Catching fraud with Benford's Law

For reasons I can't imagine, digits between 0 and 9 in financial transactions are not equally likely, even in the later digits.  They follow a distribution described by "Benford's Law," and any deviation from this pattern may signal fraud.  In the linked article, an overabundance of "4s" raised suspicions that operators with the authority issue cash refunds up to $50 without a supervisor's OK were issuing fraudulent refunds to co-conspirators.  But even without the special problem of concentrating on fake numbers just under $50, numbers randomly chosen by fraudsters will produce a pattern unlike that of natural financial digit distribution.

1 comment:

J Melcher said...

A number of tells exist.

True distributions tend to approxiamate a theoretical "Zipf distribution" where the top value is twice as frequent as the next most common, while that is twice a common as the third, etc. Fraudsters tend to flatten this curve.

"Streaks" do tend to occur in nature, but fraudsters tend to avoid them, -- reasoning, (if you can call it that) that a "streak" of say 7 heads in a row is suspicious. In fact the absence of a streak is more suspicious.

That said, a "streak" where time after time, a measurement is JUST BARELY acceptable indicates fudging.