I don't get it.

Are house prices moving up by %5.9 a year or down ?

who are we to trust ?????

This kind of depresses me. I want to be able to afford a house. But it looks like I'll never get there, even at the end of the housing bust.

Depends on where you live ams... almost anyone can afford to buy a house in the town I live in now. Most folks quite easily & without 'exotics'.

But it isn't coastal California or Manhattan.

dryfly

Depends on where you live ams... almost anyone can afford to buy a house in the town I live in now. Most folks quite easily & without 'exotics'.

Me too. In north central NE 100K will buy a very, very nice house with a few acres of land, of course the weather sucks but life has it's little trade offs.

Even Realtors out here have conceded that prices are down YOY here in LV, so what gives with OFHEO?

It is their measurement methodology, it tends to smooth & lag. It is based on transactions... something like:

(Current Price - Previous Price) divided by time between transactions.

I believe the differences are then adjusted so more recent data is weighted higher then older data... and also for inflation over the time period.

So say take two houses... each sells for $250k... house one was last sold for $50k twenty years ago (big gain), the other house previously sold for $255k just last year (slight loss).

1: $250k-50k/20 = $10k/year... then adjust for 20 years inflation & weight

2: $250-255k/1 = -$5k/year... then adjust for 1 yr inflation & weight

Sum these & all others & you get a pretty smooth & lagged curve compared to what is probably actually happening in the market at that time.

And it is probably skewed a little toward 'positive' since we've had so much more gain lately than decline that even if weighted toward 'more recent data' the past run up just overwhelms the data.

I am not EXACTLY sure that is how they do it but I understand it is something like that and that is why it misses the turns (index lags & dampens).

After the bottom is hit OFHEO will probably miss the return to appreciation too - same reasons but in reverse.

Problem is, there's probably a good reason for that, like maybe there aren't good paying jobs in tech...

Not many but some.

One of my son's friends graduated a couple years ago (IT-math double major very good grades & internship)... looked out wet, looked out east... ended up working for a small software development company making applications for agri-business (ethanol plants, co-ops, etc).

Paid about the same as what he could get on either coast (maybe 10% less locally - but close though).

The difference is the cost of living. If he had gone to Cali he would never own a house... But in fly over he is two years out of college & he and his girl are looking at buying a house... it will likely cost less than their combined annual incomes. Yes - very cheap, see Kash's chart on 'inland cities'... they started cheaper & didn't appreciate much.

But this town isn't real 'exciting' either... you have to make some choices. My guess is they will be making babies here pretty quick & won't be havin' much excitement other than tours at midnight feedings.

Not sayin' everyone should do this. People make choices - what works for some aren't right for others.

Hey Dryfly, My guess is, almost everyone in the country can afford to buy a house in CA! No doc, no down option adjusted loans are widely advertised. We're inundated with 3-5 tv infomercials DAILY telling all about opportunities to fi or refi. Even the Donald makes trips out here to advise us all of the opportunities in r.e. (Although I did hear the seminar cost was reduced from $150 to $0. or so.)
It seems the only absolute qualification one must meet here is he'she MUST BE STUPID!

I'm stuck in the Boston area because I'm tied to university life and because of family.

Well I can understand the family thing... but believe it or not there is a college or two west of the Berkshires.

The only thing that bothers me is the prices are so high if I buy a place and the market does collapse I am in deep doo doo.

If I were you, Mulik, I'd ask myself why you think this.

If what you wish to buy is a home to live in, and you can comfortably afford the payments (plus maintenance and the transaction costs and all that), and you intend to live there for quite some time, and think that intention is reasonable given your current employment and the likelihood you could get a different job of equal earning potential without moving, if you had to, then why are you in deep doo doo if prices fall after you buy?

If the only way you can afford to buy is to stretch yourself too far, or if your situation (employment, family, whatever) means that you might very well need to move in two years, then certainly you should be cautious about buying in a deflating market.

My perspective comes from someone who has been a mortgage lender, not from someone who has ever been a real estate investor. So I tend to think of homes as homes, not as "investments," and loans as ways to buy homes. I tend to think of the "investment" part, if it happens, as gravy. Just about anyone in this country who has owned a home for ten years or more has experienced periods where that home depreciated or at least didn't appreciate. It still remained a home in that time, and as long as the mortgage payments weren't crippling, most people didn't get that worked up over it. They did, after all, still have a home.

I guess this all reminds me of locking in interest rates on a loan application, which is something I do have a lot of experience dealing with from the advice-dispensing side of things. If you consider yourself risk averse--if the thought of what you lose with a higher rate bothers you more than the thought of what you gain with a lower rate--then lock the loan. If you really think rates might go down, or you're really bothered by the possibility that you'll miss a dip, then float the loan. But don't ask your lender what to do, because your lender doesn't have the right answer for you. The lender's perspective is not yours.

The OFHEO looks like they're calculating their data the way I do, using apples-to-apples comparisons of current calendar month to the same calendar month the previous year. Less noisy, less volatile, fewer bad "signals."

