The Rising Tide Problem: Why Economic Growth Alone Doesn't Lift All Boats

The Science

To study inequality, economists typically lean on one of two tools. They might crunch numbers from real-world data, tracing patterns across countries and decades. Or they build mathematical models—elegant abstractions that strip away messy human reality to isolate the essential forces at work. Both approaches have served well. But both struggle with a stubborn fact: economies are complex adaptive systems, where millions of individuals interact according to rules that are themselves shaped by those very interactions. Modeling this feedback loop mathematically is hard. Letting it emerge from simpler parts is the goal of agent-based modeling.

In this paper, Gustavo Kohlrausch and Sebastian Gonçalves at the Federal University of Rio Grande do Sul in Brazil take an agent-based approach to one of economics' most enduring questions: does economic growth help everyone, or does it primarily enrich those already wealthy? The question seems simple. The answer, their model suggests, is anything but.

Their framework builds on a tradition of wealth distribution models pioneered by economists like Jean-Philippe Bouchaud and others who treat the economy not as a single homogeneous blob but as a collection of individual agents whose interactions give rise to macroscopic patterns. What makes their approach distinctive is the combination of three elements rarely brought together: dynamic network structure, wealth exchange between connected agents, and independent stochastic growth. Most prior models treat agents as fully connected—everyone can trade with everyone else, like particles in an idealized gas. Kohlrausch and Gonçalves let connections evolve. Who trades with whom matters. And they embed this network in an economy that grows over time rather than conserves total wealth.

The model works through three alternating processes that repeat at each time step. First, the network rewires itself: agents form and break connections based on wealth, with richer agents more likely to attract new ties. Second, connected agents engage in wealth exchanges following the "Yard-Sale" rule—a framework from behavioral economics that captures a troubling empirical regularity: wealthier agents tend to win trades even when the outcome should be random. Third, each agent experiences independent growth drawn from a stochastic process with two parameters: a drift term $\mu$ representing the general health of the economy, and a volatility term $\sigma$ capturing heterogeneity in productivity. Some agents become more productive than others; some industries boom while others stagnate. The parameter $\sigma$ captures this variance.

Crucially, the exchange process includes a social protection factor $f$ that favors the poorer agent in each transaction. When $f=0$, trades are essentially symmetric despite wealth differences. When $f=0.5$—the maximum studied—poor agents have a substantially higher probability of winning any given exchange. This parameter serves as a proxy for redistributive policies: minimum wages, progressive taxation, social safety nets, union power. The researchers vary $f$ systematically to understand how institutional protection shapes economic outcomes.

Simulations run with 1,000 agents over 40,000 time steps, averaged across 1,000 independent runs to smooth out random fluctuations. The model includes a rescaling mechanism to prevent wealth values from growing without bound—the total wealth in the system expands, but individual holdings are normalized to maintain computational tractability.

What They Found

The headline finding emerges from a tension between two forces: economic growth and production heterogeneity. The drift term $\mu$ acts as a rising tide that in principle lifts all boats. The volatility term $\sigma$ creates waves of different sizes—some boats rise more than others, and some may even sink. The social protection factor $f$ mediates between them. And the interaction is non-linear: in isolation, increasing $\mu$ reduces inequality; increasing $\sigma$ amplifies it. But the effect of $\mu$ depends critically on the level of $f$. Economic growth benefits the poorest agents only when strong social protection is in place.

Consider the Gini coefficient—a standard measure of inequality where 0 represents perfect equality and 1 represents total concentration in a single agent. When $f=0.01$ (minimal social protection), the Gini index reaches nearly 1 as production volatility increases. Wealth condenses into the hands of a tiny elite. When $f=0.5$ (strong protection), inequality remains far lower even as $\sigma$ rises. The protective mechanisms buffer the system against the polarizing effects of heterogeneous productivity.

