RA insurance

1357111317

Well-Known Member
#1
Now I know some people here use RA (Risk-Adverse) indices. These plays are not EV maximizing however they take into account the EV gain vs the variance increase of deviation from basic stratagy (ie. splitting 10s).

My question or theory here is can you use a risk adverse stratagy when it comes to insurance.

Situation:

You are dealth a 20 v an ace. The TC is now 2.5 (HiLo), not an insurable count. I will analyze this using a neutral count since I don't have the tools to simulate a +2.5 count and I doubt the EV figures will be too far off.

Since the definition of variance is how far the results fall from expected I give you two possible situations.

1. You do not take insurance. You will lose your bet ~31% of the time due to dealer blackjack, Win your bet 48% of the time, push your bet 14% of the time and lose your bet due to the dealer making 21 6.57% of the time.

Your EV in this situation is roughly 10% percent with your variance being roughly 0.85

2. You do take insurance. You will push your bet ~31% of the time due to dealer blackjack, you Win half your bet 48% of the time, lose half your bet 14% of the time and lose 1.5 your bet due to the dealer making 21 6.57% of the time.

Now the Ev will be slightly less than the 10% assumed in the first situation but the variance in this second situation is roughly 0.028.

As you can see the variance on the second situation is significantly less than the first situation, without sacraficing much EV, the mark of a RA play.

What do you guys think about this one. Something like K_C's combinatorial analysis program would probably be very useful for determining the EV in and dealer outcomes in a situation like this.
 

SleightOfHand

Well-Known Member
#2
1357111317 said:
Now I know some people here use RA (Risk-Adverse) indices. These plays are not EV maximizing however they take into account the EV gain vs the variance increase of deviation from basic stratagy (ie. splitting 10s).

My question or theory here is can you use a risk adverse stratagy when it comes to insurance.

Situation:

You are dealth a 20 v an ace. The TC is now 2.5 (HiLo), not an insurable count. I will analyze this using a neutral count since I don't have the tools to simulate a +2.5 count and I doubt the EV figures will be too far off.

Since the definition of variance is how far the results fall from expected I give you two possible situations.

1. You do not take insurance. You will lose your bet ~31% of the time due to dealer blackjack, Win your bet 48% of the time, push your bet 14% of the time and lose your bet due to the dealer making 21 6.57% of the time.

Your EV in this situation is roughly 10% percent with your variance being roughly 0.85

2. You do take insurance. You will push your bet ~31% of the time due to dealer blackjack, you Win half your bet 48% of the time, lose half your bet 14% of the time and lose 1.5 your bet due to the dealer making 21 6.57% of the time.

Now the Ev will be slightly less than the 10% assumed in the first situation but the variance in this second situation is roughly 0.028.

As you can see the variance on the second situation is significantly less than the first situation, without sacraficing much EV, the mark of a RA play.

What do you guys think about this one. Something like K_C's combinatorial analysis program would probably be very useful for determining the EV in and dealer outcomes in a situation like this.
Remember that the insurance bet is INDEPENDENT of the regular BJ bet. This means are actually ADDING variance due to the bet in addition to your regular bet. There was an article somewhere about insuring good hands on extreme borderline cases, but otherwise, the EV maximizing index play is the way to go.

The thing about RA indeces is that the ones that are typically generated are the kind that maximizes Kelly betting. And what does the perfect Kelly better do? Never play in -EV situations (aka bet when the count is < insurance index).

PS: Found the articles.
http://www.blackjackforumonline.com/content/insureagoodhand1.html
http://www.blackjackforumonline.com/content/insureagoodhand2.html
 

jack.jackson

Well-Known Member
#3
Not sure if im losing in ev or not. But I really like, to take 1/4 insurance @+3, when playing two-hands(full insurance on one of the hands). If im playing 1-hand only, I will continue to take full insurance @+6(l2). Unsure if this would be considered RA insurance or not:confused:

EDIT:Actually I meant 1/4 my bet, and 1/2 insurance, at +3. So if im betting $50x$50, I will insure one of the hands, for $25.
 

