Monday, December 15, 2008

Detecting the Madoff Effect: Methodology for Fraud in Hedge Funds

As a result of the recent Bernard Madoff fraud scheme, pension funds and corporate finance managers have been put on the defensive, wondering how to detect this type of “under the radar” deceptive scam. The depth of the fraud in the case of Bernard Madoff and his ability to engage in a $50 billion undetected scheme employing “serial correlation” demonstrated the vulnerability of financial institutions to trusted individuals operating inside the security model. Bernard L. Madoff Investment Securities LLC (“Madoff”) engaged in a ponzi or pyramid fraudulent scheme in which investors were paid interests not from actual investments but from the funds deposited by other investors. But being on the defensive is not the ideal solution, as it forces financial institutions into a reactive mode, always trying to catch-up with the perpetrators, who somehow remain one step ahead. We would like to suggest a more proactive approach and corresponding methodology for detecting fraud in hedge funds.

Madoff was able to hide his scheme using a “serial correlation” reporting scheme. A serial correlation is a term used by MIT professor and hedge fund theorist Andrew Lo to describe the degree to which each month's returns in a fund mirror the results of the month before. Dr. Lo’s theory is that is a hedge fund has a nice smooth line in its rate of return every month. Upon close examination, any variation to the “smoothness” of the line constitutes a red flag, which should be look upon more carefully.

In the last year corporate finance departments, financial institutions, as well as public and private pension fund portfolios have already lost about 33% of their values due to overleveraged investment banks, the housing and credit crises. An effective, (no more than 1 to 3 weeks) and cost efficient proactive methodology to detect the Madoff effect in the hedge funds would be to apply the following methodology in the specified order to the relevant data available to you:

  • Link Analysis – Use link analysis to determine in the network Madoff is categorized by, and create a subset of that network of hedge funds.
  • Predictive Modeling – Use predictive modeling to create a score of all the hedge funds in your subset. Use Madoff’s variables as your training data.
  • Clustering Analysis – Perform a cluster analysis which includes among other variables the predictive score. Since the predictive score is a multidimensional variable when used with one-dimensional or “flat” variables you will obtain a binocular vision (or binocular summation) of your analysis and increase by 1.4 times the ability to detect the serial correlation. See, Improving Search Engine Optimization by Incorporating Predictive Analytics at http://atomai.blogspot.com/2008/12/improving-search-engine-optimization-by.html

    For verification of the analysis you could use the following factors:
  • The reputation of the independent auditors of the hedge fund identified through this methodology;
  • Control Chart using standard deviation of the yearly returns over a 3-5 year period (exclude the current year);
  • The ratio of total number of employees to the total amount of investments.

    Contact Alberto Roldan at atomanalytics@gmail.com or Sean Suskind at seansuskind@gmail.com

3 comments:

Michael B said...

Hi Alberto,

I agree with you that a defensive strategy is undesirable. BM not only had inside information and the contacts to pull this off (I doubt it was a one-man job) he played to the weaknesses of the regulatory watchdogs, the institutional investors and independent high networth individuals who thought they had found someone who could consistently beat the market. Then these investors ran after him like lemmings. How he identified and attracted these investors is worthy of an analysis.

I like your idea of applying analytics to "follow the money trail" to determine where the system broke down and where it is likely to breakdown again.

Probably the most troubling statement in your post relates to the fact that the institutional and pension fund professionals have managed to lose so much and have fail to present a clear plan that shows how they will stop the bleeding win back the public's trust. Even with the enormous resource they have and their investment in technology, this is clear evidence that even the best tools are no substitute for vigilence and sound judgement.

Thanks!

Anonymous said...

Hi Alberto, good to hear from you again.

I like the hypothesis but I'm a bit confused about Dr. Lo's conclusion. Isn't a smooth line a sign of engineered (or fraudulent returns), not a typical sign of a hedge fund? It would seem that Madoff's clients were being told numbers that were not only intuitively (in retrospect, sadly) suspicious but mathematically suspicious.

This reminds me a bit of something my Statistics professor told me when I was getting my Masters: Mendel, the pioneering geneticist, may have faked (let's say 'nudged') his data to produce results that showed his genetic studies in repressed genes were suspiciously close to the predicted numbers.

That is, Mendel did not trust his unsophisticated audience to recognize that anything far outside of say a 25% chance of two recessive genes coming together as predicted was within acceptable guidelines, so he nudged the numbers closer. The problem with Madoff is that his "onlusion" was fundamentally incorrect and the faking just hid it; in Mendel's case, the conclusion is correct and the 'faking' just reinforced the correct conclusion. He was participating in a bit of forgiveable public relations given the newness of his concept and the lack of sophistication of his audience. Or maybe his results really were within bounds - sometimes random samples really do match the known outcome.
http://www.newworldencyclopedia.org/entry/Gregor_Mendel

Unknown said...

Hi..
It is quite possibel to do in the methodology you have suggested; but in case when the books of accounts are inflated with the fictetious numbers; Analytics also fails; but under analytics sanner applications like clustering for different time periods data can be done and acutal deviations could be understood i.w, from which year could be deviations occured; and then CHAID could be applied to diffrent subsets of funds to estimate in which funds the deviation has happned...

humm just comments back

Business Analytics

Business Analytics

About Me

My photo
See my resume at: https://docs.google.com/document/d/1-IonTpDtAgZyp3Pz5GqTJ5NjY0PhvCfJsYAfL1rX8KU/edit?hl=en_USid=1gr_s5GAMafHRjwGbDG_sTWpsl3zybGrvu12il5lRaEw