Is There Really a Border Effect?

Is There Really a Border Effect? A.K.M. Mahbub Morshed *a Department of Economcs, Southern Illnos Unversty, Carbondale, IL 62901 Abstract: The observed excess prce varablty n cross-border cty pars compared to that n wthncountry cty pars has been defned as the border effect. We used a unque data set from ctes that were n the same country at one tme and were n two separate countres later on to examne the effects of the presence of a natonal border on prce varablty. Interestngly a border-le effect was detected even durng the perod when all the ctes were n the same country. We also found a large border effect when ctes were n two separate countres. However, we found no change n the prce varablty at crossborder cty pars durng the perods both before and after the ctes separated nto two dfferent countres. Ths fndng suggests that the observed systematc hgher varablty of consumer prces n cross-border cty pars mght not be due to the presence of a border as suggested n the lterature. JEL Classfcaton: F40, F41. Keywords : Border effect; Law of one prce. * Department of Economcs, Malcode 4515 Southern Illnos Unversty, Carbondale, IL 62901. Tel. No. (618) 453 5740, Fax: (618) 453 2717, Emal: mmorshed@su.edu a I would le to than Santanu Chatterjee, Scott Glbert, Susan He, and partcpants of the Mdwest Economcs Assocaton Conference 2003 and Southern Economc Assocaton Conference 2003 for valuable suggestons. The constructve suggestons of an anonymous referee are also gratefully acnowledged. 1

I. Introducton In the context of nternatonal trade, the Law of One Prce has receved more attenton lately. Usng U.S.A. and Canadan cty prce data, Engel and Rogers (1996) showed that the standard devaton of the relatve prce n U.S. and Canadan ctes s systematcally hgher for cross-border cty pars than for cty pars wthn a country. For example, the prce varablty between Detrot and Toronto s much hgher than between Detrot and San Francsco or between Toronto and Vancouver. Other researchers provded more support to these results wth data from other developed countres (Goldberg and Verboven, 2001; Parsley and We, 2001; Hasel and Wolf, 2001) and developng countres (Morshed, 2003). What causes ths falure of the Law of One Prce n the short-run remans an unresolved ssue. Although nomnal exchange rates and transport costs have been found to be sgnfcant n determnng cross-border prce varablty, stll the observed prce varablty n cross-border cty pars can not be explaned by these factors alone. Researchers have suggested that the presence of a natonal border mght be the source of ths excess prce varablty n cross-border cty pars. Ths excess varablty has been coned as the border effect. In order to dentfy the sources of ths border effect, some researchers have examned the role of prces of nontraded goods and servces (Engel and Rogers, 1996; Morshed, 2003), others have examned socal varables such as language (Engel and Rogers, 2000), whle another group of researchers has looed nto geographc features (Parsley and We, 2001). Nonetheless, all these attempts can explan only a small porton of ths large and sgnfcant border effect. Researchers generally follow a wth or wthout a border framewor to calculate border effects where they compare the prce varablty at the cty pars located n dfferent countres to that at the cty pars located wthn a country. A wth or wthout a border framewor certanly seems mportant n understandng the dfferences n prce varablty, yet to say that the presence 2

