My FXI neural net model continues to show an UP trend. Nice to know! :)
I realize that after 13 some odd years, this one sentence blog post with a FXI chart is very much out of context. This post was a simple confirmation of a model I built for trading/investing in this particular ETF. This was the early days of my learning Data Science and boy was my model wrong from the start.
I had built a simple classification model that took FXI’s time series and then tried to predict if the over all trend remained up or went down. The model was giving me confidence levels of 55%, just slightly better than a coin flip BUT I didn’t understand what the results were showing me.
If there ever was a living incarnation of the Dunning Kruger effect, it was me back then with these silly models.
The worst part about all this analysis I did? I used Cross Validation on a time series which completely destroys the tempsoral aspect. Later on I switch to RapidMiner’s Sliding Window approach and focus on directional changes for volatility.
Still, fast forward a few years and predicting stock market prices is incredibly hard and very error prone. Even with deep learning you might be able to forecast the next price tick with some degree of confidence but that edge doesn’t last long.
Market makers are a bit a different animal and machine learning does help there. They take the opposite side of a trade if and only if there’s something ready to buy or sell. The can forecast in real time what the trend looks like for a bunch of incoming ticks and know which way the market trend is going for a particular price in a very short time horizon.
So all is NOT lost when using Data Science and Machine Learning to predict ‘stuff’ in the markets. It’s usually not a stock or bond price itself but all the other support systems around the market. Especially text analytics for things like 10K’s and finding market sentiment. That works but makes everything WAY more complicated then a simple trend model like the one above.