Hacker News new | past | comments | ask | show | jobs | submit login

The real problem with these sites is that it is super easy to overfit to historical data. Try enough indicators and enough parameters and you can fit historicals to the dot. Except, no one knows how the model performs on a go-forward basis. On Wall Street, these are often easy to debug -- you just run on live data, often with live trading to see if the signal is real. However, with economic data, the velocity of new data is too low to really test models on a go-forward basis...so you just have a theoretical model, likely overfit, that has little predictive value.



Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: