Time series forecasting for decision makers

Time series analysis is key to understanding data to make the most informed decisions. Data points across time give analysts the ability to look backward as well as forward to identify trends and drill deep for a sweeping analysis. Time series forecasting makes predictions for future data and outcomes based on time-stamped past data collected over specified intervals of time. Unlike other data, time series data can show clear changes over time with an extensive data set.

Dynamic vs. static forecasting: Which is right for me and why?

One of the biggest decisions in forecasting is whether to choose a static or a dynamic model. Static models take a collection of historical data and, based on it alone, predict the future. Dynamic models take the static predictions and update them in real time with new data as it is logged.

Dynamic modeling is often considerably more complex and expensive than static modeling. In industries that operate on thin margins, dynamic forecasts may simply cost more than they are worth relative to static models. Additionally, it may not be feasible to adjust for the changing predictions of a dynamic forecast. 

Consider building a complex manufacturing facility to serve a large industrial process. Knowing small shifts in demand relative to original expectations may not help when most of the infrastructure to accommodate the process is already in place. It makes more sense to continue building the last small part of a facility for the original expectations in case the shift in the forecast is wrong because the savings in not finishing the last parts are small and even if demand does shift as expected, perhaps the excess capacity could be rented out to pay for costs or repurposed later.

However, this doesn’t mean static forecasts are always the way to go. It is also often the case that dynamic forecasts can be a significant improvement on static forecasts. In a lean supply chain, subtle shifts in demand can be important to know even if the data to discover them is only learned the day before, which can only be found using dynamic forecasting. Perhaps the most famous application of dynamic forecasting is in weather forecasting where up-to-the-second data informs the models depended upon by countless industries and individuals who would suffer catastrophically from statically forecast weather.

What makes time series forecasting a unique tool for decision makers?

A defining feature of time series forecasting is the type of time data that is utilized for analysis. In time series forecasting, measurements are taken at regular intervals over a length of time. While other techniques also record when events happen and include consideration of time in their analysis, time series forecasting requires measurements to be taken continuously and at regular intervals throughout. With this guaranteed regularity of data, analysis can find much more subtle patterns that random sampling probably would not be fortunate enough to catch purely by chance. 

The regularity also helps avoid “dark periods” where no data–or too little–happens to be collected, which leaves less chance that something important is hiding in the under-studied period.

How are events understood with time series forecasting?

Trend Analysis
Trend analysis searches for general changes or trends over time. This type of analysis can determine the underlying cause of the trend, but this is not necessary in all cases. Deterministic trends have an identifiable underlying cause, while stochastic trends are random and unexplainable by the data. 

Functional Analysis
Functional analysis understands data by searching for signs of notable events. This is the most common type of analysis.

Seasonal Variation
Seasonal variations are events occurring throughout the data at marked intervals during the year or any repeating unit of time. 

How do decision makers use time series forecasting?

A wide range of industries use time series forecasting every day to make concrete scientific predictions in their business. Some examples induce:

Financial forecasting is critical in multiple industries. Stock market analysts can study historic data to predict the closing price of a stock every day based on seasonal variation and trend analysis. 

Health forecasting is a valuable tool for predicting future health-related events like demands for services, preventative medicine, and healthcare intervention strategies. Monitoring the general population’s health with demographic data points can generate very useful predictions in healthcare forecasting. 

Engineering forecasting touches in a real way every single day. Any product that could harm someone if it failed has been designed and tested using time series techniques to ensure that it does not fail unexpectedly. Without this tool, countless activities would be drastically more dangerous, more expensive, or both, such as traveling in a vehicle, living indoors, cooking food, or wearing shoes–all of which involve products that benefit from this type of safety testing.

Some outcomes, unfortunately, can never be accurately predicted, no matter the data quality. There is plentiful weekly data nationwide about lottery winners, but no historical data could ever predict the next winner or even the next winning numbers. 

What should decision makers look for in time series forecasting tools?

Quality time series forecasting tools are flexible and can not only process existing data but also update as new data arrives. Tools should be able to understand both general trends and cyclical patterns as well as leverage all your existing analytical infrastructure. While time series forecasting isn’t infallible, it all depends on your time series data analysis.

Domo’s time series forecasting tools are the perfect extension of a world-class data package. Create complex forecasting charts over any data set within just a few clicks, leveraging the benefits of time series forecasting across your entire organization. With Domo’s advanced data governance and rules-based access, every person in the business is able to access the data they need to make data-driven forecasting decisions.

How will time series forecasting affect the future of decision making?

Data science has undergone several revolutions in recent years. The advent of big data brought new insights gleaned from large data sets that went unnoticed without access to large, complex data sets that many organizations have been collecting for other reasons entirely. The expansion of that set of tools to consider exact times as part of their modeling was the natural evolution of the field. 

Now, time series forecasting has further advanced our capabilities by no longer accepting whatever data happened to be passively collected, but instead, it specifies the timing and format of the data and with that uniformity. These benefits of controlling the collection of data so carefully will offer as much an advance in understanding as both the big data revolutions before it.