Why mean sometimes means meaningless – stochastic IAMs can help to assess Climate Change impacts

Many applied economists are concerned with average effects of  a policy intervention or changed economic conditions on economic subjects. Often linear regression is their tool of choice for conducting impact analysis. However, focusing on the mean might mask some interesting effects along the distribution of the outcome variable.

Suppose you were interested in evaluating a labour market reform that possibly has an impact on the wage distribution. Then, calculating the average effects on wages could be useless – imagine the politically sensitive case when mean wages remain constant but increase in the upper quantile while decrease in the lower quantile. That is, focusing on the mean masks the fact that the reform resulted in higher inequality. In the case of applied economics, the use of quantile regression has therefore become quite popular to get insights into the effects that occur below or above the mean.

For the same reason we have to consider the effects along the whole wage distribution in applied labour economics, we have to look in the tails when considering the economic impacts of climate change: We just might miss the relevant stuff!

When assessing the economic impacts of climate change, Integrated Assessment Models (IAM) are widely used tools. Broadly speaking, these models combine knowledge from different fields into a single framework. In their most basic version they consist of a climate system module and an economic module. The former one translates projections of population and GDP growth into levels of emissions and estimates of global mean temperatures. The economic module then takes the projected temperatures as input. It applies a damage function (which in fact is another construction site in IAMs requiring attention!) and calculates economic costs from changes such as in agricultural productivity, sea level, cyclone frequency, etc.

Quite often these economic impacts are modelled in a way that assumes that economic systems evolve deterministically neglecting elements of randomness.

While climate models generally incorporate stochastic change to account for uncertainty in climate sensitivity, there is a weakness (besides the damage function) of IAMs in the connection between climate and economic module when it comes to assessing economic costs of climate change. Temperature enters the economic module as global means and damages are calculated as a function of it. However, if we want to account for stochastic changes in the climate and economic system, temperature should enter the economic module as a random variable and allow the modeller to consider the possible damages that occur at the end of the distribution.

Empirical estimates of weather impacts indicate that temperature variability is more important than the mean. This is especially significant for agriculture and food security. For example, a study by Schlenker and Roberts (2009) find that crop yields increase steadily with increasing temperatures but decrease sharply after the temperature reaches a certain threshold.

Damages after such thresholds can be irreversible and must be captured by IAMs. This can have relevant policy implication for setting carbon taxes. For example, stochastic models that incorporate tipping points imply higher carbon taxes than deterministic models would suggest (Lemoine and Traeger 2014, Lontzek et al. 2013).

Therefore, considering the whole distribution and moving away from the mean can matter a lot. In the same way that applied labour economists have therefore added quantile regressions to their toolbox, integrated assessment modellers should widely introduce stochastic change and tipping points in their models if they aim at quantifying more realistically economic damages from climate change.

By Maike Hohberg, PhD Student, UCL Institute for Sustainable Resources