The Economic Model of Argentina in 2019 was...

But what's an economic model, and how do you find them?

Argentinean economic history can look anything but simple, even ignoring the country's world-leading inclination for inflation and defaults. Looking at "just" a dozen variables from the World Bank for the 1993-2018 period, the only stable feature is its lack of stability.

Yet what seems complex seen as bunch of plots can be much simpler if one looks at it as a time series in a 12-dimensional space of possible states instead of twelve distinct time series. "One," of course, meaning a computer, yet even a standard dimension-reduction algorithm like PCA can suggest there are patterns in the data that reflect a history with both continuities and crisis. Here's what one possible projection of this 12-dimensional space into a bidimensional approximation looks like:

It's pleasant to track the trajectory of time year by year if you have some familiarity with Argentinean history; there's an intuitive sense of similar years being close and jumps reflecting crisis points, but it's more suggestive than informative.

What we want, first, is an explicit sense of how we can group the state of the Argentinean economy in different years as more or less similar. Applying a traditional hierarchical clustering algorithm, we get

This graph (called a dendrogram) shows a way of putting together years: 1993 and 1994 had similar statistics, as did 2010 and 2011, but each pair is far from similar to the other. And, something that wasn't obvious in the PCA plot, 1999 stays on its own, as an anomalous year.

How can we understand this, and, more in general, how can we simplify this into a conceptual picture that we can use as a lens to look at the whole of the data? After all, a plot that's barely less complex than the data might be a good way to organize it, but on its own doesn't increase our understanding of the underlying reality.

We do the simplest possible thing: we squash together the more nuanced differences between years, considering 2013 the same as 1993 and 2002 the same as 2012, to keep only the three highest-level groups. (Why smash into three groups instead of seven or two? The choice is always based on a combination of what's cognitively manageable, what the purpose of the analysis is, and how little or how much of the complexity of reality we deem acceptable to leave aside in this case.)

Every year belongs then to a cluster of years, an "economic model." Thus far it's all abstract and opaque: why, in intuitive terms, can we justify to ourselves that 1993 belongs to one group and 2008 to another, and why is 1999 alone by itself? We apply yet another standard tool, a partition tree, to the model we just build, essentially asking for a set of rules that gives us not a list of what years belong to what models, but how to figure out on our own the economic model that describes a year.

The answer is either fascinatingly simply or underwhelmingly so, depending on your aesthetic preferences:

The rules turn out to be simple: - If foreign direct investment is above 6.1% of GDP, then it's a "Model B" year (so far, only 1999 meets the criteria). - If the external trade balance is below +1.5% of GDP, then it's a "Model A" year. - Otherwise, it's a "Model C" year.

These are very straightforward rules. Does the economic model they describe tell us anything useful about the aggregate performance of the Argentinean economy? We showed it on the first graph, but now we can look at it again with better understanding. The time series is Argentina's GDP, in PPP (Purchasing power parity) 2017 constant USD per capita, but normalized by the world's average. In this way - We normalize away some of the issues with inflation, exchange rates, etc. - We normalize away population growth. - We normalize away some of the issues with exogenous worldwide influences (like technological changes).

Over this time series we plot the economic model for each year - 2020 still lacking official numbers:

Leaving aside the economic model B (high FDI) as only applying for a single year, the turbulent recent economic history of Argentina can be summarized as an era of low net exports and stable or decreasing normalized real GDP, interrupted during its middle third by a phase of high net exports and stable or increasing normalized real GDP. Or, to put it in a more detailed way, an economy that fell during the decade between 1993 and 2002, recovered most of its losses by 2008, stalled for a few years, and has been falling since then.

To be clear, this doesn't mean that other economic variables, like the fiscal balance or domestic savings, are causally irrelevant. The point is rather that economic variables are tightly interrelated, and the system as a whole doesn't have as many ways of behaving as one could imagine looking at the variables separately. The Argentinean economy, the data suggests, is essentially, or rather statistically, driven by its trade balance, and many other variables, including its real GDP (as insulated as possible from unrelated factors), are sooner or later are conditioned by it.

What's next? Looking at 2019, guessing about 2020, and asking questions about 2021

What about the most recent years? The data for 2019 puts it in Model C territory, which is almost always associated with mostly stable or increased economic performance, but the pandemic hit the country during early 2020 with a huge economic impact - although not one necessarily bad for the external trade balance as such. Might 2020 be a (justifiable) anomaly in recent economic history? What about 2021? Have we entered a fourth, different economic model, or are we still switching between the two familiar clusters?

Those aren't rhetorical questions; I don't know. But I'd very much like to, and the data will tell.