You are making a large misunderstanding. The confounders are not about what the day traders themselves are using in their own models. The confounders are things that affect researchers studying whole cohorts of traders.
Day traders may well believe they can predict political events in Brazil effectively. And yet, there may be some quality about Brazilian political events that impacts the cohort of Brazilian day traders systematically differently than say Korean political events affect Korean day traders.
For example, maybe a new cabinet member was just appointed in Brazil who likes to leak new political announcements early to an old business partner with connections at several large banks and trading shops, and so regular day traders are at a significant disadvantage in just this one market.
Or maybe during the time period of the study, there were political protests that disproportionately involved young people, meaning the average age (and perhaps skill level) of the observed cohort of traders was artificially too high just in the window of time of the study. Or any number of possible things like this.
You can’t assume these effects don’t exist when looking at different cohorts that have real reasons why they might not be probabilistically equivalent to a more general population. You have to actually account for the possible sources of confounding (in this case the country of the exchange), either by proposing some prior counterfactual model, or by collecting data across different cohorts and explicitly controlling for the confounder with whatever type of model you are fitting.
To be clear: none of this requires you to know in advance what the confounders actually are, and indeed in most causal inference models you literally cannot know what they are. The best you can hope is that you measures things like, e.g. country of exchange or age of the traders, that correlates with the hidden confounders well enough that they allow you to control for the confounding effects in your model.
It is not a thing in professional statistics to expect researchers to already know the sources of confounding before postulating that there could be sources of confounding that need to be controlled prior to believing the results are generalizeable.