Wondering what the FDIC is doing with all the incoming assets? You’ll see that it doesn’t take a billion dollars to participate, they have sold several bulks under $1,000,000.
Here’s the link to their bulk sales:
http://www2.fdic.gov/closedsales/LoanSales.asp
Thet’ve been selling our residential assets for 37 cents on the dollar on average (actual vs. book value), and commercial is going out at roughly 53 cents on the dollar. A spreadsheet of a sampling of this year’s sales:
(There were more sales added to website after the spreadsheet was completed)
Jim, you’re not averaging correctly. You need to do sum(sales dollar values)/sum(book dollar values). A simple average of the percentage column weights it incorrectly.
The correct values are 30.6c for residential and 51.9c for commercial.
RTC part 2
Sorry about the error, the spreadsheet was sent to me. I’ll make sure to straighten them out.
One can invest in these as an individual if you have $1 million in assets and income of $200k+. Awwww crap I’m too poor to do it. Jim I know you only have 4 minutes of free time a day to waste but cant’ you start a vulture fund for us? Here’s a link to get started with it;)
http://www.debtx.com/
Thanks bleach, someone else directed me to debtx too, so I’ll check it out. It takes sending in an application, then they’ll see about allowing the entity to review their assets for sale.
Sorry Ted but you are wrong. You are making a very common, although critical, error that is caused by it being so easy to do in Excel.
The average is supposed to be calculated on the percentage column. The statement is about the average sale price which is a percentage of the book value. In this case (using the commercial properties), the range is 76.4% to 30.2% of book value with an average of 53.2% and a standard deviation of 14.7%. You have to do this calculation on the sale price percentage [calculated for each property not the whole lot].
What you calculated would be ‘we sold the inventory of properties [assuming this was the entire inventory] at 51.9% of book value’; this is not the average sale.
It is rather interesting that the non-performing gets more money than the sub-performing in both the residential and the commercial. I would currently assume this is due to the small sample sizes.
Looking at residential loan package #9 that LNV Corp bought at 11.4 percent of book value…76 loans for $662K! Wow; that is an average of $8710 per loan (the book value average is 76K per loan).
They must be some crappy houses!
Oh wait, these are in Dallas; that makes it less surprising; but still, wow.
Keith, equal-weighting bundles that have vastly different values and vastly different numbers of properties is really naive.
Yes, you have calculated an average, but it’s a silly average.
I second the silly average concept. If I sell a million dollar property at a 50% discount and a one dollar property at a 25% discount, then the average discount is 37.5%?
In cases of inventory, which method you use depends on what you’re trying to show. Neither method is invalid as a rule, it depends on what your trying to analyze. For example, percentage average is exactly how list price ratios are calculated. For example, houses being sold at 95% list price.
Look at this example:
House 1 LP/SP $500K/$400K
House 2 LP/SP $100K/$120K
House 3 LP/SP $100K/$120K
House 4 LP/SP $100K/$120K
House 5 LP/SP $100K/$120K
House 6 LP/SP $100K/$120K
Combining the raw dollars shows SP = LP (1.00 ratio), combining the percents shows a ratio of 1.13. The statement “Most houses are selling above list price” would be true. Of course, it doesn’t tell the whole story and never will any number.
So far I’m not offering any opinion on which method is better in this particular case. Just pointing out that both methods are statistical standards depending on the data. It would indeed be silly to use one method in the wrong case, or the other method in another case. But there are also many cases where both methods are used and it’s validity is debated.
For accounting purposes of determining profit/loss expectations, you’d tend to use the weighted method. For trend forecasting, you’d tend to use the average of percents method. Just saying.
This is why statisticians are also known as liars.
ted; The bundles were sold as a unit and those units were sold at a percentage of the book value of the unit. If you are going to bring up the individual loans into this, then nothing in this spreadsheet can help us.
Regarding sdbri‘s comment, “Of course, it doesn’t tell the whole story and never will any number.”. That is why I included the standard deviation. The average and the SD tell you a whole lot.
Also in your example, an average of 1.13 times book value with a SD of 16.3 says what needs to be said if talking about the whole lot. But since there is something interesting in your data, you would break it down into above and below some value and show the averages and SDs of those two groups. And hopefully then determine if the groups are significantly different (which they are but we know that since you were making a point with the data set).
And to no bubble here; the answer is yes indeed the average is 37.5% and to be honest you would follow that up with ‘a standard deviation of 17.7%’ and that ‘lower priced properties are selling at a greater discount than the expensive properties’. It isn’t silly at all.
The blog referred to the “bulks” being sold. We have to treat them as single entities and not consider the vast possibilities of the individual loans and thus the average of the sale price percentage is correct.
I do appreciate ted‘s point about the individual loans. Is the Fed bundling with some angle towards ensuring there are some crappy houses in there with some nice houses so that the bundle works out nicely (isn’t this what AIG was doing with their tranches?)? Or are they honest bundling of neighborhoods or at least similar type or value loans?
30.6c is the average recovery the FDIC has gotten on all of the residential loans in the spreadsheet. That, in my opinion, is the most important number. It accurately reflects the dollars recovered (and therefore the dollars lost) on all of these loans. Calculating a simple mean of a handful of bundles does not get you to that number, nor does calculating a standard deviation of the bundles.
A simple mean and standard deviation of the bundles are not even a useful estimate of future sale prices, because recovery rate and bundle size are correlated.
We’re arguing in circles here. One method, the “weighted” or aggregate average, is used by accountants. The other method, the average of percents, is used by investors who want to price future loans.
ted is an accountant and Keith is an investor.
Aggregate average tells an accountant how much money the FDIC is losing as a percentage of their portfolio. Percentage average suggests to a speculator how much to bid on a particular set of loans.
Put another way, aggregate average tells Amazon how much they’ve given away in discounts compared to revenue. But to a consumer, all they care about is the percent discounts they are likely to find on Amazon whether it’s a popular and expensive product (which has a high weight) or a rare and budget item (which has a low weight). That’s because to the consumer, each item has equal relevance when they’re making a decision between buying from Amazon and another consumer.
An investor could care less about how much the FDIC lost on the big loans. He only cares on what the FDIC seems to be discounting on the hundreds of cases they’ve handled. After all, one big loan could have been “given away” by a bad case handler, when all the other loans are more representative of what deals you’re likely to get.
As ted pointed out, both systems are limited in what they can tell you. Which as I’ve generalized, no number can tell you the whole story.
This is why in a lot of analysis (such as house median price), houses are broken up into tiers (such as for depreciation rates). There is a clear and consistent difference between the behavior among the tiers, and that difference is likely a result of the difference in tiers. Yet there is other analysis, such as List Price / Sale Price ratio, where all tiers correlate fairly closely and any difference between the tiers is unlikely due to being in different tiers (correlation is low, and cause and effect is even lower).
The most common mistake in statistics is to draw conclusions out of inconclusive data and second guess the meaning of data. People used to think alcohol *caused* lung cancer because they were very highly correlated. Turns out, the real correlation is drinkers tended to smoke. No cause and effect. For this reason, unless you can show tiering is necessary and meaningful in a situation, don’t do it. At best it’s misleading and presumes a correlation, at worst it will amplify mistaken conclusions.
Again, keep in mind that most statisticians are liars.