Supplementary material for "The Impact of Climate Disasters on Climate Action"

Supplementary Material Following the IMF, we use data from 1980 to 2020. Figure A1 shows the average number of disasters each year for each country. Figure A1: Average number of climate related disasters 1980-2020 Note: Data from EMDAT database. Climate related disasters are defined as drought, extreme temperature, flood, landslide, storm and wildfire. Figure A2 presents the average number of mitigation laws and policies for each country. Table A1 presents the impact of mitigation laws on emissions. Both the contemporaneous and the first lag of mitigation laws are significant associated with a reduction in GHG emissions. One law reduces -10.04m tons annual GHG emissions in C02 equivalents contemporaneously, and -9.46m tons in after 1 year. Figures A3 and A4 show the frequency of climate related disasters (solid line) and the cutoff of four standard deviations above the average (dashed line) in each country. Figure A3 presents the treatment group where the number of disasters is more than the four-standard deviation threshold in at least one year, while figure A4 presents the control group where disasters do not pass the cut-off point. As a robustness test we present results using cutoffs higher than 4σfrom the mean. Table A3 shows the shocks when 5σcutoff is used. Only Finland and Luxembourg in 1990 are identified. When 6σ 13 Figure A2: Average number of mitigation laws and policies 1980-2020 Note: Data from Climate Laws of the World Database. Table A1: The impact of mitigation laws on GHG emissions (1) Annual greenh b/p No. Mitigation laws -1.036e+07** (0.047) No. Mitigation law 1 -9460582.247* (0.060) No. Mitigation law 2 -9915163.278 (0.153) N 624 P-value in parentheses | * 0.10 **0.05 ***0.01 | Country and Year fixed effects are included 14 Figure A3: Frequency of climate disasters for the Treatment Group (4σcutoff) 15 Figure A4: Frequency of climate disasters for the Control Group (4σcutoff) 16 cutoffs are used, no shocks are identified. Figure 2 presents the change in the cumulative number of mitigation laws between the treated and control groups. As discussed above, the climate disaster shock occurs in 1990 which is indicated by the dark vertical line. The parallel trends assumption holds for the pre-treatment period, as both groups have the same flat trends. In 1990, the control group starts to implement mitigation laws, while the treated group does not implement laws until 1997. The increase in mitigation laws for both the treatment and the control group in 1997 is driven by all countries signing the 1997 Kyoto Accords. The parallel trends assumption is that both groups would have followed the same trajectory had the treated group not suffered from the disaster shock in 1990. Figure A5 also shows the trends of the number of climate disasters minus the mean for the treatment and control groups. The number of disasters (minus their means) are similar in the pre and post treatment group, apart from the 1990 shock. Figure A5: Parallel trends: Number of climate disasters minus country average for treatment and control group Note: Data from EM-DAT. The graph has been constructed by first finding the number of disasters in each year minus the average over the 1990 to 2020 period. These are then averaged in each year across the treatment and control group. Column 1 of table A4 presents the coefficient and significance level of the treatment effect (i.e. β2in 17 equation 1), without controlling for potential confounding factors. The treated group had -0.303 fewer cumulative mitigation laws implemented in the post treatment period (1990 to 2000) than the control group. The average number of cumulative mitigation laws in this period across both groups is 0.295, and therefore an average reduction of around -0.303 is larger than the mean. Column 2, in table A4 includes control variables. The effect of disasters on the number of mitigation laws remains significant with the same sign, although the magnitude of the effect on the number of mitigation laws is relatively smaller (-0.297) and only significant at the 10% level. Lastly, column 3 changes the dependent variable to include the cumulative frequency of laws and policies, in order to capture a potentially broader definition of mitigation action beyond just laws. As can be seen, the shock negatively impacts the number of mitigation laws and policies (by -0.606) and this remains significant at the 10% level. The average number of mitigation laws and policies implemented each year for all countries in the post treatment period was 0.705, and so the effect is slightly smaller than the average. To further check the robustness of our results we perform seven robustness tests, where we change the time frame; change the identification strategy considering as a cutoff point 5σand 3σfrom the country mean respectively; use two way fixed effects (Poisson and zero-inflated model); and estimate a placebo test by varying the start year of the treatment from 1990 to 1992, 1993 etc. The first robustness test changes the time frame to include a longer period going from 1980 to 2010, rather than our initial shorter period from 1980 to 20001. As can be seen the coefficient on the interaction term remains significant at the 5% significant level. The second robustness test changes the identification strategy. It identifies the extreme weather shock as years where there are more than 5σabove the country mean. Table A3 