Urban road congestion, capacity expansion and port competition: empirical analysis of U.S. container ports

Transcription

1 Urban road congeston, capacty expanson and port competton: emprcal analyss of U.S. contaner ports Yula Wan a,, Anmng Zhang a, Andrew C.L. Yuen b a. Sauder School of Busness, Unversty of Brtsh Columba 2053 Man Mall, Vancouver, BC, Canada V6T 1Z2 yula.wan@sauder.ubc.ca; anmng.zhang@sauder.ubc.ca b. Department of Decson Scences and Manageral Economcs, Faculty of Busness Admnstraton, The Chnese Unversty of Hong Kong, Shatn, NT, Hong Kong andrewyuen@baf.msmal.cuhk.edu.hk November 2011 Abstract: In ths study, we emprcally nvestgate the mpacts of urban road congeston and road capacty expanson on the competton between major contaner ports n the U.S. We fnd that more delays on urban roads may cause shppers to swtch to competng rval ports: a 1% ncrease n road congeston delays around the port s assocated wth a % decrease n the port s contaner throughput but a % ncrease n the rval port s throughput. Addng local roads tends to beneft the port and harm ts rval (n terms of throughput) by reducng road congeston. However, the overall mpact of road provson on ports throughput vares among the sample ports, as road capacty expanson may affect ports output through channels other than road congeston delays. Keywords: contaner port competton; road congeston; road capacty; ntermodal transport chan

2 1. Introducton Gateway seaports play an mportant role n the ntermodal transportaton network because they connect the foreland where the contaners are moved by shppng lnes and the hnterland where trucks or a combnaton of ral and trucks delver contaners to consgnees or collect contaners from shppers. Snce the 1950s, contanerzaton has mproved the effcency n ntermodal transportaton and ntensfed competton among gateway seaports. Therefore, a port can hardly mantan ts local monopoly poston by havng a large captve hnterland (e.g. [1, 2]). Furthermore, the competton s no longer just between ndvdual ports but between a number of alternate ntermodal chans (e.g. [3, 4]). In the ntermodal shppng chans, the nland leg has receved ncreasng attenton from ndustry players, government authortes and academa, as nland logstcs accounts for 40-80% of total contaner shppng costs [5]. Seaports wth more hnterland transport nfrastructure are more lkely to survve n newly establshed trade flow market [6]. As ponted out by Notteboom and Rodrgue [7], the performance of seaports s strongly ntertwned wth the development and performance of ther respectve nland networks. It s found that the shppers ocean carrer and/or port selecton process has been heavly affected by the landsde operaton attrbutes, such as land carrer servce qualty and proxmty to the hnterland [8-11]. The rapd growth of nternatonal trade (and resultng contaner shppng) has mposed tremendous pressure on the ntermodal transportaton system. As ndcated by Heaver [12], the bottleneck of such ntermodal chans has shfted from the shp/port nterface to the port/nland nterface. In the Unted States, for example, the lmted capacty of the hghway system resulted n an nablty to wthstand demand shocks, whch caused congeston on key freght transport segments [13]. Road congeston delays rase travelng tmes and fuel costs, lower the relablty of commercal truck operatons and ncrease the chance of mssng schedules [14]. All of these 1

3 factors could translate nto costs endured by shppers who select the ntermodal chan to shp ther cargoes. 1 As a large proporton of the contaners are moved nto and out of seaports by truckng, the congeston at urban roads surroundng the seaports has become one of the essental factors that nfluence a port s ablty to sustan ts compettveness. Addng road capacty s one of the most common and mportant road congeston mtgaton strateges [14]. 2 The survey conducted by Golob and Regan [15] found that addng more freeway lanes s strongly supported by U.S.-based commercal truckng frms, especally short haulers connectng martme ports and ral termnals. Port managers n the U.S., especally those on the west coast and the east coast, also consder local road capacty as the top constrant that mpedes port development and growth (e.g. [16]). Ths paper examnes urban road congeston, road capacty expanson and port competton n the context of major contaner ports n the U.S. More specfcally, the paper nvestgates the followng two questons: () whether an ncrease n road congeston reduces the contaner throughput of the port afflated to the urban area whle beneftng the rval ports nearby; and () how road capacty changes n both the port s and ts rval s respectve urban areas wll affect the port s contaner throughput. These questons are partcularly mportant when polcy makers consder expandng road capacty to allevate congeston so as to accommodate more port-related traffc and hence enhance compettveness of the port n ther jursdctons. For example, a number of major UK road mprovement schemes have been carred out n order to beneft ports around ths area [17]. De Borger, et al. [18] and Wan and Zhang [19] nvestgated these questons wth analytcal models, and t turns out that the 1 TTI [13] estmated that n 2009, the average (per urban area) congeston costs suffered by trucks n the U.S. amounted to 1,273 mllon US dollars and 215 mllon US dollars for very large (populaton over 3 mllon) and large (populaton over 1 mllon but less than 3 mllon) urban areas, respectvely. 2 Other congeston mtgaton strateges nclude: mprovng operatonal effcency, provdng truck dedcated facltes, mprovng traffc management, and managng demand (e.g. through congeston prcng). For more detaled dscusson, see FHWA [14] and Golob and Regan [15]. 2