of a natonal border causes ths ncreased prce varablty seems oversmplfed. In order to solate the effects of the creaton of a natonal border, we beleve that a complementary before and after framewor s warranted where data from ctes that were parts of the same country for a perod and became parts of dfferent countres later on are examned. Generally, t s very dffcult to get data wth these propertes, but the recent hstory of the Indan Subcontnent, n general, and Bangladesh and Pastan, n partcular, opens up an opportunty to conduct a natural experment wth a before and after framewor. Bangladesh (then East Pastan) and Pastan (then West Pastan) were two parts of the same country (Pastan) durng 1947-1971. Bangladesh became an ndependent naton n 1971. Ths brea-up of a country nto two ndependent natons allowed us to examne the effects of a new border on prce varablty. Accordngly, we have collected prce data at fve ctes each n both Bangladesh and Pastan for both the pre-1971 and the post-1971 perods. Ths unque data set wll allow us to apprase the border effect more rgorously. The geographc non-contguty of Bangladesh and Pastan wll not hnder our analyss so long as they trade goods and servces 1. Prevously, data from two physcally dsjont countres have been used to estmate the border effect. For example, Parsley and We (2001) estmated the sze of the border effect between the U.S.A. and Japan. Other researchers have used a large cross-country dataset that ncludes data from non-contguous countres (Goldberg and Verboven, 2001; Hasel and Wolf, 2001). 1 Tradng between these two regons of the country was very hgh durng the pre-1971 perod. Export-mport data were reported generally for the whole country (both East Pastan and West Pastan) durng the pre-1971 perod. However, there are also some nter-wng trade data (from East Pastan to West Pastan and vce versa). Import of East Pastan products va Karach port (the man seaport n West Pastan) and mport of West Pastan products va Chttagong port (the man seaport n East Pastan) together was about 22.9% of the total export and mport of Pastan durng the 1950s and 1960s. West Pastan s mports from East Pastan ncluded, among other commodtes, pulses, fruts and vegetables, and chles, whle East Pastan s mports from West Pastan ncluded, among other commodtes, flour, gram, mustard ol, ghee, gur, and soap. After 1971, the amount of tradng between Bangladesh and Pastan declned sgnfcantly. Stll, Pastan s an mportant tradng partner of Bangladesh. In 1991, Bangladesh mported 1.7% of ts total mports from Pastan and exported 2.3% of ts total exports to Pastan. 3

Generally the border effect s estmated by the border dummy coeffcent n a regresson equaton where prce varablty has been regressed on log of dstance, a border dummy, and other relevant varables. In ths paper, we have followed the same wth or wthout a border technque wth an assumed border between ctes n Bangladesh and Pastan even durng the pre-1971 perod. Interestngly enough, we found that ths assumed border s hghly sgnfcant for a number of commodtes. As expected, we also found a hghly sgnfcant border coeffcent for the post-1971 perod. These estmated larger border coeffcents wth better precson (larger t-statstcs) from the post-1971 data certanly ndcate the presence of a border effect under a wth or wthout a border framewor. But n a before and after framewor, prce varablty at only cross-border cty pars durng both the pre- and post-1971 perods were evaluated. Our results show that for a number of commodtes the ndependence of Bangladesh (creaton of a border) dd not rase the prce varablty at cross-border cty pars. The present paper conssts of fve sectons. In secton II, we dscuss the basc feature of the dataset and why ths dataset s sutable to apprase the border effect n a before and after framewor. We dscuss methodology n secton III. Results are reported n secton IV. At the end, some concludng remars are made. II. Data and Data Sources We have compled a three-dmensonal panel dataset consstng of annual retal prce for 14 traded goods at 10 dfferent ctes n Bangladesh and Pastan (fve ctes from each country). The pre-1971 data ranges from 1950-1971, whle the post-1971 data s for the perod 1975-1993. Fve Bangladesh ctes are Dhaa, Chttagong, Narayangan Sadpur, and Sylhet, 2 whle Karach, Lahore, Peshwar, Rawalpnd, and Salot are Pastan ctes. We nclude the followng 2 Data from Sadpur are not avalable for post-1971 Bangladesh, but we have data for Rangpur. Snce Sadpur and Rangpur ctes are from the same dstrct and not far from each other, we consder these two as the same cty for our purpose. 4

