Introduction to Data Analytics for Accounting, Second Edition by Vernon J. Richardson, Katie L. Terrell and Ryan A. Teeter

By

Introduction to Data Analytics for Accounting, Second Edition

Vernon J. Richardson, Katie L. Terrell and Ryan A. Teeter

Introduction to Data Analytics For Accounting

Table of Contents

Chapter 1

Ask the Question: Using Data Analytics to Address

Accounting Questions 2

The Explosion of Data and the Impact on the

Accounting Profession 4

Accountants Need to Develop Critical Thinking

Skills 5

Data Analytics And The AMPS Model 6

The AMPS Model: Ask the Question 8

The AMPS Model: Master the Data

(Chapters 2–4) 9

The AMPS Model: Perform the Analysis (Chapters

5–9) 10

The AMPS Model: Share the Story

(Chapter 10) 10

The Recursive Nature of the AMPS Model 10

Using Visualizations to Analyze Data and

Communicate Results 11

Software Tools Available to Perform Data

Analytics 14

Summary 15

Key Words 15

Answers to Progress Checks 16

Multiple Choice Questions 16

Discussion Questions 18

Brief Exercises 18

Problems 20

Labs Associated with Chapter 1 22

Lab 1-1 Excel: Journal Entries to Trial

Balance 22

Lab 1-2 Excel: Calculating Depreciation Using

Excel Functions 30

Lab 1-3 Excel: Creating a Mortgage

Amortization Schedule 34

Chapter 2

Master the Data: An Introduction to Accounting

Data 42

Data, Data Analytics, and Accounting

Questions! 44

Master The Data: The Second Step of the AMPS

Model 45

What is Big Data? 46

Accounting Data Sources 48

Financial Accounting Data 48

Financial Accounting-Related Data 51

Managerial Accounting Data 54

Tax Data 57

Non-Accounting Data Sources 57

Data Ethics 61

Gathering Data 62

Protecting Data 63

Some Excel Basics: The Pivottable 63

Summary 67

Key Words 67

Answers to Progress Checks 69

Multiple Choice Questions 70

Discussion Questions 71

Brief Exercises 72

Problems 73

Labs Associated with Chapter 2 76

Lab 2-1 Excel: Accounts Receivable Summary

by Customer 76

Lab 2-1 Tableau: Accounts Receivable Summary

by Customer 81

Lab 2-1 Power BI: Accounts Receivable

Summary by Customer 86

Lab 2-2 Excel: Inventory Management by

Customer Profitability 91

Lab 2-2 Tableau: Inventory Management by

Customer Profitability 95

Lab 2-2 Power BI: Inventory Management by

Customer Profitability 99

Lab 2-3 Excel: Inventory Management by SKU

Profitability 102

Lab 2-3 Tableau: Inventory Management by

SKU Profitability 108

Lab 2-3 Power BI: Inventory Management by

SKU Profitability 115

Chapter 3

Master the Data: Data Types Used in

Accounting 120

Examples of Data Types 122

Introduction to Structured Data Types: Categorical

versus Numerical 122

Additional Ways to Categorize Data Based on

Tools 127

Analyzing Data Using Both Categorical and

Numerical Variables in a Pivottable 128

Accounting Data, Data Types, and Accounting

Databases 130

Simplified Product Tables 131

Data Dictionaries and Data Catalogs 132

Summary 134

Key Words 134

Answers to Progress Checks 134

Multiple Choice Questions 135

Discussion Questions 136

Brief Exercises 137

Problems 138

Labs Associated with Chapter 3 139

Lab 3-1 Excel: Identify and Work with Different

Data Types 139

Lab 3-1 Tableau: Identify and work with

Different Data Types 144

Lab 3-1 Power BI: Identify and Work with

Different Data Types 149

Lab 3-2 Excel: Visualize Different Data Types 153

Lab 3-2 Tableau: Visualize Different Data

Types 158

Lab 3-2 Power BI: Visualize Different Data

Types 164

Chapter 4

Master the Data: Preparing Data for Analysis 172

What are the Differences among a Database, Excel,

and Data Visualization Tools (Tableau and Power

BI)? 174

Relational Databases 174

Relational Database Data Dictionaries And Entity-

