Accessibility improvement and interregional trade flows

Abstract

Accessibility improvement has been regarded as one of the main regional policy goals in the recent decades. As a result, many countries has undertaken complex transport infrastructure investment programmes worth billions of dollars. The best examples of such investment can be China or, in the European context, Spain and more recently Poland. There have been published many analyses focusing on output and employment effects of transport infrastructure investment projects. Yet, there hardly exist any studies that verify their potential influence on interregional trade flows. This paper fills the existing gap in the literature by assessing the impact of accessibility improvement on the volume and the value of interregional trade flows in Poland between 2005 and 2015. We find that there is no statistically significant relationship between travel time reduction and interregional trade. This may explain largely the surprising lack of positive relationship between accessibility improvement and regional GDP growth, reported in previous papers.

Introduction

Accessibility improvement has been regarded as one of the main regional policy goals in the recent decades. As a result, many countries has undertaken complex transport infrastructure investment programmes worth hundreds billions of dollars. The best examples of such programmes can be China or, in the European context, Spain, and more recently Poland. In fact, transport infrastructure investment has been in the core of the European Cohesion Policy almost since its emergence in the late 1970s. Yet, it was the creation of the Trans-European Networks (TENs) in 1993 that triggered a particular boost in the European transport infrastructure investment.

Following statistics provided by EUROSTAT, road transport related activities account by far for the highest share of turnover in the 28 members of the European Union (EU28)¹ as compared to the air, sea or railways. The employment in the road transport (both freight and passenger) reached almost 50% of all transportation and storage activities, exceeding 5.5 million in 2017. This is hardly surprising then that the majority of transport infrastructure investment projects accomplished in Europe have been dedicated to the roads. For instance, the report by the European Union Road Federation (2016) shows that road infrastructure investment accounted for more than 60% of all transport infrastructure investment in the EU28 member states between 1995 and 2013. The majority of financial means (76%) were devoted to motorways and express roads. The share of road investment projects has slightly decreased at the end of the last decade with an increase in spending on railways. Still, in 2019 it remained above the 55% of entire inland transport infrastructure investment in the EU28 (e.g., European Union Road Federation 2022).

Poland can serve as an example of country that has experienced a tremendous accessibility improvement in the last two decades (e.g., Rosik et al. 2015). The accession to the European Union (EU) in 2004 was followed by an increase in transport infrastructure investment, focused mainly on the development of high-speed road network. In accordance to statistics provided by the General Directorate for National Roads and Highways, the overall length of the high-speed network raised from around 700 km in 2004 to over 4600 km in 2022. This in turn, led to a significant road transport accessibility improvement. The forecasted cumulative value of the above improvement reaches over 40% in 2023 (e.g., Rosik et al. 2022). The overall development of road network in Poland between 2005 and 2020 is shown on Fig. 1 below.

Fig. 1

Source: Authors preparation

Evolution of road network in Poland between 2005 and 2020.

There exist many analyses that try to quantify macroeconomic effects of transport infrastructure investment projects worldwide. The vast majority of them focus on assessing the impact of accessibility improvement on either production or employment. Yet, at the same time, very few papers try to empirically verify the relationship between accessibility improvement and the growth of interregional trade. This may seem surprising given that the international trade literature, based on the gravity equation estimation, proves that the distance/travel time should be considered as one of the main determinants of bilateral trade flows (e.g., Bergstrand 1985; Chaney 2018). There is no reason to believe that the above should not hold for within-country trade. Still, most likely, this is the lack of the reliable data on interregional trade flows that does not allow investigating the above issue.

In the present paper, we aim to fill in the gap that exists in the literature and assess the impact of road transport accessibility improvement on interregional trade in the case of Poland. We apply data concerning the reduction in travel time and verify its relationship with bilateral trade flows between Polish regions from 2005 to 2015.² We focus on road transport infrastructure given its dominant share in interregional trade found in other papers (e.g., Llano et al. 2017; Felbermayr and Tarasov 2022). To measure the trade flows two different indicators are applied. First one measures the volume of road freight transport between particular NUTS2 level regions. The second one is expressed in pecuniary units and measures the value of trade between particular regions and industries. It is estimated using the Horridge algorithm (e.g., Rokicki et al. 2021) and modified version of the entropy-based approach proposed by by Golan et al. (1994). We find that there is no statistically significant relationship between travel time reduction and the volume of interregional trade. At the same time, however, the impact of accessibility improvement on the value of interregional trade flows is less clear and depend on the trade matrix used in estimates. Still, in the case of the preferred entropy-based trade matrix, that applies additional available information on the volume of trade flows, no specification features positive and statistically significant relationship between accessibility improvement and the value of trade flows.

