Latest Seminars

Joint Statistics Seminar - Preferential Attachment and Neutral Random Graphs: Statistically Useful Generative Models of Network Data
Prof. Benjamin Bloem-Reddy, Department of Statistics, University of Oxford

Preferential attachment (PA) and other probabilistic generative models of network growth have been popular for their ability to explain large-scale phenomena from simple interaction mechanisms.  However, PA has been of limited use as a statistical model, due to its lack of exchangeability: in a statically observed network with n edges, inference requires considering all n! possible edge arrival orders.  Moreover, in models based on forms of exchangeability, inference algorithms benefit from an edge-decoupled representation, in which all dependence between edges is captured by some latent quantity; no such representation is known for PA models.  I will describe my work toward making PA useful as a statistical model: an edge-decoupled representation for a class of generalized PA models is established, and it reveals probabilistic structure, called left-neutrality, that can be exploited for efficient inference algorithms even in the presence of unknown edge arrival order.  Furthermore, the edge-decoupled representation endows the PA model with a set of interpretable model parameters.  Finally, I will describe how exchangeability still plays a role, despite PA's non-exchangeability.

This work was done in collaboration with Christian Borgs, Jennifer T. Chayes, Adam Foster, Emile Mathieu, Peter Orbanz, and Yee Whye Teh.

 

Date 17.05.2019
Time 11:00am – 12:00noon
Venue Room 6045 (LSK Business Building)

Who Gets the Attention? The Interactions among Similar Social Media Content
Professor Bin Gu, Earl and Gladys Davis Distinguished Professor, W P Carey School of Business, Arizona State University

Date 15.04.2019
Time 2:30 - 4:00 pm
Venue ISOM Conference Room 4047, LSK Business Building

Stochastic Optimization with Decisions Truncated by Random Variables
Prof Xiangyu Gao, Department of Decision Sciences and Managerial Economics, The Chinese University of Hong Kong

Many operations management problems can be formulated as stochastic optimization problems in which the decision variables are truncated by some random variables. Examples include inventory control with random capacities and capacity allocation problems with random demands. An intrinsic challenge arises from the fact that the truncation by random variables may destroy convexity of the underlying optimization problem. We develop a transformation technique to convert the original non-convex problems to equivalent convex ones. Our transformation allows us to prove the preservation of some desired structural properties, such as convexity, submodularity, and L-natural-convexity, under optimization operations, that are critical for identifying the structures of optimal policies and developing efficient algorithms.

Date 12.04.2019
Time 11:00 am - 12:15 pm
Venue Room 1034, LSK Business Building

Inducing Compliance with Post-Market Studies for Drugs under FDA’s Accelerated Approval Pathway
Dr. Hui Zhao, Associate Professor of Supply Chain Management and Charles and Lilian Binder Faculty Fellow, Smeal College of Business, Pennsylvania State University

In 1992, FDA instituted the accelerated approval pathway (AP) to allow promising drugs to enter the market based on limited evidence, but requiring manufacturers to verify the drugs’ true clinical benefits through post-market studies. However, most post-market studies are not completed due to many incentive issues, and FDA must endure an onerous process to withdraw an unproven drug from the market when a post-market study is uncompleted. The prevalence of this non-compliance problem poses considerable public health risk, compromising the original purpose of a well-intentioned AP initiative. We address this problem by providing an internally consistent and implementable solution through a comprehensive analysis of the myriad complicating factors and tradeoffs facing FDA, including information asymmetry and moral hazard. Specifically, we adopt a Stackelberg framework in which regulator, which cannot observe manufacturer’s private cost information or level of effort, leads by imposing a post-market study deadline. The profit-maximizing manufacturer then follows by establishing its level of effort to invest in its post-market study. We develop a deadline-dependent user fee mechanism that establishes an incentive for manufacturer compliance. We show that effectiveness of the mechanism in inducing compliance depends fundamentally on what we distill as the enforceability of sanction (s), a drug-specific measure that indicates how difficult it is to withdraw a drug from the market, and the drug’s success probability (alpha): The higher is either, the higher is the probability that the mechanism induces compliance. Using data for a real drug, we calibrate our model and quantify the value of such a mechanism and its impact. We also discuss alternatives when such a mechanism is less effective. From public policy perspective, we provide guidance for FDA to increase the viability and effectiveness of AP.

