A classic question in development economics is whether (and what types of) poverty traps exist. This new paper by Balboni, @orianabandiera, Burgess, @maitreesh, @anton_heil finds evidence for poverty traps in Bangladesh. Let me try to explain... [1/N]

https://www.dropbox.com/s/4tfuhclfvh6ynnd/povertyTraps.pdf?dl=0

In a village, there are poor and rich people. One view is that these income differences reflect traits and skills. People get sorted into occupations based on their innate talent. Inevitably some end up earning less than others. [2/N]

An alternative view is that a person is poor today because that person started off poor. Had that *same* person started off rich (imagine they won the lottery the day they were born), they would today be rich. This is the "poverty trap" view. [3/N]

A stark version of the "poverty trap" view is that there is some threshold level of wealth below which one would gradually fall to some low level of income, above which one would gradually converge to the good life. [4/N]

Both views have some truth to them. But which view best approximates the world matters a fair bit. For one thing, the poverty trap view motivates temporary big push policies: attempts to push poor people over the threshold, putting them on a path to the good life. [5/N]

The paper in question uses data from a BRAC-led randomised controlled trial of one type of big push: transfers of large assets (mostly cows, $560 PPP) to the poorest women in a set of villages. [6/N]

If a poverty trap exists, these transfers should create long-lasting increases in income *only if* the transfer pushes a woman over the threshold. Otherwise, the transfer will have only temporary effects, with the woman falling back to her initial level of income. [7/N]

Now the slightly technical bit: this below vs. above-threshold behaviour has implications for how asset levels today translate into asset levels tomorrow. If a woman's asset level is above the threshold, her assets grow. If below, her assets shrink. [8/N]

Graphically we should see something S-shaped like the figure below: K is assets, K-hat is the threshold. Kp is where you end up if got trapped (starting with assets below K-hat). Kr is where you end up if you started above K-hat. [9/N]

To test these poverty trap dynamics, we would ideally randomly assign many different levels of assets (above and below K-hat), and then re-draw that figure with real data, and check for the S-shape. [10/N]

The paper doesn't quite have this -- all women got the same value asset. Instead, it uses variation in baseline assets to get variation around that threshold. Every treated woman got the cow, but for some the cow took them to $600 in assets, for others just $570. [11/N]

The killer figure is now upon us. The mapping of productive assets in 2011 to those in 2007 is S-shaped! The paper does more work to convince us this is really causal (recall: the variation on the x-axis here was not randomly assigned). This looks like a poverty trap. [12/N]

The threshold is ~$500. Now with poverty traps, people don't stay around the threshold long: those above grow away from it, those below sink below it. At baseline assets are distributed bimodally, with *very few* around $500 (2.3 below), as the poverty trap predicts. [13/N]

The authors argue that those just below the threshold cannot afford the lumpy goods (the cow itself, and vehicles to travel to market) to keep rearing livestock. Those above can, and so keep growing their incomes. [14/N]

What does this mean for development policy? Small transfers (e.g. typical microloans) may only have temporary effects. Large transfers are more likely to persistently improve income. [15/N]

Of course it's obvious that large transfers should matter more than small transfers. But the point here is that they are so much more effective that we should prefer to spend $X on large transfers than $X on many small transfers. Welcome back big push development economics. [N/N]