The remainder of the paper is organised as follows. Section "Literature review" provides the review of prior literature on the impact of transport infrastructure investment on interregional trade. Section "Methodology and data" discusses the research methodology and the data applied in our study. Section "Empirical results" presents the results of the empirical analyses. Here, we discuss both the impact of accessibility improvement on the volume of interregional trade and its commodity structure. Finally, Section "Conclusions" concludes and provides directions for further research.

Literature review

There have been published many analyses focusing on the microeconomic and macroeconomic effects of transport infrastructure investment projects worldwide. The majority of them focus on broadly understood regional economic development, with particular focus on output and employment. They differ, however, in the methodological approach adopted. The most commonly used is the Cost–Benefit Analysis (e.g., Lakshmanan 2011), followed by the general productivity approach (e.g., Aschauer 1989) and computable general equilibrium (CGE) modelling (e.g., Bröcker 2004). Only a small share of empirical studies has tried to directly examine the relationship between accessibility improvement and regional development. They find, on average, rather limited and localized impact of accessibility improvement on both production and employment (e.g., Islam 2003; Lein and Day 2008; Rokicki and Stępniak 2018).

Productivity increase is usually considered as a main channel linking accessibility improvement with regional economic development (e.g.,Venables 2007; Vickerman 2008). Consequently, for decades, transport infrastructure investment has been considered as one of key tools of European Cohesion Policy “laying the foundations for economic growth and sustainability” (European Commission 2024). More recently, additional focus was put on the social impact of transport accessibility on the overall level of well-being that is reflected in the voting behaviour of the peripheral regions (e.g., Rodríguez-Pose (2018). In this sense, improvements in regional accessibility are thought as a way to address regional divide in Europe between core and peripheral regions by improving the access to higher education or positively affecting employment prospects of low-income individuals (e.g., European Commission 2024).

Nevertheless, one could expect that accessibility increase should also have a significant effect on interregional trade. This is, at least, a conclusion that can be drawn from the theoretical models developed within the New Economic Geography (e.g., Krugman 1991, 1992). Here, trade costs (that include both distance-related transportation costs and trade barriers) act as a main determinant of trade attractiveness and regional economic development.³ Also, international trade literature, based on the gravity equation estimation, proves that the distance/travel time should be considered as one of the main determinants of bilateral trade flows (e.g., Bergstrand 1985; Chaney 2018). As a result, it may seem surprising that very few papers actually try to verify empirically the relationship between accessibility improvement and the growth of interregional trade.

In general, there are relatively few analyses that focus on issues related to the interregional trade at all. This is mainly due to the lack of the reliable survey data on interregional trade flows provided by national statistical agencies (e.g., Jackson et al. 2005). As a result, there are hardly any econometric studies focusing on interregional trade flows. On the other hand, trade matrices for regional input–output (IO) models or spatial computable general equilibrium (SCGE) models are constructed using different regionalization algorithms.⁴ Such models usually do not examine directly the structure or the evolution of regional trade flows but rather try to incorporate them in a broader analysis of the impact of different macroeconomic shocks on regional economies. Yet, as claimed by Hewings and Oosterhaven (2021), more attention should be put on the investigation of within-country trade flows per se, given the magnitudes of interregional trade volumes revealed in the existing studies. In fact, one may assume that from the point of view of regional economies, “interregional trade is most likely relatively more important than international trade” (e.g., Hewings and Oosterhaven 2021).

Existing papers confirm the significance of interregional trade flows in the case of the US states (e.g., Sonis et al. 2002; Jackson et al. 2005; Munroe et al. 2007; Hewings and Parr 2009; Szewerniak et al. 2019), China (e.g., Xu and Fan 2012; Zheng et al. 2022) or regions within particular member states of the EU (e.g., Ferreira 2008; Llano et al. 2010, 2017). Many authors underline the “border effect” as a factor that stimulates interregional trade at the expense of international one. In fact, there seems to exist a certain consensus concerning the possible impact of transport infrastructure investment on international trade. Here, an improvement of transport infrastructure in border regions should significantly reduce the “border effect” and thus would lead to an increase in international trade flows (e.g., Santamaria et al. 2021; Felbermayr and Tarasov 2022). Still, even though the theoretical literature predicts the positive relationship between within-country accessibility improvement and interregional trade, we are not aware of empirical analyses directly evaluating the aforementioned link. Instead, there can be found studies verifying the relationship between interregional trade and administrative forces or cultural differences (e.g., Zheng et al. 2022), business environment and income (e.g., Xu and Fan 2012), intra-industry trade (e.g., Munroe et al. 2007) or diesel prices (e.g., Szewerniak et al. 2019).

