http://diskworld.wharton.upenn.edu/~foster/teaching/701/class_ponzi_1.html
Statistics 701: Ponzironi
Our goal: fake a Ponzi scheme (legally?)
First story: Math league
* send out 64 emails: 32 forcasting market up 32 forcasting market down
* send out 32 emails to those who we got right in first rund: 16 up 16 down
* send out 16 emails: 8 up 8 down
* send out 8 emails: 4 up 3 down
* send out 4 emails: 2 up 2 down
* send out 2 emails: 1 up 1 down
* Now send out email asking for money for next forecast
Second story: starting many Hedge funds
* Start 64 hedge funds: 32 leverage market up, 32 leverage market fall
* Fold losing 32
* Of 32 winners: 16 leverage market up, 16 leverage market fall
* Fold losing 16
* ...
* Last 6 month track record: doubling every month
* Show track record to very rich person--ask for one million dollar
investment
Third story: Search through database
* Start with 64 rules for picking a stock (think leveraged)
* Each month or two kill off the bottom 1/2 of the rules
* After a year or so, you have one rule that has grown spictularilly
How big might the t-statistic be following such a scheme?
* Suppose you do n different tests
* Bonferonni p-value = n * regular p-value
* If JMP won't do it for you, use sqrt(2 log(n)) for significance
* If you require Bonferonni significance, you will rarely fall into the
trap of 3rd story.
