macrocephalopod

Good questions in this thread. Maybe people like @chanep @robertmartin88 @macrocephalopod @therobotjames @FabiusMercurius @financequant would have some advice?

As much as I like Julia, you have to be careful to avoid "shiny new toy" syndrome. Are there any specific aspects of your data pipeline that are slow? If it's general data processing, you should first make sure all your numpy code is vectorised.

Another important variable is your level of coding ability. Julia gives you C-like speeds at a python-like level of ease (but it is likely neither as fast as optimised C++ nor as easy as python). If you can productively whip up performant C++, why bother with something else?

My general view is that you should figure out your edge before thinking too hard about implementation. Practically, this means that you should use whatever language has the smallest "brain to test" time.

I think one of the reasons quants focus on programming languages, backtesting frameworks etc is that stuff is comfortable and well-defined. All edges come from intuition about market structure and actors. Which is murky, uncertain and intimidating. And what you need to focus on