Relationship Diagrams 176

Relational Database Data Dictionary 176

Relational Database Diagrams 177

Data Storage: Advantages of Using Relational

Databases 178

Data Integrity Benefits of Storing Data in

Relational Databases 178

Internal Control Benefits of Storing Data in

Relational Databases 178

Extract, Transform, and Load: Using Excel, Power

BI, Tableau, and Query Tools to Access Data in

Company Databases 179

Extract, Transform, and Load 179

Extract: Connecting to Data in Excel 180

Extract: Connecting to Data in Tableau 182

Extract: Connecting to Data in Power BI 183

Extract and Transform: Connecting to a Subset of

Data from a Database Using SQL 185

Extract, Transform, and Load: Using Excel Query

Tools to Access Data in Databases External to the

Company 187

Obtaining Data from the Web through Excel 187

Summary 191

Key Words 192

Answers to Progress Checks 192

Multiple Choice Questions 194

Discussion Questions 195

Brief Exercises 195

Problems 197

Labs Associated with Chapter 4 198

Lab 4-1 Excel: Working with Data in Ranges and

Tables 198

Lab 4-2 Excel: Linking Two Tables Using

VLOOKUP for State Tax Rates 205

Lab 4-3 Excel: Linking Two Tables Using

VLOOKUP for Relational Data 209

Lab 4-4 Excel: Linking Tables with a Model 213

Lab 4-4 Tableau: Linking Tables with a

Model 221

Lab 4-4 Power BI: Linking Tables with a

Model 233

Appendix 4A SQL Queries 242

Chapter 5

Perform the Analysis: Types of Data Analytics 256

The Next Step of the AMPS Model: Perform the

Analysis 258

Matching the Analytics with the Accounting

Question 258

Descriptive Analytics 260

Statistical and Summarization Tools for Descriptive

Analytics 260

Examples of Descriptive Analytics 261

Diagnostic Analytics 262

Identify Anomalies/Outliers 262

Finding Previously Unknown Linkages, Patterns, or

Relationships Between and Among Variables 263

Predictive Analytics 264

Prescriptive Analytics 265

Summary of Analyses used to Address Accounting

Questions 267

A Review of Basic Statistics and Hypothesis

Testing 268

Population vs. Sample 268

Parameters vs. Statistics: What Is the Difference? 269

Describing the Sample by Its Central Tendency, the

Middle, or the Most Typical Value 269

Describing the Spread (or Variability) of the Data 269

Probability Distributions 270

Hypothesis Testing 271

Statistical Testing 272

Statistical Test of a Difference of Means of Two

Groups 273

Interpreting the Statistical Output from a Regression 274

Introduction to Tools used in Data Analytics 275

Perform the Analysis Using Microsoft Excel

Tools/Functions 275

The Excel Data Analysis Toolpak 275

Summary 277

Key Words 278

Answers to Progress Checks 279

Multiple Choice Questions 280

Discussion Questions 282

Brief Exercises 282

Problems 283

Labs Associated with Chapter 5 286

Lab 5-1 Excel: Descriptive Statistics for the

Retail Industry 286

Lab 5-1 Tableau: Descriptive Statistics for the

Retail Industry 291

Lab 5-1 Power BI: Descriptive Statistics for the

Retail Industry 295

Lab 5-2 Excel: Using Conditional Formatting to

Perform Bank Reconciliations 299

Chapter 6

Perform the Analysis: Descriptive Analytics 304

Defining Descriptive Analytics 306

Accounting Data used in Descriptive Analytics 306

Tools And Techniques Used In Descriptive Analytics 307

Examples of Descriptive Analytics 309

Descriptive Analytics of Financial Performance

Using Tables and Graphs 309

Considering the Right Comparison Group for

Analysis 310

Descriptive Analysis Using PivotTables and Bar

Charts for Accounts Receivable Aging 311

Horizontal, Vertical, and DuPont Analysis of

Financial Performance 313

Using Descriptive Analytics to Identify Phenomena

That Might Require Additional Analysis, Including

