796 906 1 PB
796 906 1 PB
796 906 1 PB
AgEcon Search
http://ageconsearch.umn.edu
aesearch@umn.edu
Papers downloaded from AgEcon Search may be used for non-commercial purposes and personal study only.
No other use, including posting to another Internet site, is permitted without permission from the copyright
owner (not AgEcon Search), or as allowed under the provisions of Fair Use, U.S. Copyright Act, Title 17 U.S.C.
Transportation Research Forum
Many supply chain and finished goods distribution trip is accomplished by moving the truck trailer
networks involve intercity freight transportation. on rail flatcar. Projecting capacity expansion
Freight shippers secure transportation services needs for intercity trucking and intermodal is
by matching their requirements to available particularly challenging because the demands
service in an effort to minimize their total for these products are closely related; yet they
logistics costs subject to service level constraints. utilize different transportation infrastructures
For these supply chains to operate efficiently, the whose capacities are planned by different
appropriate transportation infrastructure must be decision-making entities.
in place to support them. Historically, statistical forecasts based
The U.S. government spends more than $95 on aggregate data such as regional population
billion each year on roads, and U.S. Class One growth and production indexes have been used
railroads invest more than $6 billion each year on to create forecasts of freight shipping patterns.
their infrastructure to support these distribution However, the implicit assumption employed is
networks. In making their capacity expansion that the observed transportation modal choice will
decisions, governments and railroads are faced grow proportionally with the freight volumes.
with the difficult task of projecting where In fact, data on shipper modal choice is subject
traffic growth potential is greatest. This paper to short-run capacity limitations and may not
focuses on estimating the potential demand for accurately reflect the long run market potential
intercity trucking and rail intermodal to plan the for capacity expansion because shippers may be
U.S. freight infrastructure. Intercity trucking under utilizing a mode that is at capacity. Thus,
is the largest intercity freight transportation as population and production grows, expansion
mode. Rail’s fastest-growing and most truck- of modal capacity is as influential as forecasted
competitive product is intermodal, in which the population and production on freight growth.
first and last leg of the shipment is completed by Unfortunately, governments and individual
truck, and the middle, long-haul portion of the transportation service providers generally
141
Logistics Costs
measure the observed modal choice of shippers. The following section describes contribut-
Thus, decision makers are not well-positioned to ing literature, followed by a section describing
measure the potential (uncapacitated) demand the modeling assumptions and methodology.
for freight transportation which is subject to The fourth section describes the model’s data
infrastructure capacity constraints. sources and is followed by results, extensions
Attention to short-run modal mix may result and conclusions.
in a large investment in expanded Interstate
highway infrastructure; i.e., extensive Interstate LITERATURE REVIEW
capacity allows for large truck volumes, and
these large truck (and passenger) volumes are Recent research can be divided into the areas
the justification for expanded Interstate capacity. of infrastructure capacity planning, shipper
On the other hand, to earn an effective return on modal demand estimation, and shipper logistics
investment for its capital, railroads are careful optimization.
not to over invest in their rail infrastructure
capacity and consequently grow more slowly Freight Transportation Infrastructure
and under much tighter capacity conditions. Planning
To better understand the optimal mix
of freight infrastructure, in this study the Extensive research on optimal network structure
unconstrained market potential demand for truck and predictive freight flows has been conducted.
and intermodal shipments is estimated based on Early work by Roberts (1976) and Kresge and
a detailed modeling of the individual shipper’s Roberts (1971) focused on aggregate freight
modal choice decision which is then applied to flows for the purposes of national planning of
the market at large. The extent of potential long- freight infrastructure. Bronzini and Sherman
run demand is estimated for each mode through (1983) build a single-commodity modal route-
micro-modeling of the individual shipper’s choice model based on route impedances.
minimization of total logistics costs (the sum of Crainic, Florian and Leal (1990) develop a
the transportation cost plus all inventory holding detailed optimization model which describes
costs associated with the shipment) in individual the optimal freight transportation infrastructure
transportation markets, and then the methodology given exogenous modal traffic flows. These
is applied empirically on a macroeconomic basis studies do not explicitly model the inventory
to estimate the uncapacitated market-wide holding costs of specific shippers, origin and
demand for each mode. destination pairs, and commodity types, nor
Differences between the model flows do they consider the intermodal transportation
based on minimum total logistics costs and option.
