Reductionism in Economics

Reductionism is the idea that any system or scientific inquiry can be broken down into its basic components. The idea works really well in physics, where components can be enumerated, states can be perceived and qualities can be measured. Relationships, such as gravitation, magnetism, electric currents, are well understood. For example, if two objects are moving toward each other, the speed and mass can be measured, the collision can be predicted, and the force calculated. To a certain degree, the same type of simplification can be applied to chemistry. In biology, systems become more complex and the “whole is more than the sum of its parts” becomes more significant. Moving into social sciences, psychology, sociology and economics, reductionism becomes less and less realistic.

Many social scientists, however, seem to hold onto the success of reductionism in physics as a shining example and aspire to bring to the social sciences the same rigor and objectivity that they see in other fields. I see that desire as the basis of econometrics. Econometrics is the application of statistics and mathematical modelling to empirical data in an effort to mimic the dispassionate inquiry of the physicist. But this approach builds upon a number of questionable assumptions and simplifications.

Scientific experiments, at least in theory, intend to hold all variables constant, while acting on only a single variable. In some cases, that’s possible, but it’s never possible in social investigations. Even with the will to hold all variables constant, many variables, such as weather, mood, or personal history are outside of the control of the experimenter. Further, there are variables that may not even be recognized, such as what the subject ate for their last meal, what the relationship of the subject is with the experimenter or the colour of paper that the survey is printed on. But all of these things may affect the outcome of an investigation.

Compounding the difficulty,  social outcomes are not stable. Asking two similar people (eg. who both like strawberries) about a preference (eg. for ice flavours) won’t necessarily produce the same results. Asking the same person the same question at different moments (eg. the next week) may result in different answers. The results don’t scale, either. As an example, asking an appropriate sampling of people about their ice cream preference in Calgary cannot be extrapolated to Canada or to the Western hemisphere or to the world. Even if 30% of Chinese (in China) preferred red bean ice cream, we couldn’t expect 30% of ethnic Chinese in Canada to prefer read bean ice cream.

In fact, social interactions such as market trading don’t conform to the normal distribution of Gaussian statistics. The mathematics of Gaussian statistics are well developed and well understood. They can easily be applied to many aspects of the physical world. But as soon as we enter the realm of social interactions, the normal distribution becomes a mere approximation at best. At worst, it is a totally non-functional model that doesn’t approach reality. Social interactions are, at present, unpredictable and present a type of randomness that has been referred to as “wild”. Movements in the stock market, for example, appear to obey a power law that produces outcomes with wider variance than would be predicted by a normal distribution.

In order to apply mathematical models to social interactions, linearity is assumed. Linearity describes systems that have simple (eg. a single source) input and simple output. They are not affected by concurrent feedback. Only the input produces the output; the system doesn’t respond to the output. The result is a system that is predictable and controllable. We already know that the stock market, as an example of a social system, is not predictable. It is a complex system with myriad inputs, where the system responds to concurrent feedback (eg. the market drops because the market is dropping) and where the output does not conform to a normal distribution.

Behavioural economics has shown that the basic assumption of rational man (homo economicus) underlying economic theory is false. In some cases, the simplification will render results that are “close enough.” In others, however, it will produce errors. When these erroneous assumptions form the basis of management theory, which is applied in a corporation whose shares are included in a portfolio which is, in turn, based on other erroneous assumptions, the problem is compounded and surprises become inevitable. This explains why economic theory and investment theory produce unpredictable results.