Diagnostic Analytics 317

Summary 319

Key Words 320

Answers to Progress Checks 320

Multiple Choice Questions 321

Discussion Questions 323

Brief Exercises 324

Problems 325

Labs Associated with Chapter 6 328

Lab 6-1 Excel: Accounts Receivable Aging 328

Lab 6-1 Tableau: Accounts Receivable

Aging 334

Lab 6-1 Power BI: Accounts Receivable

Aging 339

Lab 6-2 Excel: Horizontal Analysis of Financial

Performance with Sparklines 345

Lab 6-3 Excel: Vertical Analysis of Financial

Performance (with Sparklines) 351

Lab 6-4 Excel: DuPont Analysis of Financial

Performance 355

Chapter 7

Perform the Analysis: Diagnostic Analytics 360

Defining Diagnostic Analytics 362

Identifying Anomalies and Outliers 363

Diagnostic Analytic Techniques for Identifying

Anomalies and Outliers 364

Finding Previously Unknown Linkages, Patterns, or

Relationships Between and Among Variables 374

Performing Drill-Down, Detailed Analytics 374

Determining Statistical Linkages, Patterns and

Relationships Among Variables Using Statistical

Tools and Techniques 376

Hypothesis Testing Using a Difference in Means 377

Hypothesis Testing Using Regression 378

Summary 380

Key Words 380

Answers to Progress Checks 381

Multiple Choice Questions 382

Discussion Questions 384

Brief Exercises 385

Problems 386

Labs Associated with Chapter 7 388

Lab 7-1 Excel: Test of Separation of Duties 388

Lab 7-1 Tableau: Test of Separation of Duties 392

Lab 7-1 Power BI: Test of Separation of

Duties 394

Lab 7-2 Excel: Days of the Week Journal

Transactions 397

Lab 7-2 Tableau: Days of the Week Journal

Transactions 402

Lab 7-2 Power BI: Days of the Week Journal

Transactions 405

Lab 7-3 Excel: Using the MATCH() Function to

Perform Bank Reconciliations 408

Lab 7-4 Excel: Benford’s Law 413

Lab 7-5 Excel: Fuzzy Matching and Fake

Employees/Vendors 420

Lab 7-6 Excel: Sequence Check: Identifying

Missing Checks 425

Lab 7-7 Excel: Duplicate Payments 429

Lab 7-8 Excel: Looking for Fraud by Examining

Relationships within a Data File: Accounts

Payable Clerks and Company Vendors 434

Lab 7-8 Tableau: Looking for Fraud by

Examining Relationships within a Data

File: Accounts Payable Clerks and

Company Vendors 441

Lab 7-8 Power BI: Looking for Fraud by

Examining Relationships within a Data

File: Accounts Payable Clerks and

Company Vendors 452

Lab 7-9 Excel: Evaluating the Relationship between

Sales and Advertising Expense 458

Chapter 8

Perform the Analysis: Predictive Analytics 464

Introduction to Predictive Analytics 466

Classification 467

Bankruptcy Classification 468

Loan Extension Classification 470

Fraud/No Fraud Classification 472

Regression 473

Base Rates and Base Rate Fallacy 475

Forecasting Future Performance using Time Series

Analysis 476

Predictive Analytics and Hypothesis Testing 478

Predictive Analytics and Machine Learning 479

Summary 480

Key Words 480

Answers to Progress Checks 480

Multiple Choice Questions 481

Discussion Questions 483

Brief Exercises 484

Problems 485

Labs Associated with Chapter 8 487

Lab 8-1 Excel: Predicting Bankruptcy Using

Altman’s Z 487

Lab 8-2 Excel: Classifying Loan Acceptance

Using Lending Club Data 493

Lab 8-3 Excel: Estimating Cost Behavior Using

Regression Analysis 498

Lab 8-4 Excel: Estimating Activity-Based Costing

Drivers Using Regression Analysis 504

Lab 8-5 Excel: Estimating Borrower Interest

Rates Using Regression Analysis with

Lending Club Data 513

Lab 8-6 Excel: Forecasting Future Performance

(Sales and Earnings for IBM and

Netflix) 520

Lab 8-7 Tableau: Forecasting Future

Performance (Sales and Earnings for

IBM and Netflix) 524

Lab 8-8 Power BI: Forecasting Future

Performance (Sales and Earnings for

IBM and Netflix) 532

