Asset Pricing under Computational Complexity
Speaker: Peter Bossaerts (University of Melbourne)
Date: Dec 7, Friday, 2018
Time: 12 pm - 1 pm
Venue: Room 1300 at Pudong Campus
Abstract:
We often think of investments as playing roulette, with “laws” that somehow can be discovered using statistics or machine learning, and optimal policies that can be acquired through reinforcement learning. Yet many investment problems actually fall in a completely different category. Firm valuation, determining what to look for when predicting markets, even portfolio construction, are not statistical problems, but computationally complex decision problems. These require methodic approaches that resonate with the theory of computation, and individuals do tend to follow those, even to the extent that the theory, developed for electronic computers, predicts human performance. But what about markets? I show that markets ought to treat these problems as if they were statistical ones, and as a result, should underperform the average investor. Experiments confirm this prediction. Still, markets help individuals make better decisions, and the improvements appear to depend on security design. This suggests a novel aim for markets, that of transmitting crucial, even if limited, information, rather than that of revealing all available information (the Efficient Markets Hypothesis). This resonates well with Friedrich Hayek’s original conjecture of the role of markets in information transmission. And it suggest that markets can play a kind of “oracle” role as defined in the theory of computation.
Speaker’s website
Speaker: Peter Bossaerts (University of Melbourne)
Date: Dec 7, Friday, 2018
Time: 12 pm - 1 pm
Venue: Room 1300 at Pudong Campus
Abstract:
We often think of investments as playing roulette, with “laws” that somehow can be discovered using statistics or machine learning, and optimal policies that can be acquired through reinforcement learning. Yet many investment problems actually fall in a completely different category. Firm valuation, determining what to look for when predicting markets, even portfolio construction, are not statistical problems, but computationally complex decision problems. These require methodic approaches that resonate with the theory of computation, and individuals do tend to follow those, even to the extent that the theory, developed for electronic computers, predicts human performance. But what about markets? I show that markets ought to treat these problems as if they were statistical ones, and as a result, should underperform the average investor. Experiments confirm this prediction. Still, markets help individuals make better decisions, and the improvements appear to depend on security design. This suggests a novel aim for markets, that of transmitting crucial, even if limited, information, rather than that of revealing all available information (the Efficient Markets Hypothesis). This resonates well with Friedrich Hayek’s original conjecture of the role of markets in information transmission. And it suggest that markets can play a kind of “oracle” role as defined in the theory of computation.
Speaker’s website