Impact of technological change on growth trajectory of India: a multivariate-BVAR analysis

ABSTRACTThis paper examines the impact of technological change on Indian economic growth using the Bayesian Vector Auto-Regressive (BVAR) methodology. We use a comprehensive annual time series dataset covering the period of 1980 to 2019 on real economic activity, gross fixed capital formation, and employment. Technological innovation is measured by the number of patents filed by resident Indians. Technological innovation positively impacts both economic growth and gross fixed capital formation. Our findings indicate that increasing the number of patents leads to higher investment, which drives India's economic growth. However, our results also point towards the possible negative influence of technological innovation on the aggregate employment scenario in India. Our main findings are robust to alternative identification strategies and variable transformation. The asymmetric analysis also corroborates the positive influence of patents on driving investment and economic growth in India.KEYWORDS: Economic growthtechnological changepatentsBayesian VARJEL CLASSIFICATION: E44E31 AcknowledgmentWe thank the Managing Editor, Prof. Cristiano Antonelli, for providing invaluable suggestions in helping us improve the manuscript. We also thank three anonymous reviewers of our manuscript for their detailed comments and suggestions. We also thank Dr. Reshmi Sengupta, Dr. Nilanjan Banik, and Dr. Arnab Chakrabarti for their suggestions and comments in improving certain sections of the paper.Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 Antonelli and Scellato provide comprehensive discussions on the idea of "Innovation."2 Although the data is available until 2020, we restrict our sample to 2019 and exclude 2020 to avoid the problem that can arise due to the COVID-19 pandemic.3 Normal-Wishart prior constraints λ2 to the value of 1. We have estimated the baseline model with different set values for the hyperparameters. The finding suggests that IRFs are not qualitatively different due to changes in the values of the hyperparameters. However, for some choices, the confidence intervals become wider or narrower, and these are only indicative of the shape of the posterior distribution and have no statistical significance. Therefore, we can conclude that the findings from these specifications are robust to choices of different hyperparameter values for the priors.4 We have considered slightly lower values of the AR(1) parameter, such as 0.9 and 0.8, but it has a negligible impact on the results.5 The DIC value for our baseline model is -969.84.6 BVAR estimation is conducted by employing the BEAR toolbox in MATLAB developed by Dieppe et al. (Citation2016)7 Sims and Zha (Citation1999) assert that the traditional frequentist error bands may be misleading as they mix parameter location information with model fit information. The authors propose using likelihood-based bands and argue that 68% interval bands provide a more precise estimate of the true coverage probability.