"An algorithm could really do better than humans, because it filters out noise. If you present an algorithm the same problem twice, you'll get the same output. That's just not true of people."
"But humans are not very good at integrating information in a reliable and robust way. And that's what algorithms are designed to do."
- Daniel Kahneman interview "Where Humans Meet Machines: Intuition, Expertise and Learning" MIT IDE
In a recent interview Dan Kahneman talks about the problem of decision noise. When a group of decision-makers are given the same set of repeatable tasks or problems, you would like for them to generate the same answer, yet that is not the case. There is limited consistency with decision-making.
Kahneman discusses a research test done with a set of insurance underwriters who are given six sample problems. He believed that there was a surprising amount of noise given the controlled experiment with disagreement about 56% of time. Professionals with the same training and facing the same stylized problem should generate the same answer. That is not the case.
Investors may face the same problem. There seems to be a significant amount of noise around decisions that are facing similar sets of facts. Take any economic announcement as a simple case. There is a significant amount of trading volume and thus disagreement on how to interpret the new facts. Perhaps investors always think this time is different. More precisely, investors may be using different models or have different believes that will lead to different conclusions. Reality is only revealed later. This may not be a bias but just dispersion around reality. Investors may be facing too much information that creates a situation for decision noise to exist, (see Pump down the noise - Decision silence).
An algorithm will allow for similar reactions to a repeatable set of facts. There will still be error but a repeatable process will shrink the noise and allow for feedback on what may have gone wrong. Ending randomness with decision-making is a positive that can be achieved through using models.
Kahneman discusses a research test done with a set of insurance underwriters who are given six sample problems. He believed that there was a surprising amount of noise given the controlled experiment with disagreement about 56% of time. Professionals with the same training and facing the same stylized problem should generate the same answer. That is not the case.
Investors may face the same problem. There seems to be a significant amount of noise around decisions that are facing similar sets of facts. Take any economic announcement as a simple case. There is a significant amount of trading volume and thus disagreement on how to interpret the new facts. Perhaps investors always think this time is different. More precisely, investors may be using different models or have different believes that will lead to different conclusions. Reality is only revealed later. This may not be a bias but just dispersion around reality. Investors may be facing too much information that creates a situation for decision noise to exist, (see Pump down the noise - Decision silence).
An algorithm will allow for similar reactions to a repeatable set of facts. There will still be error but a repeatable process will shrink the noise and allow for feedback on what may have gone wrong. Ending randomness with decision-making is a positive that can be achieved through using models.
No comments:
Post a Comment