Chapter 9

Perform the Analysis: Prescriptive Analytics 536

Linking Back to the AMPS Model 538

Definition of Prescriptive Analytics 538

Constraints 539

Changing Conditions 539

Prescriptive Analytics Techniques 539

Marginal Analysis 540

Make-or-Buy Analysis: Making Outsourcing

Decisions 541

Cash Flow Analysis 542

Accounting Rate of Return and Payback Period 542

Net Present Value and Internal Rate of Return 542

Evaluating Future Cash Flows: Net Present Value

and Installment Payments 544

Evaluating Future Cash Flows: Capital Budgeting

and Investment Decisions 546

Goal Seek Analysis 551

Scenario Analysis 552

An Example of Scenario Analysis Using Potential Tax

Rate Scenarios 552

Sensitivity Analysis 553

Optimization 555

Summary 555

Key Words 555

Answers to Progress Checks 556

Multiple Choice Questions 557

Discussion Questions 560

Brief Exercises 560

Problems 562

Labs Associated with Chapter 9 565

Lab 9-1 Excel: Lump Sum or Annuity? 565

Lab 9-2 Excel: Evaluating Investments Using

NPV 570

Lab 9-3 Excel: Capital Budgeting Using NPV 573

Lab 9-4 Excel: Evaluating Investments Using

IRR 576

Lab 9-5 Excel: Capital Budgeting Using IRR 579

Lab 9-6 Excel: Face, Discount, or Premium? 582

Lab 9-7 Excel: What-If Analysis with Goal Seek/

Breakeven 588

Lab 9-8 Excel: What-If Analysis with Goal Seek/

Final Exam Grade 592

Lab 9-9 Excel: What-If Scenario/ Tax Rates 596

Chapter 10

Share the Story 600

The Basics of Data Visualization 602

Visualizing Descriptive Statistics and Analytics 604

Presenting Data in a Dashboard 607

Bar Charts versus Histograms 608

Visualizing Diagnostic Statistics and Analytics:

Outliers and Anomalies 610

Exploratory Diagnostic Analytics Using Data

Visualization 611

Visualizing Predictive Statistics and Analytics 612

Correlation and Regression 612

Forecasting with Time Series Data 613

Visualizing Prescriptive Statistics and Analytics 614

Sensitivity Analysis 614

Breakeven Analysis 615

Communicating your Data with Words: Executive

Summaries and Reports 616

Summary 617

Key Words 617

Answers to Progress Checks 617

Multiple Choice Questions 619

Discussion Questions 620

Brief Exercises 620

Problems 621

Labs Associated with Chapter 10 623

Lab 10-1 Excel: Create a Dashboard Using

PivotTables and Slicers 623

Lab 10-2 Tableau: Create a Dashboard 631

Lab 10-2 Power BI: Create a Dashboard 642

Chapter 11

Capstone Projects Using the AMPS Model 646

Using the AMPS Model to Address Accounting

Questions 648

Application of the AMPS Model to Your Own

Project(S) 648

Project 1: Using the AMPS Model to Address the

Question of Loan Repayment 648

Ask the Question 649

Master the Data 649

Perform the Analysis 652

Share the Story 652

Project 2: Completing Your Own Project Using the

AMPS Model 653

*Chapter 12

Financial Statement Analysis

Define Financial Statement Analysis

Ask the Financial Statement Analysis Question

Master the Data: Data Sources Used in Financial

Statement Analysis

Primary Data Sources for Financial Statement

Analysis

Perform the Analysis

Descriptive Financial Statement Analytics

Examples of Descriptive Analytics in Financial

Statement Analysis: Ratio Analysis

Diagnostic Financial Statement Analytics

Anomalies and Outliers: Comparisons to Appropriate

Benchmarks

Drill-Down Analytics to Determine Relations,

Patterns, and Linkages between Variables

Predictive Financial Statement Analytics

Regression: Predicting Market Valuation of Equity

with Net Income or Operating Cash Flows

Time Series: Predicting Levels of Business

Interruption Loss

Prescriptive Financial Analytics

Relative Market Valuation Based on Valuation of

Other Companies

Discounted Cash Flow Analysis Using Analysts’

