We study demand for symmetric and asymmetric information sources in the laboratory. Although participants respond to incentives and instrumental considerations, they exhibit a systematic bias towards symmetric sources. These patterns persist among individuals with higher cognitive abilities and better task comprehension, and they are not accounted for by their elicited subjective beliefs.
We propose an exact linear programming (LP)-based solution approach to the Guaranteed Service Model (GSM), one of the most widely applied models for optimizing safety stock placement in supply chain networks. Our approach handles any directed acyclic network and any cost function that depends on a stage’s incoming and outgoing service times. It scales polynomially in the number of nodes n in the network, pseudo-polynomially with respect to the bit size of the maximum replenishment time M, and (for fixed M) exponentially in its treewidth, which quantifies how “tree-like” a network is and can be much smaller than n. This contrasts with existing approaches, which scale exponentially in n. The proof of exactness relies crucially on showing that the join of transportation-like polytopes remains integral and is more broadly applicable to other Operations Management problems. In addition to an exact formulation, our LP-based approach enables a practical solution strategy for the GSM built on a hierarchy of LP relaxations. These relaxations provide valid lower bounds, certify optimality when integral, and can strengthen existing exact methods. In our computational study, the smallest relaxation already recovers an optimal GSM solution on every real-world benchmark instance, leading to substantial speed-ups over the state-of-the-art exact algorithm and commercial general-purpose solvers. The framework also supports sensitivity analysis and accommodates additional operational constraints. Overall, our approach builds a new bridge between Operations Management and Computer Science, providing new theoretical foundations and practical tools for managing safety stocks in complex modern supply chain networks.
By extracting information from economic news articles, this paper proposes a novel measure of narrative-based ambiguity that captures investors’ attitudes toward narrative uncertainty and exhibits strong in- and out-of-sample predictive power for the stock market returns. It reveals general ambiguity aversion except in high loss probability scenarios in which ambiguity tolerance emerges. Further tests confirm the significant pricing power and distinct information of the narrative ambiguity on top of existing ambiguity measures, such as survey- and return-based ambiguity. By aligning industry-specific narrative ambiguity, we construct a superior predictor for market returns with the predictability predominantly driven by the ambiguity from consumption-, energy-, and technology-related sectors. Our findings also carry broad implications as the predictability remains significant across international markets and other asset classes. This paper has been accepted by Will Cong for the Virtual Special Issue on Digital Finance.
Supply uncertainty, characterized by stochastic lead times and random supply quantities, has attracted increasing attention from academia, industries, and governments, particularly in the aftermath of the COVID-19 pandemic. In this paper, we consider the problem of managing lost-sales inventory systems with general supply uncertainty: stochastic lead times and random supplies. Unlike the previous studies, we assume the decision maker has no prior information on the stochastic demand and supply. We propose the first provably effective learning algorithm for inventory management problems with censored demand and supply data under general supply uncertainty. Then, we establish a cumulative regret of [Formula: see text] for this learning algorithm compared with the best constant-order policy, where [Formula: see text] is the upper bound of the random part, and L is the deterministic part of the stochastic lead times. We also conduct numerical experiments to demonstrate the effectiveness of our algorithm. Our approach lies in developing a new framework for transformed convexity. Furthermore, we address the unique challenges of our problem through new techniques, for example, estimating the long-run cost by establishing coupling and concentration results utilizing the system structures. These techniques are also of independent interest. Beyond our problem, our framework provides broad implications for other operations management (OM) problems exhibiting transformed convexity.
Unstructured text provides decision-makers with a rich data source in many domains, ranging from product reviews in retail to nursing notes in healthcare. To leverage this information, words are typically translated into word embeddings---vectors that encode the semantic relationships between words---through unsupervised learning algorithms such as matrix factorization. However, learning word embeddings from new domains with limited training data can be challenging, because the meaning/usage may be different in the new domain, e.g., the word ``positive'' typically has positive sentiment, but often has negative sentiment in medical notes since it may imply that a patient tested positive for a disease. In practice, we expect that only a small number of domain-specific words may have new meanings. We propose an intuitive two-stage estimator that exploits this structure via a group-sparse penalty to efficiently transfer learn domain-specific word embeddings by combining large-scale text corpora (such as Wikipedia) with limited domain-specific text data. We bound the generalization error of our transfer learning estimator, proving that it can achieve high accuracy with substantially less domain-specific data when only a small number of embeddings are altered between domains. Furthermore, we prove that all local minima identified by our nonconvex objective function are statistically indistinguishable from the global minimum under standard regularization conditions, implying that our estimator can be computed efficiently. Our results provide the first bounds on group-sparse matrix factorization, which may be of independent interest. We empirically evaluate our approach compared to state-of-the-art fine-tuning heuristics from natural language processing.
