Solving a System of Equations in Excel

Overview
If you have a system of linear equations, you can quickly solve it with Linear Algebra using the Matrix
formulas in Excel. This works in the Pearson Vue environment as well.

CAS Problems where this is useful:

Exam 9 - Solving for factors in a Multifactor APT Model (BKM Ch. 10)
Exam 8 - Solving for factors in Multi-Dimensional Credibility (Couret & Venter)

→ See the next two tabs for examples from the Exam 9 Problem Pack and Exam 8 Cookbook.

Key Excel Formulas

MINVERSE(A) Returns the inverse of A
MMULT(A,B) Multiplies matrices A and B

How to calculate Array formulas in Pearson Vue:

1 - Highlight the output cells
2 - Enter the formula
3 - Press CTRL + SHIFT + ENTER to calculate the array output

Example
Solve the system of linear equations below for unknown variables a, b and c:

2a + 12b + 4c = 25
3b + 12c = 9
8a + 6b + 11c = 32

In matrix notation, this system of equations is A * X = B where:

A = Matrix of coefficients
X = The unknown variables (solution vector): a, b and c
B = The vector of constants

2 12 4
A= 0 3 12 X=
8 6 11
 25
B= 9
32

Solution
In Linear Algebra, the solution for the unknown variable vector is: X = A-1 * B

Step 1: Calculate the inverse of A, denoted A -1

A-1 = Inverse of matrix A -0.0399 -0.1104

A-1 = MINVERSE( A ) A-1 = 0.0982 -0.0102
-0.0245 0.0859

Step 2: Calculate the solution vector by mutliplying A -1 and B

X = A-1 * B 2.3282
X = MMULT( A-1 , B ) X= 1.5767
0.3558

Alternative Solution
Alternatively, you can solve for X in one step by combining the two steps above:

2.3282
X = MMULT( MINVERSE( A ) , B ) X= 1.5767
0.3558

Questions?
Send me an email at: [email protected]

©Rising Fellow
www.RisingFellow.com
Linear Algebra using the Matrix

and Exam 8 Cookbook.

a
b
c
 0.1350
-0.0245
0.0061

eps above:
RF BKM 10-1
Note: This is a practice problem from the Rising Fellow Exam 9 Problem Pack

Consider three well-diversified portfolios with the information below, analyzed using a three-factor AP
by an investor:

Portfolio βA βB βC E[rp]
1 0.8 1.2 0.75 13.9%
2 0.6 1.5 1.4 16.2%
3 1.2 0.5 1.2 12.2%

T-Bill rate = 2.0%

a.
Calculate the expected return of each factor A, B, and C.

b.
The investor also sees a well-diversified porfolio, P, promising a 14.5% expected returns. It has the fol

βA = 0.9 βB = 1 βC =

Create an arbitrage opportunity using this new portfolio and some combination of two or more of the
three analyzed portfolios.

Solution
Part a
Portfolio βA βB βC
X = E[RX] = 0.119 (1) 0.8 1.2 0.75
Y = E[RY] = 0.142 (2) 0.6 1.5 1.4
Z = E[RZ] = 0.102 (3) 1.2 0.50 1.2

Solve system of linear equations with matrix algebra (see note in Discussion):

Inverse Matrix of portfolio betas:

1.2 -1.2 0.61 A=
1.1 0.1 -0.7 B=
-1.7 1.15 0.5 C=
 4.0% = E[RA]
6.0% = E[RB]
2.0% = E[RC]

Part b
Find the expected return of the new portfolio based on the APT model:
E[r] = 0.142

The investor expects 14.5%, which is 0.3% above the APT expected return, so we want to exploit this.

By inspection, the betas for each factor of the portfolio are the averages of the betas for each respective
factor in portfolios 2 and 3. For portfolio Q, set the weights of portfolios 2 and 3 at 50%:

β'A = 0.9
β'B = 1
β'C = 1.3

E[rQ] = 14.2%

If we short portfolio Q and take a long position in the investor's portfolio, the combined portfolio has:

β''A = 0
β''B = 0
β''C = 0

E[r] = 0.3%

All betas are zero and the portfolio is well-diversified, so the arbitrage will receive a 0.3% riskless retur

Discussion
Other combinations will work, but look for shortcuts like the above if you're given a similar problem.