1357111317

Well-Known Member
#4
SleightOfHand said:
Remember that the insurance bet is INDEPENDENT of the regular BJ bet. This means are actually ADDING variance due to the bet in addition to your regular bet. There was an article somewhere about insuring good hands on extreme borderline cases, but otherwise, the EV maximizing index play is the way to go.

The thing about RA indeces is that the ones that are typically generated are the kind that maximizes Kelly betting. And what does the perfect Kelly better do? Never play in -EV situations (aka bet when the count is < insurance index).

PS: Found the articles.
http://www.blackjackforumonline.com/content/insureagoodhand1.html
http://www.blackjackforumonline.com/content/insureagoodhand2.html
It is independant of your normal bet but If you have a 20 very rarely do you lose both your insurance and your normal bet. The majority of the time if you do not win the insurance bet you win the BJ bet.
 

rukus

Well-Known Member
#5
see grosjean article on RA insurance. i am of the belief that he is right and that certain card holdings lower your insurance index.
 

ExhibitCAA

Well-Known Member
#6
SleightOfHand says: "Remember that the insurance bet is INDEPENDENT of the regular BJ bet. This means are actually ADDING variance due to the bet in addition to your regular bet."

This is not true! "Insurance" is a great name for the bet, and as far as the gambler's perception of the bet, it is aptly named. Take the simplest example: if you have a blackjack, then taking "even money" (equivalent to buying insurance in a 3:2 game) lowers your variance on the hand to ZERO! When you push the blackjack, the insurance bet pays off; when you win 3:2 on the blackjack, you lose the insurance premium that you just paid. The two bets are not independent at all--they are negatively correlated.

If we were concerned only with expectation, we wouldn't care if the two bets are correlated or not. We simply try to make all of the positive-expectation bets we can, and expectation has the nice statistical property that you can simply add them up regardless of whether the two bets are correlated or not. That is, E(A+B)=E(A)+E(B), regardless of whether the random variables A and B are independent, positively correlated, negatively correlated, or whatever. But if we are talking about risk aversion, we are saying, "I care about variance--I want to reduce variance." Pages 9-10 of ExhibitCAA show that buying insurance reduces the variance for a round of BJ if you hold a good hand, such as a 20 or a BJ.

MathProf did a paper at the 13th International Conference on Gambling and Risk Taking (2006) regarding composition-dependent, risk-averse insurance. His paper goes into far greater detail than my pages 9-10, which try to make a simple point--that insurance is NOT an independent sidebet having nothing to do with your BJ hand (as many authors have stated). Insurance has a lot to do with your BJ hand, but in the sense of variance, which is something we often don't care about, because we are often preoccupied with expectation.
 

SleightOfHand

Well-Known Member
#7
ExhibitCAA said:
SleightOfHand says: "Remember that the insurance bet is INDEPENDENT of the regular BJ bet. This means are actually ADDING variance due to the bet in addition to your regular bet."

This is not true! "Insurance" is a great name for the bet, and as far as the gambler's perception of the bet, it is aptly named. Take the simplest example: if you have a blackjack, then taking "even money" (equivalent to buying insurance in a 3:2 game) lowers your variance on the hand to ZERO! When you push the blackjack, the insurance bet pays off; when you win 3:2 on the blackjack, you lose the insurance premium that you just paid. The two bets are not independent at all--they are negatively correlated.

If we were concerned only with expectation, we wouldn't care if the two bets are correlated or not. We simply try to make all of the positive-expectation bets we can, and expectation has the nice statistical property that you can simply add them up regardless of whether the two bets are correlated or not. That is, E(A+B)=E(A)+E(B), regardless of whether the random variables A and B are independent, positively correlated, negatively correlated, or whatever. But if we are talking about risk aversion, we are saying, "I care about variance--I want to reduce variance." Pages 9-10 of ExhibitCAA show that buying insurance reduces the variance for a round of BJ if you hold a good hand, such as a 20 or a BJ.

