There are different types of models used to represent a certain data. Where / when / why does one may use polynomial models ( such as cubic ones) to represent the data?
The simplest type of functions are polynomial ones. They are easy to evaluate their derivatives, integrals, zeroes and basically most notions you can associate to a function. Furthermore polynomial functions can be used to suitably approximate any other function in a specific interval by the Stone_Weierstrass Theorem. It is also quite easy to approximate suitable datasets by regression methods provided you use a high enough degree polynomial but not large enough to cause "overfitting". Now to find what degree you need there are a couple of methods, the easiest being just to conduct a visual inspection and note how many monotonicity changes you get. Notice that linear regression (if you are familiar with the term) is just a simple case of polynomial regression.