Methodology and data

Data

In this study we use two different sets of interregional trade data comes for Poland. First, in order to verify the impact of accessibility improvement on interregional trade we use a road trade data from Polish Central Statistical Office (CSO) for the years 2005, 2010 and 2015. The volume of trade flows is expressed either in thousands of tons or in millions of tons per km (tkm). We have a data on bilateral trade flows for both 16 NUTS2 level regions and 73 NUTS3 level regions.

Still, the above data do not provide any information on the value or the content of interregional trade flows. Hence, additionally we also use a second dataset derived from the 2005 and 2015 national supply and use tables by applying cross-entropy estimators. This allows us to verify whether the accessibility improvement has led to a significant changes in the commodity structure of interregional trade. The data required for the estimation of interregional trade matrices comes from the CSO.

Apart from the trade data, we also use data on regional Gross Value Added (GVA) expressed in constant 2015 prices. The data on nominal comes from the CSO while the deflators come from the EUROSTAT. The data on travel time comes from the transport model by the Institute of Geography and Spatial Organization Polish Academy of Sciences. It measures the travel time for trucks between the capitals of particular regions.

Methodology

Gravity equation

In the first stage of analysis, we verify whether our regional data fits traditional trade model. Hence, we estimate an interregional gravity equation in the simple form of:

$$lntrade_{i,j,t} = \alpha + \gamma lnGDPO_{i,t} + \delta lnGDPD_{j,t} {-}\beta lndist_{i,j,t} + \, \lambda_{t} + \, \mu \, \lambda_{t} *\nu_{i} + \, \lambda_{t} + \, \sigma \, \lambda_{t} *\tau_{j} + \varepsilon_{i,j,t}$$

(1)

where subscript i represents the region of origin, and j represents the destination region and t refers to a given year. The dependent variable lntrade is the logarithm of the road trade flow between each trade pair measured either in thousands tons or millions of ton-kilometer (tkm).⁵ The independent variables include the logarithms of distance measure between regions (lndist), the GDP of the origin and the destination (lnGDPO, lnGDPD), and the time specific effects λ. Additionally, we also control for interactions between time and both origin (ν_i) and destination (τ_j) fixed effects.

Note that unlike the majority of gravity studies, the distance measure is expressed in minutes of travel time, rather than in time invariant geographical distance in kilometers. Since the former indicator is time varying, it allows us to verify the relationship between the change in accessibility and the change in volume/value of interregional trade flows. Another difference, as compared to the gravity studies, is an exclusion of a dummy variable that controls for neighborhood regions (e.g., Anderson and Van Wincoop 2004). This is because Polish regions do not have authority to conduct trade policy (e.g., by imposing local sales or excise taxes). Consequently, there are no reasons to believe that there exist border effects that may influence interregional trade flows.⁶

In the second stage of our analysis, we assess the impact of accessibility improvement on both the volume and the value of interregional trade flows. Hence, we estimate the gravity equation in differences instead of levels. As a result, it takes form of:

$$\Delta lntrade_{i,j,t} = \alpha + \gamma \Delta lnGDPO_{i,t} + \delta \Delta lnGDPD_{j,t} {-}\beta \Delta lndist_{i,j,t} + \, \lambda_{t} + \, \mu \, \lambda_{t} *\nu_{i} + \, \lambda_{t} + \, \sigma \, \lambda_{t} *\tau_{j } + \varepsilon_{i,j,t}$$

(2)

where Δ stands for difference in logs between particular years.

Following the gravity literature, there are at least several estimation issues that should be taken into account. The first one refers to the existence of multilateral resistance terms that may affect bilateral trade flows (e.g., Anderson and Van Wincoop 2004; Head and Mayer 2014). As already mentioned, Polish regions do not have any authority in trade policy. Hence, there is no reason to believe that there might occur significant region-specific shocks affecting interregional trade. In any case, such shocks should be captured by the interaction between time and regional fixed effects that is included in our specifications. Second potential issue is the heteroscedasticity of standard errors. We deal with the above clustering standard errors by trade pairs. As shown by Cameron and Miller (2015), cluster-robust standard errors are also heteroscedastic-robust. Finally, there is also a potential endogeneity issue. We aim at dealing with the latter by taking first differences (e.g., Freund and Weinhold 2004) and by applying fixed effects panel estimator (e.g., Baltagi et al. 2015).

