Friday, March 19, 2021

Discretionary versus quant systems - Can you tell the difference? - The "I, Robot" problem

Can you construct a test to determine whether a fund is discretionary or quantitative/systematic without asking the manager directly? There is much discussion about quants versus discretionary, but how can you tell the style without the manager making the declaration? 

On the one hand, this is a distinction that should not be relevant. Performance success is all that matters, yet there is so much discussion between the relative value of quants and discretionary managers that this choice should be thought through in more detail. At a high level, what makes a systematic fund special? 

The Turing test has been pondered by many; can a computer pass for a human? How could someone tell the difference between a computer and human managing money through a set of questions? 

This question has been a repeated issue in science fiction writing like Isaac Asimov's classic "I, Robot". Can we separate the discretionary manager from the quant robot? Put differently, how can you tell the difference between a human and robot running a portfolio?

Can we distinguish human management and discretion by their mistakes? Humans will have behavioral biases. Robots can be programmed to eliminate these biases. However, robot managers can still make errors and have biases. Robots may will only use information programmed to assess. Robots may have a more difficult time adapting to changes in market regimes, but humans may also have the same problem.   

This question gets harder as we add machine learning programs to the quant field. With machine learning, a computer program is able to adapt and learn. This could be closer to the discretionary manager. A discretionary manager may be able to see new linkages between data and performance, yet big data and data science is trying to do the same thing. 

Is the key feature of a discretionary manager the ability to be creative? is the key fault the ability to make mistakes? Is the key feature of a quant system the assessment of large amounts of data? Is the key fault with inflexible behavior when faced with change? Is there a simple answer to distinguishing between robot and human portfolio manager? 

No comments: