Machine learning and gentrification | Hacker News Reader

Machine learning and gentrification

urbanspatialanalysis.com

83 pointsemilito8 comments

"The goal here is to use 1990 and 2000 Census data on home prices to predict home prices in 2010. If those models prove robust, we can use the model to forecast for 2020."

Selectively picking factors that worked in the past to predict what will happen in the future is like picking stocks based on what worked in the past- "Gee, oil and energy companies skyrocketed from 1990 to 2000, so oil and energy companies will rock from 2010 to 2020!!"

The only guarantee you'll ever get is that companies that make lots of money will succeed more than companies that don't.

When it comes to real estate: location, location, location. On average, you honestly can't go wrong with anything in the Virgo Supercluster.

Of course the real use case for this kind of thing is property investors wanting to get in ahead of the wave.

I'm skeptical about the long term prospects for gentrification though, since there's only so many gentry to go around. Perhaps instead of focusing on gentrifying places, we should be working on how to gentrify people.

That is of course a use case, but the stated use case is nearly the opposite.

> [Using this data,] City leaders and community groups could then facilitate participation from residents in decisions that affect their families and neighborhoods. Based on new forward-looking knowledge, city leaders and community groups could co-design policies that secure benefits for low-income and vulnerable families in both rising and falling markets.

Of course, if the data exists in a public arena, property investors will avail themselves of it. Unlike the present though, existing owners and tenants will have the capability to be equally in on the knowledge of the incoming gentrification effort and extract maximum value.

At a city level, this gives the opportunity for governments to react as well - diverting resources to or from areas that may no longer need them, changing police staffing in advance of shortfalls, prioritizing correction of crumbling infrastructure before a new area becomes saturated (making corrections more costly,) etc.

I suspect that the accessibility, format and cost of the data will determine which use cases this data will end up benefitting the most. If the data is open, then it could be the cause of increasing equality in these transactions, while if it were sold for $500 a zip code, would likely be more inclined to benefit only those who can already afford it.

> Perhaps instead of focusing on gentrifying places, we should be working on how to gentrify people.

That's a good way to look at it.

In general I am not a huge fan of home ownership as a primary principle (as opposed to a situational, tactical decision) but in theory it can help here: if you own in a gentrifying area you can sell or even rent out your home (amazingly my housekeeper does this in Palo Alto, having bought decades ago when the economics were different).

Unfortunately, as I think we both know, in economics, reality rarely matches theory.

Re: property investor effect.

Was thinking about this yesterday and wondering if there was some equivalent of "bitcoin days destroyed" [2] for non-bitcoin title-changes. As in, BDD is a measure of turnover in Bitcoin-land, where the movement of an asset is weighted more highly the longer it was at rest. So with Bitcoin, higher values are good, because they probably mean more hoarded bitcoins are moving.

So the thought was that changes in land title, or long-standing businesses changing owners, or long-standing tenants moving out -- that these sorts of changes might be useful in predicting gentrification.

Obviously, a land-title registry on a blockchain would give this for free, but at least in my city, land titles are very costly to search. But changes in business ownership (small businesses opening and closing at given addresses) might be doable for cheap. Could do a community survey of businesses to find out how long they've been around, then simply track the "permanently_closed" boolean flag for all businesses within the bounding box, to keep track of when things are closing.

Anyhow, still need to do some research to find out whether this would be a useful indicator, but I'm optimistic. Curious about folks' thoughts!

[1]: https://en.wikipedia.org/wiki/Community_land_trust [2]: https://bitcoin.stackexchange.com/questions/845/what-are-bit...

I think there's a correlation, but I have a feeling that it's a late indicator. I think these businesses go out of business after the gentrification, not before it, though it might still be useful.

Emphasis mine:

> Second, as previously mentioned, the predominant pattern over time is decline not gentrification. Thus, it is difficult for the model, at least in Chicago, to separate a very local phenomenon like gentrification, from a more global phenomenon like decline. Because all cities are modeled simultaneously, these predictions are also weighted not only by the Chicago trend, but by the trend throughout the sample.

Why would you do that? Housing trends in Philadelphia or NYC have nothing to do with housing trends in Chicago so modeling all cities simultaneously seems blatantly wrong.

Reply on news.ycombinator.com