A hotel bill of 1,142,351 pesos for four nights. A bottle of water at 2,000 pesos. A president campaigning with a chainsaw. Welcome to Argentina — a country where daily life feels like a stand-up routine performed by macroeconomics.
We often get asked why we omit emerging and frontier countries when we use machine learning to scan for investment opportunities. Isn’t that where promising growth might be hidden? The short answer: because our algorithms mainly work with fundamental company data — income statement, balance sheet and cashflow statement — and not macroeconomic factors or political drama.
To test this first-hand, we flew to Argentina to meet companies and get a feel for the situation on the ground. What we found confirmed why algorithms alone can’t capture demanding markets.
Buenos Dias Buenos Aires – arriving in Argentina.
Inflation as daily comedy
Inflation here isn’t a line item in an IMF report, it’s a way of life. For decades the rule of thumb was simple: spend today, because tomorrow your pesos are worthless. Supermarket anecdotes beat any CPI chart: that tiny bottle of water at 2,000 pesos tells you more than any spreadsheet.
Try putting that into a model: it registers “revenue growth” when in reality it’s just runaway prices.
Soviet in disguise
Argentina’s central bank governor described the economy to us with chilling honesty: “We effectively lived in the most soviet-like system” — but under the shell of a democracy.
For decades, a country with world-class farmland and massive natural resources ran an economy built on subsidies, price controls, and exchange restrictions. When the state dictates the price of electricity, bread and bus tickets, productivity doesn’t stand a chance. No balance sheet ratio captures that.
Zombie bankers and golf clubs
The IMF representative was even blunter: “Argentina has a textbook zombie banking system.”
For years, banks barely lent to businesses. Why bother? They could just park pesos at the central bank, collect risk-free interest, and head to the golf course by 4pm. The result: private credit at 9% of GDP. That’s not a banking system — that’s a country-club savings plan.
Now, if Milei’s reforms stick, this same system could be Latin America’s biggest turnaround story. But again — nothing in the historical fundamental company data can tell you that.
The Chainsaw Man
With Javier Milei as president came the radical break. Milei campaigned with a chainsaw in hand — a symbol of the cuts to come: ministries shut down, tens of thousands of public employees dismissed, subsidies slashed. His mantra: equilibrio fiscal — a balanced budget. Early results are striking: inflation has fallen from nearly 300% to around 30% [see red line],
and the chronic deficit has vanished.
But without political alliances, his program risks collapse. So far, Milei has been better at launching Twitter tirades than on forging coalitions and allies. Ahead of the midterm elections, everything hangs by a thread. Each headline sends markets swinging wildly — volatility remains the only certainty.
Political volatility in real time
If you want to model Argentina, forget quarterly data. You need a live ticker.
- September: Milei’s coalition loses an election in Buenos Aires province → markets tank.
- Last week: export taxes on grains scrapped by decree to attract short-term dollars → bonds rally, but fiscal revenues implode.
- Tomorrow? Who knows. Every announcement sends stock prices lurching like tango dancers after too much Malbec.
These are external factors that companies can’t influence. Take the tax cut, for example: company profits might surge overnight, but not because the business suddenly became more efficient through a technological breakthrough or similar improvement. It’s simply an external policy change — something beyond management’s control. Your model might mistakenly interpret it as a sign of intrinsic profitability.
However, the underlying data distribution has changed. And machine learning hates these changes — these regime shifts, also known as concept drift. Regime shifts, often driven by structural changes in the real world, are a challenging beast for data scientists. Argentina has several before breakfast.
Three sources of “concept drift” as the problem in which P_t(X; y) = P_t(X) x P_t(y|X) is not equal to P_t+1(X; y). See e.g. Jie Lu at al. “Learning under Concept Drift: A Review”, arXiv:2004.05785 (2020).
Why macro trumps micro
Argentina has everything an investor should love: lithium for EVs, gas and oil in Vaca Muerta, farmland that produces two harvests a year. But fundamentals don’t live in a vacuum. They live in a political system where ministers are fired on TV, subsidies vanish overnight, and the peso is held together with duct tape.
An ML algorithm scanning balance sheets might flag “cheap P/E ratios.” It won’t tell you if capital controls prevent you from repatriating profits. It won’t warn you that next week’s presidential tweet could halve your valuation.
What we learned on the ground
On our trip, companies sounded surprisingly upbeat. Airport operators could finally raise passenger fees. Energy utility Edenor told us bill collection rates remain astonishingly high — even though subsidies are gone. Oil and gas firms are pouring billions into pipelines to ship hydrocarbons from Vaca Muerta to the coast.
And on the street? Taxi drivers, waiters, hotel staff all said the same: “We support the change.” People are actually saving, comparing prices, walking one block further for a deal. After decades of financial nihilism, that’s a cultural revolution.
Try teaching that to an algorithm.
The punchline
At the end of our week, that million-peso hotel bill landed on the desk. A laughable relic of Argentina’s inflationary past — and a perfect reminder why you don’t let machine learning decide your emerging market allocation.
The takeaway:
-
ML struggles with instability, and regime change.
-
Emerging markets need boots on the ground, context, and narrative, not just code.
-
Argentina is a high-risk, high-volatility Wundertüte.
So yes, Argentina might one day be a great investment. But don’t expect your algorithm to tell you when. Until then, bring your own judgment — and maybe a calculator with extra zeros.