traded goods: flour, moogh (pulse), gram, beef, mutton, ml, ghee (processed butter), potatoes, mustard ol, onon, gur (molasses), chles, erosene ol, and washng soap. These are major consumer tems n Bangladesh and Pastan. Wth the excepton of erosene ol, all these commodtes are produced n both countres. Ths s a very unque data set for at least three reasons. Frst, we have data for a number of ctes that were part of the same country for a tme and then later were n two dfferent countres. Ths enables us to drectly determne how much the ndependence of Bangladesh, essentally the creaton of a natonal border, changes the prce varablty n cross-border ctes. Second, unle many other studes that use prce ndces, we use actual prce data. The Law of One Prce seems more naturally related to the prce of a partcular commodty than to a prce ndex. Thrd, unle cross-country studes, the qualty of the data s not compromsed by the use of dfferent datasets from dfferent sources. The same Statstcal Bureau s the source for the pre- 1971 data for both countres and the smlarty of the structure of the statstcal bureaus and the data defntons eep the potental varablty to a mnmum for the post-1971 perod. The pre-1971 prce data for all ctes were collected from 25 Years of Pastan n Statstcs 1947-72, a 1972 specal publcaton of the Central Statstcal Offce of Pastan. For post-1971 Bangladesh data, we consulted varous ssues of Monthly Statstcal Bulletn of Bangladesh, publshed by the Bangladesh Bureau of Statstcs. Post-1971 Pastan data were collected from 50 Years of Pastan n Statstcs (Volume IV) 1947-1997, a specal publcaton of the Federal Bureau of Statstcs of the Government of Pastan. We obtaned the lattude and longtude of dfferent ctes from Unted Natons Statstcs on the web at http://www.un.org/depts/unsd/demog/. We also consulted Oxford Atlas of the World, Nnth Edton, for lattude and longtude data for a few ctes. Usng the How Far Is It? 5

webste (http://www.ndo.com/dstance/), we calculated the great crcle dstance between the ctes n mles. III. Methods We, le Engel and Rogers (1996), defne P, to be the log of the prce of good n j locaton j relatve to the prce of good n locaton. For each cty par (), the standard devaton of p = P t) P ( t 1), over all relevant years t, s the measure of prce j, ( varablty. The dfference n relatve prce was taen to avod the ssue of nonstatonarty of data. As there are 10 ctes n our analyss, we have 45 cty pars; of these 45 cty pars, 25 cty pars are such that one of the ctes s n a dfferent country. We conducted our analyss based on the cross secton of these prce varablty measures. Lnear regresson technques were used to assess the mportance of, among other varables, the presence of border and the dstances between the ctes. We estmated the followng equaton usng the Ordnary Least Squares (OLS) method: n S ( p ( t)) = 1r + β2b + γ mdm + u m= 1 β (1) where S( p j, ( t) ) s the sample standard devaton of p, ( t) j, j r, s the log of the dstance measured n mles between locatons, B j, s the cross-country dummy, and m D s the dummy varable for each cty ncluded n the regresson. The country dummy varable B, denotes j whether locaton j and le n dfferent countres. For example, for the relatve prce varablty between Dhaa and Chttagong (two ctes n Bangladesh), the value of ths dummy s 0, whereas for the relatve prce varablty for ctes le Dhaa and Karach (a Pastan cty) the dummy 6

varable taes the value 1. The cty dummy 3 ncludes the cty m. D m for the cty m taes a value 1 when the cty par In order to determne the effects of the ndependence of Bangladesh, we estmated an equaton wth an ndependence dummy along wth cross terms wth other varables as regressors. Accordngly, we staced prce varablty data for both the pre- and post-1971 perods and estmated the followng equaton: S( p β 2 * B ( t)) = α + γ 2 1 + α 2 Independence + β * Independence* B + 1 * r + γ 1 r 9 9 δ mdm + m= 1 m= 1 * Independence + ψ m Independence* D m + u ( t) (2) where ndependence taes a value 1 for post-1971 data, 0 otherwse. If the creaton of a border does ncrease the relatve prce varablty, we can expect γ 2 to be postve and hghly sgnfcant. Only nne cty dummy varables and ther nteracton terms were ncluded to avod the dummy varable trap. Another way to examne the effects of the creaton of a new border s to chec the dfference, f any, between the prce varablty before 1971 and after 1971 only n cross-border ctes. To ths end, only cross-border varablty measures were staced, and we estmated the followng equaton for the staced data for cross border cty pars: S ( p 1 2 3 t ( t)) = φ + φ * r + φ * Independence + ν ( ) (3) In ths case, we expect φ 3 to be postve and sgnfcant. Snce dstances between cross-border cty pars are dfferent and ths dstance nfluences the cross-border prce varablty, we have ncluded log of dstance as an addtonal regressor n our estmaton. 3 Engel and Rogers (1996) suggested a number of reasons n favor of ncluson of the cty dummes, namely that there may be dosyncratc error n some cty data that maes ther prces more volatle on average, measurement error may exst, and the varablty n one cty may be hgh for some reasons not modeled here. 7

