Question re: Regressive analysis explain-ability

If I am able to identify an historic trend using a large sample size (to negate volatility), and the trend fails to be explained, even after exhaustive conventional and 'outside the box' theory applications, does this render the voracity and/or value of that trend moot?

I have read it many times here that simple 'bill the cop' data mining (sorry BTC), is at best useless, and at worst a predictive disaster.

Would any of you very bright 'math' guys ever accept that trend into a model with the caveat that sometimes events occur often in this world for reasons beyond our ability to understand?
Or is the rule unbreakable that demands a logical reason and 'in the lab duplicitity' for an historic trend to be considered seriously as a useful predictive component?

Really interested in the tank's take on this.

Divine intervention? What could possibly be inexplicable?

The answer is you obviously don't know enough about regression analysis to be using it. I'd otherwise try to give a more useful reply but that's the culprit here.