Forecasts

Report the Results

Reporting Descriptive and Diagnostic Analytics:

Management Discussion and Analysis of Annual

Report/10-K

Reporting Predictive Analytics: Analysts’ Research

Reports, Revenue, and Earnings Forecasts

Reporting Prescriptive Analytics: Sensitivity

Analysis

Summary

Key Words

Answers to Progress Checks

Multiple Choice Questions

Discussion Questions

Brief Exercises

Problems

Labs Associated with Chapter 12

Lab 12-1 Excel: Descriptive Analytics: Ratio

Analysis

Lab 12-2 Excel: Diagnostic Analytics: Common

Size Financial Statements

Lab 12-3 Excel: Diagnostic Analytics: Find the

Unknown Company

Lab 12-4 Excel: Predictive Analytics: Net Income,

Cash Flows, and Market Value Prediction

Lab 12-5 Excel: Predictive Analytics: Amounts,

Timing, and Uncertainty: Business

Interruption Loss

Lab 12-6 Excel: Prescriptive Analytics: Relative

Market Valuation

Lab 12-7 Excel: Prescriptive Analytics: Valuing

a Company Using Yahoo! Analysts’

Forecasts

Appendix 1: Analysts’ Forecasts and Current

Stock Price for Netflix (as of

3/9/2022)

*Chapter 13

Managerial Accounting Analytics

The Role of the Management Accountant and their

use of Data Analytics

Ask the Managerial Accounting Questions

Master the Data: Data Sources Useful in

Managerial Accounting

Internal Sources of Cost Accounting Data

External Sources of Cost Accounting-Related

Data

Example of Data Sources Needed to Address

Management Questions

Perform the Analysis

Descriptive Managerial Accounting Analytics

Diagnostic Managerial Accounting Analytics

Anomalies and Outliers: Comparisons to Appropriate

Benchmarks

Drill-Down Analytics to Determine Relations,

Patterns, and Linkages between Variables

Examples of Diagnostic Analytics in Managerial

Accounting Analytics to Assess Cost Behavior

Predictive Managerial Accounting Analytics

Managerial Accounting Analytics Using Either

Regression Analysis or Time Series Analysis

Prescriptive Managerial Accounting Analytics

What-If Analysis

Sensitivity Analysis

Optimization

Steps in Setting Up an Optimization Model

Optimizing Price (Pricing Decisions)

Report the Results

Reporting Prescriptive Analytics: Sensitivity

Using a Heat Map to Communicate Sensitivity to

Inputs

Reporting Prescriptive Analytics: Using Conditional

Formatting to Highlight Variances

Summary

Key Words

Answers to Progress Checks

Multiple Choice Questions

Discussion Questions

Brief Exercises

Problems

Labs Associated with Chapter 13

Lab 13-1 Excel: Descriptive Analytics:

Evaluating Inventory Using Inventory

Turnover and Waste

Lab 13-2 Excel: Diagnostic Analytics: Variance

Calculation and Conditional Formatting

Lab 13-3 Excel: Predictive Analytics:

Forecasting Product Demand Using

Time Series Analysis

Lab 13-4 Tableau: Predictive Analytics:

Forecasting Product Demand Using

Time Series Analysis

Lab 13-5 Power BI: Predictive Analytics:

Forecasting Product Demand Using

Time Series Analysis

Lab 13-6 Excel: Predictive Analytics:

Forecasting Product Demand Using

Regression

Lab 13-7 Excel: Prescriptive Analytics:

Profitability Scenarios Using Excel’s

Data Table

Lab 13-8 Excel Prescriptive Analytics: Price

Optimization

*Appendix A

Excel Tutorial (Formatting, Sorting, Filtering,

and PivotTables)

*Appendix B

Tableau Tutorial

*Appendix C

Power BI Desktop

INDEX I

This book is US$10
To get free sample pages OR Buy this book


Share this Book!

Leave a Comment


This site uses Akismet to reduce spam. Learn how your comment data is processed.