Hacker News new | ask | show | jobs
by AstralStorm 2165 days ago
The "large enough" argument does not work if you explicitly or implicitly filter or stratify the input to your sampling function. Just like law of large numbers does not work for heavy tailed distributions directly.

In which case your client do not represent the market, either by volume, or by number.

If you can actually know in which ways your sample is different, you can play it then against the market.

Robinhood would optimize its input for small clients with less knowledge, or clients looking for highly speculative risky plays of smaller volume. These are very specific characteristics that likely differ from general market.

1 comments

I'm not talking from the perspective of RobinHood, I'm talking about the broker/dealers who provide execution services to RobinHood. RobinHood is nowhere near large enough to have a client base which represents the market. JPM, GS, MS etc are.