BORDER EFFECTS BETWEEN U.S. AND MEXICO

Transcription

1 JOURNAL OF ECONOMIC DEVELOPMENT 53 Volume 31, Number 1, June 2006 BORDER EFFECTS BETWEEN U.S. AND MEXICO FERNANDO BORRAZ Unversdad de Montevdeo I use dsaggregated good prces data for U.S. and Mexcan ctes for nne categores of consumer prces to estmate the border effect on U.S. - Mexcan relatve prce and fnd that for all the categores of goods t s an order of magntude larger than for U.S. - Canadan prces (13.22 versus 2.25 for the CPI) However, durng a very stable perod n Mexco (May 1988 to November 1994), the wdth of the border falls dramatcally for all the categores of goods analyzed (3.22 versus for the CPI). Keywords: Exchange Rates, Dsaggregated Good Prces, Border Effect JEL classfcaton: F30, F40, F41 1. INTRODUCTION Aganst a backdrop of macroeconomc nstablty and nequtable ncome dstrbuton, n the md-1980s Mexco adopted an aggressve lberalzaton strategy that has closely ntegrated ts economy to the global market, and that of the Unted States n partcular, especally after enterng the North Amercan Free Trade Agreement (NAFTA) n As a result, both trade and nvestment flows ncreased substantally throughout the 1990s and the share of employment lnked to the world economy rose consderably. NAFTA consoldated the lberalzaton of the Mexcan economy and opened up the Canadan and U.S. markets to Mexcan producers. The mmedate result of the rapd dsmantlng of trade barrers has been a remarkable ncrease n Mexco s world trade. From 1990 to 2000, total mports and exports ncreased more than fourfold, reachng U.S. $174 and U.S. $166 bllon, respectvely. Whle the share of mports comng from North Amerca has remaned constant at around 75 percent, Mexco s exports to the regon went from 80 to 91 percent of all exports from 1990 to As a result, Mexcan producers now face fercer competton n ther home market and send a greater fracton of ther output overseas. I thank Andrea Battur for excellent research assstance. I thank John Rogers and Vctor Pacharon for very helpful comments. All remanng errors are my own.

2 54 FERNANDO BORRAZ Just as Mexco became more open to trade flows, the country revamped ts legal framework regulatng foregn nvestment, substantally openng the Mexcan economy as a whole to foregn nvestors. Whereas FDI remaned below one percent of GDP over the perod, t rose to between one and two percent from 1986 to 1993, and t has remaned above two percent snce However, current research has revealed that consumer prces are not nearly as equalzed across countres as one would antcpate n a hghly ntegrated regon. The emprcal studes fnd large departures from the law of one prce. In partcular, Engel and Rogers (1996) analyze the varablty of the relatve prces of smlar goods, usng consumer prce data from 23 ctes n U.S. and Canada. They concluded that relatve prce varablty s postve and sgnfcantly related to dstance between ctes. However, takng nto account the effect of dstance, relatve prce varablty s remarkably larger for ctes that le across an nternatonal border than for ctes that le wthn ether country. Engel and Rogers (1996) name ths the wdth of the border. There are several possble explanatons for ths large border effect n prces. These nclude barrers to nternatonal trade, such as tarffs, quotas, and non-tarff barrers such as bureaucratc red tape mposed on foregn busness, the presence of non-tradable goods and servces, labor markets that are more ntegrated and homogenous wthn countres than they are across countres, and dstrbutons networks. An alternatve explanaton reles on nomnal exchange rate varablty wth stcky fnal goods prces n local currency. When the nomnal exchange rate fluctuates so do relatve goods prce for cross border ctes pars. Rogers and Smth (2001) apply the Engle and Rogers (1996) methodology to quantfy the U.S. - Mexcan border. They show that the border effect between U.S. and Mexco prces s larger than that for the U.S. and Canadan prces. Therefore, there s a relatvely larger border effect on U.S. - Mexcan prces. However, they use aggregate prce data. In a related work, Engel, Rogers and Wang (2003) use actual prce data to quantfy a larger border effect between U.S. and Canada. They use very dsaggregated data from the Economst Intellgence Unt. The man drawback s that there s a large degree of measurement error n these unoffcal prces. Because I want to analyze the nature of devatons from the law of one prce and the U.S. and Mexcan consumer prce ndexes are constructed somewhat dfferently 1 and the basket weghts are not the same n any event, I wll study devatons from the law of one prce for smlar categores of goods and not for the general CPI. The man contrbuton of ths paper s to extend Rogers and Smth (2001) usng dsaggregated offcal consumer prce ndexes (and wth less measurement error) from ctes 1 The comparsons of CPI ndexes between countres rase the followng problems: how to handle the ntroducton of new goods, shftng consumpton weghts wthn a country, etc.

