Chapter 8: Perform the Analysis: Predictive Analytics

The AMPS Model
In the AMPS cycle, we’re now
going to continue Performing
the Analysis. One of the types
of analyses, predictive
analytics, will be our focus of
this chapter.

Predictive analytics addresses the question of

whether
something or some event will happen in the future. To the
extent that there is a strong correlation between the past and
future, predictive analytics can be particularly useful.

Predictive analytics as analytics performed to provide

foresight by identifying patterns in historical data by judging
likelihood or probability.

Whereas descriptive and diagnostic data analyses are more
deterministic generally reporting known facts (covered in
Chapters 6 and 7), predictive and prescriptive analytics
provide less deterministic output and more

probabilistic
models, judging the likelihood and probability of a future
event or outcome occurring.

With deterministic models, all events are known before hand,
with probabilistic, there is always an

element of chance of
whether or not it will happen.

Examples of probabilistic models:

• Predicting which
customers will be able to
pay back their loan. This
might help decide whether
to extend credit to
potential borrowers.
• Predicting whether a
company will need to
restate its financial
statements.
• Predicting which
companies would be more
likely to commit financial
statement fraud.
• Predicting future
performance, including
sales, net income, and
cash flows.

Exhibit 8.1 Circumplex of Types of Data Analyses

Classification is a predictive analytics technique used to

separate or classify a sample (or population) into two or more
groups or classes.

We are predicting a probabilistic outcome, or what might
happen based on our forecasts.

With classification we separate observations
into groups or classes.

• Bankrupt/Not Bankrupt.
• Fraud/No Fraud.
• Misstatement/No
Misstatement.
• Audit Client.
• Acceptance/Client
Rejection.
• Extend Loan/Do Not
Extend Loan.
• Continue as Going
Concern/Do Not Continue
as a Going Concern.

Altman’s Z can be used for

bankruptcy
classification/bankruptcy prediction.

Altman’s Z found five factors that
helps predict bankruptcy.

X1: Working capital / Total assets
X2: Retained earnings / Total assets
X3: Earnings before interest and
taxes / Total assets
X4: Market value of stockholders’
equity / Book value of total debt owed
X5: Sales / Total assets

X1: Working capital / Total assets

Measures how liquid, cash-like
assets (or liquidity level in relation to
the size of the company).

X2: Retained earnings / Total assets

Measures long-term profitability over
the life of the company.

X3: Earnings before interest and
taxes / Total assets

Measures recent, or short-term
profitability of the company.

X4: Market value of stockholders’
equity / Book value of total debt owed

Measures long-term solvency of the
company, or whether the company will
have sufficient funds to pay its debt as it
comes due.

X5: Sales / Total assets

Measures asset efficiency, or how well
assets are utilized.

Original Z-score formula:

Z = 1.2×1 + 1.4×2 +
3.3×3 + 0.6×4 + 1.0×5.

Decisions rules determine which cases fit within
each class.

If Z < 1.80: Classify as significant
risk of bankruptcy, or in
the “distress zone”

If Z > = 1.80 and
Z < 3.00:
Classify as at risk of
bankruptcy, or “gray
zone”

If Z > = 3.00: Classify as not currently
at risk of bankruptcy, or
“safe zone”

The results of the classification appear in the

base analysis. 1

Different characteristics can help determine

whether a loan should be given.
Banks and other lenders evaluate borrowers based on their
credit worthiness to decide whether they will extend a loan to
them.
Lenders use various borrower characteristics for individuals
including credit scores, work history, existing levels of total
debt, debt-to-income ratios, if they own or rent their home, as
well as other factors to determine whether to extend a loan.

Factors that influence loan extension could
include

the amount and work tenure.

Amount of the Requested
Loan:

Would you guess if
the greater the requested
loan by the prospective
borrower, the lesser or
greater the chance of
Lending Club granting a
loan?
We would expect the bigger
the loan, the less likely the
loan be accepted, so expect
a negative (−) relationship
between Loan Acceptance
and Loan Amount.

Employment Length:

Would
you guess if the longer the
employment length, the lesser
or greater the chance of
Lending Club granting a loan?
We would guess the longer
the employment length, the
more certain that the loan will
be accepted due to stability of
the job. So, we would expect
a positive (+) relationship
between Loan Acceptance
and Employment Length.

Factors that influence loan extension could
include

how leveraged the individual is.

Debt-to-Income Ratio:

Would you guess if the greater the
debt-to-annual income ratio of the prospective borrower, the
lesser or greater the chance of Lending Club granting a
loan?
We would expect the bigger the debt relative to income, the
less likely the loan be accepted, so expect a negative (−)
relationship between Loan Acceptance and Debt-to-Income
Ratio.

Classification can be used to identify

fraud or no
fraud.

Beneish (1999) conducted a
predictive analytics from 1982 to
1992, comparing companies that
were found to have committed
fraud by the Securities and
Exchange Commission (SEC) (in
Accounting and Auditing
Enforcement Releases) with a
sample of companies that did not
commit fraud.
Beneish found eight factors to be
predictive of financial statement
fraud. Using these eight factors, he
derived an index called an M-score
(similar to Altman’s Z-score
presented earlier), which is used to
predict and classify which firms
have likely committed financial
statement fraud.

