chilled fresh fish can be held. Since most ground- 

 fish harvested in New England waters are not pro- 

 cessed into frozen fish products, long-term storage 

 of New England groundfish is unlikely, and fresh 

 fish prices are likely to adjust more quickly than 

 those of most other food commodities. In addition, 

 previous exploratory analysis with adaptiver filter- 

 ing methods on the weighout file suggests that two 

 sets of round ex-vessel prices for any species k are 

 particularly important, the previous month's price 

 and the price within one month on either side of the 

 previous year. In order to account for these charac- 

 teristics and to provide both short and long versions 

 of the test, lags of 8 and 14 mo were specified. These 

 lag lengths are sufficiently long to encompass price 

 lags with monthly data. The diagnostic Q test of Box 

 and Pierce (1970) is used to detect serious serial 

 correlation. 



EMPIRICAL RESULTS 



The empirical results from the direct Granger 

 causality tests lead to somewhat unexpected con- 

 clusions for most species. The null hypothesis that 

 monthly round ex-vessel prices of cod and haddock 

 in all three ports are first formed in the Boston auc- 

 tion market is rejected in almost all instances. The 

 findings in Table 1 instead suggest that the cod and 

 haddock prices established in the New Bedford auc- 

 tion lead the prices formed in the Boston market. 

 Several factors may account for this. The New Bed- 

 ford auction's volume of landings is substantially 

 higher than that of Boston. In addition, the two 

 market times ordinarily overlap, and frequent com- 



munication occurs between economic agents during 

 the auctions. Further, the proximity of New Bed- 

 ford to Boston allows fresh fish to be easily trucked 

 to Boston from New Bedford. The markets are thus 

 physically linked, before the auctions by fishermen 

 and after the auctions by fish buyers. One element 

 of conventional wisdom may perhaps be substan- 

 tiated, however. Although the Q-test statistic in- 

 dicates severe serial correlation (and thereby 

 possibly refuting the F-test statistic), the empirical 

 results indicate that Boston cod prices do lead 

 Gloucester cod prices at the ex-vessel level for the 

 shorter lag length parameterization. 



Rejection of the null hypothesis that Boston prices 

 lead New Bedford and probably Gloucester cod and 

 haddock round ex-vessel prices and the finding that 

 New Bedford prices lead Boston prices suggest a 

 second null hypothesis for consideration. This sec- 

 ond hypothesis states that Gloucester cod and had- 

 dock prices are directly led by New Bedford prices. 

 In addition, the possibilities that Boston prices lead 

 Gloucester prices and that New Bedford prices lead 

 Boston prices suggest an additional, indirect price 

 linkage between Gloucester and New Bedford via 

 Boston. 



The results for this second null hypothesis are also 

 given in Table 1. Since this is an unplanned com- 

 parison, a Scheffe interval is used. 8 Strictly followed, 



8 An unplanned comparison occurs when in the course of exam- 

 ining results a hypothesis is tested which was not specified prior 

 to the experiment. The initial region is altered by the additional 

 information, so that the level of significance has changed. A Scheffe 

 interval allows for a more cautious test by providing a larger 

 critical value than that given by a t or F table. This pre-test bias 

 is accounted for by a conservative test. The F-test statistic now 



Table 1 .—Direct Granger causality tests for monthly fresh round ex-vessel cod and haddock 



prices. 



Cod 



Haddock 



Direction 1 



Lags 2 F-test 3 Q-test 4 



Direction 1 



Lags 2 F-test 3 Q-test 4 



B 



B 



G 



G 



B 



B 



NB 



NB 



G 



G 



NB 



NB 



->G 

 ->G 

 ->B 

 ->B 



->NB 



->NB 



->B 



->B 



->NB 



->NB 



->G 



->G 



8 

 14 



8 

 14 



8 

 14 



8 

 14 



8 

 14 



8 

 14 



2-37* 

 1.17 

 1.59 

 1.02 

 1.67 

 0.74 

 2.52* 

 1.89* 

 1.78 

 0.91 

 5 2.96 

 1.40 



35.67* 



2.61 



10.98 



12.35 



13.52 



21.85* 



8.82 



16.27 



16.46 



28.42 



7.84 



6.86 



B 



B 



G 



G 



B 



B 



NB 



NB 



G 



G 



NB 



NB 



->G 

 ->G 

 ->B 

 ->B 



->NB 



->NB 



->B 



->B 



->NB 



->NB 



->G 



->G 



8 

 14 



8 

 14 



8 

 14 



8 

 14 



8 

 14 



8 

 14 



1.69 

 1.09 

 1.07 

 1.29 

 0.08 

 0.09 

 3.15* 

 1.76* 

 0.56 

 1.36 

 5 2.90 

 1.65 



9.60 



4.49 



10.98 



15.07 



12.54 



7.06 



8.62 



8.74 



11.71 



15.09 



11.76 



8.37 



'Variable abbreviations are B (Boston), G (Gloucester), NB (New Bedford). 



2 J indicates J months lagged. 



3 Null hypothesis that past values of the causal variable do not significantly affect current values of the 

 dependent variable. An asterisk indicates rejection of the null hypothesis at the 5% level. 



4 Null hypothesis that regression residuals are white noise. An asterisk indicates rejection of the null 

 hypothesis at the 5% level. 



5 F-test statistic is significant at the 5% level, but not significant at the 5% level when a Scheffe interval is used. 



440 



