PMPanos Markou

Panos Markou

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Panos Markou has a total of 15 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Tutorial: Extracting Unstructured Text Using Large Language Models

    by Simon Spavound, Oliver Schaer, Panos Markou

    This tutorial shows how to build a two-stage pipeline in which a multimodal large language model reads each page as an image and a second model turns the extracted text into structured JSON. We cover the choices that make such pipelines reliable and affordable, including structured outputs, parameter control, task decomposition, cost-effective model selection, and validation against a human-verified ground truth.

  • Open Access

    Tutorial: Extracting Unstructured Text using Large Language Models

    by Simon Spavound, Oliver Schaer, Panos Markou

    Motivation: Unstructured text data are valuable but underutilized resources in analytics due to the lack of reliable, scalable, and cost-effective extraction methods. The growing use of Large Language Models (LLMs) is increasing the demand for clean input data from such sources, yet users lack practical guidance on incorporating these models into analytical workflows. Tutorial overview: We introduce the programmatic building blocks for working with LLMs in analytics pipelines, including API-based deployment, structured outputs, temperature control, multimodal processing, and cost-effective model selection. We then construct an end-to-end pipeline that transforms unstructured PDF documents into structured, analysis-ready JSON data. Using U.S. Food and Drug Administration Advisory Committee transcripts as a running example, we demonstrate a two-stage approach: vision-based text extraction followed by LLM-driven text structuring. Throughout, we emphasize transferable design principles, including task decomposition and parameter control. Downstream analysis and implications: We show how the resulting structured data enables quantitative analysis that would not be feasible on the raw source documents. Using extracted speaker-statement pairs, we compute speaker-level word counts and sentiment scores, illustrating how these features can characterize stakeholder dynamics in regulatory deliberations. More broadly, the pipeline provides an extensible framework for converting unstructured documents into inputs suitable for forecasting, classification, or decision-making applications. All code is publicly available.

  • Top Cited Papers

    Incentivizing Information Exchange Within Groups: The Role of Voting Protocols in U.S. Food and Drug Administration Advisory Committees

    by Panos Markou, Tian Heong Chan

    Complex and important decisions are often made with advice from a committee of experts. But how do a committee’s “rules of engagement” affect the way individuals discuss, how they vote, and ultimately the quality of their collective recommendation? Compiling verbatim transcripts from U.S. Food and Drug Administration advisory committee meetings, we study how a 2007 switch from sequential to simultaneous voting procedures changed discussions, information exchange, and decision making. Consistent with past findings, we show that, compared with a sequential voting protocol, simultaneous voting led to a reduction in the likelihood of unanimous votes. Importantly, we show novel evidence that the majority of this reduction in unanimity was mediated by changes in discussion patterns—specifically, by the increased diversity of information surfaced during discussions. We also find evidence of behavioral and linguistic changes that support our theory that voting protocols changed the incentives for members to elicit more diverse information from each other: under simultaneous voting, members exhibited greater equality in talking time, directed a greater proportion of questions to each other, and adopted language that was more positive, authentic, and equal in projecting status and confidence. Finally, we show that recommendations under simultaneous voting were more likely to be accurate, as drugs recommended and approved were less likely to encounter safety-related postmarket events. In sum, voting protocols affect the incentives for individuals to engage in robust discussions, leading to marked improvements in how information is exchanged between individuals, and in the process by which groups of experts arrive at joint recommendations. This paper was accepted by Sridhar Tayur, entrepreneurship and innovation. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.03649 .


