Time series analysis and forecasting explained


A time series essentially is a series of quantitative values. These values are obtained over time, and often have equal time intervals between them. These intervals can be quite different and may consist of yearly, quarterly, monthly or hourly buckets for instance.

Some may argue a time series is not a ‘must have’ for proper forecasting. A proper time series however does provide an important contribution to a more accurate forecasting. The availability of the right time series makes all the difference.

Two or more years of data should normally suffice as a rule of thumb. More data does not necessarily equate to better or the right input. When assessing and forecasting the of hourly inbound call volume for a call center for example, a proper forecast may rely on ‘only’ 3 to 4 weeks of data. The more specific the time series, the better any forecast generally would be.

Examples of time series are the number of units sold or the closing value of Nasdaq. The demand of one of your products and the exchange rate of the British Pound Sterling vs the American Dollar are two more examples of time series.

Data points taken over time as a time series analysis may have an internal structure that should be accounted for. This internal structure maybe an autocorrelation, trend, or seasonal variation.

What are time series analyses used for?

The use of time series models is twofold:

  1. Get an understanding of the factors and structure that produced the observed data
  2. Fit a model and proceed to forecasting, monitoring, or even feedback and feedforward control.

Time Series Analysis is used for many applications such as:

  • Economic Forecasting
  • Sales Forecasting
  • Budgetary Analysis
  • Stock Market Analysis
  • Yield Projections
  • Inventory assessments
  • Workload projections
  • Demographics projections
  • Weather patterns and forecasts

We will focus on time series’ forecasting purposes in this article. Time-series analyses can provide better insights for business decisions in not just inventory control yet also in purchasing, manufacturing, logistics, and more.

Ways to do a time series analysis

If you have your data in Excel, you can make several time-series analyses based on the data available. The Single Moving Average and Centered Moving Average are just two examples of how to undertake a time series analysis.

Time-series analyses can be either univariate of multivariate. ‘Univariate time series’ refers to a time series that consists of single observations recorded over equal time intervals in a sequential manner. ‘Multivariate time series’ therefore are based on multiple observations.

Uni variate time series can be modeled in various ways. One approach is to decompose the time series into both a trend, seasonal and a residual component. These factors may vary widely per time series.

For instance, a time series may be seen as stationary, without any significant seasonal influence. With a stationary process the mean, variance, and autocorrelation structure do not change over time.

Often their seasonality i.e. periodic fluctuations will apply. For example, retail sales generally will show a strong peak in the Christmas season and then decline after the holidays.

Whilst regression analysis may seem a perfect example of time series analysis, it is not.

In regression analysis theories are often tested of the values of one or more independent time series affecting the current value of another time series.

Time series data have a natural ordering of time. Therefore so-called cross-sectional studies are not an example of time series analysis either. In a cross-sectional study, there is no natural ordering of the observations. Think for instance of the ordering of customers' location by reference to their individual education levels: the individuals' data could be entered in any order.

From time series analysis to time series forecasting

It’s often easier and more accurate to forecast for a shorter time horizon compared to a longer horizon. The further the point in time the less accurate forecasts usually get. If you want, you can frequently update your statistical model as you gain more new information that may help to make more accurate forecasts.

The use of time series analysis is a helpful instrument in forecasting. Mere time series analysis crunch time series data in order to extract meaningful statistics and other elements of the data.

Time series forecasting goes beyond ‘just’ time series analysis. With time series forecasting a model is being used to predict future values based on previously observed values over time.