IV. Results IV. A. Border and Prce Varablty To proceed, we calculated exchange-rate-free prces n order to exclude the effects on our retal prces of the fluctuatons of nomnal exchange rates. Followng Engel and Rogers (1996), all prces were dvded by the prce of a dfferent good n the same cty to obtan exchange-ratefree prces. For example, the prce of potatoes n Dhaa relatve to the prce of flour n Dhaa s used as the real prce for potatoes. The same has been done for Karach. Ths relatve prce, for example: Prce Prce Pr ce Pr ce Potatoes Dhaa, t Flour Dhaat, Potatoes Karacht, Flour Karacht,, s a nomnal exchange-rate-free prce. If the volatle nomnal exchange rate explans all of the border effect, the border coeffcent should not be sgnfcant n a regresson (usng equaton 1) wth real relatve prces constructed n ths way. Ths s how the effects of dfferent exchange rate regmes and the presence of nomnal exchange rates are neutralzed. Average prce varablty at cty pars wthn a country and at cty pars n dfferent countres s shown for the pre-1971 perod n Fgure 1 and for the post-1971 perod n Fgure 2. The prce varablty measures n cross-border cty pars were found to be hgher for all commodtes durng both the pre- and post-1971 perods. We observed the lowest average prce varablty measures for washng soap, beef, mustard ol, and onon n Bangladesh ctes durng the pre-1971 perod. For the post-1971 perod, Pastan ctes yelded the lowest prce varablty measures for all commodtes except washng soap, gur, and erosene ol (Fgure 2). It becomes clear from a comparson of Fgure 1 and Fgure 2 that the average prce varablty n 8

Bangladesh ctes declned for all commodtes durng the post-1971 perod. Wth the excepton of gur and erosene ol, the same s true for Pastan. The prce varablty at wthn-a-country cty pars declned sgnfcantly after the regons were separated nto two ndependent countres. Fgure 1: Average Prce Varablty n Wthn a Country and Cross-border Locatons Durng the Pre- 1971 Perod (all prces are relatve to the prce of flour) 0.700 Average Prce Varablty 0.600 0.500 0.400 0.300 0.200 Bangladesh Pastan Cross-border 0.100 0.000 Washng Gram Mutton Moogh Beef Ml Ghee Mustol Potato Onon Gur Chlles Kerosene Commodtes In a typcal wth or wthout a border framewor, we compared the three seres n each of Fgures 1 and 2. It s evdent from these fgures that the prce varablty at cross-border cty pars was much hgher compared to that at wthn-a-country cty pars durng both the pre- and post-1971 perods. Moreover, the dfferences were much more pronounced durng the post-1971 perod (Fgure 2). Ths yelds larger border coeffcents and larger t-statstcs (see Table 1). Thus, the exstng methodology (wth or wthout a border) yelds a large border effect. However, for our proposed complementary experment (before and after a border) we needed to examne what has happened to the prce varablty only at cross-border cty pars durng the pre- and post-1971 perods. To ths end, average prce varablty estmates for only cross-border cty pars durng both the pre- and post-1971 perods are shown n Fgure 3. It s nterestng to note that the prce 9

varablty at cross-border locatons dd not have any jump durng the post-1971 perod, for more than half of the commodtes. Fgure 2: Average Prce Varablty n Wthn a Country and Cross-border Ctes Durng the Post- 1971 Perod (all prces are relatve to the prce of flour) 0.700 0.600 Average Prce Varablty 0.500 0.400 0.300 0.200 Bangladesh Pastan Cross-border 0.100 0.000 Washng Gram Mutton Moogh Beef Ml Ghee Mustol Potato Onon Gur Chlles Kerosene Commodtes 10

