Greetings, curious to know your thoughts. if you are searching a large data set for edge; to train validate and test. It seems that a certain percentage of results will pass the 3 part test: train, validate and test even though they do not actually correlate with the edge you are looking for (random), simply because of the large search space. If that is true, what percentage should you expect to pass the tests even though they provide no true edge.
It really depends on the market, timeframe, number of inputs, variety of inputs and also which indicators are used. It is too difficult to say that beforehand that 1% or 10% would pass.