THE MATHEMATICS OF GAMBLING By Dr. Edward O. ThorpTable of Contents SECTION ONE—CARD GAMES 1 Chapter 1: Introductory Statement. . Chapter 2: Blackjack Chapier 3: Baccarat SECTION TWO—THE WHEELS 4 Chapter 4: Roulete....... Chapter 5: The Wheel of Fortune SECTION THREE—OTHER GAMES. Chapter 6: Horse Racing . Chapter 7; Backgammon « SECTION FOUR—MONEY MANAGEMENT......... ‘Chapter 8: Mathematical Systems ..........-.1I3 Chapter 9: Optimal Beting .... “Appendices... Keeping Your Gaming Knowledge Current ...... 141About the Author wand Thorp is adjunct professor of finance nd mathematics at the University of California at Irvine, where he has taught courses in finance, probability and funetional analysis. He previously taught atthe University of California at Los Angeles, the Massachusens Instnne of Technology and New Mexico Ste University ‘Thorp’ interest in gambling dates back almost 30 years, while Ihe was ll in rauate school at TICT-A. Tt was bere that he fist formulated his dream of making money from the development ‘of a scientifically based winning gambling systema. His first sub- Jest of study was the roulewe wheel, which offered him the op- Foran ue moda pss ope teresa place With the routeme work unfinished, Thorp’satenton was diverted by the blackjack work of Paldwin, Cantey, Meisel and MeDer ‘tt He st to work on this new problem. With the aid of a com- puter, Thorp developed the basi strategy and the five-couat en- ‘count and ultimate counting strategies. He used these methodswith sucess inthe Nevada casinos. The work was fist publiciz- ed in a scientific journal and saw broad public exposure inthe 1962 book Bea she Dealer. The book underwent a revision in 1966 and it i till epuied as che lassi eaily work a Ue “ack Jack revolution” which continues to this dey Tinh late 1960s, Thorp developed with Sheen Kassout a suc- cessful method for stock market investing involving warrants that proved so profitable that Thorp turned $40,000 into $100.00 in two years. The strategy was published in Bea the Market in 1967 ‘Additionally, using this strategy and further refinements, Thorp ‘naauyes ul-niliondllsiavesanent portfolio. He is Pies dent of Oakley Sutton Management Corp. and Chairman of the the Board of Oakley Sutton Securities Corp. “Thorp has continued to advance new theories for eambling and other games, as wel asthe stock market. Section One Card Games Casino card games such as baccarat and blackjack differ significantly from casino games such as craps, roulette, and slot machines in that they are not indepen: dent tral processes-that is, the cards that already have bbeen played do affect the odds on subsequent hands, Consider for amioment the game of blackjack, where the cards used on a round are put aside and successive rounds are dealt from an increasingly depleted pack. ‘The cards are reshuffied before a round If the remain- ing unused cards would be insufficient to complete a round or earlier, usually atthe casino’s discretion, What the early research on blackjack (contained In Beat the Dealer showed and what has been confirmed repeated. |y in the intervening 29 years is that the end pack pio- vides favorable situations often enough to give the player an overall advantage. While itis foolish to keep a record of past decisions at craps In order to determine which numbers are "hot" ‘or *cold" (the dice have no memory}, an ability to keep track of which cards have been played and knowledge of their relationship to the player's expectation can be beneficial, a long as the cards are not reshuffled after every hand.The Mathematics of Gambling “The ability to keep track of the cards played does not alone guarantee gambliny success at aparticuler game. indeed, one of the chief tasks of this section will be to examine the usefulness of card counting strategies in bacearat, considering the bets offered and the nature of the game. In Chapler 2, we will comment on blackjack systems, ‘as well as statistical methods useful in detecting Casino cheating. The later subject is important to thooe ‘who play the game seriously, because cheating in- tidents can erode any small edge the player may gain through the use of basic strategy and card counting. Chapter 1 Introductory Statement Thecasno patron who dacs ory isk athe als and chor: lye who wagers a the acrck confront st sero be formate adverari, The eins hope fv tbe vantage once be offered, tte rack the parm takeout -52¥on soa tensa bu few wal op item ‘Attensa, pesmi ak ea imoorant decison the fist ene: Wha ame d lay? Even aller thscce made moxtgaesoferaddional opsons Do play individual numbers or the even-money bets in roulette? Do Ian witha par oepsnlcnk or shoul | ort he Pi Shull prove ool pops cant “Theterse ayers oftred a mune of choce ar wel eis swe ih io hts Hea yom ot Torre twin, lace, or show, nson to comings ‘number of horses in the exotic or “‘gimmich age ” ‘Alot thsechozeshave tent” anne de ayers to spina hier or mini is lo. Tey lca be at east 2The Mahemaics of Gambling partially solved through the use of mathematica theory, The Inteligent player mast have + hase understanding of the ‘mathematics behind the game or games he plays if heis to survive financially or actually profi. There are situations where the player has the advantage. The most-publicized example, of Course, is casino Blackjack. The game has beciune wuglir in recent years due to casino countermeasures, Dut blackjack can still be profitable forthe sophisticated player. There could be Several other favorable games, asthe rear will soon discover. ‘A familiarity with baste probability will allow th alert gambler to discover those positive expectancy games and exploit them where they exist. A vast knowledge of mathematics is not Fequed, Some of the finest poker payers iu te wountty never went to college, but they do havea sense of what makes a good poker hand and what their chanees of having the best hand are fcr ll the cards have been dealt. ‘Mathematical Expectation Tave already made reference othe conezr of mathematical expectation. Tis pipe central to an undersanding of the chapters to follow. Imagine or amoment con ose game with an unbiased coin (acon we assume il produce SOM heade and ah) Sipe ‘ose alo that we ae offered an opportunity tobe thatthe next Fipwillbeheads and he pao wil beeven money when we win (oe reste a SI profit in addin to the return of wager). Our ‘Mathematical expectation in iisexamples: (0+ (8{=1)=0 : ‘The mathematical expectation of say bet in any game is com. Dated by multiplying each posible gain oro theproboily DT that anor os, then adding Ueto figures. nthe prceling trample, we expec to gain nog rom pangs pame. This {Sknown asa argame,onein whieh te player has no advantage Now suppessit pyot was changed 03/2 againofS1. 50 Now suppose the payotf was changalt03/2(again of .S0in adaiion 10 our §1 be). Our expeation would change to: ‘ roduny Senet (5K1)+ (54-1 Paying this game 10D times would give usa positive expectation of $35. ‘Thetwo examples presented thusfarare admittedly simple, but ‘en this type of analysis all hati needed to evaluate a prop- ‘sition. Consider the “dozens” bet in roulette. Our expectation fori betis (12182 + 28138) - 1): ‘As another example, suppose that on the first hand of four deck Blackjack te payer bes $12, be seal 6,3, and the dealer then shows an ace up. The dealer asks the player if he wants insurance. Thisisaseperate S6bet.Itpays 12 ifthe dealer's hole cardisa tenvalue. Itpays ~$6 otherwise. A full fourdeck pack fas 64 tens and 144 non-ens. Assuming the deck is “randomly” hulled (his means that all ordering of thecards are equally prob- able), th chances ae equally likely that each ofthe 2 unseen ud sd dealers ve ard. Tass de player's expen {64/205(12) + (1411206 —6)= — 78/206 oF about ~$.38, The player should not take insurance. Different betting amountshave different expectations, But the player's expectation asa percent of the amount bets always the ‘me umber. nthe case of being onthe Red in route, thisis 18/38 ~20/38 = ~2/38= ~1/19 or about ~$.26%. This, the expectation of any size bet on Red at American dovble-re0 ronleteis ~1/19orahont ~5. 26 f the total amount bet Soto ‘2 the expectation Tor any size bet on Red, just multiply by =5.258%. Wit one exception, he other American double zero roulette bets also have this expectation per uni bet. The player's ‘expectation per units often simply called the player's dsadvan- tage. What the player loses, the house wins, 50 the house advan ‘age, house percentage, or house expectation per rit bet by the - 68The Mathematics of Gambling player is +5.26%, ‘A usefl basic fact about the player's expectation is this: the expectation for a eres of bets the total ofthe expectations for the individual bets. For instance, if you bet $t on Red, then $2, then $4, your expectations are ~$2/38, ~$4/38, and ~$8/38 ‘Your total expectation is —$14/38 or (a loss of) about = §.37. ‘Thus, if your expectation on each ofa series of bets is ~5.26% of the amount bel, then the expeciation on the whole series is ‘5.26% of the ftal ofall bet. This is One ofthe Fundamental reasons why “staking systems” don’t work: a series of negative expectation bets must have negative expectation. Repeated Trials Expectation isthe amount you fendto gain or lose on average ‘when youl. I, Bowever, doesnot explain the fluctuations from ‘epectaion that ovat in avtual wal. ‘Consider the fair game example mentioned earlier in the chapter. Ina series of any length, we have an expectation of In ‘any such serestis posible tobe ahead orbehind. Your total profit tor loss can be shown to have an average deviation from expectar tion of sbout yN. Let D = T E be we diferenceof deviation be- tween what you actually gain or lose (1), andthe exposed gin or Joss). Therefore, for 100 bess, the average deviation fan =< is about $10 Gn fact, the chances are about 68% that you'll be within $10 of even; theyre about 96% that you'l Be within $20 of ven) For ten thousand! betst'sabout$100and foramillion bets i's about $1,000. Table 241 shows what happens. For instance the last ine of Table L- says that if we match coins one rillion times at $1 per bet, our expected gain or loss is 210 (@ "tar" game). But On average, we'l be about $1,000 ahead ur bbchind, Infact, we'll be between +$1,000 and ~S1,00D about 68% ofthe time. (For a milion $1 bes, the deviation D has approximately a normal probability distribution with mean 2et0 and standard deviation $1,000.) We call the total ofthe bes in 1 series the “action,” A. For one series of one million $1 bets, the action is $,000000, However (ith column) D/A=0001, 89 ‘ Tis beswoon i ater Jit evorage size OD ie sxoected 0,01 and -0.01 200 and ~ 100The Mathematics of Gambling the deviation as percent ofthe action i very small. And about (68% ofthe Une T/A fs besween ~001 and 4-001 so asa per cent ofthe action the resul is very near the expecied result of zero. Note thatthe average siz of D, the deviation from the ex- pected result . grons—conirary io popular belief Hours, the Irerae sizeof the percenage of deviation, D/A, tends to 20, in agreement witha correct version of the “aw of averages" For $1 bets on Red at American roulette, the, resus appeat in Table 1-2. Note tht in the ast cota the spread in TIA ges cle and closer E/A = ~0526 This is where ‘We get the statement that i you pay a “long time" you'll ose bout 5.26% of the otal action Noe 00, in olin dat there appears tobe essandlesschance of being ahicad asthe number of tals goeson Infact, tcan be shown that in all negative expecta tion games the chance of being abcad tends to 2e0 a8 play ‘Using the concept of action, we