The Market Mind Hypothesis, page 24
Uncertainty precedes its risk—you don’t even know exactly what the option is currently worth because you don’t know whether the model you are using is right or wrong. Or, more accurately, you know that the model you are using is both naive and wrong—the only question is how naive and how wrong. (Derman, 2004, p. 259)
The market reflects a consensus expected state of the world, with information discounted via prices. That world is the ultimate model and both real and investment life, to nuance Soros, do not take place in a laboratory, meaning that they do not involve repeatable experiments without impact on the environment.
PPT complements EMT nicely as it suggests a model of the human mind “that makes structuring our worlds genuinely continuous with structuring our brains and sculpting our actions” (Clark, 2013, p. 14).12 But we can be more specific about how PPT’s view of minds relates to markets and their extended mind. In a nutshell, and in the spirit of Soros’ earlier quote, cognitive and economic survival both depend on successfully testing hypotheses, making the least number of errors while prospering. More formally, market participants extend their internal, or cognitive, modelling to external, or economic, modelling. The Black-Litterman model, for example, is a Bayesian application used in portfolio management to optimise portfolios, especially vis-à-vis the market. In both cognitive and economic modelling, modelling includes other participants who also happen to model. Moreover, in both cases we generate hypotheses based on fragile assumptions and imperfect theories that underlie the models. With these we attempt to bridge past and future events. Based on past times, each time we want to be better prepared for the next time. Subsequently, we attempt to bridge the mental economies with the material ones. We match the subjective experiences to the objective realities, using narratives, to make sense of what we sense.
Now the crucial part: exactly because we face a world that we share with others, and thus face its states together, there is a collective dimension to all of this modelling. I would argue that this already applies to cognitive modelling, but it certainly applies to economic modelling. Ultimately, the testing of the hypotheses is played out in markets. Freely interpreting Clark in such a setting, we:
Take a highly processed cognitive product (such as an idea about the world), clothe it in public symbols, and launch it out into the world so that it can re-enter our own system as a concrete perceptible … and one now bearing highly informative statistical relations to other such … perceptibles. (Clarke, 2013, p. 15; emphasis added)
To be clear, in markets securities represent (and allow exposure) to ideas about the world. In terms of public symbols, for example, FAANG is the acronym that refers to the first letters of stocks of five prominent American technology companies: Facebook (FB), Amazon (AMZN), Apple (AAPL), Netflix (NFLX); and Alphabet (GOOG). Their prices are Clark’s “public symbols” that transcend language, cultural, and other barriers in virtue of their nature (i.e. being numbers). They bear those “highly informative statistical relations to other such … perceptibles”. Price discovery, in PPT terms, is the market’s self-organising, adaptive capability in terms of dealing with the surprises and surprisals resulting from contrasting predictions with real world news.
The error-reducing concept in PPT is formalised via the information-theoretic version of free energy, originally developed in thermodynamics. In this instance, free energy stands for the (quantified) difference between predicted sensations (as generated by models) and the actual sensations (as experienced). The free energy principle is related to active inference (see Cognitive Note: Active Inference) and states that complex adaptive systems aim to minimize free energy. In other words, the free energy principle seems to argue that it is better for survival to avoid surprises. This led to one of the criticisms of the free energy principle, namely the so-called dark room problem. The dark room problem states that the free energy principle implicitly suggests it is best to retreat to a dark room and stay put.
