Deep Learning in the Stock Market

I did something similar( with two other people) for my work hackathon, the difference was that my program analyzed news and matched the subject of that news article to a traded stock, and then predicted from the sentiment of the news article whether the stock is gonna go down or up. Very simple, and it worked most times, but could be improved on a lot

I commend this guy for trying to improve improve his skills and writing about it.

I'm not sure why he said he's just assuming the stock market data of 1000 stocks offers the type of structure relevant for NLP and this is plausible.

But even if it did, he might be in for a world of hurt trying to predict the VIX with it. The VIX is based on the prices of near term options, not stocks. The connection to stocks comes via these SPX options using the SPX weighted index of stocks. The VIX isn't trying to be accurate though, its a standard deviation estimation for options expiring 30 days into the future.

Who knows, correlation isn't causation and this doesn't seem possible, but anything can happen and he might find something. But then you're left with a black box.

If structure can be inferred in financial data, it's bound to change in unexpected ways, at some point. I'm not familiar with NLP, but I'm guessing for all intents and purposes the structure of language is assumed to be fairly static.

buy on the rumor; sell on the news.