lists these shocks by country and year. Using the 5σcutoff decreases the number of countries that experience a shock to just Finland and Luxembourg. Figure A6 in the presents the plots of the treated and control group. Column 2 of table A6 presents the results. As can be seen, the coefficient is -0.338, almost the same as when the 4σcutoff is used, and is significant at the 1% significant level. The third robustness test changes the identification strategy so that the cutoff is 3σabove the mean. Table A5 lists these shocks by country and year. Sweden is dropped from the sample as it experiences two shocks over the time period. All the other shocks also occur in 1990, however when defined in this way, the coefficient becomes insignificant. We interpret this results as suggesting that the shocks need to be sufficiently strong enough in order to have the impact on mitigation laws. The fourth and fifth robustness test estimates a two-way fixed effects regression equation, given its use in the empirical literature. Given that we are now estimating the impact of multiple shocks over 1The significance of the interaction term when extending the horizon beyond 2010 may not necessarily imply that the impact of the 1990 shock continues to grow or re-emerge over time. Rather, it could reflect the persistence of the initial post-shock divergence between treated and control countries, which remains visible in cumulative terms even after two decades. Alternatively, the result might simply stem from the increased statistical power associated with a longer panel. 18 Figure A6: Impact of climate disaster shock on cumulative no. mitigation laws with 5σ cutoff Note: Shock is identified as years where the number of climate damages 5σabove the country mean. Full list of shocks is in table A3. 19 many years, we change the dependent variable to the frequency (rather than the cumulative frequency) of laws. This is different to the diff in diff, where there is one shock and we want to understand its total cumulative effect. Given that the dependent variable is a count variable with many potential zeros, we two types of fixed effects regression, a Poisson fixed effects regression (column 4) and a zero-inflated negative binomial regression (column 5). We include 5 lags given the potential in the baseline for the shock to have an impact on laws into the future. Columns 4 and 5 present the results. The fifth lag is significant and negative at 5% level for both estimations. All other lags are insignificant. Increasing the number of climate disasters a country experiences by 1, leads to a decline in the number of climate laws passed by -0.238/-0.266 after 5 years. The number of observations increases for this specification as we estimate this now over the whole period from 1980 to 2017, as we include all disasters and not just extreme shocks. While these effects are more muted, there is still some evidence that any shock can lead to a decline in the number of mitigation laws. This is to be expected given that not all these disasters will act as focusing events, particularly compared to the 1990 shock, and therefore the results are expected to be less significant. In column 6 we keep the baseline identification strategy but we change the dependent variable to be the frequency of mitigation laws, rather than cumulative frequency. As can be seen, the coefficient on the interaction term remains significant and negative at the 5% level. The next table presents a summary of the six first robustness tests. The final set of robustness checks is presented in Table A7, which estimates a placebo test by varying the start year of the treatment from 1990. Specifically, column 2 examines the effect of a placebo shock beginning in 1992 on the cumulative number of mitigation laws, column 3 considers the impact if the shock commenced in 1993, and so forth. The primary objective of this placebo test is to assess whether the observed impact of the 1990 shock genuinely drives the results. If the placebo tests (columns 2 to 7) yield statistically significant coefficients, it would imply that the 1990 shock may not be responsible for the observed differences between the treated and control groups. The dependent variable used in these tests is the cumulative number of laws and policies The coefficients become statistically insignificant after 1993, with the pattern of insignificance persisting through 1996. While the results suggest that 1990 is not the sole year exhibiting significance, they also indicate that the further the hypothetical shock deviates from 1990, the less significant the coefficients become. This finding underscores that the 1990 shock likely captures a genuine effect. 20 Table A8: Periods where frequency of disasters is more than 4σabove country mean Country Date No. Disaster Mean No. Disaster σNo. Disaster Denmark 1990 3 .372093 .6554989 Finland 1990 2 .0697674 .337734 Luxembourg 1990 6 .3023256 .9644856 Netherlands 1990 6 .8604651 1.125069 Table A9: Summary statistics Variable Obs Mean σMin Max Cumulative freq mitigation laws 777 1.796654 2.683443 0 15 Freq mitigation laws 777 .1441441 .4275558 0 4 Freq mitigation laws and policies 777 .5160875 1.185201 0 11 Frequency of disasters 736 1.164402 1.577524 0 10 Democracy index 777 .8656023 .0442886 .594 .924 Total rents/GDP 777 .5223525 1.665277 0 12.24815 Population 777 20741.84 24868.64 316.645 83160.87 Left-Right Index 721 -55.0749 231.9422 -999 3 Damages from disasters (%GDP) 777 .0518993 .196719 0 2.895422 21