Exploitation Versus Exploration in Technology Search Strategies: A Machine Learning Replication of Morandi Stagni et al. (2021) "A Bird in the Hand is Worth Two in the Bush"

This study replicates the instrumental variables (IV) analysis from Morandi Stagni et al. (2021), “A bird in the hand is worth two in the bush: Technology search strategies and competition due to import penetration,” which examines how import penetration affects firms’ technology search strategies, distinguishing between exploration (new knowledge) and exploitation (refining existing knowledge). We replicate this work to interrogate its assumptions, including the exogeneity of instruments (tariffs and exchange rates) affecting search only through import penetration, the linearity of effects, potential omitted variable bias from domestic confounders (e.g., firm size, R&D intensity, domestic competition), and sample biases toward larger firms due to missing data filters, which may limit generalizability in complex global trade domains. Employing the original IV method in R with publicly available datasets (NBER Patent Data, CRSP-Compustat, and Peter Schott’s trade data) for U.S. manufacturing firms from 1991 to 2006, we yield 5,076 firm-year observations from 319 firms, closely matching the original after sales adjustments. We extend the analysis using XGBoost machine learning to assess variable importance in a non-linear, multivariate context, excluding tautological variables (e.g., total_cites, patent_stock_lag). The IV replication successfully reproduces the original outcomes: increased import penetration, instrumented by tariffs and exchange rates, significantly reduces technological exploration (mean = 0.038) and increases exploitation (mean = 0.962) across 3,259 observations with non-zero citations. Surprisingly, the XGBoost extension reveals that import penetration adds minimal predictive power (not in the top 10), with firm fundamentals (e.g., xrd, sale, capx) and domestic competition metrics (total_sim, hhi) dominating rankings—China-specific penetration ranks modestly (#8–#9). This coherently links to the original’s assumptions by highlighting omitted confounders and non-linear interactions that overstate foreign shocks’ causal role, implying domestic and internal drivers primarily shape search strategies in global environments and calling for nuanced policies beyond trade protections.