historical freight flows can be interpreted as Recent research on optimal provision of
short-run deviations from a long-run optimum. transportation infrastructure has been conducted
Deviations from model predicted flows could by authors such as Conrad (2000), Nash (1993)
be due to a number of sources, including: model and Winston (1991). Nash (1993) provides a
specification or input data error, shipper modal framework for thorough cost benefit analysis
choice decision with information gaps or modal of transport infrastructure projects through case
preference, or modal capacity shortfalls. In the study examples of road and rail investment
long run, short-run anomalies in modal choice from Great Britain. Winston (1991) focuses on
will tend to be eliminated if the capacity exists empirical characterizations of efficient roadway
in that mode to accommodate shippers’ modal investment and issues such as congestion
choices. The implication for transportation pricing in the U.S. Conrad (2000) presents
service providers is that the right capacity type a comprehensive analysis of transportation
must be created in the long run to support those infrastructure investment based on detailed
shippers who will, based on their total logistics microeconomic modeling of transportation as
cost function, have sufficient potential demand an input to the economy’s overall performance.
to justify expanded capacity investment. Conrad (2000) tests the model with an
142
Logistics Costs
econometric model of the German economy. component of the shipper’s decision because
These studies presume a central (government) of data availability. Oum considers 69 regions
authority is establishing transportation and eight commodity groups; Freidlander and
infrastructure policy for all modes. Spady considers 96 commodity groups, but
only five broad geographical regions. By way of
Freight Transportation Demand comparison, this study is based on 250 Standard
Transportation Commodity Codes (STCC) for
Kremers, Nijkamp and Rietveld (2002) perform over 7,200 origin and destination zip codes in the
a meta-analysis of 25 different studies of shipper United States, 75,000 product lanes in total.
demand for various transportation modes. Finally, Fernandez, et al. (2003) have
Although slightly dated, Oum, Waters and proposed a novel mathematical general
Young (1992) provide a more comprehensive equilibrium model for the supply and demand
survey of more than 60 refereed articles that of freight infrastructure, but have not applied
estimate shipper demand elasticities. In most their work in an empirical setting.
of these studies, the demand for transportation
is typically modeled as a derived demand for Freight Transportation and Logistics
an input as a component of total costs facing Optimization
a firm. The objective of these studies is to
statistically characterize the price sensitivities Sheffi, et al. (1988) looks at transportation/
and price-service tradeoffs for shippers making inventory tradeoffs facing a shipper deciding
modal choice decisions. Statistical estimation between truck and rail. Optimal mode choice is
of demand curves is based on the assumption defined as the mode which minimizes the total
that observed quantities represent the shipper’s logistics costs (TLC) of the shipper, including
long-run, unconstrained demand for the various transportation fees as well as inventory costs
modes. At any point in time, shippers are making which arise from shipment speed, reliability and
short-run modal choice satisfying decisions equipment capacity (lot size). This method was
based on the available capacity in a mode. For designed as a decision support tool for shippers
capacity planning exercises, this quantity can making modal choice and a marketing tool for
be misleading as a long-run indicator of modal a U.S. railroad to better understand and cater
demand. Additionally, regression estimates of to shipper freight transportation needs. Sheffi
these demands are specific to mode, commodity, focuses on the microeconomic issues and does
origin-destination pair, price for competing mode, not utilize the TLC model to examine the market
equipment type, service level and prevailing level demand for each mode nor to estimate the
economic conditions (such as interest rates requirements for capacity expansion.
and rents). Thus, estimation of such elasticities A modified form of Sheffi’s TLC model
across a large number of commodities and freight is developed and tested empirically against
transportation lanes is impractical. thousands of actual historical U.S. freight flows.
Typically, the modes of transportation To the extent shippers optimize mode choice, the
have been considered in isolation in these TLC model should predict U.S. freight flows of
transportation demand-elasticity studies. truck and rail accurately. However, the extent
Notable exceptions are Oum (1979), which to which the TLC model and actual flows differ
looks at the price-service trade-off for a subset represents long-term modal shift potential.
of Canadian shippers choosing between truck
and rail, and Friedlander and Spady (1980), MODELING METHODOLOGY
which performs a similar analysis for U.S.
shippers. Neither study explicitly models the This paper models the shipper’s modal choice
inventory holding costs experienced by the decision as in Sheffi’s (1988) total logistics
shipper due to various shipment attributes, nor cost (TLC) model which models the trade-offs
the value of understanding shipper behavior for between transportation mode decisions and
establishing optimal infrastructure investment. inventory holding costs.
