| Randall R. Reyff
Manager - Market Research Atchison, Topeka & Santa Fe Railway Company |
Bengt Mutén Consultant DNS Associates, Inc. |
The Atchison, Topeka and Santa Fe Railway Company (ATSF) and the Southern Pacific Transportation Company (SP) used, in their merger application to the Interstate Commerce Commission (ICC), a computer based model to estimate likely merger- related rail traffic diversions. This was the first time that the required rail traffic diversion study submitted as evidence in an ICC proceeding was done by computer. The use of the DNS Rail Traffic Diversion Model provided significant information benefits not only to the applicant railroads, but also to the protesting parties and the ICC. This paper describes the DNS Model, how it was used in the Southern Pacific and Santa Fe merger, and the benefits that resulted from its use.
Merger-related rail traffic diversions--i.e. the rerouting of individual shipments by rail customers away from competing carriers and to the merged carrier's longer haul or single line routes--are an important aspect of a proposed merger's evaluation from either a public policy or carrier management perspective. Viewing the proposed merger from a public interest perspective, the ICC must determine the degree to which the proposal is likely to affect freight traffic competition and the the essential services of railroads currently interlining traffic with merger applicants. From the merger applicants' management perspective, anticipated rail traffic diversions will have, in most cases, a material effect on expected revenue and operating expense. In addition, the actual handling of diverted traffic may suggest changes in yard and interchange operations, local and through train operations, as well as plant and equipment utilization. Consequently, both the public interest and carrier management interests demand a disciplined study of expected rail traffic diversions incorporating a level of detail appropriate to address the public policy and management issues associated with the proposed merger.
In addition to intellectual and methodological discipline, a rail traffic diversion study should allow for flexibility with respect to testing alternative key assumptions. In the final analysis, the results of a diversion study are a forecast of the gains/losses expected to accrue to individual railroads resulting from the re-routed traffic when the merging carriers reach a mature level of post-merger operating and marketing competence. As with any forecast, the process is essentially one of observing facts, estimating probabilities based on informed judgment, reaching a conclusion, and describing the projection. Because applied judgment is at the heart of the process, and because the judgment of individual analysts may honestly vary depending upon their experience, perspective, and beliefs about what drives shipper routing decisions, diversion study methodology is strengthened to the degree that alternative beliefs can be easily tested. The ability to test assumptions and quantify resultant changes does two things: (1) demonstrates the degree to which a perceived "key" assumption does in fact result in a material difference and (2) presents a range of probable outcomes which reflect any finally irreconcilable differences in individual analysts' points of view. The overall benefit to the merger evaluation is to strengthen the role of reason, at the expense of rhetoric, in focusing the analysis.
It is important that study results be in a form which can be easily analyzed from the various perspectives of railroad management (i.e. operations, marketing, engineering, mechanical, personnel, finance, and law) as well as other interested parties. Expected traffic diversions, on both a system and division basis, can have a material impact on planning in all functional areas of railroad management. No single summary, or limited series of fixed format reports, is likely to effectively respond to the traffic diversion-related questions which evolve as both railroad management and other parties analyze a proposed merger.
The DNS Rail Traffic Diversion Model provides an analytical framework with a high degree of methodological discipline, flexibility, and accessibility to study results. The model enables an evaluation of a proposed rail merger at a policy level while incorporating adequate detail, in terms of rail network utilized as well as waybill facts recognized, to address tactical level post-merger management planning. Use of the model offers significant benefits compared to manual diversion studies which typically formed the basis for past management as well as ICC analyses of rail traffic diversions. Principal among these benefits are:
(1) Analysts are led into, in a relatively more organized fashion, iterative comparison and evaluation of perceived competitive attributes in various market settings. Underlying traffic diversion logic and criteria are examined and re-examined in light of how they affect many different comodities as well as geographic traffic segments. The model exposes, and evaluators are forced to confront and resolve, apparent contradictions during, not after, the analytical process. The result is a more complete "thinking through" of the proposed merged carrier's expected comparative advantages as they relate to the composition of the marketplace. In terms of methodological discipline, decision logic is consistently applied and, where exceptions are incorporated, they are explicitly stated and their reasoning well thought out.