Fgure 3: Average Prce Varablty for Cross-border Cty Prces n Both Pre- and Post-1971 Perods (all prces are relatve to the prce of flour) 0.700 Average Prce Varablty 0.600 0.500 0.400 0.300 0.200 Post-1971 prces Pre-1971 prces 0.100 0.000 Washng Gram Mutton Moogh Beef Ml Ghee Mustol Commodtes Potato Onon Gur Chlles Kerosene In addton to ths we observed lower prce varablty durng the post-1971 perod (Fgure 3) for moogh, beef, ml, ghee, mustard ol, onon, gur, and washng soap. Only gram, potatoes, and chles yelded hgher prce varablty n the post-1971 perod. These results mply that the creaton of a border between these two regons does not necessarly lead to an ncrease n the average prce varablty at the cross-border cty pars. We have estmated equaton (1) and also have dealt wth some varatons of dstance functons. In the specfcaton I, the log of the dstance and the border dummy were ncluded along wth cty dummy varables for each cty. In Specfcaton II, dstance, dstance squared, border dummy, and cty dummes are ncluded, whle n Specfcaton III, both rght- and lefthand sde varables n equaton (1) are dvded by log (dstance). Border coeffcents and ther t- statstcs for both the pre- and post-1971 perods are reported n Table 1. 11

We found sgnfcant and postve border coeffcents even for the perod when there was no border between any ctes for (column 2) beef, ml, ghee, mustard ol, onon, gur, and erosene ol. For the post-1971 perod, except for moogh, beef, ghee, potatoes, and erosene ol under Specfcaton II, all border coeffcents are found to be postve and hghly sgnfcant. We found larger border coeffcents and larger t-statstcs from the post-1971 data. In the context of a wth or wthout a border framewor, these results suggest that the creaton of a border does ncrease the extent of the dfference between the prce varablty at cross-border cty pars and at wthn-a-country cty pars. However, whether these ncreases n the border coeffcents and correspondng large t-statstcs have been the result of the creaton of a border needs further nvestgaton. Snce the average varablty of prces at wthn-a-country cty pars declned after 1971 for both Bangladesh and Pastan ctes and the average prce varablty remaned almost the same for the cross-border cty pars (Fgure 3), the ncluson of the same rght-hand sde varables n our regressons (for example, equaton 1) for the post-1971 data yelds larger border coeffcents. The border dummy now has essentally pced up the addtonal dfferentals of prce varablty between wthn-country and cross-border cty pars. Thus, the ncrease n sze and precson of the estmates of a border dummy n the post-1971 regressons does not necessarly mply that the presence of a border has generated such a dfference. 12

Table 1 Border Coeffcents under Dfferent Specfcatons Dependent Varable: Prce Varablty (all prces are relatve to the prce of flour) Commodty Pre-1971 Data Post-1971 Data Specfcaton I Specfcaton II Specfcaton III Specfcaton I Specfcaton II Moogh 0.051 0.198 0.022 0.124 0.077 (1.840) (3.108) (0.888) (5.416) (1.287) Gram 0.043 0.322 0.042 0.160 0.139 (1.540) (4.803) (1.437) (7.959) (2.859) Beef 0.055 0.161 0.045 0.072 0.030 (2.802) (3.565) (2.415) (4.126) (0.561) Mutton 0.022 0.128 0.020 0.100 0.084 (1.442) (3.444) (1.301) (6.598) (2.434) Ml 0.081 0.203 0.074 0.101 0.148 (4.748) (4.045) (4.861) (10.592) (4.048) Ghee 0.045 0.072 0.044 0.066 0.042 (3.905) (2.199) (3.575) (5.608) (0.921) Mustard Ol 0.067 0.014 0.054 0.074 0.157 (2.804) (0.303) (2.969) (3.681) (5.262) Potato 0.038 0.132 0.055 0.216 0.123 (1.807) (2.287) (2.723) (9.086) (1.479) Onon 0.304 0.538 0.283 0.325 0.206 (6.813) (5.149) (7.981) (21.182) (3.294) Gur 0.171 0.300 0.181 0.090 0.032 (5.002) (3.170) (6.999) (3.251) (0.539) Chles 0.017-0.011-0.009 0.437 0.553 (0.424) (-0.111) (-0.270) (14.335) (8.776) Kerosene 0.033-0.007 0.028 0.065-0.045 Ol (2.383) (-0.281) (2.375) (4.227) (-1.214) Washng 0.012 0.152 0.020 0.078 0.112 Soap (0.470) (2.973) (0.924) (4.709) (2.345) Specfcaton III 0.102 (5.946) 0.150 (9.764) 0.064 (4.390) 0.090 (7.901) 0.094 (10.350) 0.052 (4.756) 0.055 (3.414) 0.219 (7.602) 0.331 (21.951) 0.084 (3.725) 0.411 (18.213) 0.075 (5.642) 0.067 (4.641) Note: We have reported only the border coeffcents. There are 45 observatons for each of the pre- and post-1971 perods. Whte s heterosedastcty-consstent standard errors were computed and t-statstcs are reported n parentheses. In order to examne the effect of the creaton of a natonal border we estmated equaton (2), whch ncludes an ndependence dummy and other cross-terms. A small part of the results from these regressons s reported n Table 2. We expected that the coeffcent for the cross-term n equaton (2), γ 2, would be postve and hghly sgnfcant f the creaton of a new border matters. However, we found nsgnfcant coeffcents for more than half of the 13 commodtes (beef, ml, ghee, mustard ol, onon, gur, and erosene ol). The γ 2 coeffcent was found to be 13