4 answers to the questons are n general ambguous, dependng on assumptons about how ports compete and other factors. 3 Thus, t s of nterest to examne emprcally the mpacts of road congeston, as well as the effect of road capacty expanson on the port competton. We fnd that port throughput s related to urban road congeston and capacty nvestment. Specfcally, a 1% ncrease n road congeston s assocated wth % reducton n the contaner throughput of the seaport afflated wth that urban area, whle mplyng an ncrease n the throughput of ts rval port by %. Further, the mpact of road capacty expanson on contaner throughput dffers among the ports n our sample, dependng n partcular on road capactes around a port and ts rval ports. Despte the dfferental mpacts of road capacty expanson, addng local road capacty tends to beneft the port n the urban area n queston and harm ts rvals when such road expanson solely affects ports throughput va road congeston. Emprcal work whch nvestgated both port competton and hnterland characterstcs s relatvely rare. Most of the studes predcted the dstrbuton of contaner traffc volumes among competng ports wth numercal smulatons and/or multnomal logt model based ether on the total delvery costs or on the dstance to access the seaport from hnterland (e.g. [1], [22-25]). None of those studes looked nto the mpact of road capacty or road congeston delays on port competton. For example, Nr, et al. [22] consdered hghway travel tme to carry the contaners from shppers faclty to contaner ports as a factor that nfluences shppers port-choce decson. They found a strong negatve mpact of the hghway travel tme on the probablty that a certan port wll be selected by shppers. However, snce the dstance s not controlled, t s unclear whether the hghway travel tme should be consdered as an approxmate of dstance or 3 Other theoretcal studes n ths strand of the lterature nclude Yuen et al. [20] and Zhang [21]. The former nvestgates the effects of congeston prcng mplemented at a gateway port on ts hnterland s optmal road prcng, road congeston and socal welfare, whereas the latter focuses on the corrdor congeston and capacty nvestment rather than urban road congeston per se. 3

5 congeston. 4 On the other hand, the mpact of road capacty provson on the road traffc volume (combnng both commercal trucks and commuter cars) has been extensvely studed (e.g. see [27, 28] for useful revews), but ths lterature focuses on local resdents and commuter cars and does not pay any partcular attenton to port ctes and port-related actvtes. The paper s organzed as follows. In Secton 2, we brefly go through the theory regardng to the connecton between road congeston, capacty and port competton derved n the lterature. Then, we ntroduce our econometrc models n Secton 3, followed by the descrpton of the dataset n Secton 4. Secton 5 presents the major emprcal results and Secton 6 provdes the concludng remarks. 2. Road congeston, capacty and port rvalry: theory The theoretcal predcton of the mpact of road capacty on port competton rests manly on two contrbutons: De Borger, et al. [18] assumed that duopoly seaports compete n port charges (.e. a case of prce competton), whle Wan and Zhang [19] assumed quantty competton between the rval ports, albet employng the smlar model settngs as n De Borger, et al. [18]. In essence, both papers consdered two rval ports competng for the same foreland and hnterland. A partcular port faces an nverse demand, ( X, Xj), whch s a functon of ts own and ts rval port j s throughputs (.e. X and X j, respectvely). Inbound cargoes arrvng at the port wll be trucked through congestble urban roads around the port before beng shpped to the hnterland. These urban roads are shared by both local commuters and commercal trucks. Therefore, the shppers average congeston delay cost due to travelng on the roads around port,.e. D ( V, K ), s determned by the road capacty ( K ) and the total road traffc volume 4 Pope, et al. [26] assessed the mpact of addng a new secton of hghway on road congeston around the port of Hampton Roads by a smulaton model, but port competton was excluded from ther analyss. 4

6 ( V X Y, where Y denotes the local commuter traffc volume). The generalzed cost for shppers equals the sum of the port charge and the average cost of congeston delay per unt of cargo,.e. p DV (, K ), where p s the port charge, and that value for local commuters equals the congeston delay cost encountered by commuters,.e. DV (, K ) 5. 5 Note that s assumed to be less than one, meanng that commuters average congeston cost s a fracton of shppers average congeston cost. At the equlbrum, the shppers (commuters ) generalzed cost equals the margnal shpper s (commuter s) wllngness to pay. That s, X, X ) p DV (, K), Y) DV (, K), 1, 2 ( j L(, (1) where ρ L ( Y ) s the nverse demand functon for commuters usng roads around port. Road congeston and port throughput affect each other smultaneously. On the one hand, when port throughput X ncreases, t s straghtforward to show that the overall traffc V wll ncrease (though the local traffc volume Y wll reduce as the trucks compete wth commuters for road space), resultng n more congeston delays on the road. On the other hand, when the road congeston around port ncreases, the generalzed cost of usng port ncreases, and as a result, the contaner throughput of port tends to decrease, whle that of port j tends to ncrease. The mpacts of road capacty on contaner throughputs, however, depend on the mode of competton between ports. Under quantty competton, the two ports smultaneously choose ther own port contaner throughputs to maxmze ther respectve profts. An ncrease n K wll have three types of effects on port s margnal proft: 1) a postve mpact on the port charge, owng to a drect congeston reducton from road capacty expanson takng the traffc volume as fxed, 2) a negatve mpact on the port charge because of an ncrease n congeston, as road 5 For smplcty, we normalze the vehcle operatng costs, such as fuel costs and road tolls, to be zero. Ths smplfcaton does not change the man results of ths model. See Wan and Zhang [19] for the full verson of ths model.

