Im a data scientist, and the reason I have trouble believing in the $OPEN model is because ML models are only as good as the data. There was high demand in recent years, now we're entering a high rate environment.

Demand changes, rates change, the velocity of change in rates changes. Historic data would need to incorporate all of this, plus appreciation factors, average price in neighborhood over time, etc. How good is the historic data? Real estate cycles vary in length. It's a complicated model. The data cannot be all that accurate.

​

Now consider a real-world ML model. Complicated, no predictable behavior, this is true of most real world problems. Perhaps your data is amazing and you have 80% predictive power (many models get 60%). Could you imagine being 20% wrong on a house purchase price?

If we take the 60% accuracy, makes sense they lost money on 40% of homes in August.* These are large transactions aimed at occurring rapidly, the losses can compound quickly

​

​

*I'm aware that the article stating '42% of transactions were losses' didn't account for their 5% fee, but still. Losing money on the sale price is not a good business model. I think they actually lost money on ~24% of transactions. Still horrible and likely to get worse with this rate hike shock to the system. Highest rates have been since 08, and never have we seen rates increase so quickly. Something breaks. Real estate does not seem safe here and if I had an inventory of homes on my balance sheet, I'd be worried. Especially if I was planning to flip them for a profit in the short-term. Many of us in the retail space know this term all too well: bag holding.