Not as dramatic a story, of course, just a more accurate reflection of what's going on.

Sebastia

Again I think the fault is mine being a engineer I had never tried to follow up on what effects interest rates had on assets. Being out of college recently didnt help either.

I'm an engineer too and I learned something about this stuff - though not enough to be confused as an underwriter...

You have time.

Rob'ts points (above) are very good. Focus on Point #5. Payments over the long haul. That is the key.

That and how long will you plan to stay there. If you think you will be moving fairly soon (like within 3-5 years) then save more & do not buy. Rent & rent wisely (look for value - it is out there).

If you think you will be there a long time, then the ups-n-downs wash out. Then it doesn't really matter if you pay a little too much when you buy.

As long as you can still make the payments that is. Point #5.

dryfly - how many times do you want me to debunk this "example" you keep posting? OFHEO's methodology is explained on their website universe

NJ-Bob - I believe you are on to something, and if we're right, the uptick in the ABX was exactly the wrong response to the OFHEO release. The Case Shiller index uses more or less the same methodology, but with different data. The small methodological differences might explain some it, but a) they are pretty arcane and b) I doubt it. So although I'll leave open the possibility that it's methodology, I'll bet it's data.

One data difference is timing. CS uses recorded transactions, which can lag months but usually only lag a month or so (which is why December is released in late Feb.) OFHEO uses loans bought by the GSEs, which often lag by a few months. So the first release of the OFHEO index (later revised) is weighted towards the beginning of the quarter (more Oct. loans have made it in than Dec. loans). True for CS, but to a much smaller extent. But CS shows a drop for every month of the quarter, so that can't explain more than a part of it.

Another is the fact that it's GSE data, so the national weighting is different. California especially would have less weight for OFHEO than it would for CS. But you can look at the OFHEO indices for San Diego or SF and compare them, and CS is substantially more negative. Unfortunately, you can't do a perfect job on that since OFHEO doesn't release the purchase only at the MSA level, and you have to make the wild guess that the appraisal bias doesn't differ much between markets. Still, it doesn't look like the different weighting for California explains all of it.

The third difference is that OFHEO has no jumbo loans, while CS does. If expensive houses were falling faster (in percentage terms) then CS might drop while OFHEO rises. But if you look at cheap MSAs where there aren't many jumbos (Detroit or Chicago) OFHEO and CS don't get much closer (again with the warning that the comparison isn't quite right because of the purchase-refi issue).

But the last difference (you knew there was a reason you kept reading. you did keep reading, didn't you?) is that OFHEO would have very few subprimes, while CS would have almost all of them. If prices in subprime heavy neighborhoods are falling faster than prices in other neighborhoods (makes sense if subprime originations are dropping fast, it's where the demand would be disappearing the fastest) you'd get what we see, a big divergence between OFHEO and CS. And that should scare the bejeepers out of anyone long the ABX.

dryfly - why exactly would bridging various periods of time and correcting for sales mix result in damping and missing turning points?
Aren't those exactly the things that you want an index to do?

mort where do they pull the data from... the pool of sales data?

the detailed description of the data is on p. 4 of the above link to the research paper, and the description of the regression used to generate the index is on p. 7

"Then the the second question is are the quarters (the -1,0,1's) all in the same quarters? If NOT in the same quarter then that's your 'bridging' right there..."

you know, I can't even begin to parse that paragraph

you know, I can't even begin to parse that paragraph

mort - they are building a matrix of transactions... from GSE databases. -1,0,1s... in fields for the quarters for which previous purchase, no purchase & current purchases (respectively) occur. Right?

Then fit via regression.

So are all the 'previous purchases dates... the occurances of those (-1)s always in the same quarter? Or are they spread out into the past for different homes & different transactions?

Example: we are in 1Q07... say we want OFHEO for 4Q06 (quarter just passed)... do we ONLY look at transaction with previous sales in one quarter (say arbitrarily, 4Q05)?

That would result in a matrix with all the -1's in 4Q05, all the 1's in 4Q06, and zeros everywhere else. Matrix would look like this:

Past qrts to Current qtr
0...0(-1)000:1
0...0(-1)000:1
0...0(-1)000:1

So on through the set of all transactions in the matrix...

Or do they use multiple transaction durations in the sample set (samples where the previous purchases could have occurred at many times in the past)... then the matrix would look something like this:

000(-1)00:1
0(-1)0000:1
0000(-1)0:1
(-1)00000:1

With the above example only looking back six quarters... could look back lots farther.

If the transaction times of the previous purchase in the sample set are spread out over time then you will have bridging and as a result smoothing.

If not - if they have a frozen duration period (and keep it real short) they will have much less bridging & less smoothing but will then have sample bias error (by limiting their sample set to only those house that flip often).

