Forecasting –Part One (text version of audio portion)
Slide One:
This is the first part of a 3-part series on forecasting.
Slide Two:
The first part of the series presents the principles of forecasting, its importance to business functions, and the types of forecasting methods available to decision makers. The second part of the series describes two forecasting models, namely time series and causal models, and their related forecasting techniques. The third part of the series discusses the criteria for choosing a forecasting model with special emphasis on tracking the accuracy of the model.
Slide Three:
Forecasting involves predicting future events for effective planning. Accurate forecasts are of critical importance in all aspects of a business. From a strategic perspective, forecasts on the economy and demand drive decisions on new product screening, production process selection, capacity planning, facility location decisions and other corporate plans. From a tactical perspective, sales forecasts determine production planning, job scheduling, budgeting, and other operation plans. The main purpose of forecasting is therefore to support both long and short-term business planning with good estimates of what will happen in the future.
Slide Four:
What constitutes a good forecast? A good forecast should be stable, short-ranged, accurate, and proper. Predicting units of demand for groups of goods or services that share similar demand and production requirements is more stable than predicting sales revenue of individual items. It is because prices and demand of individual items often fluctuate with erratic patterns that are difficult to pinpoint. A good forecast is also short-ranged because the degree of uncertainty increases as one stretches more into the distant future. Another indicator of a good forecast is its accuracy or whether the prediction actually comes true. Lastly, a good forecast is the result of selecting a proper forecasting model to meet one’s business needs.
Slide Five:
After defining what is a good forecast, we need to learn how to generate such a forecast. To do so, a 5-step forecasting process is recommended. At the outset, we need to determine what to forecast and how the forecast will be used. In general, economic and demand forecasts are essential for business planning, as explained earlier. Economic forecasts such as inflation rates, unemployment rates, money supplies, and other economic indicators define the environment in which the business operates. Demand forecasts are projections of customers’ demand for the company’s product and services. They are the root of most if not all business decisions. After deciding on what to forecast, the next step calls for defining the forecast in more specific terms such as its time horizon, the data needed and how the data should be collected. Based on the nature of the forecast needed, an appropriate forecasting method will be selected to generate the forecast. After a forecast is made, what actually happened should be compared with the prediction so that the accuracy of the forecast can be monitored and controlled continuously.
Slide Six:
There are two types of forecasting methods depending on the type of data used. Methods based on subjective data, such as the decision maker’s opinion, intuition, emotions, experience, and value system, are called qualitative or judgmental. Examples of qualitative forecasting methods include survey, Delphi, and market research. Methods that use mathematical models that rely on historical data and/or causal variables are called quantitative. Quantitative methods are further divided into time series models and causal models, if causal variables are considered. Quantitative methods are the focus of this course and will be discussed in further detail next.
Slide Seven:
Time series models assume that the history of occurrences over time can be used to predict the future. They rely on the idea that history repeats itself and the crux is to discover those underlying patterns through historical data analysis, to map out the course for future events. The historical data are based on a series of evenly spaced points, such as weekly, monthly, quarterly, or yearly data points. Four underlying patterns can be present in the time series, namely, level, trend, seasonality or cycle, and random, as shown in the accompanying slide.
Slide Eight:
The forecasting techniques that are useful to identify the level, trend, and seasonality patterns of a time series are listed here. To forecast the level of a time series, you will learn four techniques: simple average, simple moving average, weighted moving average, and exponential smoothing. To forecast the trend of a time series, you will learn trend-adjusted exponential smoothing. To forecast the seasonality of a time series, you will learn to express multiplicative seasonality as seasonal index.
Slide Nine:
The random component of a time series is unexplained variation that cannot be predicted, but can be treated as a source of forecast errors to gauge the accuracy of the forecast. You will learn how to measure such forecast errors by using mean absolute deviation and mean square error. Finally you will learn how to monitor these errors by using tracking signal.
Slide Ten:
Causal models assume that the occurrence of other events or factors can be used to predict the future. They rely on the idea that a cause-and-effect relationship exists between a set of causal factors (or independent variables) and the variable of interest (or dependent variable). For example, the sales of a product may be related to the company’s advertising expenditure, the quality of the product, and the competitor’s prices. The relationship can be captured by a mathematical model such as linear regression, the one causal model you will learn in the next set of slides (part 2).