Maximum possible edge in Blackjack

Hi all,
I was wondering what would be your maximum house edge if hypothetically you had a computer play perfectly based on the composition of the deck? i.e. not using basic strategy but combinatorics to get the optimal play for each hand (Ex. Wizard's hand calculator ). The wizard of odds has optimal house edge in those conditions if the deck is reshuffled after each hand, but I couldn't find any stats on standard games (of course it'd depend on the penetration)
The game I commonly play is 6 deck S17 late surr with a house edge of 0.36%. Would it be possible to overcome that house edge with a fixed bet in that case?
How about if also the bets are sized perfectly according to the kelly criterion?
PS: I love this forum and this is my first post here, as surprisingly I couldn't find any discussion on this subject, probably because it's just a hypothetical but I thought it is potentially useful for online live dealers for example. If there was a discussion about it I didn't find please let me know!

Quote: tyler498

Hi all,
I was wondering what would be your maximum house edge if hypothetically you had a computer play perfectly based on the composition of the deck? i.e. not using basic strategy but combinatorics to get the optimal play for each hand (Ex. Wizard's hand calculator ). The wizard of odds has optimal house edge in those conditions if the deck is reshuffled after each hand, but I couldn't find any stats on standard games (of course it'd depend on the penetration)
The game I commonly play is 6 deck S17 late surr with a house edge of 0.36%. Would it be possible to overcome that house edge with a fixed bet in that case?
How about if also the bets are sized perfectly according to the kelly criterion?
PS: I love this forum and this is my first post here, as surprisingly I couldn't find any discussion on this subject, probably because it's just a hypothetical but I thought it is potentially useful for online live dealers for example. If there was a discussion about it I didn't find please let me know!

Hi Welcome, new member.
Wizard's calculations and basic strategy cards have to make a few assumptions and basically assume infinite deck. Some other strategy cards are based on massive sample sizes of computerised play. So often you will see a few hundredths of a percentage difference between such strategies for the same rules.
I have experimented with continuously evaluated strategy based on analysis of actual cards remaining in an 8 deck shoe. That continuous evaulation would occasionally vary from accepted basic strategy, e.g when Aces are depleted or plentiful. Such computer evaluation achieves the same sort of improvement as card counting on steroids and would be more perfect than even the best card counting system. However, that evaluation, coupled with flat betting (not even wonging in/out) would yield maybe a tenth of a percent reduction in edge. That sort of evaluation combined with aggressive wonging, would beat the game, quite considerably, just as card counting and flat bet wonging could.
The best case margin and hourly rate is very low however, and getting your ass banned from the casino would be inevitable. Technically and legally, computer assisted play is cheating and deemed illegal in most domains.

Quote: OnceDear

Hi Welcome, new member.
Wizard's calculations and basic strategy cards have to make a few assumptions and basically assume infinite deck. Some other strategy cards are based on massive sample sizes of computerised play. So often you will see a few hundredths of a percentage difference between such strategies for the same rules.
I have experimented with continuously evaluated strategy based on analysis of actual cards remaining in an 8 deck shoe. That continuous evaulation would occasionally vary from accepted basic strategy, e.g when Aces are depleted or plentiful. Such computer evaluation achieves the same sort of improvement as card counting on steroids and would be more perfect than even the best card counting system. However, that evaluation, coupled with flat betting (not even wonging in/out) would yield maybe a tenth of a percent reduction in edge. That sort of evaluation combined with aggressive wonging, would beat the game, quite considerably, just as card counting and flat bet wonging could.
The best case margin and hourly rate is very low however, and getting your ass banned from the casino would be inevitable. Technically and legally, computer assisted play is cheating and deemed illegal in most domains.

Hi OnceDear, thanks for the reply! it's what I wanted to know. I know it's cheating and honestly wasn't considering it ( I can't even run a simulation to answer my question...) but I was curious to know how much you lose by having "one basic strategy" chart for any deck composition. And one-tenth of a percent in house edge is surprisingly low! Thanks again

Agreed Once Dear.
Your experiment is interesting. In developing A5 in the 80's using a marginal computer by today's standards, Flat-betting A5 reduced the House edge of S17, 6-deck, 70%pen, DD any 2, DAS, late surr by about half ( from -0.34% to -0.18%). Yes Aces and Fives are important in Blackjack. With proper wagering schemes, (wonging, and bet-ramp) a decent 1/2% PA is possible, though not for the faint of heart. Typically 1/5% is achievable.

