Narrow the data range. Sometimes the variable data is so large that it may exceed the default value of the computer when calculating. It can be said that multiplication is converted into addition. Such as additivity of logarithmic rate of return. Compressing variable scale makes data more stable and eliminates heteroscedasticity. The coefficient obtained by logarithmic regression is the definition of elasticity in economics.

When there are negative numbers in the data, the logarithm cannot be taken directly. You should move the data properly before taking the logarithm. General horizontal data is logarithmic, not proportional data, such as change rate.

Linear regression is performed on the logarithmic data, and the previous parameter represents the percentage change rate (dlnx=dx/x), that is, elasticity, which is a good property.

Generally speaking, the logarithm of each data will not change the nature and relationship of the data, and the obtained data will easily eliminate the heteroscedasticity problem; At the same time, after taking logarithm, economic variables have elastic significance, so variables generally take logarithmic form.