et al. (2007). Winsorization is argued to produce more robust results because outliers are kept in the sample, only with lesser magnitude. The regression caps the top and bottom 2.5% of the outliers at the respective 97.5th and 2.5th percentile, decreasing the influence of spurious outliers. Next, in column (2) we ignore industries at country-level with fewer than 5 consecutive growth observations. In column (3) we show how the results are affected if we put the threshold for dropping outliers to 1% instead of 2.5%. This increases the variation in the sample and we expect decreased significance of our results. In column (4) we drop industry–year fixed effects. Finally, in column (5), we add the following controls: output of the financial sector as a fraction of total output, exports and the extent to which the economy is open.