"World economy set to lose up to 18% GDP from climate change if no action taken, reveals Swiss Re Institute's stress-test analysis "New Climate Economics Index stress-tests how climate change will impact 48 countries, representing 90% of world economy, and ranks their overall climate resilience "Expected global GDP impact by 2050 under different scenarios compared to a world without climate change: "-18% if no mitigating actions are taken (3.2°C increase); "-14% if some mitigating actions are taken (2.6°C increase); "-11% if further mitigating actions are taken (2°C increase); "-4% if Paris Agreement targets are met (below 2°C increase)"
This is a tail risk study, not a central tendency study. The damage numbers in this study are way higher than typical in the academic literature. The methodology of the study is to multiply the damage forecasts from a normal model by 5x (one case) and 10x (another case). The argument is that the normal models only account for a subset of damages, and the 5x or 10x multipliers are to account for "unknown unknowns." The headline numbers quoted above are taken from the 10x multiplier case.
It's hard to say whether using a multiplier of 5x or 10x is a reasonable. My view is that applying a multiplier of this size is unlikely to tell you anything very useful because it indicates that you have a poor understanding of what is happening in the system. If you think that your model is missing 80% or 90% of an important effect (in this case, the effect being the damage to the economy from climate change), then the best thing to do is to focus on improving the representation of the effect in your model, rather than to try to scale up the accounted-for effects by a large factor. The climate semi-skeptic Bjorn Lomborg wrote a paper a little while back using a multiplier of about 1.2x (0.73% increase in damages on top of 3.6% in the base model) to account for damages not included in the model, and not surprisingly found that things don't look so grim. To me, a multiplier of 1.2x says "we know we probably missed something, so we'll be a little cautious," while a multiplier of 5x or 10x says "we really have no idea what's going to happen."
A less important critique of the piece is that the RCP8.5 scenario is probably not realistic as a 'worst-case' scenario. It is unlikely that the world will end up being in such a bad state. But the inclusion of this scenario makes sense if you understand this as a tail risk study.
Maybe a study of this type does makes sense for a reinsurance company like Swiss Re, which I imagine might be more interested in tail risk than first-line insurance companies or ordinary consumers. But this piece is framed on the Swiss Re website, and was written up in the Times, as more of a central tendency forecast rather than as a stress test or extreme scenario.
A final thought on the Swiss Re piece is that the parts that I know a bit about seem to be cherry picked, so that makes me skeptical about the stuff in there that I don't know so well. For example, the piece discusses the Stern review (published 2006) in Appendix 2 and on page 6 as another study that shows high economic impact from climate change. But the Stern review didn't really take issue with damage functions in use in the literature. Rather, the Stern used a low discount rate so damages in 2200 got pulled back to the present. That is not what the Swiss Re study is saying, which is that climate change will cause huge economic damage as soon as 2050.