(2) Constructive criticism of any aspect of the analysis (model network, diversion logic, exceptions rules, etc.) can be incorporated and the model rerun. Successive iterations develop (a) a more refined analytical model, and (b) an understanding of when continuing refinement will no longer affect a significant amount of rail traffic.
(3) Use of the model allows analysis of significantly more waybill observations than was typically the case in manual studies. Time and manpower availability restricted manual study to a relatively small sample (often 1% or fewer) of merging carrier's waybills. While this could produce an acceptable system level estimate of traffic diversion, analysis of the diversion estimate's component parts, below certain high levels of disaggregation, was subject to increasing probability of error due to sampling variability. As a consequence, diversion study results were of marginal value to post-merger management planning. General disaggregation was typically limited to describing only major groups of expected revenue changes (e.g. TOFC/COFC, principal comodities). Geographic disaggregation of study results could not, with acceptable confidence, provide meaningful information to aid tactical planning.
(4) Waybill input data and study results are available in data base form and can be easily accessed, sorted, and analyzed depending on the evolving interests of management or other parties (e.g. the ICC).
Overall, intelligent use of the computerized rail traffic diversion model allows, in fact forces, a high level of analytical discipline in railroad merger analysis and allows a diversion study to make a continuing, rather than "one-time," contribution to merger planning. The model is complete at any point in time although in a sense never completed. It can be rerun as desired, with new data or revised assumptions. Manual diversion studies tended to be episodic and terminal as well as difficult to redo with reasonable levels of effort and logical consistency from one to the next. In contrast, the computerized model is organic and iterative. It always contains in readily available form the current best understanding of merger-related traffic diversion effect and reasoning based on the cumulative understanding and expertise of railroad marketing management.
The DNS Associates, Inc. diversion model had been previously developed and used for a number of internal studies by Class I railroads, both for mergers and acquisitions, and to study the sale and acquisition of branch lines. It allows a railroad's traffic experts to determine a set of rules for diversions of movements, which the model then consistently applies to a sample of waybills. It is a flexible model that easily can be modified to suit the needs of the specific analysis. The process can briefly be described in a number of stages.
A. Waybill Data Verification and Editing
The model will use railroad waybill data from any source, including the carriers involved or the ICC waybill sample. The model can use either a random sample or a complete set of waybills for any time period. The waybills are checked for keypunch errors and incomplete information, and those errors that can be are corrected.
B. Network Updates
The DNS Rail Network is used in routing and miling traffic in the diversion model. This network is based on the Federal Railroad Administration's rail network, but it has been updated to include recent mergers, abandonments, and line sales. A basic Canadian network has also been added. The network is continually updated, but for each diversion study it is verified for the data used. This verification process involves checking for any traffic in the waybill data sample that cannot be routed using the network. Errors that may occur are:
An origin or destination station is not mapped to the network.
A junction on a waybill is not mapped to the network.
A railroad on a waybill is not in the network.
A railroad on a waybill is not shown in the network to serve the junction or station on the waybill.
Inconsistencies between the waybills and the network are checked against published references. If the network is incomplete, it is updated. If the waybill shows an error, it is corrected if possible. The conflicts with the greatest affected carload count are resolved first, and the process is continued until a point of diminishing returns is reached. For the Southern Pacific and Santa Fe diversion study, 99.4% of the carloads were routed on the DNS network.
C. Selecting Traffic Subject to Diversion
Traffic that will not be affected by the merger is eliminated from consideration. For example, traffic carried exclusively by the merging carriers prior to the merger will not be diverted (although cost savings in handling this traffic can be considered at a later stage). In the case of the Southern Pacific and Santa Fe merger, only traffic carried all or in part on railroads west of the Mississippi was included in the study. The sole exception was coal in unit trains, which was studied separately by railroad personnel and found to be unaffected by the proposed merger.
D. The Candidate Diverted Route is Developed
Using the DNS Rail Network, the model develops a candidate diverted route. This is a route favoring the merging carriers. A minimum path algorithm is used to find a route limited to certain criteria. The originating or terminating carrier must remain unchanged, unless the origin or destination is open for switching by the merged system.
