3 ms·

This is a bit like a popular hedge fund strategy known as alpha capture. In the alpha capture strategy, funds look at the buy/sell recommendations made by analy

This is a bit like a popular hedge fund strategy known as alpha capture. In the alpha capture strategy, funds look at the buy/sell recommendations made by analysts (for example those at the major banks). Applying a systematic approach means that they can measure the accuracy of each individual analyst and choose whether or not to follow along by trading the advice from that person. The best analysts eventually get hired into funds for their picks to be exclusive. I built something (DueDilly) that works in a similar way by monitoring the performance of Reddit user's stock recommendation and tracking how each idea performs and which users do well. You can choose to trade alongside them based on their track records. My site: https://duedilly.io/ https://duedilly.io/ Previous Discussion in my Show HN: https://news.ycombinator.com/item?id=28244744 https://news.ycombinator.com/item?id=28244744 There are various academic papers that also discuss it like: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3873884 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3873884

bo1024 5y ago
I think the most interesting question is how you can tell if someone is actually good vs fluctuations due to luck. I'd think it takes a long time to get statistical significance. Or, during a bull run the optimistic traders all look great for a while, but this later proves illusory.
- Galanwe 5y ago
  You just summarized the main problem of quantitative finance: you need points to assert statistical significance. And thus why data providers with long history are paid so much.
- duedillyio 5y ago
  Truly. There's some standard approaches like beta-adjusting that can be done to put things on equal footing in different market regimes but it's only one part of what determines a stock's performance. Testing for the significance of someone's average returns for example often can be rules of thumb. For example, given a return stream of a trading strategy, the Sharpe ratio (mean return/vol) is directly related to the t-stat of the mean return being different than 0 (mean ret / (vol/sqrt(n)). Several simplifying assumptions must be made about the distributions of returns that don't line up with reality to do the above but it's mostly a heuristic that will let you say "ah, the Sharpe is X over the last 5 years? That's significantly different enough than 0 for it to be tradeable."

bo1024 5y ago

I think the most interesting question is how you can tell if someone is actually good vs fluctuations due to luck. I'd think it takes a long time to get statistical significance. Or, during a bull run the optimistic traders all look great for a while, but this later proves illusory.

Galanwe 5y ago
You just summarized the main problem of quantitative finance: you need points to assert statistical significance. And thus why data providers with long history are paid so much.
duedillyio 5y ago
Truly. There's some standard approaches like beta-adjusting that can be done to put things on equal footing in different market regimes but it's only one part of what determines a stock's performance. Testing for the significance of someone's average returns for example often can be rules of thumb. For example, given a return stream of a trading strategy, the Sharpe ratio (mean return/vol) is directly related to the t-stat of the mean return being different than 0 (mean ret / (vol/sqrt(n)). Several simplifying assumptions must be made about the distributions of returns that don't line up with reality to do the above but it's mostly a heuristic that will let you say "ah, the Sharpe is X over the last 5 years? That's significantly different enough than 0 for it to be tradeable."