Both studies vastly simplify the geographic
143
Logistics Costs
The inventory required to support a shipping shipment (lot) size is the rate of consumption of
decision is a function of lot size, shipment speed, the product at destination. Given the aggregation
and shipment time variability. The lot size of of shippers by geographic region, it is impossible
the shipment affects average inventory at both to discern the rate of demand between any single
origin and destination. Shipment speed affects shipper and receiver. Thus, it is not possible to
the pipeline inventory – the amount of inventory establish optimal lot size based on that demand
in transit. Shipment time variability affects the rate. For the purposes of this model, lot size
required safety stock - inventory at destination specification is not necessary for the following
to protect against stockouts due to transportation reasons. First, only freight traffic which is
time and demand variability. currently moving in full truckload lot sizes is
Key attributes of the shipment include: considered in this analysis, either in over-the-
• Transportation cost – the door-to-door fee road truck, or in rail intermodal, and therefore
paid to the transportation service provider(s) whose EOQ is approximately one truckload.
for moving product from origin to destina- Because of the 10-fold cost premium and one-
tion. As in Sheffi (1988), and given the as- third of a full truck lot size which is typical in
sumption of full truckload shipments this less-than-truckload service, it is unlikely that a
cost is accurately defined as dollars per large portion of traffic under consideration is
shipment, errantly moving in full-truck service when it
• Shipment size –the capacity of the transpor- should be moving in smaller lot sizes.
tation vehicle which limits the total ship- For larger-than-full-truckload lot sizes, a
ment size (expressed in tons). Depending on typical rail car is two to four times larger than
the commodity being shipped, the vehicle the full truck-lot size. Although full railcars
may “weigh out” due to legal restrictions on move at a significant discount relative to truck,
truck size, or “cube out” as the trailer space approximately 20% of all shippers have rail-
is filled. direct service, so the probability of a rail-served
• Transit time – total expected elapsed time origin and a rail-served destination is less than
from origin to destination. The shipper must one in 20 for any typical flow. It is assumed
carry the inventory cost for this “pipeline” that the minority of shippers that have both rail-
inventory in transit. served origin and destination are well versed
• Transit time variability – variance around in rail economics and are utilizing a railcar
the expected elapsed time of delivery. The shipment lot size when possible. Following
shipper must carry “safety stock” inventory this logic, the focus of this study is on historical
to protect against the uncertain timing of full-truckload-shipment lot sizes and the truck-
shipment arrival. versus-intermodal decision, excluding other lot
• Inventory carrying costs – the total cost of sizes from the analysis.
owning the shipped items while in transit A more subtle EOQ issue arises when
and at destination before sale. Inventory considering various truck sizes. There are
holding costs are a function of the value of two predominant equipment capacities: 48-
the commodity being shipped, and the in- foot and 53-foot trailers and containers. When
ventory carrying charge (typically expressed evaluating the economics of shipment, the
in a percent of the total value of the com- differences in rates for transportation for each
modity per year). of these equipment sizes is included. However,
Shippers minimize the total logistics cost of in practice, these options do not appreciably
a shipment by minimizing the transportation cost affect the EOQ decision. A full truck can be
of shipment plus the inventory cost associated defined as one that has reached the maximum
with the mode of choice. legal weight allowed (“weighed out”) or that has
used all the space in the trailer (“cubed out”).
Lot Size Each commodity type has a different density;
more dense product weighs out before it cubes
As is well known from the economic order out and moves in 48-foot equipment and less
quantity (EOQ) literature, a major determinant of dense product cubes out before it weighs out and
144
Logistics Costs
moves in 53-foot equipment. Because density Construction of Shortest Paths
of a commodity is fixed, each commodity is
projected to always move in the same equipment For each shipment analyzed, the road distance
size and thus a “full truck” is always the same and expected travel time is used for calculating
lot size regardless of the multiple options in the total logistics cost of a truckload move. For
equipment size. Thus, by limiting the analysis the rail network move, minimum total logistics
to full truckload quantities, the transportation cost of a move can be found using Dijkstra’s
buying decision is simplified to one based strictly algorithm (Dijkstra, 1959) employing a
on the transportation cost/service continuum. method similar to that used by Barnhart and
Ratliff (1993). Because of the large number
Inventory and Transportation of shipments and paths to consider and the
size of the network, this method proved to be
Queuing of inventory at origin and destination of computationally burdensome.
