Forecasting Consumer Product Demand with Weather Information: A Case Study

Describes how one can incorporate weather information into a forecasting model to improve the accuracy of demand forecasts ...it is preferable to use weather forecasts which are objectively (not subjectively) derived ... explains the findings of a study of consumer products company where weather information was used to forecast demand. Weather plays a major role in our day-to-day life. It is an important factor in planning our leisure and making our purchasing plans. Most product demand patterns have a logical dependence on weather. Therefore, it is important to assess the significance of weather on demand. If it does have an impact, we have to include it in the forecasting model. In this article, we report the results of a study of daily consumer product demand forecasts where weather was incorporated into the model. WHY WEATHER FOR DEMAND FORECASTS Why blame weather when you can plan for it? Many corporations cite weather as a reason when earnings estimates are not met. While this seems to be a logical and acceptable explanation to some investors and analysts, the fact remains that companies should plan for, rather than react to, weather's impact. Hedging for an enterprise's weather exposure is one way to mitigate weather impacts. Another way is to include it in forecasting models. In this article, we demonstrate how one can incorporate weather information into a forecasting model to improve the accuracy of demand forecasts. HOW TO USE WEATHER INFORMATION In using weather in a forecasting model, two things are important: (1) Using the right weather information. (2) Using the right forecasting model and software. Using The Right Weather Information: Before deciding to incorporate weather information into a demand forecasting model, one must consider which weather variables are most likely to impact consumer demand. Anecdotal information, common sense, and experience are likely to be the guide. Also, one must consider the locations where weather is important, as well as the types of weather that have both enhancing and detrimental effects on consumer demand. Provided that one is dealing with a sales forecasting problem for a manageable amount of first or second tier cities, adequate historical weather information is likely to be available via the National Weather Service, or through commercial weather information vendors. Correlation studies can be performed to determine how closely demand follows weather patterns. Further, one can test correlations between specific weather variables (such as Temperature, Precipitation, Dewpoint, Relative Humidity and Cloud Cover) and the demand for goods, which can guide in selecting the best variable for a given demand data. Great care must be taken in establishing a complete dataset of observed historical weather, so that the demand model can properly represent the relationships between demand and weather conditions. Also, important consideration must be given to the type of weather forecast that is used. Weather forecasts can be prepared cither subjectively or objectively. Many weather forecasting companies offer weather forecasts that have been subjectively constructed by meteorologists. Objective weather forecasts are those that are generated without human intervention. This means that the outputs are systematic and reliable, and error statistics are quantifiable and stable. Using The Right Demand Model And Forecasting Software: Selecting the best weather variable is not enough. One must use the right demand model and forecasting software. The software should automatically select the best from the user's suggested variables, consider lead and lag relationships, and account for outliers and unusual values in the data. Outliers often give very important information, provided they are fully understood and properly used. They can help to improve further the quality of final forecasts. Some important outliers (unusual values) can be characterized as: (1) Pulses - one-time unusual values, (2) Seasonal Pulses - repetitive pulses over time, (3) Level Shifts -step like changes in the mean to a lower or higher level, and (4) Time Trends - systematic increases or decreases in the mean over time. …