Optimal Wagering

                          Optimal Wagering 

                     Copyright 1991, Michael Hall 

               Permission to repost, print for own use. 

 

I think I've got some good discoveries here... even if you don't 

follow the math, you can get some useful blackjack information here. 

 

The question of optimal wagering has been brewing on rec.gambling 

for a while.  I rephrase this question as the following: 

 

* What's the optimal win per hand as a portion of bankroll and 

  what is the betting pattern necessary for this? 

 

That is, we want to maximize E/a' where E is the win per hand 

and a' is the required bankroll. 

 

E is simply defined by: 

 

E=sum{WiPiEi} 

 

where i is the situation 

     Wi is the wager for that situation 

     Pi is the probability of that situation 

     Ei is the expected value of that situation 

 

I defined a' in previous articles.  Unfortunately, I made a 

slight error, in that I left out a couple of sqrt's.  I 

hope the following is correct... 

 

         log((1/R) - 1) 

a'= ----------------------------(sqrt(s^2 + E^2)) 

        /sqrt(s^2 + E^2) + E\ 

    log| ------------------- | 

        \sqrt(s^2 + E^2) - E/ 

 

where R is the risk of ruin 

      E is the win per hand 

      s^2 is the variance of E 

      a' is the necessary units of blackjack bankroll 

 

[Incidentally, the Kelly criterion leads to a bankroll formula 

 proportional to the one above, and so Kelly betting produces the 

 same optimal wagering schemes as the ones shown below.] 

 

I tried to maximize E/a' by taking the derivatives wrt Wi and setting 

them to 0.  That got really ugly.  Then I tried to maximize E or minimize 

R using various formulations of Lagrange multipliers.  That got really ugly  

too.  I did come up with the partial derivatives, which are ugly themselves, 

but solving for the Wi's is where it gets *really* ugly. 

 

So I gave up and just wrote a program to evaluate the function given 

Wi's as input, and then I wrote a program to do a simple hill-climbing 

on this function in the space of integers between 1 and some maximum 

bet like 4 or 8.  My intuition is that hill-climbing should converge to 

the global maximum and not a local maximum of this function, but I don't 

have any proof of this.  BTW: my program does adjust for the basic 

variance of blackjack, increasing the effective bet size by 1.1 and other 

such things. 

 

For a downtown Vegas single deck 75% penetration (Snyder's tables in 

"Fundamentals of Blackjack" by Chambliss and Rogenski), here is the 

optimal betting patterns I found for spreads of 1-2, 1-4 and 1-8: 

 

     SINGLE DECK 

    DOWNTOWN VEGAS 

 

                    1-2   1-4   1-8  

  ADV  FREQ  HI-LO  BET   BET   BET 

   Ei   Pi          Wi    Wi    Wi 

-.026  .065   -5     1     1     1 

-.021  .030   -4     1     1     1 

-.016  .055   -3     1     1     1 

-.011  .070   -2     1     1     1 

-.006  .100   -1     1     1     1 

-.001  .200    0     1     1     1 

+.004  .095   +1     1     1     1 

+.009  .075   +2     1     1     2 

+.014  .050   +3     2     2     3 

+.019  .045   +4     2     3     5 

+.024  .040   +5     2     4     6 

+.029  .035   +6     2     4     7 

+.034  .030   +7     2     4     8 

+.039  .030   +8     2     4     8 

+.044  .080   +9     2     4     8 

 

The Hi-Lo column shows the approximate High-Low (or Hi-Opt I) count for 

each advantage, though you should adjust for the extra advantage from 

strategy deeper into the deck.  Note that the bet should not be raised 

until a true count of 3, unless you are using a very wide spread.  

You might fool a few pit critters by your low bet at a true count of 2. 

(Or at least you won't get nailed when you increase your bet at a true 

count of 2, like I did once.)  For the 1-2 and 1-4 spreads, the betting 

pattern is easy to remember - true count minus 1 (minimum of 1, maximum 

of 2 or 4.) [More exact results using simulations for the input data 

showed that the optimal spread for Hi-Lo here is actually to bet equal 

to the true count.] 

 

Here's the same stuff, but for 2 decks: 

 

     DOUBLE DECK 

  (BSE of -0.2% assumed) 

 

                    1-4   1-8   1-16 

  ADV  FREQ  HI-LO  BET   BET   BET 

   Ei   Pi          Wi    Wi    Wi 

-.027  .060   -5     1     1     1 

-.022  .040   -4     1     1     1 

-.017  .060   -3     1     1     1 

-.012  .080   -2     1     1     1 

-.007  .110   -1     1     1     1 

-.002  .200    0     1     1     1 

+.003  .110   +1     1     1     2 

+.008  .085   +2     3     3     5 

+.013  .055   +3     4     5     8 

+.018  .045   +4     4     7    11 

+.023  .040   +5     4     8    14 

+.028  .030   +6     4     8    16 

+.033  .025   +7     4     8    16 

+.038  .020   +8     4     8    16 

+.043  .040   +9     4     8    16 

 

