# What are the distributions on the positive k-dimensional quadrant with parametrizable covariance matrix?

Following zzk's question on his problem with negative simulations, I am wondering what are the parametrized families of distributions on the positive k-dimensional quadrant, $\mathbb{R}_+^k$ for which the covariance matrix $\Sigma$ can be set.

As discussed with zzk, starting from a distribution on $\mathbb{R}_+^k$ and applying the linear transform $X \longrightarrow\Sigma^{1/2} (X-\mu) + \mu$ does not work.

Suppose that we have a multivariate normal random vector $$(\log X_1,\dots,\log X_k) \sim N(\mu,\Sigma) \, ,$$ with $\mu\in\mathbb{R}^k$ and $k\times k$ full rank symmetric positive definite matrix $\Sigma=(\sigma_{ij})$.

For the lognormal $(X_1,\dots,X_k)$ it is not difficult to prove that $$m_i := \textrm{E}[X_i] = e^{\mu_i + \sigma_{ii}/2} \, , \quad i=1,\dots,k\, ,$$ $$c_{ij} := \textrm{Cov}[X_i,X_j] = m_i \,m_j \,(e^{\sigma_{ij}} - 1) \, , \quad i,j=1,\dots,k\, ,$$

and it follows that $c_{ij}>-m_im_j$.

Hence, we can ask the converse question: given $m=(m_1,\dots,m_k)\in\mathbb{R}^k_+$ and $k\times k$ symmetric positive definite matrix $C=(c_{ij})$, satisfying $c_{ij}>-m_im_j$, if we let $$\mu_i = \log m_i - \frac{1}{2} \log\left(\frac{c_{ii}}{m_i^2} + 1 \right) \, , \quad i=1,\dots,k \, ,$$ $$\sigma_{ij} = \log\left(\frac{c_{ij}}{m_i m_j} + 1 \right) \, , \quad i,j=1,\dots,k \, ,$$ we will have a lognormal vector with the prescribed means and covariances.

The constraint on $C$ and $m$ is equivalent to the natural condition $\textrm{E}[X_i X_j]>0$.

• Terrific, Paulo! You got both a working solution and the proper condition on the covariance matrix, which also answers this question. Log-normals prove handier than gammas, in the end. Apr 6, 2012 at 5:21

Actually, I have a definitely pedestrian solution.

1. Start with $X_1\sim \text{Ga}(\alpha_{11},\beta_{1})$ and pick the two parameters to fit the values of $\mathbb{E}[X_1]$, $\text{var}(X_1)$.
2. Take $X_2|X_1\sim \text{Ga}(\alpha_{21}X_1+\alpha_{22},\beta_{2})$ and pick the three parameters to fit the values of $\mathbb{E}[X_2]$, $\text{var}(X_2)$, and $\text{cov}(X_1,X_2)$.
3. Take $X_3|X_1,X_2\sim \text{Ga}(\alpha_{31}X_1+\alpha_{32}X_2+\alpha_{33},\beta_{3})$ and pick the four parameters to fit the values of $\mathbb{E}[X_3]$, $\text{var}(X_3)$, $\text{cov}(X_1,X_3)$ and $\text{cov}(X_2,X_3)$.

and so on... However, given the constraints on the parameters and the non-linear nature of the moment equations, it may be that some sets of moments correspond to no acceptable set of parameters.

For instance, when $k=2$, I end up with the system of equations $$\beta_1 =\mu_1/\sigma_1^2\,,\quad \alpha_{11}-\mu_1\beta_1 =0$$

$$\alpha_{22} = \mu_2\beta_2 - \alpha_{21}\mu_1\,,\quad \alpha_{21} = \dfrac{(\sigma_{12}+\mu_1\mu_2-\mu_2)}{\sigma^2_1+\mu_1^2- \mu_1}\beta_2$$ $$\dfrac{(\sigma_{12}+\mu_1\mu_2-\mu_2)^2}{(\sigma^2_1+\mu_1^2- \mu_1)^2} \sigma_1^2 + \dfrac{\mu_2}{\beta_2} = \sigma^2_2\,.$$ Running an R code with arbitrary (and a priori acceptable) values for $\mu$ and $\Sigma$ led to many cases with no solution. Again, this does not mean much because correlation matrices for distributions on $\mathbb{R}_+^2$ may have stronger restrictions that a mere positive determinant.

update (04/04): deinst rephrased this question as a new question on the math forum.

