with Ahmet Colak, Sina Golara, and Achal Bassamboo

with Ahmet Colak, Sina Golara, and Achal Bassamboo

Automobile sales platforms have proliferated with the growth of online commerce. Interestingly, these platforms vary widely in their information-sharing policies including what textual and visual information about car features they provide. In this paper, we study novel feature data from Cars.com, a leading online car sales platform. Existing auto industry research focused on exploring the base characteristics of cars, but Cars.com data includes car feature information, which enables us to explore 209,567 unique car features using a national sample from 13,869 dealerships and over 50 million daily listings. In particular, we develop a two-stage estimation framework that estimates (i) the information value of features, (ii) over- and underpricing strategies of car dealers, and (iii) sales outcomes (in terms of days on the market) that stem from feature pricing. Our results reveal that, on average, features alone can account for over 70% of price variation. Moreover, we find that high-value features mitigate the overpricing effect, particularly for new vehicles and luxury brands. Our auto features data contribute to an emerging information-sharing literature, and we document novel managerial insights about overpricing that act as omitted effects in the absence of features information.

with Ahmet Colak, Sina Golara, Achal Bassamboo

Recent studies in inventory management have focused on the impacts of varietyon sales performance. However, disruptions such as COVID-19 can impose replenishment constraints, preventing vendors from utilizing the insights gained in these studies. In this paper, we study the multi-dimensional aspects of variety during shortages in the automotive industry.

with Achal Bassamboo and Ohad Perry

We consider the optimal-control problem in two-class systems involving a new product launch. The population in each class can be in one of three states: pre-purchase, post-purchase, and no-purchase. Consumers transition from a pre-purchase state to either post-purchase by buying the product or to no-purchase by leaving the potential consumer pool. In particular, they transition to no-purchase either spontaneously or through social interactions.

with Ava Liberman, Dustin French, and Shyam Prabhakaran [Link]

Background and Purpose: Differentiating ischemic stroke patients from stroke mimics (SM), nonvascular conditions which simulate stroke, can be challenging in the acute setting. We sought to model the cost-effectiveness of treating suspected acute ischemic stroke patients before a definitive diagnosis could be made. We hypothesized that we would identify threshold proportions of SM among suspected stroke patients arriving to an emergency department above which administration of intravenous thrombolysis was no longer cost-effective. Methods: We constructed a decision-analytic model to examine various emergency department thrombolytic treatment scenarios. The main variables were proportion of SM to true stroke patients, time from symptom onset to treatment, and complication rates. Costs, reimbursement rates, and expected clinical outcomes of ischemic stroke and SM patients were estimated from published data. We report the 90-day incremental cost-effectiveness ratio of administering intravenous thrombolysis compared with no acute treatment from a healthcare sector perspective, as well as the cost-reimbursement ratio from a hospital-level perspective. Cost-effectiveness was defined as a willingness to pay <$100 000 USD per quality adjusted life year gained and high cost-reimbursement ratio was defined as >1.5. Results: There was an increase in incremental cost-effectiveness ratios as the proportion of SM cases increased in the 3-hour time window. The threshold proportion of SM above which the decision to administer thrombolysis was no longer cost-effective was 30%. The threshold proportion of SM above which the decision to administer thrombolysis resulted in high cost-reimbursement ratio was 75%. Results were similar for patients arriving within 0 to 90 minutes of symptom onset as compared with 91 to 180 minutes but were significantly affected by cost of alteplase in sensitivity analyses. Conclusions: We identified thresholds of SM above which thrombolysis was no longer cost-effective from 2 analytic perspectives. Hospitals should monitor SM rates and establish performance metrics to prevent rising acute stroke care costs and avoid potential patient harms.

with Ohad Perry, Jane Holl, and Shyam Prabhakaran

Most acute stroke (AS) patients in the United States are initially evaluated at a primary stroke center (PSC) and a significant proportion requires transfer to a comprehensive stroke center (CSC) for advanced treatment. A CSC typically accepts patients from multiple PSCs in its network, leading to capacity limits. This study uses a queueing model to estimate impacts on CSC capacity due to transfers from PSCs. The model assumes that the number of AS patients arriving at each PSC, proportion of AS patients transferred, and length of stay in the CSC Neurologic Intensive Care Unit (Neuro-ICU) by type of AS are random, while the transfer rates of ischemic and hemorrhagic AS patients are control variables. The main outcome measure is the “overflow” probability, namely, the probability of a CSC not having capacity (unavailability of a Neuro-ICU bed) to accept a transfer. Data simulations of the model, using a base case and an expanded case, were performed to illustrate the effects of changing key parameters, such as transfer rates from PSCs and CSC Neuro-ICU capacity on overflow capacity. Data simulations of the model using a base case show that an increase of a PSC’s ischemic stroke transfer rate from 15% to 55% raises the overflow probability from 30.62% to 36.13%. Further simulations of the expanded case show that to maintain an a priori CSC overflow probability of 30.62% when adding a PSC with a AS transfer rate of 15% to the network, other PSCs would need to decrease their transfer rate by 12.5% or the CSC Neuro-ICU would need to add 2 beds. A queuing model can be used to estimate the effects of change in the size of a PSC-CSC network, change in AS transfer rates, or change in number of CSC Neuro-ICU beds of a CSC on its capacity on the overflow probability in the CSC.