Science And Economics

I had largely managed to avoid writing about the latest angst in the economics blogosphere regarding mathematics, science, and economics. I am not a fan of mainstream economics, but at the same time, I question some of the broad brush attacks on economics. The quest to pretend that economics can be a science like physics is doomed, and does not take into account the nature of what is being studied.

Empirics? 

This article by Chris Dillow caught my eye, as it gave an interesting background explanation of how mainstream economics has obfuscated power relationships out of economics, and hence has some difficulties explaining inflation. (The snappy explanation is that the inflation process is the result of a power struggle between capital and labour, and that sounds too Marxist and hence was airbrushed out of economic theory.)

However, the Dillow article  had a tangential link to "Why mathematics has not been effective in economics" by Tim Johnson, which is another interesting read. In turn, it referenced "The new astrology" by Alan J. Levinovitz. The tag line of that article read:

By fetishising mathematical models, economists turned economics into a highly paid pseudoscience.

The Levonovitz article caused a stir when it was first published. In fact, I have written similar things about economists, arguing that they are filling the social role of shamans in our societies. However, I think that just tells us something about human societies: there is a demand for shamans, and so the economist shaman supply grew to fit that demand. (The post-Keynesian "Reverse Say's Law.") The tendency towards shamanism is ingrained, hence, we cannot really blame it on the contents of mainstream economic theory.

What pushed me into writing this article was this question on the Economics Stack Exchange on empirical research in economics.  It quotes an econophysicist Bertrand Roehner, who stated:

Economic journals, on the contrary, are very reluctant to publish empirical observations, especially when they have no clear interpretation within the existing theoretical framework.

The questioner then asked: "Are there economic journals which are keen to publish puzzling empirical observations, i.e. those observations which have no clear interpretation within the existing theoretical framework?"

Since the answers on the Economics Stack Exchange are supposed to be somewhat authoritative and not entirely rants, I decided to give my rant here.

Problem is Falsification, Not Empirical Puzzles

In my view, the premise that economics journals do not publish articles on empirical puzzles because of "reluctance" misunderstands the reality of mainstream economic research.

Mainstream economic theory cannot give a straight answer on whether raising the policy interest rate raises or lowers inflation, and inflation targeting by central banks has been a major research topic for decades. If the theory cannot give you the sign of the most-studied macro effect, what makes you think that any empirical data can be found that cannot be "interpreted" by mainstream economic theory?

If you insist on having a more formal argument, I would suggest generating roughly 20 years of quarterly data using a somewhat realistic stock-flow consistent (SFC) model (which is roughly all we have to work with in the real world for most countries due to structural changes), including some random noise. Then apply the quantitative techniques used by mainstream researcher to that data set, and see whether the mainstream DSGE models can "explain" the data. I have not done that experiment, but I see little doubt that the data could be "explained." (Note for any post-graduate students: that would be a fun thesis topic!)

If my guess is correct, we see that a DSGE model can "explain" a SFC model -- even though the true SFC model is based on assumptions that are antithetical to DSGE model assumptions. If we were able to test the validity of DSGE modelling, such an experiment would reject DSGE models' ability to fit the data. If DSGE modelling techniques are not "falsified" by data generated by model-generated data, why would we expect that real world data to falsify the DSGE modelling framework?

In other words, most empirical investigations of the validity of mainstream theory can say very little, and even academic journals do not want to waste column inches on them.

Can We Be Scientific?

Having done my obligatory ranting about mainstream economics, I now return to my argument which defends economists (both mainstream and post-Keynesian) from the charge that "they are not scientific."

Since some philosophers sometimes read my writings, I am going to avoid any attempt to give a highbrow definition of "being scientific," I will stick to a common sense definition as "making useful predictions."*

There is another component to being "scientific" would be to find causal mechanisms to explain behaviour. In physics, the analogy would be to come up with an explanation why gravity exists, as opposed to just observing the inverse square law. In the domain of economics, we are discussing human behaviour, and the "causal explanation" comes down to: why do people do what they do? This is normally viewed as a separate question, and studied in the other social sciences. It seems quite unlikely that we will be able to take advances in psychology (for example) to help forecast the inflation rate over the next few quarters. Correspondingly, I ignore such issues in my discussion of "being scientific."

