Introduction to Forecasting Considerations

Forecasts are predictions about future events. They provide a basis for making decisions in a timely manner, which is often in advance of the facts. There are many ways of creating forecasts, and the best method for a particular forecast depends on many factors. Consider this question: Will it rain tomorrow? The degree of difficulty in correctly predicting tomorrow's weather depends on where you live. You may live where the weather is extremely stable, with little or no variation from one day to the next. In this situation, if it is raining today, then you can be fairly certain that it will rain tomorrow. However, if you live where the weather is in constant flux, with sudden and dramatic changes, then today's rainfall is not a reliable predictor. You may just make an informed guess, based on your analysis of the current weather pattern, or you might consult an arthritis sufferer whose joints ache with changes in the weather. Nonetheless, all of these methods (today's rainfall, informed guess, or swollen joints) should over time prove to be more accurate than just flipping a coin. Now consider this question: Will it rain three months from today? Instead of basing your prediction on today's weather, you need to consider the frequency of rainfall for the forecast period in previous years. If you live where rainy seasons and dry seasons are clearly defined, then you can probably answer this question with relative certainty based on the season. Otherwise, your ability to predict rainfall on a particular day that far into the future may be no better than a coin toss. To make a meaningful prediction, you may need to expand the forecast period to a week or more. You may also need to expand the size of the area in which you are predicting rain from your neighborhood to a larger region. Finally, how important is it to correctly predict the weather on a particular day and at a particular place? If accuracy is critical -- such as planning a large outdoor event -- then an accurate forecast is worth some effort, and you might try several forecasting methods to see if their predictions converge. Regardless, you might still plan to erect a tent in case you get a downpour instead of the forecasted clear skies. This simple example demonstrates several characteristics of forecasting:

Stable patterns in historical data are more likely to generate an accurate forecast. Different methods are appropriate for different forecasts, depending on how far into the future you want to make a forecast and how stable your data is. Some forecasting methods are experiential or qualitative (informed guess or aching joints), and others are quantitative (historical data). The season may be an important factor in the forecast. Forecasting is not 100% accurate. The more precise the forecast, the more prone it is to error. Longer-range forecasts should generate data at higher levels to offset the increasing likelihood of error.

The degree of error may be offset by your tolerance for error. If you have a low tolerance for error, then you may want to make some provisions that will lessen the consequences of forecasting incorrectly.

These observations may help give you a perspective on what you want to forecast, how you want to design the forecast, and how you want to use the forecast in making decisions about your business.

Choosing a General Forecasting Approach

The first step in generating a forecast is to decide how far into the future you want to make your predictions. The approach that produces the best results for short-term forecasts is not a good predictor of long-term performance. The opposite is also true. The critical question is, of course, how far into the future these time frames reach. Is "short" five weeks or five months? Is "long" five quarters or five years? As illustrated by the rain prediction example in "Introduction to Forecasting Considerations", it all depends on a variety of factors:

What are you trying to forecast? How stable is the historical data? How are you going to use this information?

These are just a few of the questions that you need to answer in order to define the forecasting time frames for your specific business. Table 7-1 provides some general guidelines for these time periods. Table 7-1 Guidelines for Choosing a Forecasting Approach Time Frame Short Medium Long Typical Forecasting Horizon Up to 18 months 6 to 36 months 19 months to 5 years Best Approach Time Series Causal Analysis Expert Opinion

Time Series
Time series forecasting methods are based on the premise that you can predict future performance of a measure simply by analyzing its past results. These methods identify a pattern in the historical data and use that pattern to extrapolate future values. Past results can, in fact, be a very reliable predictor for a short period into the future. You can generate this type of forecast very quickly and easily, and you do not need either forecasting expertise or an in-depth knowledge of your data. The modeling techniques used by the time-

series methods are relatively simple and run very fast. Time-series forecasting is extremely useful when hundreds or thousands of items must be forecast. You may also use time-series methods to generate forecast data further into the future. However, the results will not be as accurate, because factors other than past performance have a greater impact over time. For example, you or your competitors may change the pricing structure or run advertising campaigns, competitive products may come onto the market, or shifts in the economy or political events may affect performance. You should consider the forecast data generated by time series methods to be one component of a medium- or long-range forecast, which may be adjusted by expert opinion and other factors. Analytic Workspace Manager provides access to a time-series forecasting engine, which is described in this chapter.

Causal Analysis
Causal analysis takes into consideration the external factors (the causes) that can affect a forecast, as described previously under "Time Series" . Statistical regression methods are the basis for causal analysis. They use the forecasts for several independent measures to forecast a dependent measure. This type of forecast requires considerable skill and understanding of forecasting methodology and the relationships between independent and dependent variables. A good regression model will produce the best results for medium-range forecasts. However, because of the time, expense, and expertise needed to develop a model, most businesses restrict regression analysis to a few key business measures. For the other measures, they use a combination of methods including time-series and expert opinion. The forecasting engine used by Analytic Workspace Manager does not support causal analysis. The linear and nonlinear regression methods in the forecasting engine are time-series regression methods that use historical data from a single measure.

Expert Opinion
As the time horizon for the forecast moves further out into the future, expert opinion becomes the most reliable predictor. The experts, who are usually corporate executives, have their fingers on the pulse of myriad factors that may influence future performance, such as the general direction of the market and plans for new products. Customer surveys also provide input to longterm forecasts. An equivalent computer model to rival expert opinion for long-term forecasts would be too complex to generate within a usable time frame.

Forecasting Tools Used for Production Planning

As the name suggests, production planning is simply a projection of future business--or production--activity. Many companies rely on certain mainstay products to drive company profits. To ensure that those products continue to serve its purpose, the company must rely on effective production planning. The process of production planning contains a number of steps, and forecasting is an essential step, because it requires that the company project its production needs into the future. Forecasting might seem purely hypothetical, but companies can utilize forecasting tools to provide as much accuracy as possible.
Step 1

Select a method of forecasting for production planning. Available methods include the moving average, exponential smoothing and regression analysis. The moving average takes into account production averages over a period of time and looks specifically at the average of each production period against how that average has changed. Exponential smoothing weighs the average of the most recent forecast against the current demand for the product. Regression analysis uses a chart to view the moving average as a single line of change over time.
Step 2

Determine a time period to study. Forecasting is most effective over the short term, rather than the long term. This is because long-term forecasting can quickly become inaccurate when customer demand changes or market trends adjust unexpectedly. The best time period will reflect previous company activity and what changes the company has seen over timequarterly, biannually. Bear in mind that the best forecasts for production planning tend to reflect shorter amounts of time.
Step 3

Choose reports on previous company activity to help with projecting future production. Projecting for the future requires looking into the past, and companies can utilize previous production results to make forecasts for the future. Companies can look at specifics for customer demand over certain periods of timefor instance, if demand drops during some months and rises during othersand apply this information to the forecasting method that has been selected.
Step 4

Pick market trends to apply to the forecast. Market trends must work alongside expectations of customer demand. The market will play a role in dictating the extent to which customer demand will increase or decrease. If trends indicate that the market for a certain product is about to expand, the company might use this to increase production, but if trends indicate a decrease in market interest, the company might reconsider production needs.