Research

Large language models have shown great potential as forecasting tools in finance and economics, but backtesting performance is subject to look-ahead bias if the period overlaps with an LLM’s training window. Prompt-based attempts to avoid look-ahead bias require that LLMs understand chronology. We test LLMs’ ability to understand and enforce chronological order in three types of tasks: sorting randomly shuffled historical events; conditional sorting of events defined by some conditions; and anachronism detection based on intersections of multiple timelines. Our experiments use events that we first confirm are known to the LLM; this ensures that we test chronological understanding on an LLM’s pretrained internal knowledge. Across three LLM families— GPT-4.1 (standard), GPT-5 (hybrid-reasoning), and Claude 3.7 Sonnet (large-reasoning, with and without Extended Thinking), we find that performance degrades rapidly with problem complexity but improves greatly for reasoning models with test-time extended reasoning. These patterns are important for the real-time application of LLMs in finance.

Most LLM benchmarks rank models on their ability to automate work tasks. In practice, however, models are often used to assist other (human or LLM) agents. The question that drives model selection is therefore not only which model produces the best output, but which model most improves the work of another (weaker) agent. We introduce a unified framework that evaluates the capability of models to automate and augment another agent’s performance. Across seven economically grounded real-world tasks, an assistant model writes assistance text for a standardized lower-capacity worker model, which produces the deliverable. In automation mode, the assistant produces the output directly. Outputs are scored through blind pairwise comparisons by an LLM judge panel with task-specific rubrics, replicated across ten runs. Rankings across the two regimes are only modestly correlated, and the automation winner loses augmentation on five of seven tasks. Assistance is not reliably positive. The unaided worker outranks every assisted condition on three tasks, and only one model’s guidance beats no guidance on average. These results suggest that automation ability is an incomplete proxy for assistance quality, motivating benchmarks that evaluate models according to the roles they play in human-AI and multi-agent systems.

This case introduces Calyber, a simulation-based game designed to provide a hands-on and engaging experience in developing real-time pricing and matching decisions for shared ride services, where multiple riders are pooled into a single vehicle. Students design and implement dynamic pricing and matching policies using a rich historical ridesharing data set, competing for top performance on a holdout test set. Through this case, students gain practical insight into stochastic dynamic decision making within a modern, relevant, and data-driven context. Results from previous class implementations provide strong evidence of enhanced learning and engagement.

Motivated by applications such as rapid transit and make-to-order systems, we study a fluid model of a many-server on-off system with multiple queues. Each server visits the queues in a prescribed order. A queue is “on” when it is being served and “off” otherwise. Customers of each queue arrive over time and decide whether to join. Services are exhaustive: servers must clear each queue before departing, while the planner determines how long they remain after each queue is first cleared. We characterize the customer equilibrium as the solution to a bounded linear complementarity problem and develop an algorithm with closed-form updates that computes the unique equilibrium in at most n steps, where n is the number of queues. We then develop an O(n^2)-time algorithm, independent of the number of servers, for computing the optimal service policy. We further show that, under an optimal policy, service is extended beyond first clearance at no more than m queues, where m is the number of servers. This structural result yields a sharper optimization procedure that requires at most 2n steps in the single-server case. We also extend our analysis to settings beyond exhaustive services. We illustrate these results using data from Bay Area Rapid Transit, where we optimize train schedules and station dwell times, recovering almost all lost ridership due to prolonged waiting times without adding vehicles.