Transaction Costs and Market Institutions: Grain Brokers in Ethiopia
Transaction Costs and Market Institutions: Grain Brokers in Ethiopia
Transaction Costs and Market Institutions: Grain Brokers in Ethiopia
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
19<br />
7. ESTIMATION RESULTS FOR TRADERS’ USE OF BROKERS (STEP 2)<br />
In the second step, the share of brokered transactions is estimated with Tobit for<br />
both model specifications <strong>and</strong> for purchases <strong>and</strong> sales separately (Table 4). In<br />
order to control for the effects of local use of brokers (that is, with<strong>in</strong> regional<br />
market centers), the share of brokered transactions is restricted to brokers used<br />
for long-distance trade. The set of explanatory variables used <strong>in</strong>cludes regional<br />
dummy variables to capture the effects of regional effects. In a simultaneous<br />
system of Tobit equations, where predicted values of long-distance trade are<br />
used as a regressor, st<strong>and</strong>ard errors of the estimated coefficients will be<br />
<strong>in</strong>consistent. To correct for this, st<strong>and</strong>ard errors are estimated through<br />
bootstrapp<strong>in</strong>g with 1,000 replications, a procedure that provides generally very<br />
good estimates. 11<br />
Traders’ use of brokers is <strong>in</strong>dividually rational <strong>in</strong> that higher transaction costs<br />
lead to <strong>in</strong>creased use of brokers while higher social capital reduces the use of<br />
brokers, suggest<strong>in</strong>g that the presence of brokers enables traders to m<strong>in</strong>imize<br />
11<br />
This procedure <strong>in</strong>volves r<strong>and</strong>om draw<strong>in</strong>g, with replacement, N observations from the dataset,<br />
<strong>and</strong> estimat<strong>in</strong>g the statistics for each replication. From the dataset of estimated statistics, one can<br />
estimate the st<strong>and</strong>ard error of the statistic. However, the po<strong>in</strong>t estimate used <strong>in</strong> the org<strong>in</strong>al<br />
observed statistic θ obs rather than the average ⎺θ* from the replications (Mooney <strong>and</strong> Duval,<br />
1993).