Interregional trade matrix

While assessing the relationship between accessibility improvement and trade volume, we rely on the statistical data on interregional road trade flow from the Polish CSO. However, there is no official survey-based statistical data on the value of interregional trade flows in Poland and some indirect estimation method is required. Hence, we apply two different approaches to estimate interregional trade matrix from the national supply and use tables provided by CSO. This allows us to compare the robustness of our empirical analyses.

Our first approach to the estimation of interregional trade flows is based on Horridge algorithm (e.g., Rokicki et al. 2021). The latter applies doubly-constrained gravity equation together with RAS. Existing studies claim that the combination of gravity model and RAS usually provides more accurate results as compared to other regionalization approaches, such as location quotient or commodity balance approach (e.g. Fournier Gabela 2020; Riddington et al. 2006; Sargento et al. 2012).

Still, the gravity model is strictly linked to the distance variable and does not allow to include additional information that may be available. For instance, in our case, we dispose of interregional trade volume data that could be a good proxy for the value of the interregional trade flows. This in turn, could result in more precise estimation of the latter. Therefore, we follow the ideas of previous papers that apply entropy based estimators to the estimation of interregional trade matrix (e.g., Fernandez Vazquez 2015; Fernandez Vazquez et al. 2015; Johansen et al. 2018). The problem under study is the estimation of the inter-regional trade flows for a system of region $i=1,..N$ locations. The ${t}_{ij}$ cells of the matrix are the unknown trade flows we would like to estimate (shaded in grey), where the row and column aggregates (${y}_{i\cdot }$ and ${x}_{\cdot j}$ respectively) are known and where the sums $\sum_{j=1}^{N}{t}_{ij}={y}_{i\cdot }$ and $\sum_{i=1}^{N}{t}_{ij}={x}_{\cdot j}$ contain the total interregional exports and imports respectively. The entropy-based approach relies on transforming the initial trade flows ${t}_{ij}$ by scaling them as ${p}_{ij}={t}_{ij}/T$, where $T={\sum }_{i=1}^{N}\sum_{j=1}^{N}{t}_{ij}$. An equivalent scaling applies to the total margins $\mathbf{x}$ and $\mathbf{y}$, making the elements of the scaled margins to range from 0 to 1 and, consequently, the equation $\sum_{j=1}^{N}{x}_{\cdot j}=\sum_{i=1}^{N}{y}_{i\cdot }=1$ holds now. This scaling allows for considering the elements of a target matrix $P$ (note that its cells are positive and they sum up to one) as a probability distribution to be estimated. Traditionally, entropy-based technique minimizes the Kullback–Leibler divergence (e.g., Kullback and Leibler 1951) between the target $P$ and an initial matrix $Q$, provided that the solution is consistent with the observable information. A constrained minimization problem is applied, and it can be posed as a program like:

$$\underset{{\varvec{P}}}{\mathrm{Min}}\,\, D\left({\varvec{P}},{\varvec{Q}}\right)=\sum_{i=1}^{N}\sum_{j\ne i}^{N}{p}_{ij}ln\left(\frac{{p}_{ij}}{{q}_{ij}}\right)$$

(3)

Subject to:

$$\sum_{j=1}^{N}{p}_{ij}={y}_{i\cdot };\quad i=1,\dots ,N$$

(4)

$$\sum_{i=1}^{N}{p}_{ij}={x}_{\cdot j};\quad j=1,\dots ,N$$

(5)

$$\sum_{i=1}^{N}\sum_{j=1}^{N}{p}_{ij}=1$$

(6)

In most of the cases, the regional totals x and y are not directly observable and are estimated by applying the gravity model. Hence, the Generalized Cross Entropy GCE estimator re-parametrizes possible errors as (discrete) random variables that can take $L$ different values. These realizations reported on a so-called supporting vector ${\mathbf{v}}_{\mathbf{k}}^{\mathbf{^{\prime}}}$, symmetric and centred around zero as ${\mathbf{v}}_{\mathbf{i}}^{\mathbf{^{\prime}}}=\left[{v}_{1i},\dots ,0,\dots ,{v}_{Li}\right]$, where $L$ is an odd number and ${v}_{1i}=-{v}_{Li}$. The probabilities corresponding to the ${v}_{l}$ elements are defined as ${\mathbf{w}\mathbf{^{\prime}}}_{\mathbf{i}}=\left[{w}_{1i}, \dots ,{w}_{Li}\right]$. While the elements on vector ${\mathbf{v}}_{\mathbf{i}}$ are imposed by the researcher, the corresponding probabilities ${\mathbf{w}}_{\mathbf{i}}$ are unknown and should be estimated.

