There is some public excitement in the SA property space around whether the increase in SA property prices in 2015 was “above inflation” or not. I mean, it’s the kind of excitement that only economists can get excited about, but still.

The excitement began a few days ago, when “property economist” and “advisor to Homebid” and “personal empowerment ‘Renaissance Man’ guru with a poorly-cropped photo on his personal empowerment website” Neville Berkowitz came out and said:

The man in the street is being told by the banks, mortgage originators and estate agents that average home prices increased by between 5% and 6% per annum in 2015, when in actual fact, they only increased by 0.94% per annum.

What concerns me most about these 5% to 6% per year average home price increases used by market commentators, who may be using limited samples of homes sold and transferred, is that the average homeowner believes he currently has an inflation-proof investment rising at above the inflation rate.

In 2015, however, the 0.94% average home price increase was, in fact, a real decline of 4.2% after adjusting for inflation, and well below the inflation rate of 5.2% p.a. for the year.

The outlook for 2016 is worse, with a possible drop in nominal average home prices below zero and an inflation rate (CPI) probably higher than the 3% to 6% per year range aimed at by the Sarb. CPI is expected to be in the range 7% to 10% per year for 2016.

“Market commentator” and FNB “property strategist” John Loos responded with:

If I use the FNB Major Metro Deeds dataset, but don’t apply areas of sub-segment fixed weightings then I also get to a far lower price growth rate (1.9%), similar to Mr Berkowitz, for 2015 because the reality is that the transaction volumes grew stronger at the low end in 2015, while declining on the higher end.

Yes, transaction activity shifts over time across value bands are the number one challenge of any HPI compiler. Right now, HomeBid’s simple unweighted average should show such low inflation as a result.

FNB tries to restrict this impact of activity shifts by applying fixed weightings to sub-segments and areas. HomeBid doesn’t, and I would suggest that’s probably largely the reason why their result differs from ours at least – though nobody can say for sure that someone else’s inflation rate is wrong.

My feeling is, therefore, that they need to do a bit more homework on how our various indices are compiled (and why) before pronouncing on them. They have their flaws for sure, but the flaws aren’t necessarily anywhere close to the ones that HomeBid is ‘identifying’.

Which will obviously be very edifying for the man on the street. At least he knows that the statistical debate is both debatable and statistical.

Or, you know, not.

The trouble is that this debate is really about bad math, and people don’t like math. Here’s a foodie analogy:

  • Let’s pretend that we’re really arguing over whether Johannesburg’s restaurant scene has improved.
  • Berkewitz want to go onto Zomato, add up all the restaurant ratings in 2015, and compare those with all the restaurant ratings from 2014, and see if the overall ratings are better.
  • John Loos would be saying (with more words like “compiler” thrown in) “But doesn’t the type of restaurant matter though? We might end up confusing your local KFC with some of the top-end fine-dining establishments – if the KFCs are getting worse, but the fine-dining is doing better, then the overall ratings might show that nothing has changed. Only, it definitely has.”

The Mathematical Problem with House Price Inflation

Here is what the prospective home-owner wants to know: “How do home prices do on average? Do they go up or down? Should I rent or buy?”

Here is the obvious answer: “Unfortunately, that really does depend on which house you’re looking at. All locations are different. As are other things: house size and number of bedrooms and number of bathrooms and whether there’s a pool and security of the complex and the list goes on. Each property is essentially unique.”

But that’s not an answer that most people are happy with, and economists and statisticians are expected to present something more satisfactory.

Here is Neville Berkowitz’s answer:

  1. Add together all the selling prices of all the houses sold this year. Divide by the number of houses sold. That is the average selling price this year.
  2. Add together all the selling prices of all the houses sold last year. Divide by the number of houses sold. That is the average selling price last year.
  3. Work out the change in average selling price.

So, at first glance, this might sound reasonable. Only, consider this:

  1. This year, only one property was sold. It was a R20 million mansion. Average selling price = R20 million.
  2. Last year, only one property was sold. It was a R250,000 apartment in a bad area. Average selling price = R250,000.
  3. Change in average selling price = R20 million – R250,000 = R19.75 million.
  4. Effective annual increase in price: 7,900%.

Does that make any sense?

Not at all, because we’re not comparing apples with apples. We’re not even comparing fruit. We’re comparing massive plantations with a tree in the back garden.

I mean, that doesn’t mean that you can’t do the calculation. It also doesn’t mean that you can’t quote the statistic. But at best, it’s a bit meaningless – and at worst, it’s seriously misleading.

John Loos would be saying:

  1. We’ll do something similar – but before we get there, let’s split the property market into “one-bedroom apartments” and “three-bedroom homes” and “mansions”.
  2. Let’s work out the average change in price for each of those segments.
  3. And then we can do a weighted average, based on how much of the property market is made up of “one-bedroomed apartments” and “three-bedroom homes” and “mansions”.
  4. Also, this will give people more information about the home that they’re buying.

What’s even more surprising: in his statement about the allegedly-shameful statistics of commentators like John Loos, Mr Berkowitz acknowledges this need for segregation in the property market. The paraphrased gist from

The largest number of homes transacted in 2015 was in the lowest-price category of less than R250 000, where some 85 155 homes, or 29.4% of all homes were transacted. These homes, on average, dropped 6.7% in price in nominal terms, and 11.9% down in real terms (after inflation), in 2015 when compared to 2014.

This was mainly due to the 0.5% per year increase in interest rates last year. The prospects for this price category in 2016 are even bleaker, in his view, as interest rates are expected to rise during the year by at least 1.5% per year.

At the top end of the price category range are the R10m-plus homes, which saw the highest average price increase of only 2% per year. This is based on the average price of the 2 642 homes sold and transferred nationwide in this price category in 2015 when compared to 2014.

The basic take-home message of the public excitement is this:

  1. Average prices aren’t really a good measure of average prices.
  2. If you want better measures of average prices, then you need to segregate the property market and weight the average.
  3. You can change the average by changing the way that you segregate the market.
  4. All properties are unique.

I realise that’s not especially helpful.

But if you’re looking for a general statistical conclusion: I think that John Loos’ statistics can be misleading, but I’m pretty sure that Neville Berkowitz’s statistics are more misleading. Unless they lead you toward to the conclusion that statistics are relatively untrustworthy, in which case, I agree with Neville.

Rolling Alpha posts about finance, economics, and sometimes stuff that is only quite loosely related. Follow me on Twitter @RollingAlpha, or like my page on Facebook at Or both.