The Armington Assumption

Why does a country like the United States, a massive producer of automobiles, also import millions of cars from other nations? This simple question reveals a puzzle that challenged early theories of international trade, which often treated goods like "cars" as uniform commodities. In such a world, a country should either export or import a good, but not do both. The reality of two-way trade suggests a fundamental gap in this understanding: we perceive products from different countries as distinct.

This article explores the solution to that puzzle: the Armington assumption, pioneered by economist Paul Armington. This powerful idea posits that goods are differentiated by their country of origin, making them imperfect substitutes. By embracing this concept, economic models gain the ability to realistically simulate complex global trade flows. We will delve into the core principles behind this assumption, examining the mathematical tools like the Constant Elasticity of Substitution (CES) function that bring it to life. Subsequently, we will explore its powerful applications, seeing how this virtual laboratory helps us analyze the real-world effects of trade wars, major infrastructure projects, and global supply chain disruptions.

Let's begin with a simple observation that seems, at first, like a paradox. The United States is one of the world's largest producers of automobiles. It has immense factories, a skilled workforce, and a long history of car manufacturing. And yet, every year, it imports millions of cars from Germany, Japan, and Korea. Why? If a car is just a car, why doesn't a country just produce all the cars it needs?

The commonsense answer, of course, is that a BMW is not a Ford, which is not a Hyundai. We perceive goods as being different based on where they come from. They may serve the same basic function, but they possess different qualities, brand reputations, and designs that make them distinct in our minds. For a long time, however, economic models of international trade struggled with this simple truth. Many early models treated goods like "cars" or "wheat" as uniform, homogeneous commodities. In such a world, a country would either be the most efficient producer and export a good, or it would be an importer of that good—it wouldn't do both.

This is where a clever and profoundly useful idea entered the scene, thanks to economist Paul Armington in the 1960s. He proposed what we now call the ​​Armington assumption​​: that goods are differentiated by their country of origin. This isn't just a minor tweak; it's a fundamental shift in perspective. It means we don't just consume "cars"; we consume a composite good, a blend of domestically-made cars and imported cars. These two varieties are not identical. They are ​​imperfect substitutes​​. This single assumption unlocks our ability to build models that realistically capture the two-way trade we see everywhere and, more importantly, to analyze how shocks like tariffs ripple through an economy.

So, how do we turn this intuitive idea into a working machine for economic analysis? We need a mathematical language to describe the "art of substitution." Imagine you are a consumer, and your total "transportation enjoyment" is a blend of driving domestic cars and imported cars. How do you decide on the mix?

Economists often use a wonderfully versatile tool for this called the ​​Constant Elasticity of Substitution (CES)​​ function. Think of it as a flexible recipe. You're trying to bake a "consumption pie," and the recipe tells you how you can mix domestic ingredients and imported ingredients. The most important parameter in this recipe, the secret spice, is the ​​elasticity of substitution​​, usually written with the Greek letter sigma ($\sigma$).

This number, $\sigma$, is the heart of the matter. It tells us how easily we can substitute one good for the other in response to a price change.

• If $\sigma$ were infinitely large, it would mean domestic and imported cars are perfect substitutes—like two brands of bottled water that are chemically identical. If the price of one goes up by even a tiny amount, you'd switch completely to the other.

• If $\sigma$ were zero, it would mean the goods are perfect complements. You need them in a fixed ratio, like left shoes and right shoes. No matter how cheap left shoes get, you won't buy more of them unless you're also buying more right shoes.

• The real world, for most goods, lies somewhere in between. For cars, perhaps $\sigma$ is around 3 or 4. They are substitutes, but not perfect ones. If the price of imported cars goes up, you'll buy fewer of them and more domestic ones, but you won't abandon them entirely. You, or at least some consumers, still value the unique qualities of the imported models. A low value of $\sigma$, say $0.5$, would imply you see them as very different and are reluctant to switch, even when prices change.

This CES framework, powered by the elasticity of substitution, gives us a precise way to model the choices consumers and firms make when faced with a smorgasbord of domestic and foreign products.

