Forecasting Markets with XGBoost & Ensemble Methods
This tutorial uses R (unfortunately; no offense to the R-fans, I used to be one of you, but I've been forced to switch) with scikit-learn (sklearn) to make a demonstrable improvement in forecast accuracy in equity markets using advanced ML techniques. I've created a similar model in Python, and for the $BTC (bitcoin) fans out there, I made one for $BTC price forecasts. While it may appear to work given the advancements in CV, though, if you haven't read The Black Swan (Taleb) or Fooled by Randomness, I suggest you do before relying on any model that uses empirical observations based on historical data to attempt to project the trajectory of the future. Taleb uses a great analogy in The Black Swan to convey the risk associated with the use empirical observation to predict events of the future (a rather sad one, to be honest) >> It is that of the loving Turkey who, gets fed, day-in and day-out, at the same time, by the same friendly human beings that take care of him. This occurs every day for 1,000 days and with each passing day, the Turkey becomes more confident that he can expect the same to occur again on the following day, until day 1,001, which happens to be the Wednesday before Thanksgiving. The poor Turkey, relying on observations about history, became very confident that his caring, loving human friends would feed him again, according to the norm, on day 1,001...but on that day everything he thought he knew was shattered. As it turns out, all of his prior observations were in a sense "anti-knowledge," at least insofar as their usefulness: they blinded the Turkey to his eventual slaughter, giving him a false confidence that grew stronger with sample size. Those prior observations gave the poor Turkey a misleading sense of confidence....when dealing in financial markets, let's not be turkeys, yeah? Because the resulting financial calamity can be enormous.