You are on page 1of 9

Agricultural & Applied Economics Association

Testing Market Integration Author(s): Martin Ravallion Source: American Journal of Agricultural Economics, Vol. 68, No. 1 (Feb., 1986), pp. 102-109 Published by: Oxford University Press on behalf of the Agricultural & Applied Economics Association Stable URL: . Accessed: 30/08/2013 01:34
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact

Agricultural & Applied Economics Association and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend access to American Journal of Agricultural Economics.

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions




Martin Ravallion
A model of spatial price differentials for a tradable good is proposed which avoids the inferential dangers of received methods using static price correlations. The proposed method also extracts more information on the causes of price differentials from the same data. The method is illustrated using monthly rice price data for postindependence Bangladesh, including the very substantial regional price shocks during the 1974 famine. Impediments to market integration are indicated. Key words: Bangladesh, differentials. famine, market integration, modeling spatial price

Though much maligned, static price correla- at the two locations are synchronously, identitions remain the most common measure of cally, and linearly affected by another varispatial market integration in agriculture.' By able. Possible examples include the price of a this method, bivariate correlation or regres- related third good traded in a common market sion coefficients are estimated between the or a shared dynamic seasonal structure in protime series of spot prices for an otherwise duction. Then one can readily express price in identical good or bundle of goods at different one market as a linear function of price in the market locations.2 other market, with slope unity, even though There are a number of inferential dangers in the markets are segmented. Of course, any bivariate modeling of this sort. The following measurement errors or omitted variables example expands on an important point made would yield imprecision in a test equation elsewhere in the literature on market perfor- based on the static bivariatemodel. But it remance in agriculture (Blyn, Harriss). Suppose mains that under these conditions the received that trade is infinitely costly between two mar- method fails hopelessly as a test for market ket locations but that the time series of prices integration. The likelihood of serial dependence in the residuals obtained from a static model calibrated to nonstationary time series The author is a research fellow, Queen Elizabeth House, Oxford, also leads one to be suspicious of the conclupresently at Department of Economics, Research School of sions drawn from this method (Granger and Pacific Studies, Australian National University. Newbold). Financial assistance from the Overseas Development Administration (U.K.) is also acknowledged, subject to the usual disHowever, with the same data, the static claimer. bivariate method can be readily extended into The author wishes to thank the Journal's referees for useful comments on an earlier draft. a dynamic model of spatial price differentials. Review was coordinated by Hans P. Binswanger, associate By permitting each local price series to have editor. own dynamic structure (and allowing for its Another, less common, method of testing for market integration is to calculate the spatial variance of prices and test for longany correlated local seasonality or other charrun convergence to zero or close to it (for an example, see Hurd). as well as an interlinkage with However, it can be readily shown that if prices at different mar- acteristics) other local markets, the main inferential dankets are generated by identical but independent stationary autoregressive processes then they will asymptotically converge to gers of the simpler bivariate model can be zero variance (see Ravallion, forthcoming). Thus, nothing can be avoided.3 Most important, the alternative hyinferred about the interlinkage of markets from the results of such
SA statistically significant coefficient is sometimes taken to imply market integration. Of course, if testing for a nonzero result, the null hypothesis being (invariably) rejected is market segmentation, i.e., the absence of any relationship between the two price series. One could easily reject this without accepting that the markets are integrated in the sense of having a reasonably stable price differential. This problem can be readily solved by changing to the appropriate null hypothesis for market integration, as some studies have done.

3Spurious co :elations can also be avoided by filtering the price series prior to calculating pairwise correlations; this can be done by testing for residual cross correlations amongst univariate ARIMA models of each price series (Haugh; Fase applies the method to international share prices). Rather than follow this approach here, I have preferred to formulate and test market integration conditions as nested hypotheses within an explicit multivariate model of spatial market structure. As will be seen, this permits a much greater range of null hypotheses of economic in-