Date 29.03.2019
Time 11:00 am - 12:15 pm
Venue Room 3001, LSK Business Building

Joint Statistics Seminar - Aggregate Asymmetry in Idiosyncratic Jump Risk
Prof. Viktor Todorov, Department of Finance, Kellogg School of Management, Northwestern University

We study the structure and pricing of idiosyncratic jumps, i.e., jumps in asset prices that occur outside market-wide jump events.  Using options on individual stocks and the market index that are close to expiration as well as local estimates of market betas from returns on the underlying assets, we estimate nonparametrically the asymmetry in the risk-neutral expected idiosyncratic variation, i.e., the difference in variation due to negative and positive returns, which asymptotically is solely attributed to jumps.  We derive a feasible Central Limit Theorem that allows to quantify precision in the estimation, with the limiting distribution being mixed Gaussian.  We find strong empirical evidence for aggregate asymmetry in idiosyncratic risk which shows that such risk clusters cross-sectionally.  Our results reveal the existence and non-trivial pricing of aggregate downside tail risk in stocks during market-neutral systematic events as well as a negative skew in the cross-sectional return distribution during such episodes.

 

Date 27.03.2019
Time 11:00am – 12:00noon
Venue Room 4047 (LSK Business Building)

Referral, Learning and Inventory Decisions in Social Networks
Prof Guangwen Kong, Department of Industrial & Systems Engineering, University of Minnesota

In the past decade, with the proliferation of digital social networks and social media such as Facebook, Twitter and Instagram, we have observed that businesses have been increasingly using referral programs to increase their market exposures and sales. We examine the impact of social learning in a referral program when customers' preferences are positively correlated in a social network. We characterize customers’ purchasing strategies based on their information types, and derive the demand distributions when customers are involved in social learning in a referral program. While customers’ lack of knowledge on their own preferences will introduce bias to the demand expectation, social learning reduces the bias at the expense of increasing demand variance. We investigate the firm's inventory decision when customers are involved in social learning in a referral program. We find that the stock-out of one product would suppress the demand of the other product when customers are involved in social learning. Allowing customers to make multiple referrals would reduce the negative externality of stock-out but meanwhile significantly increases the demand variance. The optimal design of referral program generates market exposure with a moderate increase of inventory cost.

Date 22.03.2019
Time 11:00 am - 12:15 pm
Venue Room 3001, LSK Business Building

The Financing Role of Inventory: Evidence from China’s Metal Industries
Dr. Jing Wu, College of Business, City University of Hong Kong

Classical inventory theory examines various factors on the matching of demand and supply. In this paper, we find the model and empirical evidence for the financing role as an essential complementary factor. The- oretically, we analyze a model to show that inventory can be leveraged to obtain financial gains thus the optimal inventory level is associated with the investment opportunities. Empirically, we first show at the country-level, China’s inventory of copper, aluminum, and zinc are driven by the investment returns after controlling for other explanations, including price trajectory, currency risk, industrial demand, and economic uncertainty. Then we confirm the result at the firm-level using data from China’s manufacturing sector, and further establish the financing mechanism especially for the metal industries. The economic impact of the financing role is significant compared to the demand- or supply-side factors, suggesting that classical inventory theory should incorporate the financing role in making the optimal policy decision.

Date 15.03.2019
Time 11:00 am - 12:15 pm
Venue Room 3003, LSK Business Building

Joint Statistics Seminar - Data Science in Academia and Business
Dr. Philippe Barbe, CMG

In this talk I will discuss the similarities and differences of data science and related fields in academia and business. This is based on 20+ years of experience in academia and the last 5 in business.

Date 14.03.2019
Time 2:00pm – 3:00pm
Venue Venue: Room 4047 (LSK Business Building)

Disaster Relief Resource Pre-position and Redeployment Facing the Adversarial Nature
Prof Sarah Yini Gao, Lee Kong Chian School of Business, Singapore Management University

Large-scale natural disaster management is among the top list of worldwide challenges. One of the difficulties faced is that minimal historical information can be used for future disaster prediction and risk assessment due to the rareness of such events. A common practice adopted is to apply a robust approach to model the risk and ambiguity-averseness. However, lack of risk information posts another layer of challenge on how to choose a proper uncertainty set. In our paper, we model the natural disaster event as an adversary where the risk is not only oblivious, but it changes when preparation and mitigation strategies change.

We adopt a two-person zero sum game theoretical framework to model the interactions between an adversarial nature and a disaster manager who can pre-position disaster relief item inventory in multiple locations and then redeploy the resources after observing the realized relief item demands at each location. Our goal is to find the optimal pre-position and redeployment strategies facing adversarial nature, and study how redeployment network can affect the optimal strategy.

