Bankruptcy

Chapter 11 Filing Data Are Noisy

By Bob Lawless

On the heels of Monday's post about the January bankruptcy filing rate, I intended to write something insightful about chapter 11 filings in different places around the United States. After playing with the data, there are at least three problems with such a post. First, it is Thursday already--egad--putting us well past the heels, ankles, or even thighs. Second, "insightful" is a high ambition; "mildly interesting" would be better. Third, chapter 11 data are noisy, making it difficult to impossible to say anything about filing rate trends.

The last point might be of interest to those of us who like to look at filing rate trends, and I thought I would give it some elaboration. The graph shows the monthly number of chapter 11 filings by legal entities (such as corporations, partnerships, and LLCs) in four different federal judicial districts. Note that the scales are different in each graph. The filing rates are from the Bankruptcy Data Project at Harvard (BDP), using data supplied by AACER. My original thought was to look at the chapter 11 filing trends in different locales, perhaps comparing these trends to local economic conditions or consumer bankruptcy filing rates. That's just not going to work, at least not without some further data manipulation.

The problem is that the chapter 11 filing figures show tremendous volatility as shown by the spikes in the chart. All of the four districts--and I picked these because they are higher population districts--have sudden and abnormal surges in chapter 11 filings. For example, in August 2008 legal entities filed 215 chapter 11 petitions in the Central District of California (CACB) as compared to only 48 in July and 31 in September. In fact, in no other month during 2008 were there more than 67 legal entity chapter 11 petitions in this judicial district. This volatility would not be a problem if it was capturing some underlying phenomenon, but it's not.

The chapter 11 counts are accurate, but the spikes are not really capturing anything meaningful. Take the example from August 2008 in the CACB. Woodside Homes filed bankruptcy there on August 22, 2008. The debtor was a home builder, and each of its developments was apparently set up as a separate corporation or LLC. Each separate corporation filed its own chapter 11. This one economic unit produced a total of 188 chapter 11s. I know this because I looked at the court docket for this month.

The problem of noisy chapter 11 data is not limited to just analyses done at the local level. In August 2008, there were only 1,090 chapter 11 cases across the entire country. Thus, the extra 187 chapter 11 cases in the Woodside Homes filing inflated the number of chapter 11s that otherwise would have been reported that month by 20% (187/903). People do pay attention. Do a Google search, and you'll find some blog and news references to the surge of chapter 11 filings in August 2008. There was no surge, just a statistical artifact of a company with 187 affiliate filings.

Anyone who uses the chapter 11 filing data needs to be aware of how businesses with many subsidiaries can create artificial spikes in the data. To eliminate the noise, of course, the best solution would be to look at the underlying court dockets and not count subsidiaries as separate filings. That is not an intractable solution, but it also requires a nontrivial amount of time and perhaps cost. One also might just aggregate the data into larger units, perhaps looking at filings by quarters or over large geographic units like circuits. This solution is cheaper but loses information, and may be impracticable given the research question. If you're using the chapter 11 data, just be careful out there.

Full post as published by Credit Slips on February 11, 2010 (boomark / email).