Technology Flows Matrix Estimation Revisited

This paper revisits the methodological problems of estimating matrices showing how technological advances--measured by industry research and development outlays--flow from industries of origin to using industries. An early effort relied upon the analysis of 15 112 US patents. Several alternative methods are explored to address methodological questions concerning the choice of carrier matrices, the handling of diagonal elements, and the treatment of capital goods flows. Technology flow matrices estimated using diverse combinations of assumptions are tested for goodness-of-fit relative to the original patent-based matrix and for their ability to "predict' productivity growth in Solowian regression equations. Although some anomalies emerge, the best results are obtained using combined first-order transactions and capital flows matrices with diagonal elements adjusted to reflect the ratio of internal process to all R&D spending. However, flow data compiled using the Leontief inverse matrix add explanatory power in productivity growth regressions.