The route selected is one with the minimum weighted mileage. The length of each link is multiplied by a factor depending on the line type. Each line in the DNS Rail Network is classified as a main line or branch line, and further sub-classified as A or B. Class A main lines mileage is multiplied by one, while class B main lines and branch lines have higher factors. Thus, the candidate route uses main line links as much as possible.
Each interchange between railroads is also assigned a weighted mileage. This mileage is a function of the total interchanges at the junction, and of the cars interchanged between a specific carrier at the interchange point. This provides candidate routes that use historically prevalent interchanges. The mileage penalties can be modified by the evaluators in cases where the traffic flows are not expected to follow historic patterns, or a specific interchange is to be encouraged.
To find candidate routes that favor the merging carriers, the weighted miles of the links of the merging carriers can be multiplied by a discount factor (.7 being used in the Southern Pacific and Santa Fe study). This provides a feasible route for potential diversion to the new system where diversion may occur, but does not allow for unreasonably circuitous routes.
E. Exclude Movements that Should not be Diverted
Rules are developed by the evaluators to exclude waybills whose candidate diverted route precludes diversion. Cases where there is no competitive advantage to the diverted route over the pre-diversion route, or where the route does not become more competitive due to the merger, are excluded. For example, moves with an overly circuitous diverted route may be excluded. Other exclusions may involve moves where the diverted route only uses one of the merging carriers. All moves not excluded in this process are considered possible diversions.
F. Predict Share of Traffic that will be Diverted
The analyst develops a matrix of the factors likely to influence shipper routing decisions on each axis, and a predicted estimate of diversion in each cell. Any information on the waybill can be used, but at a minimum the merging railroads' traffic class in the historic and candidate routes, and the accessibility of the origin and destination station by the merging carriers should be used. Other factors that may be used are traffic type, equipment type and ownership, shipper and consignee, circuity, geographic areas, or any other information provided in the waybill data.
G. Calculate Revenue Divisions
Revenue divisions are calculated using the short-line or actual rail distances between the route endpoints and junctions. The use of short-line miles, regardless of railroads, provides for identical division estimates for different carriers between the same points. In most cases DNS uses a formula allocating revenue in proportion to mileage rounded to nearest 50 or 100 miles with preference given to the originating and terminating carriers, although any consistent division algorithm can be utilized.
H. Report Movement Diversion Detail
Detailed examples of a subset of the movements are printed for the analyst to review the effects of the diversion rules and percentages. These detailed records show car and revenue counts, type of traffic, commodity, equipment group, origin and destination station, historic and candidate diverted routes, diversion percentage or reason for rejecting diversion, diversion matrix cell, and revenue divisions and changes. After review, the rules may be changed, and the process reiterated from either step three, four or five. This should be done prior to reporting any revenue totals, so modification of the rules is not biased by preconceived expectations of the end results.
I. Estimate Variable Costs of Diverted Moves
Pre- and post-diversion variable costs are calculated using Rail Form A, URCS, or carrier-developed costing methods. Cost saving from the rerouting of moves internal to the merged system may also be estimated at this stage.
J. Report Merger Effects
Revenue and cost impacts, as well as changes in operating statistics, of both the merging carriers and other affected carriers, are reported. Data and programs are available for any other reports which an analysis requires. In past studies reports have included diversions by station, line, junction, state, car type and commodity, as well as reports focusing on areas of special interest. Maps showing traffic densities and changes can also be generated by the model.
K. Repeat for Sequential or Alternative Studies
To study alternative scenarios, or the impact of sequential events, the process is repeated from step two. In the case of the Southern Pacific Santa Fe merger, five sequential diversion studies were performed to show the effects of other mergers and line acquisitions that had occurred, or would occur, between the time of the data sample and the time when the benefits of the merger would be realized.
