How should non-linear trends be modeled?

How should non-linear trends be modeled?

• A. Add quadratic or polynomial terms
• B. Use linear regression only
• C. Use smoothing only
• D. Use Naive forecast

Answer: (A)

Non-linear trends need curvature in the model, not a straight line. In regression, a linear trend assumes y = a + b·time, which is a flat slope over time. If the data bend—speed up, slow down, or change direction—you need terms that allow the line to curve. Adding a quadratic term time squared (and potentially higher-order terms) lets the predicted trend bend, capturing accelerating or decelerating patterns. This is the standard way to model non-linear trends within a regression framework, and you can check whether these terms actually improve fit and are statistically significant to avoid overfitting. Relying on linear regression won’t fit a curved pattern, since it enforces a constant rate of change. Smoothing can reveal nonlinearity, but it’s a nonparametric approach that doesn’t provide explicit polynomial trend terms in the equation. A naive forecast ignores the trend entirely, which leads to biased predictions when a trend exists.

Non-linear trends need curvature in the model, not a straight line. In regression, a linear trend assumes y = a + b·time, which is a flat slope over time. If the data bend—speed up, slow down, or change direction—you need terms that allow the line to curve. Adding a quadratic term time squared (and potentially higher-order terms) lets the predicted trend bend, capturing accelerating or decelerating patterns. This is the standard way to model non-linear trends within a regression framework, and you can check whether these terms actually improve fit and are statistically significant to avoid overfitting.

Relying on linear regression won’t fit a curved pattern, since it enforces a constant rate of change. Smoothing can reveal nonlinearity, but it’s a nonparametric approach that doesn’t provide explicit polynomial trend terms in the equation. A naive forecast ignores the trend entirely, which leads to biased predictions when a trend exists.