7,607; February 7,519; March 10,226; April 8,784; May 6,662; June 5,784; July 5,647; August 5,983; September 6,509; October 7,923; November 10,534; December 13,682; total 96,860
The pattern repeats: a Lent rise in late winter, a summer low near 5,500 to 5,700 in July and a December peak above 12,800 in both years. Year two's Lent rise came later than year one's because Easter fell later, so February was lower and March and April higher.
Method One: Three-Month Moving Average
Each month's forecast is the average of the three months before it. The January forecast for year two, for example, is the average of October, November and December of year one, 10,266 cases. The method is simple and needs no knowledge of the season, which is also its weakness.
Method Two: Seasonal Index
A seasonal index for each month was computed from year one as that month's shipments divided by the year's monthly average of 7,778 cases. The indexes run from 0.70 in July to 1.65 in December, with 1.10 in February and 1.15 in March for Lent.
Each forecast in year two is built in three steps: strip the seasonal effect out of each of the three previous months by dividing by that month's index, average those three deseasonalized values and multiply by the index of the month being forecast. The January forecast becomes about 7,389 cases instead of 10,266, because the method knows that January follows the December peak rather than continuing it. Jacobs and Chase (2020) describe this decomposition as the standard way to forecast data with a stable seasonal pattern.
Errors
Both methods forecast year two month by month, and their errors were measured three ways.
Mean absolute deviation: moving average 2,058 cases; seasonal 509 cases
Mean absolute percentage error: moving average 24.9 percent; seasonal 6.1 percent
Tracking signal, the running total of errors measured in units of mean absolute deviation: moving average 0.55; seasonal 2.86
The seasonal method's error is about a quarter of the moving average's. The moving average's worst errors come exactly where the season turns: it forecast January and February far too high, coming off the December peak, and November and December far too low, coming out of the autumn trough. In December alone it missed by 5,360 cases.
The seasonal method's tracking signal of 2.86 shows a small, steady tendency to forecast too low, because year two grew about 4 percent and the method does not include a trend. A tracking signal within about plus or minus 4 is usually treated as acceptable, but the bias suggests adding a growth adjustment. Hyndman and Athanasopoulos (2021) recommend checking forecast residuals for exactly this kind of persistent bias, since it signals a pattern the model has missed.
The moving average was always looking backward at a season that had already turned.
Where the Seasonal Method Erred
The seasonal method's errors cluster in February, March and April: it forecast February about 1,120 cases too high and March about 1,555 too low, and April about 1,031 too low. Year one's indexes assume Lent peaks in February and March; in year two, Easter came later and the peak moved toward March and April. Chopra (2019) notes that holidays with moving dates are a common source of seasonal forecast error, and the fix is to index those weeks to the holiday rather than to the calendar month.
Why Year Two Is the Test
Both methods were built only from year one and then tested on year two, which they had not seen. That holdout design matters: a method fitted and tested on the same data will always look better than it will perform. Here the seasonal indexes came entirely from year one, so year two's errors are a fair estimate of how the method will do next year. With only two years of data, there is no room for a second holdout, so the refinements below should be checked against year three as it arrives.
What the Forecast Is For
The forecast drives three later decisions: the monthly production plan, the safety stock target and the purchase of seasonal ingredients such as cheddar ahead of the winter peak. An error of 2,000 cases in December, the moving average's typical miss, would mean either empty shelves in the busiest month or freezer space rented for product nobody ordered. At about 500 cases, the seasonal method's miss is small enough for safety stock to cover.
Recommendations
The company should use the seasonal method, with two refinements: index the Lent and Easter period to the date of Easter each year, and add a trend factor so the forecast keeps up with growth. With those changes, the remaining errors should fall below the 6.1 percent shown here.
References
Chopra, S. (2019). Supply chain management: Strategy, planning, and operation (7th ed.). Pearson.
Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd ed.). OTexts. https://otexts.com/fpp3/
Jacobs, F. R., & Chase, R. B. (2020). Operations and supply chain management (16th ed.). McGraw-Hill.