  • Automotive Procurement Under Opaque Prices: Theory with Evidence from the BMW Supply Chain

    by Danko Turcic, Panos Markou, Panos Kouvelis, Daniel Corsten

    Several features of automotive procurement distinguish it from the prototypical supply chain in the academic literature: pass-through pricing that reimburses suppliers for raw material costs, market frictions that prohibit cost transparency and imbue suppliers with pricing power, and contractual commitments that span multiple production periods. In this context, we formalize a procurement model by considering an automaker that buys components from an upstream supplier to assemble cars over several production periods in an environment where period demands and raw material costs are both stochastic. Our paper clarifies how information asymmetry and market factors that amplify or weaken this asymmetry affect the firms’ procurement protocol preferences. Then, using proprietary contract and supplier data from BMW, we empirically validate this model and show that it reflects BMW’s reality: the factors that should theoretically go into automotive procurement decisions do so. Our analysis also reveals that existing contracting protocols in this context are not optimal for procurement under asymmetric information, and so we propose an alternative contracting method. We calibrate our model and estimate an automaker’s performance improvement from this optimal contract over the status quo. This paper was accepted by Vishal Gaur, operations management. Supplemental Material: The data files and online appendices are available at https://doi.org/10.1287/mnsc.2023.4880 .


  • Open Access

    Changing Course: How Analogies Direct Pivots

    by Jeremy Hutchison-Krupat, Panos Markou

    Innovation is exciting because it is full of immense possibility. However, immense possibility is difficult to comprehend. To resolve this, an innovator must interpret or frame the opportunity to reduce the many possible dimensions to just a few prioritized dimensions, which translates to a strategy and a course of action. Yet, rarely does an innovative endeavor follow its original course of action or result in an outcome that replicates what was originally envisioned. Indeed, successful innovation evolves through iterative experimentation to implement a strategy. However, the strategy itself may also require adaptation, otherwise known as a pivot. An effective pivot requires the abandonment of a current strategy, and a decision on a new, more effective, course of action. We conduct two experiments to study how experience with analogous settings enables innovators to address these challenges and pivot effectively. These experiments require individuals to make sense of a difficult problem, adopt and iteratively implement a strategy, and experience progress towards an objective. Then, new information is revealed, which renders the initial strategy ineffective, and requires individuals to pivot their strategy to achieve success. Our results show that analogies are a remarkably effective means to enable individuals to pivot. Moreover, we disentangle the properties of analogies and provide insights on which types of analogies are most likely to prompt a pivot. Our results reveal that while analogies can increase the likelihood to pivot, some types of analogies may lead an innovator to change course, but do so for the wrong reason.

  • Rival Signals and Project Selection: Insights from the Drug Development Process

    by Panos Markou, Stylianos Kavadias, Nektarios Oraiopoulos

    Project selection decisions are complex because they must balance not only financial returns, project risk, and fit with strategy, but also competitive circumstances. A rival’s project development efforts provide two pieces of information: a market rivalry signal, indicating potentially heightened competition in a market, and a technological signal, indicating a possible solution to a problem in that market. We hypothesize that these signals affect a firm’s likelihood of project selection in opposite directions, and that the timing of the signals matters for selection. We examine the drug development pipelines of the top 15 pharmaceutical companies from 1999 to 2016 to examine how rival projects drive the decision to progress a drug from preclinical laboratory trials to clinical trials in humans. Early-stage rival projects provide a stronger market rivalry signal, and they are associated with a decreased likelihood of the firm selecting its own project to compete in the same market. Late-stage rival projects signal technological feasibility and are associated with an increase in the likelihood of selection. We then exploit heterogeneity in market potential (i.e., disorder prevalence/incidence) and a molecular compound’s technology (i.e., therapeutic modality) to independently manipulate the salience of the two signals. Finally, we provide evidence on how selection based on rival signals informs project success. Information from rival projects prompts the selection of more successful drugs, but only after a threshold when sufficient uncertainty has been resolved. This paper was accepted by Jayashankar Swaminathan, operations management. Supplemental Material: Data and the online appendix are available at https://doi.org/10.1287/mnsc.2022.4642 .


