(This is post 2 of 5 on the ‘Limits to prediction’ ACtioN (applied complexity network) meeting organised by the Santa Fe Institute (SFI) and hosted by Willis Towers Watson on 9 September 2016.)
This talk concerned the application of information theory to forming predictions. Simon DeDeo’s view is that probability is not just a statement of frequencies that are determined by physical laws, but is heavily influenced by human interaction in a system – that is, the very people trying to understand that probabilities of a system are important contributors to the randomness in that system. In this context, consistency of action (as measured by the Shannon information ratio) makes a system more predictable, whereas idiosyncratic behaviour lowers the predictability. As a result, DeDeo concludes that ignorance of the world is the primary source of uncertainty, in that it is more likely to lead to inconsistent behaviour.
DeDeo referred back to a previous session (see this thread), in which David Krakauer discussed two ways of improving our power of prediction, namely better use of data and better models for representing the world. His analysis led him to conclude that it is shortcomings in the latter that currently create the greatest impediments to our ability to predict the future.
Due to the large number of uninformed actors within a system, most of life tends towards maximum entropy, ie maximum disorder. The only aspects of the world that are truly predictable are those that are constrained by physical laws. For everything else, the world can be characterised by a jumble of interactions between people with different belief systems about the world. The difference in belief systems makes people appear idiosyncratic, although here is where DeDeo is focusing much of his research to try and expose an underlying basis of determinism.
As an example, DeDeo looked at the ability to determine political inclination based on language. In the US, he noted that the increased polarisation of the two main political parties had increased the predictability of the language used – which he equates to the way that people formulate their views of the world. Once we have established a speaker’s or writer’s political leaning, predicting the next word to appear in a sentence becomes increasingly easier. In Shannon terms, once we have established an agent’s belief system, the information ratio or predictability of that agent increases significantly.
All this is very interesting, but what does it have to do with investment? DeDeo’s research has applications to investment markets. With no predictability (ie completely idiosyncratic behaviour), markets could be said to be efficient. However, it may be possible to deduce discreet belief systems within the market ecology, and discover some degree of consistent behaviour within these belief clusters, thus elevating the potential for information discovery to something above zero – and hence forming a possible viable basis for a trading strategy. .
Simon DeDeo is assistant professor at Carnegie Mellon University’s School of Informatics and Computing.