There have been several responses by cognitive scientists to this problem (e.g. Friston et al., 2012). If we accept the premise of the MMH, namely that mind and market dynamics are similar, then finance also has an interpretation of the dark room problem that can help solving it. It involves arbitrage. First, free energy in markets is limited because there is no free lunch. The driving force for efficiency in markets is the minimization of its free energy by way of discounting: prices almost instantly reflect surprises. In other words, any free lunch is arbitraged away via profit maximisation by rational investors. In the process, individual forecasting errors by participants, both up and down, get cancelled out. Still, such price adjustments can be volatile which, in itself, can surprise investors. So how can investors avoid any surprise? Basically, they should move all their money into cash, removing all investment exposure. That is economics’ equivalent to moving into a dark room. In the economic jungle, however, this is not a real option if only because inflation will eat away at one’s capital. Similarly, something else will cognitively ‘eat away’ in a dark room. Instead our innate curiosity makes us want to know and to experience. It drives discovery and exploration, but not mechanically: “curiosity is one motive which obviously cannot be [physically] reduced to uniformity of sequence … Intellectual problem-solving activities can be discussed only in terms of quality of conscious experience” (Knight, 1925a, p. 387). To be precise, consciousness is the portfolio of (option) returns from those endeavours (see Appendix A-4). If we zoom in on time, for example, I previously mentioned how duration and time play a role in investing. In PPT terms, the market has a temporal horizon and temporal depth. It means, for example, that research should be able to formally link the experience of “intrinsic time” by investors to the level of market stress (e.g. during crises).
In summary, PPT considers investors to be bounded Bayesian learners who, in their exchange, attempt to reduce mutual prediction error. In fact, nowhere does predictive processing take place at a more deeply global level than in the financial markets. It is through the economic system, with arbitrage and capital allocation in markets steering actions in the real economy, that we attempt to adapt to the state of the world, while changing it at the same time.
In offering a robust framework for cognitive dynamics PPT is well suited to improve our understanding of investors’ minds in general, and for supporting the MMH in particular.
Cognitive NoteActive Inference
Closely related to PPT is the free energy principle, a.k.a. active inference. Knight offers an early grasp of the idea: “We perceive the world before we react to it, and we react not to what we perceive, but always to what we infer” (Knight, 1921, p. 201).
In the section Background I mentioned that I would like you to also hear the MMH story from various experts in the specialised topics related to it. Active inference is one of those topics. So, while we are preparing a paper on it, for now I would like to share a brief explanation from its originator, Karl Friston. (This is private correspondence; slightly edited for clarity only). Here the explanation focuses on its interpretation of utility, whereby conventional economic utility is “utilitarian”:
“In contrast to conventional expected utility formulations, the utility in active inference covers every kind of outcome in all its attributes. Some preferences can be very precise (e.g. “I don’t want to be bankrupt”), some will be more accommodating and less precise (e.g. “I’m happy to work in Geneva or Lausanne but would prefer Interlaken”). On this view, active inference regards utility as supplying constraints in the spirit of multiple constraint satisfaction or—technically—in the spirit of Jaynes’ constrained maximum entropy principle. Read in this way, the balance between epistemic (maximising expected information gain) and utilitarian (maximising expected utility) is just an expression of the imperatives that underwrite decisions and choice behaviour; namely, to resolve uncertainty, under the constraint I do not incur any surprising outcomes that would violate my prior preferences (i.e. prior beliefs about being viable and successful).
In consequence, Andy [Clark] is absolutely right that both imperatives are met jointly at the same time at every decision point. In fact, the decomposition of expected free energy into expected information gain and expected utility is just one way of interpreting the underlying imperatives. When rearranging the terms mathematically, one can also express this as minimising expected inaccuracy (i.e. ambiguity) and expected complexity (i.e. risk). Both interpretations afford the same units of measurement for epistemic and utilitarian value (and ambiguity and risk); namely, bits of information.
On the other hand, whenever encountering a new situation, there will be a systematic change in the relative contribution of epistemic and utilitarian components; simply because there is more expected information gain earlier on. As one becomes familiar with the situation, the novelty declines and the expected information gain gives way to prior preferences and the accompanying expected utility”.
