We estimate the parameters of the model using interregional freight flow microdata from the 2005 Net Freight Flow Census in Japan. The sum of these shipments for a given period becomes the company's production, compatible with the standard definition in the production model1. The cost of each shipment is the sum of the cost of input and toll if used as follows.

The toll on the highway depends on the distance and weight of the truck and is written as rijH rH( ,q dij). This differs from the conventional definition of output variables in the transport sector; i.e. the product of quantity and distance (qijdij with our notations). 4 We assume that the location of the transport company and the origin of the trip are the same, so the wage rate is applied at the origin.

The size of the truck (defined by the weight category without load) is determined so that the truck can accommodate a load of size q.9 The fuel efficiency ( , )e q s of trucks is usually an increasing function of the total weight of the truck qw qT( ) and a function of speed s in the form of a letter U. If Z3 is large, there are not enough trucks in the region for the amount of goods that need to be taken out of the region. It is likely that the transport time t depends on the distance d between home and destination, but d is exogenous to the transport company because it is determined by the order of the shippers.

Applying the OLS estimate to (3.5), we obtain 2SLS estimates of , that are consistent under endogeneity. 3.5) differs slightly from textbook 2SLS in the sense that some endogenous variables are multiplied by exogenous variables.

## Data and Empirical Results

The 2005 Census uses 16,698 samples of domestic establishments randomly selected from approximately 683,230 establishments engaged in the mining, manufacturing, wholesale and warehousing industries. The transport distance data d can be calculated using NITAS from the origin and destination information of each shipment in NFFC. The average hourly wage of the driver in the prefecture of origin RiL is calculated using the data on monthly contractual earnings, scheduled hours worked and overtime for drivers of small, medium and large trucks.

To construct a target data set for our analysis, we first abstract from the full data set the data on the shipments that used trucks as the main mode of transportation, and then remove shipments with the following conditions: (1) Since this study focuses on the trucking industry, we exclude observations in regions, that are inaccessible via a road network. We believe that 2SLS is the appropriate estimation method in the current model and data15. 15 We implemented 2SLS estimation for different sets of instruments based on the discussion in Section 3, namely we take (3.3) and (3.4) in the first stage regression.

The difference is that we use or do not use Hˆ in the first-stage estimation of t. We expect the sign of the estimates as given in Section 3, which is also tabulated in Table 2. We can also consider the case where there is no load on the return trip .

We cannot discuss the appropriate level as it depends on the mileage parameter of trucks. As in the case of the labor coefficient 1, we expect this value to be in [1,2], because if the trucks have no goods on the return trip, they may prefer to pay the highway toll for two legs to the shippers to charge. With this result we can calculate the effect of an increase in the number of truck companies.

We choose 0.3 as the default value based on the discussion at the beginning of this section. We construct wT as indicated in the previous section, but it should include noise that may not be ignorable. In order to investigate the product-specific effects on the freight price, we also estimate the model for each product.

The table shows that the variations in unit freight costs for different combinations of q and d are quite large; e.g. from (1, 50)P 431.66 (via highways) to. We also see that the effects of changing the lot size or distance vary depending on the level of q and d.

## Conclusion

As noted in footnote 1, the majority of existing research on motor carrier cost structures is based on firm-level data, and the motor carrier industry is reported to have constant returns to scale technology. Please note that freight costs per shipment are the actual transportation costs perceived by shippers and must be taken into account when making various decisions, such as the choice of factory location and the geographic extent of the shipping destinations (i.e. market area ). To gain quantitative insights, we calculate the values of freight costs per tonne-km for different combinations of q and d, as in Table 7.

This calculation incorporates the effect of lot size through choice of truck size which is ignored in the calculation of elasticities since marginal change in q does not w qT(. Differences between the two cases include the effects of various factors working in opposite directions, such as e.g. shippers ’s higher willingness to pay (+), trucking companies’ cost savings from shorter transit time (-) and toll payment Despite these results, it is somewhat surprising that the unit freight costs have similar values as the products of q and d , q d , are the same.

This suggests that the conventional definition of production, tonne-km, has proven to be a good approximation. The paper presents a microeconomic model of interregional freight traffic, based on a careful formulation of the cost structure in trucking companies and market equilibrium, which takes into account the characteristics of transport services as a set of several characteristics. The estimation results show that the transport cost determinants included in the model have significant effects in a manner consistent with theoretical predictions.

Significant economies of scale with respect to lot size and long-range economies are shown to exist. First, time is a very important determinant of transport costs, as shown in the regression results. Shippers have an increasing willingness to pay for fast delivery, while trucking companies benefit from the opportunity cost savings of labor (drivers) and capital (trucks).

Literature on the estimation of the value of transport time savings in freight transport is relatively scarce compared to that on passenger transport. It would be worthwhile to develop a methodology to measure the value of time using freight microdata. In this regard, we should note that transit time is an endogenous variable, which shippers and trucking companies choose to optimize some objective.

Companies use advanced information and communication technologies and build their own infrastructure, such as terminals. To address this issue, a methodology should be developed to define and measure productivity in the transport sector, for which conventional methods such as total factor productivity (TFP) in the manufacturing sector are not applicable.

Relation between costs in firm level and shipment level

Descriptive Statistics

Classification and Commodity