Regards
98

Quote: tyler498

I was wondering what would be your maximum house edge if hypothetically you had a computer play perfectly based on the composition of the deck?

I don't recall an answer for this.

However, something I read a long time ago said 60% winning sessions was possible for some counting strategies, and computer-assisted perfect play could have 80% winning sessions. Don't have a reference.

Even with computer shoes people still lost tons. Plenty of big named teams ran "shoe teams" with computers when they were legal (or even after) and lost due to the regular variance of the game. I wouldn't think even with computer assistance one could win 80% of their sessions.

The maximum possible edge in blackjack doesn't come from basic strategy or card counting, but other techniques which go far and beyond. The max edge one could get from basic strategy is to get the game about break even (assuming good game, decent comps, counting comps towards winnings, etc). The max edge one could get from traditional lone wolf card counting could be around 2-3% given special conditions of wonging, penetration, etc... This wouldn't be something that would last very long or be real life applicable now days in my opinion. Next, you'd need to look at the big player approach with spotters and a team. Assuming you have the spotters playing a break even game, you still only get about a 3% advantage on the big player, but he's playing big money, so the EV returns are much higher than a lone wolf approach.

Past that there are a slue of other techniques, such as next carding, hole carding, edge sorting, shuffle tracking, ace sequencing, specific card cutting/steering, etc, etc. The edges for these other techniques can range from 10% to well over that given the different scenarios/situations, though they are found less frequently than any old game you can sit down and count at.

Are we talking about your edge on any given hand or an overall edge?

Quote: Ibeatyouraces

Are we talking about your edge on any given hand or an overall edge?

I am talking about overall edge.

Quote: Romes

Even with computer shoes people still lost tons. Plenty of big named teams ran "shoe teams" with computers when they were legal (or even after) and lost due to the regular variance of the game. I wouldn't think even with computer assistance one could win 80% of their sessions.

The maximum possible edge in blackjack doesn't come from basic strategy or card counting, but other techniques which go far and beyond. The max edge one could get from basic strategy is to get the game about break even (assuming good game, decent comps, counting comps towards winnings, etc). The max edge one could get from traditional lone wolf card counting could be around 2-3% given special conditions of wonging, penetration, etc... This wouldn't be something that would last very long or be real life applicable now days in my opinion. Next, you'd need to look at the big player approach with spotters and a team. Assuming you have the spotters playing a break even game, you still only get about a 3% advantage on the big player, but he's playing big money, so the EV returns are much higher than a lone wolf approach.

Past that there are a slue of other techniques, such as next carding, hole carding, edge sorting, shuffle tracking, ace sequencing, specific card cutting/steering, etc, etc. The edges for these other techniques can range from 10% to well over that given the different scenarios/situations, though they are found less frequently than any old game you can sit down and count at.

I totally agree, I guess I should've been more specific but I meant not including those other techniques, although it is true they would provide a bigger edge.
Was I'm wondering is, for instance if you see the Wizard's house edge calculation at WoO, I quote the methodology: "The "basic strategy with cut card" results are based on total dependent basic strategy, like the tables on this site, and the use of a cut card, which favors the dealer. "
What would the house edge be if someone ran a simulation which instead of playing by the basic strategy table, made a call to the WoO hand calculator (i.e. perfect composition dependent strategy, but not reshuffled every hand, played throughout the shoe). This is one stat I have a hard time finding.

Quote: tyler498

...This is one stat I have a hard time finding.

This shouldn't be 'that' difficult to find... You should do some research on the shoe computers and read about their origins/stories. Along the way there should be plenty of math talking about their edge/etc. The only issue here is the computers played perfect composition strategy, but they also fluctuated their bets with regards to the composition/count and then played perfectly (of course) as well. I'm TRYING to remember the edge obtained by these 'perfect composition' playing machines, but I can't recall certain enough to post. I'm fairly confident it was a bit more, but not a ton more, than regular counting.

Quick google search for blackjack shoe computers turned up an interesting article/interview with the inventor of the shoe computer: http://www.blackjackforumonline.com/content/taftint.html

Wasn't there some chatter on bjtf that DonS was working on something like this?

Quote: Ibeatyouraces

Wasn't there some chatter on bjtf that DonS was working on something like this?

Yes there was,I believe Don was working on that number along with working on the edge obtained by using the Tarzan count/system.

Quote: tyler498

I am talking about overall edge.




