Time series analysis is as vast a topic as time itself, and as such, there is far too much information to cover in one article. This article is the first part in a series to go over the many intricacies of time series analysis. Read part two to learn about time series analysis and how to build a data pipeline into your business.

Authored by Arthur Wang and Kolin Konjura

Modern growth in computational processes impacts big data and artificial intelligence (AI) which directly benefit time series analysis. With more data, increasingly efficient computation, and greater processing power capabilities; leveraging time series methods to solve business problems through forecasting and process modeling has increased. Through these improvements, business owners have a unique opportunity to harness their data through new and old time series analysis processes to better anticipate future customer insights and market trends while making business decisions.

What are time series?

A time series is data measured in chronological order. The intervals of measurement can be regular or irregular, and can vary in length. For example, time series can be monthly, weekly, daily, or even to the second. Much of the data we collect are time series, such as stock prices, machine telemetry, and temperature readings over an established timeframe. Additionally, granular data can be converted to a time series; for example, individual clicks on a website can be aggregated to measure daily website activity. Time series analysis includes several techniques and models designed for this kind of data. The growth of big data has created increasing opportunities to leverage time series analyses.

What is time series analysis?

Time series analysis is a set of methods that allow us to understand time series data. This can be as simple as understanding when seasonal trends occur, but usually involves deeper analysis to build a more complete model of the underlying processes. Building and implementing a time series model describes a fundamental correlation within the data.  Being able to identify these correlations generates insight into the data that can be leveraged to forecast and predict future values.

Why use time series analysis?

Time series analysis is especially useful for forecasting. In forecasting, time series methods often perform better than simpler methods, such as linear regression or simple extrapolation of historical data. Time series methods capture underlying structures that are common to time series data, such as trend and seasonality (1). These aspects of time series data can violate the assumptions of simpler methods, making the simpler methods potentially less powerful. By accounting for common real-world conditions, time series analysis produces forecasts that are generally more accurate.

"By accounting for common real-world conditions, time series analysis produces forecasts that are generally more accurate."

Putting time series analysis in action

Imagine a company is researching the household energy levels in various neighborhoods to identify those that do not meet green energy requirements. This company determines that an individual’s daily usage is higher during the day and lower during the night, which produces a cyclical pattern within the data. Now suppose the city the company is researching is in Minnesota. The Minnesota climate is relatively mild in the summers, but unbearably cold in late fall, winter, and early spring. The data may show that household energy levels spike during those cold months, and level out during the summer months. The data then shows cyclical trends throughout the day and seasonal trends throughout the year. Using time series analysis for modeling and predicting this situation would work well.

Now imagine a supply distribution company needs to understand local traffic patterns for their supply lines. The company needs to determine the most efficient times to move their product, avoiding things like rush hour traffic or sporting event traffic. If they are moving product on Fridays, rush hour may start at 2 pm instead of 4 pm or 5 pm. If it’s football season, there could more traffic on Monday and Thursday evenings than there is on Friday. A time series model can help the distribution company find the optimal time to move their product with efficiency.


The nature of time series data can lead to complex and arduous calculations that can make forecasting difficult. Luckily, with modern computational capabilities, implementing a time series forecast can be done with relative ease. Languages such as Python and R have ready-built packages that allow analysts to create and tune accurate time series forecasts with relatively little effort (2).

In addition to providing accurate predictive capabilities, time series models have minimal assumptions, are statistically rigorous, and most importantly, are stable. This means that if a large, unexpected event occurs, the model can react and provide insight to solutions around that event. For example, the COVID-19 pandemic has disrupted operations in several industries; time series methods are able to adjust forecasts and analysis with the new climate in mind.


As with most advanced techniques, a time series analysis is only as good as the data itself. If the historical data is sparse, then the forecasting will not be fruitful and a machine learning approach technique may be better in that instance.

The more complex the data is, the more difficult it can be to implement a time series forecast model. In these instances, it may be that a time series model is only part of the solution (3). Referring back to the supply distribution example previously explored, time series analysis may give recommendations for optimal driving time, but those times could be dependent on when the supply is ready to be shipped, inclement weather, or unforeseen accidents on the road.

The landscape of time series analysis

Today, time series analysis is a mix of proven and emerging techniques. Advances in computing have enabled rapid exploration of different models in order to select the one that performs best, rather than painstakingly selecting one up front (4). Greater processing power and developments in graphics processing units (GPUs) have also enabled the application of deep learning to time series forecasting, an approach that has recently outstripped traditional methods (5). Businesses today have a greater range of choices than ever before in how they generate powerful forecasts. Our Baker Tilly Digital professionals are knowledgeable and prepared to assist your business to determine how to implement time series analysis while leveraging your current data and keeping your future business goal at the forefront of every decision.

1.     https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc4.htm

2.     Time Series Analysis with Applications in R by Jonathan D. Cryer and Kung-Sik Chan

3.     Time Series Analysis with Applications in R by Jonathan D. Cryer and Kung-Sik Chan

4.     https://towardsdatascience.com/introducing-pytorch-forecasting-64de99b9ef46

5.     https://robjhyndman.com/hyndsight/tscv/

Contact us

arrowCreated with Sketch.
Inside the Lincoln Memorial
Next up

Treasury guidance disallows deductions related to PPP loan proceeds if forgiveness expected