In accounting fraud, by definition, a fraudster has to invent numbers to hide profits or losses.
Based on our research, any successful algorithm attempting to detect accounting fraud must analyze accounting data based on the first principle. But what is the first principle? According to Wikipedia, a first principle is a fundamental proposition that cannot be deduced any further.
A simplified example: let's say an accounting fraudster has booked only two amounts: 100 and 200—the total amounts to 300. However, to catch the fraudster, we must analyze the transactions at the lowest level possible (i.e., 100 and 200). That's the first principle in accounting fraud detection.
" In analyzing accounting fraud, the first principle states that we must investigate accounting transactions at the lowest possible level."
In Data Science lingo, we call data based on the first principle data, which is based on the "most granular level" (i.e., 100 and 200). “Aggregated data” would refer to 300 in our example.
In recent years, Elon Musk popularized the first principle approach, but the concept dates back at least 2,000 years. Aristotle defined the first principle as "the first basis from which a thing is known." Marcus Aurelius had a similar point in "What is its nature?"
The first principle is also why only auditors can catch accounting fraudsters in almost all cases. Public data is often aggregated, such as in the case of Madoff's monthly returns. Aggregated data, such as Madoff's monthly returns, violates the first principle in accounting fraud detection.
For example, we've seen several analysts pondering over the monthly returns of Madoff and trying to detect fraudulent behavior. However, in our view, that doesn't work as it violates (our) first principle in accounting fraud. Namely, monthly returns such as January 2.5% and February 0.75% are aggregated numbers.
The point is so crucial that we had to illustrate it. Based on the first principle in accounting fraud, we cannot analyze aggregated data (e.g., January 2.5% return), but need the transactions at the lowest level (first principle), such as illustrated here:
Therefore, based on our research, detecting Madoff's Ponzi Scheme based on publicly available and aggregated data was nearly impossible.
Since, by definition, a fraudster has to invent numbers to hide profits or losses, the first principle in accounting fraud states that we have to analyze accounting data at the lowest granular level possible. Based on our research, we must adhere to the first principle. If we don't, we cannot catch the accounting fraudster.
June 8, 2023 Franco Arda
PS: I don't think we need to prove scientifically that there's an infinite large difference between granular and aggregated data for an algorithm.