Where most financial experts around the world stress the importance of establishing fraud detection as a key audit standard, many companies are affected by their inability to process the unfathomably large quantities of accounting journal entries and extract useful data that not only highlights fraudulent attempts in the company’s journals but also leads to elimination of such practices from the workplace. In the accounting world, it can indeed be very challenging for a company to dig massive amount of data and highlight the fraudulent activities that are nicely covered up. The good news is that there are indeed ways of achieving the impossible through a process known as data mining.
Understanding Data Mining:
Data mining is the process of digging into a large collection of data and identifying previously unknown patterns of potentially useful information. Data mining is a scientific and unconventional discovery of knowledge and information buried deep within piles of data. By applying data mining techniques to accounting entries, you can successfully uncover fraudulent entries made in the journals by establishing patterns that are otherwise unavailable.
Data is a packet of information that encloses a fact, number or text. Digital data can be processed by a computer. Over time, organizations tend to accumulate large amounts of data in different formats and databases. Data is broadly grouped into three categories:
- Transactional or operational data: company costs and sales, payroll, accounting and inventory
- Non-operational data: Data forecasts, macro-economic data, industry sales
- Metadata: Data about the data itself, for instance, design element of the database
Patterns and associations found in these components of a company data can provide useful information. For instance, by applying data mining techniques, a company can obtain historical trends and make predictions about the future.
What is Fraud?
Fraud comprises theft of assets as well as any attempt(s) made to conceal theft. While a physical theft of assets is usually revealed through routine check and balance activities, it is the act of concealment that distinguishes fraud from ordinary theft. Owing to concealment of theft, the perpetrator is likely to commit the same crime over and over.
Precisely, there are three factors that lead to fraudulent behavior of an employee:
- Sensed, yet undisclosed financial need
- Perceived opportunity to carry out misappropriation of cash or company data
- Justification of fraud and misconduct lucidly derived
While there are many different types of fraud, ranging from healthcare fraud, mortgage fraud and commodities fraud to management fraud schemes, accounting manipulation typically originates from the company’s top management in an attempt to mislead the investors.
Understanding Fraudulent Practices through Accounting Manipulation
Fraudulent schemes that have involved accounting manipulation include:
Bill and hold: In such scheme, company bills a customer for a sale, but in the next account period, reverses the sale of goods reporting a customer return
Keeping accounting journals open at the end of current period: The next accounting period’s sales are registered in the current period, thereby showing inflated revenue in the current period and understated revenue in the upcoming period
Showing fictitious sales: This will show inflated revenue.
Delaying customer returns on company ledgers: When inventory is returned, revenue recorded against those sales must also be reversed. By delaying the record of returns until the next period, higher revenue is reported in the current period.
In addition to inflating revenue, the management may also understate expenses through understatement of cost of goods sold, cost-capitalization and reduction of depreciation expense.
Other types of accounting fraud include balance sheet fraud that aims at showing lesser debt and liabilities than the company’s actual commitment.
Applying Data Mining to Detect Accounting Fraud
As discussed earlier, data mining is applied with the objective of seeking out trends and patterns in accounting data that reveal fraud.
Business organizations typically cover the following steps to apply data mining for fraud detection:
Analyze objectives of fraud, as these will be used as data mining objectives
- Collect and understand data
- Process and refine data and prepare for algorithms
- Experiment design
- Evaluate the results
- Classification of Data Mining Methods for Detecting Accounting Fraud
Several data mining techniques are available for a business to detect accounting fraud. The exact data mining technique will vary depending on the type of underlying fraud. Most data miners use a combination of different data mining techniques to achieve their objectives. Some techniques rely on statistical values whereas others are based on artificial intelligence.
Regression-based models are most commonly used for detecting accounting fraud. Most models are based on multi-criteria decision making-method and logistic regression.
A logistic model is used to identify insurance and management fraud that may include attempts to show understatement of expense or overstatement of revenue. A logistic regression based model can also predict financial statement fraud. Other methods of finding out financial statement fraud, such as balance sheet fraud include qualitative response model that uses Probit and Logit techniques.
A Cascaded Logit model is used to investigate association between insider trading and fraud possibility.
Other regression based methods, such as statistical regression analysis are used to determine if an independent audit committee is likely to detect and reduce fraud.
Regression analysis based on Logit model can predict financial fraud through empirical analysis of financial indexes.
A joint logistic- and clustering analysis is used to detect fraudulent behavior from four perspectives: company management, financial indexes, risk and trading.
Neural Networks: A neural network mimics the human brain. The benefits of using a neural network include its adaptability and robustness. Neural networks are used to detect insurance, credit card and corporate management fraud.
Bayesian Belief Network (BBN):BBN is helpful in developing models for corporate fraud detection.
Decision Trees: It is a decision support tool that is structured in the form of a tree. It is a predictive tool used for eliminating insurance, credit card and management fraud.
Expert System: An expert system increases the chances of successful fraud detection. With an expert system, the auditors can detect several different levels of management fraud.
For instance, an expert system, such as Computer Assisted Auditing Tool (CAAT) can assist auditors in handling complex transactions quickly.
A CAAT may be employed for purposes, such as:
- Checking accounting journal entries and balances on large data systems
- Finding patterns of inconsistency or fluctuation in the findings
- Rechecking the results products from other statistical data mining techniques
- How effective is Data Mining in Detecting Fraud?
Data mining is far from being a perfect fraud detection tool. Data mining techniques basically aim at finding data outliers. However, not every outlier is necessarily a fraud. Likewise, a non-outlier may also be a fraudulent entry. These are some of the challenges that data mining still faces. That said it’s still safe to regard data mining as the most effective accounting fraud detection mechanism especially for large volumes of data are involved.
Some indications of fraudulent activity in your accounting data are as follows:
If individual variables display usual data but group of variables start displaying unusual patterns, then it is likely to be a fraudulent activity
In some transactions, the combination of accounting entries yields unusual values when compared to the reference values. Such transactions are also likely to reveal fraud.
If an individual variable displays unusually low or high value, then it is not necessarily fraud, and may also be a data entry error. Statistical tools, such as range, relative size factor and standard deviation suffice in detecting fraud in such transactions
In some cases, two or more entries are accounting for more than once, especially if the entries are lying in two different formats. Statistical methods cannot reveal such duplicate entries from being accounted, and the multiple accounting of a single entry may also lead to fraud. This is one incident where data mining technique, such as cluster analysis may be required
If two financial records that are not related to each other display a value that represents a link between the two records, it may indicate a fraud. Un-supervised data mining and link analysis may be applied.
While every data mining technique has its own merits, a hybrid model that involves two or more data mining techniques is recommended by data miners as a more effective tool for combating fraud.