Intervention (Hidden)
We picked June 2025, because it coincides with the new U.S. president’s taking office and starting to implement their policies; it is close enough to the survey and election dates (August, September, October, November) such that survey respondents are not predicting long-run performances of indicators (in which they may not be very good at or which may be subject to higher uncertainty), but at the same time sufficiently later from the survey dates so that respondents interviewed months apart are unlikely to have an advantage of a significantly shorter prediction horizon.
The (likely) presidential candidates on the ballot are DJT of the Republican Party and Kamala Harris (KH) of the Democratic Party. Their policy promises seem to differ in many aspects. For example, Trump has been advocating for business-friendly policies, such as corporate tax cuts and protectionist strategies with respect to international trade. On the other hand, Harris’s economic policies are more tuned to target welfare of low-income families. Thus, (if implemented) Trump’s policies versus Harris’s policies could affect economic indicators differentially.
We anticipate that the probability of election victory of these candidates will fluctuate frequently, perhaps daily or hourly. In fact, data from the election betting market, PredictIt, confirms this volatility.
Under one hypothesis, as candidates’ probability of winning changes over time, individual investors’ perceptions of future economic indicator performance are impacted. For example, if the expectation of a DJT win increases, then an individual may expect that more business-friendly policies will be implemented. As a result, this individual may up her/his prediction of the S&P 500 index.
Suppose that there are two observationally identical individuals, A and B, who are surveyed a few days apart. Also, assume that a political scandal involving a candidate surfaced between their interview dates, causing a change in the win probabilities. In this case, if we could compare the economic indicator expectations of A to those of B, then we may be able to identify the causal impact of the policies associated with each candidate on investor perceptions.
This is precisely how we plan to execute our survey. We want to survey individuals (through an online platform Prolific) frequently (perhaps twice or three times a week) up until the election. The identification assumption is that the individuals who are surveyed only days apart are identical and have been exposed to similar information except for the events that led to a change in the candidates’ probability of winning. We will have data on several demographic and socio-economic characteristics of the respondents so that we can control for them in regressions (or test for whether they are balanced across the period of the survey). In addition, we will know whether important announcements or political events (that may impact the election probabilities or economic indicators independently) occurred on each day. This will enable us to account for their potentially confounding impact.
While it is possible that respondents consider the presidential elections and the next president’s policies, under an alternative conjecture, an individual investor may believe that a DJT win over KH will make no difference for the index value, perhaps because she/he doesn’t think that their policies differ. It is also possible that the respondent isn’t aware of the policies or the election probabilities of the candidates. To gauge these possibilities, we plan to carry out an information experiment in the same surveys. We plan three treatments: 1. Positive; 2. Neutral; and 3. Negative. In the Neutral treatment, individuals are provided with the policy position of Trump without any comments about it. In the Negative and Positive treatments, respondents are given the same information as the Neutral treatment, and some comments about the potential implications of the policies. For example, the Neutral treatment includes “If elected, Trump is likely to implement high tariffs on imports, especially those from China.” The Negative treatment will contain “If elected, Trump is likely to implement high tariffs on imports, especially those from China. Some economists argue that this will reduce the competitiveness of the US industry.” The Positive treatment will include “If elected, Trump is likely to implement high tariffs on imports, especially those from China. Some economists argue that this will improve domestic producers’ bottom line.” Alongside these pieces of information, subjects will be given the probability of a Trump victory (based on the election betting markets) and the change in that probability in the past few days.
Specifically, the layout of the experimental design is:
As soon as respondents agree to participate in the survey, they are asked about their expectations of economic indicators (i.e. priors).
One of the three treatments is randomly assigned.
Respondents are asked about their expectations of the same economic indicators(i.e. posteriors).
Respondents fill out a survey regarding their political attitudes, personal attributes, and how/whether they used the information we provided in step 2.
A test for whether the respondent considers the next president’s policies could involve comparing answers of the same individual before and after the information treatment (those in step 3 versus 1). If the individual does not change her/his predictions, this may be indicative of either a. she/he already had the knowledge of the presidential race, candidates’ policy positions, and election probability; or b. she/he does not think that these will not influence the market indicators. In contrast, if we find that individuals update their predictions, this could suggest that the respondents believe the next president’s policies will impact the economic markers. To investigate the issue further, the question in step 4 will explore an individual's perceptions regarding whether she/he thinks that the next president’s policies will affect the economy. This step will also sift through the respondent’s knowledge and interest in politics and political events.
Our setting is also useful for implementing a well-known behavioral benchmark (Bayesian belief updating) in our analyses.
According to the Bayesian belief updating context, the posterior (post-treatment response) can be written as a weighted average of the prior (the pre-treatment response) and the information in the signal (treatment). These weights are important in that they indicate whether the individual believes the information is credible and informative. The higher the weight of the signal is, the more likely the information is believable. We plan to estimate these weights in our analysis.
We additionally plan to investigate the potential heterogeneities in the treatment effects. Specifically, besides the heterogeneities with respect to demographic (race, sex, age, and alike) and socioeconomic (education, employment, income, and similar) characteristics, we plan to examine whether the responses to the treatment vary by the individual’s prior. There are reasons why the treatment effects on those with a high prior (e.g., in this context, people highly optimistic about the S&P 500) versus those with low (pessimists) or middling priors may differ. One possibility is that those with more extreme expectations (overly optimistic or pessimistic) have stronger convictions and may be less likely to find the information in the signal to be credible. As a result, they may respond less. Alternatively, we may see larger treatment effects on the individuals with more extreme priors because they are less informed, and the information in the signal is novel and surprising to them. This analysis can be carried out with a regression where the posterior is regressed on the prior, the treatment, and their interaction. In this setting, the comparison of the coefficient of the interaction term to that of the prior (which is expected to be one under Bayesian belief updating) will enable us to analyze the information quality of the signal.
Along the same lines, we will investigate heterogeneities pertaining to individuals’ political positions (obtained through the post-experiment survey). Given the political polarization in the country in recent years and the debate over fake information dissemination, one may expect that individuals from opposing views may disregard the information in the signal or find the information implausible or incorrect. For example, an individual who registers for far-right political movements could believe that the negative news (as in the Negative treatment) or the information on, say, low election probability for DJT is incorrect. As a result, she/he may not change behavior much. Our information treatment will include not only some information but also the actual (market-determined) election probabilities, which will allow us to investigate these heterogeneities further.
Other analyses:
Focus on priors and analyze their dynamics over the election cycle.
Focus on how trust in institutions changes over the election cycle.
Focus on belief uncertainty changes over the election cycle.
Focus on posteriors and belief updating and their determinants.
Consider conducting post election survey to measure beliefs again (pending due to funding)