· CarMax (KMX)
o Funny that every analyst wanted to ask about subprime on the earning call. I get that KMX is arguably a finance company. But guys, falling subprime mix is a GOOD thing!! So what if revenue slows a little bit, to the extent that customer base is more sustainable, that’s good news.
o KMX does look expensive even with today’s drop off. From lenders perspective though, it is good to hear that subprime players are tightening standards.
· Oracle (ORCL)
o This Barron’s article said that Oracle is threatened by Hadoop. Ironic considering that Oracle oversees Java – the language that Hadoop is written in. I have also heard that Oracle is hurt by freely available database options. Well, ORCL also owns MySQL, one of the most popular free databases. If Hadoop and free databases are really the downfall of ORCL, this needs to be a business school study on making your stuff open-source and freely available.
o Hadoop does not replace a database. The Hadoop wiki says as much. Hadoop is great for unstructured data (for example if you’re mining terabytes of tweets) while traditional relational databases are good for structured data (in the row/column format). Hadoop is just a way to split up your job to various computing and data resources, each of those could be different form of data storage, including a database. In fact, Hadoop and database can be complementary - there’s just so much information in databases that someone mining data will have to link up Hadoop with relational databases.
· Follow up on last week’s post about housing stock – how to find true vacancy numbers?
o I can’t stress enough that the reported “homeowner vacancy” and “rental vacancy” numbers are just fake. There are substantial “other vacancies” that are not included in those numbers – easily 30% or even 50% of total vacancies depending on location. Here’s a helpful report that US Census put out on “Other” vacancies last year.
o How to find the true vacancy number then!? Those numbers are available in the American Community Survey. Unfortunately the US Census does not make this easy. To be useful you’re really looking for local statistics. FactFinders allows you to get this by entering the MSA’s one by one. But if you want to figure out say vacancies for say, all the exposures of some home builder, then this will take a LONG time.
o Ideally you want historical time series for each local level so you get a sense of “normal”. You can try to download the ACS summary files, but those only go back to 2005 on the Census website.
o One way to do this is with Public Use Microdata Samples (PUMS). The U. of Minnesota has a great site that let you select the variables and the vintage years you want. Load that into a database (it's easily in gigs of data) then process it however you want. The results will not match the ACS summary data exactly because these are samples of the original survey. But at least you get a sense of the vacancy mixes going further back than 2005.