Utilizing the ability of the DNS Traffic Diversion Model to process a large, and thus statistically more accurate sample, one in ten Santa Fe and Southern Pacific 1982 waybills of five or fewer carloads were included. For waybills with six or more carloads, this was increased to one in five. In order to measure gains from other carriers of loads not carried by Santa Fe or Southern Pacific, the ICC waybill sample was combined with the above records. Records with common characteristics were combined, and double counts were eliminated, resulting in 441,626 movement records. A manual study typically involved only 1,000 - 2,000 records of this sort.
The DNS Rail Network was edited and supplemented in the areas of significance to this study. After this process, 99.4% of all carloads in the sample could be routed though the network, and thus considered in the diversion study. In prior studies, a station was regarded as open if two carriers served the same 6-digit Standard Point Location Code (SPLC) and it was not specifically declared as closed. With a number of exceptions, stations north and east of the Mason-Dixon Line and the Pennsylvania - Ohio border were assumed to be closed, while other stations were assumed to be open if the SPLC was served by two or more railroads. Southern Pacific and Santa Fe provided the following information to make further refinements:
(1) For larger cities where switching districts exist allowing carriers access to shippers at stations they do not physically serve through terminal railroads and reciprocal switching, a table of SPLCs and the railroads with access to those SPLCs was provided. Thus, the model allowed diversion to the merging carrier at points accessed through switching districts, and those points were considered open in estimating the diversion impact.
(2) For trailer-on-flat-car (TOFC) and container-on-flat-car (COFC) intermodal traffic, carriers may compete even if their loading and unloading points are not in the same place. A table was developed for all pairs of intermodal ramps within 120 miles of each other, with the intervening highway distance. Also in the table was the type of loading equipment at each ramp. The model was modified to allow diversion if one of the merging carriers had a ramp within 120 miles of the originating or terminating ramp. In generating the candidate diverted route, highway miles between the ramps were weighted as 2.5 times the air miles. Furthermore, to discourage the choice of inefficient small ramps, a penalty of 250 miles was added for the use of ramps not employing mechanized loading equipment.
(3) Some of the Santa Fe and Southern Pacific waybills included shipper and consignee information. A table was provided listing shipper name and number, and showing which carriers physically serve the industry. This information was used to further estimate the diversion rate.
In previous studies using the DNS Rail Traffic Diversion Model, the revenue of a waybill carried by more than one carrier had been allocated between the carriers using a prorate by mileage blocks. A mileage block is counted for each fifty or one hundred miles, or part thereof, of carrier haul, plus one block each for the originating and terminating carriers. Santa Fe Southern Pacific used the revenue associated with settled ATSF and SP waybills to find a block distribution formula that closely matched actual revenue divisions. Using short-line miles and fifty mile blocks, the formula used not only matched the ATSF and SP revenues, but has produced total revenues for other carriers in other studies that were within .2% to 2.5% of actual revenues.
As with any historical data base, there is an inevitable lag between the occurrence of the event and the application of the descriptive data in decision making. This problem was encountered in the Southern Pacific and Santa Fe merger. Specific adjustments were made to reflect the important changes affecting western railroad traffic flows which had occurred since 1982:
(1) Eastern railroad route closings
(2) Union Pacific's acquisition of Missouri Pacific and Western Pacific
(3) Southern Pacific's acquisition of trackage rights between Kansas City and St. Louis
(4) Denver and Rio Grande Western's acquisition of trackage rights between Pueblo and Kansas City, together with the impacts of the Southern Pacific-Rio Grande solicitation agreement
These changes had to be considered before the study could estimate properly the incremental impact of the Southern Pacific and Santa Fe merger upon rail traffic flows and revenues. While these adjustments were part of a single base case adjustment study, in practice, they were run through the model sequentially, with samples of adjusted traffic being carefully reviewed to ensure that reasonable evaluations were being made. Southern Pacific and Santa Fe decided on a diversion percentage matrix with nine dimensions:
(1) Pre-diversion route, categorized by merging and non-merging carriers into 17 categories.
(2) Post-diversion route, categorized by merging and non-merging carriers into eight categories.
(3) Traffic type, divided by expedited comodities, intermodal freight, autos, and other.
