Escaping the Poverty Trap: Why Some Households Stay Poor — and Others Don’t: Guest post by Ed Jee

This is the fifth in this year’s series of posts by PhD students on the job market.

Why do some households escape poverty while others stay poor?

For decades, development economists have debated whether “poverty traps” — self-reinforcing mechanisms that keep the poor poor — actually exist. In theory, a household might be stuck below a threshold where it cannot afford the large, indivisible investments needed to grow. In practice, however, markets, credit, and random shocks might mean they escape a trap.

In my job market paper, I revisit this classic question using a new approach: combining microdata from 27 randomized cash and asset transfer programs across 17 low- and middle-income countries. Together, these programs transferred more than $50 million to 75,000 households, providing an opportunity to test whether the poor are truly trapped — and how policy can help them escape.

Asset accumulation and poverty traps

Traditional evidence on poverty traps often comes from either cross-country comparisons or single randomized field experiments. Both approaches face limitations: identification in the former is difficult without exogenous country level shocks, while the latter cannot reveal how widespread traps really are. My approach harmonizes data across dozens of RCTs — from cash grants in Kenya to livestock transfers in Bangladesh — to trace how each household’s assets evolve over time.

At the heart of the analysis is a simple equation describing how household assets change:

Δassets = production – consumption – depreciation.

If this asset transition equation has multiple steady states — points where assets neither grow nor fall — then some households can be “stuck” in low-asset equilibria, even when others prosper. This gets to the heart of what it means to be trapped: with a sufficient “big-push” we could move poor households from low asset to higher asset steady states permanently.

A new way to see poverty traps

However, households are not passive, nor homogenous. They choose how much to consume or save in anticipation of future opportunities and often differ in innate microentrepreneurial ability, talent, or “grit”. Ignoring household choices and differences in ability can lead us to overestimate the presence of traps. For instance, “joining-the-dots” of the household asset transition equation when households have different abilities can create the illusion of multiple equilibria, even when each individual household has only one unique steady state.

My empirical framework explicitly models these endogenous consumption decisions and differences in ability, allowing heterogeneous households to save, invest, or smooth consumption over time. My framework has three steps:

1. Use the randomized transfers, supplemented by instruments implied by the panel structure of the data, to nonparametrically recover the household production function.
2. Estimate households’ consumption rule as a function of their current assets, ability, and persistent productivity shock. Disentangling if a household has high ability or just a sequence of lucky shocks is tricky in short panels, so I sidestep the problem by integrating over the joint distribution of unobserved variables.
3. Put the production and consumption functions together to generate the household asset transition equation.

Simulating from a household’s asset transition equation lets me figure out if a given household faces multiple steady states, and therefore is trapped. Averaging over the distribution of households in a given study gives a measure of how many households are trapped in a given context.

Trade-offs and Limitations

My approach involves a few important trade-offs. On the production side, I can estimate the shape of the production function in each study quite flexibly. However, I need to make a simplifying assumption: that differences in household ability enter multiplicatively. In plain terms, more able households produce proportionally more output at every level of assets, but the shape of their production curve is the same. This assumption may be more reasonable in rural settings, where households have limited choices of production activities, but more restrictive in urban contexts, where people can select from a wide range of occupations and technologies.

Estimating consumption behavior is even more challenging. Ability shapes not only how much a household consumes but also how it reacts to income shocks. Because randomized experiments follow households for only a few follow-ups, it’s hard to separate persistent ability from temporary luck. I address this by using households’ earnings histories: families with consistently high income are likely more able, while those with brief windfalls are probably just lucky. By conditioning on entire income paths rather than single observations, the model infers how unobserved ability is distributed across households. When these histories generate enough variation in predicted ability, a condition known as completeness, the relationship can be inverted to recover the full consumption rule, not just its average. Completeness is essential but not directly testable: readers can think of it as the instrument relevance condition from standard econometrics—only stronger, because it requires income histories to shift the whole distribution of ability, not just its mean.

What the data say

A key result I find is that many studies show signs of poverty traps: the “average” asset transition equation in 60% of studies exhibit multiple steady states. However, within a given study, I find that on average only about 25% of households are actually trapped.

Why the difference? Two main reasons stand out:

1. Ability differences matter.

Some households are more productive than others — they climb out of poverty regardless of where they start. Others are so unproductive that even large transfers cannot help, regardless of the transfer size, in the long run they will fall back down to their single, low-asset steady state.

1. Forward looking consumption responses make a difference.

Many households face fixed costs — like buying a cow, a sewing machine, or a metal roof — and I find evidence that they temporarily reduce consumption to save up. This forward-looking behavior helps them sort onto a higher return region of the production function and avoid being trapped.

These headline figures seem to be reasonably robust, while estimated thresholds may vary in a given study, I find the average fraction trapped lies between 10-26% across specifications.

Fixed costs

To understand what drives these traps, I look at the relationship between asset ownership and proximity to the estimated trap threshold in studies that report livestock ownership. The evidence is striking: households just above the threshold are 23% more likely to own a goat and 42% more likely to own a cow than those just below it. These large, indivisible assets are exactly the types of fixed costs that can create non-convexities in production — and, in turn, poverty traps.

In contrast, ownership of smaller, divisible assets (like chickens) does not change much around the threshold. This pattern suggests that “big push” investments can make the difference between stagnation and growth.

Next, I return to two of the original studies in my sample which discuss fixed costs: Balboni et al. (2022)’s asset transfer in Bangladesh, and Haushofer and Shapiro (2016)’s cash transfer in Kenya. Combining my reduced form results with a structural model, I find my estimated average thresholds line up with the prices of the indivisible good they mention. In Bangladesh, I find an average threshold of $547 against a cow price of $488 (and a $43 difference to the threshold they estimate!). In Kenya,  my structural model predicts a threshold of $974 against a roof price of $774.

What this means for policy

If poverty traps exist, the next question is: how should policy respond? Should transfers target households near the threshold — those who might escape with a smaller push — or the very poorest, who face the greatest deprivation?

To answer this, I calibrate a household growth model using the estimated production and consumption functions. The model simulates optimal transfer policies under different budgets and household types.

The result is clear:

Transfers that target the poorest households generally achieve the highest welfare gains.

Across studies, over 95% of optimal transfer policies decrease with household assets — the poorest receive the largest transfers. Even when a planner accounts for dynamic investment incentives, the optimal policy rarely shifts resources away from the poorest toward those “near the trap.”

Interestingly, I find that a simple heuristic — targeting households with the lowest current consumption — performs almost as well as the fully optimized policy. It delivers welfare improvements of 18%, compared with 23% under the optimal rule, without needing to infer a household’s dynamic optimization problem.

Conclusion

Taken together, the evidence suggests that poverty traps do exist—but not for everyone. Across 27 randomized programs in 17 countries, many contexts exhibit the hallmarks of multiple steady states and fixed costs that limit asset accumulation. Yet within those same settings, only a subset of households appear truly trapped: most remain poor not because they cannot grow, but because their productivity or opportunities remain limited.

In this sense, the question is not simply whether poverty traps exist, but for whom they matter. Understanding which households face structural barriers—and which can respond to opportunities when they arise—remains central to designing programs that foster lasting upward mobility.

Ed Jee is a PhD student on the job market from the University of Chicago.