Time series has two main goals. The first goal is to identify the nature of the phenomenon that is represented by a sequence of observations and the second is to forecast, which means to predict the future values of the time series variable. The requirement of both these goals is that the pattern of the observed time series data gets identified and described in a formal manner. After having established the pattern it is possible to interpret and integrate it with other data. Irrespective of the understanding that has been developed for the validity of the interpretation of the phenomenon, the identified pattern can be extrapolated for the prediction of the future events.
Identification of the patterns in the time series data:
1. Systematic pattern and random noise: In the time series analysis, it is often assumed that the data comprises systematic pattern which is usually a set of some identifiable components and random noise. This makes it difficult to identify the pattern usually. In most of the time series analysis, for making the pattern more salient, some amount of filtering is surely required.
2. The general aspects of the time series pattern: in most of the time series patterns, they can be described in terms of two basic classes of component. They are trend and seasonality. The first one represents a general systematic linear or non-linear component and the second one may have a formally similar nature. Like for example a plateau that is followed by a long period of exponential growth. These two general classes of time series data can be seen co existing in real life data.
3. Trend Analysis: There are a whole lot of proven techniques that help to identify trend components in the time series data. As long as the trend is monotonous, the data does not become very difficult. However , if there is considerable error in the data, the first step in the process is smoothing. By smoothing, we mean the process of averaging out the data in such a fashion that the non-systematic components are able to cancel out each other.
4. Function fitting: a lot of monotonous time series can be sufficiently approximated by the linear function. It is important to first transform the data to remove non linearity. Mostly a logarithmic, exponential or a polynomial function can be used.
