Simple Models and Housing Prices

We are taking a new slant on modeling and forecasting.  At the risk of over-simplifying, let us start with two schools of thought. 

The professionals take classes in various modeling methods, learn techniques in real-life situations, and then go into the world to predict, to explain, and to advise.

The naysayers disparage the work of the professionals.  They point out the errors in the forecasts, without seeming to make any of their own.  They proclaim that all models include errors.

This represents a clear divergence of thought and a challenge for investors.  If there is no expertise involved in modeling and forecasting, than anyone's opinion (or anecdote) is as good as anyone else's.  It is the Wild West democracy of the blogosphere.  Great fun, but is it profitable?

We recently read a comment from a journalist praising the economic blogs as "better than the economists."  The reason?  She saw a more accurate prediction of the housing situation on a blog than she could recall from a newspaper.  The problem?   You can find a blog predicting almost anything.  Will her "winning blog" be right the next time?

Our Take

There is an important concept which every investor should understand:

Every pundit does modeling.  Every pundit makes forecasts.

The difference is that professional modelers have learned techniques that recognize and measure probable errors, determine an appropriate level of complexity for the model, and accept responsibility for the results.

An Illustrative Example

We have an excellent ARIMA model for volatility forecasting.  It has a twist or two.  It is not very popular with options traders, although we have used it successfully.  Why not?  The model results include an error band.  The traders see the band and (correctly) say that the model is not very accurate.  The inaccuracy is because volatility is difficult to forecast, not because the model is bad.

The options trader uses simple heuristics.  He mentally parses some stock history, the recent market action, and a knowledge of upcoming events.  The options market includes his opinions and the rest of the market makers in the pit.  There is definitely a wisdom of the crowd.

But there is a problem:  The crowd does not specify an error band or confidence interval.  The actual agreement among peers, strongly influenced by the flow of paper, has no formal analytic basis and no "confidence interval."  No one knows if it is correct, and no one keeps records on the performance of the pit forecasters.

A Current Example

There are many pundits — too many to cite — who believe that housing prices must decline to a certain level before any stability is possible.  They typically use a simple heuristic like home prices as a ratio to income.  They compare historic values with more recent ratios.

We applaud approaches like this.  One can get a lot of mileage out of a single variable.  Strong predictive models use as few variables as possible.  The approach is supported by logic — the need the homeowner's  paycheck to cover many things.

A great starting point.

So the question becomes, "Are any other variables relevant?"

Anyone who has bought a home knows that affordability is not merely the price of the home but the size of the payment.  Much of the subprime lending problem came from loans that artificially reduced payments.

So it is obvious that a variable is missing.  The rate of interest on a fixed, 30-year mortgage is certainly relevant.  Comparing all of history with a time when rates are close to historic lows is quite misleading.


We are looking for an astute economist who examines historic affordability in terms of payments, not merely price-to-income.  Meanwhile, we expect that housing sales will stabilize sooner than most believe, since no one seems to grasp this crucial point.

You may also like


  • RB September 11, 2008  

    Hmm.. I had responded “astutely” to the post of June 19 2008, but it seems to have been deleted. At the time, at the national level it pointed to another 15-20% to go in today’s dollars including the effect of interest rates.

  • RB September 11, 2008  

    Addendum: the down payment is also another variable factoring into affordability, though hard to quantify, that becomes important at times such as today.,0,537937.story

  • Jeff Miller September 11, 2008  

    RB – My editors only kill spam that typepad misses, and they certainly would not delete something of yours!
    You pointed out research from PMI’s Berson, but it was in a comment to a different housing post:
    As you noted, that report was subject to varying interpretations. It is a source with a more sophisticated model, and they release a report quarterly. Here is the summer report, more pessimistic than the prior one:
    There should be another one soon. More later.
    Thanks for reminding me of this.

  • RB September 11, 2008  

    I had seen that post appear — and I also emailed it to you. It had a lot of hyperlinks for my data sources, which is probably why it got deleted again. I assumed it was something to do with spam-software though.

  • TR September 11, 2008  

    Jeff, I agree that GIGO is a big problem with real estate models. Data lag, uneven reporting, and problematic sources (incl. NAR) make the situation worse. Another multifactor model worth looking at is from John Burns RE Consulting.

  • gaius marius September 11, 2008  

    as RB notes, there’s yet another missing variable — underwriting standards. if there is any lesson to be learned from this bust, it is that prices and valuation metrics are completely, totally and always subordinate to underwriting standards.
    when credit was easy and fraud rampant, prices flew to levels double any historical measure of affordability.
    now that credit is difficult and diligence keen, prices can fall a very long way indeed — as far beneath mean valuation metrics as they were above, and far below traditional measures of “affordability”, which will be largely irrelevant until banks are recapitalized and unsold housing stocks decrease. and this to say nothing of credit demand, which appears to be suffering under a sea change in social mood.
    it’s important to remember that “mean” valuations are a midpoint, not a floor — by definition, the metric carves out as much area beneath as above the mean.

  • Bill aka NO DooDahs! September 11, 2008  

    Actually gaius, the MEDIAN carves out as much area beneath as above.

  • Jeff Miller September 11, 2008  

    RB – I am not sure what happened to your comment with the links. We found it in the spam filter and restored it. I did receive your email (and responded) and I appreciate the sources, which I check periodically.
    Thanks again,

  • Jeff Miller September 11, 2008  

    TR — Thanks for the Burns link. That is one of RB’s missing links 🙂

  • Jeff Miller September 11, 2008  

    There are some great additional comments here about how many variables we really need to add. The problem with adding variables comes from the lack of data — or at least data from different eras. I suspect (but cannot prove) that lending standards are not just a cause, but also an effect of the availability of capital and a securitization process.
    It is also quite possible that we will overshoot to the downside — or not.
    This is a work in progress, but it seemed to be a good illustration of some modeling problems.
    Thanks to all for the thoughtful reactions.

  • shrek September 11, 2008  

    This is a rare event with no prior data points therefore no one knows what the eventual outcome is.

  • Lord September 11, 2008  

    I agree. Other factors are whether we are entering a recession since job uncertainty can hinder a recovery even if incomes are adequate, and how many have the credit to qualify if they were previously involved in the crisis. Energy prices have also skewed the data towards cities away from suburbs. The market may be turning up in some non-boom areas but may have another year to go in the boom areas.
    Another common technique is to accept the data but to make a radical interpretation of it. Unemployment down? People have to work harder. Unemployment up? People want more leisure.

  • bendlund September 12, 2008  

    Another way to look at it is to examine home inventory. As long as home inventory levels remain elevated, prices will continue to fall. This doesn’t provide a useful guess now as to how far prices will fall before reaching a trough, but it does provide a real time indication as to whether we’re near a trough now, and I suspect that it will be more reliable for this purpose than looking at whether prices have hit some pre-determined target.