Today’s article will showcase a subset of Pandas’ time-series modeling capabilities. I’ll be using financial data to demonstrate the capabilities, however, the functions can be applied to any time-series data (application logs, netflow, bio-metrics, etc). The focus will be on moving or sliding window methods. These dynamic models account for time-dependent changes for any given state in a system whereas steady-state or static models are time-invariant as they naively calculate the system in equilibrium.

In correlation modeling (Pearson, Spearman, or Kendall) we look at the co-movement between the changes in two arrays of data, in this case time-series arrays. A dynamic implementation would include a rolling-correlation that would return a series or array of new data whereas a static implementation would return a single value that represents the correlation “all at once”. This distinction will become clearer with the visualizations below.

Let’s suppose we want to take a look at how SAP and Oracle vary together. One approach is to simply overlay the time-series plots of both the equities. A better method is to utilize a rolling or moving correlation as it can help reveal trends that would otherwise be hard to detect. Let’s take a look below: