We've already seen a lot of bs numbers in the cramdown debate. The Mortgage Bankers Association keeps pushing its ridiculous figures. And now Todd Zywicki has joined the fray with an op-ed in the Wall Street Journal a couple of weeks ago.

Professor Zywicki that claimed that "A recent staff report by the Federal Reserve Bank of New York estimated a 265 basis-point reduction on average in auto loan terms as a result of the reform."

One little problem. That's not what the Fed staff report found. Professor Zywicki was off by 250 basis points (a doozy of a mistake!), as well inserting a causal link not supported (and arguably contradicted) by the Fed staff's study. The study states that "The decline in the average auto loan spread was 15 basis points lower after BAR for unlimited exemption states, a 5.7 percent decline relative to the mean over all states (265 basis points)." In other words, the average rate spread is 265 bp. The decline in rates, to which Zywicki was referring was only 15bp, and that was only in states with an unlimited homestead exemption.  That it was not 265 bp was abundantly clear from the regression tables.

But that's not all. It's not as if Professor Zywicki simply mistook a 15 bp drop for a 265 bp drop.  That 15 bp isn't what it appears to be.  The study used two statistical specifications and looked at states with limited and states with unlimited homestead exemption to see what impact there was on auto loan rates post-BAPCPA, which enacted an anti-auto cramdown provision (the infamous "hanging paragraph" that says that there's no bifurcation of claims for cars purchased primarily for personal use in the previous 910 days).

In one specification it found nothing with statistical significance regardless of the homestead exemption level, which means that it couldn't rule out the possibility that the change in rates was random.

In the other specification, post-BAPCPA there was a marginally statistically significant 15 bp drop in five-year auto loan rates in states with unlimited homestead exemptions. There was no statistical significance in the drop in other states. What's funny about this is that homestead exemptions have no bearing in Chapter 13–exemptions are only available in Chapter 7. So if the study had aggregated all states for its regression, it seems unlikely that it would have gotten stronger statistical significance.

So we have at best very weak evidence of a 15bp drop in rates. But it doesn't follow that the drop was due to the anti-auto-cramdown provision. The study also found a significant decline in auto-loan delinquencies in the short period after BAPCPA. The most plausible story, I think, is that surge in bankruptcy filings before BAPCPA's effective date cleared out the pipeline of troubled loans so that post-BAPCPA auto loan default rates were lower. My guess is that they've climbed right back up.  Notice that this has nothing to do with cramdown. This has to do with moving forward some filings that would have happened later. So we have a 15bp drop that might not even be statistically significant and only in some specifications and only for states with unlimited homestead exemptions, and it probably isn't attributable (or at least most of it) to the anti-cramdown provision, but instead to BAPCPA causing a filing pile-up. So where did Professor Zywicki get this 265 basis point number from? That's the spread that exists between five year auto loans and five year Treasuries. It has nothing to do with bankruptcy.

Sometimes a little common sense is needed when looking at numbers, too. In December 2005, auto loan rates were at around 6.63% (663 bp). If 265 bp was right, it would have been a 40% decrease in auto loan rates! Whatever impact bankruptcy has on credit costs, I don't think there's anyone who could honestly argue that 40% of the cost of auto loans is due to the ability to cram down loans on cars purchased primarily for personal use within the previous 910-days with a purchase money security interest. There just aren't that many folks filing for Chapter 13 bankruptcy, much less who fit into this particular set of circumstances, to have this kind of impact on pricing, regardless of the loss severities.

Yet another case of baloney numbers shaping the bankruptcy debate. I hope the WSJ runs a correction on this.  Now there's some fact-checking for you. 

[Update 3.6.09:
Based on correspondence with Don Morgan, one of the NY Fed study's authors and Professor Zywicki, a few new points emerge:

First, I misread the study too.  The 15bp finding is in a regression that measure the "difference-in-differences" in the spread between auto loans and Treasuries pre- and post-BAPCPA for states with and without unlimited homestead exemptions.  The study does not report the post-BAPCPA rate drop in auto loan rates.  The author, however, tells me that it turns out to be 46-56 bps, and to have strong statistical significance. So let the record stand corrected on this.  

It's possible that BAPCPA resulted in lower auto loan rates.  But in order to make a reasonable causation argument, one must first explain the similar or larger rate drops in 2000-2001 and in 2003 and in 2007 that have nothing to do with BAPCPA.  Otherwise, the causal argument is reduced to the fallacious post hoc ergo propter hoc variety.  

The chart below, taken from the NY Fed study shows with the solid and dotted lines the spread between auto loan rates and Treasury's for states with and states without unlimited homestead exemptions.  They move in sync, and they clearly fall after BAPCPA.  But they also fall equally sharply before and after BAPCPA.  Auto loan rate spreads over Treasury jump around a lot, and the mere fact that they fell after BAPCPA doesn't prove anything.    

(fwiw, Chart 5 appears to be incorrectly labelled in the study.  The study says that the "Left axis measures interest rate on new automobile loan (5 year) minus rate on government bond (5 year)."  If so, then 15bps would appear to be roughly the right measure.  Instead, the rate spread must be the right axis in bps, and the left axis must be measuring the difference in the auto-loan-treasury spread between limited and unlimited homestead exemption states.)  

The problems with Professor Zywicki's causality argument don't end there.  Any causality argument must also distinguish between general impacts of BAPCPA (e.g, delinquency pipeline cleared out) and the auto-cramdown provision.  This type of event study cannot provide support for that.  The rate drop could be due to the hanging paragraph, but there's no responsible way to make that claim without addressing these other factors, and the NY Fed study doesn't attempt to do that. The fact that Professor Zywicki was off by 209-219 bps, rather than by a full 250 bps (something he couldn't have known from the study) doesn't absolve him of making an untenable causal claim. 

The  bankruptcy policy debate should happen on the basis of the best possible evidence.  If more restrictive bankruptcy laws result in cheaper credit, that's a very important policy consideration, and for the integrity of the policy debate, we need to be working off the best numbers available. I've updated this post to make sure that the correct numbers are clear.  I'm still hopeful that Professor Zywicki will make clear that he doesn't stand by either his 265 bp claim or his untenable claim of causality.]

[Updated 3.7.09

Professor Zywicki has corrected on the 265 bp claim.  He still seems to be making causal assertions, however, such as that the study finds "the impact of eliminating cramdown was a reduction in
interest rates of 56 or 46 basis points."  That's not quite right.  The study can't test the elimination of cramdown; it can only test the impact of BAPCPA as a whole.  In fairness, Zywicki later refers to the study finding the impact of BAPCPA, rather than the specific cramdown provision.  Regardless, Professor Zywicki still has no response to all of the equally large jumps up and down in the auto loan rate to Treasuries before and after BAPCPA, which casts serious doubt on any causal story.]