shipment is the same regardless of mode because A greedy heuristic method performed
lot size is fixed in this study. Accordingly, the admirably because of the special structure of the
model can focus strictly on pipeline inventories problem. For the trucking option, the problem
in transit driven by transit time, and safety stock is trivial. A single arc from origin to destination
of inventories because of variability of transit includes the truck rate plus the total inventory
time. holding costs for the transportation time. For
To explicitly measure the cost of service the rail network, each path from origin to
variability, the model measures the cost of destination consists of three legs: an origin dray
holding safety stock inventory at a destination and destination dray connected by rail long haul.
to protect against stockouts. As in Sheffi (1988), Because the cost for dray is typically four times
the requirement of the shipper is specified the cost per mile of the rail component of the
in terms of a required “fill rate” which is the move, it is usually true that the shortest-cost path
percentage of the time that demand is met from through the rail network is via the origin ramp
in-stock inventory. The required fill rate is and destination ramp closest to shipper origin and
assumed to be 98%, meaning that shippers plan destination. However, because of differentials in
on the 98th percentile arrival time to meet their rail rates and service that are origin-destination
customers’ orders.1 The 98th percentile measure specific, occasionally a longer dray is warranted.
of transportation arrival time then includes all Thus, the nearest five origin and destination
inventory required to be held to account for both ramps are considered as candidates for routing,
longer average duration as well as any arrival and the lowest cost path based on that subset of
time variability in the transportation mode. the paths is chosen for comparison to the cost of
Thus, the model captures both pipeline and a truckload move.
safety stock. As modified from Sheffi (1988), For capacity planning purposes, a
the total logistics cost (TLC) per unit is: conservative approach is taken for estimating
potential rail demand. First, intermodal options
(1) TLC = TC + V * i * t are limited by considering only rail ramps within
500 miles (one day’s trucking distance) of origin
where TC is the transportation cost per ton, V is and destination. Typically, a one-day dray is
the value of the product per truckload (value per the maximum that is economically advisable.
ton times tons per shipment), i is the inventory Second, any intermodal option that has a 98th
holding costs (measured as a percentage), and percentile transit time that is in excess of two
t is the expected 98th percentile transit time of days longer than the trucking option in that lane
the shipment. is excluded. Although longer transit times can
be accommodated by additional inventories,
145
Logistics Costs
this constraint is imposed to recognize an upper The model compares the total logistics cost
bound on a shipper’s willingness to accept in this lane and chooses the option with the low-
longer shipment times, regardless if such time est total logistics costs. In this illustration, the
and increased inventory is justified economically lower transportation cost of intermodal justifies
by lower transportation costs. In both cases, the holding additional inventories and the intermo-
rail options are limited and costs of rail are dal option is superior in terms of total logistics
overstated, thereby reducing the estimated costs. Although in this illustration truckload is
total opportunity for shippers to convert to two days faster than intermodal and therefore
intermodal. requires 33% less inventory ($90 for rail, $60 for
Figure 1 illustrates the calculus for a ship- truck), the transportation cost premium does not
ment choice from Los Angeles to Chicago for justify the reduced inventory requirements. By
a fictitious shipper. The illustrative commodity, modeling this decision thousands of times across
general mixed freight, has an average value per multiple geographic regions, commodities and
ton of $2,000 according to estimates from the shipper proximity to rail ramps, the model can
Bureau of Transportation Statistics, Commodity help identify the traffic flow patterns that seem to
Flow Survey. Using an inventory holding cost diverge from TLC model recomendations.
of 18% per annum and 15 tons per truckload,
the cost per day of a truckload of inventory in DATA SOURCES AND EMPIRICAL
transit is $15 per day per truck ($2,000 per ton METHODOLOGY
* 18% per year * 15 tons per truck/365 days per
year). The truck transportation price is the door- Year 2000 U.S. freight traffic flows were
to-door truckload rate ($2,350); the rail price is analyzed to evaluate the potential demand for
the sum of origin and destination dray, rail line each mode evaluating each shipper’s actual
haul, and Intermodal Marketing Company (IMC) modal choice decision against the total logistics
fees ($1,820). cost (TLC) model results. If the TLC model
146
Logistics Costs
indicates an alternative mode mix than was of origin and destination of shipments. Reebie
recorded in the traffic flow database, then it and Associates’ Freight Locator® data identifies
is interpreted as an indicator of feasible modal a sample of specific producers’ total output by
shift and the basis for considering incremental commodity and specific location at the street
rail capacity. address level of detail.