The United States Food and Drug Administration (FDA)’s 510(k) pathway allows manufacturers to gain medical device approval by demonstrating substantial equivalence to a legally marketed device. However, the inherent ambiguity of this regulatory procedure has been associated with high recall among many devices cleared through this pathway, raising significant safety concerns. In this paper, we develop a combined human-algorithm approach to assist the FDA in improving its 510(k) medical device clearance process by reducing recall risk and regulatory workload. We first develop machine learning methods to estimate the risk of recall of 510(k) medical devices based on the information available at the time of submission. We then propose a data-driven clearance policy that recommends acceptance, rejection, or deferral to FDA’s committees for in-depth evaluation. We conduct an empirical study using a unique data set of over 31,000 submissions that we assembled based on data sources from the FDA and Centers for Medicare and Medicaid Services. Compared to the FDA’s current practice, which has a recall rate of 10.3% and a normalized workload measure of 100%, a conservative evaluation of our policy shows a 32.9% improvement in the recall rate and a 40.5% reduction in the workload. Our analyses further suggest annual cost savings of approximately $1.7 billion for the healthcare system driven by avoided replacement costs, which is equivalent to 1.1% of the entire United States annual medical device expenditure. Our findings highlight the value of a holistic and data-driven approach to improve the FDA’s current 510(k) pathway.
This study investigates how the platform’s disintermediation governance policy (i.e., the policy that disciplines complementors and consumers circumventing the platform to transact directly) affects complementors’ product innovation and launching strategies. We leverage a change in the Communication Guideline on Amazon.com (i.e., the focal platform), which prohibits complementors (i.e., sellers) from sending external website links, telephone numbers, and email addresses to buyers in direct messaging, as a policy shock to Amazon complementors. By conducting a difference-in-differences analysis, we find that the governance policy significantly reduces product innovation on the focal platform among complementors with direct selling channels as measured by the number of new products and the innovativeness of new products. These main effects are mitigated by complementor reputation on Amazon and complementor multihoming. Mechanism analyses show that complementors with direct selling channels increase the number of new products launched through those channels after the policy change (i.e., the switching effect). Moreover, complementors tend to strategically switch innovative and high-value products away from the focal platform while leaving less innovative but price-competitive products on it to attract Amazon consumers before diverting to their direct selling channels (i.e., the bait effect). Our results have implications for platform governance policies regarding disintermediation and its impact on complementor product innovation and launch strategies both on and off the focal platform.
We study periodic review perishable inventory systems with a fixed product lifetime, positive replenishment lead times, and a general issuance policy under the average cost criterion. The optimal replenishment policy for such systems is notoriously complex and computationally intractable because of the curse of dimensionality. To address this challenge, we propose a class of projected inventory level (PIL) policies, which maintain a constant expected on-hand inventory level, and compare them with conventional base-stock (BS) policies that maintain a constant inventory position. For both backlogging and lost-sales systems, we show that the best PIL policy is asymptotically optimal with large unit penalty costs for a broad class of unbounded demand distributions. When demand is bounded and the unit penalty cost is sufficiently large, we prove that the best BS policy is optimal under first-in-first-out issuance, whereas the best PIL policy is optimal under last-in-first-out issuance (under certain conditions). Furthermore, we show that both policies are asymptotically optimal as the demand population size grows large, and their optimality gaps diminish exponentially fast in backlogging systems under a broad range of issuance policies. To facilitate computation, we introduce a class of approximate PIL (APIL) policies and extend most theoretical results for PIL to the APIL policy. Numerical results show that both PIL and APIL policies perform very close to optimal and significantly outperform BS policies.
We examine the decision to become an entrepreneur in the context of online business. Using data on entrepreneurs from a large e-commerce marketplace in China, we find that individuals are more likely to start online businesses after observing the salient success of nearby stores. Leveraging the platform’s store rating system and detailed geographic information in rural China, we show that an upgrade in a local store’s rating leads to a significant increase in new store entries nearby. This effect diminishes with physical distance and intensifies with the salience of the upgrade. However, these entry decisions appear to be suboptimal; entrants driven by local upgrade events underperform in sales and are more likely to exit, especially during downturns in overall market conditions. Our findings are consistent with salience-based theories of decision making, in which individuals overweight highly visible local information.
People and organizations often set goals to self-motivate and achieve better outcomes in challenging tasks. But goals, and their effectiveness, might depend on what people expect to happen. Do goals reflect expectations or do goals set expectations? How do goals and expectations affect performance? These distinctions are important for motivation and intervention design. We run an online real-effort task to answer these two questions by introducing exogenous variation in goals and expectations. First, we find that goals mostly reflect existing expectations rather than set expectations. Second, practicing an easier version of a task leads to higher expectations and higher performance. Eliciting a goal leads to higher performance as well. However, controlling for expectations, changing the difficulty of the goal has no discernible effect. These results suggest that people benefit from being optimistic and setting a goal, but they cannot fool themselves into expecting and doing more simply by choosing a higher goal.