can now understand the famous “lw of averages.” Thissays, oughly, thatif you makea dona series of bets and record both the action (A) and your toa profitorloss (1), then the faction T/A isapptoximatelythesame BsthefacionE/A where Eisthe otalof the expected gain o1ss foreach bet. Many people misunderstand this.” They think Ua it says the E and T are approximately the same ater a long sere of bets, Thisis false. In fact, the differen between Eand tends to get larger as A gets baer, ‘Now. the ordinary player probably won't make a milion $1 bes, Butte asin probably wise that mary and more. From the casinos point of view, it doesn't matte whether one player makes all the bets or whether a seis of players dos. In ether {be is profit inthe long runs assured and wil be very clos 1 5.26% ofthe ation. With many players, cach making some of the 100,00 bets, some may be lucky and win, but these wll senerally be compensated for by others wb ese more than the expected amount. For instance, if each of 00 players take turnsmakings hundred bets, Table I-2tllsusthat about 68% ofthe time ther result wil be between +3474 and’ $15.28, : -$51621,00 an -453601.00 | 0,001The Mathematics of Gambling About 16% ofthe time the player wins more than $4.74 ("“tucky") land about 16% of the time the player loses more than $15.2 (unlucky), But players cannot predict or control which group they'll bein. "This same “law of average” applies to mote complicated sequences of bets. For instance, suppose you bet $10 on Red at roulete (E= ~.53), then bet $100 on “players” at Baccarat (E= $1.06, then bet $10 on a hand ina single-deck blackjack ‘game where the ten-count is 1S ten, 15 others (E= +830). The total Eis $.53 $1.05 +8.50 = ~$.09, The total A is $10+$100 + $10 = $120, Ifyoumakea long series of bets and record Eand ‘Aas wea your gains and losses foreach one, then just asin the ‘coin matching example (Table 1-1) and the roulette example (Table 1-2), the fraction D/A tends to er0 s0 T/A tends 10 E/A. "Thatmeans that over, say, lifetime, your total lossesasa percent cof your total action will tend tobe very close to your total expecta- ‘on as a percent of your action. Tf you want a good gambling life, make positive expectation bets, You ean, as a fist approximation, think of each negative expectation betas charging your account withataxia theamount of the expectation, Conversely, each positive expectation bet right be thouzht of as crediting your account with aprofitin the amount ofthe expeciation. Ifyou only pay tax, you gobroke. If you only collet credits, you get rich. Chapter 2 Blackjack Blackjack, or twenty-one, is 2 card game played throughout the world, The casinos in the United States currently realize an anual net profit of roughly one bilion dollars from the game. ‘Taking a prcelearnings ratio of 1 as typical for present day com- _mon stacks, the United Stes backjock operation might he com= pared toa $15 billion corporation. "To begin the game a dealer randomly shufles the cards and players place their bes. The numberof decks doesnot materially affect our discussion. It generally is one, two, fou, six or eight ‘There are a maximum and minimum allowed bet. ‘The players’ hands are dealt afer they have placed ther bets. Each player then week ie choice of atratewy forimproving hishand. Finally, the dealer plays out hishand aocordingto a fixed strategy whic doesnot allow kil, and betsaresetled.Inthecase ‘where play begins from one completerandomly shuffled deck, an. approximate best siategy (Le, one giving greatest expected ‘etum) was first given in 1956 by Baldwin, Cantey, Maise, and MeDermot.The Mathematics of Gambling ‘Though the rules of blackjack vary sigh, the player follow. ing the Raldwin greup strategy typically has the tiny edge of “+-10%. (The pessimistic igure of ~ 62% cited inthe Baldwin's group's work vas erroneous and may have discouraged the authors from further analysis.) These mathematical reslts were in sharp contrast t0 the earlier and very different inuiive strategies ‘grerally eeommended by cant experts, andthe associated player isadvantage of two or tree percent. We cll the best strategy Sosinst a complete deck the hase srategy. Determined in 1965, itis almost identical with the Baldwin group strategy and it gives the player an edge of +13 against one deck and ~.53 against out decks. ‘ithe game were always deal from a complete shud derky ‘we would have repeated independent tras. But for compeling Practical reasons the deck snot generally reshuffled after each ‘oundof play. Thus as successive rounds are played from agiven eek, we have sampling without replacement and dependent ‘wal IKis necessary to show te playersmosorallofthecardsused ‘ona given round of play before the place their bets for the next ‘ound, Knowing that certain card are missing from the pack, the plier ean, in preciple, repeatedly recalculate is optimal strategy and his corresponding expectation. (The strategies for various ‘ard counting procedures, and tei expectations, were detemined ) ¢-2 914% ez 1am,The Mathemaics of Ganbing sure that his win rat ten uns per hour and his average bet Seep ren tone munpony bing ced See eeu or onesenh ofthe Ue would cane his advan- Bie BE cheated more than ten pecans othe me woud prob- Symon bininto aloe Sar i he rcal wordisprobebly moreefTectv than inthe typothettal example just ced, because Ue calculations fr that ‘Rape assure ceaing i equalyUkly fr small bes and big Seas my experience, the bettor Se auch more Healy to be hesied on lage bes than on smal ones. Therefore, the dar Sho cheats wi maou effeney wil wait uni a player Thakes is top bet, Supose that bet totals five nis. the cheat Shifts the odd to 80 percent in favor Of the house, ie expected Joi 2-1/2unis, and just four cheatingeffortsperCOhands wil ‘acl a profesional players advantage. A cheating rate of five ‘Sen hand er 100 wi pt his player ta sveredsarantage ‘We can sce fom ts that a comparasvely small amount of hating applied tothe larger handscan havea ignificant impact. ‘onthe game'soutcome. Ts ives youan isa of what look for Athen you aren he easnos and ink tat someting may be Missing Cards: The Short Shoe have heard complaints that cards have been missing from the ‘pack in some casino blackjack games. We'll discuss how you ‘ight spot this cheating method. In 1942, | wrote on page SI of Beat the Deoler, “Counting the. .card, is aninvaluabe asst in the detection of cheating because a common devices to remove one or more ards from the deck." Lance Humble discusses cheating methods for fourdeck ‘games dealt from a shoe in his Intemational Blackjack Club nevaleticr. He say, “The house can take certain cards such a8 tens and aces out of the shoe. This is usually done after several rounds have boen dealt and after the decks have been shuffled several imes, Its done by palming the cards while they are being 2 Blokiact Shute and by hing them on he dealer's person, The dealer then disposes ofthe Eas won he goes of hs teak” Bat cheating this way in ited othe cain, Players have been known to remove small” cards from te pack it he edge ther way The casino can sp this ply ty aking the pack aed cue isthe player wunly bs tose satsical methods. 1 the cheating ads, the method is known asthe stot shoe “Let's say the dea i dealing fom a shoe containing four docks of sats each In 52 candy tere shoud be fetal cars ‘the tens, jacks, quoens and kings. Logical in four decks of 208 cards, there should be 64 eneaue cards eal al ofthese ‘ens from now on. Casino rarely trinowe the acre novice layer sometimes count these Suppose the shift bos or pit boss ake uta total often tes; some of each king, ofcourse, ntl kings or quens. The shoe's Shortened fu 6 es 1034 tens, and he four decks from 208 cards to 198 card “The lossof these entensshitsthe advantage fomtbeplaerto the der or house The atin steven hae othe onnal 176 =2.35 to 144/S4=2.67, an this pamsaliticover One percent forthe house. How can you discover thelack of tens without the dealer knowing i? lve une eto that sed ou're playing at te black jacktabl sin the st chairon the dae’ ight. Beta sal fied mount throughout whok pack of four dest. After the dealer Putte ct card back ony ay, tn reno the way nto the four shuffled decks and returns the dks into the shoe then ready youself to coun the crs, Play your band mechanical, only pretending intrest in your good or bad fortunes. What youre terete is fding oUCs te amber of tensa ie OE Fourdesk shoe. Lessa theshit boss ha emovedten tens. (Report are that. they seem to love removing exact ten frm sfonde shoe) ‘Winn he white cut card shows atthe face ofthe shoe, e's ay thatthe running count of ters has reached 2. Tat means mathematically tha if all 6 tn-alue card wer inthe shoe, aThe Mathematics of Gambling then, ofthe remaining 15 cards behind the cu card, as many a3 12 of them would be teas, which mathematically is very unlikely. This is how one detects the missing ten tens because the dealer never shows their faces but just places them face down on top of thestack of diucarded cardstohisright, which then proceedsto shuffle face down inthe usual manner preparatory to another ourdeck shoe session. Although at first the running counts not easy to keepin areal ‘casio situation, a secondary cificuly is estimating the approx- mate number of card let behind the cut card afterall the shoe hnas been dealt. To practice this, take any deck of S2cards and cut off what you think are ten, 18 or 20 cards, commit yourself to some definite number, and then count the cards to confirm the closeness of your estimate. After awhile, you can look ata bunch ‘of cards cut off and come quite close to their actual number. Insurumary, count thenuumber of tens seen from the Desig of a freshly shuffled and allegedly complete shoe. When the last card is seen and it is time to reshuffle the shoe, subtract the ‘Bumber often sen from the number that are supposed to bein the shoe—68 fora four-deck shoe —to get the numberof unseen ten-value cards which should remain. If $8 ten-value cards were seen, there should be ten tens among the unused cards. Then estimate te numberof unseen cards. You have to besure toad fo the estimated residual stack any cards which you did not see during the course of play, suchas burned cards. Step fouristoask ‘whether tne numberof unseen Lenvalue cards is remarkably large for the numberof residual cards. If 50, consider seriously the possibility thatthe shoe may beshort. For instance, suppose there are IS unseencards, ten of which ae supposed tobe ten-values. A ‘computation shows thatthe probably that the last 1S cards of @ ‘wellshufed four-deck shoe will ave a least ten en-value cards {5 0.003247 or about one chance in 308. Ths the evidence against the casino on the basis of this one shoe alone isnot overwhelming. Butif we weretocount down the same shoe several times and each ime were to find the remaining cards suspiciously ten-tch, then the evidence would become very lacjac ‘rong. Suppose that we counted down the shoe four times and ‘har each time there were exactly 15 unseen cards Suppove thet the number of unseen tens, assuming a ful four decks, was nine, 1, ten, and [3 respectively. Then referring to Table 2-4, the pro- bbls to six decimal places are H() = 014651 wo have nine ‘or more unseen ten, and fr atleast I, ten, and 13 respectively, the chances are Hl) = .000539, Hd0) = .003247, and H(3) = 000005. These correspond io odds of about 168, 1855, 1308 and 17200000 respectively. The odds against all hese ever hap- ‘ning ogether is much greater sil. In this example, the evidence ‘Strongly suggests that upto nine tele cards