The above GCE formulation may correctly address the issue related to the possible errors in regional totals that can be estimated using the gravity model. However, it does not effectively deal with the possibility of having other factors that are affecting the interregional trade flows. Therefore, in this paper we use an alternative entropy formulation proposed by Golan et al. (1994), which rely not on the aggregates but on the cross-moments between the interregional trade flows and some additional covariates that are not considered in simple gravity-type equations. In particular, we apply moment-constrained GCE estimator that allows including additional sources of information to the derivation of interregional trade matrix from the national supply and use tables. Here, we are assuming that there is a set of covariates in a ($N\times H$) matrix Z that are expected to be correlated with the ${t}_{ij}/T$ scaled trade flows. Examples of these covariates can be income differences, stocks of infrastructures or indicators of specialization, among others. This information can be incorporated to the GCE estimator by following variant of the estimator proposed in Golan and Vogel (2000) or Golan (2018), which make use of the cross-moments equations:⁷

$${\text{Z}}^{{\text{t}}} {\tilde{\text{y}}} = {\text{Z}}^{{\text{t}}} \left[ {{\text{Pi}} + {\text{u}}_{{\text{y}}} } \right] = {\text{Z}}^{{\text{t}}} \left[ {{\text{Pi}} + {\text{W}}_{{\text{y}}} {\text{v}}} \right]$$

(7)

$${\tilde{\text{x}}}^{{\text{t}}} {\text{Z}} = \left[ {{\text{i}}^{{\text{t}}} {\text{P}} + {\text{u}}_{{\text{x}}}^{{\text{t}}} } \right]{\text{Z = }}\left[ {{\text{i}}^{{\text{t}}} {\text{P}} + {\text{ v}}^{{\text{t}}} {\text{W}}_{{\text{x}}}^{{}} } \right]{\text{Z}}$$

(8)

that serve as constraints in the following optimization program:

$$\underset{P,Q}{\mathrm{Min}}\,\, D\left(\text{P},\text{W},\text{Q},{W}^{0}\right)=\sum_{i=1}^{N}\sum_{j\ne i}^{N}{p}_{ij}ln\left(\frac{{p}_{ij}}{{q}_{ij}}\right)+\sum_{i=1}^{N}\sum_{l=1}^{L}{w}_{li}^{x}ln\left(\frac{{w}_{li}^{x}}{{w}_{li}^{x0}}\right)+\sum_{i=1}^{N}\sum_{l=1}^{L}{w}_{li}^{y}ln\left(\frac{{w}_{li}^{y}}{{w}_{li}^{y0}}\right)$$

(9)

subject to:

$$\sum_{i=1}^{N}{z}_{ih}{\widetilde{y}}_{i\cdot }=\sum_{i=1}^{N}{z}_{ih}\left[{p}_{ij}+\sum_{l=1}^{L}{w}_{li}^{y}{v}_{li}\right];\quad h=1,\dots ,H$$

(10)

$$\sum_{j=1}^{N}{{\widetilde{x}}_{\cdot j}z}_{jh}=\sum_{j=1}^{N}\left[{p}_{ij}+\sum_{l=1}^{L}{w}_{lj}^{x}{v}_{lj}\right]{z}_{jh};\quad h=1,\dots ,H$$

(11)

$$\begin{aligned} \mathop \sum \limits_{j = 1}^{J} p_{ij} & = 1;\quad i = 1,..,n \\ \mathop \sum \limits_{l = 1}^{L} w_{li}^{y} & = 1;\quad { }i = 1, \ldots ,N \\ \mathop \sum \limits_{l = 1}^{L} w_{lj}^{x} & = 1;{ }\quad j = 1, \ldots ,N \\ \end{aligned}$$

(12)

Once the GCE program above is solved, the estimates of $P$($\widehat{P}$) are recovered, being this solution the closest one to matrix $Q$ while being consistent with cross moments between $x,y$ and $Z$ given in the left hand side of Eqs. (10) and (11). Without these pieces of information, the solution of the GCE estimator will be matrix $Q$, but this equation “pushes” the solution to be consistent with these observed moments.

Empirical results

Development of high-speed road network and resulting improvement in transport accessibility should have a significant impact on trade flows. Still, it is not clear whether in the case of a small open economy placed in the Central Europe, accessibility improvement would result in an increase of interregional or rather international trade flows. Actually, an improvement in interregional transport connections in Poland should also lower the costs of international transportation. This applies both to the east–west and south-north European transport corridors. As a consequence, local producers could find it more profitable to expand their trade network internationally instead of deciding to increase exports to other Polish regions. Especially, if there exist transport service supply constraints (e.g., due to the limited amount of trucks or drivers).

