The internet is now flooded with “predicting stock market prices using LSTM”. I went through 9 articles that I found on websites like medium, KDnuggets, etc. And I realized almost 6-7 out of them showed promising results. But none of them showed their real-life use-case; the question is is it beneficial?
LSTMS predict T+1th term by previous k terms of time-series, say k=2, So we need to have T and T-1 to predict T+2 so suppose my X for input is Open, High, Low, Close, Volume, and Y would be the next day’s Close
=> I need to have OHLCV values of the previous two days to predict the next day’s Close
=> If I need to know tomorrow’s Close, I need to wait till the current day ends to get its OHLCV values, i.e. it predicts only one candlestick in the future.
=> One article was like using previous k days Close to predict next days Close.
So for T+1, I would give it T’s and (T-1)’s Close as inputs and for T+2; T’s, and (predicted_T+1)’s Close as input.
Again, this is not useful because if there is an error in T+1, it gets propagated in the following sequence
Also, logically, we cannot predict movements based on only closing prices. OHLCV may work because it might detect a few candlestick patterns like Doji, Harami, etc., and consider volume.
LSTM’s in those articles showed good predictions because they predict only a single value instead of a range of values. So are they worth a single candlestick? Because even indicators like moving averages give a proper estimate for a single candlestick.
I think there should be a different approach to using LSTMS; I didn’t find any article. Please comment down below if you see something. I am not an expert in these fields — I am just putting forward the conclusions I got to after my explorations with this topic, so feel free to point out anything I missed. Thank you for your time.
