As a comment on my previous article ("Science and Economics") André asked, "If we are unable to test macroeconomic theory, how will we know that it works?" If we use a wide definition of "test," we are able to do so. However, this notion of "testing" would probably raise eyebrows among physical scientists, who perhaps assume that "forecasting" and "testing" would be the same thing in this context. It is possible to look at macro in a rigorous way, but we need to drop the embedded assumption that rigorous means the same thing as acting like physicists.
My arguments here should not actually be surprising to economists, as they are effectively a hidden background assumption in their worldview. Instead, this viewpoint is aimed at non-economists who want to treat macroeconomics like other fields of knowledge.
Rigorous Depends Upon ContextDifferent academic disciplines have different standards of rigour. A paper in pure mathematics is written in a different way than a paper in applied mathematics. Historians follow standards for argumentation, which have varied over time. Philosophers seem to be very rigorous, but they do such a good job of confusing non-philosophers that nobody is entirely sure.
The previous observations should be extremely obvious, but the discontent with the status of economics has lead many non-economists (like myself!) to stick our noses in. In order to say anything useful about macroeconomics, we need to understand the context of the field of study.
(Once again, I am ignoring all areas of economics that are not macroeconomics. The traditional definition of economics covers too much ground. Furthermore, I am not covering some areas in macroeconomics such as ethical questions and economic history, which should have standards similar to the related disciplines in the humanities.)
How to be RigorousSo long as we are avoiding ethical questions, the main way to be rigorous in macroeconomics is to follow the three steps.
- Pick a mathematical model. (See note below.)
- Determine the properties of that mathematical model.
- Ask ourselves whether observed data have those properties (and under what context)?
Many people have an aversion to mathematical models in economics, and might object to that characterisation. (I would note that very few of my articles have equations, although I do use time series charts to stand in for equations.) I am halfway sympathetic to that view; it may be possible to have a relatively rigorous "verbal model." However, the idea is that a reader should be able to at least translate the argumentation into statements about the tendencies of economic time series, even if the time series are not specified precisely.
Difference From Forecasting
The steps described above may seem straightforward, and the same thing as what people think of what a physical scientist might do. However, there is a key difference; what physical scientists typically do is closer to what I term "forecasting." The steps in forecasting are:
- Pick a mathematical model (with some free parameters values).
- Fit the mathematical model to observed data. (In other words, fix the parameter values in some systematic fashion.)
- Use the fitted model to forecast future observed data points. Some means of statistical testing will exist to determine whether forecasts are accurate enough so that the model is considered "correct."
The difference between the two methodologies is simple. In forecasting, you are assuming that you have the particular model of the real world. In my "rigorous" method, you are just seeing whether real world data has properties similar to the data generated by a family of mathematical models.
As should be obvious, forecasting is a much stronger condition. However, there is no reason to believe that we should be able to make such forecasts.* Instead, we are stuck with weaker explanatory power: we can say that observed data seem to be consistent with certain classes of models, but we cannot say that any particular model is "correct." If we take any particular model from within a class of models, it is entirely likely that it will have properties that are unlike observed data. For example, most models will abstract out certain areas of the economy, and thus will miss dynamics associated with those areas.
In any event, one may note that we are starting from assumptions, then work towards observed data. The belief that we can take random time series and have a macroeconomic model fall out of the sky as a result is a pipe dream. (This presumably can be shown by attempting such a task against reasonable macro models; that is, we can use rigorous methods to demonstrate this.)
Squabbling Economic Tribes
My description of the rigorous methodology should not raise too many eyebrows among trained economists, whether they are of the post-Keynesian or "mainstream" tribe. (I am ignoring some of the smaller tribes, like the Austrians, for reasons of space.) The bitter arguments between post-Keynesians and the mainstream might be considered debates about the application of the rigorous method (modulo the complaints about mathematics in economics, which I think can be dealt with).
My high level description of the source of disputes is that the mainstream wants to assume forecastability, and so they embed far too many assumptions in their models to allow them to generate forecasts. The quality of mainstream economic macro models speaks for themselves.
My feeling is that many non-economists do not grasp that the economic arguments are occurring in this analytical context, and instead are thinking in a mode more similar to the physical sciences.
What Can We Do With Rigorous Economics?
It is abundantly clear that people want forecasting models. When you tell them that is not going to happen, they tune you out, and prefer to listen to someone who appears to have such forecasting power. Wishful thinking is an essential part of human nature.
However, we can drop the wishful thinking, and ask ourselves: even if we cannot predict the future, what can we do? We can certainly do many things.
- We can use reasonable macro theory to make conditional economic forecasts. This style of forecasting is more an art than a science. This is actually what most "market economists" are effectively doing (whether they realise this or not). Unfortunately, the partial models used are incoherent when taken together, and so we cannot decide which is "correct."
- We can design institutions in a sensible fashion. This is discussed in the next section.
- We can determine what not to do. For example, we could use apply the Kalman filter techniques used to determine the natural rate of interest to practically any reasonable macro model, and we would discover that the natural rate of interest "explains" the model outcomes -- even for models where interest rates have no effect on behaviour. The reason -- the estimation method for the natural rate assumes that the natural rate of interest drives behaviour, and hence the output of the Kalman filter has to explain outcomes. Unfortunately, we have just changed the forecasting problem from forecasting the economy to forecasting the movements in the natural rate of interest.
After the crisis of the Great Depression, the political consensus in the developed countries shifted towards the view that the government needs to smooth out the effects of the business cycle. The question then arises: how is this to be done?
The answer from modern mainstream economists is that we create committees of mainstream economists that will make optimal decisions to achieve optimal outcomes. As should be obvious, such optimal control techniques are reliant upon having a good model of the economy. (It should be noted that even control systems engineers abandoned optimal control in the 1960s because it was obvious our models were not perfect, and optimisation techniques are inherently destabilising in the presence of model error.)
Luckily, modern mainstream economics did not exist when the welfare state institutions were created in the aftermath of the Great Depression. We can take any number of reasonable economic models, and show that if the central government runs fiscal policy in a automatic fashion against the business cycle, the effects of the business cycle are dampened. And this is exactly what modern taxation systems and things like unemployment insurance achieve. Even if we cannot predict the future exactly, we can look at the oscillations in real GDP, and see that they became less severe after the implementation of the welfare state.
What Else Can We Do?
We can multiply the areas where we can draw some conclusions. For example, even if the effect of fiscal spending during an expansion is uncertain (the great "what is the fiscal multiplier debate?"), that uncertainty should not be a great surprise. That said, the economic vandalism that was inflicted upon Greece by the Troika was so extreme that it was almost certainly going to be a disaster.
For readers interested in financial applications of economics, all we need to do is find areas where the odds appear stacked in our favour, even if we cannot quantify the odds. Although it would be nice to be able to predict every 25 basis point move in the 10-year Treasury yield, it would also be nice to inherent a billion dollars from a long-lost uncle.
Much of the anguish over economic theory would disappear if we stopped asking economic theory to do what we wish it would do, but instead asked: what can we do with the theory?
* Certain types of "forecasts" can be made. For example, we presumably can make forecasts conditional upon certain economic states (avoiding recession), or that series will follow accounting identities (within measurement error). Such trivial forecasting power is expected, as based upon a rigorous analysis of macroeconomics. Stronger forecasting is not believed to be possible, based on arguments similar to those presented in the previous article. If we were to assume the position of actors in an economic model, strong forecasting powers requires outlandish assumptions about our knowledge and the positions of all the actors in the model.
(c) Brian Romanchuk 2017