Trade volume

Table 1 compares the overall amount of interregional and international road trade volume between 2005 and 2015. It can be observed that the evolution of interregional and international trade flows is completely different. On one hand, interregional trade flows, measured in thousands of tons, increase between 2005 and 2010 by more than 30% and slightly decrease between 2010 and 2015. At the same time, however, there is a constant increase in trade volume expressed in tkm (by almost 35% between 2005 and 2010 and over 27% between 2010 and 2015). This indicates that interregional trade is not necessarily expanding in terms of volume but clearly increases in terms of distance. Completely different evolution is experienced in the case of international trade flows. Trade volume measured in thousands of tons increases almost 130% from 2005 to 2010 and almost 30% from 2010 to 2015. The tkm indicator increases by around 120% between 2005 and 2010 and 20% between 2010 and 2015. Hence, international trade increases in terms of volume but not necessarily in terms of distance.

Table 1 Evolution of interregional and international trade flows in Poland between 2005 and 2015 (road transportation only)

The next step of our analysis consists in verifying whether the interregional road trade data can be considered as a good proxy for the interregional trade flows. Table 2 below shows the results of the simple gravity estimation for the trade volume measured either in thousands of tons (columns 1–3) or millions of tkm (columns 4–6). It can be observed that in both cases the results are in line with the expectations. It is, the volume of trade is positively correlated with the economic mass of the origin and destination. At the same time it is negatively correlated with the travel time. Hence, we may assume that our data can be used to analyze the impact of accessibility improvement on interregional trade volume.

Table 2 Simple gravity equation FE estimation for the period 2005–2015 at the NUTS2 level

Table 3 shows the results of FE estimations based on changes instead of levels that allow verifying the relationship between accessibility improvement and the volume of interregional trade between 2005 and 2015. Columns 1 and 4 display the results for the 2005–2015 period where the dependent variable is expressed in thousands of tons and mln tkm respectively. We may observe that, although the signs of all explanatory variables are as expected, the coefficients not statistically significant, irrespective the specification. We have previously found, however, that the interregional trade rose between 2005 and 2010 but not necessarily between 2010 and 2015. Therefore, the results are likely to differ among particular periods. The above assumption is verified in columns 2 and 5 (2005–2010 period) and in columns 3 and 6 (2010–2015 period). Indeed, it appears that accessibility improvement is positively correlated with interregional trade in the former period. However, there is no such a relationship in the latter one.

Table 3 Accessibility improvement and interregional trade between 2005 and 2015 at the NUTS2 level

There is at least one possible explanation of different results for particular periods, related to an increasing deficit of professional drivers. As shown in the report by PwC (2016), the volume of road transport in Poland increased by more than 80% between 2005 and 2015. However, the amount of professional drivers hired in the transport sector stabilized after 2010. As a result, the estimated deficit of drivers in Poland already reached 20% of optimal number of workers in 2015. Existing constraints on the supply side seem to be a good reason for producers to optimize their trade patterns and stick with the most profitable routes. The above may explain stagnation of interregional trade flows and an increase in the volume of international trade between 2010 and 2015.

Still, one may wonder whether the stagnation of interregional trade volume in Poland between 2010 and 2015 is a kind of general pattern or rather, it is that certain trade routes have grown at the expense of others. In this sense, a significant accessibility improvement could be considered as a motivation to increase trade volume with some locations. Especially, if the observed decrease in travel time is heterogeneous across regions. Table 4 shows the improvement in travel time between particular NUTS2 regions during the 2005–2015 period. It can be observed that in many cases the cumulative decrease in travel time does not exceed 5%. Consequently, it is likely that an overall accessibility improvement may have nonlinear effects. It is, in the case of regions with smaller decrease in travel time we could observe a decrease in trade flows, while the regions with the higher accessibility improvement could experience an increase in the volume of interregional trade.

Table 4 Percentage change in travel time between 2005 and 2015

Note that the spatial distribution of accessibility improvement closely follows the construction of new motorways depicted in Fig. 1. This is due to the European Union investment projects related to the creation of the Trans-European Transport Network (TEN-T). The highest reduction in travel time (20%) is due to the completion of the north–south A1 motorway linking Gdańsk (Pomorskie) with Łódź (Łódzkie). There is also a high reduction in travel time observed between Pomorskie and Kujawsko-Pomorskie (with the capital city Toruń) or between Pomorskie and Śląskie (with the capital city Katowice). The latter connection is particularly important as it links the most industrialized Polish region with the largest Polish seaport. In addition, a significant improvement in accessibility follows the completion of the A2 motorway linking Warsaw (Mazowieckie) with the western border of Poland and the S8 express road connecting Warsaw with Wrocław. The former allowed for improved connections between Mazowieckie and some of the biggest Polish cities such as Łódź (Łódzkie) and Poznań (Wielkopolskie). The latter decreased travel time between the capital region Mazowieckie and Dolnośląskie, which has been highly industrialized and probably the most dynamic Polish region in the last two decades.

