I don't think home prices are unpredictable. However, I think the issue with predictions is that you can have a long track record of success that's just due to random trends, and separating that out from actual data-driven foresight is pretty difficult.
For the decade '95-'05, you could simply predict "home prices will go up" in many areas in the US and be correct. There were lots of people making big money on speculation and flipping homes. Many of them probably thought they were pretty smart for consistently predicting the future, and kept on reinvesting in the expectation of yet more returns. That was a good strategy... until it really, really wasn't.
Similarly, '11-'21 or so has seen consistent, strong appreciation in a lot of housing markets, and people have made lots of money predicting that prices would keep going up. Will that keep happening? If we keep having highly-priced assets, low interest rates, and relatively low housing starts, sure. If some of those factors change, maybe not.
I was reading recently about different distributions (normal, power-law, Erlang) and how we make Bayesian predictions based on our assumptions around how events are distributed. For example, if you think housing boom durations are normally distributed with a mode of 8 years, you would guess that a boom that had lasted 1 year would probably end around year 8-9, and a 10-year boom would likely end within a year or so. If you think they have an Erlang distribution, you would consistently predict they would last a couple more years, no matter how long they've gone on currently. And if there's a power law distribution, you would predict that they would last some multiple of the current duration, say 1.3x- i.e. the longer a boom has gone on, the longer it is likely to go on!
I lean towards a normal distribution in this case- every boom runs its course. The boosters and Youtubers who talk about flipping and house hacking seem to prefer a distribution where the boom will always continue.