Representing Mean and Volatility Views in Entropy Pooling

Summary

The document asks how to express ordinary statistical views in the scenario-based form used by Entropy Pooling. It sets up a matrix of simulated outcomes for two variables, assigns equal prior probabilities to the scenarios, and describes a view matrix whose columns represent functions of the market. The example views concern the first variable’s mean and standard deviation.

The central issue is how to turn those views into scenario-level columns that can be used in the entropy-pooling constraints. The document provides a small numerical scenario set to make the dimensions concrete, but it does not give the transformation or a worked solution. It therefore introduces the representation problem rather than teaching the full procedure; implementation details such as how the standard-deviation condition is encoded remain unresolved.

Key ideas

• Entropy Pooling updates scenario probabilities to reflect views while retaining a prior scenario distribution.
• Each row of the scenario matrix represents one joint outcome, and each column represents an asset or variable.
• The view matrix contains scenario-level functions used to express constraints on the market distribution.
• The example asks how to encode a mean and a standard-deviation view for one variable.

Tags

# Non-Parametric Entropy Pooling View and Constraint Matrix Structure


# Non-Parametric Entropy Pooling View and Constraint Matrix Structure












Meucci has a compelling system to merge a prior distribution with subjective views about risk drivers called Entropy Pooling. The original paper is here. In the non-parametric (scenario-based) approach, we take a $J \times N$ matrix $X$, where $J$ is the number of simulations, and $N$ is the number of assets. Then we associate a $J \times 1$ vector of probabilities $p$ to each simulation outcome. From there we create a $J \times K$ matrix of views $V$, where $J$ again represents the number of scenarios, and $K$ represents a generic function of the market. I can't quite visualize what the $J \times K$ matrix would look like for standard views. Let's say that I have simulated 10 joint scenarios for a two-variable standard normal distribution.

$$ \mathbf X= \begin{pmatrix} -.85 & 1.02 \\ 1.62 & .36 \\ .31 & .93 \\ -.53 & .36 \\ 1.1 & -.84 \\ -1.49 & .30 \\ 1.45 & -.31 \\ .37 & -.64 \\ .96 & -.29 \\ 1.01 & .97 \\ \end{pmatrix} \space \space \mathbf p= \begin{pmatrix} .1 \\ .1 \\ .1 \\ .1 \\ .1 \\ .1 \\ .1 \\ .1 \\ .1 \\ .1 \\ \end{pmatrix} $$

I have two views: asset one will have an average of $.5$, and asset one will have a standard deviation of $.1$. How do I transform these standard views into a conformable $J \times K$ matrix?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.