Estimating technology performance improvement rates by mining patent data

•Patent data can be used to predict yearly technology improvement rates reliably.•Monte Carlo cross-validation backcasting chooses the most reliable predictor.•The centrality of cited patents in the citation network is the most reliable predictor.•Patent centrality accounts for 64% of the variance of technology improvement rates.•Managers and policy makers can use this method to support long-term planning. The future direction of technology development depends on the relative yearly rate of functional performance improvement of different technologies. We use patent data to identify accurate and reliable predictors of this rate for 30 technologies. We illustrate how patent-based predictors should be normalized to correct for possible confounding factors introduced by changing patenting dynamics. We test the accuracy and reliability of various predictors by means of a Monte Carlo cross-validation exercise. We find that a measure of the centrality of domains’ patented inventions in the overall US patent citation network is an accurate and highly reliable predictor of improvement rates.