Polling JSE market data

[HavocXphere]

Well, it's definitely a hard project - for me it was enlightening to discover how hard it is to predict the market successfully. You actually have to both predict and react to something first to cash in on it, which means some pretty exotic ML.

The more interesting question is whether a private individual has any chance of competing with the big algo houses (or worse Goldman with co-location).

 

[cguy]

The more interesting question is whether a private individual has any chance of competing with the big algo houses (or worse Goldman with co-location).

Quantitatively/automatically, probably not. For an individual, it makes much more sense to research a select few stocks and symbols extensively and understand the industry behind these businesses to see who to to bet on (or short). That is, if you want to beat the market, otherwise just buy some etfs and wait.

 

[cguy]

There are some projects which try to measure sentiment in a stock together with looking at the traditional company stats like P/E ratio, leverage etc. For example analysing what could happen to Disney stock because Johnny Depp now has restraint order for alegedly abusing his wife. So not just reacting to the possibility of a backlash against Disney as a family orientated brand, but measuring reaction and trying make an accurate prediction on how this will affect Disney revenue if at all. And doing so without introducing own bias.

It's clear that stock prices are driven by more than a company's actual value or even the value of their brands or even their patents, but also so by perception. So need to do more than focus on stats if you want to predict the market. And you have to do across many companies.

So you need algorithms which can analyse social media as well as traditional media. A fun challenge

Yeah, there has been a lot of interest in this lately. The big issue is the low signal to noise ratio on social media.