An international math superstar's assessment of sims as models of human behavior: "Absolute rubbish!"

_
Last week's edition of Newsweek includes an article which I interpret as vindication for my longstanding skepticism about the validity of computer simulations as predictors of human behavior.

Paul Wilmott is described as a leading quant, meaning that he specializes in developing new principles of quantitative finance that can be safely applied to the complex world of economics.

It seems he agrees with me that because mathematics cannot predict human behavior, mathematicians recklessly created models that ignored it entirely, with the disastrous results that we are all now living through.

Economics and gambling are alike only when it comes to the speculative nature of stock trading, but human nature has a powerful influence in both fields, and ignoring it is both deceptive and dangerous.

To celebrate Mr. Wilmott's dismissal of most financial models as "absolute rubbish" I decided to boost my database of blackjack and baccarat outcomes past the half-million mark by writing a new "sim" of my own to give me 150,000 new rounds.

That might seem foolish or hypocritical, but I had every reason to be confident that target betting would come out ahead this time, just as it always has.

I would love to be able to claim that there was not a loser among the 20 models of 7,500 rounds or 100 shoes apiece, but it didn't work out that way.

Here's the bottom line:


Given my dismissal of sims as proof of anything other than the disingenuous agenda of house-trained mythematicians, I can hardly trumpet the above summary as a victory for the principles of target betting.

But like my critics, I have to be grateful for the speed and convenience of computer game simulations while doubting their relevance to real play in real time by real people risking real money.

It would take me at least a month to record and collate 150,000 outcomes of baccarat, which has to be the most laborious table game ever devised, and even my soon-to-be-replaced laptop was able to come up with this data set in barely 15 minutes.

The first thing to notice is how similar most of these sets are to the baccarat data independently supplied by Messrs. Jones and Rodriguez.

As regular readers will know, my quarrel with sims is based on their requirement that the "player" they pretend to emulate must bet exactly the same way whatever the current conditions. Like a suicidal lunatic, in a word.

No defensive tactics or damage control can be applied in the otherworld of runaway sims. And perhaps the silliest conceit of all is that, in a sim, it must be possible to bet from the house minimum to the house limit in one place.

The defenders of the status quo insist that damage control is in the long run irrelevant, because the inexorable nature of the house advantage makes the probability of a beneficial effect slightly less than a 50-50 proposition in the same way that each bet is slightly more likely to lose than to win.

What this assumption willfully ignores is that progressive betting changes the odds from less than 50% to better than 90%, as shown throughout the earlier blackjack/baccarat data set and in the new outcomes summarized above.

That's an improvement from less than 1 chance in 2 to a much friendlier 10 chances in 11!

Target betting will always show a profit if a recovery series lasts for 10 rounds or fewer, even if more bets were lost than won in that series.

In the latest set, 91.4% of all series fell into that "easy win" category.

That's not to say that series lasting longer will lose - there was just ONE "bust" in 28,798 series in the new sample, indicating a win rate of 99.9965%.

My experience is that real play almost always offers better conditions than any simulation, even if that substitute for reality is based upon actual outcomes recorded in a casino.

The problem is, of course, that even "real" results become sim-like when they are plugged en masse into a model, with a betting strategy then strictly applied with no reliance on a human response to potential threats to the bankroll.

It is a classic rock-and-hard-place dilemma, confirming that the only truly relevant test of a betting strategy is real play by...etc.

The gambling industry knows that essential truth better than any mathematician!

Paul Wilmott happens to be the loudest and most visible among countless 20/20 visionaries who were either silent or unheard in the run-up to the economic tsunami that, according to an estimate I read on the Net today, has wiped out more than $1.3 trillion in Americans' personal wealth in less than a year.

The only good news is that a big chunk of that washed-away wealth was in the hands of casino moguls like Steve Wynn, Sheldon Adelson and Kirk Kerkorian after they had leeched it from the misguided fools who actually earned it before flushing it away in Las Vegas.

To the thousands of Wall Street wonks whose jobs were sucked away by the ebb tide, I can only suggest that target betting in a casino offers them a better shot at rebuilding their bankrolls than their former workplace ever will again!

And target betting is also a whole lot less complicated and dangerous than any of the fancy fiscal formulas that flew ever higher before succumbing to the force of gravity and falling to earth.

What sims ignore is that once a player recognizes that progressive betting is the only way to win consistently, a wide array of tactics become available to him or her.

And while a million-dollar bankroll may be ideal, it is not essential: the principles of target betting can be adapted to more modest resources, and will always offer the player better long-term odds of success than flying by the seat of his pants.

Luck might serve him better in the short term. But luck never lasts, for either side of the gambling equation.

