DataCrunch’s Risk-Informed Financial Machine Learning
In financial markets, machines, like humans, tend to load on risk to maximize returns: quantitative finance and machine learning
in particular, are not guarantee of absence of biases in their implementation. The Hedge Fund industry has been identifying a number of common and risk dimensions of return that, biased as they may be, are a great way to build uncorrelated financial products (quants would use the word “orthogonal”) that evolve independently of the big swings of financial markets, while living inside them. This is analogous to string theory: there are trivial dimensions in our reality, space, and time. But then there are other small ones, hidden under our present capabilities to perceive them and move along them in a smart way. The good news is that “Microscopic alpha today is much more abundant than macroscopic alpha has ever been in history. There is a lot of money to be made, but you will need to use heavy ML tools” [1].
Crowdsourcing Machine Learning is one such tool, that has to be carefully crafted to get the best out of a herd of talents: even acknowledging the virtue of complexity [2], problems have to be well posed, and “in nonexpert hands ML algorithms can cause more harm than good” [3]. One of the main reasons for this is the pressing need to align the alpha generator incentive with the consequential capital allocation rationale and its constraints [4]. This issue is critical and can only be partially mitigated with feature engineering and financial data labeling.
0 Comments