METHOD
Nineteen soybean importing countries used
by Stallings in constructing his trade-weighted
real exchange rate index for U.S. soybean
markets were used to estimate six equations
with annual data for U.S. soybean exports to
the EC-9, Japan, Spain, Taiwan, South Korea,
and the remainder of the 19 countries (rest of
world, ROW), which collectively imported 93
percent of U.S. soybean exports during Stall-
ing’s 1983-1985 base period (countries and trade
weights in Appendix).
Previous studies of export demand for U.S.
soybeans have included as their explanatory
variables the price of soybean meal as a substi-
tute for soybeans (Houck and Mann; Houck et
al.), income or livestock in the importing
countries (Houck and Mann; Houck et al.;
Helmberger and Akinyosoye; Chambers and
Just), and exchange rates (Anderson; Chambers
and Just). We specified our soybean import
demand equations as input demand equations
with the prices of U.S. soybeans and soybean
meal, an exchange rate index, and pork produc-
tion (as a measure of output) as explanatory
variables.
We chose pork production as a representa-
tive of livestock production that uses soybean
meal in foreign countries, excluding ruminant
meat production that uses forages more exten-
sively than high-protein concentrate rations in
foreign countries. Poultry production, especially
broiler production, also uses soybean meal in
feed rations. However, the largest importer of
U.S. soybeans, the EC, uses substantially more
oilseed in pork production than in poultry meat
production (Leuck).
The six equations were specified as linear
combinations of the exogenous variables and
estimated in the form
(11) SBX. = bf, + b1.SBP + bs,iSMP + b4.PORK +
ɪ V __ H Zl Ol 1
b4iEXRi + uj,
where
SBXi = U.S. soybean exports to the ith mar-
ket (i = 1,...6);
SBP = U.S.soybeanprice,Rotterdam($/met-
ric ton * 1∕U.S. CPI);
SMP = U.S. soybean meal price, Rotterdam
(ditto);
PORKi = pork production in the ith market;
EXRi = real exchange rate index for the ith
market: (foreign currency units∕for-
eign CPI)/($1/U.S. CPI) indexed to
1980 = 100. For the EC and the ROW,
the individual country’s real exchange
rate indexes were weighted by Stall-
ing’s trade weights (shares of U.S.
soybean exports) before summing to
aggregate indexes for the EC and the
ROW (countries and weights in Ap-
pendix). The exchange rate indexes in
these six markets, when weighted by
Stalling’s market shares, sum to Stall-
ing’s trade weighted exchange rate
index, usedinthe OLS and2SLS single
equations for all 19 markets;
bji = parameters; and
ui = normally distributed random errors.
Calendar year U.S. soybean exports,
1963-1986, were the dependent variables
(United Nations, Commodity Trade Statistics').
Soybean and soybean meal prices, exchange
rates, and CPI indexes came from the
International Monetary Fund’s International
Financial Statistics and Taiwan’s statistical
counterparts (Central Bank of China; Council
for Economic Planning and Development). Pork
(pigmeat) production came from computer tapes
from the Food and Agriculture Organization of
the United Nations. Zellnerisunrestricted seem-
ingly unrelated regression (SUR), using annual
data, provided individual estimates of the
parameters on the variables for all six equations.
Testing for Simultaneous Equation Bias
Before estimating our equations by SUR, we
tested the market equation that represented
the largest share of 1983-85 U.S. soybean ex-
ports (the EC, which averaged 36 percent) for
simultaneous equation bias between soybean
and soybean meal prices and U.S. soybean
exports using a test developed by Wu and de-
scribed by Chow (p. 314).
To test whether U.S. soybean and soybean
meal prices were exogenous to EC imports of
U.S. soybeans, we obtained instrumental vari-
ables for soybean and soybean meal prices whose
estimated values were specified as a function of
U.S. soybean exports to the EC, plus the addi-
tional explanatory variable of the price in t -1.
We used the instrumental variables as ex-
ogenous variables in the EC import demand
equation and obtained 2SLS estimates for the
EC equation. We also obtained OLS estimates
of the EC equation. Wu’s statistic for testing for
differences between 2SLS and OLS estimates
in econometric equations,
<H0: B2S = bOLS against Ha‘ B2S ≠ BOLSX
is
W = n(B2s - B0bs)'V(q)-1(B2s - Bols),
131