Copyright 1986 American Agricultural Economics Association

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions


Testing Market Integration


potheses of market integration and market segmentationcan then be encompassedwithin a more general model and so tested as restrictedforms. A dynamic model also has the advantage that one can distinguishbetween the concepts of instantaneous market integrationand the less restrictiveidea of integration as a long-run target of the short-run dynamic adjustment process. This distinction seems important.In many settings it will be implausiblethat trade adjustsinstantaneouslyto spatialprice differentials, and so one would be reluctantto accept short-runmarket integrationas an equilibriumconcept. But, given enough time, the short-runadjustmentsmight exhibit a pattern which converges to such an equilibrium.If short-runintegrationis rejected, then it would be nice to know if there is any long-runtendency toward market integration.By permitting the investigatorto answer this new question, a dynamic model can extract more information about marketsfrom the same data used by the static model. This paper offers an approachto testing agriculturalmarketintegrationalong these lines and illustrates it using data on the interrefor rice in Bangladesh gionalprice differentials duringthe turbulentpostindependenceperiod, 1972-75. The periodof analysis has been chosen to include the substantialregional price shocks which occurredduringthe 1974famine (Seamanand Holt, Alamgir,Sen). ElsewhereI have examinedthe intertemporal performance of rice marketsin Bangladeshduringthis period (Ravallion 1985b). Why Study MarketIntegration? It is well-knownthat, underregularly assumed restrictions on the continuity, slope, curvature, and domain of utility and production functions, a competitive equilibrium for a complete set of markets will exist and be efficientin the Paretiansense. In general, this will also hold for the spatialcompetitiveequilibriumof an economy consistingof a set of Nregions among which trade occurs at fixed transportcosts (Takayamaand Judge). Such an equilibriumwill have the propertythat, if
terest than is possible with simple correlations. However, the prewhitened bivariate correlation method is likely to be useful in some applications, particularly if one has no a priori basis for identifying a model of market structure; for an example, in a similar setting, see Ravallion (forthcoming).

trade takes place at all between any two regions, then price in the importing region equals price in the exporting region plus the unit transport cost incurred by moving between the two. If this holds then the markets can be said to be spatiallyintegrated. Marketintegrationis by no means sufficient for the Pareto optimality of a competitive Even when based on a sound emequilibrium. piricalmethodology,the conclusion that markets are well integrateddoes not, of itself, imply an efficient spatial allocation (see, for example, Newbery and Stiglitz). Nonetheless, one can be interested in empiricallytesting for spatial marketintegration withoutwishing to rest the case for or against Pareto optimality on the outcome. Measurement of market integrationcan be viewed as basic data for an understanding of how specific marketswork. For example, in the present setting, a study of the dynamics of marketintegrationshould throw some light on one of the oldest questions concerning famines in market economies: how long can an initiallylocalized scarcity be expected to persist? Policies of noninterventionwith markets duringfamines have often been advocated or defended along the following lines: given that the necessary transportinfrastructure exists, the unaidedresponse of grain tradersto the induced spatial price differentialswill quickly eliminate any localized scarcity. For example, this assumption was a cornerstone of the governmentof India'sfaminerelief policy duringmuchof the nineteenth and early twentieth centuries (Bhatia, Ambirajan).Against this view, it has often been argued that markets will be too slow to respond; for example, Ambirajanreports that duringthe severe famineof the mid1870s, the local government in Madras rebelled against the government of India's policy, arguingthat "if time were given to the market,the necessary grainwould eventually come, but time was what could not be given" (Ambirajan, p. 95).4 Since independence,governments of India and other countries in the subcontinenthave tended to adopt highly in4 The assumption that a localized shortfall will be reflected in spatial price differentials has also been questioned. For example, Sen has pointed to the possibility that the shortfall may be associated with a fall in incomes and, hence, demand (due to, for example, the effect of a drought on employment) and that this will mitigate the upward price response needed to encourage imports. Elsewhere I have examined the theoretical and empirical grounds for this argument (Ravallion, forthcoming).