However, this source does not indicate the
Traffic Flow Data destination of shipments. Thus, the two data
sources were combined: total traffic flows from
Transearch® Freight Market Data from Reebie Transearch and precise origin and destination
and Associates is used to characterize current of shipments from Freight Locator. It doesn’t
traffic flows in the United States. The flow file matter which shippers would utilize each mode;
indicates total tons by commodity class shipped only the geographic areas within BEAs that
between basic economic areas (BEA) in the generated and received freight are important
United States. Only the shipments that moved by for planning. A cumulative probability density
full truckload, private truckload, and intermodal function (CDF) is thus created for each BEA that
(truck and rail combined) are considered in this expressed the probability of shipment from each
analysis, eliminating all air, water, less-than- ZIP code in the BEA as a function of the total
truckload and full railcar movements. The outbound shipments of the ZIP code based on
U.S. Department of Transportation Bureau Freight Locator estimates. The same CDF is used
of Transportation Statistics Commodity Flow at the destination of each shipment, based on the
Survey indicates that these modes account for assumption that the ZIP codes that produce the
70% of all the tons moved in the United States economic activity receive freight in the same
each year. The analysis is limited to shipments relative quantities.
that travel between BEA’s in the contiguous 48 ZIP codes within the origin and destination
states, which accounts for more than 90% of BEA’s are randomly sampled for each
the intercity truckload traffic overall (based on commodity that flows between two BEA’s
summary analysis of Reebie Transearch data). based on the commodity’s volume-weighted
This data represent more than 932,000 CDF, using the centroid of the ZIP code as the
commodity-origin-destination combinations location of the shipper and receiver. In effect,
over which freight has flowed historically. high-volume origin-destination pairs are more
As a matter of practicality, only the origin- frequently sampled because the more flows
destination pairs that average more than two between two BEAs, the more shipper locations
trucks per week in 2000 are included in the within the BEA’s were sampled. This volume-
analysis. This limitation reduces the number of weighted random sampling more narrowly
freight flow combinations to just over 75,000 defines likely shipper locations and more
(8% of the total commodity origin-destination exactly specifies shipment and total logistics
pairs in Transearch), but excludes only 14% of costs. Because of the large number of origin-
the tons in the base data file. As such, the most destination-commodity combinations, extensive
important flows in the United States are included sampling is conducted. Resampling does not
while vastly reducing the number of records to appreciably change the results.
process.
Truck and Rail Data Requirements
Shipper Sampling
To maintain comparability to trucking, a door-to-
Basic economic areas can encompass wide door intermodal rail cost is calculated based on
geographic regions and as such are inadequate for four factors: rail rate, dray (origin and destination
determining total logistics costs for intermodal truck transportation), equipment costs, and
movements. Proximity to the rail ramp is a management fees.
critical determinant of the economic feasibility Rail data from the Burlington Northern
of the rail option. A second data source was Santa Fe Railway (BNSF) was collected on the
required to give a more precise understanding location of more than 40 rail ramps in major
147
Logistics Costs
metropolitan areas in the western United States for products in transit is used in this study,
for calculating dray distances from customers to because equipment costs are included in the
rail ramps. The cost of dray per mile by location transportation rate and no warehouse is being
and equipment type is derived from BNSF used while product is in transit. Data provided
experience. A minimum dray cost is specified in the U.S. Department of Transportation,
for drays of short distance, and a maximum dray Bureau of Transportation Statistics Commodity
distance is constrained based on the customer’s Flow Survey was used for the value of major
service expectation (500 miles, or one-day’s commodity groups per ton.
dray, maximum).
BNSF provided 2001 intermodal rail rates EMPIRICAL RESULTS
and historical 98th percentile service levels (hours
in rail transit) for all rail ramp pairs for which it The TLC model was run on the Transearch
provides service. In the cases when the rail rate database of 75,047 potential shipment lanes
does not include equipment in the rate, the total is over which 825 million tons of freight flow.
