Consider a $2\times 1$ random vector $$ V\equiv \begin{pmatrix} V_1\\ V_2 \end{pmatrix} \sim \mathcal{N}\left(\begin{pmatrix} 0\\ 0\end{pmatrix}, \begin{pmatrix} 1& \rho\\ \rho & 1 \end{pmatrix}\right) $$ with $\rho\neq 0$ and $\rho\in [-1,1]$ defined on the probability space $(\Omega, \mathcal{F}, \mathbb{P})$. I am asking your help to solve the following exercise on identification in econometrics.

Consider the system of equations $$ \begin{cases} \mathbb{P}(V_1\geq V_2-b_1; \rho)=p_1\\ \mathbb{P}(V_1\geq V_2-b_2; \rho)=p_2\\ \mathbb{P}(V_1\geq V_2-b_3; \rho)=p_3\\ \end{cases} $$ where the unknowns are $(b_1,b_2, b_3, \rho)$ while $(p_1,p_2,p_3)\in [0,1]^3$ is known.

It is assumed that this system has at least one solution (i.e., the model is correctly specified).

I want to show that this system has at least two solutions denoted by $(b^*_1,b^*_2, b^*_3, \rho^*)$ and $(b^{**}_1,b^{**}_2, b^{**}_3, \rho^{**})$. Below I have reported what I have tried and my doubts.

I started my analysis with deriving, for any $b\in \mathbb{R}$, the parametric expression of $$ \mathbb{P}(V_1\geq V_2-b). $$

From the assumption above we have

  • $V_1 \mid V_2=v_2\sim \mathcal{N}(\rho v_2, (1-\rho^2))$ for any realisation $v_2$ of $V_2$

  • $V_2\sim \mathcal{N}(0,1)$

Hence $$ \mathbb{P}(V_1\geq V_2-b)= E(\mathbb{P}(V_1\geq V_2-b \mid V_2))= \int_{\mathbb{R}} \Big[1-\Phi\Big(\frac{v-v\rho-b}{\sqrt{1-\rho^2}}\Big)\Big]\phi(v) \,\mathrm{d}v, $$ where $\Phi, \phi$ are the cdf and pdf of the standard normal.

Therefore the system of equations becomes $$ \begin{cases} \displaystyle\int_{\mathbb{R}} \Big[1-\Phi\Big(\frac{v-v\rho-b_1}{\sqrt{1-\rho^2}}\Big)\Big]\phi(v) \,\mathrm{d}v=p_1\\ \displaystyle\int_{\mathbb{R}} \Big[1-\Phi\Big(\frac{v-v\rho-b_2}{\sqrt{1-\rho^2}}\Big)\Big]\phi(v) \,\mathrm{d}v=p_2\\ \displaystyle\int_{\mathbb{R}} \Big[1-\Phi\Big(\frac{v-v\rho-b_3}{\sqrt{1-\rho^2}}\Big)\Big]\phi(v) \,\mathrm{d}v=p_3\\ \end{cases} $$ From here I do not know how to complete the exercise. Is non-uniqueness of solution obvious because I have three equations and four unknowns?


$\def\Φ{{\mit Φ}}$It suffices to prove that for any $-1 \leqslant ρ \leqslant 1$, there exists $b_1, b_2, b_3 \in \mathbb{R}$ satisfying$$ P(V_2 - V_1 \leqslant b_k; ρ) = p_k. \quad k = 1, 2, 3 $$

Because$$ \begin{pmatrix} V_1 \\ V_2 \end{pmatrix} \sim N\left( \begin{pmatrix} 0 \\ 0 \end{pmatrix}, \begin{pmatrix} 1 & ρ \\ ρ & 1 \end{pmatrix} \right), $$ then\begin{align*} V_2 - V_1 = \begin{pmatrix} -1 & 1 \end{pmatrix} \begin{pmatrix} V_1 \\ V_2 \end{pmatrix} &\sim N\left( \begin{pmatrix} -1 & 1 \end{pmatrix} \begin{pmatrix} 0 \\ 0 \end{pmatrix}, \begin{pmatrix} -1 & 1 \end{pmatrix} \begin{pmatrix} 1 & ρ \\ ρ & 1 \end{pmatrix} \begin{pmatrix} -1 \\ 1 \end{pmatrix} \right)\\ &= N(0, 2(1 - ρ)). \end{align*} Since the distribution function of $N(0, 2(1 - ρ))$ is $F(x; ρ) = \Φ\left( \dfrac{x}{\sqrt{\smash[b]{2(1 - ρ)}}} \right)$, then$$ b_k = \sqrt{\smash[b]{2(1 - ρ)}} \Φ^{-1}(p_k). \quad k = 1, 2, 3 $$

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.