Research

Research

Two research programs: management and statistics.

Management research

Strategy, innovation search, organizations, and AI coordination.

Innovation & competitive search

Job market paper

Search Incursions: A Relational Theory of Disequilibrating Inventions and Recombinant Advantage

Tejas Ramdas and Gautam Ahuja

Figure 1. One invention, different strategic consequences: Shizuoka Seiki and Combustion Research
Figure 1. One invention, different strategic consequences: Shizuoka Seiki and Combustion Research
Figure 13. Relative recombinant advantage and firms’ incursion roles
Figure 13. Relative recombinant advantage and firms’ incursion roles

How does a new invention redistribute firms' opportunities to enter one another's inventive territories, and their exposure to incursions by others?

Summary

Each novel invention can be viewed as a recombination of existing knowledge elements. By bridging previously less connected elements, an invention changes what constitutes “local” knowledge for firms and reshapes the interfirm matrix of search costs. These changes can enable firms to enter the established inventive territories of other firms; an event we call a search incursion. For each invention (I), we theorize how each firm's position in the technological knowledge network, relative to the invention and other firms, shapes its opportunities to undertake incursions and its exposure to incursions by others. To test our theory, for each focal invention, we reconstructed the technological network of U.S. patents granted during the preceding fifteen years, then introduced the invention and its citations to prior patents, holding the existing knowledge network and firms' prior knowledge positions fixed. Modeling firms as purposeful but error-prone searchers on this network, we formulated each firm’s search as a Markov decision process, and employed policy iteration to estimate how the invention changed the minimum expected cost of searching from each firm's knowledge position toward that of every other eligible firm. Calculating the invention-induced changes in the interfirm search-cost matrix allowed us to estimate the relative recombinant advantage conferred by the invention upon each firm. Separately, we used patent text and synthetic-patent constructions to identify subsequent search incursions, finding support for our predictions concerning which firms undertook incursions, which firms became susceptible to them, and which particular firm-to-firm incursions materialized.

Working paper

Shaping the Search Landscape: Generative Inventions and Follow-On Invention

Tejas Ramdas

Prior knowledge relationships and the timing of follow-on search. The central artifact illustration is conceptual.
Prior knowledge relationships and the timing of follow-on search. The central artifact illustration is conceptual.

How does the first firm to build on a breakthrough shape the inventive paths and opportunities available to firms that follow?

Summary

I develop competitive shaping search to explain how a firm's early choices of knowledge elements and linkages can affect later search around the same generative invention. Horizontal shaping makes a derivative path less attractive to other firms. Vertical shaping positions the first searcher to benefit from future recombination possibilities. Using patent data, I relate knowledge relationships established before the breakthrough to the timing of later follow-on invention.

Work in progress

Generative Inventions and Changing Recombination Spaces

Tejas Ramdas and Y. S. Wang

How a generative invention can reconfigure interfirm knowledge flows. Network changes are hypothetical.
How a generative invention can reconfigure interfirm knowledge flows. Network changes are hypothetical.

How breakthroughs change the combinations of knowledge available to firms and the interfirm networks through which knowledge is used.

Coordination in AI collectives

Working paper

Spontaneous Coordination and Common Knowledge in AI Collectives

Tejas Ramdas and Michael W. Macy

Graph coloring and protocol switching under finite acknowledgement chains.
Graph coloring and protocol switching under finite acknowledgement chains.

When can independently acting AI agents coordinate their choices, and how do their information and expectations affect collective outcomes?

Summary

We study coordination among large language model agents in two experiments: distributed graph coloring and joint protocol switching after finite chains of acknowledgements. The experiments vary models, network conditions, and what agents know about one another's information. Collective outcomes vary substantially: repeated adjustments can leave conflicts unresolved, and additional acknowledgements do not consistently improve joint action. The findings motivate direct evaluation of coordination alongside individual model performance.

Earlier collaborative work

Revise and resubmit at Academy of Management Journal

Substantive and Subjective: A Multi-Method Investigation of B Corp Certification on Customer Ratings

K. Qiao, Wesley D. Sine, Tejas Ramdas, and M. Ross

Statistics research

Statistical methods, explainable AI, finance, and constitutional text.

Statistics, finance & law

Accepted at ICON

The Genesis of Constitutions: A Natural Language Processing Approach

Tejas Ramdas, Patrick Huang, Nuno Garoupa, Martin T. Wells, Yun-chien Chang, and Tom Ginsburg

Decomposing constitutional language into earlier reference contributions and novel vocabulary. Conceptual schematic.
Decomposing constitutional language into earlier reference contributions and novel vocabulary. Conceptual schematic.

A method for estimating how constitutional language combines contributions associated with earlier texts and vocabulary absent from those references.

Summary

Using 569 constitutions adopted between 1900 and 2020, the study models each constitution's word-frequency profile as a combination of earlier reference constitutions and a novel component. Chronological restrictions and separate analyses of rights provisions allow continuity and departures from earlier constitutional language to be examined together.

Revise and resubmit

Algorithmic Selection of Iconic Constitutions

Tejas Ramdas, C. Huang, Nuno Garoupa, Martin T. Wells, Yun-chien Chang, and Tom Ginsburg

Genetic search for reference constitutions, balancing originality and coverage. Illustrative, not estimated results.
Genetic search for reference constitutions, balancing originality and coverage. Illustrative, not estimated results.

Selecting reference constitutions by balancing originality with coverage of language in subsequent constitutions.

Summary

The procedure selects sets of four reference texts from 180 constitutions enacted through 1899 and evaluates their coverage of 581 later constitutions. It identifies alternative balances between originality and coverage, making the choice of reference texts explicit and repeatable.

Working paper

Bellwether Trades: Characteristics of Trades Influential in Predicting Future Price Movements in Markets

Tejas Ramdas and Martin T. Wells

From trade sequences to forecasts and trade-level influence. Conceptual schematic.
From trade sequences to forecasts and trade-level influence. Conceptual schematic.

Explainable AI identifies which trades influence a model's price forecast and how that influence depends on trading context.

Summary

I measure a neural network forecast's local sensitivity to individual trades, then examine how influence varies with trade size, venue, timing, asset type, and surrounding transactions. The method connects the model's predictions to economic heterogeneity in the information supplied by trades.