MGT 240 Module 5 Demand Forecasting Example

Out of Chlorine in June, Overstocked in September: Choosing a Forecasting Method for a Pool Supply Chain

Student Name

Business Administration Program, Aspen University

MGT 240: Operations Management

Instructor Name

Month Day, Year

Out of Chlorine in June, Overstocked in September: Choosing a Forecasting Method for a Pool Supply Chain

Bayou Pool Supply, a composite company, runs six stores in the Houston, Texas, area selling chemicals, equipment and parts to homeowners and pool service companies. Chlorine tablets in 25-pound buckets are its largest product by revenue. The owner orders them every two weeks from a distributor, based on memory, last year's totals and the weather forecast. In June 2025, a hot spell emptied shelves for nine days and service companies moved accounts to a competitor. By September, the stores held about 2,400 unsold buckets, tying up roughly $190,000. This paper compares forecasting methods for weekly demand and recommends a forecasting process.

The Demand Pattern

Weekly chain-wide sales over four years show a strong seasonal pattern. Demand averages about 150 buckets a week in January, rises through spring, peaks near 1,100 in July and falls through autumn. The timing of the rise shifts by two or three weeks depending on when warm weather arrives. Total annual demand has grown about 4% a year as new homes with pools are built. Two promotional weeks each spring produce spikes.

Methods Tested

Four methods were fitted on 2022 through 2024 and used to forecast each week of 2025, two weeks ahead, matching the ordering cycle. The naive method uses the most recent week's sales. The eight-week moving average averages the last eight weeks. The seasonal naive method uses sales in the same week last year, adjusted for growth. Holt-Winters seasonal exponential smoothing estimates a level, a trend and a seasonal index for each week, updating them as new sales arrive. The owner's past orders, converted into implied forecasts, were included as a fifth comparison.

Results

What the Research Says About Methods

Makridakis and Hibon (2000) reported the results of the M3 competition, in which many forecasting methods were tested on 3,003 time series. Among their conclusions were that statistically sophisticated methods did not necessarily produce more accurate forecasts than simpler ones, that combining methods often helped and that accuracy depended on the forecasting horizon and the measure used. Their findings support using a well-understood method such as Holt-Winters rather than a complex model the business cannot maintain.

Seasonal Methods and Honest Testing

Hyndman and Athanasopoulos (2021) explain that exponential smoothing methods with seasonal components suit series like Bayou's, where a repeating pattern sits on top of a changing level. They stress evaluating forecasts on a test set the model did not use in fitting, since a model can match past data closely and still forecast poorly. The holdout year used here follows that advice.

Judgment and Adjustment

Fildes and Goodwin (2007) studied how companies adjust statistical forecasts and found that adjustments were very common, that many were small and that small adjustments often reduced accuracy, while large adjustments based on specific information, such as a known promotion, were more likely to help. They recommended that adjusters record their reasons and that companies review whether adjustments improved forecasts. Bayou's owner reacted to every weather report, which explains why his forecasts overshot after the June heat.

Would Combining Help

Because Makridakis and Hibon found that combining methods often improved accuracy, the analyst also tested a simple average of the Holt-Winters and seasonal naive forecasts. Its error in 2025 was 10%, slightly worse than Holt-Winters alone, because the seasonal naive forecast repeated the late spring of 2024. The combination was not adopted, but the test will be repeated each year, since one year is a small sample and the result could change.

Forecasting by Store

Chain-wide forecasts are more accurate than store forecasts, because random variation at individual stores partly cancels out. Store-level Holt-Winters forecasts had errors between 12% and 19%. The process will therefore forecast chain-wide demand first and divide it among stores using each store's share of sales over the past eight weeks, which adjusts for stores that are gaining or losing customers. Store managers will see their store's forecast each Monday and may flag local events, such as a new subdivision opening or a large service company switching suppliers, which the owner can then treat as a recorded adjustment.

A Forecasting Process

Each Monday, the system will produce a Holt-Winters forecast for the next four weeks by store. The owner may adjust it only for three reasons: a promotion, a heat wave forecast of three or more days above 98 degrees and a known change such as a new service company account. Each adjustment will be recorded with its reason and size. Each quarter, the owner and store managers will compare adjusted and unadjusted forecasts against actual sales and keep only the kinds of adjustments that helped.

From Forecast to Orders

Forecasts will feed orders with safety stock sized to the remaining error. Holt-Winters' lower error allows the chain to carry about a third less safety stock than under the owner's method while reducing stockout risk in early summer, when the timing of warm weather matters most.

Limits of the Analysis

One test year cannot show how the methods behave in an unusual season, such as a cool summer or a hurricane that closes stores for a week. The chain will track errors continuously and revisit the method if Holt-Winters' error rises above 13% for a full quarter.

Value of Better Forecasts

If the improvement had been in place in 2025, the chain estimates it would have avoided most of the nine days of June stockouts, worth about $85,000 in lost sales and the two lost service accounts, and would have ended September with about 900 rather than 2,400 excess buckets, freeing about $120,000.

Conclusion

Bayou's chlorine demand is seasonal, growing and sensitive to weather, which makes simple averages a poor fit and the owner's judgment unstable. Testing methods on a holdout year shows Holt-Winters seasonal smoothing cuts error nearly in half compared with the owner's approach. Research supports simple, well-tested methods and disciplined adjustments, which together give the chain a forecasting process it can run every week.

References

Fildes, R., & Goodwin, P. (2007). Against your better judgment? How organizations can improve their use of management judgment in forecasting. Interfaces, 37(6), 570-576. https://doi.org/10.1287/inte.1070.0309

Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd ed.). OTexts. https://otexts.com/fpp3/

Makridakis, S., & Hibon, M. (2000). The M3-Competition: Results, conclusions and implications. International Journal of Forecasting, 16(4), 451-476. https://doi.org/10.1016/S0169-2070(00)00057-1