Experimental Design
Our objective is to manipulate uncertainty about future CO2 prices (the second moment), while keeping average expectations (the first moment) constant across treatment groups.
We have two treatment conditions. Respondents in both conditions are asked to imagine a survey among 100 energy experts (such as analysts) in which the experts are asked about their expectation regarding future CO2 prices. The respondents in our survey are then asked about their estimate of the expert survey's outcome. Specifically, they are asked to indicate the narrowest price range in which the majority of experts falls with their price expectation. The two treatment conditions are different with respect to the survey response categories that we provide. In the first condition, we provide four categories with very narrow price ranges and an "other" category for respondents who believe that the expert majority does not fall in any of the four provided categories. In the second condition, we provide four very wide price ranges along with an "other" category. The mean price in the provided categories is the same across the two conditions: the provided price ranges always center around 90 EUR, which is the average expectation that experts had in an actual expert poll. The underlying idea is that the provided narrow price ranges are suggestive of a high level of agreement among the experts (i.e., low uncertainty), while the provided wide price ranges are suggestive of a low level of agreement among experts (i.e., high uncertainty).
Our main outcome variable of interest is a survey item in which respondents are asked how they would invest a hypothetical amount of 10,000 EUR. This survey item has four response categories that correspond to different ways to invest the amount: i) green investment class, ii) conventional investment class, iii) consumption/bank deposit, and iv) "other". We are primarily interested in the amount and share that respondents would invest into the green investment class. The first step of our analysis will be to compare the two treatment conditions with respect to the mean amount/share of money that is invested in the green investment class. We will also look at the extensive margin decision of whether any amount is invested into the green class or not. Among other methods, we will use simple OLS to compare groups.
We are also interested in heterogeneity, i.e., does the treatment affect different groups of respondents differently. In light of evidence that risk behavior is different across men and women, gender is one source of heterogeneity that we consider. We also consider financial literacy (proxied by education) as well as climate and political preferences, and risk aversion. These variables come from previous waves of the GIP.
After our treatment intervention (after, rather than before, to not prime people and to not generate anchors), we have three additional questions (aside from our main outcome variable of interest): i) belief about the likelihood that the future CO2 price will indeed be close to the average expectation of 90 EUR, ii) estimate if green investment products will have higher or lower returns than conventional investment products, iii) estimate if green investment products will bear higher or lower risk than conventional investment products. We will also consider heterogeneity of treatment effects with respect to these variables. In addition, we will use the first variable to assess if the treatment manipulation worked. This then motivates us to have a special focus on respondents who, depending on their treatment, indeed find it very likely or very unlikely that the price will be close to 90 EUR.
The survey wave of the German Internet Panel (GIP), in which we implement our survey experiment, was in the field in November 2022. We expect to obtain access to the data in January 2023 (as verified in email by GIP staff on Dec 16, 2022). We have 3785 survey respondents in this wave and the computer randomization assigns respondents evenly to the two treatment conditions.