It seems that everytime we get a major economic release, there is almost an instantaneous headline and commentary generated. This often causes me to say to myself "Hey, thats bigger/smaller/stronger/weaker than I was expecting."
Then I go to the original source of the data (BLS, Commerce Dept, CBO, FRB, etc.) and find out that headlines were woefully wrong, and the data is in actuality inopposite situated.
Perfect example: Yesterday’s New Home Data. Commerce released the New Homes Sales numbers, which were dutifully reported Friday thusly:
"Sales last month rose 4.8 percent, after a 7.4 percent rise in November. Those two increases, however, were not enough to salvage the full year, which had sales of 1.06 million units, down 17.3 percent from 2005. That was the biggest decline since 1990, when sales fell 17.8 percent."
How reliable are those numbers, according to the Commerce Department itself?
Let’s start with sales of new one-family houses in December 2006: They were reported as 4.8% above the November rate.
In reality, the mathematical change was statistically no different than zero.
Why? Margin of error. It was ±12.2% — much higher than the reported gains. This means the actual increase or decrease in new home sales (according to the commerce department itself) was in a range ging from as low as -7.4% to as high as +17%.
Same with the year over year decline: December 2006 was 11.0% percent below December 2005 numbers — but it was ±11.7% . This represents a range of -22.7%, to plus +0.7%. And that’s before we even get to the now well documented cancellation issue, which according to the major builders themselves, excessively high cancellation rates may be overstating new sales by as much as 30%.
Statistical meaning: If you read the Commerce Department footnote, you are advised: "If a data range contains zero, the change is not statistically significant; that is, it is uncertain whether there was an increase or decrease." I doubt very many media sources want to report month after month that "reported changes in new home sales are statistically meaningless." Thus, tools such as myself are forced to plow into boring government reports, reading footnotes when I would much rather be whining about other things. Such is the life of a curmudgeon.
There was one statistically significant
number released yesterday: Sales for the full year: 1,061,000 new homes were sold in 2006, and that is a 17.3% decrease (±3.4%) below the 2005 figure of 1,283,000.
This means the range of new home sales for all of 2006, according to Commerce, was as good as -13.9%, or as bad as -20.7%.
Now those are some statistically significant numbers . . .
>
Sources:
NEW RESIDENTIAL SALES IN DECEMBER 2006
JANUARY 26, 2007 AT 10:00 A.M. ESTU.S.
Census Bureau, Commerce Department,
Department of Housing and Urban Development
http://www.census.gov/const/newressales.pdf
http://www.census.gov/newhomesales
Durable Goods, Housing Deliver Signs of Strength
MICHAEL CORKERY and MARK WHITEHOUSE
January 27, 2007; Page A3
http://online.wsj.com/article/SB116981819353189003.html
The 2005 figure of 1,283,000 must have also had a margin of error. So when you are comparing it with 2006 figure shouldn’t you also take that into consideration?
The data on housing is just wrong. All of the data coming from the housing sources is so biased that consumers are mislead with a false picture of reality (like new home sales not counting cancellations or sales being counted at contract signing, not at closing). Therefor the consumer must ignore the data and return to fundamentals of home affordability. If you can commit to 10 years, if you can put down 20%, if PITI (principle, interest, taxes, insurance) consumes no more than 30% of your take-home income, then buy. The trouble is with those constraints, there are no SFH on the market on either Coasts. Although the data is hard to find, try and get rental data for similar types of housing. If there is more than a 10% gap between the cost of renting and the cost of owning don’t buy — your market is overpriced.
Rental data is easy to find on Craigslist.
nice spin barry. how about units sold in 2006 is the 4th highest on record? how about since july with the exception of 1 month the inventory spread has narrowed on a y-o-y basis?
How can there be such a high margin of error? Are they surveying builders or looking at the transactions being recorded?
Um, richard, it isn’t the level, it’s the direction/trend.
The same tripe is tried when interest rates rise…..”but they are the lowest in blah,blah, blah years….” Ask the market what they think of bond rates this week. They freaked at the move from 4.75 to 4.95
If the rate is moving the wrong direction ( in this case falling sales) then the YOY isn’t the issue. It’s the TREND, which according to all is DOWN. 13-20% down completely trumps any healthy looking YOY number.
Using YOY is like overinflating your tires and running over a nail. You may still have too much pressure, but you know the direction the tire is going regardless of comparisons to the overinflated state.
“I still have 35 lbs of pressure” is irrelevant to the fact that your tire is going flat.
Nice analogy Craig!
Thanks Barry for digging through the data again.
The gov’t data is in complete contradiction to the data coming from the home builders. This is a classic foible of human nature. How to decide what to believe. Those who have an agenda will believe the set of data that supports their world view. Those who don’t will look for other data points to help them decide between the two.
Said it before, will say it again. I will believe the guys who have Sarbanes Oxley on their backs. Ceo’s don’t like jail, which is where they can be sent off to if they lie about their numbers.
