Events & Seminars

Supply diversification and price postponement are two common mechanisms for dealing with supply yield uncertainty. In this talk, we investigate the interaction between the two aforementioned strategies and provide insights on how to effectively integrate them in combating supply yield risk. Specifically, we study a firm's pricing and sourcing decisions under supply yield uncertainty, and compare them under two distinct pricing schemes to investigate the impact of price postponement: (1) ex ante pricing - the firm simultaneously makes the sales price and sourcing decisions before production takes place; (2) responsive pricing - the pricing decision is postponed until after the yield realization. We find that the effect of price postponement on the optimal sourcing decision varies. With one unreliable supplier, responsive pricing mitigates the overage and the underage risks imposed by yield uncertainty, and results in a lower [higher] optimal order quantity than that under ex ante pricing when the procurement cost is low [high]. With two unreliable suppliers, when the sole- sourced supplier's reliability is low [high], responsive pricing promotes [discourages] supply diversification; when the sole-sourced supplier's reliability is moderate, responsive pricing promotes [discourages] supply diversification when its unit procurement cost is low [high]. The composition of supply portfolio also has a fundamental impact on such strategic interaction: When the supply portfolio consists of one unreliable and one reliable supplier, diversified sourcing is never optimal under ex ante pricing, but may be optimal under responsive pricing. Finally, we conclude by comparing our results with those obtained under random capacity model and discussing several related extensions to provide additional insights in mitigating supply yield risk.

Gig economy firms benefit from labor flexibility by hiring independent workers in response to real-time demand. However, workers' flexibility in their work schedule poses a great challenge in terms of planning and committing to a service capacity. Understanding what motivates gig economy workers is thus of great importance. In collaboration with a ride-hailing platform, we study how on-demand workers make labor decisions. We are interested in both improving how to predict the behavior of gig workers and understanding how to design better incentives and policies. Using a large comprehensive dataset, we first develop an econometric model to analyze workers' labor decisions in response to incentives while accounting for their personal goals, sample selection, and endogeneity. Our careful analysis has revealed behavioral insights that can inform better incentive and regulatory design. To further capture platform competition, we leverage our proprietary spatial data and the publicly available trip records to develop and estimate a structural model of gig workers' sequential dynamic decisions in the presence of alternative work opportunities. Our simulation-assisted estimation provides insights into workers' switching behavior and potential policies to better manage the flexible workforce.

The talk draws from two papers in dynamic contract design. The first paper considers a basic model, in which a principal incentivizes an agent to exert effort to increase the instantaneous arrival rate of a Poisson process. The effort is costly to the agent and unobservable to the principal. Each arrival yields a constant revenue to the principal. The principal, therefore, devises a mechanism involving payments and a potential stopping time to maximize her profits.

The second paper builds on the first paper's framework and considers a more complex setting where a principal hires an agent to run a local service store. Customers request service in one of two ways: either via an online or a traditional walk-in channel. The principal does not observe the walk-ins, nor does she observe whether the agent exerts (costly) effort to increase customers' arrival rate. This creates an opportunity for the agent (i) to divert cash (that is, to underreport the number of walk-in customers and pocket respective revenues) and also (ii) to shirk (that is, not to exert effort). This leads to a novel so far unexplored double moral hazard problem. We also present dynamic contracts that maximize the principal’s profit.

In both papers, we derive the optimal contracts which have simple and intuitive structures. Further, in the second paper, we extend the model to allow the principal to either (i) monitor the agent or (ii) manipulate the relative attractiveness of the online- against the walk-in- channel (by allowing the use of dynamic price discounting). Both tools help the principal alleviate the double moral hazard problem: we derive optimal strategies for using those tools to guarantee the highest profits.