The new RNG summary offers a wealth of useful insights that are not rendered totally meaningless by the unreliability of sims.

For instance, a standard Martingale and my souped-up versions of double-up and Oscar's Grind all did well against the new sample, each reversing the effect of the negative expectation that would have doomed a flat or random bettor challenging the same outcomes.

There may be skeptics out there who will say that 150,000 rounds is "not a representative sample" but they will be tooting a hollow tune: very few "recreational" gamblers will place that many bets in an entire lifetime.

The bottom line is that all of the results I have reported in this blog are impossible according to the conventional wisdom, meaning that either the conventional wisdom is wrong or I am a fraud.

In the new non-fraudulent sample summarized here, 14 of 20 sets of 7,500 outcomes ended with more bets lost than won, and more than 64% of all series "should have" ended with a negative or neutral result.

Instead, more than 73% of all series scored a profit after just one win, a minor miracle made possible by a combination of the OL, 2L, 3L and MSL rules of target betting (click on the rules link at the top of the blog for a refresher course).

The average end of series (EOS) win came in less than six bets.

As expected (and required by the overall house edge), more EOS attempts failed than succeeded. But because the win target is constantly updated to match the loss to date, that was not, in the end, a problem.

Target betting with the above parameters applied is an aggressive approach suited only to a large bankroll, but the actual exposure was barely 50% of the recommended $1 million.

The average win per 7,500 rounds was more than $110,000 or about $1,000 per shoe of baccarat.

Target betting turned a negative expectation of 0.97% (low for baccarat, but that can happen) into a positive result equal to +1.54% of total action.

SM, SM+ and OG (see below) all did better than target betting, in percentage terms, but delivered smaller funny-money profits.

Target betting is intended to win more when it wins than it loses when it loses in order to overcome the inescapable house advantage in casino games of chance, and against the new sample, the overall average win value exceeded the average loss value by 5%. That was more than enough to trump the house edge.

The AWV/ALV percentage for the 19 winning sets was 115%.

The average bet for target betting was $529, although only 5% of all bets (7,440 out of 150,000) exceeded $1,000.

The averages for SM, SM+ and OG were $51, $217 and $74 respectively.

The final target betting win against this irrelevant and unreliable sample of outcomes (it came from a sim, so how else can I describe it?) was equal to a little more than $800 per hour of play.

Exposure of the bankroll exceeded 10% in 9 of the 20 sets of baccarat outcomes. I will apply "bare bones" parameters against the same data sample sometime soon and post the results.

The "expected" result from this set, given a -0.97% actual value (AV) and target betting action of $79.35 million, was a LOSS of $773,000 in contrast to a WIN of $1.22m.

The overall win rate for target betting was 99.9965% or 1 loss in 29,798 series.

My critics will continue to insist that none of this could have happened without sleight of hand. That is because in many cases their livelihood depends on denial.

More about the alternatives to target betting featured in the above summary...

SM is the much-maligned standard Martingale (-1, -2, -4, -8, +16), which is very effective but almost always unplayable because of the vigilant paranoia of pit personnel.

SM+ is my adaptation of SM to deliver a little more than $5 in profit from a $5 opening bet: -$5, -$10, -$25, -$50, +$100). It is also virtually unplayable, although scrutiny can sometimes be dodged in a busy casino by betting no more than 3 wrong bets at any one location.

OG is for Oscar's Grind, which as far as I know is the only betting strategy ever endorsed in a mass-market "how to" book about gambling.

That alone makes the method suspect, since gambling books as a genre are intended to encourage people to play more while reinforcing the message that losing is ultimately inevitable.

I applaud OG for freezing the bet after a loss, thus helping to confound the enemy. But author Tom Ainslie's rules (NB=PB+1u after a win, bail out after a 20u loss, and do not apply a win progression) are in my view a prescription for certain failure.

Mr. Ainslie claims he has paid for countless Caribbean gambling trips with Oscar, which makes him a very lucky fellow. Whenever I model his rules, I drown in red ink!

My "OG" applies a win progression of PB+1u after an opening win, and as with target betting makes the streak-ending loss the LTD. After that, the mid-series win response is PBx2 to a max of LTD+1u rather than Mr. Ainslie's wimpy PB+1u.

As I said earlier, a player who embraces the unique efficacy of progressive betting as an antidote to the progressive (or regressive) process called losing has multiple options that range from mildly confident to supremely cocky.

Any one of them is better than playing and losing the way the gambling industry expects you to.

An important reminder: The only person likely to make money out of this blog is you, Dear Reader. There's nothing to buy, ever, and your soul is safe (from me, at least). Test my ideas and use them or don't. It's up to you.
_