2 ms·
Like most software, quant investing models are usually separated from production systems when being created and undergo integration testing, back testing, etc.,
by mlthoughts2018 7y ago
Like most software, quant investing models are usually separated from production systems when being created and undergo integration testing, back testing, etc., and have all sorts of automatic and manual revert capabilities in the deploy systems.
Not much different than shipping something broken to millions of users and needing to triage a rollback.
Of course there are outlier firms that play fast and loose and can have risk of huge sudden losses that can’t be stemmed, but most firms do risk management as a core part of their purpose for existing, and this applies to risk of bad code / bad model just like anything else.
- onlyrealcuzzo 7y agoYeah, but even with integration & unit tests, people regularly miss some corner case. Again, maybe I just don't know enough about how it works, or maybe I just know enough about what I do -- but every once in a while you make some catastrophic mistake that takes done Google or Facebook for a few minutes -- the only cost you incur is opportunity. The same mistake at Knight Capital cost them like $500M in 15 minutes or something crazy. So, taken a step further, the little bugs that regularly make it into Google that cause 1% of requests to 500 on some service is a big deal, the opportunity cost can exceed your salary for years. I imagine the same thing is true in trading, except it's not opportunity -- it's real money. That just seems way more stressful to me, but maybe it just doesn't happen very often, or the test coverage is taken much more seriously, and it's better.