negatve but nsgnfcant for gur. Thus, the observed hghly sgnfcant border dummy n post- 1971 wth large t-statstc does not necessarly represent a large border effect. Table 2 Regresson Results wth Cross-terms as Addtonal Regressors (Equaton (2)) Dependent Varable: Prce Varablty (all prces are relatve to the prce of flour) Commodty Dstance Border Border*Independence Adjusted R 2 Moogh 0.011 0.051 0.072 0.88 (0.803) (1.840) (2.017) Gram 0.025 0.043 0.118 0.91 (2.173) (1.540) (3.422) Beef 0.004 0.055 0.017 0.91 (0.389) (2.802) (0.658) Mutton 0.018 0.022 0.078 0.90 (2.585) (1.442) (3.663) Ml -0.008 0.081 0.021 0.89 (-0.973) (4.748) (1.056) Ghee 0.016 0.045 0.021 0.93 (3.159) (3.905) (1.271) Mustard Ol 0.008 0.067 0.007 0.94 (0.698) (2.804) (0.225) Potato 0.025 0.038 0.178 0.88 (2.623) (1.807) (5.629) Onon -0.006 0.304 0.021 0.96 (-0.304) (6.813) (0.450) Gur 0.008 0.171-0.082 0.88 (0.468) (5.002) (-1.859) Chles 0.054 0.017 0.419 0.96 (2.961) (0.425) (8.224) Kerosene Ol 0.011 0.033 0.032 0.94 (1.359) (2.383) (1.528) Washng Soap 0.019 (1.490) 0.012 (0.470) 0.066 (2.245) 0.87 Note: Whte s heterosedastcty-consstent standard errors are computed and t-statstcs are reported n parentheses. There are 90 observatons n these regressons. Shaded values represent sgnfcant coeffcents. IV. B. Prce Varablty at Cross-border Ctes and the Independence of Bangladesh We then staced the prce varablty at only cross-border cty pars for both the pre- and post-1971 perods. Results of the OLS estmaton of equaton (3) are reported n Table 3. It s nterestng to note that only for gram, potatoes and chles do we observe a postve and sgnfcant dummy. Negatve but hghly sgnfcant ndependence coeffcents for mustard ol, onon, gur, moogh, and washng soap suggest that the creaton of a natonal border reduces the 14