Now let's put our new machine to work. Consider a small country that decides to impose a 10% tax, or ​​tariff​​, on all imported goods. This makes imported cars, TVs, and clothing more expensive for its citizens. What happens to the quantity of goods it imports?

Your first instinct is likely correct: imports will fall. The higher price will push consumers to substitute away from the now-more-expensive foreign goods and toward their domestic alternatives. This is the ​​substitution effect​​.

But wait, there's a complication. The government is now collecting a large amount of money from this new tariff. What does it do with this revenue? In many economic models, to isolate the effect of the price change, we assume this money is returned directly to the citizens as a lump-sum rebate. Suddenly, everyone has more money in their pocket! With this extra income, they will likely want to buy more of everything—including, perhaps, some of those imported goods. This is the ​​income effect​​.

So we have a battle of two forces: a substitution effect driving imports down, and an income effect potentially pushing them back up. Which one wins, and by how much? This is a question that requires a ​​general equilibrium​​ perspective, where we solve for all these interacting effects at once. It sounds terribly complicated.

Yet, through the beauty of mathematics, the answer that emerges from the model is one of stunning simplicity. If we start from a baseline of zero tariffs and introduce a small tariff change, $\Delta \tau$, the resulting proportional change in the volume of imports ($\frac{\Delta M}{M_0}$) is given by an elegant little formula:

$\frac{\Delta M}{M_0} \approx -a \cdot \sigma \cdot \Delta \tau$

Let's take this beautiful result apart. It says the fall in imports is determined by just three things:

1. $\Delta \tau$: The size of the tariff itself. A 20% tariff will have twice the impact of a 10% tariff. This is completely intuitive.

2. $\sigma$: The elasticity of substitution. If goods are very easy to substitute (a high $\sigma$), then even a small tariff will cause a large drop in imports. If they are poor substitutes (a low $\sigma$), the drop will be much smaller. This, too, makes perfect sense.

3. $a$: The share of spending that was initially devoted to the ​​domestic​​ good. This is the most subtle and surprising part of the result! Why does the domestic share, not the import share, appear here? It reflects the balance between the substitution and income effects. When the domestic share $a$ is large, it means imports are a small part of the economy. The income effect from the rebated tariff revenue (which is levied only on imports) is therefore small compared to the overall economy. The price signal from the tariff—the substitution effect—dominates. Conversely, if imports were a huge part of the economy (low $a$), the tariff revenue would be enormous, creating a large income effect that would partially offset the substitution effect.

Let's make this concrete with an example straight from a model simulation. Suppose for a particular good, consumers are quite flexible, with an elasticity of substitution $\sigma = 3.0$. And let's say that in this country, consumers initially spend 60% of their money on the domestic version ($a = 0.6$) and 40% on the imported one. If the government imposes a 10% tariff ($\Delta \tau = 0.10$), the predicted change in imports will be:

$\frac{\Delta M}{M_0} = - (0.6) \cdot (3.0) \cdot (0.10) = -0.18$

The model predicts that imports will fall by 18%. A different economy, perhaps one that is more reliant on unique foreign goods (low $\sigma$) or one that is already highly open to imports (low $a$), would see a much different result. This simple formula, born from a complex system, gives us a powerful intuition for how trade policy works.

We have a powerful intellectual machine. But to use it to offer advice on real-world policy, we can't just admire its design; we have to feed it real numbers. Where do we get the values for parameters like the elasticity $\sigma$ and the expenditure share $a$? This question takes us to the frontier of economic modeling and introduces a critical lesson about scientific certainty. There are two main philosophies for building these models.

The first approach is ​​calibration​​. This is like taking a high-resolution photograph of an economy in a single year—say, using the 2023 national accounts data (the Social Accounting Matrix, or SAM). The modeler then "tunes" the parameters, like the share parameter $a$, to ensure the model perfectly replicates the economy of that snapshot year. Elasticities like $\sigma$ are often borrowed from other specialized statistical studies. This method is pragmatic, relatively fast, and results in a single, deterministic prediction. If you run the tariff simulation, you get one answer: "imports will fall by 18%."

