Supposing to have two random dependent variables $X$ and $Y$ how can one calculate the covariance between the two variables $X$ and $X \cdot Y$? That is


Trying using the definition does not lead to a simple result, so are there some properties about this particular covariance?

(For istance if $Y$was a constant then it would be

$$\mathrm{Cov}(X,XY)=Y \mathrm{Cov}(X,X)=Y\mathrm{Var}(X)$$

But this is not the case)

Nevertheless the following reasoning makes me think that it should be $\mathrm{Cov}(X,XY)=Y\mathrm{Var}(X)$ also in this case.

Infact take two indpendent variables $a$ and $b$ and say I want to calculate the error $\sigma_C$ on the quantity $$C=ab-a=a(b-1)$$

Since I can see $C$ in two ways (as highlighted above) I can calculate $\sigma_C$ in two ways:

$C=ab-a \to \mathrm{ab \,\, and \, \, a \,\, are \,\,\, dependent} \to \sigma_C^2=\sigma_a^2 b^2 +\sigma_b^2 a^2 +\sigma_a^2-2 \mathrm{Cov}(a,ab)$

$C=a(b-1) \to \mathrm{a \,\, and \, \, (b-1) \,\, are \,\, not \,\, dependent} \to \sigma_C^2=\sigma_b^2 a^2 +\sigma_a^2 (b-1)^2 $

These two must be equal, so

$$\sigma_b^2 a +\sigma_a^2 (b-1)^2=\sigma_a^2 b^2 +\sigma_b^2 a^2 +\sigma_a^2-2 \mathrm{Cov}(a,ab) \implies \mathrm{Cov}(a,ab)=b^2 \sigma_a^2$$

  • 1
    $\begingroup$ See this paper: George W. Bohrnstedt and Arthur S. Goldberger, 1969. On the Exact Covariance of Products of Random Variables. Journal of the American Statistical Association Vol. 64, No. 328, pp. 1439-1442 jstor.org/stable/2286081 $\endgroup$ Aug 28, 2016 at 9:53
  • 1
    $\begingroup$ Note that $\mathrm{Cov}(X,X\,Y)$ is a number while $Y\,\mathrm{Var}(X)$ is a random variable (non-constant if $Y$ is non-constant and $\mathrm{Var}(X)>0$, so the equality $\mathrm{Cov}(X,XY)=Y\mathrm{Var}(X)$ does not make sense. $\endgroup$ Aug 28, 2016 at 9:56

2 Answers 2


This is a nice problem for testing the development code in the next version of mathStatica.

Note that $\text{Cov}(X, XY) = \mu_{1,1}(X, XY)$ (i.e. the covariance operator is the {1,1} central moment), which is why I am requesting the {1,1} CentralMoment of {X, X Y} ... when the variables are {X,Y}:

enter image description here

where $\mu_{2,1} = E\left[(X-E[X])^2 \;(Y-E[Y])^1\right]$

If $X$ and $Y$ are independent (information not stated in the problem), then the answer simplifies further:

enter image description here


My first reaction is that you won't be able to find this value without knowing the dependence structure between $X$ and $Y$. This is further verified by using law of total covariance as shown below, \begin{align*} Cov(X,XY)& = E[Cov(X,XY\mid X)] + Cov(E[X\mid X],E[XY \mid X])\\ & = E[X^2Cov(1,Y \mid X)] + Cov(X, X\, E[Y \mid X])\\ & = 0 + Cov(X, X\, E[Y \mid X])\\ & = Cov(X, X\, E[Y \mid X]). \end{align*}

$Y$ goes away due to the expectation, so your claim cannot hold. If you know the expectation of $Y \mid X$ then you can find this. If not, then you won't be able to solve this.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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