 

 

Here's the same stuff, but for 8 decks: 

 

      EIGHT DECKS 

(NEGATIVE COUNTS PLAYED) 

                    1-8   1-16   1-32 

  ADV  FREQ  HI-LO  BET   BET    BET 

   Ei   Pi          Wi    Wi     Wi 

-.030  .010   -5     1     1      1 

-.025  .010   -4     1     1      1 

-.020  .020   -3     1     1      1 

-.015  .060   -2     1     1      1 

-.010  .130   -1     1     1      1 

-.005  .510    0     1     1      1 

 .000  .130   +1     1     1      1 

+.005  .060   +2     8     8     10 

+.010  .030   +3     8    15     20 

+.015  .015   +4     8    16     30 

+.020  .010   +5     8    16     32 

+.025  .010   +6     8    16     32 

+.030  .005   +7     8    16     32 

 

 

      EIGHT DECKS 

(NEGATIVE COUNTS NOT PLAYED) 

                    0-8   0-16   0-32 

  ADV  FREQ  HI-LO  BET   BET    BET 

   Ei   Pi          Wi    Wi     Wi 

-.030  .010   -5     0     0      0 

-.025  .010   -4     0     0      0 

-.020  .020   -3     0     0      0 

-.015  .060   -2     0     0      0 

-.010  .130   -1     0     0      0 

-.005  .510    0     1     1      1 

 .000  .130   +1     1     1      1 

+.005  .060   +2     4     5      8 

+.010  .030   +3     8    10     16 

+.015  .015   +4     8    15     24 

+.020  .010   +5     8    16     31 

+.025  .010   +6     8    16     32 

+.030  .005   +7     8    16     32 

 

What follows are statistics on all these different optimal spreads. 

The bankroll requirements assume we want to have a 20% chance of 

losing *half* the bankroll before winning *half* the bankroll. 

One you lose half the bankroll, I'd advise cutting the bet size 

in half.  (Note that the desired risk of ruin has absolutely no effect 

on the optimal betting pattern - it just changes the bankroll 

by a constant amount.) 

 

                                       UNIT^2    UNITS 

              % BANK GAIN  UNIT GAIN  VARIANCE  REQUIRED 

               PER HAND     PER HAND  PER HAND  BANKROLL 

 DECKS SPREAD|  E/(2a')        E        s^2       2*a'    

-------------*-------------------------------------------- 

1-Deck FLAT  |.001420%      .0050?      1.27      352 

1-Deck 1-2   |.008027%      .0165       2.47      206 

1-Deck 1-4   |.014170%      .0348       6.16      245 

1-Deck 1-8   |.018132%      .0695      19.19      383 

2-Deck 1-4   |.002765%      .0170       7.55      615 

2-Deck 1-8   |.006787%      .0433      19.92      638 

2-Deck 1-16  |.009916%      .0946      65.16      955 

8-Deck 1-8   |.000251%      .0064      11.77     2550 

8-Deck 1-16  |.000673%      .0162      28.00     2401 

8-Deck 1-32  |.001033%      .0328      75.24     3177 

8-Deck 0-8   |.000675%      .0086       7.82     1263 

8-Deck 0-16  |.001047%      .0169      19.33     1600 

8-Deck 0-32  |.001288%      .0326      59.57     2532 

 

 

Some things to conclude, given the above table: 

 

  * A 1-2 spread on a single deck is more than 6 times more profitable 

    than a 0-32 spread on 8 decks!  Even flat betting a single deck 

    is probably better. 8 decks stink! 

 

  * It takes a 1-16 spread on double decks to beat a 1-2 spread on single 

    decks!  (Can this be true?) 

 

  * A 1-8 spread buys you 29% more income over a 1-4 spread on 

    a single deck, but you'll probably lose more than that from 

    the extra countermeasures. 

 

  * Given a $6,125 bankroll, you could spread $25-$100 on a single 

    deck, making $86.8/hour (.014170%*6125*100).  This is probably 

    overly optimistic, since it rare that you can freely spread 

    1-4 on a 75% penetration downtown Vegas game. 

 

  * You need about a 1-32 spread on 8 decks before you can get away 

    with playing through negative counts.  A 1-8 spread gets killed 

    sitting through negative counts, as the high bankroll requirement 

    shows. 

 

One thing that might be fun is playing around with the above 

betting spreads.  They are optimal, but how weird can you get 

without sacrificing much of the E/a'? 

 

I'd like to acknowledge Blair for getting me to think in terms of 

percent bankroll win. 

 


Comments

Popular posts from this blog

BOTTOM LIVE script

Evidence supporting quantum information processing in animals

ARMIES OF CHAOS