Correction and apology

Austin G Mackell
2 min readNov 1, 2022


I was mixing my datasets, which was sloppy and wrong.

Early in October I decided to add a “falls so far” value to my daily updates on Sydney house prices. I was calculating these using data from CoreLogic. But I couldn’t (and still can’t) find (not for free on the open web, anyhow) a direct estimate from them about the price of Sydney houses at or near their peak (on feb 13). So Ifudged it and used data from Domain. I was assuming that the two major sources of real estate data would have similar-enough estimates for the median house prices in Australia’s most valuable housing market. As a recent release from CoreLogic shows, that was wrong. Core Logic’s estimate as of the end of october is about 1.25 million. The most recent estimate from domain, for september, is over 1.45 million. This difference of over 200k means my estimates of falls so far were slightly high.

More significantly wrong was my prediction on 15 October that gains made during the pandemic would “be erased by years end” since it used a high estimate (Domain’s) for the pre-pandemic level, and a low estimate (Corelogic’s) for the current price and trajectory. I haven’t re-cacluclated when the threshold will be reached, but I know my previous estimate was several months early.

Since the main point of the excersize I am engaged in is not actually to track house prices, but to test my company’s software and demonstrate our vision for transparent, evidence based, journalism, this whole thing was very uncool of me.

Starting from today, estimates will be based on Corelogic data exclusively.

You can see me retrace my steps and realise my error in the video bibliography below. Let me know if you spot any other mistakes!

UPDATE: Note that this doesn’t mean my estimates of total losses were neccessarily high — there’s also an extent to which my estimates downplay the scale of the falls so far — they use an index that includes apartments, which have been less volotile than houses. But using one data set is better than two, and we can now be certain that these estimates er on the conservative side.