7 capacty expanson nduces more local commuters to travel, and 3) a combnaton of mpacts on the margnal change of congeston delay,.e. the second-order effects. If the net mpact on the margnal proft s postve, ncreasng K wll lead to port ncreasng throughput at each level of port j s throughput, thereby shftng port s reacton functon outward. However, an ncrease n K does not affect port j s margnal proft and hence port j s reacton functon remans unchanged. Because quanttes are strategc substtutes, the reacton functons are downward slopng, and thus when port s reacton functon shfts outward, port s throughput ncreases, whle port j s throughput decreases. When demand and congeston delay functons are all lnear, Wan and Zhang [19] proved that the net mpact on the margnal proft s postve, and, therefore, a port s throughput tends to ncrease when the road capacty nearby ncreases (.e. dx */ dk 0 dx * j / dk 0). ) 6 but decrease when the road capacty around the rval port ncreases (.e. Under prce competton (where the two ports smultaneously choose ther own port charges to maxmze ther respectve profts), however, the mpacts of road capacty on ports throughputs are n general ambguous. Note that n such a case, port charges are strategc complements and thus the reacton functons are upward slopng (see fgure 1). As both K j enter the reacton functons, both ports reacton functons wll shft f demand functons and congeston delay functons are all lnear, an ncrease n K and K changes. If K wll lead to the reacton functon of port shftng outward and that of port j shftng nward. De Borger, et al. [18] proved that at the equlbrum an ncrease n K wll reduce port j s charge whle rasng 6 Note that the superscrpt * denotes the equlbrum outcomes. 6

8 port s charge f port s output level s low. 7 The mpact on port s equlbrum throughput s gven by dx* X dk p X dk dp * p j dp* j X K K. (2) It can be shown that the second term on the rght-hand sde (RHS) of equaton (2) s negatve, whle the thrd term, the drect mpact of capacty expanson on throughput, s postve. If the capacty ncrease leads to an ncrease n port charge, then the frst term wll be negatve as well. 8 Accordngly, the overall mpact of an ncrease n road capacty on the port s throughput s undetermned. The mpact on the rval s equlbrum throughput s gven by dx* j X j dk p X dk dp * p j j dp* j X K K j. (3) All the three terms on the RHS of equaton (3) have sgns opposte to those n equaton (2). Agan, an ncrease n port s road capacty has ambguous mpact on ts rval port s contaner throughput. [Insert fgure 1 here] 3. Econometrc specfcaton The rest of the paper addresses two questons. Frst, how does urban road congeston affect contaner throughput? In partcular, we emprcally test whether road congeston n the urban area near a port wll reduce the port s contaner throughput, whle ncreasng ts rval s throughput. Second, how does the change n urban road capacty affect the competng ports contaner throughputs? Although the mpact of road capacty change on throughput s 7 De Borger, et al. [18] argued that t s qute plausble to have port s charge ncrease n road capacty K. 8 De Borger, et al. [18] showed that X p 0, X p j 0, X K 0, and X j K 0. 7

9 analytcally ambguous under a prce competton model, as dscussed n the prevous secton, the quantty competton settng suggests that road capacty expanson s lkely to beneft the port nearby and harm ts rval. On the other hand, prevous surveys [15, 16] found that both commercal truckng frms and port managers beleve that ncreasng road capacty s an effectve soluton to road congeston, whch, n turn, facltates port development. Gven these survey fndngs and the ambguous results n analytcal studes, we emprcally nvestgate whether an ncrease n road capacty n an urban area wll have a postve mpact on the contaner throughput of the port nearby and a negatve mpact on that of the rval port. Our estmaton s based on a panel dataset that ncludes eleven major contaner port areas n the U.S. from 1982 to To examne the relatonshp between urban road congeston and contaner throughput, we estmate the followng model: where lnx t 0 1lnT t 2lnP t 3lnD t 4lnDRt u t, (4) X t = the contaner throughput for port n year t, T t = the value of U.S. nternatonal trade n year t, P t = the populaton n the catchment area of port n year t, D t = the road congeston delay n the urban area around port n year t, DR t = the road congeston delay n the urban area of port s rval port n year t, u = the fxed effect for port, t = the error term for port n year t. Note that we apply a log-log regresson model n order to reduce the magntude of heteroskedastcty whch s lkely to be an ssue n cross-sectonal data [29]. Besdes, the 8

10 coeffcents n (4) can subsequently be drectly nterpreted as elastctes. It s expected that 1 s postve because martme traffc s manly derved from nternatonal trade, and, therefore, contaner throughput should be postvely correlated to the nternatonal trade volume. As the populaton affects the consumpton of mported goods and the capacty of producng export goods, ports wth larger populatons n ther own catchment areas are lkely to have hgher contaner throughputs. Thus, 2 s expected to be postve, as well. In fact, the contaner throughputs for the selected U.S. ports have an ncreasng trend, as do other ndependent varables, as the economes and the shppng demand have a growng trend n general. Therefore, by ncludng the value of U.S. nternatonal trade and the populaton n the port s catchment area as the ndependent varables, we control the ncrease n contaner throughput owng to the economc growth and the shppng demand ncrease. Moreover, as mentoned n Secton 2, an ncrease n contaner throughput wll ncrease the severty of road congeston around the port, whle congested roads may redrect shppers to the rval port. Thus, we may have an endogenety problem: D t s lkely to be correlated wth the error term and wll lead to nconsstent estmaton. We deal wth ths problem through nstrumental varables: the urban populaton ( pop) t and area ( area), t or the urban populaton densty ( dens) t whch s a rato of popand t area. t Roads n urban areas whch accommodate larger populatons or populaton denstes are lkely to be more congested. Besdes, people who lve n larger urban areas may need to travel longer dstances, thereby creatng more traffc on the roads. Thus, road congeston should be correlated wth those nstrumental varables. Conversely, because contemporary contaner ports serve destnatons deep nto the hnterland nstead of beng confned to the mmedate urban area around the ports, such factors as the populaton densty of the assocated urban area s less lkely to affect contaner throughput 9

