Evaluating the modal split of overland transportation of general cargo in Brazil using a market share model

This paper presents a methodology developed to evaluate the modal split of land transport of general cargo in Brazil and possible reasons for the prevalence of road transport over rail, using a market share model. No studies were found using the market share model for planning cargo transport, in spite of its generic nature which makes it readily applicable to any market, service, utility or commodity. The methodology was applied in three transport corridors, two groups of general cargo and enabled us to establish that Brazilian rail operations are still uncompetitive in terms of logistics costs when transporting general cargo of high aggregate value. Moreover, transport supply is a determinant factor for the greater use of road transport in Brazil’s general cargo transport matrix, and the demand for general cargo transport in Brazil is elastic in relation to this factor. For low aggregate value general cargo (AV1), a 1% reduction in the gap between road and railway supply leads to an increase in the railway market share of 4.5% through TC1 (São Paulo Porto Alegre São Paulo), 4.9% through TC2 (Santos Brasilia Santos), and 3.3% through TC3(São Paulo Rio de Janeiro São Paulo). For high aggregate value general cargo (AV2), this elasticity was more pronounced. A 1% reduction in the gap between road and railway transport supply leads to an increase in the railway market share of 4.2% through TC1, 11.9% through TC2, and 9.6% through TC3. www.journal-of-transport-literature.org JTL|RELIT ISSN 2238-1031 B T P S Brazilian Transportation Planning Society


Introduction
This paper presents a methodology developed to evaluate the modal split of land transport of general cargo in Brazil and possible reasons for the prevalence of road transport over rail, using a market share model.
Using railways to transport general cargo 1 in Brazil is still not a prevalent practice. Based on ANTT (2004), ANTT (2011) and ANTAQ (2010), the authors have estimated that railways transported a mere 7.9% of all Brazilian general cargo in 2011; approximately 25.5 million tons. In contrast, the more popular road transport mode accounted for 85.3% of cargo transportation in the same year; approximately 275.9 million tons.
Brazilian transport policy guidelines are attempting to establish a more equitable transport matrix with priority being given to the use of more economical and environmentally sustainable transport modes, particularly rail and waterway. The Brazilian government is therefore endeavoring to exploit the comparative advantages of different modes of transport and to integrate and combine their operations to achieve safer and more economical intermodal handling of goods.
Transport studies use various methodologies but new advances have been made in economics that can be useful for transport analysis. Train (2003) indicates that there are three kinds of approaches used in discrete choice models: disaggregate models, aggregate models and those using the dependent variable in the estimation process. The second approach is widely used in market economic structure analysis as, for example, in the works of Basuroy and Nguyen (1998), Nevo (2000) and Train (2003). No studies were found using Nevo's (2000) approach, however, in spite of its generic nature which makes it readily applicable to any market, service, utility or commodity. 1 The main general cargo handled in Brazil comprises construction supplies, steel products, agricultural supplies, processed foods, beverages, electronics, home appliances, and automotive products (ANTT, 2004). In this context, a methodology based on a market share model has been designed aimed at identifying and measuring factors that influence the greater use of road over rail transport in Brazil. The methodology is a useful tool for sector analysis and for planning cargo transport studies.
Section 1 presents the structured scientific methodology used for analysis, gaining an understanding of the problem and unfolding the research. Section 2 describes the study undertaken to analyze three important Brazilian transport corridors. In Section 3 there is a discussion of the results and lastly, the Section headed "Conclusions" presents the paper's final remarks. Figure 1 is a schematic illustration of the methodology. Details of the stages are set out in 2.1 to 2.6.

Stage 1 -Initial Settings
The initial settings in Stage 1 define the scope of the analysis, including the type of cargo and transport modes to be analyzed, a description of a representative sample, and a characterization of the analysis, including level and key aspects.

Stage 2 -Attribute Selection
Stage 2 selects the attributes to be incorporated to the model. They should reflect possible factors determining the use of the modes of transport under analysis and are investigated using research techniques such as literature reviews, document searches, field research, and interviews.

Stage 3 -Indicators and Attribute Measurement
Stage 3 identifies the indicators needed to measure the previously selected attributes.

