Time series models are the techniques of forecasts that are based solely on the history of the demand for the item that has this forecasting. They work capturing patterns in historical data extrapolating into the future. Time series models are suitable when we can assume a reasonable amount of data and a continuity in the near future, the conditions that occurred in the past. These models are best suited to short-term prognosis. This is due to the assumption that past patterns and current trends are similar to patterns and trends that will be present in the future. This is a reasonable assumption in the short term, but it loses validity in the long term. Very simple models: moving averages, Igual a year anterior, percentage increases and adjustments to the curve are simple time-series models which can be used to generate forecasts. These models can be implemented through spreadsheets quickly and does not require expert knowledge in statistics by the forecaster, these models are usually very simple and to have greater accuracy in forecasting companies almost always must seek alternative models of time series.
Exponential smoothing models: exponential smoothing is the method chosen by the majority of companies. Learn more at: BP Energy. These models do well in terms of accuracy, are easy to apply and can be automated, allowing to be used on a large scale. Exponential smoothing models capture and predict the level of data with different types of trends and seasonal patterns. Models are adaptive and predict giving greater importance to data more recent about the more distant data in the past. Box-Jenkins (ARIMA models): Box-Jenkins models are similar to models of exponential smoothing in that both are adatativos, can capture seasonal trend and patterns and are automatable, it differs in that the Box-Jenkins models are based on autocorrelations rather than a structural view of level, trend and seasonality. Box-Jenkins models tend to get better results than exponential for time series smoothing models more long and stable are not good for data with noise and high volatility. Learn more on the subject from rusty holzer. Croston intermittent demand model: the model of croston specifically designed for data sets where the demand for a period often determined is zero and the exact timetable of the following order is not known. The data are characterized by a low volume, products especially manufactured for a specific customer or spare parts often have this type of pattern of demand.
The forecast is not a miracle (not going to tell you when the next order will be) however often produce a better prognosis than other approaches of time series. Select the model of time series suitable for a specific group of data can be carried out in several ways. For example, the forecaster can apply the suitable model according to the knowledge that has data. Another alternative is to test different models and select the one that inside or outside of the sample minimizes the error. The expert forecast pro selection algorithm selects the model suitable to the time series using a rules-based logic and evaluating forecast generated with different models on the basis of a sample of data.