11 drectly. Therefore, the abovementoned nstrumental varables are legtmate. As long as the endogenety problem s properly resolved, we expect 3 to be negatve and 4 to be postve. Recallng the theoretcal model, because road capacty s the only exogenous varable n the theoretcal model, we are nterested n how the port s equlbrum throughput wll change as the road capacty changes. To examne the relatonshp between urban road capacty and contaner throughput, we estmate the followng model: where lnx t 0 1 lnt t 2 lnp t 3 lnlmt 4 lnlmr t lnlmt 6 lnlmr t 7 lnlmt lnlmr t v t LM t = the road capacty n the urban area near port n year t,, (5) LMR= t the road capacty n the urban area near the port s rval n year t, v = the fxed effect for port, t = the error term for port n year t. Both De Borger, et al. [18] and Wan and Zhang [19] showed that the port s equlbrum throughput s a functon of the port s own road capacty and ts rval s road capacty, but the functonal form depends on the structures of the demand functons and the congeston cost functon. In general, the mpacts of road capacty on throughput are non-lnear and the road capacty elastctes are also not constant but rather are functons of capactes, as well. Wthout an explct functonal form, to capture the feature that the road capacty elastctes may depend on LM t and LMR, t we apply the second-order Taylor expanson around the pont at whch both lnlmtand ln LMR t are zero. Consequently, the coeffcents 3 and 4 only reflect the elastctes at the (0, 0) pont, whch does not have a practcal meanng. Instead, we need to estmate the elastctes wthn the range of the observed ponts. The own-road capacty elastcty 10

12 and rval-road capacty elastcty can be estmated by takng partal dervatves of ln LMR t respectvely. That s, X 3 2 lnlm lnlmr t t ln ln 5 7 LMt X t, and ln LMt and t 2 lnlmr t lnlmt ln ln (6) LMR t In the theoretcal models, t s assumed that road capacty affects ports throughputs only through the change of road congeston. However, n realty, road capacty expanson may affect the ports competton through other channels. For example, holdng the road congeston unchanged, an ncrease n urban road capacty may mply an mprovement n the accessblty to the hnterland by, for nstance, havng shorter routes and thus attractng more shppers. On the other hand, an ncrease n urban road supply may suggest an mprovement n the connectvty to the rval port nearby, consequently ntensfyng the port competton. Furthermore, an nvestment n urban road capacty may be accompaned by an nvestment n other transport nfrastructure n and outsde of the urban area, reducng the cost of usng other transportaton modes, such as ar, ral and long-dstance truckng. As a result, local frms may change ther sourcng destnatons, as well as transportaton modes, and these factors may eventually affect the demands of martme shppng. Therefore, t s of nterest to dstngush the effect of road capacty nvestment through road congeston from other channels. In vew of ths, we consder the model that controls the road congeston whle estmatng the mpact of road capacty nvestment through other factors: where lnx t 0 1lnT t 2lnP t 3lnD t 4lnDRt 5lnLMt 6lnLMR lnlmt 8 lnlmr t 9 lnlmt lnlmr t t = the fxed effect for port, and t = the error term for port n year t. 11 t, (7)

13 4. Data Our emprcal study s based on the panel data consstng of eleven U.S. ports, and the observatons are taken from 1982 to 2009 annually. 9 Four of the eleven ports are on the west coast, namely, Seattle-Tacoma (SeaTac), Portland, Oakland and Los Angeles-Long Beach (LALB). 10 The other seven ports are on the east coast: Boston, New York-New Jersey (NYNJ), Baltmore, Hampton Roads, Charleston, Jacksonvlle and Mam. Among these ports, SeaTac and LALB are conceptual aggregatons of two ports located n neghborng ctes. For example, the Port of Los Angeles and the Port of Long Beach are consdered to be one port, despte ther separate managements, because these two ports are located close to each other such that the traffc from/to the two ports probably uses the same road system. Thus, t s napproprate to treat the two ports as rvals nstead of a sngle port, as n the present study, we look at the mpacts of road system on ther throughputs. On the other hand, ctes n whch the two ports are located are conventonally pooled nto the same urban area n the datasets we used for urban road congeston and road capacty. For the same reasons, we also combne the Port of Seattle and the Port of Tacoma as a sngle port. For each port, we consder ts rval port beng the one n our sample wth the shortest dstance to that partcular port. Table 1 gves the lst of the ports and ther respectve rvals. Competton between ports through the rvalry between alternate ntermodal chans served by commercal trucks s most lkely to be affected by road congeston and relevant to the context of ths study. Among the ports on the same coast, an overlap of hnterlands served by trucks s most lkely to occur between ports located closest to each other, because trucks are often used to 9 By lookng at the U.S. ports, we can focus on the nteracton between the port throughout and the road system. As the Jones Act lmts feeder servces between U.S. ports, U.S. ports have lttle transshpment and have a strong nland orentaton [30], and thus ther throughputs are more lkely to be affected by port access condtons. 10 These ports mght also compete wth Canadan west-coast ports, namely, Vancouver and n the future, Prnce Rupert. 12

14 transport short-to-medan-haul cargos. Thus, we only consder the closest port as the rval n our study. On the other hand, Luo and Grgalunas [1] argued that competton between the westcoast ports and the east-coast ports for contaner traffc mght be present, due to U.S. ntercontnental ralways. In effect, the west coast and the east coast are probably competng for the Amerca-bound Asan cargos destned for coastal and nland regons n the east of the contnent and lower Oho Valley [6]. However, urban road congeston barely affects such competton drectly, because these cargos arrve at the east coast ether drectly through the allwater route va the Panama Canal or by beng transted by double-stack trans from the west coast. Snce our focus s on local road condtons and port competton, competton between ports of the two coasts, f any, s less relevant n our dscusson. [Insert table 1 here] The contaner throughput data (n 1,000 TEUs) are taken from the Amercan Assocaton of Port Authortes. As the present study nvestgates the mpact of road systems on contaner throughput, the throughput data should be adjusted by subtractng the amount transported nto and out of the nteror regons by trans and/or nland waterway. However, as the modal splt ratos for nland transportaton at each seaport are not avalable, we approxmate the throughput transported by truckng usng the data provded by the U.S. Commodty Flow Surveys. These surveys report the tons of cargoes shpped nto and out of a certan urban area by transportaton modes. Based on ths nformaton, we calculate the fracton of cargoes shpped by trucks. 11 We multply ths fracton wth the contaner throughput to estmate the number of TEU s transported by trucks, whch s the dependent varable ( X t) n the econometrc models. 11 Unfortunately, ths seres of surveys s only conducted once every 4-5 years, and we have data for 1993, 1997, 2002 and 2007, but our dataset has observatons from 1982 to Therefore, we assgn the fracton calculated from the 1993 survey to all perods n , the 1997 fracton to all perods n , the 2002 fracton to all perods n and the 2007 fracton to the remanng years. 13