Stage 4 -Data Collection, Estimates, and Assessment of Attribute Indicators
In Stage 4 data is collected for assessing the attribute indicators. This information is necessary for the subsequent model fitting.

Stage 5 -Model Fit
In Stage 5, the modeling technique, the model specification, and its functional form are first defined, and, if necessary, the appropriate software to fit the model is researched. The indicator values measured in Stage 4 provide the data for calculating the model's coefficients.
With the model defined, the signs of the coefficients estimated for each variable are checked for theoretical consistency. Then, statistical analysis is performed to evaluate the significance level of both the model and the variables.

Stage 6 -Scenario Analysis
Stage 6 analyzes the impact on demand stemming from changes to the parameters of the transport modes under consideration, using the fitted model. Changes in demand and the subsequent analysis are the result of a direct application of the model fitted in Stage 5.
Once Stage 6 is complete, the researcher may want to review the model, which is represented in the proposed procedure by means of a link between Stage 6 and Stage 1.

Methodology applied to a study
In sections 2.1 to 2.6, we apply the methodology illustrated in Figure 1.

Stage 1 -Initial Settings
During this stage, the scope of the Brazilian general cargo industry is analyzed with the aim of developing a model intended for the cargo transport strategic planner. The idea is to fit a modal split model that would consider the key attributes determining the choice between road and railway transport in Brazil, taking into account transfer operations between plants, ports, and distribution centers.
The demand for Brazilian general cargo allocated to the main transport corridors in the country was identified based on official data from the National Overland Transport Board (Agência Nacional de Transportes Terrestres) (ANTT, 2004). This demand was divided into two groups: general cargo of low aggregate value (AV1) and general cargo of high aggregate value (AV2). General cargo from the AV1 group has an average value of 25,000.00 R$/TEU 3 .
The main products are construction supplies, steel products, and agricultural supplies. General cargo from the AV2 group has an average value of 75,000.00 R$/TEU. The main products are processed foods, beverages, electronics, home appliances, and automotive products.
For the subsequent analysis, the three transport corridors with the greatest flow of general cargo were selected: TC1 (São Paulo -Porto Alegre -São Paulo); TC2 (Santos -Brasilia -Santos); and TC3 (São Paulo -Rio de Janeiro -São Paulo). Figure    Railways transported 10.0 million tons of general cargo through these three transport corridors in 2011 (ANTT, 2011); approximately 39.3% of the total Brazilian general cargo moved by rail in that year. The rail cargo in the three corridors was subdivided as follows: 3.2 million tons through TC1; 3.9 million tons through TC2; and 2.9 million tons through TC3.
The cargo transported in these corridors was mainly general cargo of low aggregate value (7.7 million tons) and non-containerized general cargo (8.8 million tons).

Stage 2 -Attribute Selection
The literature review summarized in Table 1 was performed to select the attributes to be incorporated in the model. It showed a high incidence of cost-related attributes (32 observations) and service levels (44 observations). Thus, based on the review and prior knowledge of the problem under consideration, it was decided to choose one cost attribute, logistics cost, and one service level attribute, transport supply, as explanatory variables in the model.  Baumol and Vinod (1970) x T samboulas and Kapros (2000) x Cullinane and T oy (2000) x Shingal and Fowkes (2002) x x x x 10 Staton and Senna (2003) x Effron and Rose (2003) x  Kutanoglu and Lohiya (2008) x

Stage 3 -Attribute Indicators and Measurements
The Table 2 shows indicators and measurements used to determine the time series of logistics costs and the levels of road and railway transport supply on the selected transport corridors.
The measurement of logistics costs was taken to be the sum of the transport, stock, handling, and storage costs between a source and a destination, as proposed by Baumol and Vinod (1970). To measure transport supply, we assumed that the supply of both road and railway transport for general cargo in a specific transport corridor is proportional to the amount of general cargo handled by the transport mode in the corridor in question.

Stage 4 -Data Collection, Estimates and Assessment of Attribute Indicators
During this stage, we collected the required data, representing logistics operations already performed for the transport of general cargo. Table 3 also shows the sources used to gather costs and transport supply statistics. From this collected data, it was possible to estimate and assess the indicators of the selected attributes, thereby creating the database we required to fit the modal split models. This database contained biannual time series of road and railway, market share, logistics costs, and transport supply data.

