In a previous blog, we mentioned that Marcos Lopez de Prado has been named “2019 Quant of the Year” by the Journal of Portfolio Management (see this previous blog for more details). Today (6 February 2019), Lopez de Prado was cited in two financial news reports.
In the first report, from the Financial Times, Lopez de Prado argues that the “black box” paradigm for artificial intelligence (AI), as is used by Amazon, Google, Netflix and others, is poorly suited to finance. Instead, he recommends the “causality” paradigm, which is used more by large scientific laboratories such as the Lawrence Berkeley National Laboratory.
In a scientific research lab, the goal of the research is to derive or deduce a theory of how some causal agent produces an empirically measured effect. Scientists use machine learning to narrow the search for potential causal agents. In other words, scientists use machine learning for refining and perfecting theories, not for building black boxes, since these black box approaches seldom work well in real scientific research.
Lopez de Prado argues that black box approaches do not work well in finance either. For example, constant arbitrage forces quickly subdue any signal in financial datasets, and so these datasets tend to be relatively noisy. Without a crisp theory to guide the analysis, machine learning methods usually confuse noise with signal, leading to ever-present statistical overfitting and consequent losses. As Nobel laureate economist Lars Harsen explains, “Data seldom, if ever, speaks for itself.”
See the Financial Times article for additional details.
A second report, in E-Financial Careers, summarizes some conclusions from Lopez de Prado’s recent book Advances in Financial Machine Learning, the first chapter of which is summarized here.
In this chapter, Lopez de Prado warns that “Machine learning does not fail, researchers fail.” One reason that machine learning projects often fail in finance is that investment organizations typically place trained quants into an isolated, “silo” environment, where they are unable to collaborate with others having similar training in the field. “Because nobody fully understands the logic behind their bets, they can hardly work as a team and develop deeper insights beyond the initial intuition.”
What’s more, this siloed approach often backfires due to the sheer complexity of developing truly effective algorithmic trading strategies. Lopez de Prado compares this to a single person trying to build an auto from scratch: “One week you need to be a master welder, another week an electrician, another week a mechanical engineer, another week a painter, … try, fail and circle back to welding. It is a futile endeavor.” All too often, the end result is frantic and futile search for investment opportunities, eventually settling for false positives or “overcrowded” avenues with underwhelming outcomes.
See the E-Financial Careers report for additional details.