(4) Origin classified as exclusive to, open to, or closed to the merging carriers. Specific facilities at open origins were further classified as served directly or by reciprocal switching, where such information was available.
(5) Destination classified in the same manner as the origin.
(6) Car type classified as special or standard equipment.
(7) Car owner classified as merging railroad, competing railroad, or neutral railroad.
(8) Significant improvement in rail service. Significant improvement was defined by how much shorter the diverted route was than the historic route. The improvement had to be at least 200 miles for expedited, auto and intermodal traffic, and 600 miles for all other traffic.
(9) Traffic to or from Oregon using the Overland route though Utah was treated separately.
While not all cells in this nine-dimensional matrix were used, the resultant table had 45,000 entries. A base value was calculated for each combination of pre-diversion and post-diversion route, and multipliers were applied for each of the other criteria to develop diversion rates ranging between 1% and 100%. For finished automobile traffic, the diversion rates were modified to be either 0% or 100% to reflect routing practices of automotive manufacturers.
To illustrate with an example, if the move historically originated by a merging carrier, was carried overhead by a non-merging carrier, and then terminated by another non-merging carrier, and the overhead carrier is eliminated in the diverted route, then the base diversion estimate is 49.5%. If the move is a non-expedited commodity, originated at a station not served exclusively by the merging carriers, and terminated at a station not served by the merging carriers, a multiplier of .78 is applied. If special equipment is not used, the car owning carrier is irrelevant, i.e. a multiplier of 1.0. If the new route saves over 600 miles, the rate is further multiplied by 1.2, producing a composite estimate of diversion of 46.3%.
The rules deciding if a diversion to the candidate route would be likely were fine tuned in a number of iterations. The evaluators devised special rules to exclude diversions where the merged system will not have a competitive advantage over the two carriers prior to the merger. These rules took into account which carriers served the origin and destination, the origin and destination states, the carriers used in the historic and candidate routes, and the interchanges used in both routes.
To find the net gain from the traffic diversions, the variable costs of the diverted moves were estimated using Rail Form A techniques. Modifications to the costing algorithms were made to take into account the over the road or "rubber" interchanges of trailers at Chicago and St. Louis, and for cost savings gained through less inter-train and intra-train switching.
After the four base case diversion runs were completed, reports were produced showing the estimated revenue impact of those mergers and trackage rights acquisitions. These results compared favorably with ICC findings in the Union Pacific-Missouri Pacific-Western Pacific merger case. This comparison strongly confirmed in the aggregate what the Southern Pacific and Santa Fe traffic evaluators had been observing in detail; the model was implementing diversion decisions and calculating their associated revenue change in a reasonable and appropriate manner.
The primary product of the model was the estimated gross and net revenue gain to Southern Pacific and Santa Fe:
| Gross Revenue Gain | Net Revenue Gaini |
| $237,500,000 | $65,900,000 |
The model also estimated the traffic and revenue loss for each affected carrier.
The DNS Rail Traffic Diversion Model produced rerouted traffic records which were used in developing the operating plan for the merged system. This allowed Southern Pacific and Santa Fe to make accurate estimates of facility savings possible through the merging of operations.
The DNS study also measured the net public benefits resulting from the projected rail traffic diversions to the Southern Pacific and Santa Fe system. As a result of projected traffic diversions, rail shippers on a national basis will save about 6,445,000 loaded car miles annually. Also as a result of rail traffic diversions, over 413,400 carload interchanges will be avoided each year for about 19,350,000 tons of present interline rail traffic. These increased routing efficiencies will provide benefits to rail shippers using the nation's rail transportation network.
The primary purpose of the 1982 traffic diversion study was to support, as part of the Applicants' direct case, the merger application of the Southern Pacific Transportation Company and the Atchison, Topeka and Santa Fe Railway Company before the ICC. Verified statements describing the process and showing the results of the study were presented. These statements were then challenged by reply studies and statements by protesting carriers.