MathProf did a paper at the 13th International Conference on Gambling and Risk Taking (2006) regarding composition-dependent, risk-averse insurance. His paper goes into far greater detail than my pages 9-10, which are simply to make a simple point--that insurance is NOT an independent sidebet having nothing to do with your BJ hand (as many authors have stated). Insurance has a lot to do with your BJ hand, but in a variance sense, which is something we often don't care about, because we are often preoccupied with expectation.
ORLY? May I ask how much it affects the insurance count? Is it enough to affect it by a whole TC? Or is it something fairly small, like .1-.2 TC?
 

ExhibitCAA

Well-Known Member
#8
The great thing about expectation-maximizing analysis is that there is only one answer. Once we open up the door to risk aversion, then there are many different models, Kelly utility being only one of them. So, while it may be true (CVSim can probably tell us) that the Kelly index for insuring a certain hand (you'd need a different index for each hand) may be only slightly lower than the expectation-maximizing index, empirical evidence shows that many APs are more risk averse than suggested by Kelly. The variance of the net payoff for a round of BJ is GREATLY reduced when you hold a good hand and buy insurance (see p. 10 of Exhibit CAA). If you hold a BJ or a 20, buying insurance greatly reduces variance. To some gamblers and some APs, this reduction in variance is worth the price they pay (so most gamblers take even money on BJs).

I think that many fake APs use variance, RoR, BR, and risk aversion as a big rationalization for degenerate behaviors and "unlucky" results, so I think that straying from expectation maximization is a dangerous path. However, I have been thinking of an exception lately, though this exception is vague and conceptual, with no hard numbers (yet!). Let's say that you have a win ceiling, and possibly a CTR constraint. For example, you can win only $8000 in the joint before heat comes, and you also do not want to CTR, so you can buy in only about $9000. Let's say that you're already ahead $5000, and have bought in $6000 already, and you've got a $1000 bet on the layout, and you get a BJ against a dealer's Ace. Should you take even money?

I am considering the idea that if you can essentially "lock in" the win on this hand, getting you closer to the win ceiling without buying in any more cash, and allowing you to get out the door before heat comes, it might be worth doing. Obviously we must be careful not to carry this idea too far, but my point is that larger factors such as heat and "hit management" come into play.

(In practice, you would simply buy insurance when the dealer has blackjack, but this is a theory thread.)
 

jack.jackson

Well-Known Member
#9
Ive been insuring *11vA, and BJs @+2 for one, of my two hands and full insurance at +4(l2) for single hands only, for years.


*Full insurance on one of two-handed play.

+2 for 11vA is also my index for H17 games.(A Juicy Double makes it worth insuring early)
 
#10
ExhibitCAA said:
SleightOfHand says: "Remember that the insurance bet is INDEPENDENT of the regular BJ bet. This means are actually ADDING variance due to the bet in addition to your regular bet."

This is not true! "Insurance" is a great name for the bet, and as far as the gambler's perception of the bet, it is aptly named. Take the simplest example: if you have a blackjack, then taking "even money" (equivalent to buying insurance in a 3:2 game) lowers your variance on the hand to ZERO! When you push the blackjack, the insurance bet pays off; when you win 3:2 on the blackjack, you lose the insurance premium that you just paid. The two bets are not independent at all--they are negatively correlated...
I like this practice of using insurance more on +EV hands than -EV ones because it gives the appearance of ploppy behavior.

Suppose you are playing two hands, and get a natural and a 16 vs. dealer Ace. There is a place where you would insure the natural and not the 16. Or just take half insurance on each. Thus, if you were playing one hand and it had an EV somewhere between that of a 16 and a natural, it seems there might be some room for partial insurance too.

Maybe some other sidebets with negative correlation to the BJ bet could be used to control variance too. Super Sevens comes to mind, as it "insures" against the -EV event of receiving a 7.
 

nightspirit

Well-Known Member
#11
The paper from MathProf mentioned by ExhibitCAA is available on the GC-pages at bj21.com. Some months back when I tried to get access to the paper it wasn't available anymore but maybe it's up again.
 
Top