We Can Use Economics Make Some Predictions. So What?

I have little experience (or interest in) many areas of economics, such as microeconomics. My comments here are aimed only at the questions of predictions in macroeconomics.

The first thing to realise is that there are embedded relationships between various economic time series. If we pick series at random (without understanding those relationships), we will eventually be able to find series that we can fit mathematical models to. However, most economists have an understanding of those underlying relationships, and so are unimpressed by such empirical efforts.

One example would be the relationship between currency outstanding (notes and coins) and nominal GDP in Canada from 1964-2015. As can be seen, the ratio of currency outstanding to nominal GDP was stable. Although this may not appear to be the most useful empirical relationship, we can find many other series whose relationship to GDP follows smooth trends that are amenable to modelling.

The empirical observation that currency outstanding runs around 3.0-3.5% of Canadian nominal GDP is somewhat interesting, but it begs a lot of questions. Why is this happening?

• The sensible viewpoint is that private sector entities withdraw currency from the banking system based on their nominal incomes. Since nominal GDP equals nominal Gross Domestic Income (by definition), this implies that currency holdings will be stable as a percentage of nominal GDP.
• The less sensible viewpoint is to argue that currency holdings determine nominal GDP, and that the central bank can set the level of nominal GDP by (somehow) changing currency holdings.

The upshot is that the empirical relationship exists by itself is meaningless; we need an overall theory to fit the observations into. We also need to take into account that theory needs to be useful for policy purposes: what happens when changes are made to policy parameters? Furthermore, we need to understand the data series themselves in order to judge that an empirical observation is not useless. (For example, it is possible to rediscover the relationships implied by accounting identities.)

I could multiply examples of such partial theories. For example, modern fixed income pricing theory is a highly sophisticated framework for pricing all sorts of fixed income securities (many of which probably should not exist). The theory itself is rock solid. However, the empirical behaviour revolves around the correlation of forward rate movements. Meanwhile, the fair value of the forward rates is what market participants believe what policy rate will be set by the committee of very human central bankers in the future. No matter how deep the mathematics gets, the whims of central bankers remains as the ghost in the machine.

Form of Models Informed by Experience

An additional thing to keep in mind is that the way economists build models reflects experience with the subject matter. For example, we know that yesterday's 10-year Treasury yield is a pretty good estimate for today's 10-year Treasury yield. If we feed forward yesterday's bond yield into today's model estimate, our model will often look much better than the sort of bond valuation models that are typically produced by strategists. (Link to the "World's Simplest Bond Valuation Model.") To the inexperienced eye, it seems reasonable to use a model that has a much tighter fit to observed bond yields (by incorporating yesterday's yield).

The reason why practitioners are happy with valuation models with relatively large pricing prediction errors is that you make money in the bond market by predicting large changes in the bond yield, not predicting the current level (which you can just look up on Bloomberg). Instead, you want a model that indicates the direction of the next big move in yields, and feeding in yesterday's bond yield generally does not help in that quest.

In this example, the real test of the model is: can we make money with it? (I would not put forth "The World's Simplest Bond Model" as being the potential secret sauce for a hedge fund, but someone could develop a better model that looks similar.) We know how to determine the bond yield today exactly: we look up a quote. In order to make money, we need a model that has errors that are pointed in the right direction (most of the time). That is not a standard modelling criterion in the physical sciences or engineering.

Can Comprehensive Economic Models Make Predictions?

In macro, all roads lead to a comprehensive economic model. The question then arises: can such models make useful unconditional predictions?**

(When I refer to a comprehensive macro model, it is a model that attempts to simulate all the major aggregates of the national accounts. Other economic time series should be inferred from the simulated series. As a result, the empirical relationships between arbitrary time series end up being the result of the dynamics of the comprehensive model. We may have some pretty ugly simplifications, such as throwing out foreign interactions and just using a closed economy model.)

Unfortunately, this is probably not the case. One of the key drivers of the economic cycle is fixed investment. If we drop the assumption of mainstream economics that everyone is the same, the determination of the level of fixed investment in a capitalist economy is extremely difficult. Firms invest if they believe that the investment will be ratified by future profits. Meanwhile, fixed investment is a major driver of aggregate profits (as per the Kalecki Profit Equation). Firms invest if they believe that other firms will also invest.