Modern traceability technologies promise to improve supply chain management by simplifying recall procedures, increasing demand visibility, or ascertaining sustainable supplier practices. Managers in the dozens of traceability initiatives developing such technologies face a difficult question: which companies should they target as early adopters to ensure that their technology is broadly employed? To answer this question, managers must consider an extended supply chain effect that is inherent to traceability technologies. Namely, the benefits obtained from traceability are conditional on technology adoption throughout a product's supply chain. This effect, together with the fact that supply chains are interlinked in complex networks, makes the problem of choosing early adopters complex and difficult to solve. Our first step in tackling the question of selecting the smallest set of early adopters is to introduce a new model of the dynamics of traceability technology adoption in supply chain networks. Similar to extant diffusion models, our model specifies new adopters based on past adopters. Unlike other models, however, it incorporates extended supply chain effects. We show that the problem of selecting the smallest seed set is NP-hard and that no approximation to within a polylogarithmic factor can be obtained for any polynomial-time algorithm. Nevertheless, we introduce a procedure that identifies an exact solution in polynomial time under certain assumptions about the network structure. We provide evidence that our procedure is tractable for real-world supply chain networks. Our results further provide insights into the relationship between network structures and the optimal set of firms to target. In particular, they suggest that small, isolated firms may be favored over large, highly connected ones.

We model the creation of a new venture with a novel drift-variance diffusion control framework in which the state of the venture is captured by a diffusion process. The entrepreneur creating the venture chooses costly controls, which determine both the drift and the variance of the process. When the process reaches an upper boundary, the venture succeeds and the entrepreneur receives a reward. When the process reaches a lower boundary, the venture fails. The entrepreneur can choose between two different controls and wishes to determine the policy that maximizes the expected total reward minus total cost. We derive closed-form expressions under which the optimal policy will be dynamic versus static. The results reveal a subtle trade-off between the cost of the two controls, their drift and their variances.

Service systems’ behavior can be affected by multiple factors. In the case of intensive care units (ICUs), which admit patients from four primary loci (the emergency department (ED), scheduled patients, planned transfers from other ICUs, and unplanned transfers), it is known that admission rates of some patients decrease as occupancy increases. It is also known that, for at least some conditions, ICU admission is not just a function of patients’ illness. Instead, a significant proportion of the variation in ICU admission rates is due to hospital, not patient, factors. In this paper, we employ two years of data from patients admitted to 21 Kaiser Permanente Northern California ICUs from the ED. We quantify the variation in ICU admission from the ED under varying degrees of ICU and ED occupancy. We find that substantial heterogeneity in admission rates is present, and that it cannot be explained either by patient factors or occupancy levels alone. We use a structural model to understand the extent that intertemporal externalities could account for some of this variation. Specifically, we identify the discount factor in the structural model from observed data using a novel econometric approach. We find there is large heterogeneity in the discount factors across hospitals, suggesting they behave very differently when balancing the short and long-term considerations. Using counterfactual simulations, we show that, if hospitals had more information regarding their behaviors, and if it were possible to alter hospital admission processes to incorporate such information, hospitals could reduce ICU congestion safely. This type of intervention can be implemented via a simple heuristic policy that achieves most of the benefit.

Classical statistical learning distinguishes between offline learning and online learning. Motivated by the idea of bridging the gap between these two different types of learning tasks, this work investigates the impact of pre-existing offline data on the online learning in the context of a dynamic pricing problem. We consider a seller offering a single product with an infinite amount of inventory over a selling horizon. The demand in each period is determined by the price of the product according to a linear demand model with unknown parameters. We assume that the seller has some pre-existing offline data before the start of the selling horizon, and wants to utilize both the preexisting offline data and the sequentially-revealed online data to minimize the regret of the online learning process. We characterize the joint effect of the size, location and dispersion of the offline data on the optimal regret of the online learning. Our results reveal surprising transformations of the optimal regret rate with respect to the size of the offline data, which we refer to as phase transitions. In addition, our results also demonstrate that the location and dispersion of the offline data have an intrinsic effect on the optimal regret, which is quantified via the inverse-square law.