are missing, There can't be more than nine missing, of course, because we say all ‘but nine on one countdown. ‘Ifthe casino shufls after only 104 cards are seen itis not so cay tote if ten en-au cards were removed. A mathematical roof of this fs contained in the “This discussion should make t clear that the method suggested is generally not able to easy spot the removal often-value cards| ales he shes counted sea tines or del down cs 10 ‘One of the interesting ironies of the short shoe method of cheating playersis that neither the shift boss nor the ptboss—the laser bringing the decks of cars wo the dealer's uble—ued ell the dealer tha isshoeis shor. Thus, the dealer doesn’t necesarly hhavetoknow that he'scheatng. Afterall he's just dealing. I'san {pen question how many deals know ta the dling from ‘Reports are that the short shoe is a frequent method that casinos use in cheating at blackjack using more than one deck. “The tables with higher minimams (say $23) are more tempting ‘candidates for short shoes than those with the lower minimums. ‘An experienced card counter can improve the method by coun= ing hoth tens and non-iens. Then hell now exactly Bow many ‘unseen cards there are, as well as unseen tens. Table 2-4.can then ‘be used wit greater confidence. Inpractce, you don’t need to count through ashe while bet- 2 see Arent peThe Mathemats of Ganbing Table 24 ting (and dhus losing money inthe proces) to find out that the ‘asin is cheating. Ifyou suspect foul play, count whe sanding ‘ehind the player tothe dealer’ ight, "You might easly catch a short shoe by simply counting ll the cards that reused, whether or not you see what they are, Then if the remaining cards, atthe reshufl, are few enough so you can 6 Blactjack ‘accurately estimate their number, you can check the total count. For istance, you count 165 eards sed and you estimate dat 31 ‘& 3cards remain. Then there were 196 = 3 cards athe than the 206 expected, o the shoeis shor. ‘Acting countermeasiteisto put hack a4, $or6 for each ace or ten-value card removed. Then the total number of cards ‘emains 208, and thecasno getsan even greater advantage thanit ‘would froma short shoe. ‘Cards do get added to the deck, and there's u spooky caine cidencetoillusirat this. On page SI of Bea the Dealer, rotein 1942, “One might wonder a this point whether casinoshave also ‘tid adding car tothe deck. have only seen it done once. Iis very risky. Imagine the shock and fury ofa player who picks up his hand and ses that not only are both his cards Ss, but they are also both spades." And then 15 years later in 1977, a player ina ‘one-deck game did getahand with two ofthe same Garde Sot Spades. Walter Tyminsk’s casino gaming newsleter, Rogue et ‘Noir News, reported on page3 ofthe June 15,1977 issue, “What ‘would you dof the player at your ight ina single blackjack game hhad two $ of spades? Nicholas Zaika, a bail bondsman from ‘Detroit had tht experience at the Saharan Las Vegason May 24 ata $5 minimum table. “zatka wasnt inthe best of humor because he had reportedly lost $594,000 at other Sabara tables, ty far the largest loss he has ever experienced. Zaika had the blackjack supervisor check the carde and there were $3 cards in the deck, the duplicate be- inthe 5 of spades. The gumer has engaged the services of Las Vegas attomey George Grazadei to pursue claims he feels that be has against the casino. The Sahara denies ayy wrongdoing and says that it i cooperating fully with the investigation... Players arentlikely 10 Introduce an exzaS because the presence ofthe extra’ favorsthe Ione and not the player” ‘Suppose instead o ust counting tens used and totaleard wed, ‘you kept track of how many aces, 25,35, queens, kings, andso on ‘were used. This extra information should give the player a better ”‘The Mathematics of Ganbling ‘chance of detecting the short shoe. The ultimate proof would be to count dhe number of each of the 5 types of eas which have been used. Mathematical readers might wish to investigate effec- tive statistical of other ways of using information for detecting shoes in which the numbers of some of the cards have been changed. 2s Baccarat “The games of baccarat and chemin de fe are wel nown gan ‘bing games playee fr high stakes in several pars of the world. Baccarat ssid to be a card game of din origin that was troduced into France shout 1190 A.D. Two forms of he game ‘evelped One form wae ciled tocar andthe ober ws alld teminde fer, The mos bac diferene betwen hese two games is simpy that dhtee hands are dealt n bacarat (called bocarat en bang in Englare) ad two hand ae dealt in chemin de fet (Cale acearaschemin de fet in England and Nevada}. The cards ae tough aine are each wort ther face vale and the cards ten, jack, queen and king ae each worth 20 pins. ‘Ahan is evluniad asthe um modo tm of i card, 8, om iy the las digit ofthe (oui counted. The objet ofthe game isto be as close to eight or aine as posible with two cards, oF 2s close to nine as posible wih at host thre cards fone does Mothave eight or nine on is rst two cards, Then Ge high hand “The games of baccarat and chemin defer Became popula in »‘The Mathematics of Gambling public casings all over Europe, as wells in private games, about 4850. Arthe present time, one or both ofthese games are well known in London, southern France, the Riviera, Germany and the United States. A form of chemin de fer, which we shall cal [Nevada baceart, bas boon played in afew Nevada casinos since 1938 ‘The rules, structure and format of the three games have strong. similaritios,T studied Nevada baccarat with William E. Walden iment intensively beeause the casinos whee ts played were reali Iy accessible. Our techniques can be carried over tothe other forms of baccarat and chemin defer ‘We were originally motivated by the abwervation that baccarat ‘and chemin de ferbave several points of resemblance tothe game Of blackjack, or twenty-one. ‘The fact that practical winning strategies for twenty-one have been discovered suggested that there might also be practical winning strategies for baccarat and chemin de fer In contrast woth situation in twenty-one, we found that there ae no curcent practical winning strategies forthe main part of tho game, ie, for the money Ranker and Player bets. Rules and Procedures ‘To besin the Nevada baccarat game, cight decks of cards are shuffled and a joker is placed face upnear the end, ibe cards are then pa into a wooden dealing box called ashoe. The fist cardis ‘exposed, and its values noted, face cards being counted as tens. ‘Then this uuniber of cards is discarded, or “bumed.” "The table has twelve seats, occupied by an assortment of customers and shi. sili a house employee who bets money land pretends t0 be a player in order 10 attract customers of ‘Stmulae pay. We refer to them indiscriminately a “player.” ‘There are wo principal bets, called “Banker” and “Players.” ‘Any player tray make either ofthese bets before the begining of fay round of play, oF “coup.” To begin the eveing’s pay, two ofthe players are singled out, ‘Oneis termed The Banker and the others termed The Player. The seats are numbered counterclockwise from one totwehe. Player 0 Becca ‘number one i inially The Banker, unless he refuses. n this case the opportunity pase vounterlackwise aroun the table wn someone accepts. The Players generally chosen tobe that player, other than The Banker, who has the largest bet on the Payer. We hhave not noticed an oovasion when there were no bets on The Player. When we played, there were shils in the game and they ‘generally bet on The Plver (except when acting as The Banker, ‘when they generally bet on The Banke). “The Banker fais the shoe and deals as long as the bet “Banker” (which we also refer to asa bet on The Banker) does not ose. When the bet “Players” (which we also refer to as abet ‘on The Player) wins, the shoe moves to the player on the right. ‘This player now becomes The Banker. I the coup isa tie, the players are allowed to alter their bes in any manner they wish. ‘The same Banker then deals another coup. To begin a coup, The Danker and The Player are dealt (wo ‘ards each. As we foted above, the cards ace through nine are cach worth her fae value and tens and facecardsare each worth zero points. Only th last distin the totais counted. ‘After The Banker and The Player each recive two cards, the ‘roupier faces their hands. If ether two-card foal equals 8 or 9 (Cermed a natural 8 ora natural 9, asthe case may be), all bets are sealed at once. If nether The Player nor The Banker have a natural, The layer and The Banker then draw or stand according to the set of| rules in Table 3-1. ‘The high hand wins. Ifthe handsare equal, thereisateand no money changeshands. Plaersare then freeto change their betsin any desired manner. Ifthe coup being played is complete when te joker is reached, the shoe ends and the cards azereshuled. (Otherwise the coup is Fist played out to completion. ‘Then the she ends and the card ae reshuffled. However, the casino may reshule the cards at anytime between coups.The Mathematics of Gambling Table 341 Player having 05 draws acare 67 stands 89 tums cards over Bankerhaving draws when does not draw when Te Player craws The Player draws o none, 0.8 1 one, 09 2 one, 08 3 one, 07,9 8 4 none, 27 01.89 5 one, 47 0388 6 67 ‘one, 06, 8,9 7 stands stands 8 tums cards over tune cards ovor 8 tums cards over tums cards over ‘The Main Bets ‘Two main bets agsinst the house can be made. One can bet on ther The Banker or The Player. Winning bets on The Player are paidat even money. Winning betson The Banker arepaid0.95of fhe arnount Ux. The five percent tax which i imposed on what ctherwise would have been an even-money pay-off is called ‘igoish.” For eight complete decks, the probability that The ‘Banker wins is 0.458597, and the probability of a ties 0.095156. “The basic ida ofthe calculation ofthese numbers isto consider all possible distint six-card sequences. The outcome for each sequence is computed and the corresponding probability of that 2 acca sequence is computed and aocumulated in the appropriate regter. Numerous shortcuts, which simplify and abbreviate the caleulation, were taken. “The house advantage (we use advantage as a synonym for mathematical expectation) over The Player is 1.2351 peoent. The house advantage over The Banker is 0.458597X5 peroent—1.2351 percent or 1.0579 percent, where 2.2980 percent {sthe effecivehousetax on The Banker's winnings tes azenot counted as trials, then the figures for house advantage should be multiplied by 1/0.904844, which give a house advantage per bet that is nota ti, over The Banker of 1.1692 percent and over The Player of 1.3680 percent. The effective house taxon The Banker in this situation is 2.5341 percent. ‘Weattempted to determine whether or not the abnormal com- positions of the shoe, which arise as successive coups ae dealt, sve rise to fluctuations in the expectations of The Banker and “The Player bets which are sufficient to overcome the house ede. Itturns out that this oocsionally happens but the fluctuations are notlarge enough nor frequent enough tobe the basis of apractical ‘winning strategy. This was determined in two ways. Fist, we varied the quantity of cards of a single numerical value. The results were negative. "We next ingured as o whether ifone were able to analyze the ende-eck perfectly (eg. the player might receive radioed instruc- tions from a computer), there were appreciable player advancages con ether bot aignificant part ofthe time. We selected 29 xs of 13 cards each, each set drawn randomly from eight complete decks. There were small