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions

104 February 1986

Amer. J. Agr. Econ.

terventionist policies concerning food grain markets (see, for example, Bhatia, chap. 12; George). An empirical assessment of the speed of market adjustmentto spatial price differentials may help resolve this longstandingdebate.
Modeling Spatial Market Structure

The specificationof an econometric model of spatialprice differentialswill depend, in part, on assumptions about spatial market structure. Here I shall assume that there exists a groupof local (rural)marketsand a single central (urban)market.While there may be some tradeamongthe local markets,it is trade with the central market which dominates local price formation.Dependingon the numberof local markets and their sizes, one can also posit that the central market price is influenced by various local prices. Thus, the static pattern of price formation amongN markets,where market I is the centralmarket,may be summarized by a modelof the form
(1) Pl = fl(P2,P3, fi(Pl,X,)

Clearly, this approach has its limitations. Since spatial price differentialsbecome more aggregated,it produces inferentialdifficulties when investigatingthe linkagelocation of any revealedimpedimentto trade. Indeed, if there is a large number of local market linkages, then (dependingon what other local non-price variablesare relevant) it may become impossible to identify even the indirect radiallinkage. As always, the merits of the model need to be judged in its specific applications. Since the main aim in estimatingthe model is to test alternative hypotheses to do with market integration, its econometric specification should not prejudice the outcome. This is most easily assured if the alternative hypotheses can be nested within a more general model and so tested as restricted forms. For estimation,it is also convenient to
assume that the functions

(2) Pi =


,* PN,XI)

(i = 2,

. .

, N)

where X, (i = 1, .

. , N) is a vector of other

influences on local markets. The functions f. N) can be thought of as solutions (i = 1,..., conditionsfor marketequiof the appropriate librium, taking account of the main spatial choices and the costs of adjustment facing traderswhen decidingwhere to sell. The derivation of these functions does not seem to pose any new theoreticalproblemsor insights, and so I shall take them as given. At first sight, this model only seems well of marsuited to a simple radialconfiguration kets in which each local market is directly linked with the central market.In most applications a more plausible configurationis one in which some local markets only trade with the central market via other markets. However, providedone is willing to forgo identification of at least some of the nonradiallinkages, the radialmodel given by equations (1) and (2) can often provide a useful characterization of more complex marketstructures.By subsuminglinkages between intermediatelocal markets, an (implicit)binary relation can be obtained between each local market and the central market, and so the radialmodel is preserved.

can be given a linear representationby introducingan appropriatestochastic term. The econometric version of equations (1) and (2) shouldalso embody a suitabledynamic structure;as is well known, dynamic effects can arise from a numberof conditions in the underlyingbehavioral relations includingexcosts (for pectationsformationand adjustment a survey of the possibilities see Hendry, Pagan, and Sargan). these considerations,the followCombining ing econometric model of a T-periodseries of prices for N regions is assumed:

f, (i = 1, . . . , N)

(3) P,


+ alP1t-1


+ XSC1 +
(4) =




+ a.Pit1


+ Xitci + eit


(i = 2, . . . , N)

errorprocesses where the e's are appropriate and the a's, b's, and c's are fixed. A wide range of possible hypotheses about interregional trade can be formulated and tested as parameterrestrictionson equations (3) and (4). In discussingthese, I shall concentrate on equation (4) since, in many applica(I tions, equation (3) will be underidentified. shall returnto this point later.) In terms of the parametersof equation (4), the following hypotheses will usually be testable.
(a) Market segmentation. Central market

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions


TestingMarketIntegration 105

prices do not influence prices in the ith local marketif

(5) b 0 (j = 0 ... , n),

reestimated with long-run integration imposed. For example, under long-runintegration, equation(4) can be writtenin the following equivalentform:
(11) APit

in which case the data would be better described by the correspondingrestrictedform of equation (4). crease in the central market will be immediatelypassed on in the ith marketprice if
bio = 1.

= (ail



(b) Short-run market integration. A price in-

j=2 n-1

+ bioAP1t

Of course, there will also be lagged effects on future prices unless, in additionto (6): , n). (j = 1,... (7) aij = byij= 0 If both (6) and (7) are accepted as parameter restrictions,then one can say that marketi is integratedwith the central marketwithin one time period. A weakerform of marketintegration will also be tested in which the lagged effects need only vanish on average:

P1,t-) aij(Pit-j j + - 1+ + bik) Plt-j j=1 (bio k=l aik

+ Xici + ei,.



+ b.

= 0.

(c) Long-run market integration. A long-run

is one in which marketprices are equilibrium constant over time, undisturbedby any local stochastic effects. So consider the form that
equation (4) takes when Pit = P'i, Pit = = 0 for all t; then and


P*I Z (9)
P*i = j=o
n bi



ai1 It can be seen that marketintegrationnow requires that

n n



a +

b = 1.

