A serendipitous mistake

Greetings patrons,

It’s time for another research update. In this letter, I describe how an embarrassing error led me to some exciting new research.

House prices and debt

In my January research update, I told you how I was working on a post about house prices. Well, I’m still working on that post. Here’s why.

The backstory: for over a decade, renegade economist Steve Keen has been writing about the link between house prices and debt. The chart below shows the latest iteration of his research. In the United States, house prices are strongly correlated with levels of household debt.

Last fall, I was reminded about this debt-to-house-price connection while reading Hilliard MacBeth’s book about the Canadian real-estate bubble. (Full disclosure: Hilliard is a patron.) Intrigued by this research, I set out to deepen it.

The idea was fairly simple. Steve Keen has connected US house prices to levels of American household debt. My goal was to do the same, but for every country with available data. And so I started scouring the internet for international data about house prices and debt.

After a few days of work, I had the results I wanted. Across countries, housing unaffordability (house prices measured relative to average income) is strongly correlated with the household debt load (debt relative to GDP). The chart below makes the case.

Here, each point represents an observation in a specific country, with color differentiating the various countries. The horizontal axis plots the level of household debt measured relative to GDP. The vertical axis show house prices indexed against GDP per capita. The message is that as debt levels rise, houses grow increasingly unaffordable.

A whopper of a mistake

With the evidence in hand, I started to write up my post about house prices and debt. But then came some nagging questions.

The chart below shows my data for house prices measured relative to GDP per capita. As I eyeballed this data, I realized that the apparent rise of house prices was unbelievably large.

According to the chart, since 2005, many countries have seen their house prices more than double relative to average income. Now it’s one thing for the nominal price of houses to double over a decade. That’s plausible. But it’s quite another thing to see house prices double against average income in under a decade. That’s such a steep rise of unaffordability that it defies belief.

As it turns out, the cause of this belief-defying trend is that it’s based on an embarrassing mistake. In my haste to gather evidence, I’d apparently downloaded GDP-per-capita data that was inflation adjusted. Then I’d compared this inflation-adjusted data to house-price data that was not inflation adjusted. The result was that my measure of ‘housing unaffordability’ was fundamentally flawed.

Then things got worse. When I corrected my mistake — calculating housing unaffordability using nominal GDP per capita — my results turned to mud. The previously tight relation between housing unaffordability and household debt turned into a blob of uncorrelated data.

Back to the drawing board

When I first discovered my fatal error, I was mad that it sabotaged my write up about house prices. But after a few days of reflection, my frustration turned to confusion.

You see, I thought I had a good theory of what drove house prices — they were supposed to be a simple function of household debt. My flawed data had ‘confirmed’ this theory. But the real data announced that there was more to the story. And so I had to go back to the drawing board.

When I’m confused, my first instinct is to look at a long-term time series. That’s because history has an unforgiving way of offering clarity.

When there’s a need for long-term economic data, that almost always means looking at trends in the United States. (A side benefit of imperial hegemony is good data.) And so I set out to create a long-term times series of US house prices measured relative to average American income. The chart below shows the results.

Clarity!

This evidence makes two things clear. First, the relative value of US houses is not primarily a function of household debt. (A long-term plot of household-debt-to-GDP looks nothing like the one above. Instead, debt levels are mostly related to short-term changes in house prices.)

Second, when it comes to US house prices, there appears to be two distinct epochs. Prior to 1972, US house prices tended to fall relative to average income. This is behavior that we expect from consumer commodities, which tend to become more affordable with time. But after 1972, US house prices tended to (slowly) appreciate relative to average income. This is behavior we expect from assets like corporate stocks.

Housing: From commodity to asset

As I pondered the US data, I realized that there was a bigger story to tell. The historical evidence demonstrates a fundamental shift in how Americans think about housing — a shift that I suspect will also be evident in other rich countries.

For my grandparent’s generation, a house was like an expensive toaster. It was something that you bought to use. But for my generation, housing is unequivocally an investment. It’s something you buy expecting that the value will appreciate.

Everyone with a modest sense of history recognizes this transition. But few people have a good explanation for why it occurred. Well folks, I think I have an answer. But I won’t spoil it. You’ll have to wait for my forthcoming post. But suffice to say, I don’t expect a return to the days when house prices fell relative to income.

Scientific uncertainty, rhetorical confidence

Now that I’ve told you about my stumbling path to clarity, you have a glimpse into the unspoken part of science. You see, my finished piece on house prices will be written in a confident tone — one that implies that I knew what I was doing all along.

This confident tone is a rhetorical trick — a white lie that makes science more digestible. For example, when explaining his theory of General Relativity, Einstein didn’t burden the reader with all of his unsuccessful attempts at theorizing gravity. Instead, he explained the finished theory as if it had been delivered to his brain on a silver platter. The smörgåsbord of failure was left untold.

A big part of scientific training is learning about this failure smörgåsbord. The effect is that during grad school, students become better scientists … but horrendously worse writers. A big part of post university training is learning how to write well, which requires pretending that the failure smörgåsbord doesn’t exist.

It’s a delicate act to balance these conflicting goals. Too much confidence leads to propaganda, and too little leads to undigestible prose. On good days, I think I come close to getting it right.

Until next time

That’s it for this update. I hope you’ve enjoyed the journey into my research foibles. And thanks again for supporting my work!

Best,

Blair