The second approach is ​​econometric estimation​​. This is a more statistical philosophy. Instead of a single snapshot, the modeler uses a long movie of the economy—perhaps 30 or 40 years of data. They then use statistical techniques (like regression) to find the parameter values that provide the "best fit" over that entire history. This method has two profound consequences. First, the estimated model likely won't perfectly match the data for any single year, including our 2023 "snapshot," because it's trying to be true to the long-run average behavior. This means the "baseline" from which the policy experiment starts is different from that in a calibrated model.

More importantly, the statistical process doesn't just give us a single number for $\sigma$; it gives us a best guess and a measure of its uncertainty, often expressed as a ​​confidence interval​​. It acknowledges that we can never know the true value with perfect precision. When this uncertainty is propagated through the model, the output is not a single number, but a range of possible outcomes. The report no longer says "imports will fall by 18%." It says, "we are 95% confident that imports will fall by somewhere between 15% and 21%."

This is not a failure of the model; it is a mark of its scientific integrity. It reminds us that our models are not crystal balls. They are powerful lenses for understanding the world, built on clever ideas like the Armington assumption. They reveal the hidden mechanisms that link policy to outcomes, but their predictions carry the honest signature of uncertainty that is inherent in the study of any complex system.

Now that we have disassembled the clockwork of a Computable General Equilibrium (CGE) model and inspected its gears and springs—the households maximizing their well-being, the firms making their products, the markets clearing, and the clever Armington assumption that gives each part its unique identity—it's time to put it all back together. We have built a magnificent, intricate model of a national or global economy. What is it good for?

The answer is: it’s a laboratory for the imagination. It’s a flight simulator for an economy. We can’t actually start a trade war, build a continent-spanning railroad, or block a major shipping canal just to see what happens. But in the world of a CGE model, we can. By changing a parameter, we can nudge the machine, and then watch, with the precision of a physicist, as the effects ripple through every single gear. We can ask "what if?" and get a rigorous, quantitative answer. This is not just an academic exercise; it is a vital tool for navigating the complexities of our interconnected world, linking economics with policy, engineering, and geopolitics.

One of the most classic and contentious questions in economics is about the benefit of free trade. For centuries, nations have been tempted to erect barriers, usually in the form of tariffs—taxes on imported goods—to protect their domestic industries. What does our economic simulator have to say about this?

Imagine a simplified world of two countries, say "Country A" and "Country B," each producing their own signature product. Thanks to the Armington assumption, we know consumers in both countries have a taste for both the domestic and the foreign variety. Now, suppose Country A decides to impose a tariff on goods from Country B.

The first, most obvious effect is that the price of Country B's product rises for consumers in Country A. They will likely buy less of it. So far, so simple. But the story has only just begun. The tariff generates revenue for Country A's government, which, in our models, is often returned to the people. This extra income changes their spending patterns. Furthermore, as consumers in A shift away from the now-expensive foreign good, the demand for their own domestic product might rise. This could drive up its price.

But the most crucial part of the story is the "general equilibrium" effect, which spreads through the global system. The reduced demand from Country A for Country B's product means the producers in B now have a surplus. To sell it, they might have to lower their producer price on the world stage. And what about retaliation? Country B, seeing its exports taxed, may decide to impose a tariff of its own on Country A's goods. This sends a whole new series of price shocks rippling back.

Untangling this web of action and reaction by intuition alone is nearly impossible. But a CGE model can do it. It simultaneously solves for the new set of prices and trade flows that brings the entire world economy back into balance. When economists run these simulations, a powerful and consistent lesson emerges: while one country might briefly hope to gain an advantage (what is known as an “optimal tariff”), the likely outcome of a trade war is that both sides lose. The barriers gum up the works of the global economic machine, making it less efficient. The result is a net loss in overall welfare—a lower real income—for the citizens of both countries. The model transforms a heated political debate into a clear, albeit simplified, quantitative result.

