Silly house price headlines

a mathematical view

July 2018

There's something rotten in our most basic housing data.

Like most property anoraks I follow the four main housing indices fairly closely. They don't tell you anything useful about the price of your home, but they do generate plenty of press and set the national mood.

Are prices up or down? Am I making free money? Can I get a new kitchen.

The raw numbers behind the headlines look like this on a chart. While each index follows a fairly similar (upward) path, there's a glaring problem:

The "movement" of the indices rarely match. When one is reporting rises, another is reporting falls and each time these turn into a shouty headline to further bemuse us.



Which headline is right? The sheer noise makes these numbers hard to ignore. They're tabloid gold, but add nothing to the market.

Who marks their homework?

The black line on the chart is HM Land Registry's sold prices - what we might consider "the truth". This figure is reported much slower than the other indices which leaves Rightmove, Halifax and Nationwide to get all the headlines because they report more quickly.

The problem is nobody goes back and checks who turned out to be right.

On first sight Halifax looks like the most accurate index because it runs closest to the black line, but look again: It's very often completely out of step.

In the five reported months since January, the movement of the Halifax index was wrong every time.

A way to check index accuracy

A 5-minute bit of maths can tell us very accurately how closely each housing index correlates with HM Land Registry. It's called the Pearson Coefficient, and you can do it very quickly in any numbery software. Excel implements it with the CORREL() function, while PostgreSQL offers CORR().

This coefficient outputs a very simple result between 0 and 1 (or -1).

  • A result of 1 says two series are completely correlated i.e. when one series goes up, so does the other.
  • A result of -1 says the numbers are inversely correlated i.e. when one goes up, the other goes down
  • A result of 0 (zero) says the two series have no link at all.

I've gathered raw data from the four main indices going back 14 months and the correlation coefficient stacks up like this to May 2018:

  • Halifax = 0.187
  • Nationwide = 0.791
  • Rightmove = -0.459

Assuming 0 = Useless and 1 = Useful, we can visualise the result like this:

This puts the headlines in a very different light.

The gap between Nationwide and Halifax is stark. Both are based on a small subset of the mortgage market, but Nationwide's price movements match "the truth" much more closely than Halifax.

The result may be a little unfair on Rightmove; they reset their methodology in January, and report asking prices as opposed to sold prices. Still - they're happy to take the free headlines.

There could be many more arguments on differences in methodology but ultimately, all the major indices are happy to garner free press from their numbers. They deserve comparison because they set themselves up for comparison.

Save us...

The UK's national house price index may be an interesting macroeconomic measure, but it's of almost no use to buyers and vendors who read newspapers. Indeed it's particularly unhelpful to agents, who often have to deal with unrealistic expectations from vendors who's only knowledge of the housing market is a Daily Mail headline.

The local facts that really matter are always very different from the national narrative. Runaway house price inflation is concentrated in the southeast, and the more recent story of price drops is also very London-centric. That tells us nothing about Manchester, Exeter or Newcastle.

The UK is lucky to have a land registry that provides excellent price data, if only we took the time to unsilo it from analysts. I've spent a lot of time wrangling similar (but hopeless) data in Spain, France and Portugal - and arrived at the conclusion that we really should appreciate our Land Registry much more.

Unfortunately I rarely see house prices reported in detail.

It's a shame. A better informed market is a more efficient market.

For the Excel nerds out there, here's a screencast of how to do the calculation.

Source data: Correlations.xlsx (10Kb)