positive expectations in only two Instanoss out of 58. Once The Player had a 3.2% edge and once ‘The Banker had a0.1% edge. "Wenext proved, by arguments too lengthy andintricateto give here, that the probability distributions describing the conditional expectations of The Banker and The Player epread out at the ‘number of unplayed cards decreases. Thus there are fewer advan- tageous bets of each fype, and they are less advantageous, a the numberof unplayed cards increases above 13. The converse oc- aThe Mathematics of Gambling (0102 1101 yoed uy siUjed Os 44 0} Zt Woy peonpas AUER. ° e eo00- 1 = soo0- i e- e000- z- ue t100- (Panyu) 7-¢ 21%, I e- 2000- ars n100+ so00+ 00+ sroo0+ ewzono+ wunog weg (UexueA)SonIeA tag seYUEG erewpoxdyuy —wodexnejey — yo Be IUENY (0g 181g) eHEORG Te TR coor scoo+ oot Lioo+ 2000+ zioo- 2000- sooo wnoo- we000- yea sofia uy eb6ubuD en eoThe Mathematics of Ganbling curs asthe numberof unplayed cards decreased below 13. "Tne observed practical minimum ranged from eight 0 17 in ‘one casino and from 20 up in another. The theoretical minimum, ‘when no cards ae burned, is six. Thus the resis for 13 unplayed Cards seem 10 conclusively demonstrate that no practical win- ‘ing strategy is possible forthe Nevada game, even with a com puting machine playing a perfect game. "To see why, consider te accompanying Table 3-2 (pp. 4-35), based on Table 2 of Walden's thesis rom this Table we se the effect of removing one of any card from the eight-deck baccarat pack. Proceeding in the way we developed the theory of Blackjack, we get relative point vals ‘which are sted in the next to the lat column. The last column ives a simpler approximate point count system. ‘We would now like to know how powerful a point count system in baccarat is compared with porst count systems in black- jack. Todo this we compute theroot mean square (RMS) value of the cohumn called “Change in Advantage of Banker Bet." ‘We do this by equaring each of those numbers, ecunting the square for zero-value cards four times because there are four times as many. Then we add these square, divide by 13 and take the square root. The resulting root mean square or RMS valueis (O64, That measures how lat the deck shifts [roms base tar ing value fora full pack. “Taking one card ofa given rank (we think of there being 13 ranks) changes the fraction of ay ofthese 18 ranks by an amount 532/416-31/415 which equals .00222. If we divide the RMS value by this value we get (288 asa measure of how rapidly thisadvan- tage ofthe two bes shifts from the starting value as the compas tion ofthe deck changes. 'Now we aze going to compare this with the situation in Diack- jack, Table 23, for one-deck blackjack, can be treated in the Ses way to aus how far the advantage changes in blackjack. ‘The Table is from Peter Gifin's book, Theory of Blackjack Revised, page 44, Weget RMS valu of 0.467%. Thecorrespond- ing change inthe fraction ofa single rank when one cardis drawn, 6 Baccarat Table 3.3 (One-Deck Blackjack Subtract One Card of Value | Change In Advantage of Player -081% 035% oases 055% 059% 046% 028% 00% cor oanon> 18% 51% 15 4/52/51 or 0181. The ratio of RMS value to this value is 2580, If we divide this by the corresponding resut in baccarat ‘wewet 8.97, which ell usthatasthe ruecount in Blackjack varies the change in player avantage or disadvantage shifts nine times as fas in blackjack as it does in bacaral. aThe Mathematics of Gambling [Note that dividing the RMS value by the “change in fraction” ‘ofa single peck adjusts the one deck blackjack figures and the ‘iehcdeck baccarat Figures so they are comparable. If we had tsed, for example, an eight dock blackjack table instead, we still ‘would have hada final ratio of about nine times. “This allows us to translate how well a point count in baccarat ‘works compared with onein blackjack. ln baccarat we start out ‘with orethana 1% disadvantage and with eight decks. Imagine f blackjack game with an eight deck pack and a 15 disadvan. tage. Now imagine play continues through the blackjack deck. “The blackjack deck advantage from a ~ 1%e starting level might, ‘on very rare occasions shift 9% to a + 8% advantage. ‘As often as this happenin blackjack would be the approx imate frequency with which we would get 1/9th as much shift in baccarat; meaning froma ~ 1%) advantageto OMe advantage or break even forthe banker bet, Since there are two bets, banker and player, the player bet would also be breakeven or beter about as often. "The conclusionis that you night expect to breakeven or better ineight dock baccarat about twice as often a you would expectto have an BM edge in eight deck blackjack. How often would you hrave a 1%) advantage in eight deck baccarat? About twice as ‘oftenasyou would gota 7Mscdgein blackjack. Theobvimsean- ‘lson sthatadvantagesin baccarat are very small, they are very rare and the few that occur are nearly always in the lst five 0 20 cards inthe pack. ‘The Tie Bet In adtion to wagers onthe Player or Banker hands, the casinos anker and Player hands Ihave the same total, this bet gains nine times the amount bet. Otherwise the bet i lost. The probability ofa tie is 9.51565, hace the expectation ofthe bet is ~4884%. tis ear, however, thatthe probability and thus the expecta- tion of a tie depends on the subset of unplayed cards. For in- stance, in theexireme ard improbable event that theresidual deck 7 consists solely often valucards, the probability ofatieisequlto tneand the expectation inne Thveard etiing alginate Potentially advantagsous. ‘Using computer mulation random subs of diferent sizes vere selected from a complete 16card (ht deck) pack. The Fesuls. were dsappoting fom a money-making perspec= tivemthe advaniages which oocur with complete knowledge of ‘he wed cards aeliited tothe extreme end ofthe pack a are generally not large. Practical card counting trates area best ‘urgn, and at bes precarious, forthe are easly eliminated by shuffling the dock with 25 cards remaining.Section Two The Wheels I doest’t require an extensive mathematical back: (ground to look at the 38 identically-sized spaces on an American roulette wheel (note the 354 payoff on a single number) and conclude that the game is Unbeatable. With @ 1/98 chance of having a number come up on the next spin and the 354 payoff, tis easy to caloulate the often-quoted expectancy of the player of 8.26. The odds for other wheels, especially tne Wheel of Fortune, appear even more against the player. The unbeatabllty of the roulette wheel Is based on the mechanical perfection of the wheel—auch a con clusion is based on the assumption that the ball has, an equal chance of landing in each pocket. This may ‘or may not be true although Allan Wilson, in The Casino Gamolers Guige, and others glve fairly convincing evidence for the existence of blased wheels—wheels. sufficiently biased to overcome the house advantage. The very mechanical perfection of the wheel, however, would suggest the applicability of the laws. ‘of physics to prediction of the next number, whether the game is roulette or the Whee! of Fortune. Just as the future position of a planet can be predicted quite ‘accurately, So can an understanding of the physical laws at work minimize the uncertainty surrounding the a‘The Mathematics of Ganbling resting place of the ball or the final position of the wheel. Its not possible, of course, to obtain an exact predic tion, But this Is not absolutely necessary to assure a profit. As Marvin Karling has pointed out in his book Payching Out Vegas, “Simply being able to predict which half of the wheel the ball will plunk into would the player such a whopping edge that he could go Yor the chandeliere....and make it? The following two chapters investigate the promise of this approach to beating the wheel as well as discussing some of the difficulties that might arise im plementing such a strategy in the casino environment. Chapter 4 Roulette It was the spring of 1955. 1 was finishing my second year of ‘graduate physics at U.C.L.A. In the course of the next year 1 ‘would make three decisions that would shape my life forthe next, 28 years. married (my present wife, Vivian), Lehanged my field ‘of stody from physics to mathematics, and I bopan 10 toy with the fantasy that Icould shatter the chains of poverty through a scietifially-based winning gambling system. 1 was living in Robison Hall the student-owned cooperative. ‘For $50a month and four hours work a week, we got our foom, ‘and board. I had lived in the co-ops for neatly six years of ‘undergraduate and graduate work, ona budget of about $100 a month, Part of this came from scholarships and, in the early years, 1 got some help from home. But {was basicaly self Supporting like most ofthe other 200 or so co-op resident Tettended clases and studied from $0 to 60 hours a week, generally including Saturdays and Sundays. had read about the psychology of learning in order to be able to work longer and. harder. found that “spaced learning” worked well: study for an “6The Mathematics of Gambling hour, then take break of at east en minutes Ghower, med te, tianda; te). One Sunday afternoon about 3 rm Teameto the oop dining room fora tea break. The sun was steaming through the ig glass windows. (Robison, designed by Richard Neutra inthe 30s, as very alia fr that ie. K had so mary big sheet glass windows that it was on called “the glass house) My head was bubbling with physis equations, and several of my ood friends were siting around chating nur muta! poverty the converaton ready tured (0 fan- tases of ey money. We began lo spaulae on wheter there teas wy fo eat he rouleue wel In adion tome, he group included math majors Mel Rosenfeld and Andy Bruckner (00% profesor of matherais at U.c. Sania Barbara), Tom Sct, and engineringmajorRck Rushall. After al these years it'shard tobe sre of exactly who sid what, but we began the discussion by acknowledging that mathematical rsteme were impossible Tildemonsrate isin a future chapter. “Then we kicked around the de of wheter erouies could control where the ball willand well enought significantly affect, the odd. [wil show later hat is simpossibe unde he wal coaltons of the game. (The incredible thing is that logical Feasoning could even be used to sete sucha question) twas a ‘hort brainstorming sep to wondering whether wheel were i= Parfet enough o charge the odds to favor the player. Thos in {he group who "ktew" asured me tha the whel ae veriable Jenled watches of perfection, carefully machined, balanced and Mabnlaned. ‘This 6 fase. Whees af somettnes ampetet ‘enough the can be beaten, Thad no experience ith gambling, ‘oF with casinos, or with roulette wheels so T accepted the ‘echanial perecon of route wheel ‘But mechanical pereson, fora physics, means predictably. ‘oucan'thaveit beth way, argued. If tbesewhees are very perfect the odds wil change enough so we can beat ther. I they ie prfet enough we ean pret n proce) approxmatey ‘where the ball willand, Suddenly the orbiting roulette bal eet lke the planetsinther stately and reise, prdicablepaths. In “« oul ‘ymin there was thatinutve “cick” of discovery that would experience again and aga. Unknowingly, Thal just when Uae fist step on a long journey in which I would discover winning ‘stems suchas those for Blackjack and for the options market, and I would accumulate a wealth I never imagined. ‘One side argued thai isa long way from prediction in princ\- ple to practical prediction. My group said that over and over, the sory of scence has been a rapid lp from a theoretical vision (E=MC) to an unexpecied pracucal result (oulear power plants. By now our inital group