Analysis of R&D Investment Efficiency in Technology-Intensive Industries: Evidence from Malaysian Listed Companies (2010–2023)
Main article
Abstract
The capacity to transform research and development (R&D) expenditure into tangible innovation outputs represents a fundamental determinant of competitive positioning in knowledge-intensive economies. Existing efficiency assessment methodologies, however, are often constrained by restrictive parametric assumptions and limited ability to model the non-linear, heterogeneous pathways through which R&D inputs generate innovation outcomes. This study applies a comprehensive machine learning framework—comprising Ridge Regression, Lasso Regression, Random Forest, Gradient Boosting, XGBoost, LightGBM, and Support Vector Regression—to evaluate R&D investment efficiency across eight technology-intensive sectors of the Malaysian economy using a panel of 12,418 firm-year observations drawn from 1,124 listed companies on Bursa Malaysia during the period 2010–2023. A rigorous validation architecture employing 70:30 stratified train–test partitions, 10-fold cross-validation, RobustScaler preprocessing, and conservative hyperparameter search spaces is adopted to safeguard against overfitting. Statistical comparisons across sectors are conducted using paired t-tests and Mann–Whitney U tests with Bonferroni correction. The Gradient Boosting ensemble delivers the highest predictive performance (test R² = 0.961, RMSE = 0.0259), with the difference between training and test performance below 0.012—indicating negligible overfitting. R&D intensity emerges as the dominant predictor across all ensemble models (feature importance = 0.328), with patent count (0.217), firm age (0.108), and firm size (0.094) forming secondary influences. Pharmaceutical and biotechnology firms exhibit the highest average efficiency (0.312), while automotive parts manufacturers record the lowest (0.167); this range (0.145) is substantially wider than patterns reported in comparable Chinese listed-firm studies, indicating meaningful inter-sectoral heterogeneity within Malaysia's technology economy. Temporal analysis reveals a steady efficiency improvement averaging 0.55 percentage points per annum between 2010 and 2023, coinciding with the National Policy on Science, Technology and Innovation (NPSTI) implementation phases. The findings indicate that ensemble machine learning, when buttressed by disciplined validation protocols, offers a methodologically robust lens for innovation efficiency assessment in emerging Southeast Asian economies, and that sector-targeted innovation policies—rather than broad-based incentives—are more likely to yield productivity gains in the Malaysian technology ecosystem.
