1,000A more direct, but qualified, candidate is Buchheit, Dalton, Downen & Pippin (2012), DOI: https://doi.org/10.2308/bria-50151
165 accounting students forecast plant profits. Among participants receiving outcome feedback, average absolute error fell from 4.04 to 3.13 million with accuracy-linked pay, versus 4.91 to 4.74 million without pay. The planned comparison of improvements was p=.019 (Table 1B; means in Table 3A).
Important caveats: feedback was individually randomized, but incentives varied by session, so this is not a clean individually randomized pay treatment. The full three-way interaction was not significant (p=.182). I would cite it as evidence for a context-dependent benefit, especially learning with feedback, not a universal claim. Author-uploaded full text: https://www.researchgate.net/publication/256485156_Outcome_Feedback_Incentives_and_Performance_Evidence_from_a_Relatively_Complex_Forecasting_Task
For balance, Remus, O’Connor & Griggs (1998) found no evidence of an incentive effect in their time-series forecasting experiment: Publisher abstract
Its kind of adjacent to what you are looking at, but I found this article very interesting: https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2019.3294
There is of course also this: https://en.wikipedia.org/wiki/Skin_in_the_Game_(book) which might give you some good leads.