Algorithmic Intermediation and the International Transmission of U.S. Monetary Policy

The Risk of Algorithmic Monoculture in Global Finance

Why do some emerging markets experience sudden, violent capital flight when the U.S. Federal Reserve changes interest rates, while others remain relatively stable? As investment funds increasingly rely on AI-driven models to manage portfolios, the mechanics of these international money flows are shifting. Many fear that the rise of automated trading inherently increases market fragility, but a new study suggests the danger is more nuanced.

The researchers argue that the threat is not the presence of algorithms themselves, but their similarity. When many funds use nearly identical AI models, they tend to make the same mistakes at the same time. This paper shows that while diverse algorithmic models can actually stabilize markets during normal periods, they become a powerful engine for instability during financial crises. This leads to massive, synchronized money outflows from emerging markets.

The Danger of Shared Errors

The core concept introduced by the authors is "algorithmic homogeneity"—a measure of how much forecast error is shared across different algorithmic funds. Think of it like a group of navigators on a ship. If every navigator uses a different map and a different compass, their individual errors will likely cancel each other out. However, if every navigator uses the exact same digital GPS software, a single software bug will cause the entire crew to steer the ship into the rocks simultaneously.

In the context of global finance, the authors define $\phi$ (phi) as the fraction of forecast-error variance that is common across algorithmic funds. When $\phi$ is low, the sector is diverse and stabilizing. When $\phi$ is high, the sector is homogeneous and prone to "herding." In this state, trades reinforce one another and intensify the movement of capital across borders.

The Mechanics of Global Spillovers

To analyze this, the authors developed a two-region macro-financial framework—a Dynamic Stochastic General Equilibrium (DSGE) model. This model simulates a "center" (the United States) that sets monetary policy and a "periphery" (an emerging market) that receives cross-border lending.

The model distinguishes between traditional funds, which are "sluggish" because they face costs when adjusting their loan books, and algorithmic funds, which react rapidly to new signals. The authors demonstrate that the impact of a U.S. monetary shock on the periphery depends entirely on the level of homogeneity.

As shown in, the "amplification factor"—the ratio of how much a shock is magnified under high homogeneity versus low homogeneity—scales with the square root of the ratio of those two states.

For a benchmark comparison where high homogeneity is 0.8 and low homogeneity is 0.2, the factor is 2.0. This means the volatility of the response doubles. Crucially, the authors find that homogeneity doesn't necessarily change the average amount of money moving. Instead, it changes the variance (the spread or unpredictability of the response).

The most striking finding, however, is that this risk is "state-dependent." The authors derive a threshold, $\phi^$ (phi star), which represents the level of homogeneity at which the market shifts from being stabilized by algorithms to being amplified by them. According to Proposition 3, this threshold is not constant. As local economic fundamentals deteriorate—meaning the economy moves further from its steady state—the threshold $\phi^$ collapses.

As illustrated in, a level of algorithmic similarity that appears harmless during tranquil times can suddenly cross this falling threshold during a crisis.

Once the threshold is crossed, the same algorithmic sector that helped price fundamentals accurately during peace begins to drive violent, correlated outflows during stress.

Rethinking Market Oversight

These findings fundamentally change how we might interpret the growing influence of non-bank financial intermediation (the sector housing hedge funds and money market funds). The authors report that the algorithmic sector can act as a "perfectly informed, stabilizing intermediary" in the limit of total diversity. This means that the rapid speed of AI trading is not the enemy. Rather, the lack of architectural variety is.

This perspective offers a new lens for viewing systemic risk. Instead of focusing on the sheer volume of money managed by AI-driven vehicles, regulators should perhaps focus on the "cognitive diversity" of the market. If the goal is stability, the objective is to prevent a "monoculture" where every participant is reading the same signals through the same mathematical pipelines.

Where the Evidence Stands

While the theoretical framework is robust, the authors are transparent about the limitations of their empirical testing. They tested their predictions using a panel of 19 emerging markets from 2000 to 2024. However, the statistical significance of the "state-dependent" effect relies on specific design choices.

Specifically, the authors note that the results are most significant when the "stress regime" is defined by the VIX (a measure of market volatility) rather than by fixed calendar years. Furthermore, the statistical significance weakens if the 2008–09 global financial crisis—the most informative period of stress in their dataset—is removed from the sample.

Finally, the authors admit a gap between their math and their data. Their theory predicts a change in the variance (the second moment) of capital flows. However, their empirical test primarily measures a shift in the conditional mean (the average outflow). While the results are consistent with the theory, they do not constitute a definitive proof of the variance-driven mechanism.