To test the above hypothesis we have applied a data at the lower level of territorial aggregation (NUTS3 regions). This allowed us to dispose of more observations at a more detailed scale. We have tested several specifications with different cut-offs. Still, we were not able to find the rate of reduction in travel time that would become statistically significant. Table 5 shows the results of selected specifications. Columns 1 and 4 refer to the full sample, columns 2 and 5 refer to regions with travel time improvement greater than 6% and columns 3 and 6 refer to regions with travel time improvement lower than 6%. It can be observed that for the full sample an increase in the volume of trade is positively correlated with an increase in regional GDP only, irrespectively the indicator used as a trade volume measure. Accessibility variable is not statistically significant and the coefficient has a very high standard error. Still, it appears that there exist certain differences between locations with accessibility improvement above and below the 6% threshold. In the case of the former (columns 2 and 5), trade volume is positively correlated with an increase in trade partners GDP. In the case of the latter (columns 3 and 6), this is exporter’s GDP growth that seems to stimulate trade volume. At least in the specification with millions of tkm. It also seems that accessibility improvement may potentially have a bigger impact on trade volume of regions below the 6% threshold. The value of the coefficient is much higher as compared to the regions above the 6% threshold. However, the coefficient is not statistically significant. Therefore, we find that accessibility improvement has no statistically significant impact on interregional trade volume in Poland between 2005 and 2015. Still, this is in line with the results of the previous study for Poland by Rokicki and Stępniak (2018) who proved that there was no correlation between accessibility improvement and production growth.

Table 5 Accessibility improvement and interregional trade between 2005 and 2015 (with 6% accessibility improvement cut-off) at the NUTS3 level

Trade value

Having analyzed the impact of accessibility improvement on the volume of interregional trade, we turn to verify whether there exist any correlation between travel time reduction and the value of interregional trade flows. Table 6 shows the value of trade in different manufacturing products between 2005 and 2015, based on doubly-constrained gravity estimations.⁸ It can be observed that the value of trade flows increases constantly in the case of all product categories both during the 2005–2010 and 2010–2015 periods. Hence, there is a clear difference as compared to the volume data. In the case of the latter, there was a small decrease in trade flows, measured in thousands of tons, between 2010 and 2015. As a result, we may wonder whether the results of formal econometric analysis could also differ.

Table 6 Evolution of interregional trade flows in manufacturing products in Poland between 2005 and 2015 (in PLN million, constant 2015 prices). 2008 CPA product classification

Table 7 provides the results of FE estimates based on interregional trade matrix, obtained using the doubly-constraint gravity model and RAS. It can be observed in the first column that accessibility improvement is positively correlated with the value of interregional trade flows in manufacturing goods between 2005 and 2015. The value of the coefficient is -0.66 and is statistically significant at 5%. Positive and statistically significant impact of accessibility improvement is also found for value of trade in coke, refined petroleum products, chemicals and pharmaceutical products (groups 15–17), basic metals and fabricated metal products (groups 20–21) and motor vehicles and other transport equipment (groups 25–26). In the case of other manufacturing products even if the sign of the travel time coefficient is as expected, it is not statistically significant.

Table 7 Accessibility improvement and interregional trade in particular products at the NUTS2 level between 2005 and 2015 (gravity based trade matrix)

The above results may indicate that, although accessibility improvement had no positive and statistically significant impact on the volume of interregional trade between 2005 and 2015, there exist positive correlation between travel time reduction and the value of trade. Even though, it applies to the selected groups of manufacturing products only. Still, the question is to what extent our conclusions can be influenced by the possible estimation bias during the regionalization of national supply and use tables. To verify the latter, we repeat the analysis applying interregional trade matrix, obtained using the entropy-based approach.

Table 8 shows the results of FE estimates based in the interregional trade matrix relying on the latter methodology. Indeed, they are quite different from the ones reported previously in Table 7. In particular, no specification features positive and statistically significant relationship between accessibility improvement and the value of trade flows. In this sense, the results are in line with the ones related to the trade volume. Higher R-squared statistics may also indicate that entropy-based trade matrix is more accurate for our estimates. Consequently, we may claim that our results do not prove the positive impact of travel time reduction on both the volume and the value of interregional trade flows in Poland between 2005 and 2020. This in turn, may explain to a great extent the surprising lack of positive relationship between accessibility improvement and regional GDP growth, reported in previous paper by Rokicki and Stępniak (2018).

