Forecasting Process Summary

The accumulation of time-stamped data into time series data is based on a particular frequency. For example, time-stamped data can be accumulated to form hourly, daily, weekly, monthly, or yearly time series. Additionally, the method for accumulating the transactions within each time period is based on a particular statistic. For example, the sum, mean, median, minimum, maximum, standard deviation, and other statistics can be used to accumulate the transactions within a particular time period.

For automatic forecasting, accumulation is the most important decision because the software makes most of the remaining decisions. If weekly forecasts of the average of the transactions are needed, then the accumulation frequency should be weekly and the accumulation statistic should be the average.

Accumulating the transactional data on a relatively small time interval can require a long forecast horizon. For example, if the data are accumulated on an hourly basis and if it is desired to forecast one month into the future, the forecast horizon is very long and the width of the confidence limits will be very wide toward the end of the horizon. In this situation, the forecast content or usefulness of the forecast will be low.