Myooligan: Value of Precision - Preliminary Results

[Myooligan]

A week or so back there was a discussion about using index numbers to the first decimal place. Don has been reviewing/reworking a spreadsheet I made that deals with that subject, and it should eventually make it to these pages in one form or another. But in the meanwhile, I thought it'd be interesting to show what happens when you go the other direction - become less precise than accuracy to the "ones" place.

The tables below compare the SCORES of precise, risk-averse index numbers (roughly 150 indices) to another set in which the index numbers are averaged between single deck and multi-deck, and then rounded to the nearest "5,"(In other words, index numbers are 0, +-5, +-10, +-15, etc). Four exceptions: Insurance, 12v3, and A,8v3 = +3, and 13v3 = -3. I kept the precise numbers intact on these 4 decisions because the plays are "volatile" and the index happened to be "in the middle," thus not well-suited to rounding.

When I had to make decisions about rounding up or down, I favored the single deck numbers, because that's the game I play most often. But it's noteworthy, because even with these low "loss" percentages, these are still "compromise" sets of index numbers, landing somewhere in between Reno, Vegas, and AC figures. So, if you were merely rounding to the nearest five, and not also "compromising," you might be able to do even better.

 

H17 NDAS 1D 

	 Benchmark	 Rounded        Loss  

1-5	 $189.30 	 $187.87        1% 

1-4	 $160.27 	 $158.66        1% 

1-3	 $120.65 	 $118.99        1% 

1-2	 $66.71 	 $65.28         2% 

1-1	 $7.83 	         $7.30          7% 

 

H17 NDAS 6D - play only @ TC >= 0 

(Only 1 variation from 1D matrix: Insurance = +5) 

	 Benchmark	 Rounded        Loss	 

1-16	 $167.84 	 $163.38        3% 

1-12	 $162.91 	 $157.91 	3% 

1-10	 $161.86 	 $157.18 	3% 

1-8	 $157.04 	 $152.20 	3% 

1-6	 $148.25 	 $143.34 	3% 

 

S17 DAS 6D play only @ TC >= 0 

(30 play variations from 1D matrix) 

	 Benchmark	 Rounded        Loss 

1-16	 $210.43 	 $209.12 	1% 

1-12	 $208.95 	 $207.46 	1% 

1-10	 $207.61 	 $206.05 	1% 

1-8	 $204.21 	 $202.55 	1% 

1-6	 $197.49 	 $195.87 	1%
						

[Don Schlesinger]

Now that we know how costly being off by 5 is (not!), I can't wait to learn how much tremendous extra edge Francis is getting for precise, decimal-point indices. Has to be hundreds of ... pennies in a ... lifetime. :-)

Don

[Sun Runner]

Thanks for confirming what I already thought (hoped) was the case.

From reading Snyder's Blackbelt in Blackjack, and paying attention to what the smart guys say, I became convinced that the principles behind HILO Light had real merit.

Since that time, I've only been using the Catch 22 indices (for the most part) and I've rounded all tags to 0, 2, and 4.

Insurance will always be 3.

It's been working for me.

[Norm Wattenberger]

Neat stuff. Index accuracy really doesn't help much unless the game really sucks or you're flat-betting a SD game. Glad to know that with my ageing memory

[MGP]

Very nice study

Thank you very much for answering my question. This clearly confirms that it's not worth the programming hassle for changing counts as a hand is played out.

I guess it really was possible to sim it out after all.

Thank you again very much!

Sincerely,
MGP

[Francis Salmon]

Your idea is quite interesting but you should use appropriate indices in order to compare.Risk-averse indices are useless for practical play.It would mean that you have to use a different index for every bet level.It's funny that the same people who say that index precision doesn't matter claim that you have to allow for risk.
Of course you should use EV-optimizing indices and even more important they should be correct. From the four examples you gave two are totally wrong: The index for A,8 v. 3 is +5.5 and not +3 (risk-averse would even be higher)whereas the index for 12v3 is only +1.3 and not +3.
If 50% of your indices are wrong,I'm not surprised that the results are what they are.
You should use different indices for one deck and for shoe games and not just take the average for everything. Mind also the difference between H17 and S17 as well as NDAS and DAS.
Besides,I don't understand why you would make exceptions for the "in-betweeners".

Francis Salmon

[Saboteur]

He intentionally used indices known to be off-the-mark. Then, he showed the results of such an approach. The small loss-percentage figures, as I understand it, verify that a precise index is not important.

He's not disputing that he used the wrong indexes. That was the whole point of the exercise. Or am I missing something?

[T. Hopper]

The exact index varies at different points in the shoe. So even a so-called "exact" index is nothing more than a weighted average.

[Francis Salmon]

He claimed to compare with precise indices but as we can see from the examples he gave, many of them are way off the mark.

Francis Salmon

[Saboteur]

You might be the only one here with precise indexes. If you'd be willing to share them, maybe another sim could be run.

If you own simulation software, it might bolster your viewpoint if you could run the sims and make the parameters and the results available to everyone. If you don't own the software, you might suggest the parameters for a sim to be run by someone else.

[Myooligan]

I hate hi-lo.

The system was Zen and the indices were created with SBA. As Saboteur said, the whole point is that, even after compromising and rounding, the losses were negligible.

But Francis is correct that these are the losses as compared to "useless" risk-averse indices, rather than the "appropriate" ev-maximizing ones, so READERS BEWARE!!!

> Your idea is quite interesting but you should use
> appropriate indices in order to compare.Risk-averse
> indices are useless for practical play.It would mean
> that you have to use a different index for every bet
> level.It's funny that the same people who say that
> index precision doesn't matter claim that you have to
> allow for risk.
> Of course you should use EV-optimizing indices and
> even more important they should be correct. From the
> four examples you gave two are totally wrong: The
> index for A,8 v. 3 is +5.5 and not +3 (risk-averse
> would even be higher)whereas the index for 12v3 is
> only +1.3 and not +3.
> If 50% of your indices are wrong,I'm not surprised
> that the results are what they are.
> You should use different indices for one deck and for
> shoe games and not just take the average for
> everything. Mind also the difference between H17 and
> S17 as well as NDAS and DAS.
> Besides,I don't understand why you would make
> exceptions for the "in-betweeners".

> Francis Salmon

[Myooligan]

> The exact index varies at different points in the
> shoe. So even a so-called "exact" index is
> nothing more than a weighted average.

I think that's a really good argument against using decimal index numbers. Kind of an obvious one, too -- surprised it didn't come up earlier in the discussion. Since Francis doesn't like "to use a different [risk-averse] index for every bet level," I wonder how he feels about memorizing a different decimal value for every penetration point.

[Francis Salmon]

> The exact index varies at different points in the
> shoe. So even a so-called "exact" index is
> nothing more than a weighted average.

Right! But is there any reason why this weighted average should be rounded to a whole number,especially if it's somewhere in between?