A logarithmic model is above an exponential model throughout the observed data window. What warning should you give before forecasting far beyond the data?