(Someone who is more familiar with Keynes' General Theory could tell you what chapter(s) that argument came from; I would need to re-read the book to give a more specific reference.)

You then need to ask yourself: do you know what the level of fixed investment (and profitability) will be in the future? If so, all firms would have access to this forecasting mechanism, and so perhaps all activity could be coordinated. Unfortunately, there is no evidence that such a forecasting mechanism exists, and so actual investment is going to be the result of multiple decisions by multiple decision-makers, based on internally contradictory logic. Needless to say, such a situation is not going to be amenable to a clean mathematical model. Without such a clean model, quantitative predictions are hard to come by.

Can we go the other way: observe a whole bunch of smaller empirical regularities, and then attempt to back out the comprehensive model? The argument is that this is not feasible, as economic outcomes are the result of economic actors attempting to forecast aggregate behaviour. We are not working with elementary particles that mechanically react to local forces.

Although I have severe reservations about mainstream economics, the theory does reflect the reality that outcomes are the results of humans forecasting macro outcomes. This means that they are correctly dismissive of arguments made by non-economists who argue that we need to take random empirical regularities and attempt to build a coherent macro theory out of them.

Peak Oil: The Horrible Example

If you read enough "economics is flawed" articles, you will run across the sentiments that economics needs a multi-disciplinary approach, which needs to take into account things like limitations of physical resources. Allegedly, the discipline of physical scientists and their appreciation of conservation laws would put analysis on the right track. This ignores the skeleton in the closet: the Peak Oil movement.

As a disclaimer, I would consider myself to have been a member of the "Peak Everything" crowd, but probably should refer to myself as a "extreme resource limitations worrier." There was a lot of excellent Peak Oil analysis available, but the good analysis was tainted by the association with proponents of "oil economic determinism." The logic went roughly like this:

1. Economic growth and oil production growth over the past century rose together. Oil is important for all economic processes (due to the high energy return on investment), and hence, its use caused the economic growth.
2. Conventional oil production will peak (and I believe that this indeed happened; the debates arise over what to include in "conventional oil.").
3. Therefore, when oil production declines, the size of the economy will decline (negative GDP growth).

Once again, that is my summary of arguments made by the more exuberant members of the online Peak Oil community, and not the measured arguments of the Peak Oil analysts. I largely ignored the oil economic determinism arguments when they were popular, so my explanation may be missing some details. However, it should be noted that these arguments were the result of a multi-disciplinary collaboration by applied and pure physical scientists, and were using analysis techniques that would be considered best practices in their disciplines. (My training was as an engineer, and I did not see technical problems with the arguments, rather the economic theory was faulty.)

When the peak in oil production arose, and there was no ongoing economic apocalypse (only the relatively brief Financial Crisis), enthusiasm for "Peak Oil" waned, and various websites declared victory and shut down.

Meanwhile, mainstream and post-Keynesian models essentially predicted what happened: there was no reason for a sharp change in behaviour solely because of a change in the sign of the derivative of oil production. Eventually, resource constraints will matter, but standard economic models could easily be modified to take those into account. In summary, theories that are based on the assumption that human reactions to expected macro outcomes were much more successful than theories which suggested that humans react mechanically to particular economic variables.

Concluding Remarks

The idea that economics can be like physics is based on an extremely shaky assumption: that we should expect be able to make unconditional economic forecasts. Once we understand reasonable economic theory (such as post-Keynesian theory), we see that such strong predictability in macro is not going to happen. The defect of mainstream economics is that it embeds too many assumptions in how actors react to macro uncertainty, not that it is unscientific. That is, we need to make some assumptions, we just have to avoid making too many (or obviously incorrect ones).

Footnotes:

* Which may or may not be considered "philosophical pragmatism", which may or may not be considered highbrow.

** We can make conditional predictions: "we believe that X will happen if we avoid recession." Such a strategy appears more feasible, but it begs a lot of questions (what are the odds of a recession?).

(c) Brian Romanchuk 2017