If this parameterrestrictionis accepted, then the short-runprocess of price adjustmentdescribed by the model is consistent with an in which a unit increase in central equilibrium price is passed on fully in local prices. Notice that acceptance of the short-runrestrictions implies long-run market integrationbut that the reverse is not true. If the long-runmarketintegration restriction is accepted, then more efficient estimates of the remainingparametersand more powerful statisticaltests will be possible if the model is

This is a member of the class of error correlation models discussed by Hendry and Richard,and Hendry, Pagan, and Sargan.By interpretation,changes in local prices are attributedto changes in central prices and past spatial price differentials;the latter variables allow for the possibility that the marketsare not observed in an integratedequilibrium at a given point in time, and so there is feedback from prior disequilibria. (See Salmon for a control theoretic interpretationof error correction models.) There are a number of possible variations aroundthese three main hypotheses, particularly to do with the nonprice influences on each local market.The existence of significant localized marketcharacteristicsalso indicates that arbitrageis imperfectin eliminatingprice differentials.Thus, strongerintegration conditions can be formulatedby addingc = 0 to the above restrictions. I shall returnto this point in discussing the results for Bangladesh. One problemthat may arise when applying the above approachis multicollinearity among the regressorsof the unrestrictedmodel given by equation(4). The inferentialdifficultiesthat can arise are well known; a high standarderror on, for example, the coefficientfor central marketprice may be due to its highcorrelation with lagged local prices ratherthanweak market integration.However, such collinearityis a poor reason for adoptinga more restrictive model specification.The bias inducedby omitting relevant variables is well known. Also, the removal of such variables from a model can actually worsen the precision in estimation of the remaining regressorseven when the two sets of variablesare highly (albeit imperfectly) correlated(Davidson et al. give an exrestricample). Ratherthan impose parameter tions to reduce collinearity,a better approach in this setting is to first test for long-runinte-

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions

106 February 1986

Amer. J. Agr. Econ.

gration; since this restriction involves all of the price variables, a good deal of the collinearity danger is avoided (possible correlations with Xi remain). If long-run integration is accepted, then it should be imposed on the model with subsequent tests based on a restricted form such as equation (11), for which the collinearity problem is likely to be a good deal less severe. If long-run integration is rejected, then this avenue is best closed off and so one should apply extra caution in rejecting the short-run integration conditions.