3.5
Integrated Information Theory
Integrated Information Theory (IIT) is a theoretical framework for understanding consciousness as integrated information. In particular, it identifies the essential properties of consciousness (called axioms) from where it subsequently infers the properties of physical systems that can account for it (called postulates). This is the reverse of the approach by other theories which usually start from the brain and then reason how it could give rise to consciousness. IIT’s main developers are Giulio Tononi, Christof Koch, and their collaborators.13
IIT considers such physical systems to consist of components (e.g. neurons) that are in a state, while being able to change that state. The basic premise is then that consciousness emerges from the integrated information within a system, not from the individual components of that system. This implies that consciousness is not just the sum of the activity of individual neurons, but rather arises from the way in which these neurons interact and communicate with one another. According to IIT, consciousness is an intrinsic property of certain physical systems, such as the brain, and emerges as a result of their unique organisation and dynamics.
IIT proposes that there are certain fundamental properties that a system must possess in order to support consciousness. These properties include:
Integrated Information: a conscious system must contain a high degree of integrated information, which refers to the amount of information that is generated by the system as a whole and cannot be broken down into its individual components. According to IIT, the degree of integrated information present in a system can be measured using a mathematical formula and is known as the phi coefficient (Φ).14
Exclusion: a conscious system must also be capable of excluding certain types of information. In other words, it must be able to filter out irrelevant or distracting stimuli in order to focus on the most important information.
Composition: a conscious system must be able to combine different types of information in order to form complex, meaningful representations.
Causal Power: a conscious system must be able to influence its own future states and the states of other systems through its causal interactions.
IIT proposes that these properties are necessary, but not sufficient, for the emergence of consciousness. In other words, a system that possesses these properties may not necessarily be conscious, but a system that lacks these properties cannot be conscious.
The two key elements of IIT, differentiation and integration, are defined in terms of axioms and postulates. Here I will attempt to translate and apply one of these to an appropriate economic setting with an example of a differentiation axiom, emphasising its contingency (in option terms: to become ‘in-the-money’). Imagine you are a fundamental investment analyst on a company (factory) visit before issuing your updated investment recommendation to clients. Your experience of that visit may include phenomenal distinctions specifying numerous spatial locations. These concern several positive concepts, such as a factory (as opposed to no factory), a machine (as opposed to no machine), a conveyor belt (as opposed to no conveyor belt), a black colour (as opposed to no black), and higher-order combinations of distinctions, such as a black conveyor belt (as opposed to no black conveyor belt). Vice versa, it can also specify many negative concepts, such as no truck (as opposed to a truck) on the company’s parking lot, no computer (as opposed to a computer) in their offices, and so on. Similarly, an experience of pure darkness and silence—for example, after getting stuck in the company’s broken elevator—is the particular way it is. It has the specific quality it has (no machine, no conveyor belt, no black, nor any other object, colour, sound, and so on). This ‘optionality’ makes your visit “what it is like” and differs it from other experiences. IIT similarly extends such reasoning to differentiation’s postulate, respectively integration’s axiom and postulate.
With its focus on consciousness, information is interpreted differently in IIT than in Shannon’s theory of communication. Shannon information is extrinsic to any subject. It measures the amount of signal versus noise. It is not compositional nor qualitative, and it does not require integration or exclusion. In contrast, IIT distinguishes a qualitative and quantitative aspect to the information content of an experience. The qualitative aspect, the quality of the integrated information, concerns the form of the associated conceptual structure.15 IIT treats conceptual structures as patterns or shapes which underpin specific kinds of phenomenality. Translated in portfolioism terms, these are the pay-off structures of consciousness’ options. The spatial nature of your factory visit as visual experience can be related to the cause-effect structure of grid-like mechanisms in the visual cortex. Ultimately, all qualitative features of every experience correspond to patterns specified by a system of elements in a state, whereas the quantity (of integrated information) is calculated via Φ.
