The R&D Productivity Puzzle: Innovation Networks with Heterogeneous Firms

We introduce heterogeneous R&D productivities into an endogenous R&D network formation model, generalizing the framework of Goyal and Moraga-González (2001). Heterogeneous productivities endogenously create asymmetric gains from collaboration: less productive firms benefit disproportionately from links, while more productive firms exert greater R&D effort and incur higher costs. When productivity gaps are sufficiently large, more productive firms experience lower profits from collaborating with less productive partners. As a result, the complete network – stable under homogeneity – becomes unstable, and the positive assortative (PA) network, in which firms cluster by R&D productivity, emerges as pairwise stable. Using simulations, we show that the clustered structure delivers higher welfare than the complete network; nevertheless, welfare under this formation follows an inverted U-shape as the fraction of high-productivity firms increases, reflecting crowding-out effects at high fractions. Altogether, we uncover an R&D productivity puzzle: economies with higher average R&D productivity may exhibit lower welfare through (i) the formation of alternative stable networks, or (ii) a crowding-out effect of high-productivity firms. Our findings show that productivity gaps shape the organization of innovation by altering equilibrium R&D alliances and effort. Productivity-enhancing policies must therefore account for these endogenous responses, as they may reverse intended welfare gains.

💡 Research Summary

The paper extends the classic endogenous R&D network formation model of Goyal and Moraga‑González (2001) by introducing firm‑specific R&D productivities. In the baseline model all firms face the same quadratic R&D cost function, implying symmetric returns to collaborative R&D. The authors replace this homogeneity with heterogeneous cost parameters ϕ_i, which they normalize into relative productivities θ_i∈(0,1] (θ_i=ϕ_min/ϕ_i). A higher θ_i means that a unit of R&D effort yields a larger reduction in marginal production cost.

The game has three stages. First, firms choose bilateral R&D alliances (costless links) and the equilibrium concept is pairwise stability (Jackson & Wolinsky, 1996). Second, given the network, each firm selects its R&D effort e_i, incurring cost ϕ·e_i² where ϕ is the cost coefficient of the most productive firm. The marginal cost of production becomes c_i= \bar c – θ_i e_i – Σ_{j∈N_i} θ_j e_j, i.e., a firm’s own effort and the efforts of its partners reduce its cost proportionally to their productivities. Third, firms compete à la Cournot in a linear‑demand market, choosing quantities q_i. The authors solve the third stage analytically, obtain best‑response functions, and then solve the second stage by backward induction, proving existence of a unique interior equilibrium with positive outputs and efforts.

A key comparative‑static result shows that when two firms of different productivities are symmetrically positioned in the network, the low‑productivity firm gains a larger profit increase from forming a link than the high‑productivity firm, even though the high‑productivity firm contributes more to cost reduction. Consequently, the benefit of linking is monotone in the partner’s productivity: the higher the partner’s θ, the larger the gain for the linking firm. This creates asymmetric incentives: low‑productivity firms always want to link to high‑productivity partners, but the reverse is not guaranteed.

Because of this asymmetry, the complete network—pairwise stable under homogeneity—can become unstable when the productivity gap exceeds a threshold. The authors define two critical values, θ̲ and θ̅. If θ>θ̲ the complete network remains stable; if θ<θ̅ the positive‑assortative (PA) network, in which firms only link to firms of the same type, becomes stable. For intermediate θ both structures can be pairwise stable, yielding multiplicity of equilibria.

Analytical tractability is limited for larger n, so the authors resort to simulations. They consider n∈{4,…,20} firms, two types (high productivity θ=1 and low productivity θ∈(0,1]), and vary the share of high‑productivity firms ρ. For each (n,ρ,θ) they enumerate possible adjacency matrices, compute equilibrium efforts and quantities, and test pairwise stability. The simulation results reveal three main patterns:

1. Network structure – When the productivity gap is small, the complete network dominates; as the gap widens, PA networks appear and eventually become the unique stable configuration.

2. Welfare comparison – PA networks generate higher total welfare (consumer surplus + producer surplus) than the complete network because they avoid inefficient links that impose high costs on high‑productivity firms.

3. Inverted‑U welfare with respect to ρ – Welfare rises with the fraction of high‑productivity firms at low ρ (more efficient firms improve the system), peaks at an intermediate ρ, and then falls when ρ becomes large. The decline is driven by a “crowding‑out” effect: many high‑productivity firms compete for the same market, reduce their R&D effort, and the marginal benefit of additional productivity diminishes.

These findings constitute the “R&D productivity puzzle”: economies with a higher average R&D productivity may experience lower welfare because (i) the equilibrium network shifts from a dense complete graph to a clustered PA graph, or (ii) an excess of high‑productivity firms crowds out each other’s R&D incentives.

Policy implications are profound. Standard productivity‑enhancing policies (subsidies, tax credits) that raise average R&D efficiency could unintentionally alter the stable network topology, leading to welfare losses rather than gains. Effective policy design must therefore account for endogenous network formation: either by encouraging balanced collaborations across productivity levels or by mitigating the crowding‑out effect through measures that preserve R&D effort incentives for high‑productivity firms.

The paper concludes by suggesting empirical avenues: cross‑country or industry‑level studies that jointly observe productivity dispersion, R&D alliance patterns, and welfare outcomes; and dynamic extensions that trace how networks evolve over time in response to policy shocks. Overall, the work offers a novel theoretical lens on how heterogeneity in innovation capabilities reshapes both the architecture of R&D alliances and the aggregate benefits of technological progress.