Knowledge gathering from complex systems: not from first principle

Excerpt

An interesting paper published by the US National Bureau of Economic Research in October this year suggests a strong causation between PM2.5 particulate pollution in lower Manhattan and the movement of the stock market averages (WSJ 2016). The paper claims that higher PM2.5 in the ambient air depresses stock, and vice versa. Further, the correlation with aggregate air pollution is weak; PM2.5 is where the action is. The explanation is that these fine particles enter the blood stream and trigger physical distress and a decline in mental acuity, as a result the induced fear among the traders in the stock market leads to a loss of confidence in stocks. The authors claim that the result is statistically valid and is in sync with findings of other researchers in similar inquiries. Wise men have opined that correlation is not causation, so at first glance, the result may appear incredible. But considering the larger context of the impact of fine particulates on physical and behavioral health, further research is warranted. More comprehensive medical rationalization would also be helpful in understanding the underlying mechanisms. The result attracts attention not merely because of the mental hazard inferred in this study but also because this is an interesting practical example of learning from aggregate effects, which is the reverse of learning from first principles. This is relevant because statistical treatment of big data through techniques such as machine learning is gaining strength as a field of study. It opens up the enticing possibility of learning the characteristics of a complex system by big data techniques more easily than by first principle techniques. At the very least, learning from aggregate behavior will aid in sharpening the scope of basic research to enlighten the mechanisms of the aggregate findings. …