I need to calculate the price of exchange option between 2 assets $S_1$ and $S_2$ The formula is given here Wiki: Magrabe formula or here Quant Stack Exchange. In the derivation of the formula it is assumed, that price is the discounted expecation of future claim under risk-neutral measure $Q$: $$p=e^{-r_{f}}\mathbb{E}_{Q}(S_1-S_2)^{+}$$. and under risk-neutral measure it is assumed, that: $$dS_{i,t}=\mu_{i}dt+\sigma_{i}dW^{i}_{t}$$ $$\mu_{i}=r_{f}\text{, for }i=1,2$$.
My question is:
Why do we calculate expectation under risk-neutral measure $Q$ and not real-world measure $P$ ? In real world return on each asset can be different, i.e. $\mu_1\neq\mu_2\neq r_f$. Then it would make sence to use real-world measure, under which this condition is fullfilled.