Essays on Freight Traffic Management in Indian Railways

Indian Railways (IR) is the world's fourth-largest railway network. In 2023–24 alone, IR transported more than 7 billion passengers and over 1,400 million tonnes of freight. Freight operations contribute to the majority (~70%) of the IR's revenue. Despite being the most cost-effective and energy-efficient mode of transporting freight, the share of IR in the freight transportation market has been steadily declining. One of the primary reasons for this decline is the low speed of freight trains, which has remained constant at around 25 kmph for the last five decades. Unlike passenger trains, freight trains typically operate without fixed schedules, with their movement decisions made in real time by traffic controllers. Thus, to better understand the causes of low speed, it is important to study the decision-making process that governs freight train movements.
In the first essay, we empirically study how section controllers make decisions for freight trains. Using a detailed dataset from the Deen Dayal Upadhyay division of IR, we construct a minute-by-minute snapshot capturing network congestion along with freight and passenger train movements. Utilizing this, we estimate two discrete choice models corresponding to the section controllers’ hold and stop decisions. We explicitly account for spatial dependence across these decisions by using a copula-based approach and address potential endogeneity concerns using network instruments. Our results show that passenger trains are given high priority, which significantly affects the hold and stop decisions for freight trains. Even when the subsequent track is free, freight trains are held in the anticipation of an upcoming passenger train (strategic idling effect). Additionally, the presence of trailing freight trains increases the probability of the lead freight train being released (push effect). Further, we find that section controllers are more likely to hold freight trains at high-capacity stations, such as junctions.
In the second essay, we use the estimates from the first essay to conduct a detailed simulation-based counterfactual analysis aimed at improving freight train speeds. Specifically, we study two sets of counterfactuals: capacity-based and non-capacity-based. In the capacity-based counterfactual, we simulate Freight Only Corridor (FOC)—dedicated corridor reserved exclusively for freight trains. We find that FOC leads to a 37% reduction in average dwell time, and a 31% reduction in stoppages, resulting in a 16% increase in average speed. Given the high cost of capacity interventions, we also examine two operational, non-capacity-based counterfactuals: threshold-based releases and capacity consolidation. As cost-effective alternatives, we find that a 60-minute threshold-based release policy increases average speed by 13%, while a moderate capacity consolidation of 30% results in a 13% improvement in average speed. Together, these findings highlight practical and effective interventions for improving freight train speeds.
In the third essay, we model a railway station as a finite-buffer priority queue with two customer classes: passenger trains (priority) and freight trains (non-priority). The objective is to characterize optimal service‐control policies for freight trains with the aim of minimizing the waiting costs. We analyze two cases: visibility and no-visibility, where visibility refers to the information of the time and type of the next train arrival. For the no-visibility case, we formulate the problem under both long-run average cost and infinite-horizon discounted cost objectives, and derive structural results. Under the long-run average cost objective, we find that the optimal policy is to always release (serve) the freight train. For the infinite-horizon discounted cost objective, we identify conditions under which a threshold policy, based on the number of freight trains waiting at the station, is optimal. In the visibility case, our simulation study shows that the optimal policy follows a two-dimensional monotone switching curve, defined by the number of freight trains waiting at the station and the expected arrival time of the next passenger train. This curve characterizes the threshold at which the policy shifts from holding to releasing the freight train. We hope these policies can serve as practical guidelines for managing freight train movements while minimizing delays and maintaining passenger train priorities.