RSM318H1 Lecture Notes - Lecture 4: Unsupervised Learning, Supervised Learning, Data Mining
Document Summary
Historical average = average historical expected return from first observation to most recent observation. Will run a regression using first 240 observations. Regression model will be updated every period to estimate next period. Data mining: typically uses very simple machine learning techniques on very large databases. Computers are too slow to do anything more interesting. Misguided statistical procedure of manipulating data until a relationship is found. Lines between machine learning and data mining more blurred nowadays. Ml only helpful if there is sufficient data. Data in the past must be representative of future. If regime change such that historical data no longer applies, ml not a good guide to decision making. Machine learning is not the same as machine doing learning. Human is still doing the learning, but simply telling machine what to do. Goal is to learn a mapping from inputs to output, given labeled set of input-output pairs.
