Skip to Content
Author's profile photo Vikas Ohri

Forecasting and Time Series Decomposition

Forecasting is a vital cog in effective business planning be it forecasts of stock requirements or call volumes of a call center.

Obtaining the seasonal , trend and cyclic components or decomposing the time series data can provide insights into the business analysis for enterprise functions varying from sales forecasting, demand planning or customer satisfaction. Sharing a data product which provides these insights.

Learning from the past data patterns to extrapolate the future predictions  and at the same time ignoring random data or data that is unlikely to repeat is the key for useful forecasting.

Time series data like sales data of quantities of items units sold can be regarded as comprising of a seasonal component, a trend-cycle (contains both trend & cycle) component and anything else regarded as remainder component. These components can be model as additive or multiplicative  i.e. they add up or multiply to constitute the time series data.

Time series  data  for example Product Units Sold per month can exhibit a huge variety of trends, seasonality and cyclic patterns  in the data. Business analysts are continuously trying to identify factors causing Change  ( either increase or decrease) in direction of trend.It is useful to extract those patterns to understand the business and improve forecasts. Since Forecast is a prediction of values in future , it is useful to attribute a confidence interval around these values . Shared below is link to application i created for quick visual analysis of time series data . Forecasting of your enterprise data with Exponential smoothing and Seasonal and Trend decomposition using Loess  can be done as well Time series decomposition by this tool.

Past 25 years of monthly utility average price of California have been already included to study the data, from  data source U.S. Energy Information Administration (EIA) .  Monthly trends plots show power prices have increased in summer months by providing the increase or decrease of average monthly values compared to data average. Currently it can also accept, spreadsheet .XLSX format with ,monthly periodic data of uni variate time series for data analysis purposes, as shown in the table picture below.

Product Year Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
A 2014 100 89 87 93 101 115 116 90 85 78 75 60
A 2011 89 90 87 91 102 103 125 96 80 89 71 67
A 2012 102 79 83 95 105 92 78 59 93 71 80 72
A 2013 99 103 78 99 84 78 86 90 73 89 79 64
C 2014 100 89 80 77 90 70 74 63 72 78 85 98
C 2013 95 99 75 76 80 65 68 70 61 74 88 100
C 2011 95 99 75 76 80 65 68 70 61 74 88 100
D 2014 100 89 87 93 101 115 116 90 85 78 75 60

Using Native ECC & Netweaver tools creating time series data like quantities of product sold by time period (say monthly),  as shown in above table  is fairly easy and commonly available in out of the box reports.

Few terms that may help the reader & user are explained below.

A trend exists when there is a long-term increase or decrease in the data. It does not need to be linear. A seasonal pattern exists when a series is influenced by seasonal factors (e.g., the quarter of the year, the month, or day of the week). Seasonality is always of a fixed and known period.A cyclic pattern exists when data exhibit rises and falls that are not of fixed period over a fairly long time periods usually in years. Average length of cycles is longer than the length of a seasonal pattern . Seasonally adjusted data is if the seasonal component is removed from the original data.

Importantly note, although in the data product , exponential forecasting option automatically chooses best fitting model, there are several  models or methods to do forecasting , which one to use is essentially determined by the data characteristics and out of sample accuracy testing .

Assigned Tags

      Be the first to leave a comment
      You must be Logged on to comment or reply to a post.