cross-border prce varablty a negatve border effect ndeed. We also found negatve but nsgnfcant ndependence coeffcents for beef, ml, ghee, and potatoes, whle for mutton and erosene ol, postve but nsgnfcant ndependence coeffcents were found. Ths experment ndcates that the creaton of a natonal border mght not create any addtonal prce varablty at cross-border cty pars. Table 3 Effects of Independence on the Cross-Border Prce Varablty Dependent Varable: Only Cross-border Prce Varablty (all prces are relatve to the prce of flour) Commodty Independence Log (Dstance) R 2 Moogh -0.046 0.0009 0.15 (-2.907) (0.015) Gram 0.058-0.161 0.34 (4.009) (-2.200) Beef -0.016 0.007 0.02 (-1.066) (0.111) Mutton 0.007-0.060 0.09 (0.894) (-1.542) Ml -0.022-0.137 0.15 (-1.533) (-2.469) Ghee -0.010-0.136 0.13 (-0.722) (-2.644) Mustard Ol -0.123 0.213 0.73 (-10.370) (4.756) Potato 0.041-0.029 0.11 (2.336) (-0.443) Onon -0.079-0.154 0.55 (-6.830) (-2.962) Gur -0.148-0.184 0.68 (-9.551) (-3.628) Chles 0.281 0.008 0.93 (24.384) (0.169) Kerosene Ol 0.008 0.021 0.02 (0.721) (0.399) Washng Soap -0.062 (-4.522) -0.174 (-2.920) 0.40 Note: Whte s heterosedastcty-consstent standard errors are computed and t-statstcs are reported n parentheses. IV. C. What Explans the Border Effect? In order to understand why the presence of a natonal border does not ncrease the crossborder prce varablty, we need to as the reverse queston: Why do we expect that the 15

exstence of a natonal border wll ncrease the cross-border prces varablty n the frst place? The prevalng wsdom s that the presence of arbtrage opportunty n the maret would tend to equalze prces through tradng of goods and servces. If the ctes are n dfferent countres, we expect more frcton n tradng, resultng n an ncrease n prce varablty at cross-border cty pars. However, we dd not observe any jump n prce varablty at cross-border cty pars after the creaton of a natonal border between Bangladesh and Pastan n 1971. Even lower trade volume after 1971 along wth the presence of a natonal border dd not rase the prce varablty at cross-border cty pars. These results ndcate that some degree of maret segmentaton may perpetuate even wthn a country. V. Conclusons Observed sgnfcant and systematcally hgher prce varablty at cross-border ctes were comparable to what s found for ctes located n the same country seems perplexng. These results hold even after the contrbutons of nomnal exchange rates and transport costs are accounted for. To examne whether the presence of a natonal border s responsble for ths or not, we collected prce data from a number of ctes that were part of the same country (Pastan) for a perod of tme and part of a dfferent country (Pastan and Bangladesh) later. We conducted a natural experment usng ths unque data. We found larger prce varaton at crossborder ctes than that n wthn-a-country cty pars, both before and after the ndependence of Bangladesh. However, we found no sgnfcant dfference n the prce varablty at cross-border cty pars for ths poltcal change. Ths suggests that the conventonal border dummy approach may have detected somethng other than the effects of a border. In the case of the natonal border between Bangladesh and Pastan, t seems that for a number of commodtes there s no sgnfcant border effect. We understand that t s hard to generalze these results, yet t suggests that the unfcaton of countres s not necessarly a recpe for quc convergence of prce 16

varablty, whch mght be a valuable pece of nformaton n the context of polcy mang n the European Unon. References Engel, Charles, and John H. Rogers (1996). How Wde Is the Border? Amercan Economc Revew, December, 86(5), 1112-1125. Engel, Charles, and John H. Rogers (2000). Relatve Prce Volatlty: What Role Does the Border Play? n Intranatonal Macroeconomcs, Gregory Hess and Erc van Wncoop, eds., (Cambrdge Unversty Press), 92-111. Goldberg, Pnelop K., and Fran Verboven (2001). The Evoluton of Prce Dsperson n the European Car Maret, Revew of Economc Studes, 68 (4), 811-48. Hasel, Jonathan, and Holger Wolf (2001). The Law of One Prce: A Case Study, Scandnavan Journal of Economcs 103 (4), 545-58. Morshed, AKM Mahbub (2003). What Can We Learn From a Large Border Effect n Developng Countres? Journal of Development Economcs, 72, 353-369. Parsley, Davd C., and Shang-Jn We (2001). Explanng The Border Effect: The Role of Exchange Rate Varablty, Shppng Costs, and Geography, Journal of Internatonal Economcs, 55, 87-105. Whte, Halbert (1980). A Heterosedastcty Consstent Covarance Matrx Estmator and a Drect Test for Heterosedastcty, Econometrca, 48(4), 817-38. 17