3 BORDER EFFECTS BETWEEN U.S. AND MEXICO 55 n U.S. and Mexco. In ths paper I use consumer prces ndexes for several goods from Mexcan and U.S. ctes to estmate border effects on relatve prce between Mexco and U.S. Also, I analyze the border durng the subperod May 1988 to November 1994 known as El Pacto Perod. Durng ths perod the peso-usd exchange rate was qute stable. 2. DATA I use consumer prce data for 14 ctes n USA and 14 ctes n Mexco for dsaggregated consumer prce ndexes. The data cover the perod from January 1984 to December The sample ends n 1997 because the U.S.. Bureau of Labor Statstcs changed the Consumer Prce Index base n Table 1 lsts the good n the study. Gven that the data s very mportant to understand the contents of ths paper I explan t n detal. Table 1. Categores of Good n Dsaggregated Consumer Prce Indexes Used Good Unted States Mexco 1 Food at Home Food purchased from stores 2 Food away from home Restaurants and cafeteras 3 Footwear Footwear 4 Shelter Shelter 5 Fuel and other utltes Electrcty and fuel 6 Household furnshngs and operatons Furnture and home equpment 7 Medcal care Medcal care 8 Publc Transportaton Publc Transportaton 9 Prvate Transportaton Transportaton by your own The data from Mexco was obtaned from the Bank of Mexco ( The data from USA was obtaned from the Bureau of Labor Statstcs (stats.bls.gov) All of the prce data for both countres are seasonally unadjusted. The data for the exchange rate was obtaned also from the Bank of Mexco. The dstance n mles between ctes was download from I work wth the shortest drvng dstance between ctes. The 14 Mexcan ctes are Acapulco, Chhuahua, Guadalajara, Hermosllo, Juarez, Matamoros, Merda, Mexcal, Méxco Dstrto Federal, Monterrey, Tampco, Tjuana, Veracruz, Vllahermosa. For all of these ctes the prce data are monthly. Monthly prce data for U.S. are avalable only for four core ctes: New York, Phladelpha, Chcago and Los Angeles. In addton for fve ctes, data are released n odd numbered month: Boston, Mam, St. Lous, Washngton DC and Baltmore. For

4 56 FERNANDO BORRAZ fve other ctes data are avalable n even numbered months: Dallas, Houston, Pttsburgh, Detrot and San Francsco. For each good I calculate the log of the ntercty ( j, k) relatve prce, that s Log[ CPIt ( j) / CPIt ( k)], where t denotes tme. Because I apply ths formula, I only calculate the relatve prces for ctes that report the CPI n the same tme t. In order to calculate the volatlty of changes n prces I compute the standard devaton of the 6-month changes n the log of the relatve prce (Standard Devaton of Log[ CPIt+ 6( j) / CPIt+ 6( k)] Log[ CPIt ( j) / CPIt ( k)]. Thus, when I use only the Mexcan ctes and the core U.S. ctes, for each good there are 153 ntercty prces ( 18*17 / 2). Addng the fve even month U.S. ctes adds another 100 prces ( 18*5 + 5* 4 / 2), and addng the fve odd month U.S. ctes adds another 100 prces. The total number of cty pars for each good s 353. Note that I don t match the U.S. ctes whose data s reported even month wth those ctes that reported data odd month. Also note that I have 91 cty pars for Mexcan ctes ( 13*12 / 2), 66 for U.S. ( 4*3/ 2 + 2*(4*5 + 5* 4 / 2)), and 196 cross border par ctes ( 14*14) Summary Statstcs Table 2 reports summary statstcs. For each of the 9 goods and the Consumer Prce Index I report the average standard devaton for pars of ctes that are: (a) both n USA, (b) both n Mexco, and (c) one n each country. Table 2 reveals that the volatlty of prces between Mexcan ctes s hgher than that between U.S. ctes pars. 2 The volatlty of prces between U.S. ctes are , close to the estmaton of Rogers and Smth (2001) of For the Mexcan ctes I obtaned a volatlty of , aganst ther estmaton of Ths may ndcate that t s more costly to transport goods between Mexcan ctes than between equ-dstant ctes wthn the U.S.. However, cross border ctes pars have much hgher volatlty. Note that for the CPI the cross border volatlty s ten tmes hgher that the volatlty between Mexcan ctes. However for the categores of goods analyzed all cross border volatlty ranges only from 1.5 to 5 tmes the volatlty of Mexcan ctes. Note for the dstance between ctes that the nter-natonal cty pars are on average slghtly farther apart than are the ntra-natonal pars. 2 The only excepton s n the footwear, where the volatlty s hgher n U.S. par ctes.

