Corporate research spending depends on who copies whom in corporate networks
A statistical model of corporate alliances reveals that American companies divide into distinct peer-driven and self-reliant groups when deciding how much money to channel into innovation.
When a corporation decides how much capital to pour into developing new technologies, its executive board rarely makes that choice in total isolation. Companies monitor what their peers and research partners spend, adjusting their own budgets to protect market share or capitalise on shared technical breakthroughs. Yet measuring that mutual influence has frustrated economists for decades. Organizations do not form alliances at random; they selectively pair with partners whose existing capabilities and unobserved ambitions make collaboration attractive. When researchers observe two partners spending heavily on laboratory work, teasing apart whether one influenced the other or whether both were drawn together by an unmeasured drive to innovate has proved notoriously difficult.
Untangling that knot matters directly for public policy. Governments deploy billions of dollars each year in research subsidies and corporate tax credits under the assumption that public incentives will spill across corporate networks. If policymakers misjudge how firms choose partners or assume every business responds to peer pressure in the same manner, public spending risks flowing to companies that absorb ideas without passing them on.
Why is peer influence so difficult to measure in business networks?
Corporate decisions follow a chain of strategic choices that entangles partner selection with internal budgeting. First, a firm weighs whether to establish a joint development pact, relying on both public financial statements and private assessments of common technical ground. Once an alliance takes shape, engineers and managers trade knowledge across organizational borders. Next, corporate treasuries observe their collaborators' commitments and recalibrate their internal investments, balancing the pressure to keep pace against their own internal hurdle rates and borrowing costs. Because unobserved organizational traits shape both the original partnership and the eventual spending decision, standard statistical regressions misread correlation as peer influence.
A working paper posted to the arXiv preprint server by economists Duong Trinh of the University of Graz and Santiago Montoya-Blandón of the University of Glasgow offers a unified way to separate those intertwined forces.12 Posted in June 2026 and the (which may not have been peer reviewed) study proposes an econometric framework named the Selection corrected Heterogeneous Spatial Autoregressive model.2 The method evaluates partner selection and corporate spending simultaneously, allowing peer sensitivity to vary across unobserved types rather than assuming an identical response across the entire economy.12
How do corporate reactions diverge across research networks?
Companies split into two sharply contrasting behavioural groups when confronted with peer spending and tax changes. Applying their model to an innovation network of 1,150 American firms, Trinh and Montoya-Blandón found that businesses do not share a single, uniform sensitivity to their partners.2 Instead, the data uncovered two latent categories operating within the identical alliance network: a peer-driven cohort and a self-driven cohort.
The peer-driven segment accounts for approximately 34 percent of the sample. These firms exhibit a network spillover estimate of around 0.215, meaning their research budgets respond aggressively to the outlays of their collaborators. They also show a weaker own-price elasticity of research spending of near -2.2, indicating that direct tax costs matter less to them than keeping step with alliance partners. The remaining 66 percent of firms belong to a self-driven type. For this majority, the network effect drops to around 0.127, while their own-price elasticity sharpens to near -9.5, demonstrating intense sensitivity to internal tax prices rather than external peer activity.
What does the model reveal about targeted public support?
Targeting research incentives toward central corporate transmitters generates substantially greater innovation spillovers across the wider economy than scattering subsidies broadly. Trinh and Montoya-Blandón demonstrated that firms with the highest capacity to absorb peer innovation are rarely the companies that excel at broadcasting it outwards.1 Organizations that act as network amplifiers tend to occupy highly central positions in high-technology clusters.1 When public tax credits target these central transmitters, the resulting expenditure propagates across corporate boundaries, while subsidising self-driven absorbers leaves neighbouring budgets largely unchanged.
To arrive at these findings, the authors built a two-stage Bayesian data augmentation system estimated via Markov chain Monte Carlo simulation.12 Rather than treating the corporate network as fixed, the system's first stage uses a probit link function to model the probability of an alliance between any two firms, accounting for unobserved factors that influence both connection likelihood and corporate outcomes. The second stage assigns firms to latent groups through a finite mixture structure, avoiding arbitrary rules based on industry tags or company size.12
The mathematical framework comes with defined boundaries. The results represent statistical projections derived from a cross-sectional sample rather than a live empirical tracking of evolving balance sheets over decades. The model assumes that unobserved drivers follow specific distributional structures and normal errors, which may overlook sudden macroeconomic shifts. Because the paper is an unrefereed preprint, the estimates have not yet undergone formal peer scrutiny.
For economic policymakers, the immediate challenge lies in turning these latent classifications into practical screening criteria. While econometric algorithms can distinguish between peer-driven transmitters and cost-sensitive absorbers in historical datasets, tax authorities must determine which observable corporate markers reliably signal that a firm will propagate its state-subsidised innovations to its peers.
The record
8 sentences trace to 2 sources.
-
1
A
arxiv.org Paper · arxiv.org Heterogeneous Peer Effects with Endogenous Network Formation Preprint · may not have been peer reviewed See the source
- 2 Heterogeneous Peer Effects with Endogenous Network Formation Not published online