Critics of IIT have raised a number of objections to the theory. In one of his interviews with Robert Kuhn for the series “Closer to Truth”, the well-known mathematician Gregory Chaitin discussed IIT and admitted that he “didn’t understand the math”. One of the specific criticisms is that the Φ measure is difficult to calculate in practice, as it requires knowledge of the complete causal structure of the system being studied. Some have pointed out, using the ‘unfolding argument’, that IIT requires recurrent processing. The unfolding argument claims that any recurrent process can be unfolded into one without recurrence which would—problematically for IIT—result in zero Φ. Specifically, using theorems from the theory of computation, and applying the unfolding argument, Doerig et al. (2019) show that causal structure theories like IIT “are either false or outside the realm of science”. Others have argued that IIT does not provide a satisfactory explanation for why consciousness arises from integrated information, or how integrated information could give rise to subjective (i.e. phenomenal) experience.
Despite these criticisms, IIT has been influential in shaping the way that consciousness is studied and understood in the field of neuroscience. It has also led to several new experimental techniques for measuring the degree of integrated information present in a system, including methods for measuring the coherence of brain activity and the degree of synchronization between different brain regions.
Applied to the economic system, it should be clear that prices and other market data can be considered (integrated) information in the IIT sense. Next, I will list characteristics of the market which, following a selection of IIT’s requirements and predictions (per Tononi, 2015), suggest it is a candidate for a conscious system.
IIT predicts that loss and recovery of a system’s consciousness are associated with a breakdown, then restoration of that system’s capacity for information integration. The capacity for information integration in the market is determined by the number of available, and actively quoted, securities, based on the concept of complete markets (Arrow and Debreu, 1954). In the unrealistic situation where markets are fully complete, information is perfectly integrated because it is possible to create any portfolio of securities that has a payoff for any conceivable state of the world (making the market fully deterministic). These securities can be considered the markets’ neurons which signal information, primarily via their prices. Like neurons, they are clustered, in this case across regional, sectorial, and individual asset class, reflecting specialised (economic) functions. In terms of the analogy to the physical brain, the closest we get to some kind of market cortex is the network of exchanges where securities are listed that can be traded by investors. Ultimately, the level of consciousness of the market correlates with the quote activity of securities which, in turn, relates to the conscious engagement of agents with idd-minds that provide the liquidity to the market. As I explained, in 2008 the market came close to a near fatal ‘stroke’ and ‘the ceasing of consciousness’.
When securities expire, de-list or otherwise become permanently non-tradable that part of the market’s cortex is removed, whereas if their prices become stale (e.g. due to temporary circuit breakers) that part of the market’s cortex becomes inactive. Each has different implications for information processing.
In general, the activity states that matter most for the market’s mood are the price changes, i.e. volatility, in those sectors that have the biggest market capitalisation, i.e. sectors in which attention of the collective investor is, literally, invested most heavily. These are likely candidates, for example, for the market’s “minimum partitions”.
Nevertheless Tononi and Koch (2015, p. 13) argue that IIT “aggregates”, like groups, are not conscious. By extension, this means that the market cannot be conscious (see also List, 2016). Apart from the points raised above, the MMH tackles this particular IIT view from various angles. First, I repeat my earlier argument: if we consider the mind to be extended, with cognition distributed, we cannot then suddenly cut off consciousness at the skull. Second, I point to the simple fact of scale: there is no known system where information about the world, including agents’ beliefs, expectations, and feelings, is more extensively integrated (and available as data) than the market. Third, we need to seriously consider the possibility that the market is significantly conscious because a partition of its components is not without consequences (as the GFC showed) and the existence of its mood challenges the exclusion axiom. The latter is the synergy of information the system realises. Specifically, it states that the information realised by a system as a whole is not simply the sum of the information generated by its individual parts (i.e. content), but rather the result of the exchanges between those parts (i.e. process). Finally, it is strange that Koch elsewhere has stated that collective consciousness is conceivable in the form of the internet (Reese, 2018). In Chapter 10 I will posit why the market is a much stronger candidate than the internet.
