That comes down to what you think drives the stock market. There are many factors, not all are observable. That's why no one really knows where AAPL stock will be at the close tomorrow. Will it be up or down? No one really knows. Take the most informed market participants on Wall Street, ask them where the market will close tomorrow, up or down? Their average prediction accuracy will be a little higher than 50% if they just say "higher" every day (market goes up more often than down). Few can predict much better than that over a long period of time. I don't see how an AI can improve on that, at least with long-term predictions. Short-term trading, maybe. Btw, there's already AI robo-funds, most I have seen such as this: https://www.qraftaietf.com/ Underperform the market.
Prediction is probability. If my quantum driven AI system can accurately predict 60% of the time, you have a winner. You don't need to be the fastest runner from a charging bull, just not the slowest.
You don't seem to get it. You don't need to make a long-term prediction. All that you need is to accurately forecast the NEXT bar or candle. Not 10, 20 or 30 bars in the future. No, just 1 bar is enough. Do that every time and you will become uber-rich. Get it?
Actually, there is a clear market for long term predictions among big institutions. HFT has liquidity capacity constraints. They would all love to just have to predict the next millisecond or even minute if it was all they had to do.
In the one side you got someone developing A.I to beat the market. In another there's institutionals with proffesionals developing at which suppose to beat that which is suppossed to beat the market. And then there is human greed and get rich quick in the middle. Yh, good luck.
They won't have to... the vix will go to less than 5 and the indexes will look like pancakes for the most part. If one operator with 10's of billions aum has one, you can bet three or four more won't be far behind, and they'll be battling it out for 1/100's of a penny. This volatility we all enjoy so much... it'll be gone. Like the sound decaying off a tuning fork.
Speculation. The training set isn't current nor connected to real-time data. When an api gets rolled out or one trains on their own data, then something may come of it. Chatbot can give very intelligent answers, but some have been proven to be factually incorrect. Came across this summary of 30+ classes of HFT algos running co-currently; but upon further DD, some answers are re-hashes.
Of course but not with an NLP algorithms whose only purpose of existence is to make sense of linguistic and purely textual data.
Why do you think hft pay vasts amounts of money for alternative data. It's to push up that edge by some positive delta. You are mixing up entirely different domains. Long term investors suffer significant risks that hft algorithms don't and vice versa. Different types of risk. The normalized metric to compare is "edge" not total return or win probabilities on individual trades.
Well but that is by choice, each institution defines their own edge and niche and allocate capital accordingly. Even Fidelity employs hft algorithms and hired dedicated domain experts for execution purposes. Sure, there is a liquidity constraint baked into hft but entities make choices.