Yesterday’s post was an April Fools' joke. However, the simulation results for the casino dice game of craps were real. As commenter Patrick figured out, the catch was that the strategy called for betting negative amounts.
The “system” starts out with $100 bets and goes up by a dollar after a win and down by a dollar after a loss. However, we will lose slightly more often than win. So, the bet amount will keep going down until it hits zero and then become negative.
Of course, casinos won’t allow gamblers to bet negative amounts. This is effectively like reversing the role between casino and gambler. It’s no wonder that a gambler who uses the system starts out losing while the bets are positive and ends up winning after the bets become negative. The only way I know to beat a casino at craps is to get them to gamble at your craps table.
This kind of hidden problem with experiments and simulations doesn’t just appear in cooked-up April Fools' jokes. A study by Schleef and Eisinger on asset allocation schemes concluded that investors should increase their stock allocations over time. However, this conclusion was just an artifact of a problem with how their simulations sampled historical returns. Poorly constructed studies can produce all kinds of crazy results.