A dynamic log-linear regression model to forecast numbers of future filings at the European Patent Office

• We model patent filings from R&D expenditures and GDP split into trends and cycles.
• 38 source countries are used and standardisation of data via logarithms is discussed.
• Differenced data and business cycles improve the goodness of fit and the forecasts.
• Adding future GDP shocks, within certain margins, has limited effects on forecasts.
• Models fitted only on 7-year time windows fit well and confirm a 5-year lag for R&D.

An econometric model is applied to forecast future levels of patent filings at the European Patent Office out to 2019, using historical data from 1990 to 2013 with 28 source country terms. Descriptors include Research and Development expenditures and Gross domestic product, where the latter is split into trend and business cycles components. The model is applied to logarithmically standardised data.The effects on the forecasts of additional future positive and negative stimuli to the GDP components are considered. Reasonable forecasting accuracy is found. Using a series of shorter historical data windows may give improved accuracy for short term forecasts.