calculated based on the transportation-only rate Because of shipper or receiver distance from an
plus an imputed lease rate per day of trailer or intermodal ramp with BNSF service, 16,226 of
container equipment. these lanes had feasible rail service defined. The
Finally, a 7% management fee is assessed maximum dray distance constraint of 500 miles
for the management of each intermodal and the western U.S. geographic limitations of
shipment. Because intermodal is inherently BNSF railway eliminated many origins and
more complicated a product to buy and execute destinations from rail consideration. Further,
than trucking, intermodal marketing companies, rail origin-destination pairs are limited because
or IMC’s, have emerged. Similar to freight less than 30% of all intermodal ramp pairs in
forwarders and truck brokers, the IMC’s handle this study have defined service. For example,
the intricacies of an intermodal shipment for the although BNSF has an intermodal rail ramp
shippers to mirror the transactional simplicity in both Phoenix and Seattle, no intermodal
of purchasing truck. Thus, from a shipper service is defined for that origin-destination
perspective, a third party manages the hand- pair. Only 146 million tons have feasible access
offs between the modes, making the product to intermodal ramps, thus 82% of all potential
truck-like in all ways but service speed and freight traffic is excluded from rail consideration
variability. because no rail service exists.
Full truckload costs are based on the North For markets served by BNSF, the model
American Truckload Rate Index, a comprehensive indicates opportunities for modal shift in both
source of truck rates, and financial statements of directions. Table 1 shows aggregate results in
major trucking companies. mode shift opportunity.
Summary results from Table 1 indicate that
Inventory Carrying Cost Data 50 million tons could be shifted from truckload
to intermodal, representing a reduction of 44% of
Estimates of inventory carrying costs are existing truck freight shipments (or equivalently,
required for comparing transportation modes a 145% increase in intermodal) in lanes where vi-
with different service speeds and reliability. able intermodal service exists. It should be noted
The cost of holding inventories includes interest; that despite the large modal shift potential in the
insurance; warehousing; shrinkage and damage; high-volume, western U.S. lanes under consider-
and management, handling, and administrative ation, this conversion represents only 5% of the
costs. Although exact inventory carrying costs total U.S. truck and rail freight market.
vary by product and company, these costs are Figure 2 shows the modal shift potential for
typically estimated in the 20-30% of commodity the markets addressed, expressed as a percentage
value per year range in Delaney (2000). This of the total existing flows: 62% of the flows
means that for every $1 in inventory, a cost of stay in existing modes, but 38% should shift
20-30 cents is assessed per year for holding this modes, 36% of which is shifting from truck
inventory. A 17% per annum carrying costs to rail. Caution should be taken in interpreting
148
Logistics Costs
Table 1: Summary of Total Logistics Cost Model Results for Origin-Destinations
with Rail Options
���������
���� �������������� �������� ����� ����
����� �������� ������� ����� ����������� ���������� �����������
����� ����������� ���� ���������� ���������� �����������
�������� ������ ��� ����
these results. By the nature of the modeling used, a point of comparison for the order of magnitude
each market and commodity is either better off of this savings, the BNSF railway as a whole has
using truck or rail. In practice, because of a revenues of approximately $10 billion, with only
number of factors such as capacity availability, $2.5 billion of that coming from its intermodal
varying commodity values, shipper-negotiated product.
rates, distances from rail ramps, and shipper The detailed nature of this modeling effort
preferences, a mix of modes is the norm. Only allows BNSF to observe in which geographic
the potential for modal shift is demonstrated, regions to plan its expansion and base its
actual shift will vary depending on modal marketing effort. Tables 2 and 3 show the largest
capacities, pricing, and the level of marketing truck-to-rail shift opportunities by origin and
effort employed. destination state; Tables 4 and 5 do the same by
The average savings in total logistics city. Recall that only BNSF rates were used in
costs from shifting modes, including inventory this analysis, thus, the majority of opportunities
carrying costs, is 21%. The total potential savings are in western states. To explain opportunities
for shippers from modal shift is $1.5 billion. As identified to or from eastern states, BNSF has
149
Logistics Costs
some limited “through rates” with eastern carriers Table 7 shows the top modal conversion by
for which BNSF can sell intermodal product that commodity type. Note that fresh vegetables are
extends beyond its rail network. Additionally, the top conversion commodity, and that these
dray can be conducted from BNSF’s eastern most products tend to have a short shelf life (translated
points (e.g. Chicago, Kansas City, and Memphis) as a high inventory holding cost in this model).
to reach some eastern locations. Two additional days transit could significantly
Although California emerges as both a top reduce their salability. Other high-conversion
origin and destination for rail opportunity, it is commodities tend to be products of higher value.
important to note that top rail traffic opportuni- Model sensitivity to increased inventory holding
ties are not within California, but to and from costs is measured below.
