3 ms·
> It’s perfectly possible for any individual to find tradable effects just like RenTec does: find data (maybe weather reports), look for a relationship to futur
by bko 5y ago
> It’s perfectly possible for any individual to find tradable effects just like RenTec does: find data (maybe weather reports), look for a relationship to future securities prices (how about agriculture company stock prices) and build a predicative model. Repeat until you find one that works.
This made me chuckle. Analyze data and figure out how it correlates to future prices. Why has no one ever thought of this?
- typon 5y agoActually, in a weird way he's right. It's very easy to find correlation with prices. If these correlations hold for future bets is an entirely different question.
- bko 5y agoI was following Quantopian because it answered the question. Can you find a signal. They did a great job at giving you the tools to test your strategy and unsurprisingly a lot of people found a lot of signals. They even had standard best practices on using test and validation sets to ensure you weren't data mining or overfitting. So they created a hedge fund with the best performing strategies, and it did very poorly. Backtesting is a non-trivial matter and is incredibly hard to do correctly. You can lie to yourself or do it poorly. You can basically fit to the test set or create enough iterations to find one that does equally well on train, test and validation, but without a lot of experience, intellectual honesty and the right incentives can you turn that into a strategy. That's essentially what RenTec built. A backtesting and risk control system so they can look at data in a sanitized way, test strategies and ensure they're profitable going forward.