This buffering works not just on average inequality but on the distribution of wealth across percentiles. For the richest 1% of agents, increased production volatility dramatically concentrates wealth—over 95% of total wealth flows to the top percentile at moderate volatility levels when social protection is weak. Economic growth, captured by higher $\mu$, partially offsets this tendency, but its effect is modest compared to that of $\sigma$. More interestingly, the middle class—the bottom 10% to 50% of the wealth distribution—sees benefits from economic growth only when social protection exceeds a threshold. Below that threshold, growth in $\mu$ cannot compensate for the wealth erosion caused by high production variance.

The network structure adds another layer of insight. Assortativity—how agents with similar wealth connect to each other—responds to these parameters in revealing ways. At low social protection, the network becomes strongly disassortative: wealthy agents connect to poor agents, not to other wealthy agents. This makes intuitive sense: when the rich win trades, their counterparts lose wealth and connections. The system evolves toward a star-like topology where one agent sits at the center, connected to everyone, while everyone else connects only to the center and perhaps a few others. As $f$ increases, this disassortativity softens. Networks become more mixed, with wealthy agents also linking to each other. Economic growth itself reduces disassortativity by equalizing wealth across agents.

The temporal dynamics reveal something important: wealth exchange and stochastic growth interact in non-trivial ways over time. Early in simulations, systems with wealth exchange can actually show higher inequality than systems with pure stochastic growth—trades between agents favor the already-wealthy initially. But as time progresses, the exchange model diverges depending on $f$. With weak protection, it becomes more unequal as trades accelerate wealth condensation. With strong protection, it becomes more egalitarian as trades stabilize a middle class. The comparison is stark: the model without exchanges shows wealth condensation for any $\sigma >0.05$; the model with exchanges and strong protection maintains a broad distribution.

Figure 7 from the paper illustrates the network topology at equilibrium. Left: moderate production volatility (σ = 0.05) with strong social protection (f = 0.5). The network is relatively dispersed, with wealth and connections distributed across many agents. Right: high production volatility (σ = 0.25) with weak protection (f = 0.01). The network collapses into a star: a single agent at the center holds nearly all wealth and connections, while everyone else clusters at the periphery. This is wealth condensation made visible.

The cumulative wealth distributions tell a similar story. For $f=0.01$ and moderate volatility, essentially only the richest 1% of agents hold any wealth at all—the bottom 99% are effectively bankrupt. Increasing social protection to $f=0.1$ dramatically changes this: even at the same volatility levels, roughly 34% of agents retain wealth above minimal thresholds. The difference between these scenarios is not the underlying economic fundamentals—same $\mu$, same $\sigma$—but the rules governing how agents interact when wealth changes hands.

Why This Changes Things

The model's core insight—economic growth benefits the poor only when social protection is strong—arrives at a moment when this proposition is fiercely contested in democratic politics worldwide. It speaks to debates over minimum wage laws, progressive taxation, and the social safety net that have moved from academic journals to campaign trails. The finding does not adjudicate these debates. The model is an abstraction, not a policy simulation. But it identifies a structural logic that transcends the specifics of any particular economy: growth and redistribution are not substitutes. They are complements. Without the complement, the effect of the former may be captured by the latter.

This matters for how we think about the "rising tide" metaphor that titles the paper. The phrase suggests that general economic growth should lift all boats regardless of their starting position. The model suggests this is conditionally true: it lifts all boats if, and only if, the rules of exchange favor the less fortunate. Without those rules, growth may lift yachts while swamping dinghies.

The finding about production heterogeneity is perhaps even more quietly important. Standard economic discourse often focuses on $\mu$—the rate of growth—as the primary policy target. Kohlrausch and Gonçalves's model suggests that $\sigma$—the variance in productivity gains across sectors and individuals—may matter more for inequality. A growing economy where productivity gains are highly unequal may produce worse distributional outcomes than a slower-growing economy with more uniform gains. This reframes the policy question: it's not just about growing the pie, but about how the slices of the pie vary in size.

The network topology results carry implications for social cohesion and economic resilience. A star-like network, where one agent (or firm, or region) sits at the center with all connections while everyone else clusters at the periphery, is fragile. The periphery depends entirely on the center; if the center stumbles, the whole structure destabilizes. A more distributed network, where wealthy agents connect to each other as well as to poorer agents, is more robust. It allows wealth to circulate rather than accumulate. The model suggests that social protection policies don't just affect wealth distribution—they affect the structural architecture of economic relationships.