California. Table 6 shows top origin-destination Figures 3 and 4 show the the distribution of
state pairs. preferred mode by transit distance. In general,
Table 4: Top Modal Conversion Opportunities Table 5: Top Modal Conversion Opportunities
by City and State of Origin by City and State of Destination
150
Logistics Costs
as would be expected, rail is superior over long correspond well with the data reported by the
distances. The handoffs between dray and rail U.S. Department of Transportation Commodity
give rail short hauls a higher cost structure and Flow Survey, confirming the model’s predictive
lower service levels. The average distance for capability.
which rail was the desirable option was 2,070
miles; the distribution of these distances is shown Implications for Railroad Executives and
in Figure 3. Public Policy
On the other hand, truck-preferred
shipments averaged 1,122 miles with the Overall, the model results indicate that a
distribution depicted in Figure 4. It is worthy of substantial amount of current long haul
note, however, that considerable overlap exists truck freight could be moved with lower
between truck and rail in the 750-2,000 mile total logistics costs by intermodal rail, while
range. These shipment distance distributions relatively little freight currently using intermodal
151
Logistics Costs
Figure 3: Distribution of Shipment Distance for Intermodal Rail Preferred
would be moved more cost effectively by investment in the freight rail infrastructure would
truck. For railroad executives who face tight help supply the rail capacity for the freight that
capacity constraints and challenges justifying could move via rail, and may be justified through
infrastructure investment to their shareholders, the savings in reduced highway investment to
these results suggest that there is potential for support that freight movement.
intermodal price increases which may help to
justify more investment in expanded rail capacity Sensitivity Analysis
to accommodate these shippers.
For public policy makers, extensive The predictive model was evaluated for sensitivity
investment in highway capacity based on to key assumptions and model parameters.
historical truckload shipments and growth rates The primary interest is in understanding the
may result in a misallocation of funds. Much likelihood of the predicted modal shift from truck
truck freight could be moved via intermodal with to rail. First, the inventory carrying cost estimate
lower total logistics costs for shippers and lower was doubled from 17% to 34% to evaluate the
societal costs (e.g. congestion and pollution) if impact on modal shift from truck to rail. (Transit
the rail capacity existed. Well-placed public time and inventory holding costs enter into the
152
Logistics Costs
total logistics cost function in the same fashion; distribution of major population centers in the
similar results apply to increasing the rail transit eastern United States, the expected results may
time.) From doubling the inventory carrying be markedly different in new geographies and
costs, 427,000 gross tons that had shifted from markets. Due to the generally shorter lengths
truck to intermodal were better off staying truck of haul in the East, the potential truck to rail
(approximately 1% of the predicted shifted conversion would be much smaller. Another
freight). As would be expected, conversion evaluation of network expansion that could be
reversal to truck occurred in the higher valued evaluated is to increase the density of intermodal
commodities. The robustness of the finding with ramps and connectivity between those ramps to
respect to this parameter is somewhat expected evaluate the modal shift in markets where no
because the transit time difference between truck intermodal service currently exists.
and rail averaged 1.3 days in the routes where This methodology could be extended to
intermodal was preferred and was at most two evaluate different lot sizes and service levels.
days more than truck. (Recall, any rail service Although the focus here is on full truckload lot
that took more than two days longer than truck sizes only, this model has been applied to broader
service was excluded from modal conversion.) ranges of rail freight transportation services such
Accordingly, the transportation cost constituted as rail boxcar, which has three times the capacity
the majority of the total logistics costs (roughly of truck. By including inventory queuing costs at
80%), and therefore had a larger impact on modal origins, the shippers’ propensity to shift to larger
choice than inventory costs. lot sizes can be evaluated. More information
The model was further tested for sensitivities would be required of specific shippers’ shipment
to the estimated transportation price to allow activity in each lane to evaluate shipment
for shipper-specific volume discounts for full aggregation potential, and is thus impossible in
truckload or under-estimate of IMC brokerage this macro setting.
fees or dray costs. To account for these potential Finally, this modeling and data analysis
errors, any modal shift that was based on a total approach may be applied to other areas of
logistics cost savings of less than 20% from study, such as an optimization-based estimate
full truckload costs was subtracted from the for shipper price and service responsiveness.