5 BORDER EFFECTS BETWEEN U.S. AND MEXICO 57 Table 2. Average Prce Volatlty for 6 Month Change Cty Pars Good US US MX MX US MX CPI Dstance(mles) In table 3 I present summary statstcs for the volatlty of relatve prces n El Pacto Perod. 3 Durng that perod of stablty n the Mexcan economy the volatlty of prces s lower than for the full perod. Also, the volatlty of ctes pars across borders s near one thrd of the volatlty between 1984 and Table 3. Average Prce Volatlty durng El Pacto Perod Cty pars Good US US MX MX US MX CPI From May 1988 to November 1994.

6 58 FERNANDO BORRAZ Table 4 shows that the volatlty of the exchange rate drops substantally durng El Pacto Perod. Durng ths perod the volatlty of the Mexcan nflaton decreases substantally, n conjuncton wth an overall mprovement n macroeconomc and fnancal condtons n Mexco. Table 4. Exchange Rate Volatlty for 6 Month Changes 1984: : : : Note: The columns dsplay the mean value of the standard devaton of changes n the nomnal exchange rate. 3. ORDINARY LEAST SQUARES ESTIMATION Followng Engel and Rogers (1996), my regressons attempt to explan the standard devaton of P( j, k). Engel and Rogers (1996) hypothesze that the volatlty of the prces of smlar goods sold n dfferent locatons s related to the dstance between the locatons and other explanatory varables, ncludng a dummy varable Border, for whether the ctes are n dfferent countres. In partcular I estmate the followng cross-secton regresson: where V P ( j, k)) = β r + β B + α( m) D + ε, (1) ( 1 j, k 2 j, k m j, k r j, k s the log of the dstance between locatons. I expect β 1 to be postve, as the varaton n transport cots should be larger the greater the dstance between locatons. 4 As n the gravty model of trade I expect a postve relatonshp between relatveprce volatlty and dstance. Countres are more lkely to trade wth neghbors because the transportaton costs are lower. I also nclude a dummy varable n the equaton for each cty n the sample, That s, for cty par ( k) D m. j, the dummy varables for cty j and k take on value of 1. The ncluson of separate dummes for each cty allows the standard devaton of prces to vary from cty to cty. There are a few reasons why I allow the level of standard devaton to vary from cty to cty. Frst, there may be dosyncratc measurement error n some ctes that make ther prces more volatle on average. Second, there may be dfferences across methodologes for gatherng data. Thrd, for the ctes that report prces only bmonthly, there may be addtonal volatlty that s ntroduced by measurement errors from the less frequent observaton of prces. Fourth, an emprcal 4 Note that dstance proxes for more than just transportaton costs.

7 BORDER EFFECTS BETWEEN U.S. AND MEXICO 59 motvaton comes from Table 2, whch ndcates greater relatve prce volatlty for Mexcan ctes than U.S. ctes. B, the border varable, s a dummy varable for whether locatons j and k are j k n dfferent countres. I expected the border to be postve, and I am nterested n comparng the wdth of the U.S. - Mexcan border to the U.S. - Canadan border estmated n Rogers and Smth (2001). Table 5 shows the OLS results whle consderng the standard devaton of the 6-month changes n prces as the dependent varables and the log of the dstance between ctes and the border varable as ndependent varables. Table 5. OLS Regressons Good Log Dstance Border Adj. R (0.27) (0.35) (0.11) (0.26) (1.70) (0.23) (0.34) (0.44) (0.31) (0.39) ) (0.16) (0.20) (0.18) (0.24) (0.18) (0.24) (0.14) (0.19) CPI (0.16) (0.20) Notes: The sample perod s 1984:1-1997:12 for the 28 ctes. Heteroscedastcty consstent standard errors are n parenthess (Whte (1980)). The dependent varable s the standard devaton of the 6-month Changes n the log relatve prce. The ndependent varables are: the log of the dstance between ctes n the partcular par (mles) Border whch equals unty f the ctes n the par le across an nternatonal border. The estmaton ncludes 28 ctes dummes. Coeffcents and standard errors have been multpled by 100.

