90% of stock trades are through algorithmic trading most of which is machine learning...

and if machine learning algorithms are to perform well, they must use training data. So if one were to train a neural net or some algorithm to trade the stock market and make money, one would train it with existing stock market data.

After training, they lock in the weights, and that's the neural net that is used as an input-output machine and given money to play in the market. How many different machine learning algorithms do you think are actually used, that actually have enough money to sway the market? There are likely some bots that are essentially throwing around billions of dollars an hour for the largest algorithmic investing firms, and these bots are precisely trained and tested against real-world scenarios.

The bots learn from the past. So the more the stock market is traded with algorithms (which has been the trend) the more the future will look like the past, but averaged out across those few dozen or hundred different algorithm designs that essentially drive the stock market today.

TL;DR: The more the market is traded with algorithms that are trained from the past behavior of the S&P 500, the more the future market behavior will begin to look like an averaged-out version of the past.

Reference chart of the S&P 500, note the 3 humps of around 2000, 2008, and today

Submitted April 19, 2016 at 12:22AM by magnora7 http://ift.tt/1qVkThG