select... Topics Organization People Activities Issues

Vol 25 - No 08 - October 1999 MAD, SAD or just BAD Stock control is an exact science, but unfortunately the answers are never quite right. My letter earlier this year - March issue) suggested that safety stock for an item could not be calculated using the results from a group, an opinion reinforced by Mike Thomson's letter in the September issue... I am still unsure however whether we can get an accurate service level within a group simply by applying the group MAD to each, even though the service level for each item will be wrong. Mike does introduce another issue which I think is important to those who want to do accurate forecasting, namely that of data accuracy. In using a historical forecast we have to resort to old data. In most businesses the demand is changing significantly due to the customer base, market forces, product life cycle and economic conditions. What this means for a large number of companies is that the demand does not consist of a number of cause and effect related events, but of random demand, largely caused by impulse purchasing, review levels triggering and random replenishment of unserviceable items. By the time we have collected enough data the product has been superceded - and worst of all we collect data on sales not demand, ignore public holidays and social events which modify demand. Therefore we have to be careful to make the right compromise. Theoretically we are not using enough historical data to establish a) a statistically significant value for MAD or SAD b) whether the distribution is Gaussian or not. What we do know is that statistically that the deviation should be measured using positive deviations only (thus throwing away about half of the already scarce data). The measured standard deviation is therefore likely to be equally as inaccurate as the MAD. The Gaussian distribution approximation is mainly made using the customer service factor tables for the Gaussian distribution, rather than in the measurement of deviation. Moving to older data degrades the accuracy of the information so extra information does not help. Forecast accuracy by SKU is an advantage in some circumstances, but elsewhere we find that the most accurate forecasts are the aggregate ones (particularly the budgets). Dividing the history into smaller time buckets gives more data but leads to the extra deviations due to the Friday effect or the 4th week of the month effect. Whether on the basis of this dodgy data we use MAD or standard deviation is less important than these other factors. It is obviously most important to use an exponentially weighted MAD in practice and the value of weighting factor chosen will have most bearing on the safety stock. (The difference between this weighed MAD and an exponentially weighted variance would appear to be a minor complication). The formulae which we use for reorder levels are wrong, we know that. The more refined formulae are not worth using because they are more difficult for the users to understand and only give marginal theoretical improvements. My own spreadsheet based package gives a simple and accurate solution which the users can understand, apply with confidence and save inventory. Improving the forecast accuracy will decrease the importance of the safety stock calculation. We are indebted to RG Brown for his major contribution to inventory control. In this context I would favour the continuing use of MADs for ease of understanding and weighting. Incidentally the operating problems with Triggs Tracking signal do not necessarily consign it to the scrap heap (as it did with the EOQ formula) because it can be modified. Tony Wild, MIOM, Midas Consultants Page: 5 Words: approx. 400 Related Topics: Forecasting Inventory management
	To read the articles you'll need Acrobat Reader. To download Acrobat Reader click the following button: To Save and Print the articles click the appropriate on-screen buttons or select Save or Print from the File menu at the top of the Acrobat window.