• One way of slighly extending this is to consider the natural exponential family $$f(\mathbf{X}|\mathbf\theta)=h(\mathbf{x})e^{\mathbf\theta^T\mathbf{X}-A(\mathbf\theta)}.$$ Then the mean and covariance are the gradient and Hessian of $A$. If $h$ is a polynomial (with real exponents > -1) then $A$ is the log of a polynomial (with real exponents), and the variance and Hessian are rational functions. I think this gives enough freedom to represent any mean and covariance matrix. Mar 31, 2012 at 17:42
• @deinst: (+1) Do you have an example where this exponential family representation can be exploited straightforwardly? Mar 31, 2012 at 19:30
• Maybe I don't quite understand the problem. But, consider a bivariate random vector $(X,Y)$ with the same marginal $F$ with full support on $\mathbb R_{+}$ and having mean $0 < \mu < \infty$. How can such a bivariate distribution have correlation $\rho$ close to -1, for example? Heuristically, though I haven't carried this out, it seems that if $\mathbb P(X > 2 \mu) > 0$, then a contradiction regarding the support must arise. No? Apr 1, 2012 at 15:00
• There are certainly constraints on the covariance matrix $\Sigma$ when the support is $\mathbb{R}^k_+$, covered via the Stieltjes moment condition. Anyway, I do not see why a correlation close to -1 is excluded a priori. Apr 1, 2012 at 15:17
• Right, this is related to what I was getting at. Regarding the correlation, consider my example. If $X$ and $Y$ have the same marginal $F$ with mean $\mu$ and a correlation of exactly -1 and $\mathbb P(X > 2 \mu) > 0$, what must the value of $Y$ be for all such realizations of $X$? (+1 on both question and answer. I like this.) Apr 1, 2012 at 15:24

OK, this is a response to Xi'an's comment. It is too long and has to much TeX to be a comfortable comment. Caveat Lector: It is virtually certain that I have made an algebra mistake. This does not seem to be quite as flexible as I first thought.

Let us create a family of distributions in $\mathbb{R}_+^3$ of the form $$f(\mathbf{x}|\mathbf\theta)=h(\mathbf{x})e^{-\mathbf\theta^T\mathbf{x}-A(\mathbf\theta)}$$ Let $\mathbf{x}=(x,y,z)$ and $\mathbf\theta=(\theta_1,\theta_2,\theta_3)$. Let $$h(\mathbf{x})=c x_1^{e_1-1}x_2^{e_2-1}x_3^{e_3-1}+d x_1^{f_1-1}x_2^{f_2-1}x_3^{f_3-1}$$ be a two term polynomial where $e_i, f_i$ are real numbers greater than 0 for all $i$. Then we find that $$A(\mathbf\theta)=\log\left(c\frac{\Gamma(e_1)}{\theta_1^{e_1}}\frac{\Gamma(e_2)}{\theta_2^{e_2}}\frac{\Gamma(e_3)}{\theta_3^{e_3}}+d\frac{\Gamma(f_1)}{\theta_1^{f_1}}\frac{\Gamma(f_2)}{\theta_2^{f_2}}\frac{\Gamma(f_3)}{\theta_3^{f_3}}\right).$$

Now, for convenience let us define $$c'=c\Gamma(e_1)\Gamma(e_2)\Gamma(e_2)\theta_1^{f_1}\theta_2^{f_2}\theta_3^{f_3}$$ and $$d'=d\Gamma(f_1)\Gamma(f_2)\Gamma(f_2)\theta_1^{e_1}\theta_2^{e_2}\theta_3^{e_3}$$

Now, as the mean of our distribution is the gradient of $A$, we have $\mu_X=\frac{e_1c'+f_1d'}{\theta_1(c'+d')}$, $\mu_Y=\frac{e_2c'+f_2d'}{\theta_2(c'+d')}$, and $\mu_Z=\frac{e_3c'+f_3d'}{\theta_3(c'+d')}$. And as the covariance is the Hessian of $A$, we have $$\sigma_X^2=\frac{(e_1c'+f_1d')(c'+d')+(e_1-f_1)^2c'd'}{\theta_1^2(c'+d')^2}$$ and $$\text{Cov}(X,Y)=\frac{(e_1-f_1)(e_2-f_2)c'd'}{\theta_1\theta_2(c'+d')}$$ (the other terms of the covariance matrix obtained by changing subscripts in the obvious way).

This does not seem to be quite enough flexibility to get any covariance matrix. I need to try another term in the polynomial (but I suspect that also may not work (obviously I need to think about this more)).

• Four parameters $(\theta_1,\theta_2,\theta_3,c)$ for five constraints...? Apr 1, 2012 at 15:24
• @xian There are the 6 exponents $e_i$ and $f_i$ as well. Apr 1, 2012 at 15:28
• I am slightly (?) confused: you did not process the exponents as parameters of the exponential family. But indeed you can change those powers as you wish towards getting the 9 moment equations right. Apr 1, 2012 at 15:55
• @Xi'an You are correct, I did not process them as parameters of the exponential family. Doing so would have made the family no longer a natural family, and including them would have just muddled the algebra for comuting the moment equations (which was muddled enough to begin with). Apr 1, 2012 at 16:10