auto.arima gives bad prediction

Question

I\'m used auto.arima for time series prediction in the following way:

data_ts <- ts(new_cases_vector,
           start = c(2020, as.numeric(format(inds[1],