It takes quite a while to wipe out all the optimism from a secular bull market. Look at the Nasdaq from April 02 to Oct. 02. Many thought they had seen the bottom but the bear wasn’t done yet. Or how about Apr. 01 to June 01. Nope not done yet. Ok for sure then Oct 01 to Jan 02. Sorry still not done. The first leg down always takes at least a couple of years to complete. Housing just hasn’t had enough time to wipe out 20 years of optimism. Now how about Oct of 02? That was the bottom of the bear market for sure right? Well lets take a look at the Nikkei from 89-02. How about the Dow from 66-82 or gold from 80-99. I’m going to have to say no we have a long way to go yet and much disappointment has yet to be felt before the next secular bull market can start. It’s actually easy to spot a bubble. Just look around and see if everyone you know is participating. Was everyone buying houses in 05 sometimes multiple houses…yep. Did most of these people know anything at all about real estate…nope. Did everyone know for certain that real estate only appreciates…yep. Was the media finding countless reasons why this time was different and that a paradigm shift had occured…yep. A classic bubble if I ever saw one. This will end exactly the same way all bubbles end and not before no matter how much people want the bottoming process to be over quickly.
Even if the total units sold is 4th best, that’s NOT a good sign, the number of units should always be going up in an expanding economy with a growing population.
It’s like when administration officials say tax receipts are going up because of tax cuts. Well, they went down, now are growing back at trend rates, they’re not going up because of tax cuts the muthpeice isn’t adding the important ‘they always go up’ addendum.
As for inventory, both my neighbors of my second home in SoCal took their homes off the market over the holidays. But more importantly, as sales decline the months of sales numerator goes down, thereby dropping the number (e.g. if only one house is sold this year, then only one house ‘on the market’ is 12 month’s inventory)
More to the point, months sales is a flawed measure for new homes, since new-home sales remain at historically high levels versus population growth, a proxy for household formation. Cf http://www.bignose.org/blog/index.php?/archives/185-New-home-sales-rate-and-population-growth.html
You can’t sell new homes to people who don’t exist. Unless you think there’s evidence of a secular increase in demand for second homes starting on, oh, 2001 and peaking in, say, 2005. Then months sales might be appropriate, though you’d still have to explain why all those people started buying vacation homes in 2001.
Hey, Mabel, our tech stocks blew up — let’s buy a place on the water with what’s left!
BR said Why? Margin of error. It was ±12.2% — much higher than the reported gains. This means the actual increase or decrease in new home sales (according to the commerce department itself) was in a range ging from as low as -7.4% to as high as +17%.
But is the probability of +4.8% the same as that of -7.4% or +17% just because they are all within the margin of error? If not, then there is at least some validity to the +4.8% and it can’t be completely dismissed.
I’m not that savvy on statistics so could someone explain this?
Do the inventory data have a similar margin of error? Cancellations eventually get factored back into the sales numbers (with a couple of months lag) but if inventory is in fact going down then houses are selling (existing homes can be taken off the market or rented, but not new homes).
They’re back! Remember all the screaming and hollering about illegal aliens by the politicians before the election when fruits and vegetable prices went to the moon? Well, after the election prices started to decline dramatically until the recent freeze in California. They gave muchos non-income documentation mortgages in Miami and LA to illegals 4 years ago which caused a bottoms up boom in house prices in those cities. Will they start giving loans to them again to reduce the home glut? And how about the 36 month no payment refi mortgages for those who can’t pay their mortgage now? And how about the banks that are telling their mortgagees to put their houses up for sale if they can’t make the payments in lieu of foreclosure? Do the banks show the increased mortgage principle immediately on their books? Does this show up then as increase in loans or money supply?
One issue the media was discussing yesterday was that the real estate market seemed to be stabilizing. Some of the evidence they gave was that inventory is declining. However, there are a number of reasons why the inventory is distorted on the low side.
– Inventory declines every year during December as people take their homes off the market during the holidays. Many of these homes will go back on the market in the spring.
– In my area which is Northern California I am seeing homes that were for sale now becoming for rent.
– As Barry pointed out cancellations are not included.
“Applied to an opponent, it means the habit of impudently claiming that black is white, in contradiction of the plain facts. Applied to a Party member, it means a loyal willingness to say that black is white when Party discipline demands this. But it means also the ability to believe that black is white, and more, to know that black is white, and to forget that one has ever believed the contrary. ”
To Michael C.
“But is the probability of +4.8% the same as that of -7.4% or +17%”
A probability cannot be assigned to a range that falls within the margin of error. It is either inside or outside the margin. If inside, then the data is statistically meaningless.
That is the beauty of maths. Arguments can’t win against it, however well crafted and spinned they are.
Unless of course, one has a forceful desire to lose money. To quote Ed Seykota:” Everyone finds what he/she’s looking for in the market”
Francois
Michael C, I believe you are correct. It all depends on the distribution of that variance. I doubt the govt or companies announce the variance distribution. But, even if we assume a worst-case random distribution within that variance (rather than a bell-shaped curve), then, according to the odds, the real number isn’t all the way at the low-end of the margin of error.
