The only thing I can speak to is crop yields. Satellite imagery S flyovers won't really teach you a lot outside of a binary yes/no r field is alive.

The technology farmers use teams yields to the square foot. I'm betting you can buy that from John Deere and others. That's probably where they get that?

I don't know how effective the project was because I left before it finished but we were working on using the normalized difference vegetation index, normalized difference water index, and a bunch of other data points to try to predict crop yields.

Farmers might get that kind of stuff from flyovers for the extra resolution but that was cost prohibitive for our purposes except when validating data from the satellites.

Yeah good luck getting accurate NDVI measurements down to the square foot from satellites.

Luckily corn fields go on for miles so we didn’t need that resolution.

Fields vary wildly within relatively small spaces. Just because it looks uniform doesn't mean it is uniform.

Our 100 acre bottom field this year averaged 274 bushels of corn per acre. Some places were around 360, some were around 50.

If you're trying to predict crop yields to beat the market, granularity is wildly important.

This year in university I took a course where I had to imagine creating a startup and we ended up mocking up an agritech business that dealt with drones and satellites data. We discovered that Sentinel 2, a European satellite, has a resolution for ndvi that goes as low as 10/20m. This improves nasa's modis sitting at a resolution of ~250m.

Furthermore, this is free. There are companies providing paid images with higher resolution from their own satellites (Constellr comes to mind). I'm sure the resolution isn't going to be exact to the square foot, but for some cases you probably don't need such a high resolution

> major ports might get daily fly bys to track containers and boat traffic

Shipping traffic has transponders these days, no different to air. Don't even need drones.

Don’t forget the easiest dataset to use: they buy consumer transaction data directly from Visa and Mastercard.

With that in hand it’d be pretty easy to know if, say, Apple will beat or miss revenue for a given quarter.

> With that in hand it’d be pretty easy to know if, say, Apple will beat or miss revenue for a given quarter.

Wow, seems so easy! That's not how the market works though. You've got to know if Apple's going to beat the analysts' expectations of how Apple is going to do. All the analysts have access to that data and that's table stakes.

Apple's guidance on revenue is like 30% of the equation.

Typically, because they make so many bets, a hedge fund needs to be only 53-54% accurate with their predictions to make money.

The stock of every company will move a decently large amount on the day quarterly revenues are published, whether it's Apple or a smaller firm. You can make a lot of money via options maturing on such days. When you need to be right only 53% of the time, these signals _really_ add up.

>Typically, because they make so many bets, a hedge fund needs to be only 53-54% accurate with their predictions to make money.

Wouldn't that imply that every bet the fund makes is roughly of the same $$ value? I would figure more senior traders or more successful ones would vary the size of their bets relative to certainty or uncertainty of an event outcome (could be quarterly earnings, a political/geopolitical event, etc)?

Don't think of it like buying a stock on the spot market like retail investors typically do. Many hedge funds place their bets by using the options market, which limits the downside on a bad bet (see e.g. the butterfly strategy [1]).

The size of the bet also depends on a lot of factors. Many hedge funds don't work like monolithic entities, rather they have a bunch of portfolio managers (PMs) who have their own allocated pools of assets from investors. These PMs decide on placing their bets based on lots of different things like their liquidity position (e.g. you don't want to lock up your cash in one position even if you're sure it'll make money when there's a risk of being margin called because of another trade) in addition to signals such as the one dicussed above.

[1] https://en.m.wikipedia.org/wiki/Butterfly_(options)

The use of aerial imagery for institutional asset tracking is older and much more widespread than we'd think.

Back in 2005/2006 my university maths professor (in Finland) had an established side gig in the US. He had polished a process to manufacture aerial drones - still called UAVs at the time - and had put together a fairly slick software pipeline to combine GPS tracking with digital imagery. One of his longer-term clients at the time was Harvard University; they had contracted his firm to get routine data on how their land endowments were doing.

An associate professor at the CS department has had a similar thing going on since ~2005. His company does drone imagery for land owners in Finland. Rather surprisingly a notable fraction of his business at the time was coming in from families and corporate offices wanting "just" nice looking shots of their various farm buildings.

At the time the university in question was rather well known for their computer vision unit.

Try 1910's. Reminiscences of a Stock Operator describes Jesse Livermore collecting unique public data on shops and use for trading. The founder at my first trading job gave me this book to read, and I am eternally grateful.