×

Integrative method for modeling multiple asset classes

  • US 20030110016A1
  • Filed: 06/29/2001
  • Published: 06/12/2003
  • Est. Priority Date: 06/29/2001
  • Status: Active Grant
First Claim
Patent Images

1. A method for combining two or more risk models to create a risk model with wider scope than its constituent parts. The method insures that the combined risk model is consistent with the component models from which it is formed. The method consists of the following steps:

  • letting C1 denote a class of algorithms for constructing estimates of a covariance matrices from time histories of data;

    letting C2 denote a class of asset classes;

    for x in C2 let C3(x),denoting a class of multi-factor risk models for x;

    for y in C3(x) denoting its parts as follows;

    factor exposures X(y,t) at time t;

    factor returns f(y,t) at time t; and

    specific covariance matrix D(y,t) at time t;

    factor covariance matrix F(y,t) at time t;

    giving the following components;

    two or more asset classes x1, . . . , xn, et x denote an asset class which is a union of these given asset classes;

    for each asset class x1 giving a risk model yi in C3(xi);

    letting Y(t) be such that the decomposition;

    (f

    (y1,t)
    f

    (y2,t)


    f

    (yN,t)
    )


    f

    (t)
    =(Y1

    (t)
    Y2

    (t)


    YN

    (t)
    )


    Y

    (t)






    g

    (t)
    +(φ

    1


    (t)
    φ

    2


    (t)


    φ

    3


    (t)
    )


    φ



    (t)
    which results in residuals φ

    (t) such that correlation (φ

    1

    j) is nearly zero if i≠

    j; and

    constructing a risk model for x as follows;

    forming X(t)=diag(X(y1,t), . . . , X(yn,t));

    forming D(t)=diag(D(y1,t), . . . , D(yn,t));

    applying a method from C1 to estimate a covariance matrix G(t) from a history of the g(t)s; and

    applying an optionally different method from C1 to estimate a covariance matrix Φ

    (t) from a history the φ

    (t)s;

    Then X(t)[Y(t)G(t)Y(t)t

    (t)φ

    (t)]X(t)t+D(t) is a risk model for x. Insure the risk model is consistent with each component, asset class risk model as follows;

    Let F1(t) be the block diagonal matrix obtained from Y(t)G(t)Y(t)′



    (t) by setting all elements to zero except those of the blocks corresponding to each asset class. Each such block represents the covariance among the factors explaining risk for a particular asset class. Let F2(t) be the block diagonal matrix whose blocks contain the asset class factor covariance matrices, F(y1,t) in the same order as they appear in F1(t);

    the off-block diagonal elements are zero. Given a real symmetric positive semi-definite matrix M, let M1/2 denote a square root of M so that M1/2(M1/2)=M. There may be several choices for M1/2. Let M

    1/2
    denote the inverse of M1/2, or in the event that the inverse does not exist, let M

    1/2
    be the pseudoinverse. Then X(t)(F21/2F1

    1/2
    (Y(t)G(t)Y(t)′



    (t))(F21/2F1

    1/2
    )′

    )X(t)′

    is a risk model that is consistent with the component asset class models.

View all claims
  • 6 Assignments
Timeline View
Assignment View
    ×
    ×