danieldwilliam | On the Quality of The Science of the Dismal Science

This post started out as a reply to a comment on this entry. It got a bit long and became a post of its own. Below I attempt to excuse the poor performance of macro-economics as a science and demonstrate the better peformance of micro-economics.

It is certainly the stated aim of the science economics to understand economic activity so that testable predictions can be made and be seen to be correct. In that sense it is a science.

It’s ability to do that in macro-economic situations is relatively poor. I offer the following explanations and excuses.

The range of possible outcomes is much greater than the range of interesting outcomes. A prediction that, next year, people will buy and sell goods and services much like they did last year but perhaps a little more of them instead of the whole shebang descending into outright anarchy is of less use to the people who use economic predictions than whether GDP growth in the UK is 1.1% or 1.2%.

Newtonian physics ability to predict that the sun will rise tomorrow is much better, but less use, than its ability to predict whether you need to wear a sunhat or a rain hat.

Secondly, macro-economics is looking at very complex systems, made up of other complex systems. The interactions of these complex systems is difficult to understand because it’s quite difficult to isolate them experimentally. (1) Even it were possible to get whole economies in the lab there is so much data that it’s difficult to analyse. So, the macro-economists ability to do primary science is very limited. They can use a whole slew of real data to try and understand what has happened in historical situations but there are problems with that. Firstly, there is lots of debate about which data are real, influential and complete, which confounding, which complementary. Some of the data is erroneous. Some is corrupt. There are lots of inputs and lots of outputs and the lines between them are not well understood. It’s also not exactly clear whether the inputs are actually inputs or really outputs or just noise. The second…

The second is the third big reason why macro-economics struggles. All the situations it is using to create a model of how economies behave are unique. The 1929 Crash and Depression has a lot of similarities with our own current situation but it’s different enough that it is unique (2). The next Great Crash and Depression will be different again. You can gain some insight by comparing sets of similar events and trying to isolate differences in input. This is, to use a technical term, bloody difficult. And again, there are the usual arguments about the quality of the data and it’s interpretation. What you see when you see macro-economists arguing is two people with limited data arguing about whether a unique event was Type 1 or Type 2 with an occasional heckle suggesting it’s Type 3 and someone suggesting that the Types should really be Class A, Class B and Class Pyjama.

In historical terms we’ve had perhaps 6 big recessions or depressions in the UK and maybe 800 adjustments of the Bank of England base rate. That’s not a lot of data to build a workable theory. Not compared to the number of times balls have been dropped.

If you want a historical analogy for the science of macro-economics – picture yourself as a 17^th Century physician trying to explain the Great Plague of London.

Micro-economics, which is where I play, has a much happier situation. It’s trying to create theories and models that predict the behaviour of smaller systems with fewer participants. We’re looking at the behaviour of individuals and firms. People are buying and selling things all day every day so there are lots and lots of unique data sets. You can isolate some interactions in a lab and replay them changing the variables one at a time. For those that you can’t there are sufficient volume of interactions that even if we can’t isolate them in the lab we can do meaningful statistical analysis. We can analyse the behaviours, the inputs and outputs, we can drill into the underlying psychology or the maths. And we can test lots of different, seemingly unique situations, again and again until we are able to discover universal rules.

Micro-economics predicts that if the price of a good rises then the volume of sales will fall. You might well say, well, no shit Sherlock, but up until Galileo conducted his experiments people thought that the large cannon ball would hit the ground before the smaller one. So micro-economics predicts that if a price rises, volumes will fall. It offers an model for predicting by how much volumes will fall and what the consequent impact on total revenues will be. It offers a theory for explaining why some goods, Giffen goods, behave differently and sales volumes increase when the price increases (3) and you can use this to predict which goods are going to be Giffen goods. If you see a good behaving with a strange price elasticity of demand you can run experiments to see if the good is a Giffen good or a Veblen (high status goods).

The theory of transaction costs offers a predictive model for where the boundaries between different organisations will lie based on the comparative cost of performing specific operations in house or hiring out the work. It tells me that firms will be reluctant to hold rare and expensive specialist services on their books 365 days a year if they can hire in consultants when they are needed and only when they are needed. It tells me that small firms therefore won’t have internal consultancy departments but that very large firms might do for some specialist roles. And lo, MLW and f4f3 both land contract work doing specialist labouring.

Most excitingly for me, we can do economics experiments on non-humans and get decent results predicting the way they behave and how that behaviour will change over time if we change one of the input variables. The maths of economics appears to be coded into the genes of species.

So, macro-economics has problems with prediction because it’s ability to gather experimental data is poor. It’s a problem intrinsic to the nature of the beast it is studying. The theories are untested and a bit wild. This won’t really improve until we have much more data. Check back in 2512. Micro-economics is much better at predicting outcomes using a theory. Theories can be built and quickly tested in the lab and in the wild.

One way of looking at stock and commodities markets is as a series of experiments where scientists pretending to be traders use a predictive model to work out if a price is going rise or fall.

(1) an example of an attempt to perform economics on an experimental basis is North Korea. A second, perhaps happier example, is Cuba.

(2) e.g. no China and India growing, or not, at 10% per annum.

(3) Giffen goods, fascinating and one of my first loves as an economist.