| Course | BU 407 Quantitative Decision-Making |
|---|---|
| Unit | Unit 3 |
| Paper type | Forecast selection record |
| Length | About 1,010 words, 6 pages |
| Format | APA 7 student paper |
| School | Herzing University |
| Program | BS in Business Analytics |
| Updated | September 2026 |
Free sample paper for BU 407 Unit 3
Naive, Three-Month Average or Exponential Smoothing? A Forecast Selection Record for Two Years of a Composite Dairy Cooperative's Yogurt Demand
Student Name
Business, Herzing University
BU 407: Quantitative Decision-Making
Instructor Name
February 18, 2026
Naive, Three-Month Average or Exponential Smoothing? A Forecast Selection Record for Two Years of a Composite Dairy Cooperative's Yogurt Demand
The Question and the Data
The fictional Northwoods cooperative needs a monthly forecast of yogurt demand to plan production at its Eau Claire plant and to support the capacity decision in Units 1 and 2. This record runs three forecasting methods over the same 24 months of demand history, January 2024 through December 2025, scores them on the same error measures over the same months and selects one. The series, in thousands of cases, comes from the cooperative's shipment records and belongs to the composite case.
The history is: 316, 317, 328, 315, 318, 341, 328, 325, 348, 345, 347 and 325 for 2024; and 356, 344, 346, 356, 358, 368, 363, 371, 366, 357, 377 and 361 for 2025. Plotted as a line, the series climbs from the low 300s to the high 300s over two years, an upward trend of roughly two to three thousand cases a month, with month-to-month noise of about ten thousand cases. There is no clear seasonal pattern; the peaks do not recur in the same months each year, so seasonal indices are not needed for this series.
Method 1: Naive
The naive forecast uses last month's actual demand as next month's forecast. It needs no parameters and serves as the baseline every other method must beat; a method that cannot outperform the naive forecast adds complexity for nothing. For February 2024 it forecasts 316, the January actual, and so on through the series. Its first forecast covers February 2024.
Method 2: Three-Month Moving Average
The three-month moving average forecasts each month as the average of the previous three months. It smooths noise but lags any trend, because it gives equal weight to months that are up to three months old. Its first forecast covers April 2024, the average of January through March 2024: (316 plus 317 plus 328) divided by 3, or about 320.3. Its forecast for January 2026 is the average of October through December 2025, about 365.0.
Method 3: Simple Exponential Smoothing
Simple exponential smoothing forecasts each month as a weighted average of the last actual and the last forecast, with a smoothing constant, alpha, setting the weight on the actual. The method was run with alpha equal to 0.3, a common starting value, initialized with the first actual as the first forecast. Hyndman and Athanasopoulos (2021) explain that larger values of alpha make the forecast respond faster to recent changes and smaller values make it smoother. Alternative values from 0.1 to 0.6 were also tested, and the results are reported below. With alpha at 0.3, the forecast for January 2026 is about 364.5.
The choice of alpha deserves a note. The error table shows that raising alpha from 0.3 to 0.5 improves every measure, because a series with a trend rewards a method that reacts quickly to the most recent months. Beyond 0.5 the gains stop and reverse slightly, as the forecast starts chasing noise. A forecaster who tuned alpha on the same months used to score the methods would be flattering exponential smoothing, since the tuned value was chosen to fit those months; that is one reason the comparison reports both the standard value and the tuned one.
Comparing the Methods on the Same Months
All three methods were scored over the same 21 months, April 2024 through December 2025, the months for which every method has a forecast. Four measures were used: mean absolute deviation, the average size of the errors in thousands of cases; mean squared error, which punishes large errors more heavily; mean absolute percentage error; and bias, the signed average error, positive when forecasts fall short of actual demand.
Using the same 21 months for every method matters. If the naive forecast were scored from February 2024 and the moving average only from April 2024, the two would be judged on different months, and a noisy early month could favor or penalize one method by chance. Starting every comparison at the first month all methods share removes that distortion. The measures also answer different questions: mean absolute deviation shows the typical miss in cases, mean squared error shows whether large misses occur and bias shows whether a method leans consistently high or low.
| Method | MAD | MSE | MAPE | Bias |
|---|---|---|---|---|
| Naive | 11.19 | 194.0 | 3.22% | +1.57 |
| Three-month moving average | 8.86 | 116.0 | 2.54% | +4.13 |
| Exponential smoothing, alpha 0.3 | 10.10 | 149.3 | 2.87% | +7.09 |
| Exponential smoothing, best tested alpha 0.5 | 9.03 | 132.3 | 2.58% | +4.09 |
The Winner, Defended Against the Baseline
The three-month moving average wins on every accuracy measure. Its mean absolute deviation of 8.86 thousand cases is about 21 percent smaller than the naive forecast's 11.19, and its mean squared error is about 40 percent smaller, showing that it avoids the naive method's occasional large misses. It also beats exponential smoothing at alpha 0.3, and narrowly beats the best smoothing constant tested, 0.5. The comparison that matters most is with the naive baseline, and the moving average clears it comfortably.
The winner has a clear weakness, however, which the bias column reveals. Every method except the naive forecast under-forecasts on average, because the series is trending upward and averaging methods look backward. The moving average under-forecasts by about 4 thousand cases a month. The naive forecast is nearly unbiased because it always uses the latest, highest-recent value, but its errors are larger. Makridakis et al. (2018) found in a large comparison that simple statistical methods often match or beat more complex ones, but also that no method is best everywhere, which argues for testing rather than assuming.