In general, finding an equilibrium pre-position and redeployment strategy is challenging. For a general redeployment network structure, we obtain an equivalent reformulation of the problem using a conic program. We then show that when the redeployment structure has a special form, such as a “k-chain” structure, closed-form equilibria to these problems can be obtained. We further explore how the redeployment network structure affects the game, showing that a sparse redeployment network structure can already capture the value of redeployment. We finally demonstrate the framework via a practical case study of Yushu Earthquake in the numerical study.

Date 01.03.2019
Time 11:00 am - 12:15 pm
Venue Room 3003, LSK Business Building

Statistics Seminar - A Self-Normalized Approach to Sequential Change-point Detection for Time Series
Mr. Wai Leong NG, Department of Statistics, The Chinese University of Hong Kong

This paper proposes a self-normalization sequential change-point detection method for time series.  In monitoring for parameter changes in real time, most of the traditional sequential monitoring schemes utilize a CUSUM-based test statistic, which involves a long-run variance estimator.  However, the commonly used long-run variance estimators require the choice of bandwidth parameter which could be sensitive to the performance. Moreover, the traditional schemes usually suffer from severe size distortion due to the slow convergence rate to the limit distribution in the early monitoring stage. In this article, a self-normalization method is proposed to tackle these issues.  We establish null asymptotic and the consistency of the proposed sequential change-point test under general regularity conditions.  Simulation experiments and applications to railway bearing temperature data are conducted for illustrations.

 

Date 27.02.2019
Time 11:00am – 12:00noon
Venue Room 4047 (LSK Business Building)

Short-term Housing Rentals and Corporatization of Platform Pricing
Dr. Mehmet Gumus, Desautels Faculty of Management, McGill University

In recent years, we have seen the emergence of a number of platforms that facilitate short-term peer-to-peer rentals of assets as part of the bigger collaborative consumption or shared services movement. Although all platforms are similar in that their primary goal is to facilitate asset sharing and the above two channels, they differ in other aspects. In this paper, we focus on one such aspect - how they decide on the price to charge their customers. Specifically, in a platform like AirBnB, the price is effectively set based on a market mechanism that matches supply and demand. But, some other platforms like corporatestays.com and guestbnb.ca are more active. They take turnkey control of the assets and determine the price on "behalf" of the owners that maximizes their profits based on them being paid a share of the price. Our primary goal in this paper is to understand the implications of this difference in pricing strategy for the direct stakeholders of the platform such as customers, owners and the platform as well as indirect stakeholders such as long-term rentals and hotels.

Date 25.02.2019
Time 11:00 am - 12:15 pm
Venue Room 3005, LSK Business Building

Joint Statistics Seminar - High-dimensional Q-learning for Dynamic Treatment Regimes
Prof. Rui Song, Department of Statistics, NC State University

Dynamic treatment regimes are a set of decision rules and each treatment decision is tailored over time according to patients’ responses to previous treatments as well as covariate history. There is a growing interest in development of correct statistical inference for optimal dynamic treatment regimes to handle the challenges of nonregularity problems in the presence of nonrespondents who have zero-treatment ef-fects, especially when the dimension of the tailoring variables is high. In this talk, we propose a high-dimensional Q-learning (HQ-learning) to facilitate the inference of optimal values and parameters. The proposed method allows us to simultaneously estimate the optimal dynamic treatment regimes and select the important variables that truly contribute to the individual reward. At the same time, hard threshold-ing is introduced in the method to eliminate the effects of the nonrespondents. The asymptotic properties for the parameter estimators as well as the estimated optimal value function are then established by ad-justing the bias due to thresholding. Both simulation studies and real data analysis demonstrate satisfac-tory performance for obtaining the proper inference for the value function for the optimal dynamic treat-ment regimes.

Date 19.02.2019
Time 3:00pm – 4:00pm
Venue Room 6045 (LSK Business Building)

It Takes Two to Tango: The Effects of Internal and External Information Integration on Healthcare Process and Outcomes
Prof Hillol Bala, Kelley School of Business, Indiana University

Date 18.02.2019
Time 2:00 - 3:30 pm
Venue ISOM Conference Room 4047, LSK Business Building

Membership-Based Free Shipping Programs: A New Vehicle to Gain Competitive Advantage for Online Retailers?
Mr Geng Sun, Naveen Jindal School of Management, University of Texas at Dallas

Date 15.02.2019
Time 3:15 - 4:45 pm
Venue Room 3003, LSK Business Building

Mechanism Design in Large Cloud Computing Systems
Mr Yingda Zhai, University of Texas at Austin

Date 13.02.2019
Time 3:15 - 4:45 pm
Venue Room 3003, LSK Business Building