15 The data for the value of U.S. annual nternatonal goods trade ( T) t are from the U.S. Census Bureau. To decde the populaton n the catchment area ( P), t we frst need to decde for each port the relevant catchment area n whch truckng plays a major role n transportng contaners between the catchment area and the port. Based on the data provded by the U.S. Commodty Flow Surveys, among all the cargoes moved by trucks, those cargoes shpped wthn 50 mles account for over 75% and 50% n terms of weght and value respectvely; those shpped wthn 250 mles account for 95% and 76% respectvely. Therefore, we consder areas wthn both the 50-mle radus and wthn the 250-mle radus of the port to be the port s respectve catchment areas. Populatons n countes wthn the respectve radus of the port are summed up as an estmaton of the populaton n the catchment area. The county level populaton data are obtaned from the U.S. Census Bureau. The Annual Urban Moblty Report 2010 prepared by Texas Transportaton Insttute (TTI) provdes urban road congeston delay measurements for major urban areas. Ths report records road congeston measures aggregated across all the freeways and arteral streets n the urban area. Although data that only nclude roads heavly used by contaner trucks ft to our research queston better, these data are, unfortunately, not avalable. A number of congeston measures are avalable n the report, such as congested travel, congested system, the number of rush hours, annual total delays, annual total congeston costs, cost per peak traveler, delay per peak traveler (DPPT) and travel tme ndex. Among them, we choose DPPT to approxmate road congeston n our study because t s hghly correlated wth the other congeston measures and seems to ft the defnton of congeston delay n our analytcal model the best Congested travel, congested system and the number of rush hours do not provde any nformaton about the level of congeston faced by each traveler. Annual total delays and annual total congeston costs are affected by the populaton sze. Cost per peak traveler s affected by the assumpton of fuel costs and commuters value-of-tme. 14

16 Although road capacty can be theoretcally defned as the maxmum traffc volume that can pass through per unt of tme, such exact measurements are not avalable. Rather, ths paper uses lane-mles to measure the amount of road avalable n an urban area and, therefore, approxmate road capacty. Ths s a common approach n the lterature, whch estmates the road traffc demand nduced by addng road capacty (e.g. [31-34]). The Annual Urban Moblty Report 2010 reports lane-mles of freeways and arteral streets. Ths report also provdes urban area populaton and square-mles, whch are the nstrumental varables, pop t and area, t for the estmaton of equatons (4) and (7). 5. Emprcal results Tables 2, 3, and 4 present, respectvely, the major results of regresson models n (4), (5) and (7) wth the panel data descrbed n Secton 4. The coeffcents for ln(t), ln(p250) and ln(p50) are the estmated elastctes of contaner throughput wth respect to U.S. nternatonal trade value, 250-mle catchment area populaton and 50-mle catchment area populaton respectvely. As expected, all of these coeffcents are postve and statstcally sgnfcant n all models. The nternatonal trade elastcty of contaner port throughput ranges from 0.50 to 1.22, and most of them are below 1, meanng that n the U.S., the percentage growth n contaner throughput s smaller than the percentage growth n nternatonal trade value. On the other hand, we found that a 1% ncrease n the populaton of the 250-mle catchment area s assocated wth a % growth n contaner throughput, whle the elastcty on populaton n the 50-mle catchment area ranges from 0.53 to In general, the estmated coeffcents of the 250-mle catchment area populaton are larger than the coeffcents of the 50-mle catchment area populaton. The travel tme ndex s the rato of peak travel tme and free-flow travel tme that measures congeston on each trp and each mle of travel. DPPT, however, reflects both the per-mle congeston and the length of each trp. 15

17 [Insert tables 2, 3 and 4 here] Table 2 shows that the result for the relatonshp between urban road congeston and port throughput s consstent wth our expectaton (see table 2). In models D1 and D2 (wth large and small catchment areas, respectvely), the OLS coeffcents of own road congeston, ln(d), are negatve and statstcally sgnfcant, whle the coeffcents of the rval s road congeston, ln(dr), are postve but not statstcally sgnfcant. However, after usng the nstrumental varables for ln(d) (.e. the natural logarthms of urban populaton and urban area for models DIV1-2 and the natural logarthm of urban populaton densty for models DIV3-4) and applyng the two-stage least square method (2SLS), the sgns of all the coeffcents reman unchanged, whch are all statstcally sgnfcant, and ther magntudes ncreased. The Durbn-Wu-Hausman tests suggest that the endogenety ssue exsts, and thus the nstrumental varable approach should be adopted. In models DIV1-4, we observe that the mpact of rval s road congeston s smaller than the mpact of own road congeston. In partcular, a 1% ncrease n own road congeston mples a reducton n contaner throughput by %, whle a 1% ncrease n rval s road congeston mples an ncrease n contaner throughput by %. Ths fndng s consstent wth the theoretcal assumptons for demand functons: when the road congeston around a certan port (and thus the full prce of usng ths port) ncreases, only a porton of the shppers who choose not to use the port wll swtch to the rval port because the rest of them wll choose not to shp the goods at all. The mpacts of urban road capacty expanson are estmated by the fxed effect OLS on lane-mles (models SO1 and SO2 wth large and small catchment areas respectvely n table 3). Note that because equaton (5) s a Taylor expanson around the pont where both ln(lm) and ln(lmr) are zeros, whch s out of our sample range, the coeffcents of these two varables do 16

