Bank Failures: Fun With Curve Fitting

[GliderGuider]

The common wisdom is that the number of US bank failures is going to take a dramatic rise this year. I enjoy messing with Excel, and I wondered if the raw numbers show a mathematical trend (without factoring in any economic fundamentals) that is worth extrapolating out a few months.

To see, I took the data from the FDIC Failed Bank List, sorted it into monthly counts, and graphed the result for the last two years. I found a trend line that fits pretty well (the correlation of 0.86 is fairly good), and extrapolated the curve out six months. From a purely mathematical perspective the trend is very worrying. It will be interesting to revisit this as the year unfolds.

[Zynx]

1. I would caution that it is generally dangerous to extrapolate trends like that.

Note how poorly the past six months have been fitting the trend.

[GliderGuider]

2. I agree. I promise not to base any fiscal policy on it. nt

[Gman]

3. A trend for the last 6 months

looks like it would be headed downward but the R² would be uselessly low.

[Po_d Mainiac]

4. Insert a third line representing the FDIC fund balance...

I'll speculate that...As the fund balance line puts pressure on the zero line, the break of actual failures from the trend curve will occur :shrug:

[leahcim]

5. If you are fitting an exponential...

...I an not impressed. Any exponential will fit, and project either a dramatic increase or dramatic decrease in the future. For that matter, any polynomial will too when extrapolated into the future of the fit data (but slower than an exponential).

The trend line looks like a great fit up until about may 09, and after that it fails to track the data at all.

Also, arguably, the number of insured bank failures is not as relevant as the total amount of insured funds in those banks. By that measure, WaMu is a spike that dwarfs everything else, but it doesn't show here.

[GliderGuider]

6. I agree with all your points.

Idle hands and Excel are the devil's playground. I'm pretty content to see what the real world tells us over the next six months, it's bound to be more interesting than a trendline on a graph. I just get impatient sometimes.

[stevebreeze]

7. Calculated risk does a nice blog entry every fri on bank failures.

http://www.calculatedriskblog.com/