More from our BLS linear regression series

In December 2020, we launched our BLS linear regression series of blog posts exploring how to use LinkUp job data to calculate nonfarm payroll (NFP) predictions. In our last post predicting the BLS NFP, we used a simple seasonal adjustment to improve our predictions. As with our original prediction, we were able to come closer to the reported number than consensus. We will again build off this, using a more advanced method of seasonal adjustment.

We started with the raw data set from which we filtered down in our original blog post published in December, and expanded upon in January. This time instead of using the manual seasonal adjustment method utilized in the prior post, we will be using the stats model seasonal_decompose function. First, let’s loop through all of the different features we have created to see who correlates the most with our BLS set.

It appears the average active change is still working the best for us. An added benefit to using the stats model function is that we can see what trend and seasonality we are removing.

We will take the residual and use that for our independent set. This gives us an output of 938,401 jobs being added. This is likely not not going to be the case, but it does look like recent data points have started to move back towards the numbers we are accustomed to seeing. Let’s attempt to remove some of the outliers that could be skewing the results.

This levels out our regression line, compared to what was observed in previous posts.

This gives us a prediction of 251,995 jobs added. A number that is certainly good news for the economy, if it is close to being correct.

You can find previous posts from our BLS linear regression tutorial series here:

Using simple linear regression for BLS estimate (12/3)

Using average active linear regression for BLS estimate (1/7)

And if you’re interested in the data behind this tutorial, please contact us to learn more.