Suppose I am trying to do a linear regression $Y = k_1X_1 + k_2X_2$. Then the statistical decision model we want to find $k_1, k_2$ such that $E(Y - k_1X_1 - k_2X_2)^2$ is minimized. However suppose $X_2$ is not random in the sense that it is something can be manipulated only by me and does not have a distribution. For instance I am trying to determine the relationship between the price of something that I sell to the revenue I get. The price is set by me a priori, so it does not have a distribution(Assume also that the relationship is not deterministic here and the revenue is influenced by some random factor $X_1$). How should I model this situation? Should I just treat $X_2$ as a parameter like $k_2$? That feels so strange.
Also my goal is to choose the optimal price so that my expected revenue is maximized.
Yes. If $X_2$ may be fixed, then we will treat it as parameter in $E(Y - k_1X_1 - k_2X_2)^2$ for all fixed values of $X_2$. In other words, we will find $\min_{k_1}\mathbf{E}(Y - k_1X_1 - x_2c)^2$ and get optimal value for fixed $X_2 = x_2$. We also may find $\min_{x_2 \in D}\min_{k_1}\mathbf{E}(Y - k_1X_1 - x_2c)^2$ if we want optimal value for all $x_2 \in D$ where $D$ is all possible values of $X_2 = x_2$.