In past merger proceedings the individual waybills of the typically small manual traffic diversion sample were often vigorously debated as to whether the facts supported diversion to the merging carriers. Certainly the intensity of the debate which often focused upon minutiae or matters of opinion as much as fact, resulted in large part because of the small sample size itself. Each waybill where a different decision could be successfully argued had the potential of adding or subtracting relatively large amounts of revenue on an expanded basis. In the Southern Pacific and Santa Fe case the debate or discussion of traffic diversion focused less on individual records than on supporting rules, underlying diversion logic, and the sequence of study iterations. While individual movement discussion was not entirely eliminated, it was much reduced.
One protesting carrier, the Union Pacific, while not endorsing the model as necessarily the best or only way to conduct a diversion study, used the package already available at DNS Associates, Inc., to prepare a countervailing view of diversion impact. They ran the model using the same data and most of the rules, but changed some basic assumptions and the sequence of iterations. The results were then presented to the ICC. Southern Pacific and Santa Fe in rebuttal evaluated Union Pacific's criticisms, accepted some and rejected others, and reran the model and submitted the results to the case record. Other carriers requested computer tapes showing all the movements and diversions in order to conduct their own studies, using their own programs. The use of a model allowed DNS to provide them with tapes as well as paper output, thus improving their study approaches.
One of the concerns of the public in a rail merger is the potential for reduced competition. The ICC has to weigh the benefits of the merger to the participants, and to society as a whole, against the possibility of economically meaningful loss of competition resulting in decreased service or increased cost to the shipping public. The data base produced by the DNS Rail Traffic Diversion Model was eminently suitable for identifying potentially affected traffic. The
updated "open/closed" station tables were used to identify specific traffic flows where meaningful competition could possibly be affected by the Southern Pacific and Santa Fe merger.
This analysis was performed both to the specifications of Southern Pacific and Santa Fe and independently for the U.S. Department of Transportation. It was the underlying basis for determining geographical market segments in which Southern Pacific and Santa Fe agreed to grant Burlington Northern ratemaking authority in order to eliminate perceived competitive problems of the merger. The Agreement's purpose was to guarantee Burlington Northern continued or new access to certain rail traffic in the event that Southern Pacific and Santa Fe were to exercise undue market power to increase rail rates.
The use of a computer processed traffic diversion model proved to have all the expected advantages over a manual diversion study. The model's results were accepted as evidence in the case despite some strenuous but unsupported arguments that the model was just an unverifiable "black box." The DNS Rail Traffic Diversion Model played a major role in the evaluation of the Southern Pacific and Santa Fe merger. It provided for a disciplined analysis with a high level of consistency compared to manual diversion studies. It allowed speedy revisions as new information became available or in response to certain protesting carriers criticisms. The larger sample size allowed disaggregate analysis with a high level of confidence, including the use of the model for competitive analysis. After the merger is completed, re-study of then current Southern Pacific and Santa Fe traffic could provide a data base appropriate and valuable for merged system market planning as well as operations analysis and budgeting. We expect that in future rail merger litigation, the parties will take advantage of this or similar rail traffic diversion models.
Cerveny, Harold and Keith Molan. Interstate Commerce Commission Finance Docket 30400, Vol. UP-MP-28. January 1985.
Fine, Sidney and Rebecca Owen. "Documentation of the ICC Waybill Sample." Washington, D.C.: The Office of Policy and Analysis, Interstate Commerce Commission. November 1981.
Guerin, Frank M., Robert Keyes, Randall R. Reyff, and Newton D. Swain. "Market Impact Analysis." Interstate Commerce Commission Finance Docket 30400, Vol. SFSP-14. March 1984.
Harris, Robert G. "A Statistical Analysis of the FRA Waybill Sample: Interim Report." Washington, D.C.: U.S. Department of Transportation, Federal Railroad Administration, March 1976.
Kornhauser, Alain L. "Elementary Theory of Traffic Diversions: A Tool for Analysis of Restructured Railroad Networks." Paper presented at the 58th Annual Meeting of the Transportation Research Board. January 1979.
Reyff, Randall R. "Rebuttal Verified Statement." Interstate
Commerce Commission Finance Docket 30400, Vol. SFSP-49. July 1985.
Return to Mutén &
Associates, Inc. Home Page