Recommendation
The cooperative should use the three-month moving average for monthly yogurt planning, giving a January 2026 forecast of about 365 thousand cases, and should add about 4 thousand cases to each forecast to offset its trend lag until a better method is tested. Because the series is trending, the next step should be a method designed for trend, such as Holt's linear trend method, which Hyndman and Athanasopoulos (2021) describe as extending exponential smoothing with a second equation for the slope. The forecast should be rescored every quarter as new months arrive. Northwoods Dairy Cooperative and its demand history are composites created for this course (Render et al., 2018).
References
Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd ed.). OTexts. https://otexts.com/fpp3/
Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2018). Statistical and machine learning forecasting methods: Concerns and ways forward. PLOS ONE, 13(3), e0194889. https://doi.org/10.1371/journal.pone.0194889
Render, B., Stair, R. M., Jr., Hanna, M. E., & Hale, T. S. (2018). Quantitative analysis for management (13th ed.). Pearson.
BU 407 Unit 3 instructions, in plain terms
The Unit 3 assignment asks students to apply several forecasting methods to the same demand history and choose one. Most versions supply a series and want a naive baseline, a moving average and exponential smoothing applied, error measures calculated over a common set of periods and a justified selection. Many rubrics ask how the smoothing constant was chosen and whether seasonality needs to be addressed. Instructors expect the history to be examined for trend and seasonality before methods are applied and the winner to be compared with the naive baseline, not only with the other methods. A typical submission runs a few pages around one error table, with forecasting sources cited in APA 7, and it should give the forecast for the next period from the chosen method.
How the BU 407 Unit 3 example is put together
The sample opens with the business question and prints the 24-month history, then describes its trend and lack of seasonality. Separate blocks describe the naive forecast, the three-month moving average and exponential smoothing, each with its parameter, first forecast month and next-period forecast, and the smoothing block explains how alpha affects responsiveness. A comparison section explains why all methods are scored on the same 21 months and what each error measure shows. The results table reports four measures for four versions. The winner is defended against the naive baseline and the bias is interpreted. The recommendation gives a forecast, a correction and a next method to test.
BU 407 Unit 3 rubric: what full points look like
Herzing's criteria for the forecasting unit commonly include examination of the data, correct application of methods, a fair comparison and a justified selection. Examination earns full points when trend and seasonality are assessed before methods are applied. Application is judged by correct forecasts with stated parameters and first periods. The comparison scores highest when every method is scored over the same periods with several error measures, including bias, as the sample does. Selection rewards a winner defended against the naive baseline and honest discussion of its weaknesses. Instructors also credit explaining how the smoothing constant was chosen. Clear tables and citations complete the grade.
Common BU 407 Unit 3 mistakes, and how to avoid them
A common Unit 3 mistake is scoring each method from its own first forecast, so methods are compared on different months. Another is choosing a smoothing constant by trial on the same data used for scoring without saying so, which flatters that method. Students often report only one error measure, which can hide large misses or systematic bias. Some papers skip the naive baseline, making it impossible to tell whether any method adds value. Others apply seasonal indices to a series with no seasonality. Finally, look at bias; a method that is accurate on average but always too low will leave shelves short in a growing market.
Get a BU 407 Unit 3 example written to your instructions
Attach the Unit 3 instructions, the BU 407 rubric and whichever demand series your assignment supplies. We write a custom example that charts the history first, keeps a naive baseline in the comparison, scores every method over identical periods and defends the parameter it settles on. First custom sample free, returned in 24 to 48 hours.
More BU 407 papers and business analytics samples
- BU 407 Unit 1: Payoff table write-up
- BU 407 Unit 2: Expected value comparison
- BU 407 Unit 4: Optimization model summary
- BU 407 Unit 5: Sensitivity table analysis
- BU 407 Unit 6: Transportation model record
- BU 407 Unit 7: Simulation output brief
- BU 407 Unit 8: Quantitative recommendation file
- BU 351 Unit 3: Descriptive analysis
- BU 454 Unit 3: Comparison and ranking charts
- BU 455 Unit 3: Storage and format decisions
- BU 451 Unit 3: Cleaning and transformation in code
BU 407 Unit 3 questions, answered
How do I choose the smoothing constant?
By testing rather than by taste. Run two or three values over the same history, report the error measures for each, and keep the one performing best on the periods you scored. Say what you tried. A constant near zero smooths heavily and reacts slowly, while a high one chases noise. Graders in many sections award the search itself, not only the value you end on.
Does the record need seasonal indices?
Only if the series shows seasonality and the instructions call for them. Where a repeating pattern is visible in the chart, a method ignoring it will lose to one that does not, and saying so is part of the selection. Where the history is short or flat, note that you looked and found nothing rather than leaving the question unmentioned.
Can I use sales figures from work?
Sales history sitting in an employer's system belongs to the employer, whatever access you happen to have. Where an assignment opens a door to data of your own, get permission first, remove anything identifying, or build a composite series with the shape your industry shows and note that you did. The method work scores the same either way.
Where can I find a free BU 407 Unit 3 sample paper?
A free copy of the BU 407 Unit 3 forecast record, including the error table, is shown above. If you have a different demand series, send it with the prompt, and a free custom selection record will be worked through for you, typically within about two days.
How do I choose the smoothing constant in BU 407 Unit 3?
Try several values and compare their errors on the same periods, then say how you chose. The sample reports alpha 0.3 as a standard starting value and alpha 0.5 as the best tested, noting that tuning on the scoring months flatters the method.