Prophet Based Forecasting of Daily Tiktok Affiliate Revenue Using Operational Variables: The Cutetastic Case

Abstract

This study evaluates the ability of the Prophet model to forecast daily TikTok affiliate revenue for the Cutetastic_ account and examines the contribution of operational variables to forecast accuracy. The dataset comprises 120 days of daily performance records from November 1, 2025, to February 28, 2026, including date, total revenue, number of video posts, video views, product clicks, and products sold. Three modeling scenarios were evaluated: univariate Prophet, Prophet with extra regressors, and Prophet with extra regressors combined with log transformation and parameter adjustment. The data were split chronologically into 106 training observations and 14 testing observations, and model performance was evaluated using MAE, RMSE, and MAPE. The results show that the univariate Prophet model produced a MAPE of 53.86%, whereas Prophet with extra regressors reduced the MAPE to 21.55%. The best performance was achieved by Prophet with extra regressors, log transformation, and parameter adjustment, with an MAE of 8,977,172.95, an RMSE of 11,549,223.21, and a MAPE of 19.31%. These findings indicate that historical revenue patterns alone are insufficient to capture the volatility of TikTok affiliate revenue. Operational variables improved forecast accuracy in this case study, particularly when treated as inputs for conditional forecasting or simulation-based forecasting. Therefore, the results from the regressor-based models should not be interpreted as assumption-free forecasts, but as short-term projections that depend on the availability or simulation of operational inputs.