MJ: KO vs Hi-Lo: Question and Analysis

[JohnAuston]

> No bet jumps greater than double cannot be
> accurately calculated in this manner. This
> only reduces the number of bets jumps
> greater than double.

Norm ( I believe ) is making the point that the BJRM canned-sim-based charts only show what happens to the canned data when the bet jumps between two consecutive counts are restricted to no more than double.

It does NOT reflect what happens when the count increases or decreases more than that, from one round of betting to the next, as can happen when you actually play. Results from that would produce different data values than the charts I show.

This sort of nuance underscores the difficulty of doing a "fair" comparison between different counting systems. The first set of charts are "fairer" than the second set.

For "quick and dirty", the BJRM charts will not be far off, but for pure research you would need to use Norm's software to do "fair" custom sims of bet restriction, and real world chip denominations.

I'm just showing the BJRM data adjusted artifically. The second set of charts would probably have been better if I had not check-boxed the "2x Jumps max" .

[Magician]

> For "quick and dirty", the BJRM
> charts will not be far off, but for pure
> research you would need to use Norm's
> software to do "fair" custom sims
> of bet restriction, and real world chip
> denominations.

I'd love to be able to import CVData sims into BJRM like I can import SBA sims. Any chance of that happening?

[Norm Wattenberger]

BJA page 211 paragraph 11.

> What do you have in mind here, Norm? What
> does "in this manner" mean to you
> such that you don't think that BJRM is
> eliminating all of the bet jumps?

> Don

[JohnAuston]

> I'd love to be able to import CVData sims
> into BJRM like I can import SBA sims. Any
> chance of that happening?

I think Norm has made it possible, by creating a readable output file that I could write code to parse, but I haven't followed up on that, as of yet. It is a time issue for me. It will take a lot of time, and would have to be a true labor of love, as was BJRM itself, since the effort/$$ is less than minimum wage. :-(

I'll think about it.

[Don Schlesinger]

> Norm ( I believe ) is making the point that
> the BJRM canned-sim-based charts only show
> what happens to the canned data when the bet
> jumps between two consecutive counts are
> restricted to no more than double.

> It does NOT reflect what happens when the
> count increases or decreases more than that,
> from one round of betting to the next, as
> can happen when you actually play. Results
> from that would produce different data
> values than the charts I show.

> This sort of nuance underscores the
> difficulty of doing a "fair"
> comparison between different counting
> systems. The first set of charts are
> "fairer" than the second set.

> For "quick and dirty", the BJRM
> charts will not be far off, but for pure
> research you would need to use Norm's
> software to do "fair" custom sims
> of bet restriction, and real world chip
> denominations.

> I'm just showing the BJRM data adjusted
> artifically. The second set of charts would
> probably have been better if I had not
> check-boxed the "2x Jumps max".

Actually, I really DO know what I wrote on p. 211 (!), but I didn't realize that the "2x" button on BJRM behaved the way it does.

I thought that it simply looked at the previous bet and, when the button was checked, it bet either what the TC called for (optimal) OR two times the previous wager -- whichever was less -- with no other restrictions.

Guess not, huh?

Don

[JohnAuston]

> I thought that it simply looked at the
> previous bet and, when the button was
> checked, it bet either what the TC called
> for (optimal) OR two times the previous
> wager -- whichever was less -- with no other
> restrictions.

> Guess not, huh?

It bets what the tc/rc is set to. If the count jumped more than one "level", so that, say, a 6 unit bet is allowed at the TC that is now 2 levels above the one on the previous round ( which was, say, 2u), it goes ahead and bets the 6u, instead of only 4u.

I'm guessing that, at 6-deck 75% penetration, and TC change upward by 2, from round to round , is not all that likely. So, for the pictures I posted, the data is probably "good enough for most folks".

1 and 2 deck sims would be more susceptible to this factor, of course, and 8-deck even less so than 6.

[MJ]

Interesting point you make. In fact for a 6 and 8 deck game the KO Preferred Strategy only uses 13 index plays. The following 5 index plays are EXCLUDED from the 6 and 8 deck game and only used for the 1 and 2 deck game:

13 vs 2
13 vs 3
12 vs 4
12 vs 5
12 vs 6

As you pointed out the KO Preferred Strategy replaces splitting 10s against 5 and 6 by doubling 8 vs 5 and 6. I'm not sure how much thats costs in terms of EV but it is certainly something. If most Hi-Lo SIMS use the FULL I-18 including the 10 splits then it is no wonder that Hi-Lo has a higher SCORE then does KO Preferred Strategy. It contains 5 more index plays then does KO! NO FAIR! :-)

However it should be noted that the KO Preferred
Surrender Strategy has 3 more index plays then the Hi-Lo Fab 4. Here they are:

16 vs 8
14 vs 10
14 vs A

So KO does have something working in its favor assuming the KO Surrender Strategy was used in the SIM.

I suppose it does make sense to compare both systems like this because that is how most people will use them. But the comparison John does below shows that when an "apples to apples" comparison is used KO performs nearly identical to Hi-Lo.

How can I incorporate the 5 index plays above into a 6 deck game? Why did the authors only apply those plays to single and double deck?

-MJ

> Most Hi-lo sims use the I-18 indices. Most
> KO sims use the KO Preferred strategy
> matrix. There is so much info at bjstats.com
> that I was unable to determine exactly which
> sims you were looking at.

> This isn't really a fair apples-to-apples
> comparison since the I-18 includes splitting
> 10's against dealer 5 and 6, while the KO
> Preferred replaces these with doubling 8 vs
> 5 and 6, two considerably less powerful
> indices.

> The KO creators did this because they felt
> that splitting 10's draws too much heat. In
> addition, it is a high variance play that
> many counter just plain don't like to make.

