What is Benford's Law?
Definition: A statistical pattern in which smaller leading digits occur more often than larger ones in many naturally occurring sets of numbers, used as a screening test for unusual or manipulated figures.
Benford's Law is the observation that, in many collections of numbers that arise from real-world processes, the first (leftmost) digit is not spread evenly across 1 to 9. The digit 1 leads roughly 30 percent of the time, while 9 leads fewer than 5 percent of the time. The pattern is named after the physicist Frank Benford, who documented it in 1938 across many unrelated data sets, and it had been noticed earlier by the astronomer Simon Newcomb in the 1880s. It is also called the first-digit law or the law of anomalous numbers.
The law matters in finance because invented or altered numbers tend to look different from genuine ones. People who fabricate figures usually spread leading digits too evenly or favour particular values. Forensic accountants, internal and external auditors and tax authorities therefore compare the leading-digit frequencies of a ledger with the Benford pattern as a first screen, to decide where to look more closely. A departure is a prompt for questions, never proof of wrongdoing.
The formula
The expected probability that the leading digit equals is:
- is the expected share of numbers whose first digit is , written as a decimal.
- is a digit from 1 to 9.
- is the base-ten logarithm.
Applying the formula gives the expected shares: 1 is 30.1 percent, 2 is 17.6, 3 is 12.5, 4 is 9.7, 5 is 7.9, 6 is 6.7, 7 is 5.8, 8 is 5.1 and 9 is 4.6 percent. The nine values add up to exactly 100 percent. For the digit 1 the working is . For the digit 9 it is .
The law extends to the second digit, the first two digits together and the last two digits. Second-digit shares are much flatter (0 leads that position about 12 percent of the time, 9 about 8.5 percent), which is why auditors often run several digit tests rather than one.
Worked example
Suppose an auditor tests 500 supplier invoices from one year. Under the law the expected count of invoices starting with 1 is , about 150. If only 100 invoices start with 1, the observed share is percent, well below the expected 30.1 percent. If a large number of invoices start with 4 or 5, perhaps because amounts were set just under an approval limit, the excess shows up as a spike on a digit chart. The auditor would then sample those invoices and check them against purchase orders and receipts.
Analysts judge the gap with a significance test such as a chi-square test or a z-statistic for each digit, because even genuine data deviate somewhat by chance, and large samples make small gaps statistically significant. The test statistic is a screening aid: it ranks which digits, accounts or vendors deserve attention.
Where it is used
The accountant Mark Nigrini popularised the law in auditing during the 1990s, and it is now built into audit analytics software. Typical uses include:
- Testing accounts payable and expense data for invented or split transactions, one route into occupational fraud and embezzlement.
- Screening journal entries and revenue figures for signs of financial statement fraud, alongside journal entry testing.
- Reviewing tax returns and declared deductions for patterns that may indicate tax evasion.
- Supporting wider forensic accounting techniques and anomaly detection programmes in banks.
When the law applies and when it does not
The law holds best for data that span several orders of magnitude, are not capped or floored, and are generated by a mixture of processes, such as invoice amounts, populations, transaction values and accounting balances. It fits poorly in several situations:
- Small samples, where random variation alone produces large digit gaps.
- Assigned or sequential numbers such as invoice numbers, account numbers, telephone numbers and postal codes.
- Data with a built-in minimum, maximum or fixed price, such as a standard fee or a narrow range of salaries.
- Data rounded to convention, such as amounts quoted in round hundreds.
A genuine data set can therefore fail the test for innocent reasons, and a careful fraudster who knows the law can imitate it. Conformity does not show that figures are genuine. For this reason the test is one signal among many in an audit risk assessment and in a wider fraud detection programme, and its results are followed up with documentary evidence rather than reported on their own. The Association of Certified Fraud Examiners includes digital analysis of this kind in its training.
The law is explained by scale invariance: if a set of numbers follows the pattern, changing its currency or unit leaves the pattern intact. It also reflects the way quantities grow multiplicatively, which links it to the probability distributions used in finance and to the lognormal distribution.
Ref 1010001601