Get startedGet started for free

Fundamentals of time series data

1. Fundamentals of time series data

Hi, welcome to this course about time series analysis in Power BI. My name is Kevin, and in this video, we will cover the fundamentals of time series data.

2. Course pre-requisites

Before we jump into the course, we want to make sure we are level-set on the requirements for the exercises that follow. We will be using a variety of more intermediate functions and calculations, such as CALCULATE, IF statements, and SUMMARIZE. If you are not familiar with these functions, we recommend going to the Intermediate DAX course on the DataCamp website before continuing.

3. Why is it important?

Before we dive into specific aspects of time series data, let's take a step back and ask ourselves why this is important. Time is a dimension in every dataset, whether we only have data for a snapshot or data across several years. Looking at the same observations over time can yield some interesting insights, such as how has my data changed over time? Based on historical data, what should my data look like? What will it look like in the future?

4. Defining time series data

Time Series Data is a set of observations (i.e. data points) that have been collected about the same subject over a period of time. For example, this could be the temperature at a particular location over a month. Generally speaking, there are two main components to the time dimension we need to take into consideration: span and interval. Span is the total time period we have data for, while interval is the amount of time between each observation.

5. Tracking changes over time

Time Series datasets have variation, just like any other dataset. The kinds of variation that a particular dataset has can fall into various patterns. Secular variation can be thought of as a global trend in variation. This, most commonly, is shown through a consistently increasing (or decreasing) value over time. Seasonal variation will have a regular pattern that lines up with the seasons of the year. Cyclical variation has a regular pattern year over year but does not necessarily correspond to a season. These datasets have similar behavior at each point of the year. Random variation is a catch-all category, as it will capture any dataset that does not display a clear regular variation pattern like the others. In reality, many datasets will have some randomness to their variation.

6. Short term time series analysis

When analyzing data over time, we need to remember the total time span we are interested in. Short-term analyses look at data within the last year. There are two common groups of analyses: 1) Analyzing data over the last x periods. For example, last ninety days, last two quarters, etc. 2) Analyzing the current period to date, where year to date is the most common approach. In the example we have here, we have both an analysis where we calculate just the last 2 months of data, but also show what the year to date value is. Both of these analyses can show important information, depending on what question you are asking.

7. Long term time series analysis

The second set of analyses falls into a long-term analysis, which typically spans at least one year. With these analyses, we are trying to understand historical data and if it was a good predictor of where we are currently. An common application of this analysis is calculating year over year change for a particular metric. We can see in this chart that for the month of January, we can observe a measureable change in value, which can provide us some interesting insights into company performance.

8. Forecasting the future

We go beyond descriptive analyses at the other end of time series data and move into predictive analyses. This forecasting analysis can be very powerful, using historical data to model the future. Examples of forecasting range from basic linear modeling all the way to advanced Machine Learning techniques. An example provided here shows how one would forecast the quantity of stock trades for a particular company, showing that even the noisiest, fast-moving datasets can be forecasted with the proper methods.

9. Let's practice!

So with that, let's dive into some exercises to solidify our understanding of time series data.