conversion potential. (For example, a shipper
may not be inclined to change modes for less CONCLUSIONS
than a 20% savings in total logistics costs, thus
smaller savings amounts may not produce modal Optimization-based methods are used to
conversion.) A total of 16 million of the predicted forecast potential modal conversion in specific
50 million tons (just under one-third) of freight geographies. By modeling the shipping decisions
do not shift from full truckload to intermodal at a microeconomic level and applying them to
in this scenario. Thus, the transportation cost a large-scale shipping database, the potential
estimate is a more important determinant of demand for intermodal rail service can be
model results than inventory holding costs. Still, estimated. Deviations from model predictions
the general findings hold for the majority of the may have many causes, such as model error,
freight that is predicted to shift modes even with bounded information, rail capacity, previous
higher transportation costs. shipper satisficing, previous experience and
preferences, union contract issues, or a host of
EXTENSIONS others. Model results are tested for sensitivity to
input data errors and find the model predictions
The analysis could be expanded in both to be robust. The resulting modal conversion
geographical extensiveness and intensiveness. marketing opportunities that are based on this
The results here are based solely on the western methodology are identified and implications
United States and existing intermodal service. discussed.
Because of the more compact geographic
153
Logistics Costs
Endnotes
1. A 95th percentile assumption would advantage rail; a 99th percentile would advantage truck.
The 98th percentile assumption is a middle ground and consistent with Sheffi (1988).
References
Barnhart, C. and H. Ratliff. “Modeling Intermodal Routing.” Journal of Business Logistics 14(1),
(1993): 205-223.
Bronzini, M. and D. Sherman. “The Rail Carrier Route Choice Model.” Transportation Research
17(A), (1983): 463-469.
Crainic, T., M. Florian, and J. Leal. “A Model for the Strategic Planning of National Freight
Transportation by Rail.” Transportation Science 24(1), (1990): 1-24.
Fernandez, L., J. Enrique, and A. Soto. “A Multi-modal Supply-Demand Equilibrium Model for
Predicting Intercity Freight Flows.” Transportation Research (B) 37(7), (2003): 615-640.
Freidlander, A. and R. Spady. “A Derived Demand Function for Freight Transportation.” Review
of Economics and Statistics (1980): 432-441.
Kresge, D. and P. Roberts. “System Analysis and Simulation Models.” J. Meyer ed. Techniques of
Transport Planning. Brookings Institute Washington, D.C. (1971).
Class 8 Solutions, Inc., North American Truckload Rate Index, Woodbury, MN.
Oum, T. “A Cross Sectional Study of Freight Transport Demand and Rail-Truck Competition in
Canada.” Bell Journal of Economics 10(2), (1979): 463-482.
Oum, T., W. Waters, and C. Yu. “A Survey of Productivity and Efficiency Measurement in Rail
Transport.” Journal of Transport Economics and Policy 33(1), (1999): 9-42.
154
Logistics Costs
Oum, T., W. Waters, and J. Yong. “Concepts of Price Elasticities of Transport Demand and
Recent Empirical Estimates: An Interpretative Survey.” Journal of Transport Economics and
Policy 26, (1992): 139-154.
Roberts, P. “Factors Influencing the Demand for Goods Movement.” Paper presented at the
Annual Meeting of the Transportation Research Board
Board, Washington, D.C., 1976.
Sheffi, Y., E. Babak, and H. Koutsopoulos. “Transportation Mode Choice Based on Total
Logistics Cost.” Journal of Business Logistics 9(2), (1988): 137-154.
Acknowledgements
We thank the BNSF Railroad for providing the Reebie and Associates traffic flow and shipper data,
as well as their own price and service data, and we thank Scott Webb of BNSF for his insights in
our modeling efforts.
Michael F. Gorman earned his Ph.D. from Indiana University in 1994. He is an assistant professor
in decision sciences, University of Dayton and president of MFG Consulting, Inc. He has a 10-year
career working for the Burlington Northern Santa Fe Railway where he held positions in supply chain
solutions, eBusiness, operations research and finance. His field of specialization is in transportation
asset pricing, investment and allocation decisions under uncertainty. He has served as treasurer,
secretary, vice chair and chair for the Rail Applications Special Interest Group (RASIG) of the Institute
for Operations Research and Management Science (INFORMS).
Daniel G. Conway earned his Ph.D. from Indiana University in 1992. He is an associate professor
in the management department at the University of Notre Dame. Previously, he was an assistant
professor at Virginia Tech and at the University of Florida. His field of specialization is supply chain
management, project scheduling, and risk management.
155