8 60 FERNANDO BORRAZ The coeffcent on the log dstance s postve for the 9 goods, but t s sgnfcant at the 5% level for only 3 goods. 5 The coeffcent on the dummy varable for the border s of hypotheszed sgn and hghly sgnfcant for all 9 goods. The nterpretaton of the coeffcent on the border dummy n ths regresson s the dfference between the average standard devaton of prces for cty pars that le across the border less the average for pars that le wthn one of the two countres, takng nto account the effect of dstance. For the CPI, the coeffcent on the dummy Border s slghtly hgher than the value of 9.76 found by Rogers and Smth (2001). I have confrmed the fndng of Rogers and Smth (2001) of a postve large and sgnfcant border effect usng dsaggregated consumer prce data. Ths mean, that the border effect s sgnfcant not also for the CPI but for dfferent categores of goods. It s nterestng to explore the cross-sectonal evdence n more detals. In partcular, some of the tmes (shelter) are nontraded, others (publc transportaton) are nontraded and perhaps subject to regulaton, yet others (footwear) are traded. These features would be expected to provde a useful way of dfferentatng the cost of arbtrage from the exchange rate volatlty explanaton. In the Table 5 the larger border effect s observed for tradable and nontradable goods; however dstance s n general nsgnfcant n explanng prce varablty among tradable. Therefore, the evdence suggests that an mportant part of the larger border effect s explaned by nomnal exchange rate varablty wth stck fnal good prces. The most remarkable fndng n Table 5 s that for all the categores of goods the border effect t s of an order of magntude larger than for U.S. - Canadan prces Sub-Sample: El Pacto Perod El Pacto Perod runs from May 1988 to November The man feature of ths perod s the overall mprovement n the macroeconomc condtons n Mexco. Trough ths perod the standard devaton of the peso-usd was smlar to the standard devaton of the Canadan Dolar-USD. Durng the stable peso perod we see that the coeffcents on the border dummy are notably smaller than n the entre sample perod. For the CPI the dummy varable falls from to For the CPI, the coeffcent on the dummy Border s slghtly hgher than the value of 2.55 found by Rogers and Smth (2001). 5 Wthn countres, the log dstance s sgnfcant and postve for the goods 1,3,4,5,8,9 and CPI n Mexco, and for the goods 1,3,8,9 n U.S..

9 BORDER EFFECTS BETWEEN U.S. AND MEXICO 61 Table 6. OLS Regressons El Pacto Perod Good Log Dstance Border Adj. R (0.07) (0.11) (0.08) (0.12) (0.10) (0.14) (0.18) (0.24) (0.38) (0.47) (0.09) (0.13) (0.08) (0.10) (0.19) (0.23) (0.15) (0.22) CPI (0.06) (0.09) Notes: The sample perod s 1988:5-1994:11 for the 28 ctes. Heteroscedastcty consstent standard errors are n parenthess Whte (1980) The dependent varable s the standard devaton of the 6-month change n the log relatve prce. The ndependent varables are: the log of the dstance between ctes n the partcular par (mles), Border whch equals unty f the ctes n the par le across an nternatonal border. The estmaton ncludes 28 ctes dummes. Coeffcents and standard errors have been multpled by CONCLUSIONS I use consumer prces ndexes for several goods from ctes n USA and Mexco, to quantty the extent to whch prces fals to equalze across countres aganst a baselne of the sze of the falure across regons wthn countres. That s, I compute the Enge-Rogers wdth of the border measures for Mexco. Focus on 6-month changes and over the sample perod , I show that the border effect n U.S. - Mexcan prces s an order of magntude larger than that obtaned by Engel and Rogers (1996) for U.S. - Canadan. In addton, durng El Pacto Perod the border decreases dramatcally and becomes only slghtly hgher than the U.S. - Canada border.

10 62 FERNANDO BORRAZ The major concluson of the paper s that the border effects between U.S. and Mexco matter for relatve prce varablty for several categores of smlar goods. That s, I extend the Rogers and Smth (2001) result that valdate the border effect only consderng CPI. Also the evdence suggests that an mportant part of the larger border effect s explaned by nomnal exchange rate varablty wth stck fnal good prces. REFERENCES Engel, C., and J. Rogers (1996), How Wde s the Border? Amercan Economc Revew, 86, Engel, C., J. Rogers, and S. Wang (2003), Revstng the Border: An Assessment of the Law of One Prce Usng Very Dsaggregated Consumer Prce Data, Internatonal Fnance Dscusson Papers, Number 777, Board of Governors of the Federal Reserve System. Mark, N. (2001), Internatonal Macroeconomcs and Fnances: Theory and Emprcal Methods, Blackwell Publshers. Obstetlf and Rogoff (1996), Foundatons of Internatonal Macroeconomcs, The MIT Press. Rogers, J., and H. Smth (2001), Border Effects Wthn the NAFTA Countres, Internatonal Fnance Dscusson Papers, Number 698. Board of Governors of the Federal Reserve System. Whte, H. (1980), A Heteroskedastcty-Consstent Covarance Matrx Estmator and a Drect Test for Heteroskedastcty, Econometrca, 48(4), May Malng Address: Unversdad de Montevdeo, Prudenco de Pena 2440, CP: Montevdeo, Uruguay. Tel: , Fax: E-mal: fborraz@um.edu.uy. Manuscrpt receved February, 2005; fnal revson receved May, 2006.