The finding that wealth exchanges can initially increase inequality before stabilizing is also noteworthy. It suggests that markets—treated by some as inherently equalizing mechanisms—may first concentrate wealth before redistributive forces assert themselves. This has echoes in debates over "trickle-down" economics: the idea that helping the wealthy first will eventually help everyone else. The model's temporal dynamics suggest this sequence is not guaranteed. The wealth concentration phase may become permanent if protective mechanisms are weak or absent.

For the growing literature on complex systems approaches to economics, this paper demonstrates the value of moving beyond mean-field models where everyone interacts with everyone. The network structure in the model isn't merely decorative—it shapes which agents trade with which, which in turn shapes the distribution of wealth. A fully connected economy and a network-structured economy with the same aggregate parameters can produce very different outcomes. This suggests that the topology of economic relationships—the structure of supply chains, the geography of trade, the clustering of industries—may be as important as the aggregate numbers.

What's Next

The model necessarily abstracts from many features of real economies. It assumes homogeneous agents except for wealth and a random risk factor; real economies contain firms of different sizes, workers with different skills, and institutions with different capacities. It assumes identical rules for all exchanges; real economies contain a variety of contract types, regulatory frameworks, and cultural norms governing transactions. It assumes a single aggregate growth rate $\mu$ that applies uniformly; real economies experience sector-specific growth, technological disruption, and geographic variation.

The authors note that they tested their results against a model without wealth exchanges—purely stochastic growth—to understand the independent effects of the exchange process. This comparison is valuable, but real economies contain both exchanges and growth simultaneously, plus other factors (investment, saving, inheritance) that the model does not include. Future work might productively extend the framework to incorporate these elements.

A particularly important extension would involve spatial or geographic structure. The current model treats the network as purely social—connections form based on wealth, not on physical proximity. But many economic phenomena have geographic dimensions: urbanization, regional inequality, the concentration of tech industries in particular cities. Adding space to the model might reveal how growth and protection interact with geography to shape inequality across regions.

The comparison with empirical data is implied by the paper's attention to universal patterns in wealth distributions—the two-class structure of upper tails and middle bulges that appears across countries. Future work might calibrate the model's parameters against real wealth distributions to test whether it reproduces observed patterns and, more importantly, whether it can predict how policy changes would affect real distributions. The Yard-Sale rule already captures behavioral biases in trading, but richer behavioral assumptions—loss aversion, reference-dependent preferences, limited attention—might bring the model closer to empirical regularities.

The finding that $\sigma$ may matter more than $\mu$ for inequality deserves further scrutiny. This emerges from a model with specific functional forms and parameter ranges. Whether it holds under different assumptions—different distributions of productivity shocks, different elasticities of substitution, different capital-labor ratios—remains an open question. It invites empirical investigation: has the rise in inequality since the 1980s been driven more by increased variance in productivity gains or by lower aggregate growth rates? The model suggests the former; testing this proposition against data would be valuable.

The policy implications should be read with appropriate humility. Models illuminate mechanisms, not magnitudes. The model shows that growth and protection interact in this way under these assumptions. It does not tell us how strong social protection must be, or what specific policies achieve it, or what tradeoffs might be involved. Those are empirical and political questions that go beyond what any model can answer. But the model does suggest that treating growth and redistribution as alternatives rather than complements is a category error—a misunderstanding of how economic systems actually work.

The network visualizations in Figure 7 provide perhaps the most intuitive entry point to the model's logic. The contrast between the dispersed network of the protected economy and the star-like structure of the unprotected one is not just a technical result. It is a picture of two possible futures: one where wealth circulates and connections diversify, and one where wealth accumulates and connections collapse onto a single point. The rising tide does not choose between these futures on its own. The rules of exchange—who wins, who loses, who is protected—shape which future arrives. That is the model's core message, and it remains as relevant outside the model as inside it.