Stage 5 -Model Fit
Logit The market share model, adapted from the researches works of Basuroy andNguyen (1998), Nevo (2000), and Train (2003), was chosen to fit the modal split models for each of the selected transport corridors and groups of general cargo. As Nevo (2000) has shown, the market share model is developed from an adaptation of logit-type models, assuming that the probability of choosing a transport alternative is equal to the market share of that alternative.
Thus, the market share of an alternative is given by Equation 3. The time series data shown in Table 3 was used with the GRETL® software to fit the market share model for each transport corridor and general cargo group.

Stage 6 -Scenario Analysis
To analyze the sensitivity of the coefficients obtained in Stage 5, we calculated the difference between the rail transport market share estimated for the second half of 2011, and the rail transport market share for a hypothetical situation resulting from changes in the values of the explanatory variables of the model.
The calculation of the rail transport market share was performed based on Equation 5, considering that the sum of the rail and road market shares is 100% (Equation 6). From these two equations, we obtained Equation 7, which determines the rail transport market share of a given transport corridor.

Results and Discussion
An analysis of the results obtained at each stage of the applied procedure showed that the handling of general cargo in Brazil is heavily concentrated in the southern and southeastern regions of the country. As a result, the three analyzed transport corridors (results from Stage 1) handle 61.6% of the total cargo transported. An analysis of the general cargo transported through these corridors by rail showed that the average aggregate value and the level of containerization of these cargos are both low, suggesting that rail logistics costs and rail transport supply are still uncompetitive for goods with higher aggregate value.
The literature review detailed in Table 1 shows that these authors agree with the selection of attributes used to model the modal choice, usually selecting the attributes of cost and service level. However, there is no consensus regarding the most suitable indicators to measure these two attributes. An analysis of the problem proposed in this paper suggests that logistics costs and transport supply are adequate indicators, although, in fact, only transport supply has shown satisfactory statistical results.
The collection of data, data analysis, and measurements performed in Stage 4, as shown in Table 3, showed little variability over time for the observed values of the variables Market Share, Logistics Costs, and Transport Supply for any of the transport corridors and general cargo categories analyzed.
The average market share analysis of road and rail operations for general cargo categories, during the observation period (Figure 3), showed that, with the exception of low aggregate value general cargo through TC2, the remaining transport corridors have a road market share above 90%. This confirms the greater tendency to use road transport for general cargo in the country. The average rail transport market share of 76.6% through TC2 for the transportation of general cargo in the AV1 category shows that a significant amount of agricultural and construction supplies is transported by rail through this particular corridor.

Figure 3 -Market share in transport corridors 8
An analysis of the average logistics costs of road and rail operations for general cargo categories during the observation period ( Figure 4) showed that, for the three transport corridors analyzed, rail logistics costs are competitive when transporting general cargo of low aggregate value (AV1). This finding reflects the fact that some Brazilian rail operations have been designed for door-to-door delivery of these cargo categories, with railway branch lines connecting factories to storage centers, thereby avoiding any additional cargo transfer or road transport costs.
For the high aggregate value general cargo group (AV2), rail transport has a small competitive advantage in terms of logistics costs, but only through TC2, where the average logistics costs of road-railway intermodal operations is 1.2% lower than the unimodal road operation. This advantage reflects the competitiveness of rail freight for average distances above 600 km, as is the case in this transport corridor. In contrast, the transport figures for these cargos through TC2 and TC3 show that road-railway intermodal transport is still uncompetitive in them. This finding reflects a scenario in which railway fares are still uncompetitive when compared to road freight charges and, furthermore, there are additional cargo transfer and road transport costs associated to the rail operations.