sonably significant. The average price differentials also accord roughly with the distances to Dhaka, although their standard deviations are very high indeed. It is hard to believe that transport costs could vary so much. This leads one to doubt the existence of reasonably stable price differentials determined by transport costs, and so the picture given by the correlation coefficients may well be spurious. Three variables were selected as likely nonprice influences on local markets, aiming to capture the seasonality in rice production, any localized effects of the 1974 famine and any local time trend. The main (aman) harvest is from mid-November to early January and Spatial Price Differentials in Bangladesh 1972-75 seems to be quite adequately described by a single dummy variable for November and DeIn this section the general approach outlined cember. It is widely agreed that the worst above will be applied to monthly district-level months of the 1974 famine were August, Sepdata on rice prices (coarse quality) for Bangla- tember, and October; a single dummy variable desh in the post-independence period as moni- was included for these months. tored by the Bangladesh Directorate of AgAs has been noted, there will often be ricultural Marketing. A thirty-six-month series difficulties in identification of the parameters was used, from July 1972 to June 1975. This of equation (3). But equation (4) will generally period was chosen for its unusual price turbu- not pose such problems. The simultaneity in lence; there were numerous localized scar- the system can be dealt with easily using an cities (particularly during the famine year appropriate instrumental variables estimator 1974) and disturbances to trade and communi- such as the least squares estimate of (4) obcations (for a detailed economic history of the tained by replacing P1, by its predicted values period see Ravallion and van de Walle). Cer- from the reduced-form equation obtained by tainly this should be a hard (but important) using equation (4) to eliminate P1, (j = 2 ... test of the market integration restrictions. N) from (3). This is the method of estimation The demand for rice in Dhaka, the capital used here. Reduced-form estimates of P1, and (by far) largest city, would appear to be were obtained from a regression of this varithe dominant influence on interdistrict rice able against its own lagged values, the lagged trade in Bangladesh. And this is reflected in values of prices in all other districts and the the country's heavily monocentric transport seasonality, famine, and time trend variables. system. Aside from the old transport linkages All seventeen districts of Bangladesh in the to Calcutta in West Bengal (the previous colo- data set were used although (because of the nial capital of Bengal prior to partition in number of degrees of freedom needed) only 1947), the main road, rail and water routes are one-month lags were included. focused on Dhaka. The radial model seems The way transport charges are levied offers well suited to this setting. a guide to appropriate model specification in It was decided to concentrate on Dhaka's this setting. During the winter of 1983-84, I trade linkages with the main surplus districts asked a number of wholesale rice merchants at of its rural hinterland: Mymensingh, Bogra, Badamtali in Dhaka how the rice was transRangpur, Dinajpur, and Sylhet. During 1974 ported and how much it cost to do so. I was these five contiguous districts of northern told that by the main modes of rice freight Bangladesh had the five highest levels of food transport (road and ferry), the transport cost grain production per capita among the nine- for any trip largely depends on the number of teen districts of Bangladesh, while the neigh- trucks involved. And so, when comparing boring district of Dhaka had one of the lowest given markets, transport cost can be treated (rank 16), see Alamgir (table 6.29) and Sen as fixed per unit quantity rather than an (table 9.7) for details. Table I gives summary ad valorem charge.5 Thus, an equilibrium in data for these districts. Even taking first differences, the correlation coefficients between For example, transport cost from Rajshahi to Dhaka was reck- to be Taka 22-25 per maund. This is the cost of hiring a truck Dhaka and local prices are quite high and rea- oned

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions


Testing Market Integration


Table 1. SummaryData for Main Bangladesh SurplusDistricts,1972-75

District Mymensingh Distance to Dhaka Mean monthly price differential deviationof Standard price differential Simplecorrelationof of first Correlation
differences price with Dhaka

Bogra 150 20 32
.93 .57

Rangpur 200 19 44
.81 .43

Dinajpur 250 22 39
.87 .43

Sylhet 200 14 22
.96 .48

120 12 20
.97 .79

etc.) in Takaper maund,n = 36. (Dhaka-Mymensingh, Note: Distancesin road miles, price differentials

which price differentials depend solely on transportcosts would fix the unit price difference ratherthan, say, the percentage(log) difference. For, the econometric model to be at least consistent with the possibilityof such an equilibrium under linear parameter restrictions it should itself be linear in nominal prices. Any laggedeffects in the model are likely to arise from sluggishness in price adjustment, and expectations fordelays in transportation mation under price uncertainty. On a priori grounds, maximum lags greater than six months due to these causes seem highly implausible and also strain the data in terms of degrees of freedom. The unrestrictedmodel, equation (4), was estimated for n = 6 assuming an AR1 error. This gave an excellent overall fit and a reasonover six lags for ably flat residualcorrelogram each district.6ResidualvarianceF-tests were then appliedto the three mainrestrictedforms described in the previous section. The significance of the two postulated nonprice influences was also examined. Whenever the long-runintegrationrestriction was accepted the model was reestimated in the form of equation (11) and short-run integration was then retested. The test results are given in table 2. Summaryand Discussionof the Results To summarizethe results: (a) Marketsegmentationperformspoorly as a restrictedform of the general model for all districts.
for one day (Taka 4,000) divided by the quantity of rice carried (about 76 bags, at 2.25 maund per bag). 6 All residual correlograms comfortably passed the Box-Pierce test at the 5% level (for further discussion of this test see Harvey).

