Forecasting in Practice

Collaborate In building forecasts. Collaboration with one s supply chain paitners 

Share only the data that truly provide value. The value of data depeuds on where one sits in the supply chain. A retailer finds point-of-sale data to be quite valuable in measuring the performance of its stores. However, a manufacturer selling to a distributor that, in turn, sells to retailers does not need all the point-of-sale detail. The manufacturer finds aggregate demand data to be quite valuable, with marginally more value coming from detailed point-of-sale data. Keeping the data shared to what is truly required decreases investment in IT and improves the chances of successful collaboration. 

Be sure to distinguish between demand and saies. Often, companies make the mistake of looking at historical sales and assuming that this is what the historical d and was. To get true demand, however, adjustments need to be made for unmet demand due to stockouts, competitor actions, pricing, and promotions. Failure to do so results in forecasts that do not represent the current reahty. 

---

These methods are based on subjective opinions obtained from company executives and consumers. These methods are ideally suited for forecasts where there are no past data or past data is not reliable due to the changes in environment (e.g., peacetime data on spare part needs for military aircrafts are not useful for forecasts during war time). Qualitative methods are commonly used for strategic decisions where long-term forecasts are necessary. The qualitative approaches vary in sophistication from scientifically conducted consumer surveys to intuitive judgments from top executives. Quite frequently qualitative methods are also used as supplements to quantitative forecasts. In a 1994 survey of forecasting in practice, 78% of top 500 companies responded that they always or frequently used qualitative methods for forecasting (Sanders and Manrodt, 1994). 

Estimates of sales income and other types of forecasts are usually based on the opinions of experts. Experts should be able to estimate maximum, minimum, and most hkely, or modal, values for a quantity. The modal value is not necessarily midway between the minimum and maximum values, since many distributions are skewed. An expert may be asked to estimate the probabilitv of the occurrence of certain values on each side of the mode. Wken experts are questioned separately, the procedure is known as the Delphic method. Strictly speaking, this method requires that the opinion of each expert be assessed by a coordinator, who then feeds the resiilts back to see if the opinions of one expert are modified by those of others. The process is repeated until agreement is reached. In practice, the procedure is too tedious to be repeated more than once. 

A particular problem is the number of events that should be simulated before the results are stabilized about a mean value. This problem is comparable to the question of how many runs are required to simulate a Gaussian distribution within a certain precision. Experience shows that at least 1000 sample arrivals should be simulated to obtain reliable simulation results. The sample load (samples/day) therefore determines the time horizon of the simulation, which for low sample loads may be as long as several years. It means also that in practice many laboratories never reach a stationary state which makes forecasting difficult. However, one may assume that on the average the best long term decision will also be the best in the short run. One should be careful to tune a simulator based on results obtained before equilibrium is reached. 

Demand-oriented models investigating demand and classical forecasting of demand quantities in the chemical industry can be found for example in practice-oriented industry cases (Franke 2004). 

In practice, the smallest customers are not planned individually but are grouped into a cluster e g. called other customers therefore, single customer forecasts are considered on a customer cluster level, where a large customer forms a single customer cluster and small customers are grouped. 

Waters, A.J., M.J. Bader. J.R. Grant, G.S. Forbes, et al. Images in Weather Forecasting A Practical Guide for Interpreting Satellite and Radar Imagery, Cambridge University Press. New York, NY, 1997. 

The objective function of a supply network design model can either minimize costs or maximize profits. In practice the production function is often required to assume that all forecasted demands have to be met. In this constellation cost minimization and profit maximization lead to identical results and consequently cost minimization models are used. From an economic perspective this simplification can be justified in cases where a high share of fixed costs allows the assumption that any product sale con-... 

Lewis CD. 1981. Forecasting. In Lewis CD (ed), Operations Management in Practice. New York Wiley. 

It is also good practice to compare the statistical forecast to a naive forecast. (The naive forecast is a simple technique where the forecast equals the volume of goods sold in the prior forecasting period.) Naive forecasts, in some situations, can be surprisingly difficult to beat, yet it is very important that the organizations ensure that software and a statistical modeler improve on the naive model. The focus needs to be on continuous improvement. If the software, modeler is not able to do this, it makes sense to implement better software, improve the skills of the modeler, or just use the naive model as a baseline forecast. 

The SI unit for pressure is a derived unit that reflects the definition of pressure. The SI units for force and area are the newton and the square meter, respectively. So pressure is measured in N m . This unit has also been named the Pascal, Pa. One newton is roughly the force exerted by gravity on a one-quarter-pound object. When spread out over a square meter, this force does not produce very much pressure, so the Pa is a relatively small quantity. Typical atmospheric pressures are on the order of 10 Pa. Thus a more practical unit is kPa, which is the unit used in weather forecasts in Canada and other metric countries. ... 

Forecasts are d5utamic and they are updated as more information becomes available. Hence, after selecting an appropriate forecasting method and the forecasts based on that method, it is important to continuously monitor the forecast accuracy. For this, one can use one or more of the forecast errors discussed in Section 2.9. In practice, another measme, called Tracking Signal, is also commonly used for monitoring forecast accmacy. 

There are several published results of real world forecasting. We discuss briefly a few of the forecasting applications in practice. For more details and additional applications, the reader is referred to the text by Taylor (Taylor, 2007). 

---