> It could be argued that the systems should
> be compared in this way because it is the
> way most people play them. However, I doubt
> that the majority of counters routinely
> split 10's every time the count calls for
> it.

> There are also differences in the KO
> Preferred surrender strategy and the Fab 4
> surrender indices. It is even possible that
> the sim in question did not include the KO
> surrender indices at all.

> At any rate, as others have mentioned, you
> can certainly trust the numbers in BJA3.

[MJ]

Hey John thanks for the charts. Well there you have it...Hi-Lo only outperforms KO by a $1.50/Hr using similar index and surrender plays. Finally a true "Apples to Apples" comparison.

However, most people will not use the I-18 when playing KO because there are only 13 plays included in the book for a 6 and 8 deck game. Also the KO Preferred Surrender Strategy includes more plays then the Fab 4. Assuming playing KO in RC mode are all the index values given in Systems 101 based upon penetration? If there are several index numbers for one index play then that sort of defeats the purpose of KO.

For example, I've been told BJRM gives many index numbers based upon penetration when splitting 10s vs 5 and 6. A single index number should be generated for EACH I-18 play for KO. This way it makes things simple and the user doesn't have to know 20 different numbers for a single play deviation.

-MJ

> Further showing the closeness of these two
> simple systems, here are the real world
> versions of the two charts ( normal chip
> denominations and no bet jumps greater than
> double).

>

[Magician]

> I think Norm has made it possible, by
> creating a readable output file that I could
> write code to parse, but I haven't followed
> up on that, as of yet. It is a time issue
> for me. It will take a lot of time, and
> would have to be a true labor of love, as
> was BJRM itself, since the effort/$$ is less
> than minimum wage. :-(

Alternatively, one or both of you opened your file format and allowed a third party to write a converter. Or you could both add XML support, then a third party could define a transform. Just trying to think of ways that don't involve you doing all the work.

[Parker]

> So KO does have something working in its
> favor assuming the KO Surrender Strategy was
> used in the SIM.

Maybe. There are more indices, but they are also rounded, giving up a tiny amount of EV in the process.

> I suppose it does make sense to compare both
> systems like this because that is how most
> people will use them. But the comparison
> John does below shows that when an
> "apples to apples" comparison is
> used KO performs nearly identical to Hi-Lo.

My point all along has been that this is not how most people use the system. I really question how many Hi-lo players who profess to use the I-18 actually split 10's each and every time the count calls for it.

> How can I incorporate the 5 index plays
> above into a 6 deck game? Why did the
> authors only apply those plays to single and
> double deck?

The problem is that when using an unbalanced system in running count mode, the most accurate indices are those at or near the pivot point. This is one of the reasons KO works as well as it does -- many important indices are near the KO pivot. The further we stray from the pivot, the more the indices are affected by penetration. The 5 indices in question are so far from the pivot in a 6 or 8 deck shoe as to be useless, adding no significant EV. Furthermore, these are negative indices - they only come into play when we have a minimum bet out, reducing their usefulness even more.

[MJ]

> The 5 indices in question are so far from the
> pivot in a 6 or 8 deck shoe as to be
> useless, adding no significant EV.
> Furthermore, these are negative indices -
> they only come into play when we have a
> minimum bet out, reducing their usefulness
> even more.

If that is the case then howcome John's SIM shows Hi-Lo I-18 performs nearly identical with KO I-18?
The only difference between the BJRM SIM and the one from bjstats.com are the 10 splits and the 5 additional index plays which you say "are useless adding no significant EV".

Clearly you can see from the BJRM chart Hi-Lo I-18 outscores KO I-18 by only $1.50/Hr. Where as before Hi-Lo I-18 outscored KO Preferred by $6.00/Hr. Those 5 extra index plays and 10 splits certainly must be worth something. How else can you account for the extra $5.50/Hr that KO earns?

Lastly, does BJRM give the index numbers based upon penetration or does it give an average number for each index play in RC mode? If not would it be possible to get an average index number taking different penetrations into consideration? Thanks for assistance.

-MJ

[MJ]

Thanks for the clarification Don. I understand that once N0 hands have been played then EV = 1SD. So if your behind one SD then your earnings are $0.00. But assuming a normal distribution on the bell curve, 1 SD to the left is = 34%. So how do you come up with a 15.86% chance of being behind once N0 hands have been played? Thanks for any help.

> I didn't write what I meant to write. Sorry.
> When you play NO hands, your e.v. now equals
> one s.d. So, if you were to experience a
> one-s.d. LOSS after playing NO hands, you
> would still break even (not lose anything).
> This probability is what is equal to 15.86%.

> In other words, after playing NO hands, you
> have only a 15.86% chance of being behind.
> Some will consider that the "long
> run." Some require even more hands,
> equal to two s.d.s, for example. Sorry for
> the confusion.

> All.

> There is no such thing as a
> "certainty" of achieving a win
> rate. The longer you play, the smaller s.d.
> is, expressed as a percentage of e.v. That's
> all the "certainty" you get. :-)

> See above. It's a measure of how long it
> takes for e.v. to catch up to s.d. -- which
> isn't a trivial value.

> The reciprocal of 3 is 1/3. The reciprocal
> of SCORE is NO, once you multiply that
> reciprocal by one million.

> I'd use the charts in BJA3.

> Don

[Don Schlesinger]

> But assuming a normal distribution on the bell
> curve, 1 SD to the left is = 34%.

The area between one s.d to the left and the mean (in the middle) is 34%. Same for the distance from the mean to one s.d. to the right. that leaves 1 - 2*34 = 32% for the TWO tails: 15.86% to the left of one s.d. and 15.86% to the right of one s.d.

Don