Information Games: Strategic Crowding and Firm Repositioning in Language-Model Space

Information Games. Strategic Crowding and Firm Repositioning in Language-Model Space. By Marcus Gawronsky and one other, from the University of Cape Town. Published on the fifth of November.

When a firm responds to a new market opportunity, its choice of capacity changes the future discounted free cash flows available not only to itself but to its competitors and to investors. In competitive markets those investment decisions are publicized, in news articles and in analyst reports, and that coverage registers changes in capital structure, product mix, firm leadership, risk exposure and operating procedure. A written record of how firms move already exists. What it lacks is a way to be read as positions relative to one another.

The bridge comes from search. Modern search engines rely on large language models to measure how relevant a source document is to a given query. Those same measurements let the authors represent a firm's news coverage as a cloud of dense vectors, and the information geometry of those clouds reflects how dissimilar firms are from each other over time. Crowding together and moving apart become distances in a language-model space.

On top of that representation sits an economic model. Building on Xu and colleagues, the authors develop a model of costly competitive adjustment, in which firms adjust their capacity in response to free cash flow shocks. To analyse the sign and the magnitude of the adjustment this model predicts, they introduce a novel sandwich estimator and apply it through multi-agent simulation.

The findings from simulation are then validated against real coverage: a sample of news and analyst reports from Nasdaq-listed companies, spanning two thousand eighteen to two thousand twenty-two. The abstract does not quote the estimates themselves, so what can be described is the structure of the test. The simulation predicts a direction and a size of adjustment, and the document geometry of listed firms is where that prediction is checked.

Underneath the economics is a change of instrument. A search engine's sense of which documents are relevant to which becomes a ruler for competitive position. Pair that ruler with a model of what adjustment costs, and the coverage of listed firms turns into evidence about how they reposition when their cash flows are shocked.