  • Financial and Operational Risk Management: Inventory Effects in the Gold Mining Industry

    by Panos Markou, Daniel Corsten

    Financial and operational risk management are central concepts at the intersection of finance, operations, and commodity risk management. Yet, empirical evidence on their effects on inventory is lacking. We use a fine‐grained data set comprising the financial and operational risk management decisions of gold miners from 2003 to 2011 to empirically assess the effects of risk management on inventory. Faced with volatile gold prices, miners may manage (output) risk financially by committing to sell future gold production and lock in prices. They may also manage (input) costs operationally by varying the quality of ore they extract and process, thereby altering the costs they incur and influencing inventory holdings. In addition to affecting profitability, we show that these two risk management strategies have implications for inventory holdings. We find that a one‐standard deviation increase in financial risk management (FRM) is associated with an 14.3% decrease in inventory, as FRM decreases the option value of delaying processing inventory. On the other hand, a one‐standard deviation increase in operational risk management (ORM) is associated with a 3.5% increase in inventory. We also find evidence that, in this context, FRM and ORM could be viewed as complements.


  • Open Access

    Process Analysis Simulation

    by Panos Markou

    This interactive exercise emulates a process wherein a virtual Card is created in a series of steps. It can be used to expose students to ideas and concepts in process analysis, management, and improvement. It may be used in classes focusing on operations management, project management, and software development. The simulation is run on the project-management web application Trello. It allows for simultaneous, real-time observation and collaboration between players. Students can observe other students working at the same time they are performing specific tasks. This exercise is suitable for students in either full-time or executive MBA programs, as well as executive education programs. Excerpt UVA-OM-1718 Dec. 16, 2020 Process Analysis Simulation Introduction A potential recruiter is looking to invite a team of high-performing individuals to a new software-development initiative they are initiating. Time and quality are of utmost importance to the recruiter, so they have developed a competition to test velocity and attention to detail. This task involves creating a set of virtual Cards. You and your team will be competing against several other teams for the positions. Production Process . . .

  • Open Access

    Information Asymmetry, Rents, and Cost Risk Allocation: An Investigation of the BMW Supply Chain

    by Danko Turcic, Panos Markou, Panos Kouvelis, Daniel Corsten

    This paper studies a dynamic procurement process used by many European automakers and BMW in particular. An automaker (she) must procure components from an upstream supplier (him) to assemble cars in a given production period. Demand for vehicles is stochastic, and retail prices are endogenous. The automaker may or may not be privy to the supplier's raw material product cost, which is also stochastic. In the former case, we say that information is symmetric; in the latter case, information is asymmetric. (Existing work in this area is limited to static models with deterministic demand and symmetric information.) The contracts BMW uses are the material-plus-surcharge (MPS) and the fixed price contract (FP). In that environment, we explain how to choose between the contracts that BMW uses currently, and we identify a new, optimal contract for the asymmetric case that would net the automaker an additional profit boost. We conclude by assessing the empirical support for our predictions using a unique data set of contractual arrangements provided to us by BMW. Our analysis sharpens our understanding of contract choice in dynamic procurement settings and identifies a list of key factors in making such choices.


  • The effect of corporate social performance on the financial performance of business‐to‐business and business‐to‐consumer firms

    by Jonathan Luffarelli, Panos Markou, Antonios Stamatogiannakis, Dilney Gonçalves

    There exists a widespread managerial belief that higher corporate social performance (CSP) increases both firm value and sales. Although numerous studies provide evidence of a positive effect of CSP on firm value, whether CSP can impact sales remains largely unknown. Can CSP influence sales? Is this effect contingent on the product‐market profile, that is, on whether firms operate in business or consumer markets? We use a panel dataset comprising 23,769 firm‐year observations to help address these questions. We find that higher CSP has a strong negative effect on sales for business‐to‐consumer firms but an insignificant or economically trivial effect for business‐to‐business firms. However, we also find that higher CSP has a positive effect on firm value for both types of firms. Taken together, these results demonstrate that higher CSP results in higher firm value but can hurt sales. We discuss the theoretical contributions and managerial implications of these findings.