Solve the system of linear equations quickly using matrices.

Here's a quick recap from your old Linear Algebra class:

In matrix notation, this system is A * X = B where:

A = Matrix of betas for the portfolios
X = The factors: A, B and C
B = The matrix (array) of risk premiums for the portfolios
To solve for X (factors), we need to get the inverse of A, denoted A-1.

The solution for the factors is: X = A-1 * B

In Excel, use the following formulas:

MINVERSE(A) - Returns the inverse of A
MMULT(A,B) - Multiplies matrices A and B
Note: Hold CTRL + SHIFT + ENTER for array formulas

Here's a link for more on solving systems of linear equations in Excel:

https://www.excel-easy.com/examples/system-of-linear-equations.html

Recipe
BKM 10 - Multifactor APT Arbitrage
analyzed using a three-factor APT model

expected returns. It has the following betas:

1.3

mbination of two or more of the

E[RP]
0.119
0.142
0.102

Factors
#NAME? = E[RA]
#NAME? = E[RB]
#NAME? = E[RC]
 6.0% = E[rA]
8.0% = E[rB]
4.0% = E[rC]

urn, so we want to exploit this.

of the betas for each respective

os 2 and 3 at 50%:

o, the combined portfolio has:

will receive a 0.3% riskless return.

similar problem.
 Couret & Venter - Multi-Dimensional Credibility
Note: This is a recipe from the Rising Fellow Exam 8 Cookbook

Problem
Given the following information for Class 294, which is part of Hazard Group B:

Claim Counts by Injury Type - Class 294

Year F PT TT
2014 5 15 600
2015 10 20 650
2016 5 20 660

• Hazard Group B ratio F:TT VB = 0.005

• Hazard Group B ratio PT:TT WB = 0.01

Variance of Injury-Type Ratio Across HG B Classes

Measure V W
VHM 0.00026 0.00043
EPV 0.46 0.88

• Covariance across classes for HG B Cov(V, W) = 0.00035

a. Calculate the multiplicative factors b and c that can be used with multi-dimensional credibility to
estimate the F:TT ratio, vi, for Class 294.

b. Calculate the predicted F:TT ratio, vi, using multi-dimensional credibility and parameters b and c fro

Solution
Part a – Multiplicative Factors

1) Solve for the unconditional variance of each variable along the matrix diagonal, Var(V i) and Var(W

𝑉𝑎𝑟(𝑉_𝑖 )=(𝐸𝑃𝑉_𝑉)/𝑚_𝑖 + 〖𝑉𝐻𝑀〗 _𝑉 Var (Vi) =

Var (Wi) =
𝑉𝑎𝑟(𝐹:𝑇𝑇)=(𝐸𝑥𝑝. 𝑃𝑟𝑜𝑐𝑒𝑠𝑠 𝑉𝑎𝑟)/(#𝑇𝑇 〖𝐶𝑙𝑎𝑖𝑚𝑠〗 _(𝐶𝑙𝑎𝑠𝑠 𝑖) )+𝑉𝑎𝑟 𝐻𝑦𝑝𝑜𝑡ℎ𝑒𝑡𝑖𝑐𝑎𝑙 𝑀𝑒𝑎𝑛

2) Set up the matrix equation with the covariance matrix to solve for the multiplicative factors.

𝑣_𝑖 ) ))=( ■8(𝑉𝑎𝑟(𝑉_𝑖

VHMV)&𝐶𝑜𝑣(𝑉_𝑖,𝑊_𝑖
-> )@𝐶𝑜𝑣 (𝑉_𝑖,𝑊_𝑖 )&𝑉𝑎𝑟 (𝑊_𝑖 ) ))∙(■8(𝑏_𝑣@𝑐_𝑣 )) → (■8([email protected]))=( ■8(0.000501&
 B = A x X
bv cv
0.00026 0.000501 0.00035 bv
0.00035 0.00035 0.000891 cv

3) Solve the system of equations for the multiplicative factors.