I totally agree, I guess I should've been more specific but I meant not including those other techniques, although it is true they would provide a bigger edge.
Was I'm wondering is, for instance if you see the Wizard's house edge calculation at WoO, I quote the methodology: "The "basic strategy with cut card" results are based on total dependent basic strategy, like the tables on this site, and the use of a cut card, which favors the dealer. "
What would the house edge be if someone ran a simulation which instead of playing by the basic strategy table, made a call to the WoO hand calculator (i.e. perfect composition dependent strategy, but not reshuffled every hand, played throughout the shoe). This is one stat I have a hard time finding.

You've heard the answer but apparently don't believe it. The answer is - with computer perfect play and taking into account all of the cards that have been dealt, you might pick up a fraction of one percent, maybe as much as one percent.

There simply are not that many "close call" decisions in blackjack, most of your decisions will not be altered by any reasonable distribution of cards left in the card deck. Assuming that the cut card is in a reasonable position and the shoe isn't dealt down to the last few cards, its like this: averaging over many shoes, you would alter < 5% of your decisions, each time gaining (on average) < 5% advantage over the normal decision. So the gain in EV from computer-perfect play is < 0.05*0.05 or < 0.0025. That's a ballpark number.

Quote: gordonm888

You've heard the answer but apparently don't believe it. The answer is - with computer perfect play and taking into account all of the cards that have been dealt, you might pick up a fraction of one percent, maybe as much as one percent.

There simply are not that many "close call" decisions in blackjack, most of your decisions will not be altered by any reasonable distribution of cards left in the card deck. Assuming that the cut card is in a reasonable position and the shoe isn't dealt down to the last few cards, its like this: averaging over many shoes, you would alter < 5% of your decisions, each time gaining (on average) < 5% advantage over the normal decision. So the gain in EV from computer-perfect play is < 0.05*0.05 or < 0.0025. That's a ballpark number.

Thanks for the reply! But yeah I did believe it. The answer from OnceDear made sense and I didn't question it. The discussion just kept going as to whether there is some simulation result to this.
I agree with your reasoning, I was thinking the same thing until I played single deck, and now I think the numbers are conservative, based on the fact that the difference in edge for single deck between composition dependent stategy and basic strategy is 0.15%! and that's just for the first hand, and the composition of that hand is already taken into account in many hands for basic strategy (the cases where you don't have different combinations) so the impact on the first hand is smaller than on the next hands. That's what got me expecting a much higher number (Before OnceDear's answer)

Depends on what you mean by “perfect play”. The old BJ computers used simple combinatorics. That’s not really perfect.

First, perfect play should be based upon maximizing bankroll growth, not edge. That means variance must be taken into account in every decision. Turns out this is a large problem. You need to look at all depths into the shoe, their frequencies, and the frequencies of remaining cards at those depths. More complex than it sounds as the rules of the game result in non-randomness in remaining card frequencies. For example, the last card dealt in the previous hand is more likely to be a high card than a low card. To be perfect, you would also need to consider the difference a play could make on the probability of getting an additional round near the cut card.

Having said all that, Eric Farmer published partial results of a study in which Gronbog and DonS were involved. Don arranged the study. (My only part was to get copied on all the correspondence and make an occasional useless remark.) The study was not perfect perfect; but far better than the old BJ computers. It showed about a 75% gain in win rate over HiLo with Kelly optimal betting and given a particular set of specifications. Note, this is less than a 75% gain in EV.

Quote: QFIT

Depends on what you mean by “perfect play”. The old BJ computers used simple combinatorics. That’s not really perfect.

First, perfect play should be based upon maximizing bankroll growth, not edge. That means variance must be taken into account in every decision. Turns out this is a large problem. You need to look at all depths into the shoe, their frequencies, and the frequencies of remaining cards at those depths. More complex than it sounds as the rules of the game result in non-randomness in remaining card frequencies. For example, the last card dealt in the previous hand is more likely to be a high card than a low card. To be perfect, you would also need to consider the difference a play could make on the probability of getting an additional round near the cut card.

Having said all that, Eric Farmer published partial results of a study in which Gronbog and DonS were involved. Don arranged the study. (My only part was to get copied on all the correspondence and make an occasional useless remark.) The study was not perfect perfect; but far better than the old BJ computers. It showed about a 75% gain in win rate over HiLo given a particular set of specifications. Note, this is less than a 75% gain in EV.

Thanks for the reply! Now I understand why I couldn't find that exact number easily. It does sound quite complicated to calculate!