Joined: Oct 8, 2010
  • Threads: 6
  • Posts: 207
October 13th, 2010 at 1:19:07 PM permalink
I just made a quick little program in java to simulate a betting strat. It is extremely basic, just pass line bets, full odds, and then come bets with full odds. Most of my simulations I did with 3/4/5 odds.

My reason for creating the simulation was to see if the swings I saw in the last times I've played are actually common or not and what are the odds that I go up to 10 tables in a row with $400 and never double my money at a single one.

I initially had 2 conditions on the trials. I would start with $400 and stop at $1200. I played around with the numbers and in the end it was coming out that most of my trials were winning. Granted the house odds are very low on this strat over the long run, but this was just surprising to me. Most trials I did were 1 million shooters, a few at 10 million, and 1 at 1 billion. I'd go higher but I need to look up how to even create a static long of 200 billion in java. :) I did remove the stop limit at the end and just let it ran. Previous trials simply added up all the money I left the table with and compared it with what I bet. The later ones just kept a running bankroll.

Now my question. It has been a long time since I took statistics. What is a reasonable sample size for this simulation? Should I be doing number of dice rolls vs shooters?

At this point I'm just trying to verify my simulation. I simply don't trust the numbers I'm getting which leads me to think I've got some sort of logic issue. If someone wanted the code I'm happy to post that too, it is about 350 lines long.
Joined: Nov 30, 2009
  • Threads: 86
  • Posts: 555
October 14th, 2010 at 9:24:13 PM permalink
A million trials is a lot and any negative EV should manifest itself within it. Perhaps you don't have a good random number generator? Or maybe your strategy is playing off of some entropy built into the RNG, so that your betting style changes against numbers that haven't manifested themselves over time. Hard to say.

If the strategy is that good, you can always try it out in real life. I suspect a real life test would prove faulty at some point.

But then again, if your program is designed to quit after the first win and martingale out losses starting on a $5 could be betting the passline/don't pass 5, 10, 20, 40, 80, and 160. What are the odds of losing 6 straight pass line bets?

I think some key considerations are:

1.) how does the variance on your bankroll look with your strategy?
2.) what kind of loss limits and win limits are built into your strategy? I would expect the above sample to actually have some decent results.
3.) from the more technical and less intuitive perspective, how is your RNG?
4.) did you simulate dice rolls instead of shooters?

Lastly, I don't think any system would survive 1mm rolls. 1bn is an over the top test, but 1mm could exploit RNG issues. If you think your system is that good, try it out in real life. Nothing beats experience right?

  • Jump to: