Returns to Capital in Sri Lanka

RETURNS TO CAPITAL IN
MICROENTERPRISES: EVIDENCE
FROM A FIELD EXPERIMENT
Chris Woodruff, UC San Diego
(With David McKenzie and Suresh de Mel)
October 2006
The project


We estimate returns to capital for a
set of very small household
enterprises.
No paid employees, capital of less
than $US 1000 (~lower 25% of
distribution of self employed in Sri
Lanka)
Returns to capital in microenterprises -Why do we care?



Large portion of urban labor market is
self-employed. About one-third in Sri
Lanka
What is the potential for growth of
these enterprises?
Even absent sustained growth, what is
the potential for increasing incomes
among these individuals?
Why might returns be high or low?

Low returns:
– Minimum scale of investment / nonconvex production sets

High returns:
– Capital constraints
– Risk / uncertainty
Evidence on returns to capital in
enterprises
Among others:
 Banerjee and Duflo 2003 (74%)
 Bigsten et al (30%)
 Udry and Anogol 2006 (60%)
 McKenzie and Woodruff 2006
(10%/month)
What is wrong with existing evidence?



Some from cross section: Worry about
conflating ability and capital investment
Some from loan programs: Measure only for
the self-selected sample that applies for
credit
McKenzie and Woodruff suggests that
returns are high in the broad sample of
firms, yet take up rates for loan programs
are low
The Experiment


Randomized experiment where we provide
grants to enterprises to create exogenous
variation in capital stock
Selected 618 firms in three districts in
southern Sri Lanka (Kalutara, Galle,
Matara)
–

Sample drawn from block-to-block census in
selected GNs
Surveyed first in March 2005, then
quarterly since (5 waves used in the paper
– now up to 7 waves)
The Experiment

Firms in three zones:
1) Suffered direct damage from tsunami
2) In coastal zone, but no damage
3) Farther inland

In this paper, we exclude firms
directly affected by the tsunami
The enterprises


All had less than 100,000 SLR ($US1000) in
capital (not counting land and buildings) in
baseline survey
Half in retail, the other half in
manufacturing / services (clothing, lace,
bamboo, food products)
Sri Lanka: capital shock

After the first and third round of the survey,
randomly selected firms were given capital
shock
– ~$100 or ~$200, in cash or equipment
– 59% of firms received treatment

Larger treatment is:
– About 75% of median capital stock
– About 6 months of reported earnings

Use grants rather than loans because we
want to measure the full spectrum of firms
Capital Shock

Baseline survey asked what firms
would purchase if they had:
% Inventories
5,000
68%
10,000
59%
15,000
49%
Most profitable
17%
Median most profitable investment is 25,000; 2/3rds
say less than 30,000; 20,000 is enough for 42% of
the firms to make their most profitable investment
Capital Shock


About 55% of the in-kind treatments
were inventories
Of the cash treatments invested in the
enterprise, about 2/3rds were spent
on inventories
Pre-treatment means
Post-treatment means
Estimate FE regression
5
 i ,t  Amounti ,t   t  t   i   i ,t
t 2
Also consider revenues, log profits
Results
Table 4: Effect of Treatment on Revenues and Profits
(2)
(4)
Log
revenues Revenues
FE
FE
Treatment amount (0, 1, 2)
(7)
(9)
Log
profits
FE
Profits
FE
0.177***
3058***
Hours worked
-0.0003
12.7
0.001
3.3
Wave effects
yes
yes
yes
yes
1812
383
1812
383
1795
383
1795
383
Observations
Number of groups
Median
7000
0.126*** 531.369**
3000
Results
Effect of Treatment on Profits
5 rounds
5 rounds
7 rounds
1% trim
Treatment amount (0, 1, 2)
Wave effects
Observations
Number of groups
7 rounds
1% trim
Profits
FE
Profits
FE
Profits
FE
Profits
FE
568.7**
614.4**
712.5***
557.1***
yes
yes
yes
yes
1857
383
1834
383
2575
383
2531
383
Interpretation


10,000Rs treatment increases profits
by 560-712Rs. i.e. a 5.6-7.1% return
Log specification gives around 4%
return.
Non-linearities in returns?
Tests for Linearity of Impact
Profits
5 rounds
5 rounds
1% trim
7 rounds
Treated amount 10,000 Rps
813.3**
731.5**
851.2*
Treated amount 20,000 Rps
1022.4** 1172.2***
7 rounds
1% trim
575.9**
1357.0**
1100.9***
F-test p-value 20,000 = 2*10,000 effect
0.480
0.623
0.725
0.929
Observations
Number of firms
1858
383
1835
383
2576
383
2534
383
Cash vs Equipment
Treatments



Cash was given without restrictions, told
they could purchase anything they wanted,
for themselves, household, business, or
other
Asked owners what they had done with
treatment
Approximately 58% of cash was invested in
business, additional 12% saved, 6% used to
repay loans.
Cash vs. equipment treatments
Cash vs Equipment Treatments
5 rounds
5 rounds
7 rounds
7 rounds
Cash Amount
Profits
FE
523.4*
1% trim
Profits
FE
527.0**
Profits
FE
889.1**
1% trim
Profits
FE
633.7***
Equipment Amount
613.2**
700.5***
537.3
482.0***
F-test of equality p-value
0.811
0.501
0.413
0.543
Observations
Number of groups
1857
383
1834
383
2575
383
2531
383
Higher profits or higher
reported profits?

Possible concern is that this might just reflect a
change in reporting of profits:
1) Perhaps treated firms trust us more and so are less likely
to underreport
– Address this by looking at reporting done after treatment
for sales in periods before treatment
i) Asked March 2005 sales in Round 1, and then re-asked
about these in Round 2, after some firms were treated.
=> No significant differences between groups in ratios
ii) Second group of firms were treated in November,
interviewed at start of October and in January. Compare
ratio of October sales (asked Jan) to September sales
(asked Oct) for treated vs untreated
=> again find no significant difference between groups.
Higher profits or higher
reported profits?
2) Perhaps then firms overreport profits after
treatment, since they want to show us that giving
them money is good?
- If this was the case, we would expect them to
overreport the share of the cash treatment invested
in business
- But on average say about 55-65% invested in
business, and return we get from cash treatment is
2/3rd return from equipment treatment
=> Appears that treated firms are not overreporting
Results so far:




Real returns 5-6% per month
No evidence inconsistent with linearity
of returns
Only small decay over time
Returns are much higher than interest
rates on micro loans (3-7% per year)
– What “explains” this gap?
Females vs Males



Many microfinance organizations
concentrate on lending to women
Is there any evidence to support them
having higher returns?
See that women invest lower share of
the cash treatment in business on
average (67% vs 88%, p=0.08).
Males vs Females
Gender and Treatment Effect
5 rounds
5 rounds
7 rounds
7 rounds
Amount
Profits
FE
782.4**
1% trim
Profits
FE
835.9***
Profits
FE
807.9**
1% trim
Profits
FE
795.5***
Amount*Female
-667.9
-588.3*
-467.3
-715.9**
1857
383
1834
383
2575
383
2531
383
Observations
Number of groups
What do the firms say are
their constraints?
Table 7: What do Firms Report as Constraints to Growth?
% of firms reporting
that this is a constraint
Lack of Finance
Lack of Inputs
Lack of Demand
Lack of Market Information
Lack of Clear Ownership of Land
Economic policy uncertainty
Costs of hiring new employees
Poor quality roads
Lack of trained employees
Legal regulations
Poor quality electricity and phone
High crime rates
High tax rates
92.7
53.8
34.5
15.9
15.7
15.1
11.2
8.1
6.8
6.3
3.7
1.6
1.0
CREDIT
MARKET
RISKINESS
OF RETURNS
How do firms finance
existing business?
Only 3.1% have bank account
 89% got no start-up funding from bank or
microfinance
 71% relied entirely on own savings and
family for start-up funds
 83-100% of firms making purchases of
equipment between waves used only own
savings and family to finance this
 Internal capital market of household is
major source of funds.

Heterogeneity of returns





Model of capital constraints, risk and uncertainty.
Household has endowment of assets, earns money
from market labor of other household members.
Can finance capital stock through borrowing, and
through its internal capital market
With well-functioning credit and insurance markets,
will choose capital stock such that marginal return
to capital = market interest rate
With missing markets, marginal return to capital will
exceed market rate
Heterogeneity of returns

Predictions:
– Returns lower when capital constraints less
severe



More workers in household (baseline)
Lower entrepreneurial ability (measures: education;
digit span test; self-efficacy; time solving maze)
Higher wealth (durable assets)
– Returns higher when more risk and uncertainty


CRRA estimated with lottery exercise
Uncertainty from subjective distn of profits
Heterogeneity of returns
S
ln i ,t  Amount i ,t    s Amount i ,t  X s ,i ,t  hoursi ,t
s 1
 5

   t  t      s ,t  t  X s , i , t    i   i , t
t 2
s 1  t  2

5
S
Heterogeneity of returns
Table 9: Treatment Effect Heterogeneity
Dependent Variable: Log Profits
(1)
0.126***
(2)
0.238***
Interaction of Treatment Amount with:
Number of Workers
-0.0946**
Treatment Amount
Asset Index
Broad Entrepreneurial Ability
Risk Aversion
Uncertainty
(3)
0.131***
(6)
0.118***
(7)
0.125***
(8)
0.201**
-0.00428
0.0926***
-0.00995
-0.136
Conclusions





Shocks to capital were large and random,
and hence uncorrelated with ability
Suggested returns 4-8% per month
No evidence of non-linearities in returns
Returns higher where capital constraints
bind tighter
No evidence returns affected by risk,
uncertainty