18 not have practcal meanngs. 13 However, by estmatng equaton (5), we can acheve a better understandng of the hgher-order mpacts of road capacty changes. In fact, almost all of the second-order terms are statstcally sgnfcant, suggestng that hgher order effects should not be gnored. Recall equaton (6),.e. our defntons of road capacty elastctes of contaner throughput. The estmaton of 5 s postve, although that of 7 s negatve, mplyng that the own-road capacty elastcty ncreases n own road capacty whle decreasng n ts rval s road capacty. As a result, ports wth much larger road capacty relatve to ther rvals are more lkely to beneft from road capacty expanson. Regardng the mpact of the rval s road capacty expanson, because the estmaton of 6 s negatve, the rval-road capacty elastcty decreases n both own road capacty and rval s road capacty. Therefore, ports are lkely to be harmed by ther rvals road nvestments, unless the road capactes of both competng ports are low. In addton to the general observatons dscussed above, based on the results stated n table 3, we further estmate the road capacty elastctes of throughput for each port and each year covered n our sample and analyze the behavor of ndvdual ports. Except NYNJ, Mam and LALB, each port n our sample has negatve own-road capacty (LM) elastcty on average (table 5). However, n general, LM elastcty has a slghtly ncreasng trend over tme (fgures 2 and 3). The rval-road capacty (LMR) elastcty of throughput s lkely to be negatve, as well, except those of Charleston and Jacksonvlle. We observe a clear downward slopng pattern for each port s LMR elastcty over tme (fgures 4 and 5). Therefore, over the past three decades, own road capacty expanson tended to be more benefcal to or less harmful for the afflated port, whle rval s road capacty expanson tended to become more harmful. 13 We also ft the pure frst-order models by removng all the second-order terms n equaton (5). Assumng there s no hgher order effects, the coeffcents of ln(lm) and ln(lmr) can be nterpreted as, respectvely, the own-road capacty and the rval-road capacty elastctes. We fnd that the coeffcent of ln(lm) s negatve and statstcally sgnfcant, whle the coeffcent of ln(lmr) s negatve but not statstcally sgnfcant. 17

19 [Insert table 5 here] [Insert fgures 2, 3, 4 and 5 here] We further estmated the regresson model whch ncludes road congestons as explanatory varables n equaton (7). Table 4 shows the results for models SOD1-2a wthout IVs and models SOD1-2b and SOD1-2c wth IVs as n table 2. We fnd that a port s contaner throughput s negatvely correlated wth ts own road congeston whle postvely correlated wth ts rval s road congeston. Ths fndng s consstent wth our fndngs from estmatng equaton (4). Moreover, after controllng road congeston, the relatonshp between contaner throughput and road capactes perssts n two respects. Frst, the sgns of the coeffcents of road capactyrelated varables reman the same (table 4). Second, more mportantly, the prevously mentoned patterns of estmated road capacty elastctes over tme do not change, ether (table 5). However, t s worthwhle to have a closer comparson between the results of regresson models n (5) and (7). The LM elastctes decrease (become more negatve or less postve) after controllng for road congestons (table 5). Ths result can also be observed from fgures 2 and 3, as all of the curves of SOD models (.e. after controllng road congeston delays) are below the curves of SO models (wthout controllng road congeston delays). Ths fndng suggests that an ncrease n own road capacty expanson s more (less) strongly assocated wth contaner throughput reducton (ncrease) when road congeston s controlled. In other words, ths fndng may mply that road capacty expanson s postvely correlated wth contaner throughput va the changes n road congeston, whle the mpact of road capacty expanson on throughput through other channels (as mentoned n Secton 3) s negatve. Regardng the LMR elastctes, we fnd that the curves representng LMR elastctes of SOD models are probably above the curves of SO models n fgures 4 and 5, mplyng that the rval s road capacty expanson s less (more) 18

20 strongly assocated wth contaner throughput reducton (ncrease) gven the road congestons beng controlled. One possble nterpretaton for ths result s that rval s road capacty expanson may have a negatve correlaton wth a port s contaner throughput va the adjustment n road congestons but a postve correlaton wth a port s contaner throughput through channels other than road congeston delays. Therefore, gven these results, we mght ndeed dstngush the mpact of road capacty expanson on throughput through road congeston and other channels. 6. Concludng remarks In ths paper, we emprcally study the mpacts of urban road congeston and road capacty expanson on seaport contaner throughput, based on data derved from a sample of major contaner ports n the U.S. We fnd that a port s contaner throughput s statstcally sgnfcantly assocated wth the congeston delays on ts own urban roads, as well as delays on ts rval s roads. Specfcally, a 1% ncrease n own road congeston mples a reducton n contaner throughput by %, whle a 1% ncrease n rval s road congeston mples an ncrease n contaner throughput by %. Thus, by mtgatng road congeston n the urban area nearby, the port management would be able to effectvely compete wth ts rvals. Urban road capacty expanson, on the other hand, has dfferent mplcatons on contaner throughput. We fnd that own urban road capacty s postvely correlated wth the contaner throughput of the Port of NYNJ, the Port of LALB and the Port of Mam whle negatvely correlated wth the throughputs of the other ports n our sample. Furthermore, except the Port of Charleston and the Port of Jacksonvlle, the contaner throughput of a port tends to be negatvely correlated wth ts rval s road capacty. The relatonshp between road capacty and contaner 19