(b) At the other extreme, short-runmarket integrationwithin one month cannot be reasonably accepted for any districts except Sylhet and in this case only for the weaker form. For all otherdistricts,residualvariances are considerablyincreasedby assumingshortrun marketintegration,and so the assumption is hardto justify for these data.7 (c) The parameter restriction implied by long-runmarketintegrationperformsslightly better but is still difficultto accept for three of the five districts. Short-runintegrationconis tinues to be weak when long-runintegration imposed for the other two districts. (d) Local seasonalityis in evidence for four districts. For each of these, winter harvest prices were significantly lower than Dhaka prices. (e) The price series for all districts exhibit quite strong localized effects of the 1974 famine. In all cases, local prices were (ceteris paribus) significantly higher than Dhaka prices. Indeed, for two districts(Mymensingh and Rangpur)the price differential actually changed directionbrieflyduringthe famine. (f) There is evidence of a local (positive) time trend for two districts, Bogra and Rangpur. These results suggest some quite significant impedimentsto trade between Dhaka and its main ruralsupply areas, with the possible exception of the northeasterndistrict of Sylhet. The northwesterntrade corridorto Rangpur and Dinajpurseems highly restricted. While
7Nor do collinearity problems appear to be the reason. One check for this is to examine the significance of individual coefficients in the unrestricted model. Over all unrestricted equations, 14 of the 30 coefficients on lagged own price had absolute tratios over 3.0 while this was the case for 12 of the 35 coefficients on Dhaka price and its lagged values. Clearly, quite a high degree of resolution is possible in spite of multicollinearity. Full details of the unrestricted estimates of equation (4) are available from the author.

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions

108 February 1986

Amer. J. Agr. Econ.

Rice Markets of Bangladesh Table 2. Tests for the SpatialIntegration

District Hypothesis 1. Market segmentation 2. Short-run integration 3. Short-run integration (weak form) 4. Long-run integration 5. No local seasonality 6. No local famine 7. No local time trend Mymensingh 18 60 (52) 7.1 5.4* 13 11 2.3* Bogra 7.0 56 33 16 21 23 20 Rangpur 14 236 99 184 9.5 117 92 Dinajpur 13 63 25 32 12 31 .45* Sylhet 16 15 (12) 6.0* 9.1" .50* 16 8.4*

Notes: The unrestricted model is equation (4) for n = 6 with an AR 1 error estimated by maximum likelihood using the Beach-Mackinnon method for encorporating the stationarity condition into the likelihood function (see Beach and MacKinnon). The table gives F-tests of the linear restrictions on this model implied by each hypothesis. (1), (2), and (3) are distributed as F(7,13), F(13,13), and F(2,13), respectively, while the rest are F(1,13). All restrictions are rejected except * (at .01). The figures in parentheses in row 2 are the values of F when long-run integration is imposed.

these are relativelyremote areas, it is also nomarketintegratable how poorly the short-run tion restrictionsperformfor the much closer districtof Mymensingh. It should be emphasized that such test results do not reveal the sources of spatialprice differentials.In particular,it cannot be concluded that the markets are noncompetitive; the Bangladeshgovernmentis also known to exert considerable influence on the private grain trade. For example, having first observed that Dhakaprices were not fallingduring the 1983-84 winterharvestperiod, I asked a numberof the wholesale rice merchantsat Badamtaliwhy this was so. (This was done duringa related survey reportedin Ravallion 1985a.) I was told that government officials and police were cordoning rural markets to prevent trade with Dhaka. The government was thoughtto be doing this so that rice could be obtainedat the government'sprocurement price, then below the Dhakaprice allowingfor transportcosts.8 is only Of course, governmentprocurement trade. Most one of the risks facinginterdistrict years see quite heavy flooding in Bangladesh and in 1974it was very bad. This is certainto have added to transportcosts. Also, it adds the risk of losing all or part of the trader's consignmenten routeand, given thatrice traders in Bangladesh are unlikely to be able to insureagainstsuch a loss, this will also reduce expected returnsfrom trade at given prices.
8 This was later reported in the Bangla niewspaper Daily Ittefaq (12 Dec. 1983), although apparently the cordoning was not government policy at the time (The Bangladesh Observer, 28 Dec. 1983).

Conclusions A procedurefor testing marketintegration has been proposed which avoids the well-known inferentialdangersof received methods using spatialprice correlations.The price series for each local marketis permittedto have its own autoregressivestructureand a dynamic relationshipwith marketprices in a tradingregion. Alternativehypotheses concerningthe extent of integrationcan then be encompassed and tested as restricted forms of the general model. The new method also extracts more informationon the natureof spatialprice differentials from the same data. In particular, the dynamicapproachpermitsa clear distinction between short-run market integration, and integrationas a long-runtendency in the short-runadjustmentprocess. An application of this approachto monthlyrice price datafor deBangladeshsuggests some quite significant partures from the conditions for both shortand long-run market integration. And these are not revealed by tests using static correlations.
[Received March 1984; final revision received January 1985.]

