FISHERY BULLETIN: VOL. 87. NO. 2. 1989 



P„ - P„_, = (a;, -1) (P„_, - Pi,_i) + I a,j(P„_, - P„_j) 



j-i J 



J = 1 A- = 1 



X (-Pit-j - Pu-j-i) + CiXit + ea- 



(9) 



Changes in local market prices, P^, - P,t-\< are 

 then attributable to changes in central market prices 

 and past spatial price differentials between local and 

 central market prices. The latter variables allow for 

 the possibility that the markets are not observed in 

 an integrated equilibrium at a given time period, so 

 that there is feedback from prior disequilibria. 



Ravallion (1986) proposed the following sequence 

 of nested F-tests for the different null hypotheses. 

 First, test for long-run integration. If long-run mar- 

 ket integration is not rejected, then it should be im- 

 posed on the model with subsequent tests based on 

 a restricted form such as Equation (9). If the null 

 hypothesis of long-run market integration is re- 

 jected, then short-run market integration and mar- 

 ket segmentation are tested. 



Specification Issues 



Central and local market prices in Equation (1) 

 might be simultaneously formed. This possibihty 

 leads to a simultaneous equation problem, so that 

 parameter estimates could be biased and inconsis- 

 tent. Ravallion (1986) noted that the simultaneity 

 in the system can be easily dealt with by using an 

 appropriate instrumental variables estimator. This 

 paper uses the two-stage least squares estimate of 

 Equation (1) formed by replacing P^, with its pre- 

 dicted values from the reduced form equation ob- 

 tained from a regression of Pj, against its own 

 values lagged one period, the values of prices in all 

 local markets lagged one period, all dummy vari- 

 ables, and the time trend variable. 



Several nonprice influences (X,,) are possible. 

 First, the influence of seasonality is accounted for 

 by quarterly dummy variables for winter, spring, 

 summer, and fall. Second, the possibility of long- 

 term effects from increasing U.S. exports coupled 

 with continued reductions in Japanese harvest are 

 captured by a Hnear time trend. Third, a dummy 

 variable for the years 1984-86 is included to cap- 



ture any effects from the reduction of Japanese 

 sablefish catch within the U.S. 200 mile zone and 

 concomitant increase in U.S. harvests and exports 

 of sablefish that experienced an important increase 

 beginning in 1984 (Hastie 1988). 



Lagged Effects 



Any lagged effects in the model are likely to arise 

 from sluggishness in price adjustment, delays in 

 transportation, cold storage inventory holdings, and 

 expectations formation under price uncertainty 

 (Ravallion 1986). A maximum lag of six months was 

 chosen. This relatively long lag length allows for 

 Tokyo's fall prices to influence Alaska's spring 

 prices and ex-vessel markets after Alaska harvest- 

 ing has tapered off over the winter months. 



A 6 mo lag length also accommodates the effects 

 of commodity flows from the Alaska's spring 

 harvests on price formation in the Tokyo market. 

 The peak Alaska harvests occur in late spring and 

 early summer. The major Tokyo wholesale pur- 

 chases (approximately 65% of the year's total) are 

 concentrated from late May through October. After 

 these purchases, cold storage inventories become 

 particularly important in order to accommodate the 

 major Japanese consumption, which takes place in 

 the fall and winter months. Thus, the major inven- 

 tory holdings in the marketing chain occur at a 

 higher level than that which our study examines and 

 should not directly affect the model.' 



U.S. inventory holdings and transportation lags 

 make only a minor contribution to price formation. 

 Sablefish are shipped frozen. Cold storage holdings 

 prior to export are relatively small and declining in 



^While multicollinearity from current price and six lagged prices 

 could present a problem in estimates and tests of significance for 

 individual regression coefficients, the hypothesis tests in this paper 

 are for the joint effects, requiring tests on the joint confidence 

 region, so that multicollinearity presents far less of problem than 

 it ostensibly might appear. 



344 