21 throughput va the changes n road congeston delays s largely consstent wth the predctons obtaned from the quantty-competton analytcal model. That s, va the change of road congeston, an ncrease n road capacty mples an ncrease n output by the port nearby but a decrease n output by ts rval port. Therefore, addng more roads mght be an effectve strategy to mprove a port s compettveness, provded that road capacty expanson solely affects road congeston whle havng lttle negatve mpact on the port through other channels. Local governments and port management should be cautous when decdng to provde more roads so as to reduce hnterland congeston and ncrease throughput, as addng road capacty mght be harmful to the port overall, although benefcal n terms of mtgatng road congeston. Our study offers a couple of avenues for further nvestgaton when better and more detaled data are avalable. Frst off, we do not have lane-mle data for roads around the ports or roads heavly used by port-related traffc. It s lkely that extra capacty n the overall urban areas that we consdered s added at the areas less used by commercal trucks and f so, t may amplfy the effects of road capacty expanson va factors other than road congeston. Second, we approxmate the modal splt shares wth data from the U.S. Commodty Flow Surveys, whch nclude all domestc freght movements but are not lmted to port-related contaner traffc. There mght be a large share of truck-ral ntermodal transport for port-related contaner movements, but we are not able to dstngush ths possblty from truck-only contaner movements. 20

22 Fgure 1 An llustraton of the prce competton case 21

23 Fgure 2 Estmated LM elastcty of throughput (250-mle catchment area) Baltmore Boston Charleston Hampton Roads Jacksonvlle LALB Mam NYNJ Oakland Portland SeaTac Year SO1_LM SOD1a_LM SOD1b_LM SOD1c_LM Graphs by pd 22

24 Fgure 3 Estmated LM elastcty of throughput (50-mle catchment area) Baltmore Boston Charleston Hampton Roads Jacksonvlle LALB Mam NYNJ Oakland Portland SeaTac Year SO2_LM SOD2a_LM SOD2b_LM SOD2c_LM Graphs by pd 23

25 Fgure 4 Estmated LMR elastcty of throughput (250-mle catchment area) Baltmore Boston Charleston Hampton Roads Jacksonvlle LALB Mam NYNJ Oakland Portland SeaTac Year SO1_LMR SOD1a_LMR SOD1b_LMR SOD1c_LMR Graphs by pd 24

26 Fgure 5 Estmated LMR elastcty of throughput (50-mle catchment area) Baltmore Boston Charleston Hampton Roads Jacksonvlle LALB Mam NYNJ Oakland Portland SeaTac Year SO2_LMR SOD2a_LMR SOD2b_LMR SOD2c_LMR Graphs by pd 25

27 Table 1 Ports studed and ther respectve rval Port Contaner throughput n 2009 (n 1,000 TEU s) Average market share a (%) Rval port Dstance to the rval port (mles) Boston NYNJ 190 NYNJ 4, Baltmore 170 Baltmore NYNJ, Hampton Roads 170 Hampton Roads 1, Baltmore 170 Charleston 1, Jacksonvlle 197 Jacksonvlle Charleston 197 Mam Jacksonvlle 326 SeaTac 3, Portland 145 Portland SeaTac 145 Oakland 2, LALB 344 LALB 11, Oakland 344 a. West-coast port market share: S X X j {U.S. Pac fc coast ports} East-coast port market share: S X j j X j j j {U.S. Atla ntc coast ports} 26

28 Table 2 Fxed effect OLS and 2SLS on congeston delays: equaton (4) OLS 2SLS D1 D2 DIV1 DIV2 DIV3 DIV4 ln(t) ** (0.0769) ** (0.0595) ** (0.0982) ** (0.0910) ** (0.1744) ** (0.1349) ln(p250) ** (0.2123) ** (0.2507) ** (0.3999) ln(p50) ** (0.0925) ** (0.1030) ** (0.1277) ln(d) * (0.1171) ** (0.1302) ** (0.3644) ** (0.3642) ** (0.9264) ** (0.6815) ln(dr) (0.1375) (0.6028) ** (0.2655) ** (0.2598) ** (0.6408) ** (0.4631) Constant ** (1.7150) ** (0.6028) ** (2.1445) ** (0.7159) ** (3.5085) ** (0.9415) N F Adjusted R Instruments ln(area) and ln(pop) ln(densty) Durbn-Wu-Hausman ch-sqare test H0: ln(d) s exogenous Ch 2 (1) p-value * sgnfcant at α=0.10; ** sgnfcant at α=0.05; Values n the brackets are robust standard errors for OLS and standard errors for 2SLS. Fxed effect coeffcents are omtted. 27

29 Table 3 Fxed effect OLS on road capactes (lane-mles): equaton (5) SO1 SO2 ln(t) ** (0.1021) ** (0.0889) ln(p250) * (0.2414) ln(p50) ** (0.1181) ln(lm) ** (2.1502) ** (2.1587) ln(lmr) ** (2.3505) ** (2.1940) [ln(lm)] ** (0.1737) ** (0.1749) [ln(lmr)] ** (0.1968) (0.1829) ln(lm)*ln(lmr) ** (0.2546) ** (0.2200) Constant ** * (9.3778) (9.5978) N F Adjusted R * sgnfcant at α=0.10; ** sgnfcant at α=0.05; Values n the brackets are robust standard errors. Fxed effect coeffcents are omtted. 28