Figure 4 -Comparison of logistics costs 9
An analysis of the transport supply side ( Figure 5) showed that the supply of road transport for general cargo in Brazil has been growing steadily over time, being historically much more significant than the supply of rail transport for this type of cargo which has remained virtually unchanged. As a result, the gap between road and rail transport supply, which was An analysis of the fitted models (Table 4) shows that all six models had an adjusted R-Square value above 0.60. An analysis of the coefficients of the logistics costs variable ( 1) showed that these reached a significance level above 90%, but only for Models III and VI. However, in these models, these coefficients were positive, which is inconsistent with the expected result that the logistics costs variable would be inversely proportional to the market share variable. This may reflect the fact that general cargo demand was evaluated in an aggregated manner, segmented into two groups of general cargo only. Furthermore, the results may be a reflection of the sample size, the lack of data variability and the existence of endogenous variables that were not incorporated to the modeling. Thus, the data analysis performed in Stage 4 did not consider points where there is a migration of general cargo from road to rail as a result of decreases in railway logistics costs.
An analysis of the coefficients of the transport supply variable ( 2) showed that these attained statistical significance for all six fitted models. Moreover, in all the models, these coefficients were positive, which is consistent with the expected result that the transport supply variable would be directly proportional to the market share variable. These results show that the transport supply variable was the key explanatory variable in the fitted models, reflecting the fact that the largest market share for the road transport of general cargo is highly correlated to the large supply of road transport and low supply of railway transport for this market niche. In Stage 6, we analyzed the changes in the rail transport market share relative to the transport supply variable. 12 To perform this analysis, the gap between road and rail transport supply was reduced by 1% in each of the analyzed transport corridors by assuming that the road transport fleet has remained unchanged since the second half of 2011. Changes in the rail transport market share thus obtained are shown in Figure 6.
The analysis shows that for both low aggregate value (AV1) and high aggregate value (AV2) general cargo, demand was elastic relative to fluctuations in the values of the transport supply variable.
For low aggregate value general cargo (AV1), a 1% reduction in the gap between road and railway supply leads to an increase in the railway market share of 4.5% through TC1, 4.9% through TC2, and 3.3% through TC3. For high aggregate value general cargo (AV2), this elasticity was more pronounced. A 1% reduction in the gap between road and railway transport supply leads to an increase in the railway market share of 4.2% through TC1, 11.9% through TC2, and 9.6% through TC3. These numbers may reflect the high potential of rail transport of general cargo in the AV2 category through TC2.
TC1 has a similar attractiveness for both AV1 and AV2 cargos, with potential to increase by 447 tons per car for AV1 group cargos and 408 tons per car for AVA2 group cargos, respectively. These numbers may stem from the fact that the railway supply for both AV1 and AV2 groups through TC2 are very similar.
TC3 is the least attractive for rail transport of the AV1 group with potential to increase by 357 tons per cargo fleet unit only. However, for AV2 cargos, TC3, as in TC2, is very attractive.
Here, TC3 could potentially increase by 843 tons per cargo fleet unit. This number may also reflect the high potential for the rail transport of general cargo from the AV2 group through TC3.

Conclusion
The aim of this paper was to evaluate the modal split of land transport of general cargo in Brazil and possible reasons for the prevalence of road transport over rail transport, using a market share model.
Applying the methodology developed in this study has provided an insight into the land transport of general cargo in Brazil, providing a unique and novel portrait of the recent behavior of demand, logistics costs, and transport supply for this market niche.
Results have shown that the preferential use of road transport, to the detriment of rail transport, for general cargo in Brazil is due to the low competitiveness of the railways in terms of logistics costs for transporting general cargo of higher aggregate value, which leads us to suggest that Brazilian railways are not yet ready for the large-scale transportation of such goods.
In addition, another determining factor for this road transport predominance is the existence of a growing supply of road transport in contrast to the small and virtually unchanged supply of rail transport. Based on the supply elasticity analysis, this scenario is likely to change, should the rail transport supply in Brazil come to be stimulated and promoted.
The major limitations of the developed model were the unavailability of time series data concerning the handling of general cargo on Brazilian roads and railways prior to 2006, and the lack of demand data classified according to a wider range of goods categories. These limitations precluded a more disaggregated analysis of the goods that are part of the scope of general cargo in Brazil.
The methodology presented in this paper is general enough to be applicable to other related research. It is suggested that future studies on the topic should: analyze why the use of railway containers for transporting general cargo in Brazil is so limited; analyze the rail transport offer for general cargo, and investigate the feasibility and impact on demand and on the environment of establishing intermodal logistics operations for the transport of general cargo.