3 of 10

What I really like about the paper

Very rich and detailed data
Cleanly and logically link empirical results to theory:

How important are monitoring costs to finance?

Greater increase in firms with ex-ante higher monitoring costs

How are entry and exit affected?

Exit of older and less productive entreprenuers; entry of new entreprenuers

How important are non-convex costs?

No differential treatment across industries ranked by average firm size

Does financial development facilitate better worker-firm matching?

Why does wage inequality increase with financial development?

Limited mobility across cities, demand for high-skilled workers grows

4 of 10

Suggestions on Economic Interpretation

Put variables in per-capita terms

Inverse hyperbolic sin has some unattractive properties
Credit per capita or GDP per capita or share employed in formal sector all very well defined, even when zero
Also help a lot with interpreting effect sizes

My best guess at magnitudes:

“Treated” municipalities get .2 additional bank branches on average, size of municipality is 10,000

1 new bank branch in a town of 50,000
How much credit does public bank introduce (appears to double amount of credit)
Baseline level of credit/deposits often zero in ‘treatment’ villages so percentage increases difficult
Number of firms increase by 9 percent (but this is ITT number), so should we think 45%? Seems high

5 of 10

What Should Be the “Headline Comparison”?

Prior to program, 40% of cities did not have a public bank prior to 2004
Authors call these cities “treatment” cities, though only 35% ultimately receive branches

“Treatment” cities go from 0 public sector branches to .35 public sector branches
“Control” cities go from 1 or more public sector branches to 1 or more public sector branches

I would be just as interested (or more interested) in a comparison of

Villages with no bank branch which go to one bank branch compared to
Villages with no bank branch which stay at no bank branch

Authors rightly point out that this would be a very endogenous comparison
But no more or less endogenous than the comparison the authors are running now
There may be some hope of understanding determinants of which municipalities receive a bank branch

6 of 10

More clarity on matching

We start with all 4,260 cities and compute quintiles of population and share of skilled workers over the pre-reform period. We then match with replacement each treated city with all control cities in the same quintile. We also match treated and control on employment, number of firms, average wage, and wage inequality measured by the Gini index, and select the three closest control cities.

My guess: divide cities into 25 groups [5 quintiles x 5 quintiles]
Look for Euclidean distance, maybe weighted sum of difference of z-score by employment, number of firms, average wage, and Gini, select three closest matches

But it’s not entirely clear

7 of 10

Identification concerns

Economically meaningful differences between matched “treatment” and “control” cities

These differences may limit ability of group * year fixed effects to serve as effective controls

Measure	Treat	“Control”	Difference
Wage(m)	431	488	13%
Employment(m)	645	1169	81%
Population(m)	10,652	11,829	11%
Local GDP	42,598	81,302	91%
Per Capita GDP			91%
Manuf Share	.15	.23	53%

8 of 10

How to Address Identification Concerns

“Treatment” and “Control” villages are substantively different and the parallel trends assumption may not hold after 2004 (significant convergence)

Consider synthetic controls approach to finding nearest neighbors
Show that size and per capita GDP do not predict variation in post-2004 trends in the large set of villages/cities that had a public-sector bank before 2000
“Placebo test” for wage distribution comparing richest half of already-banked villages to the poorest half of already-banked villages

Pre-trends in credit aren’t as informative because many “treatment” villages have no credit

Look for longer history of pre-trends with other data (census) that goes back further

Government in general may be targeting rural areas with other programs/transfers

Examine central or state transfers to municipalities over time
(They do this for bolsa familia, which is distributed by Caixa bank)

9 of 10

What is special about government owned banks?

Cole (2009) – Used regression discontinuity to compare villages in which bank branch was nationalized to villages in which bank branch stayed private

Public sector banks:

Lent more in total
Lent more to agriculture, same amount to industry
Charged lower interest rates
Had many more non-performing loans
Slowed down economic transition from agriculture to industry

C.f. Carvalho (2012)

Brazilian government-owned banks expand employment in politically attractive regions near elections

10 of 10

Conclusion

Really interesting policy experiment
Extraordinary data set
Extremely ambitious paper: tests a very wide range of theories

Feature, not a bug, but referees may disagree

More clarity on identification
Exploit additional variation of banked vs non-banked branches

Event-year regressions

More clarity on economic magnitudes

1 of 10

2 of 10