Solve system of linear equations with matrix algebra (see note in Discussion):

Inverse Matrix of covariance matrix: Multiplicative Factors

2752.5 -1081.5 bv = 0.337
-1081.5 1547.6 cv = 0.260

Alternatively, you can solve for the factors as a system of equations using algebra.

Part b – Predicted Class F:TT Ratio

4) Calculate the frequency relativity for each injury type as a ratio to TT claims.

𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑖𝑡𝑦= (#𝐼𝑛𝑗𝑢𝑟𝑦−𝑡𝑦𝑝𝑒 𝐶𝑙𝑎𝑖𝑚𝑠)/(#𝑇𝑇 𝐶𝑙𝑎𝑖𝑚𝑠) Vc = 0.0105

Wc = 0.0288

5) Use the multiplicative factors to calculate the predicted frequency relativity for the injury type to
claims. Remember that the multiplicative factors will vary by injury type.

〖𝑣 _𝑐 〗 ^𝑝𝑟𝑒𝑑=𝐸[𝑉]+𝑏_𝑣 (𝑉_𝑐−𝐸[𝑉])+ 𝑐_𝑣 (𝑊_𝑐−𝐸[𝑊]) 𝑣_𝑐𝑙𝑎𝑠𝑠^𝑝𝑟𝑒𝑑= 0.0117

𝑣_𝑐𝑙𝑎𝑠𝑠^𝑝𝑟𝑒𝑑= 𝑉_𝐻𝐺+𝑏_𝑣 (𝑉_𝑐𝑙𝑎𝑠𝑠−𝑉_𝐻𝐺 )+𝑐_𝑣 (𝑊_𝑐𝑙𝑎𝑠𝑠−𝑊_𝐻𝐺 )

Discussion
We’re excluding Minor and Major claims so that this is a more realistic problem. If we include
Minor and Major claims, the matrix algebra in steps two and three would be based on a 4x4 matrix
with a system of four equations to solve.

One tricky thing to remember in step two is that Cov(V i, vi) is set to the VHMV. In order to solve
for the bw and cw parameters for the wi ratio, we would use the matrix equation below and Cov(W i, wi)
would be set to the VHMW. This is how we get different credibility parameters for calculating the
different predicted vi, wi, xi and yi ratios.
 To solve for the wi ratios we would use the formulas below:
𝐶𝑜𝑣 (𝑉_𝑖, 𝑤_𝑖 ))¦(𝐶𝑜𝑣 (𝑊_𝑖,𝑤_𝑖 ) ))= ((𝑉𝑎𝑟 (𝑉_𝑖))¦(𝐶𝑜𝑣 (𝑉_𝑖, 𝑊_𝑖)) (𝐶𝑜𝑣 (𝑉_𝑖, 𝑊_𝑖) )¦(𝑉𝑎𝑟 (𝑊_𝑖)))∙(𝑏_𝑤¦𝑐_𝑤 )

(.00035¦.00043)= (.000501¦.00035 .00035¦.000891)∙(𝑏_𝑤¦𝑐_𝑤 )

Couret & Venter Assumptions

The uppercase Vi, Wi, Xi, and Yi indicate the observed ratios for injury-types. The lowercase v i,
wi, xi, and yi indicate the population hypothetical mean ratios for the ith class.

Couret & Venter use some simplifying assumptions in the model, discussed on the bottom of pg. 78
in the paper. The observed ratio Wi is assumed to be the hypothetical mean w i plus a random
error term. Based on this simplifying assumption, we use Cov(W i, wi) = Cov(wi, wi) = VHMW.

Source
Couret & Venter – pg. 77-79

More Practice
CAS 2019 – 1
lti-dimensional credibility to

ility and parameters b and c from part a.

trix diagonal, Var(V i) and Var(Wi).

0.000501
0.000891

the multiplicative factors.

[email protected]))=( ■8(0.000501&[email protected]&0.000891))∙(■8(𝑏_𝑣@𝑐_𝑣 ))
ng algebra.

relativity for the injury type to TT

ype.

problem. If we include
uld be based on a 4x4 matrix

VHMV. In order to solve

equation below and Cov(W i, wi)
meters for calculating the
ypes. The lowercase v i,

ssed on the bottom of pg. 78

mean w i plus a random
Cov(wi, wi) = VHMW.