So, BR, the statistic has a value and a range of error. Does a statistic lose its “significance” simply because the range of error includes “0”? The govt footnote may say so. But, I disagree with the footnote’s characterization/semantics. Why? Well imagine that the number was -50% ± 5%. It could have been. But, it wasn’t. The fact that it was 4.8% ±12.2% rather than some worse number is indeed significant.
Also, December had 20 business days, while November had 22. So, with 10% fewer business days, December did what it did. Which does not seem bad to me. (Or do these statistics “normalize” monthly figures based on the number of business days?)
(BR:
A sale can take place on a weekend, so business days in the month is not relevant to New Homes Sales)
Being house-poor, credit-rich myself, I have a commitment bias to seeing the housing market tank. But, I don’t see how to explain away the apparent bit of strength in Dec’s #’s.
Sure, warmer weather probably helped. If we were looking at Dec YOY, I would cite “warmer weather” as the major factor explaining Dec’s rise. But, Dec was colder than Nov and still eked out an increase.
Maybe the best I can do is cite the reduction in the growth rate (down 2.6% from Nov’s 7.4% to Dec’s 4.8%) as evidence of slowing. Remember that builders cut prices significantly just a few months ago? Well, maybe the effect is wearing off.
Also, there is that cancellation effect. Was Dec up 4.8% from a cancellation-corrected November figure or from an uncorrected/raw/misleading November figure? If the industry/govt compared Dec sales (including cancellations) with Nov sales (minus cancellations), then that’s the explanation for the “increase”. But, if this was an apples-for-apples comparison, the curmudgeon in me worries Dec went up because demand for new homes really did rise.
When do we learn about Dec pricing??
(BR: There are no cancellation corrections — my understanding is that this data does not get corrected)
To Francois and Michael C.
the uncertainty quoted is due to _sampling error_ ONLY, i.e. it is a statistical uncertainty. In this case one almost always assumes a “bell shaped” probability distribution around the centre value i.e. values far away from the middle are increasingly less likely that the centre value. In this case, a 7.4+-12.2% (90% confidence) means that the central value
7.4+- 6.1% 68% likely (1 “std. deviation”)
7.4+-12.2% 90% likely (2 )
7.4+-18.3% 95% likely (3)
etc
so in this case one can say that it is about 15% likely that the increase was 0 or negative. BUT NOTE, that the uncertainty does NOT include that due to the uncertainties in their “method”, including any corrections, etc.
So the “true” uncertainty is likely even LARGER.
Which is why the CB advises only to look at the “trends” over about a 4 month or more period, since (assuming constant uncertainties) the 4 month moving average should have a SQRT(4)=2x smaller uncertainties.
[check : 12.2% uncertainty per month averaged over 12 months is 12.2/SQRT(12) = 3.3%, which is about what the CB quotes for the yearly uncertainty. check.]
BIT OF ADVICE: don’t try arguing with a journalist about this. Useless waste of your time. That guy from Business Week who fessed up yesterday to ignoring the error bars is the FIRST time I have ever seen that happen in over a decade.
@Michael C –
Piggy-backing on Francois – The issue is that it’s a SAMPLE, and not the whole population. Associated with a sample is a margin of error, and all you can say is that there is an X% chance that the real value W falls between Y and Z, based on that margin.
When Y and Z include 0, you can’t conclusively say that the real value (W) is actually positive or negative in that light; you can’t conclude a difference. Of course, you could skew Y and Z such that they include only positive numbers – but X% would be much lower (assuming normal), and thus less significant.
Confidence Intervals – How to Understand Confidence Intervals
Statistical Hypothesis Testing
>> (BR:A sale can take place on a weekend, so business days in the month is not relevant to New Homes Sales)
Oh, right. In that case, I wonder: do Saturday and Sunday contribute more heavily to home sales than during the week? If so, then notice that Dec had 5 weekends while Nov had just 4.
For those of you playing the “Real Estate Crash Game” at home, there are two very reliable stats to watch in your neighborhood: Foreclosures and vacancies. The pain is just beginning.
Well, I tried to play a Housing Bubble! game like slugbug!, but it turned out too fast paced and tedious. I’ve resorted to Boilerroom fishing. I intentionally left my home number off the do-not-call-list. I made up a plausible but totally fake mortgage situation for myself to give out to the boilerroom callers. I get many calls a week, different company name each time. I generally stall and ask barely related questions to the first line troopers for awhile. Sometimes I even ask them how their day is going — they really like that. If it is obvious from the accent, I’ll ask them what part of India they are calling from and how the weather is this time of year there. The fun starts when you get handed off to the closer. I really take special pleasure in wasting these guys’ time. Eventually, it is “uh oh, here comes a tunnel…” *click*
Dr StrangeMoney, have you ever recorded those calls and really gone over the top? Maybe sick a little Mike Derucki on them and see how they respond!
http://en.wikiquote.org/wiki/Jerky_Boys#Mike_Derucki