96
where the terms are defined as for the models pertaining to the 31-40
size class. The calculated Box-Pierce statistics for both size classes
support the hypotheses that the residual series are white noise at the
.05 level. The above filter models are integrated (homogeneous degree
one) autoregressive with no moving average component. The Sims univer
sal filter" given as (1-.75B)^ (Bishop, 1979) was applied to each series
but did not transform the series into white noise. A lag of twelve
periods was used on all tests at the monthly level. Empirical research
on price movement in the shrimp market has not shown statistically
significant periodicity other than seasonal fluctuations (Thompson and
Roberts, 1982). Therefore, a full year was assumed sufficient to cap
ture all pertinent price responses. The number of observations used for
each causality test differed slightly due to the lag structure of the
ARIMA models used to filter the corresponding price series.
The 31-40 size class
The Haugh-Pierce tests for causality between the residual series of
ex-vessel and wholesale prices indicated that unidirectional causality
exists such that ex-vessel price causes (leads) wholesale price (Test I
in Table 1). The Haugh-Pierce statistic relevant to the null hypothesis
that ex-vessel does not cause wholesale price was larger than the tabu
lated critical chi-square value at 12 degrees of freedom and, thus, the
null hypothesis Is rejected at the .05 percent level. However, the null
hypothesis that wholesale does not cause ex-vessel price was not
rejected. Thus, unidirectional causality exists such that ex-vessel
causes wholesale price. A lag of one period was significant. The
current cross correlations were significant for both lead/lag alterna
tives in Test I, indicating instantaneous causality exists such that
current ex-vessel causes current wholesale prices, and vice versa.