Alamgir, M. Famine in South Asia: Political Economy of Mass Starvation. Cambridge MA: Oelgeshlager, Gunn, and Hain, 1980. Ambirajan, S. Classical Political Economy and British Policy in India. Cambridge: Cambridge University Press, 1978. Beach, C. M., and J. G. MacKinnon. "A Maximum

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions


Testing Market Integration


LikelihoodProcedurefor Regressionwith AutocorrelatedErrors."Econometrica46(1978):51-58.

Bhatia, B. M. Famines in India: A Study of Some Aspects of the Economic History oflndia (1860-1965). Lon-

Analysis of Economic Time Series." Int. Statist. Rev. 51(1983):111-63. Hendry,D. F., A. R. Pagan,andJ. D. Sargan."Dynamic
Specification." Handbook of Econometrics, vol. 2,

don: Asia PublishingHouse, 1967. as a Measureof MarBlyn, G. "Price Series Correlation

ket Integration." Indian J. Agr. Econ. 28(1973):5659.

ed. Z. Grilichesand M. D. Intriligator. Amsterdam: North-Holland PublishingCo., 1984. Hurd, J. "Railwaysand the Expansionof Marketsin India, 1861-1921." Explorations in Econ. Hist.

12(1975):263-88. Davidson, J. E. H., D. F. Hendry, F. Srba, and S. Yeo. "Econometric Modelling of the Aggregate Time Newbery, D., and J. Stiglitz. "Pareto InferiorTrade." Rev. Econ. Stud. 51(1984):1-12. Series RelationshipBetween Consumers'Expenditureand Incomein the U.K." Econ. J. 88(1978):661- Ravallion, M. Markets and Famines. Oxford: Oxford 92. UniversityPress, forthcoming. Efficiencyof Traders'Price Fase, M. M. G. "The Linkageof Stock ExchangeMar. "The Informational kets BetweenCountries."Econ. Letters7(1981):363Expectationsin a BangladeshRice Market."Oxford 69. Bull. Econ. and Statist. 47(1985a):171-84. of Rice Marketsin Bangladesh . "The Performance George, P. S. "Some Aspects of Public Distributionof Duringthe 1974Famine."Econ. J. 92(1985b):15-29. Foodgrains in India." Econ. and Polit. Weekly Ex-ante Ravallion,M., and D. van de Walle. "Measuring 19(1984):A106-10. Information: Whatthe NewspapersSaid Duringthe Granger,C. W. J., and P. Newbold. "SpuriousRegres1974 Famine in Bangladesh." Rev. Public Data Use sion in Econometrics."J. Econometrics 1112(1974): 20. 12(1984):169-84. Harriss, B. "There Is Method in My Madness:Or Is It Salmon, M. "ErrorCorrelationMechanisms."Econ. J. Vice Versa? MeasuringAgricultural 92(1982):615-29. MarketPerformance." Food Res. Inst. Stud. 17(1979):197-218. Seaman, J., and J. Holt. "Marketsand Famines in the ThirdWorld."Disasters 4(1980):283-97. Harvey, A. C. The Econometric Analysis of Time Series. Oxford:PhilipAllan, 1981. Sen, A. K. Poverty and Famines: an Essay on Entitlement and Deprivation. Oxford: Oxford University Haugh, L. D. "Checking the Independence of Two Press, 1981. Time-Series: A Univariate Covariance-Stationary Residual Cross-CorrelationApproach." J. Amer. Takayama,T., and G. G. Judge. Spatial and Temporal
Statist. Assoc. 71(1976):378-85. Price and Allocation Models. Amsterdam: North-

Hendry, D. F., and J-F. Richard. "The Econometric

HollandPublishing Co., 1971.

This content downloaded from on Fri, 30 Aug 2013 01:34:27 AM All use subject to JSTOR Terms and Conditions