Effect of climate and geography on worldwide fine resolution economic activity

which do not allow a robust statistical characterization across geographies and meteorological conditions. Role of country fixed effects While the focus of this work is on the role of exogenous geography and climate variables to explain GCP-PC, we want to try to distinguish whether the results obtained here are the effect of within-country variations or a world-wide (cross-country) variations. Such an assessment would assist with the assessment of the (relative) role of institutions on economic productivity. A standard approach to testing the within- versus cross- country effects is the use of dummy variables, whereby each country in turn is given the value of one, while all the others are set to zero. These variables have the purpose of fixing the within country effects, even if set- ting a single (unit) value may not be a good approximation for some countries given their het- erogenous institutions. This notwithstanding, together with the fact that we would need nearly two hundred dummy variables (corresponding to the number of countries considered here) which would make the interpretation of results extremely complex, the RF model does not lend itself to mixing categorical (dummy-variables) and continuous variables (geography and climate variables). An alternative approach is to make use of one of the six institution-related (continuous) indicator listed in the Introduction [18]. As mentioned, these are only available as country averages, and therefore are not directly comparable to the climate and geography fine resolu-

tion data. Also, while they may represent within-country fixed effects, they do not distinguish between countries with similar values of the chosen institutional indicator. Even more criti- cally, however, any such indicator is both endogenous and highly correlated with GCP-PC. As in [4], we consider here Rule of Law as the institutional indicator, by setting the same value for each grid point within its assigned country. Rule of Law has a correlation coefficient of 0.65 with GCP-PC. When Rule of Law is included as an additional predictor in our RF model, and ranked together with climate and geography variables, it yields (for the global case) an nMAE three time smaller (hence better) than for latitude alone (the top geography/climate predictor), and a correlation coefficient of 0.94 (compared to 0.62 for latitude alone). Stated otherwise, Rule of Law alone has the same explanatory power as the combined first 4–5 geography and climate predictors. This means that any additional geography/climate predictor, when added to Rule of Law, improves the fit with GCP-PC only marginally. While such a test is useful in proving that an institutional indicator such as Rule of Law can explain GCP-PC to a high degree of precision within our modelling framework, this does not provide a definitive explanation of country level fixed effects–given the inability of Rule of Law to discriminate between countries with similar values, such as Germany and United Kingdom. More fundamentally, however, Rule of Law (as any other institutional indicator), being endog-

enous, is inextricably connected with GDP and it is therefore arduous to relate it to the exoge- nous climate and geography variables considered in this work. Conclusions and discussion This work has investigated the role that exogenous factors represented by a wide range of cli- mate and geography variables have on worldwide economic activity as measured by the per capita Gross Cell Product (GCP-PC) at a fine, 1˚ by 1˚, geographical resolution. We considered two main cases: all global grid points and upper tercile of GCP-PC. We find that eight out of the top ten predictors are in common in these two cases. However, an interesting distinction is that latitude is the top predictor in the first case, but it is less important in the second case (it ranks third). We have also seen that month-to-month variations of meteorological variables, PLOS ONE Effect of geography on fine resolution economy PLOS ONE | https://doi.org/10.1371/journal.pone.0229243 March 2, 2020 16 / 20