4 ms·

I went to a recruiting talk by Two Sigma a while back where they discussed the same thing as an example of "look at this cool stuff we get to work on". They had

I went to a recruiting talk by Two Sigma a while back where they discussed the same thing as an example of "look at this cool stuff we get to work on". They had slides with satellite images of some parking lot and showed how they could segment cars, deal with car-like things on roofs, map regions of parking lots to the right stores, handle different conditions like snow, etc. From a technical perspective it's actually an interesting problem because there are so many things to consider. Really makes you notice how disadvantaged retail traders are, though. I think traditionally a lot of hedge fund strategies are kept pretty secret, but this one is so out of reach and technically difficult for most people that they don't really care if it gets out.

mc32 7y ago
How do they deal with underground parking in urban areas especially?
- mbreese 7y ago
  You wouldn’t. But just because a prediction can’t be 100% accurate because of missing data doesn’t mean that the prediction isn’t useful. Really, what they are likely comparing is the relative changes in parking lot occupancy over time. So it’s not necessary to have complete coverage of all parking lots. You just need enough coverage to make the prediction with a certain level of confidence.
- bluGill 7y ago
  They don't need to. People who park underground, walk or use public transit are close enough to people who park above ground that they get enough data without that. In most cases just one WalMart is enough to infer everything you need to know about them all - if you select the right one for your question. (you wouldn't select just one because then you need to ask if there is something specific about that one, but a statistical sample is enough). Unless your thesis is people who park underground shop differently. That is a valid idea and you cannot answer it via this study. However you can study it by other means if you believe this is important.
- floatrock 7y ago
  buy the feed from the nest camera across the street from the entrance
- mgfist 7y ago
  You can count cars going in and out.
  - mc32 7y ago
    Aren’t they using LEOs for this? How do they get it to hover and capture traffic patterns? Or you might be saying having alternative methods to count traffic like cameras or other ground based counters.
- ThrustVectoring 7y ago
  They don't, just as they don't get a sense of the average purchase amount per trip. If above-ground parking lots get 20% more traffic, you can expect roughly 20% higher retail sales with some error bars based on the historical correlation. The game here is essentially to get noisy signals about companies earlier than the official ones provided by the company. Take satellite pictures of parking lots, do phone surveys of consumers, count trucks leaving supplier companies, whatever. If it's a good predictor, you know upcoming price moves early, and can likely make significant profits off of it.

mc32 7y ago

How do they deal with underground parking in urban areas especially?

mbreese 7y ago
You wouldn’t. But just because a prediction can’t be 100% accurate because of missing data doesn’t mean that the prediction isn’t useful. Really, what they are likely comparing is the relative changes in parking lot occupancy over time. So it’s not necessary to have complete coverage of all parking lots. You just need enough coverage to make the prediction with a certain level of confidence.
bluGill 7y ago
They don't need to. People who park underground, walk or use public transit are close enough to people who park above ground that they get enough data without that. In most cases just one WalMart is enough to infer everything you need to know about them all - if you select the right one for your question. (you wouldn't select just one because then you need to ask if there is something specific about that one, but a statistical sample is enough). Unless your thesis is people who park underground shop differently. That is a valid idea and you cannot answer it via this study. However you can study it by other means if you believe this is important.
floatrock 7y ago
buy the feed from the nest camera across the street from the entrance
mgfist 7y ago
You can count cars going in and out.
- mc32 7y ago
  Aren’t they using LEOs for this? How do they get it to hover and capture traffic patterns? Or you might be saying having alternative methods to count traffic like cameras or other ground based counters.
ThrustVectoring 7y ago
They don't, just as they don't get a sense of the average purchase amount per trip. If above-ground parking lots get 20% more traffic, you can expect roughly 20% higher retail sales with some error bars based on the historical correlation. The game here is essentially to get noisy signals about companies earlier than the official ones provided by the company. Take satellite pictures of parking lots, do phone surveys of consumers, count trucks leaving supplier companies, whatever. If it's a good predictor, you know upcoming price moves early, and can likely make significant profits off of it.