of people agreed tha the idea had merit and might well work. The novel debate attracted listeners, some of them cyical. They challenged us to prove the ‘dea worked. The ten minute “study break” had runinto acouple of hours. We adjourned with the half definite idea of “doing something. Tn the following works Ute idea kept couig back (o ae: ‘measure the position and velocity ofthe roulette ball ata fixed time and (maybe) you can then predict its Fture path, including ‘when and where the ball willspral into the rotor. (The rotors the spinning circular central disc where the bal finally comes to rest in ‘numbered pockets.) Also measure the rotors postion and velocity ata (possibly diferent) fixed ime and you can predict the rotors Totton forny future ime, But dix you wil kaow wha section ‘ofthe rotor wil be there when te ball artves. So you know (ap- proximately) what number will come up! ‘Youcan see thatthe system requiresthat bets beplacedafterthe balland rotor are set in motion and somehow timed. That means that the casinos havea simple, perfect countermeasure: forbid bets after the balls launched. However, I have checked games youghout the world, including Reno, Las Vago, Loudouy Venie, Monte Carlo, and Nice. Only ina few eases were bets for- bidden after the ball was launched. A common practice instead was ta call "no more bets” a revolution of two before te ball dropped into the center. ‘The simple casino countermeasure meant that there were {60 problems: (find out whether exact enough predictions could be “6The Mathematis of Gambling ‘made o get a winning edge, first in theory and then nthe casino itself, and (2) camouflage the sysem s0 te esis would be unaware of it use, If we could solve the prediction problem, the ‘camouflage was easy. Have an observer standing by the wheel ‘ecoeing the nuraers that came up. aspart of system.” Many do thissoit doesn't seem out of place. But the observer also wears concealed computer device with timing switches. Hs eal job is to time the ball and rotor. (Much ister we setiled on toe operated Switenes, leaving bot aun fre ud the open.) The computer Would make the prediction and transmit it by radio tothe bettor ‘The bettor, atthe far end of the layout, would appear to have no ‘connection tothe observertimer. The bettor would have a poor ‘view of balland rotor and would not pay much attention othem. ‘To futher break any ink between timer and betor, I would have several ofeach, with identical devices. They would each comeand ‘0 “at random, ‘The important bets have tobe place after the balls launched. A betior who only bet then, and who consistently won, would soon become suspect. To avoid that, planned to have the bettor aso make bets before the ball was launched. These would be limited so ther negative expectation didn't cancel all the positive empectation of the other bets. I became a radio amateur (ONGVVND when Twas 13 back i 1945 wea thee ween" easy ‘ovioe-lass test), so I thought I could build the radio ink and other electronic gadgetry. “This left me with the prediction problem to solve. More than a year passed without much time for roulette: I got my Master's Aegree in Physics June 1955)and wrote the ist part of my Ph.D. thesis on nuclear shell structure (Mayer-Jensen theory). The ‘mathemaseal problems that fran inc le mein Ue fll of 193300 take graduate math courses. I needed so many that I got my PhD. in math instead! And carly in 1956 T got masted. [had been working 2¢ 2 tutor and one of my “students” was T:T. ‘Thomton. He was an independently wealthy, knowledge-oving bachelor of about 4, who had degress in English and chemistry. "Now he was geting a degree in mathematics, jus forthe pleasure “ Rosle fit, He was an excellent student who didn’t need a tutor but had fred me simply to lear faster and more efficent. We shared bits and pieces of our hopes, dreams, and enthusiasms. ‘After Ihad mentioned the roulette project, Iwas suprised and touched by his gift of a halPsized wheel. it was black plastic (Cake?) made in France. [learned later that it cost the enor- ‘mous sum of $25. Though I had thought about the roulette system off und on, hei oF tis wld (oretaneut 1938, Lvl) got me to work more seriously on it. My fis idea was to se a Fhome movie camera to film the orbiting bal. I then plotted the ‘amount the ball had traveled versus the numberof the frame of the flm. expected thatthe pictures were taken at auniform rate ‘of 24 (2) frames per second s0 T could plot (angular) distance traveled versus time as in Figure 41. Instead of a smooth graph Tike the solid lie iu Figure 4-1, wy fast fn showed pevular| ‘wavy structure, lke the dashed line. ‘After thinking aboot this, I guessed that this was because the ‘camera didnot runat uniform speed. BY takinga movie of aston- ‘watch that timed in hundredths of a second, I found thatthe camera did vary in speed. Photo stores confirmed this. The distortion of the curve in Figure 41 is analogous tothe way a ‘musical tones disoried bya phony tunable whise speed vats ‘My uext move wasto take a movie ofthe rotating ball and the stopwatch. This gave me an accurate ime foreach frame. (still hhave aol ofthese pictures, postmarked January 16,1959.) But there was stil some “ripple” tothe curves, (Later Feared that ‘evea igh tt would eause this.) Worse, I found tha the curves ‘Were not consistent fom spot 10 spin. The sluwation was something like Figure 42. Thismeant the ball behaved diferently {rom spin to spin, This meant that the distance it raveled varied even with the same intial velocity. This doomed predicabilty fon my wheel. ‘found with further experiments that my half-ied wheel was really very ieregular. The track Was curved lke a tube andthe bal, "‘The Mathematics of Gambling Th amy Pp aunty,‘The Mathematics of Gandling “rattled around” eratcally, upand down, asi orbited. Thestick ‘bake surface was moulded, not machined. Theballalso skidded and bounced. And there was a horizontal junction which added Inveglaries to the track. But follsined wheels were not He that. In December 1988, 1 ‘made my fist visit tothe casinos. | observed several regulation ‘wheels and found thatthe ball moved smoothly in is track. Also thetrack was a pair of flat-beveled, carefully-machined surfaces, ‘nota tube. When [Saw how good theeasino wheels were, I was ‘more convinced than ever that prediction was possible. But I ‘ede full-sized whee and some good laboratory equipmentto ontinve. How cowl Fay fori? Igot my Ph.D. i June of 1958 fand was teaching at UC.L.A. Though my’ wife was finally able to stop working, we had no savings and I barely supported us. couldn't ask her to go back to work to buy me a roulette wheel and to france my pipe dren. ‘But! persisted, Isimlated the study of the problemof whether theroultteball would, forte same starting velocity, travelabout the same dicinoe along the track. I set up a litle veeshaped incined tough. 1 would start a marble from a fixed height (a ‘mark onthe trough) and measure how far across the flor it roll fed, I was encouraged but not surprised to find that te distance ‘the marble went could be predicted closely fom the suring “One memorable evening when my in-laws were due for dinner, {ran overtime on marhle experiment. They came into the kit- chen wondering why I hadn't come to gret them at the door. ‘They found me rolling marbles down alitle wooden trough and acrossthe floor. Allover the floor werelitle distance markersand pees of tape. Tn early 1959 Vivian and T spent time with Mel and Judy Rosenfeld, working on radio link for the casino test of my yetto becompleted roulette stem. Wetook model airplaneradio con- trol equipment and altered it somewhat. We succeeded in getting ‘ workable but somewhat inconvenient radio link. “Then around March or April of 1959, pushed the roulette Dro- 0 Rovere jet aside. Twelve man years of blackjack calculations arived, couresy of Baldwin, Caney, Masel sx MeDeimott T taal convinced myself (as described in Beat te Dealer that I could ‘devise a winning blackjack card counting ssstem and now Iset to ‘work on this intensely. The impractical marble roller now sid he ‘could beat the casinos a blackjack. What nex? ‘wrote my blackjack computer programs in the summer and fal of 1989. Testing, then debugging followed, and then from late 1959 through early 1960 ay eotapucer production ruts produced he bas results that gave me the fve-count sytem in early 1960. ‘Then during 19401 worked out most ofthe ter-count system and the ideas forthe ultimate stratepy.Lalso made the computer runs and worked out the methodology so that all of today’s so-called “one parameter” blackjack systems could be readily devised by anyone versed inthe use of computers. In December 1960, The Notices of the American Mulhematical Souety carted che abstract of my upcoming talk, “Fortune's Formula: The Game ‘of Blackjack.” Life would never be the same again. The intense professional and public interest aroused by the abstract, even before the talk, led me to seek quick publication in a scientific Journal. I chose to try the Proceedings ofthe National Academy Of Sciences I needed amember of the Academy to communicate (Qe. approve and forward for rexouuseaded publication), 80 T fought out the one mathematics member of the Academy at M.LT,, Claude Shannon, Claude Shannon: Genius ‘Shannon, then in his early fortes, was ard is one ofthe most famous applied mathemativians ir dhe work, As owe genius among many, he was relaively unnoticed as a graduate student—unti he handed in his master's thesis. It developed the rrathematical thenry of switching electrical networks (et telephone exchanges) and became the landmack paper inthe sub- Jest. After receiving his doctorate, Shannon worked at Bll abs for several years and then became world-famous for papers a‘The Mathematics of Gambling establishing the mathematical foundations of information theory. T was able to arrange a short appointment early one chilly DDeceanber afternoon, But the seeetary warned me that Shanon ‘wasonly going to bein for afew minutes, not to expect more, and that he didn't spend Eime on subjects (or people) that didn’t ‘neret im (enlightened selPinerest, I thought to myself Feeling both awed and lucky, I arived at Shannod’s office for my appoinuneat, He was @ thnnish alert man of middle height and build, somewhat sharp featured. His eyeshada genial crinkle andthe brows sugzested his puckish incisive humor. 