unreliable, but Brazilian and other coffees are traded on international markets, e.g., New York.
If the use of export tax rebates caused a bidding up of the Brazilian export price, the effect
should be evident in the New York market since the price there should reflect the export price
plus a reasonably constant amount for shipping and insurance. A simple model of the
relationship to be tested is:
5) PSANTOS4 = β0 + β1Alpha + β2PCOMPETITOR + β3RS + ε.
The dependent variable is the New York price per pound of Santos 4, a major Brazilian
Arabica coffee traded on the New York market. PCOMPETITOR is the price of a similar Arabica
coffee sold by a competitor, e.g., Colombia (MAMS) or Central America (Other Milds). The
model assumes that the prices of similar coffees move together over time since they are close
substitutes in consumption, save for substantial variations in relative supply (RS) and the
potential market distortion created by Brazil’s use of a unit export tax rebate (Alpha). Although
each of the independent variables in Equation 5) is almost certainly endogenous, I thought
Ordinary Least Squares likely to provide more robust estimates than Three Stage Least Squares
because of the limited availability of truly exogenous instruments and thus used both. I report
estimates for OLS regressions of Equation 5), an analogous model using first differences and
also for a system estimated using 3SLS. The approaches provide similar estimates of β and all
estimates are highly statistically significant.
I ran two regressions for each specification, one using the price of MAMS and the other
using the price of Other Milds as the competitor’s price. The supply variable attempted to
capture the effect of unexpected changes in relative supply on the assumption that prices are
more affected by unexpected rather than expected changes. To form this variable, the ratio of
Brazilian to other Latin American exports was regressed on a constant, a time trend, and an
16