MT296 Unit 2 instructions, in plain terms
Unit 2 asks students to forecast a demand series with at least two methods and compare their accuracy. Use a real or composite history long enough to show any pattern, ideally two full years for seasonal data. Build each method on one part of the data and test it on another the method has not seen, so the comparison is fair. Measure error in more than one way: average size in cases, average size as a percentage and a tracking signal that reveals bias. Then explain where and why each method erred, not just which one won. Recommend improvements aimed at the errors found. Six pages of mostly tables, with sources in APA 7, is a typical length.
Inside the MT296 Unit 2 example
The exercise lists all 24 months and notes the Lent rise, summer low and December peak, with year two's later Easter. The moving average is explained with its January forecast. The seasonal method's indexes and three-step forecast are described, with Jacobs and Chase, and the January figure is compared, under a margin note on one number. An error section reports three measures for each method and explains the moving average's misses at the turns, with the highlighted line about looking backward. The tracking signal's bias is linked to growth, citing Hyndman and Athanasopoulos. Sections on the seasonal method's Easter errors with Chopra, why year two is the test and what the forecast drives lead to two refinements.
Where the points sit in the MT296 Unit 2 rubric
Forecasting exercises are graded on correct application of each method, fair testing, accurate error measures and insight into the errors. Building a method on one period and testing on another earns more than fitting and testing on the same data. Instructors reward reporting several error measures and explaining what each shows, especially bias through a tracking signal. Locating where a method fails, such as at seasonal turns or moving holidays, demonstrates understanding beyond the numbers. Recommendations that target the errors found show the comparison has a purpose. Clear tables, consistent arithmetic and correct citations complete the rubric, along with a short note on what the forecast will drive.
Common MT296 Unit 2 mistakes, and how to avoid them
The most frequent error is judging a method on the very months it was fitted to, which makes it look better than it will perform. Another is reporting only one error measure, so bias goes unnoticed. Some exercises compare methods without explaining why one did better. Others ignore holidays that move from year to year. Make sure seasonal indexes come only from the training period. Report errors month by month as well as on average, since averages hide where a method fails. Note any holiday that moves between years, and say how the method treats it. Share your demand history, and a writer can build both methods, compute the errors and explain where each one went wrong.
Get an MT296 Unit 2 example written to your instructions
Attach the Unit 2 demand series, or describe the product if data must be found, together with the prompt, its rubric, and any method the instructions name. Two methods are run and compared on held-out months in the sample, all arithmetic shown. It comes free as a first custom sample, inside 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 MT296 papers and BS in Business Administration samples
- MT296 Unit 1: Discussion Board Post
- MT296 Unit 3: Inventory Model Application
- MT296 Unit 4: Safety Stock Calculation
- MT296 Unit 5: Supplier Scorecard
- MT296 Unit 6: Seminar Reflection
- MT296 Unit 7: Transportation Mode Comparison
- MT296 Unit 8: Facility Location Analysis
- MT296 Unit 9: Sales and Operations Plan
- MT296 Unit 10: Applied Supply Chain Report
- MT282 Unit 2: Material Property Analysis
- MT281 Unit 2: Delivery Method Comparison
- MT262 Unit 2: Cultural Framework Application
- MT260 Unit 2: Group Development Analysis
MT296 Unit 2 questions, answered
Which two forecasting methods should I compare?
Whatever the prompt names comes first. If the choice is yours, pair a simple method with one that handles the data's main feature: a moving average against exponential smoothing for steady demand, or against a seasonal method when the series clearly repeats each year. Comparing two methods that share the same blind spot teaches very little.
What is a holdout period, and do I need one?
A holdout is data kept out of model building and used only to test forecasts. Building seasonal indices from year one and testing on year two is a simple version. Without one, error measures describe how well a method fits the past it was built on, which overstates how well it will forecast months it has never seen.
How are large forecast errors best explained?
Look for a business cause before a statistical one: a holiday that moved, a promotion, a stockout that capped sales, or a new customer. Name the likely cause, say how the method could be adjusted, and keep the adjustment out of the comparison itself so both methods are still tested on equal terms.
Where can I find a free MT296 Unit 2 sample paper?
The full MT296 Unit 2 demand forecast exercise is printed here. It forecasts a pierogi flavor's monthly shipments with a moving average and a seasonal index method and compares their errors on a holdout year.
Why did the seasonal method still err in the MT296 Unit 2 sample?
Easter fell later in year two, so the Lent peak moved from February and March toward March and April. Year one's indexes assumed the old timing, so the forecast was high in February and low in March and April.