Let's turn from erecting barriers to tearing them down. Not tearing down tariffs, but overcoming the natural barriers of geography. How much is a new bridge, a high-speed rail line, or a fiber-optic cable worth to society? This is a question not just for economists, but for engineers, urban planners, and governments.

CGE models offer an elegant way to think about this using the concept of "iceberg costs." Imagine that for every 100 widgets shipped from one region to another, some fraction "melts away" in transit. This "melt" represents all real transportation costs—fuel, time, and logistics. A new high-speed rail line is, in essence, a technology that reduces a lot of that melting.

In our model, this is simply a reduction in the transport cost parameter, $\tau$. Let's say we model two regions connected by a slow, old railway. We solve for the equilibrium and record everyone's welfare. Then, we run the simulation again, but this time with a smaller $\tau$ to represent a new, efficient bullet train. What happens?

Again, the chain of events is complex and beautiful. The direct cost of shipping goods between the regions falls. This makes imported goods cheaper in both places, a direct benefit to consumers. But it also means producers in each region now face stiffer competition from their cross-regional rivals. At the same time, they also have better access to the export market. The relative producer prices of the goods will adjust. Perhaps the region with a productivity advantage will see its economy boom, while the other must adapt.

The CGE model tots up all these competing effects. It tracks the change in consumer prices (the cost of living), the change in producer incomes, and the change in trade volumes. The end result is a clear number: the percentage change in real income for each region. These models can show that the benefits of infrastructure are often far greater than the simple sum of saved shipping costs. By making markets more integrated, they can unlock new efficiencies and specialization, leading to a significant increase in overall economic well-being. This provides a powerful, data-driven argument for ambitious public and private investments.

For decades, globalization has woven the world into an astonishingly complex tapestry of supply chains. A smartphone might be assembled in one country from components made in a dozen others. This system is incredibly efficient, but as recent events have shown, it can also be a fragile. What happens when a single, vital thread is snipped?

Think of the 2021 incident where a single container ship, the "Ever Given," got stuck in the Suez Canal, blocking one of the world's most critical trade arteries. The effects were felt globally, from delayed cars to scarce garden gnomes. A CGE model is the perfect tool to understand these cascading failures.

In a more sophisticated model, we can have multiple regions and, crucially, multiple types of goods (say, "manufactured goods" and "raw materials"). Now, we simulate the port blockage. This isn't a general increase in transport costs; it's a massive, sudden spike in the iceberg cost $\tau$ for a particular good, or for all goods passing along a specific route.

The model immediately shows the consequences. The delivered price of the blocked goods skyrockets in the importing regions. Firms and consumers who rely on them are faced with a choice. Can they substitute toward a domestic alternative? The Armington assumption allows for this, but not perfectly—a German car is not a perfect substitute for a Japanese one. Can they substitute towards a different type of good? Perhaps, but this is also limited.

The shock doesn't stop there. Producers of the blocked good see their export orders vanish, and their income plummets. Meanwhile, producers of substitute goods, both domestic and in other, unaffected countries, experience a sudden surge in demand. The CGE model simulates this frantic, economy-wide scramble. It traces how a highly localized physical disruption—one stuck ship—can propagate through the price system, causing widespread economic disruption and a net loss of global welfare. It highlights the hidden trade-off of our hyper-efficient world: the efficiency of "just-in-time" systems comes at the cost of the resilience of "just-in-case" inventories.

From the political theater of a trade war, to the concrete and steel of a new railway, to the geopolitical tremor of a blocked canal, CGE models provide a unified framework for analysis. They are not a crystal ball; the real world is infinitely more complex than any model. Their predictions are only as good as the data and assumptions we feed them.

But they are an unparalleled tool for thought. They are a way to make our assumptions explicit and to follow their logical consequences with unflinching rigor. They force us to see the economy not as a collection of separate issues, but as a deeply interconnected system where every action, no matter how small or distant, can have ripples that are felt by everyone. By allowing us to experiment in this virtual world, CGE models help us to better understand, and perhaps better navigate, the beautiful and fragile economic world we all share.