Research On Short Time Series Forecasting And Analysis

Forecasting is a process of digging out the underlying rules of future things on the basis of observing and analysing of the objects during the history and former status. The theories and methods of forecasting can be widely applied into various fields of both nature and society, developed into subjects like social forecasting, weather forecasting, medical forecasting, biology forecasting, disaster forecasting, military forecasting, economic forecasting, etc. In market economy, economic activities are changing all the time, with high complexity and uncertainty. In order to get rid of the risks of decision-making and improve management and foreseeable ability, people pay more and more attention to economic forecasting. Either the macroscopical or microcosmic decision-making is closely attached to reasonable economic forecasting. Forecasting and decision-making are two important components of management. The key of management lies in decision-making and its precondition is forecasting. In commercial circumstance, leaders need to make various kinds of decisions which mean a lot to the success of their corporations. Market demand, productive ability and other aspects need to be forecasted during operation. Via forecasting, leaders could understand the supply-demand relationship, grasp the direction and trend of market, further adjust the competitive stategies, production scale and gain more benefits.Before the emerging of modern forecating methods and computer technology, managers carried out forecasting only based on their subjective judgement which absolutely lacked accuracy compared to quantitive forecasting technology. As information technology develops and computers come into common life, forecasting methods have evolved to make it possible that massive data can be stored, gathered, processed and analysed. As a vital part of quantitive forecasting, time series analysis has been highly developed during the last decades, and shaped into mature mechanism of forecasting. Time series is a serial data of observing results according to time sequence. Most of data that are gathered by common corporations are time series, like Earning Per Share, text message income per month, revenue of voice service per month and so on. In recent years, as the rapid growing of imformation software and hardward industry, data analysis is playing more and more important role in decision-making process of government and corporation. Time series analysis is widely used by many fields. The general purposes mainly include:1. Forecasting future observing data through the relativity in unique time series.2. Understanding the relativity between data in series through analysing several interrelated series to improve the accuracy of forecasting.3. Dividing the series into several main parts (trend ingredients, seasonal factors, and circulation factors, irregular factors) to understand the dynamic action of series.4. Checking the similarity of theoretical patterns and series to discuss the accuracy of pattern demonstating the phenomenon.5. Estimating the influence of special policy or events.We propose a procedure to forecast short seasonal time series. It is a modification of method developed for forecasting series with stable seasonal patterns. The new method is motivated by the observations that seasonal patterns may be evolving over time and that short time series arise in many situations. The new method would be more effective when seasonal patterns are not stable and only a small amount of data is available. Real data analyzed in the literatures and new data from telecommunication industry will be collected and analyzed by the proposed procedure.