30 Table 4 Fxed effect OLS and 2SLS on congeston delays and road capactes (lane-mles): equaton (7) SOD1a SOD2a SOD1b SOD2b SOD1c SOD2c ln(t) ** (0.0985) ** (0.0819) ** (0.0961) ** (0.0917) ** (0.1392) ** (0.1152) ln(p250) ** (0.2476) ** (0.2393) ** (0.3615) ln(p50) ** (0.1203) ** (0.1076) ** (0.1346) ln(d) ** (0.1158) ** (0.1103) (0.3283) ** (0.3311) ** (0.7837) ** (0.6023) ln(dr) ** (0.1450) ** (0.1409) (0.2493) ** (0.2485) ** (0.5656) ** (0.4322) ln(lm) ** (2.1013) ** (2.0873) ** (1.9259) ** (1.9003) ** (2.6011) ** (2.1947) ln(lmr) ** (2.4403) ** (2.2758) ** (1.9312) ** (2.1166) ** (2.5001) ** (2.2694) [ln(lm)] ** (0.1761) ** (0.1704) ** (0.1854) ** (0.1728) ** (0.2383) ** (0.1854) [ln(lmr)] ** (0.2174) (0.2022) ** (0.1969) (0.2007) ** (0.2651) (0.2165) ln(lm)*ln(lmr) ** (0.2818) ** (0.2449) ** (0.3026) ** (0.2832) (0.4181) ** (0.3146) Constant ** (9.9426) (9.9384) ** (8.2794) (9.0437) ** ( ) ( ) N F Adjusted R Instruments ln(area) and ln(pop) ln(densty) Durbn-Wu-Hausman ch-sq test H0: ln(d) s exogenous Ch 2 (1) (p-value) ( ) ( ) ( ) ( ) * sgnfcant at α=0.10; ** sgnfcant at α=0.05; Values n the brackets are robust standard errors for OLS and standard errors for 2SLS. Fxed effect coeffcents are omtted. 29

31 Table 5 The average elastctes estmated by equatons (5) and (7) Ports Boston NYNJ Baltmore Hampton Roads Charleston Jacksonvlle Mam LM LMR LM LMR LM LMR LM LMR LM LMR LM LMR LM Models elastcty elastcty elastcty elastcty elastcty elastcty elastcty elastcty elastcty elastcty elastcty elastcty elastcty LMR elastcty Eq.(5) SO SO Eq.(7) SOD1a SOD2a SOD1b SOD2b SOD1c SOD2c Ports SeaTac Portland Oakland LALB Models LM elastcty LMR elastcty LM elastcty LMR elastcty LM elastcty LMR elastcty LM elastcty LMR elastcty Eq.(5) SO SO Eq.(7) SOD1a SOD2a SOD1b SOD2b SOD1c SOD2c

32 References 1. Luo, M., and Grgalunas, T., 2003, A spatal-economc multmodal transportaton smulaton model for US coastal contaner ports. Martme Economcs and Logstcs, 5, Cullnane, K., and Song, D.-W., 2006, Estmatng the relatve effcency of European contaner ports: a stochastc fronter analyss. In: Port Economcs: Research n Transportaton Economcs, Cullnane, K. and Talley, W. (eds), Vol. 16 (Oxford, U.K.: Elsever JAI Press), pp Suykens, F., and Van De Voorde, E., 1998, A quarter a century of port management n Europe: objectves and tools. Martme Polcy and Management, 25(3), Robnson, R., 2002, Ports as elements n value-drven chan systems: the new paradgm. Martme Polcy and Management, 29, Notteboom, T.E., 2004, Contaner shppng and ports: an overvew. Revew of Network Economcs, 3(2), Fan, L., Koehler, M.M., and Wlson, W.W., 2011, Intermodalsm and new trade flows. In: The Blackwell Companon to Martme Economcs, Talley, W.K. (ed.), (West Sussex, U.K.: Wley- Blackwell), pp , forthcomng. 7. Notteboom, T.E., and Rodrgue, J-P., 2008, contanersaton, box logstcs and global supply chans: the ntegraton of ports and lner shppng networks. Martme Economcs and Logstcs, 10, Slack, B., 1985, Contanerzaton, nter-port competton and port selecton. Martme Polcy and Management, 12(4), Lrn, T.C., Thanopoulou, H.A., Beynon, M.J., and Beresford, A., 2004, An applcaton of AHP on transhpment port selecton: a global perspectve. Martme Economcs and Logstcs, 6(1), Ugbonma, C., Ugboma, O., and Ogwude, I.C., 2006, An Analytc Herarchy Process (AHP) approach to port selecton decsons emprcal evdence from Ngeran ports. Martme Economcs and Logstcs, 8(3), Yuen, A., Zhang, A., and Cheung, W., 2011, Port compettveness from the users perspectve: an analyss of major contaner ports n Chna and ts neghborng countres. Research n Transportaton Economcs, forthcomng. 12. Heaver, T.D., 2006, The evoluton and challenges of port economcs. In: Port Economcs: Research n Transportaton Economcs, Cullnane, K., and Talley, W. (eds), Vol. 16 (Oxford, U.K.: Elsever JAI Press), pp TTI, 2010, Urban Moblty Report 2010, Texas Transportaton Insttute (TTI), FHWA, 2004, Traffc congeston and relablty: lnkng solutons to problems. Federal Hghway Admnstraton (FHWA), the Unted States, Golob, T.F., and Regan, A.C., 2000, Freght ndustry atttudes towards polces to reduce congeston. Transportaton Research Part E: Logstcs & Transportaton Revew, 36E(1), Malon, M.J., and Jackson, E.C., 2005, North Amercan contaner port capacty: an exploratory analyss. Transportaton Journal, 44 (3), Pettt, S.J., and Beresford, A.K.C., 2008, An assessment of long-term Unted Kngdom port performance: a regonal perspectve. Martme Economcs and Logstcs,10,