From reading the article - and watching how slowly OFHEO index moves - my guess is they bridge quite a lot.

let me modify that slightly. you've never claimed a reason consistent with the index being created via a regression on log prices. back when you were stating that they took some sort of average over durations you almost got to the point of claiming that averaging these averages produced the smoothing, but now that we're past that, what mechanism do you assert causes this "smoothing."

but now that we're past that, what mechanism do you assert causes this "smoothing."

The bridging causes the smoothing.

Take a look at my early example... Say there are two houses selling right now... both for $250K (just a convenient coincidence).

House #1 was previously bought 20 years ago for $50K then sold this year for $250K... Net gain.

House #2 was previously bought last year for $255K and sold this year for $250K... Net loss.

1: $250k-50k/20 = $10k/year.

2: $250-255k/1 = -$5k/year.

What does that tell us about housing prices at this very moment? Does it tell us houses have gone up or houses have gone down?

If you don't like using houses like House #1 in the dataset and only use houses like House #2 because #2's previous purchase was more recent, then you have made a selection bias right from the get go against houses with ownership stability.

If you include House #1 in the data set... then you have 'bridging problems'... because maybe House #1 went down in value last year too, but it is unknown since there was no transaction to uncover it.

If you use a lot of homes like House #1 in a large data set you will smooth out the response because of the hidden data that never gets captured. In that 20 year period prices went up and down quite a lot but house #1 missed it all.

If you throw out all homes like House #1 and only look at recent transactions you get a skewed sample for another reason BECAUSE houses like House #2 that sell often are not necessarily representative of the entire universe of all houses. We all have homes in our neighborhoods that sell often - the reason isn't always 'random'.

I believe OFHEO includes some if not a lot of homes like Home #1 - though not necessarily with twenty year old transaction histories - as well as those with the most recent transaction histories. I believe they use a staggered mix to try and reduce the sample bias error.

I believe they try and reduce the bridging errors by weighting transactions histories differently with those with more recent transactions histories higher.

That is how I interpret 'weighted repeat sales' from the article.

In short - if they reach back into the past to multiple start times they will have smoothing (signal dampening). If only because of the lost pricing information they passed over.

If they only use short selling periods (all recent) they will have sample bias & noise.

I believe OFHEO tries to split the difference between these two risks and I believe they do a good job. I don't know a better way to get a meaningful 'global number' but believe there are still issues with the index nonetheless... the biggest being the smoothing (or signal dampening).

szara - there is a discussion of this phenomonon in the press release.

in the example that I gave the index isn't "smooth." It rises for 19 years, and tanks in 1.

You keep asserting that the index is fitting slopes, and that this causes smoothing. But you're presenting no evidence that the index is fitting slopes, and the paper describing the creation of the index explicitly states that they are fitting endpoints - all the independent variables between the dates of sale are zero. For a property that rises 21% over two years, a pattern of 10%, 10% does not fit the observation any better than does a pattern of 21%, 0%, or 0%, 21%, or for that matter, -10%, 34% or 34%, -10% (give me a few basis points for rounding). If observations on sales that occur in between this property's sales indicate that a pattern other than 10%, 10% fits better, than their is nothing in the fitting of this observation to bring the result back towards 10%, 10%.

Oh and the 'evidence' of this is in the regression fitting of the (-1,0,1) matrix... that is in effect the equivalent of fitting the vector array made up of all {Xi(ti2)-Xi(ti1)} with the Xi(ti2)'s corresponding to the 1's and the Xi(ti1)'s corresponding to the -1's... the 0's all the intervals before & between.

It is algebraically impossible for the sum of {Xi(t2i)-X(t1i)} to be different from the sum of{Xi(t2i)} minus the sum of {Xi(t1i), as you assert.

I have to think about that one - it wasn't a good example. I know you are right if X is linear & there are no sample selection issues for X(t) period to period.

But if X(t) is nonlinear, or if there are sample selection issues creating different samples period to period, then I'm not so sure.

The price/valuation curve OFHEO is trying to map probably isn't linear.

And there are sample selection issues (the units that sell vs. those that don't sell spread out over different intervals).

Regardless it was still a bad point on my part.

I know you know a lot about regression & statistics in general... can see that. But I think you are missing or underestimating the impact from the hidden appreciation of the previously unsold units carried forward that don't show up until that particular unit is sold.

Until a transaction takes place that appreciation is 'zeroed' out in the regression.

Realize I am not interested in the OFHEO index vs some other index but rather OFHEO vs what is really happening in the actual real estate market & does OFHEO always measure that well. That is what you want an index to do - track reality.

I think it lags & is dampened compared to what is happening in the 'real world' and I think we are seeing that now with OFHEO still increasing but real markets probably already heading down.

Even though I don't think it is a terrible index, I think OFHEO has missed this turning point. I believe it will also miss the initial change toward a future rebound - whenever that happens.

But I clearly don't have a complete enough grasp of the mechanics to explain why I 'feel' this way. I have to read more & crunch some numbers before I will be able to do that.

Have you any other references besides this paper - ones where they actually walk through the calculation with data sets?