Beneish selected eight factors that might
predict the fraud class. 1

• Increase in receivables as
compared to last period.
• Decline in gross margin as
a percent of sales.
• Decline in asset quality
index (noncurrent assets
as compared to current
assets).
• Increase in sales growth in
current period.
• Decrease in depreciation
expense.
• Decrease in selling, general
and administrative costs.
• Increase in debt (leverage).
• Higher total accruals to total
assets.

Regression is a predictive analytics technique that allows the
accountant to

estimate a specific dependent variable
outcome value based on independent variable inputs

Regression is used in management
accounting to determine

cost behavior.

Firm Cost Behavior
Overhead Cost Drivers

Firm Cost Behavior

How are firm costs
dependent on the level of
production for a company
that makes composite
decking?

Dependent variable (DV):
Total production costs

Independent variable (IV):
Board feet produced

Overhead Cost Drivers

What drives overhead costs
for a company that makes
deliveries?

DV: Daily Overhead Costs

IVs: Deliveries made, Miles
driven, Delivery time, Weight
of all deliveries

Regression can be used to predict

interest
rates and social media sentiment

Interest Rates

What interest rate do we give to
a borrower?

DV:
Interest Rate Given to Borrower

IVs:
Borrower Income, Level of Debt
Compared to Income,
Credit History,
Rent/Own,
Credit Score

Social media sentiment and
stock returns

Does sentiment predict returns?

DVs: Different measures of
stock market return

IVs: Different measures of social
media sentiment (positive versus
negative words; number of
positive posts versus negative
posts, etc.)

Regression is used to evaluate

risk and
return.

Risk and Return

What is the relationship
between investment risk taken
and expected investment
returns?

DVs: Different measures of
expected investment return [for
example, stock prices (or
returns), bond yields, or returns
on real estate investments, etc.]

IVs: Different measures of risk
[for example, stock price
volatility, volatility of earnings,
beta, bankruptcy risk (also
known as default risk), level of
debt (leverage), etc.]

Base Rates are defined as

probability of an event occurring
based on a related historical average

Base Rate Fallacy occurs when

the prediction places too
little weight on the base rates of the past and instead uses
different or new information.

One of the most important things accountants do is forecast
future sales, earnings, and other measures of performance.
Time series analysis is a tool/technique used to

predict future
values based on past values of the same variable.

One of the reasons accountants carefully measure the past
is to be able to predict the future.

Example of Time Series Analysis Using Quarterly Data
for Microsoft

Persistence is

the repeatability (or continuity) and durability
of the financial statement variables (for example, earnings,
sales, cash flows) over time. Simply, it estimates if the trend
of the past financial performance continue will into the future.

Persistence: Dechow & Schrand

Dechow and Schrand (2004)
looked at the persistence of
various financial statement
variables over the 1987–2002
time period. Their basic model
used current financial
performance (for period t) to
predict financial performance
for the next period (time t+1):
Performance_(t+1) = B₀ + B₁

(Performance)
Where B₁ is our measure of
persistence.
Note accrual-basis measures
of performance exhibits
higher persistence than
cash-basis measures of
performance.

Persistence is higher for figures on the income
statement. 1
Exhibit 8.5 Persistence of Various Financial Statement Items

Hypothesis testing is sometimes used in conjunction with
predictive analytics. Using hypotheses, accountants can

understand the impact of different decisions.

Hypothesis testing helps us understand Lowe’s
Home Improvement credit card users.
For example, imagine Lowe’s Home
Improvement wants to understand
the future impact of Lowe’s
continuing to offer its own credit
card.
Management suggests three
hypotheses be tested to
support/reject Lowe’s having its own
credit card:

H₁: Lowe’s customers using a Lowe’s
card will have a higher average sales
amount than those customers not paying
with a Lowe’s credit card.
H₂: Lowe’s customers using a Lowe’s
credit card make more Lowe’s visits than
those customers not paying with a
Lowe’s credit card.
H₃: Lowe’s customers using a Lowe’s
credit card buy higher-margin products
than those customers not paying with
a Lowe’s credit card

Comparing multiple models with hypothesis
testing shows impact. 1

Using statistical tests such as a t-test of a difference in
means or through use of regression analytics to see what the
different is for customers using or not using a Lowe’s credit
card.
This will allow accountants and data scientists to test these
hypotheses and start to understand the impact of having their
own credit card or not having their own credit card.

Machine learning is

the ability of a computer (or machine) to
automatically learn on its own without being explicitly
programmed to do so.

A form of machine learning in accounting is

robotic process
automation (RPA) referring to software programs that
automate certain repeatable tasks.

In other words, computer scripts are written to regularly
access various accounting systems, extract data, process
and reconcile them and then report “All is well” or “Here’s
some exceptions to investigate further.”