1told the blackjack story briefly and showed himmy paper, Wechanged the tide from ““A Winning Strategy for Blackjack” to “A Favorable Strategy for Twenty-One" (more sedate and respct- able, I reluctantly accepted some suggestions for condensation, land we ngroed iat Pd send him the retyped revision right away for forwarding tothe Academy. ‘Shanon was impressed with bth my blackjack results and my method and cross-examined mein deal, both to understand and to find possible flaws. After my few minutes were up, he pointed ‘ou in closing that I appeared to have made the big theoretical breakthrough on the subject and that what remained to be lscovered woul be ane in the way of deals and clsboration. ‘And then he asked, “Are you working on anything else in the ‘gambling area?” T decided to sill my other big seeet and told him about roulette, Several exciting hours later, as the wintery sky turned dusky, we finaly broke off with plans to meet again on the ‘ouletle project. Shannon live in a huge old three story wooden house on one of tie Mystic Lakes, several miles from Cambridge. His basement was a gadgetcer's paradise. It had perhaps a hun- red thousand dollars worth of lecironic, electrical and ‘mechanical items. There were hundreds of categories, like ‘motors, transistors, switches, pulleys, tools, condensors, transformers, and on and on. ‘Our work continued there, We ordered a regulation roulette 2 Rosle wheel from Reno and assembled other equipment including (frost imporur) a sirobe Het and a lage cock will a second hand that made one revolutionin one second. Thedial was divided into hundredths of asecond and stil finer time divisions could be ‘esimated closely. Weset upshopin “the billiard room.” where a massive old dusty late billard table made a perfect solid stable mounting forthe roulette whee. Analyzing the Motion ‘My original plan was o divide the various motion of ball and rotor into pars and analyze each one separately. They were: ‘The ball islaunched bythe croupier. It orbits on horizontal ‘track on the stator nti it slows down enough to falloff this (Gloped) track towards the center (rotor), Assume a first that (@) the wheel is pertecly level aid (D), the velosty of the ball ‘depends on how many revolutions it has left before falling of. Referring o Figure 42, (b) means that ever spin would produce ‘the same curve, nat different ones lke my hatfste whee. Pt another way, this means that if you timed one revolution of the ball onthe stator, you could tell how many more revolutions and ‘how much more tire uni the ballleftthetrack. I these assurp- tons turned out to be poor, we would attempt to modify the analy, * Next analyze the portion ofthe ball orbit from the time the ‘all leaves the tack tit crosses from the stator to the eo. 1F the wheelis perfectly leveland there ae no obstaces,thenit seems plausible that this would always take the same amount of time, (We later leaned that wheels ar often significanly ted. This Tut, when st oocurs, can affect the analysis subsianually, We even tually learned how t0 use it to our advantage.) There are, however, vanes, obstacles, or deflectors on this portion of the wheel The size) number, and arrangement vary from wheel ta wheel ‘Onaverage perhaps half the time these haveasignifcant effect, ‘on the ball. Sometimes they knock it abruptly down into the sThe Mathenais of Gambling ‘otor, tending to causeit to come to rest sooner. Thisistypcal of | "Seta deflectors (ones approximately perpendicular to the ball’ path). Other times they “stretch out” the ball’ path, caus- ingit to enter the rotor at amore grazing angleand to cometo ret later, on average. This typical of “horizontal” deflectors (nes approximately parallel tothe ball’s path). * Assume the rotor is stationary (Not rea), and beat that situa tion first. Reasoning: if you cant beat a stationary rotor, you can't beat the more complex moving rotor. Here the uncertainty {s dueto the ball being "spattered by the frets (the cividersbe- {ween the numbered pockets). Sometimes ball wil hia fret and bounce several pockets on, other times it will be Knocked backwards. Ori may be stopped dead. Occasionally the ball wil ‘bounce outtotheedge or the rotor and move most of arevolution ee before allng back nto the inner ringof pockes. Thus even if weknew where the ball would enter the rotor, the “spatering” from the frets causes considerable uncertainty regarding where it finally stops. This tells you that there is no possible reliable “paysical" method for predieting ahead of time which pocket the bal is going to land in, uniess the whee! is prosaly defective or ‘rooked. That makes the roulette method "used" inthe movie ‘fhe Honeymoon Machine,” where the players forecasted the ‘exact pocket, an impossibly. It aso tells you that succesful pliysical prediction can at most forecat with an advantage which sector of the wheel the ball will end in * Ascurme now tha the rotors moving. Generally the ball and rotor move in opposite directions; increasing the velocty of the ball relative tothe rotor. We'llasume this isalways thecase. Ive ‘never seen or heard ofa casino spinning ball and roto inthe sare direction. If this were done, the relative motion of ball and rotor Wwould be even less than with a stationary rotor and prediction Note tha this change depends onthe root veloty. Since that ‘aries from time to time and from eroupie to erouper this adds further complexity. It turns out thatthe velocity of the rotor changes very dowly, sits posdbleto predict with high accuracy Ey Reale ‘which part of the rotor willbe “there atthe preted time and place that te bal aves de stator. ‘wallow take you through a simplified version of what we fist ttiedtodo, Later, with that overview to guides, I'explainsome of the modifications we had to make and describe our casino experiences. First, let's consider pat 1, the motion of the ball onthe track. ‘The actual function x, which describes the numberof remain- Jn revolutions x versus the remaining te 6 is theoretically very complex. ‘Our first problem, and the key one, was to predict when and ‘where on the stator the ball would leave the track, This problem ‘was Rey because once we knew this, everything else except rotor velocity was a “constant.” And rotor velocity is easy to measure in advance and incorporate into the prediction, as we shall se. ‘Our metiod was to measure the time of onc ball revolution, Ifthe time were short, the ball was “Fas” and had along way to go. the time were “ong,” the all was “slow” and would son fall, from the track. "We hit microswitch asthe ball passed a reference mark on the stator. This started the electronic clock. Ths was at time ¢ (ogo) with 2 revolutions ogo. (There ate many such “marks” sallable on all actual casino wheels.) When the bull posscd the reference mark the second tine we hit the witch again, stopping the electronic clock. That was at atime (eft to go) before the bal eft the track) with xy revolution left. The clock measured 4) ~ fy the time T for one revolution (s0 2) — ay = 1)* ‘Movie Experiments “The function xr which we ar wing inthis uration i ot the actual one. The actual x can be determined by a "movie experiment” ikethe ones described exer which Tain 19590 ‘ny balt-ie wheel. To do tit experiment today. get fullsize roulette wheel, alarpeclock which reads accurately in huncedkhs ‘of second or bate, and a video camera or movie camera. Then take a movie of the orbiting ball. The sucoesive frames give 5s See Apreni ,.5t sexe Arena re‘The Mathematics of Gambling values for ‘and xf, which ean be plotted to get an x()cuve ike tat of Figure 43, Several moves shoukd be made to see how ‘much the 2() curve varies from one spin another. This uncer- tainty is asouree of erors in determining 7, that discus later fon. These #1? erors can be incorporated into the theny in the same way asthe ming errors. Tey each cause some uncertainty in theprediced X,/7)value. The data from the movi experiment can be improved the camera framesaresynchronizedtoastrobe sothat themouonof both balland lockis"stopped” rather than bluny. I didn't do this in my original movies, so I got a short blurry ae, instead ofa ball, in each frame. Ian appropriate clock is not evelable, yon can nse a high ‘quality phonograph turntable instead. These rotate at very “uniform speeds which can be verified for your turntable with & strobe. Now gcta stiff paper disc and mark the edges in equal Smal nis. Number these ruts (Mach as you would “‘reula™ ruler) for ease in reading, Now place a thin fixed pointer jst x Figure 43 2 " 19| Rosle ove the disc. When the dis rotates, you have avery accurate Glock whose hand Tix and whose face moves. I ou we & paper dsc of polar coordiate graph pape (ued, petaps, oan ‘ld record), there wl be 36 equal spaced degree marks. ‘AL 330) pun cach mack 1/300 see. AtAS tps each tmarkis/20se. anda Tx pam, each marks 1/468 se. Ona "inch dls, the 30 marks wilbespaced abouta tenho aninch nat so ado mars ean be used othe pictues can simply ‘erent toa fraction of an intra Resord et ds wih equally spaced “spokes,” for use witha tbe for esng tata, re also avaiable andcan be wed Timing Errors ‘Shannon and I used the switch which measured Tto Nash a strobe as wellas start and stop he lok. We discovered the gh nd thesttoe las "stoppedthetallatcacho the wo instants ihesvitch wast Tis lowed us oso how much the ball was ofthe reference mark, Since we Kaew approximately how fast the ball was moving, we could tel about how much inte we tree ery ort in hiting the sich, This enabled uso cork the times recorded onthe clock, thereby making te data much nore accurate. We als learned from the sua feedback hoo bezome much more accrae at timing ‘ere’ an ilstration. Suppose the tack ofthe whe! was 25 inches in diameter (don't have any ofthis equipment now so Tintemembering back ovr 20yearsandrocalingabout what the Siz, velols, te seemed 0 be, They be cose enough (0 be representative and good enough o show you how to do i all ain beter for you you wat to) Suppose the bali inch in Siancter and 7: heme foronerevoluten, 0.8 seco. Then the track 78.54 inchs in eng, or 98.17 ball imate I the bal center is one diameter away from the reference mark when the stobe ashes, then the Sing err is about 1/98.17 of Tor bout $1000 of aseond. Tere willbe one ofthese errors when de svitch first hit and another when iis it the second time ‘With pace we were ale o reduce each ror apical (oot ”The Maemaic of Canbing smsan square) ize of one bal ameter or abut 8/100seonds Recoag to te tcoy of er he two cors oper ge typical (fot mean square) sizeof 2000 or about 1.2000 seconds “Thee eros woud be unobsevable in casino play, 90 we collin correct for hen tere, Tec queton w how do they aff the prediction? A Simple Casino Countermeasure Itshould be cleat that fr thismethod to work, we have totime ‘the ball (and rotor) before placing our potentially winning bes, (Earlier bets are losing, on average, so are only camoufage.) "Ths theeasino must allow us o continue to bet for atime afer the ball iced, Thave obsered roulette wheels ll over the world: Monte Carlo (our final oa), Nevada, Puerto Rico, Nee, Venice, and London, The practice has been, generally but not always, to allow bes until he ball as almost ready to falloff the track. This was much longer than we needed. Be wamed again, though; all ie casino needs to dot prevent our methodisto for bid bets once the ball is launched. That simple perfect countermeasure isthe Achilles heal ofthe system and a major reason why I never made a total effort to implement it. (People ‘who use the system in casino play say the casinos don’ catch on ‘nd don't use the countermeasure, But ifthe player isnot really ‘careful, I would expect the casino to catch on.) “The ball iming erors caus: errrsin predicting both the ime and place the bal leaves the track. Even ifthe pial path of the ball Gown the stator ito the rotor i abways the same in time and clxance, this stl yields eros in predicting when and where on the rotor the bal enters. * ‘se Appi See Apps Roulewe Error Analysis, We have long ist of sources for error inthe prediton ofthe tal’ finl postion, They re FF Rotor ting —use 14 pockets to hte BBall timing —wse 55 pockets to lustre, 3 Variaonsin bal “pas” on rotor ee Figure 41), Bor ize ‘unknown, cali. B4 Ball path down stator: eror due primarily to “deflectors” sodas wh hetypeand pom Use en pockeo ES Vartion in distance al raves ono. ror due primary to frets between pockets "spattering” bal, plus cgeasional very long paths aon the fim of the rotor “outde” the pockets, Uses pocket fo lust. .B6 Tied whee. (We dnt know about this yet.) Fr iluseaive pupoes, ssme the errors pprotimatly obey the nora probability dsuibution. Then the andar devition (yp at) of he am of sl enue ge a ot thesum of llth squared errors. Fo instance, using “pockets” {6 our ul, combined errs B+ ES have pial sar yO" + Jy = Yas = 91 pockets, Now ad onthe ing econ: E+ E, +E, + Ey hive ypicl ste ya? + $57 6° + 75 Funai 108 poetews Thos the Yming oor in tis example ase ver lit adlional enor: just 108 ~92, or 16 pocket Ofeoute, wehaven'taddedin 3 yet an, i Xisbigenough it ould ruin everything. Possible variations inthe bal orbit {etavior on testator were forusto measire because we undithardto telat exactly what point teal ost contact th the ute wall ofthe wheel. Wels eared from both our ovn Inbexperionecs and from watching in the esnor why she or ‘ated somewhat. Once a runken, igarmoking bettor knocked Iisashootothetrck, This washard oc ut, Tigotontheball. andspead out onthe trac. That immediately changed thetall’s 2 see Apeni .wThe Mathematics of Gambling ‘behavior, Skin il from our fingersorthecrounie's wouldslowiy “poison” ball and track and seem to affect the orbit behavior. If we or the coupler gave the ball los of axial ‘spin” Grn the sense of tennis of Ping pond), it could take several revolutions fround the rac before his abnor pin encrpy was converted {orbit eneruy. (We named this effet after the famous quantum mechanics concept of spin-orbit coupling.”) On the other Fhand, the ball might he launched with no sin or backspin, soit ‘would skid for a while before spin and orbit got “into synch.” Advantage Versus Error ‘Obviously, the greater the error, the less the avantage I we assume the total prediction ercor E's (approximately) normally distributed, then we can construct a table showing the player's fepected gin or lee ae 2 funtion of E. "Fable 44 gives the results for abet on the best pocket and also fora bet on the best “octant.” The best octat isa set of five pockets, two on each side of the best pocket. "The Table shows that, when the prediction ervt is normally sistributed, the typical forecast error (tandard deviation) must be IG pocketsor es, in order forthe bettor to have an advantage, ‘This 16/38, or about 0.2 revolurions. Thisistrue both fr bets On the best pocket and the best octant. Since the best octant {eludes four pockets that aren't quiteas good asthe bes, the a= ‘vantage is somewhat less fr a given typical eror E. However, as ‘Ye wllsce later in dscassing tne Kely-Bretman syste for money management it generally better for a small to medium-sized ‘bankroll to bet the best octant. Kimmel and the Dealer's Signature ‘Stephen Kimmel aserted that a dealer who works eight hoursa day, 50 weeks a year, tendsto spinthe bal and rotorin ahabitua, feglar way. Ts Would niake possible accurate preditione—a beton ten pockets, Kimmel contended, wouldhavea 0" chance of sucess. His views were contained in an article “Roulette and Randomnest” in the December, 1979 issue of Gambling Times, @ Rowen Tie 4 i =| [Ldon’t believe Kim's approach works. Here’s why’: there arethreeimportant conditions that must remain roughly constant ‘througout pay forthe payer o take advantage of theregulaity fof the dealer's signature. Thee conditionsare (the rotor velocity should be approximately the same each time the balls spun, 2) ‘thespinning ball should make approximately the same number of | revolutions each time, and (3) the inital postion of the rotor ‘When the dealer launches the ball should be approximacely the ‘ame each time. This third condition, which is not mentioned in ‘Kime!’ art, is crucial.The Mathematics of Gambling way of lustration, suppose thatthe rotor veloty was early hesgme ass une andar thea spuntballesaaly {hesame numberof relations ineach instance. Suppose frie thatthe ball spun evacly eh evelutions and the rotor four Fevoluions duane the me: Given hase assumptions, the all ‘ould ind aout 12 revolutions beyond the point where it was Tnunched Inter wore, ithenumber 13 was pasing bal as the dene released the bal wuld arse 12 evoltions later, Rise fo te spinning rotors prone the mabe 13. ‘You can sce, however that the umber 2 on te o.t a6 Aico he baller theinstant twas raed thebllvould hen Stop near hat umber 12 revolts ltt Ifthe dealer reese te bl witout regard which umber nthe pining ocr closet the launch pont, the bal would ‘tndomly alone rior [revolts ater. nthiscase there ‘fous no predsibity wasoevs,cten though the rotor ostyb absolutely fed and the aurber of ball evolons stat Anyvaraceinrotor velocity or number of alrevali- {fons wold firther gurantee random outcome, Because Kin fhe id not docs varatons nthe pot of rekase 1 not teievein hismethod. There a eter approach oth sania analysis of rule, Watt's dee aot count the nunber of revolutions the bal takes on te slafr from the ine of fase unl i eresses onto thevoter, Note how consant that number of revolutions. The flea poor observation canbe sais stated a some erage numberof revlon ps an err tem. ‘Neu coun the nunber of rexohons the rotor makes ring thetine hebalisonthesstr This wllveyouanoter average {orn number of rar reitony, ps sand cor tr Fray, coum ow far the bal aves on the rotor ater it has {rosea te ier betwen the rr and ator. You can sie Store thc ris as some averene numberof revoluons oF posits pls an ero en Tn ovr fortis approach to work, i is neesary thatthe save 10% Of the suns ofthe squares ofthe error terms bees Rostese than 17 pockets. The proof ofthis appears in Table 4-4 which shows what the rate of return is, given various root mean square ‘errors. That table demonstrates that a positive return is possible ‘only when that root mean square error is less than 17 pockets. ‘Now forthe improved method. Inthe unlikely event thatthe oot mean square eor i less than 17 pockes, then —and only then—you have achanoeto win. Thekey lis in sing the postion ofthe rotor when the balls Innched as your starting point for reiting whee Ue ball wil all out ou the whee, For example, suppose you find that fora certain dealer the ball, travels eight revolutions with a root mean square eror of five Dockets, Suppose also that curing thistime. the rotor travel four revolutions, with a oot mean square eror of six pockets. And suppose sil further that once the balls ontherotor, travels 13 pockets with a root mean square error of eight pockets. Given tse supposiions, you can predict una Ue all wil vel eight revolutions plus four revolutions plus 13 pockets from the launch ‘Pition, or 3 pockets beyond that point. The oot mean square ‘or is the scuare rot of five squared pls six squared plu eight ‘uared. This tums out to be 11.2 pockets, well within the required eror of less than 17 pockets. In this case, the prediction sjstem would work. "However [think you will find that when you collet this data, theerrors at each stage are several times as large as Ihave used in this example. My own observation is that the dealer errr in the ‘number of revolutions forthe ball spin sabout 20 pockets forthe ‘more consistent dealers; iti much larger witha less consistent ‘one, [algo noticed tat the rotor velocity isnot nearly as constant ‘as Kimmel would like, Thats because the dealer givesit an extra ick every few spins to rebuild ies velocity. ‘tis also true that the deflecting vanes on the sides of the rotor ‘add considerable randomness o the outcome, as do the frets or spacers between the pockets. The upshot i Gat I don't believe that any dealer is predictable enough tocause aroot mean square error of les than 17 pockets. I'm wing to examine proof to the contrary, but I would be very surprised if anyone could ever pro- aThe Mochomais of Gambling uceit, If a dealer Gutfuly practic spinning the balla fixed number ‘of revolutions, and if a motor drive spun the rotor at constant ‘elocity,andif wehavea very good way of deciding exactly which ‘numer is pposit te ball just sit istleased, it might be barely possible to gain smal prediction advantage. fconsder even that very unlikely. Tn closing, I'l give you the perfect casino countermeasure t0 the srategy ofthe dealer's signature, pretending forthe moment thatthe sategy worked. Fist, the casino halts the betting before ‘thedealer spins the ball. Second, the dealer costs his ees r looks away from the wheal when he releases the ball so that he has no Knowledge of which number on the rotor is closes o the ball ‘when it is launched. Then, for the reasons explained above, the result will be perfectly random. Chapter 5 The Wheel of Fortune Inthe last chapter, [described a system for winning st roulette ‘based on physical prediction. That system was developed largely in 1961 and 1962 in colaboration with Claude Shannon at MIT. (One by-product was an even simpler system for physical predic- tion of the Wheel of Fortune. A story about me and blackjack catd-counting in Lifemagazine, March 27, 1964, reported on this ina section entitled “Beating the Wheel of Fortune wth the Big Toe.” ‘While I wast the Fifth Annual Conference on Gambling and Risk Taking at Caesars Tahoe in October of 1981, collected data ‘ona Wheel of Fortune at Carsars. I wanted to see whether their ‘wheels could sill be predicted in the samme way. "My Casio C-80 watch has a digital stop watch feature which times to 1100 ofa second. Ivsed it to time one revolution ofthe wheel and then recorded how many revolutions it went. I col- lected the data in Table 5-1 atthe Wheel of Fortune nearest to ‘Caesars! cashier cage To see how predictable the Whee! was, I looked for a 6“e fer te 30 oie tee ‘Table $1 “Genes Tahoe Pr ian Pr egsag Oecinat reed a0 og at eer br The Whe! of Forune ‘mathematical cuve which would best fit these data points, A. ‘uve which worked wellwasR = A times Ttothe Dpower where ‘A= 121,545 and B = ~2.11153. Inthe equation, T isthe time for the wheel tomake onerevolution and Ris the numberof addi- tonal revolutions which then travels, Iacuiivly, if Tis shor, the whee dd one revolution quickly soit wil 0 far and R willbe large. ButifTislong, the wel was low and wil stop soon so R will be small, “The letter pin the third column of the Table (“raw data") stands for “"pegs.” The wheel has pees separating the payott numbers, Asthe wheel rotates, the pegs push past flexible “ap per.” This gradually slows the whos, When the whee stops, the ‘winning number i the one with the flapper between its peg. ‘Theraw data column gives3.5 + 22p for observation number 1. This means thatthe wheel traveled 35 revolutions plus 22 pegs for further numbers. Since there re $4 numbers lit went 3.5 “+ 22/54 ot 3.907 revolutions in all. That is shown under “decimal” in column 4. ‘The prediction P is made ftom the equation. The “eror” P-R {isthe amount the prediction is off from what actually happened. Strictly speaking, what Lam calling «predictions onlyafittothe data. The fit approaches “true” fitmore closely as more datais| Included. However, there ix generally a difference batween the “crue” fc andthe actual fited equation. ‘New data tends {0 custer around this slightly different ‘unknown tre fit, so it wil tend to deviate from the actual fit 10 ‘the data by this extra amount. Thus, we expect future data to be redicied by the equation not quite as wel asthe data in Table S-1. ‘The error P—R has a standard deviation "typical size”) of 0567 revolutions, or 3.2 number. The true curve location (tan- ad deviation of the curve) s probably within .0169 revolutions (0.9 numbers, on average. Considering this and he greatest ‘vive and negative values in te column, eror in “pegs” sug- ‘ests thatthe prediction will almost always be within tive "pegs" (or positions of the actual outcome. ‘Table $2 shows the actual arangement of numbers on the oThe Mathematics of Ganbling wheel, They ae listed in order, clockwise a sen by the player. ach number eves the profi pe unit bet. Thus, player wii Bets ‘9n2 wins $2 