I’ve been invited to give the keynote lecture at the Society for Nonlinear Dynamics and Econometrics Annual Symposium next week at the Federal Reserve Bank of San Francisco. I plan on presenting results of some ongoing research with Mike Owyang of the Federal Reserve Bank of St. Louis on regional propagation of business cycles. This is the first of two posts that describe some of our findings about how recessions differ and seem to propagate across different states and regions.
The building block for this research is the same basic framework used to characterize expansions and contractions by the Econbrowser recession probability index. While that index is based on quarterly national GDP growth, Mike and I are looking instead at the quarterly growth rate of employment at the level of individual states. Our starting point was to replicate and update the results from an earlier paper, Business Cycle Phases in U.S. States, which Mike published with Jeremy Piger and Howard Wall in the Review of Economics and Statistics in 2005. The original Owyang, Piger and Wall piece essentially just implemented for state-level employment growth the same algorithm that I apply to quarterly GDP to get the regularly reported Econbrowser recession probability index. They conducted this analysis on each state individually, one at a time, without trying to use information from other states.
What one finds in that exercise is that recessions and individual states can look pretty different from each other. For example, the graph below displays the recession probabilities for individual states around the time of the most recent recession. Dark blue indicates strong evidence of expansion, while the brighter the shade of green, the stronger is the statistical confidence that the indicated state was in recession at that time. The figure dynamically displays the changes between 1999:Q3 and 2001:Q4 (and yes, I’m proud of myself for figuring out how to create these pictures).
What one sees in that episode is some initial weakness in the lower Mississippi basin, which had spread throughout the southeast and much of the upper Mississippi by 2000:Q4. Remember that, according to the NBER, the recession did not begin at the national level until 2001:Q1. As 2001 progressed it came to affect almost every state, though a few states in the Midwest and Rocky Mountains seem to have avoided it, even at the 2001:Q3 trough.
On the other hand, there are other interesting episodes such as 1985, when the collapse in world oil prices produced a boom for most of the U.S., but shows up as a regional recession in the key oil-producing states (such as Louisiana, Oklahoma, North Dakota, Texas, and Montana) which later spread to their immediate neighbors.
The intriguing thing about these pictures is that they seem to suggest some clear patterns and relations between the states, even though there was no assumption of any such structure in the estimation, which ignored everything about Louisiana in forming an inference about Texas. Mike and I have been looking at a broader statistical characterization of how the recessions in different states appear to interact with each other, using the full panel of data from all 48 contiguous states to form an inference about what is going in any given state at any given date. Next week I’ll describe how we did that and some of what we found.