Table 8 Accessibility improvement and interregional trade in particular products at the NUTS2 level between 2005 and 2015 (entropy based trade matrix)

To summarize, we find that there is no statistically significant relationship between accessibility improvement and an increase in the volume of interregional trade flows in Poland between 2005 and 2015. It appears, however, that there exist certain differences between locations with higher and lower accessibility improvement. We also find that the impact of accessibility improvement on the value of interregional trade flows our findings are less clear. Estimations based on the trade matrix obtained with the Horridge algorithm (with doubly-constraint gravity equation and RAS) suggest the existence of positive influence of travel time reduction on trade of certain manufacturing products. However, estimations applying the entropy-based trade do not confirm positive and statistically significant relationship between accessibility improvement and the value of trade flows. Given the fact, that the latter approach applies additional available information on the volume of trade flows, we assume that these estimates are more accurate.

Conclusions

This paper assesses the impact of road transport accessibility improvement on interregional trade in the case of Poland. We apply the data on the reduction of travel time and verify its relationship with bilateral trade flows between Polish regions from 2005 to 2015. We focus on road transport infrastructure given its dominant share in interregional trade found in other papers (e.g., Llano et al. 2017; Felbermayr and Tarasov 2022). To measure the trade flows we apply two different indicators. The first one measures the volume of road freight transport between particular NUTS2 level regions. The second one is expressed in pecuniary units and measures the value of trade between particular regions and industries. The data on trade value is obtained by regionalizing national supply and use tables, using either the Horridge algorithm (e.g., Rokicki et al. 2021) or the modified version of the entropy-based approach proposed by by Golan et al. (1994).

Our results indicate that there is no statistically significant relationship between accessibility improvement (or travel time reduction) and an increase in the volume of interregional trade flows between 2005 and 2015. This applies to trade flows between both the NUTS2 and NUTS3 level regions. It appears, however, that there exist certain differences between locations with higher and lower accessibility improvement. We find that the lack of significance of accessibility improvement is mainly due to the stagnation of interregional trade volume in Poland between 2010 and 2015. This in turn, may be attributed to an increasing deficit of professional drivers. In fact, the estimated deficit of drivers in Poland already reached 20% of optimal number of workers in 2015. Existing constraints on the supply side seem to be a good reason for producers to optimize their trade patterns and stick with the more profitable international trade routes.

In the case of the impact of accessibility improvement on the value of interregional trade flows our findings are less clear. It is, the estimations based on the trade matrix obtained with the Horridge algorithm (with doubly-constraint gravity equation and RAS) suggest the existence of positive influence of travel time reduction on trade of certain manufacturing products. However, the application of entropy-based trade matrix leads to different conclusions. In particular, no specification features positive and statistically significant relationship between accessibility improvement and the value of trade flows. Given the fact, that the latter approach applies additional available information on the volume of trade flows, we assume that these estimates are more accurate. Therefore, we may claim that our results do not prove the positive impact of travel time reduction on both the volume and the value of interregional trade flows in Poland between 2005 and 2015. This in turn, may explain largely the surprising lack of positive relationship between accessibility improvement and regional GDP growth, reported in previous paper by Rokicki and Stępniak (2018).

Our findings confirm the necessity to change the focus of transport infrastructure policies in countries such as Poland. Until now, the majority of transport investment projects, funded under the European Cohesion Policy programs, have focused on high-speed road and railway network (e.g., Rosik et al. 2015). Such an investment has been considered as a key tool fostering trade and economic development. It appears, however, that more focus should be put on the social impact of transport accessibility in order to address regional divide in Europe between core and peripheral regions (e.g., European Commission 2024). This implies that the priority should be an improvement of intra-regional and local connections rather than the investment in the high-speed transport network.

Since Poland is one of the European countries that presents higher levels of territorial inequalities, ensuring the availability of trade statistics at the subnational—regional and local—level could help us answer certain crucial research questions: the impact of R&D and/or institutional quality on Polish trade, given the development of statistics at the subnational levels for these determinants. In addition, what about estimating regional trade flows of sustainable transport given the uprising academic interest in this topic due to 2030 Agenda? This is not a comment, but to encourage you to conduct future research. If the authors manage to get data on exports from Polish regions to other EU regions, I suggest they work on testing their hypothesis that “Existing constraints on the supply side seem to be a good reason for producers to optimize their trade patterns and stick with the more profitable international trade routes.”