On the surface, today's New Home Sales number was great (as always tends to happen just before a presidential election): a print of 389K seasonally adjusted annualized units sold in the US (ignoring the 37.3% collapse in the Midwest), which was a 5.7% increase from August's downward (unlike initial jobless claims, when one is attempting to report an increase, the last number is always revised downward) revised 368K (was 373K). This number was the highest adjusted print since April 2010, which makes for great headlines. So far so good, until one looks beneath the headline and finds that the 389K number (to be revised lower next month), is based on a September unadjusted number of 31K in actual sales, consistting of 3K sales in the Northeast and MidWest each, 16K in the South and 9K in the West. This is the unadjusted number, which as last week's BLS fiasco with Initial Claims showed, applying seasonal adjustments is the easiest and best way to manipulate any data set (for more see X-12 Arima's FAQ - all 257 pages of it). This was the lowest print since February's 30K, the same as August's 31K, and well below the 35K from May 2012.
But wait, there's more: when one looks at the stage of construction (analogues to Housing Starts and Permits where one just needs a piece of paper to fabricate buying interest, and can be cancelled days later with zero sunk costs), the 31K unadjusted number consisted of 10K Not Started, 10K Under Construction, and 11K homes actually completed. To summarize: the Census Bureau took the sale of 11K actual new completed homes and extrapolated an annualized, seasonally adjusted number of 389K, feeding the media a number that is the "highest since April 2010." Instead, one look at the NSA chart below and one can see that grinding along the bottom is the best one can actually say.