Why can extrapolation be risky even when a fitted line has a strong correlation in the observed data? Identify the statement supported by the appropriate theorem.