The predictability of any epidemic is highly uncertain, especially regarding a new emerging pathogen such as SARS-CoV-2. We studied the predictability of the incidence series of COVID-19 in Chile (whole country) and three regions with different population sizes. The analysis included the period intervened by vaccination campaigns and when new variants of the SARS-CoV-2 virus arrived. This study also is focused on possible differences in predictability between epidemic and "inter-epidemic" periods. We studied the daily incidence of COVID-19 in Chile for the Metropolitan, Biobío, Arica, and Parinacota regions from March 2020 to February 2022, with the recently proposed method of the third moment of incidence. We assessed the predictive capacity with the corrected mean arctangent absolute percent error. The predictability of the daily incidence of COVID-19 was on the limit between good and reasonable for the entire epidemic process. The third moment of incidence produced reasonable predictions for regions with large population sizes and insufficient predictions for smaller regions. We found lower prediction capacity during the start of the pandemic and the epidemic caused by the Omicron variant. The third incidence moment method is suitable for short-term forecasting of COVID-19 with an error of around 30%. This forecast represents a short time of predictability in mainly chaotic dynamics. The predictability decreased only slightly due to pharmacological interventions and the income of new virus variants. We found low predictability in the initial periods of the epidemic and during the Omicron epidemic outbreak.