foreach $I bet. The number marked 40, and called (Caesars, pays 4010 1 and the number 40B, called Cleo, also pays 40to I. Abeton oneof them doesnot win the other One comes here ‘are 24 “ones” in Table $-2, Thus, if each of the $4 ruber comes up oes, one win 3 times nd ees 30 nes for uty of 6 usa 3 un bas, ra expen ate of Se/st= 21791118, Sima action lad 10 Table 53. Forte player who dost prec, te hose eg erm. Tisha eae to a0 ‘Table 5:2 Table 53 Fie eis 12 mt a || Mame Howe tion perro ail 3 fot tame zorad 5 vast 22200 pereert| 6 {eet Seam Bers 120i) mw fist germ ere rs zl) mB tit 2am ‘Now lets se what the player avantage might be from predic- tons. Suppose for he sake of discussion thatthe final wheel post tion isalays within five numbers of thepreicted whee position. For any prediction in the eleven number sp centered around 40,, We snould Devon 404. Ip s4spins where each final pustion ‘occurred once, we will place 11 bets on 0A and win one of them for again of 40 10 = 30 units “Thesiscusion isthe same for 40R. For any prediction in either of the eleven number strips surrounding each 20, twenty-1¥0 ‘number in all, we bet on 2. In weny-¢wo bets we expec on 20 units twice and lose one unit twenty times for a net gai: 6 The Whe! of orane ‘werty unis. This leaves 54-44 or ten predicted postions where ‘weneed instructions, “There are four 10s inthis ef-over set often positions. Suppose webs the 10 each time one of these postions ispredited. Itseems plausible to suppose that we would win ten units four timos and Jose I unit sx times for ane gain of 40 —6=34 units. (Actua, since the 10sin this cas are either the predicted number or witht ‘one position ofthe predicted number, we expect to do beter sil, ‘Finally, in Sunt bes wenet units from 40, 30 wuts ror 408,20 units from the two 20s, and 4 units from the four 1, for a total of 114 units/$4 units ora 211% rate of return. Te may he poeshle ta improve bosh thesming proce and the method of exloiing predictably. Tis would improve the Is. ‘We see now that the Caesars wheel can be predicted well, «enough so that wecan Dearie te casino wil let us pu down bes after the wheel has been set in motion.Section Three Other Games While we have so far concemed ourselves solely with casino games, some of the principles we have used are equally applicable to other gambling situations. In this section, we apply mathematical theory to horse race betting and backgammon. Horse racing Is the number one spectator sport In ‘America and a large amount of its success In this regard can be attributed to the wagering opportunities. ‘The racegoer becomes a participant in the spectacle. While we offer no surefire system, we do suggest an approach that shows promise for the gamblerinvestor. Backgammon is an exceedingly complicated game from a mathematical point of view. Because of the ossibility of repeated restarts by the counters, the game is potentially infinite. This impedes analysis, but We offer several insights into the end game and the doubling cube, parts of the game where optimal strategies can be computed.Horse Racing While so far Ihave limited myself to discussing casino games, the concepts presented epply equally to other gambling games, suchas horse racing. At the racerack, ones offered. variety of| sifferent wagering possiblities. The payer ean wager on one or more horses to win, place, oF show, as well as combine any ‘numberof horsesin the various exotie bets aly double, exactas, uinellas, etc.) The goal of the gambler atthe racetrack is 10 isolate in each race those bets that yield a postive return, after considering the part-mutucl takcout and breskage. ‘One approach wth applications in horse racing involves thewse of a technique called hedging. Hedzing, offen used in the securities and financial markets, involves taking two oF more investment postions simultaneously. The risks should cancel out ‘and an excess rate of expected return should remain, Why “Hedge?” In the securities and finance markets, to hedge isto take two ‘of more investment postions simultaneously. The risks shouldThe Mathematics of Gambling cancel out and an excess rate of expected return should remain, ina real horse race (or any parrnutue contest for that mat~ ter), the tre probabilities are not known. If we knew the true probabilities Or had better estimates than the pur-mutuel pool Effers, we might find horses with » postive expectation. Then ‘we could simply be directly on those horses instead of develop- the following method for the daily double. “There isa plausible argument which upholds the pai-mutuel pool's estimate ofthe rue horse winning probabilities: “dere ‘Were a method of predicting horse winning probabilities, and these probebiites differed enough from the pasi-mucvel pool's estimate to give the prodicior an advantage, then he would place bets and by s0 doing would cause the pari-mutuel pool odds to sift in such a way as to reduce that advantage. With many betors and ‘much information and available computing power the overall e- fect iso reduce such advantages so they are small or even become sisudvanages.” Tn other Words, “Ifyou could beat the casinos at blackjack, then they would change the game so yo coulda. Thus, there fsa any sytem for beating them. 1f we assume that pri-mutuel pool probabilities re true prob- abilies then the horse hedge system doesnot improve our edge ‘ver the rack Lake! You might think thar makes the horse hedge dea useless, but this not true. Consider the daily double pool: ‘The payoffs should be consistent with the probabilities in the individual race win pool; but in general, they aren't consistent ‘Ths, we have a chanoe to use the probablities based on the individual race win pool. The Daily Double ‘Let's apply the horse hedge ides tothe daily double bet. The same idea, with some modifications, also applies to exacts, pic ‘ixand similar bets and to exacts, quinlls and trifecta in ja ala. For a litle background on the daly double, I quote from the book Science in Betting: The Players and the Horses, by E.R. a Siva, and Roy M. Dorcus Hone Racing In daily double betting, any horse inthe first race can be ‘combined with any horse in the accond race, and {0 Win ‘the betior must successfully select the winners of bath races Some bettors combine all ofthe horses in the it two races. If there are ten horses in each race, in order to cover all possible combinations of horses, one would have to buy one hundred tickets at $2.0 each. If by chance long-odd horses ‘won both rae, it would be possible to make a profit on that single daily double. However, uch a situation is act common throughout a week or a season. One daily double $2.0 ticket at Det Mar ecenly paid $2,878.60, another paid 9685 and there were in addition thee others during this season which paid over $200—yet actual return fOF this season were only $6808. The average numberof horses was leven inthe first race and ten inthe second. To have con bined all these horses in all the daily doubles for this season ‘would have cost $200 per race, and since there were fory- {wo days in this season, the total cost would have been $9,240, producing a loss of several thousand dollars, Notice that they consider beating equal amounts on each horse, From one season at Del Mar, they found that $9,240 in total bets ‘Were returned; $6808 for a payback fraction of 074 or a loss of| 265 of te amount bet. Thu, betting equal amounts a eal cour bination did not work. strated in Table 61 is the horse hedge method for daily doubles ina real race. The Table list the winning probabilities based on odds forthe frst race at Del Mar on August 13,1980, ‘The horses are listed according to post postion inthe frst col- ‘umn. The second column has the haricapping odds given inthe LA. Times on the morning of race day. The third colun is b= tained from these odds by taking the right hand number in the second column and dividing by the sum of the two nombers. For example, 30-1 gives a probability of U31 = 00323. For te horse inthe {3th post position, 7-2 gives a probability of 29 = 02222. When there is no track tak, the probabilities calculsted this way must add up to 1.00The Mathemanes of Ganbing Table 61 ‘When there isa track take, the probabilities calculated from the final payotf odds at race time will equal more than 1.00. In fact, they add to WK, where Kis the action of the pool which is rowmed wo the betrs. This rue is not quite exact due to the irregular effects of breakage, bu the effets are generally small and not worth discussing, Th order to correct for probabilities that donot add up to 1.00, we add them, deducting horses which may have been scratched. ‘We then use the final total and divide it into the preliminary pro- babies so that it equals 100. (Corrected probabilities appear in columa four.) Colum five gives the final odds on various horses. Columa six bas corresponding uncorrected probabilities and column seven lis coneted probabilities. Notice tat column six adds to 1.263, by dividing this into 1.00 we get 07880 which corresponds to ‘tack take of 21.30% for this particular race. 8 Hore Ring Column tree equaled 17237 before deducting the horses which vwete later scratched, making the track tke 100 large. The SOM {or Del Mar i typically about 1.20; therefore the handicapper’s setting ofthe odds was not consistent. On average, the odds were set 10 low in this race forthe horses. When four Borses were scratched, the odds on the remaining horses gave probabiiies which equaled 1.3105. That i the typical sum at Del Mar. ‘Table 6-2 presents the probability calulations forthe second race. The four column appears 10 equal 09999, But shows 110000, because the entries have Been rounded of to four paces. ‘he final outcome of the daily double: horse 2 woa the fist race: horse 1 won the second race: and a winning $2 ticket paid back $3860 or $19.30 per unit bet. The amount bet on each of the 15 x 8 or 120 combinations is proportional to the product of the corresponding probabilities. For caunple, if we use de correc probabiides based onthe rmoming odds, we have 526 for horse 2 in the first race and. 725 for horse 1 inthe second race. The product ofthese two ‘numbers is 0263. That means we bet 0263 of cur teal uni het (on the combination ‘which aetwally won the daily double. Te 62 Sst ee 7The Mathematics of Gambling “There, we ines euro S190 > hs robe or 5080 ‘Ta which meus we ot 8.2% or ur bet te hs ose the ilo, he probes te 1922 and 2313. Thelepro- thts O45 anf we wold seve bt cnt % S83D or 4380 sia unt, ova los of 4.2% (On page i, Da Sv and Doras warn you that Tn dig any stata work on aly oat, he reader fst be cael ot wo ne the actual cosing ods of he fore, slid te dy flowing eats mesa ears from aewspapers or fom he Form orth Tera, ice ese stint od ue nt wae the ly dale a fr lher the fis ort sand aces.‘ bene tut ly only upon the protbl ods for stil sy ‘daily double ting, eds wich are given nthe Mor tng Line atte tac inthe Fore or Telegraph under fet aaicapes sen Suoe, Anal, oka oF Bien nthe tack programs artermore in deg ih these probable athe bet tor mu rember ta thy yO ay ot ered tthe laine closing son the cba Forth ist rac oly the ata oS that we would we in atc gy be fy close thes alos we were acta Fatt tate watching the tncboar) As post we cans dices withthe hoe deen as trols io We daly diab, For cmp there ea mini $2 bet. node approx imate the varus probes of te typi oe handel oro Combinon, we a 0 make several hunted S2 bis wich feist tani, Anterprble shat he inal pasate! pol odds ar unin, Even if we did ko the Sis ont nivel mec te ne pei of he ial fries winning nthe respective races wal tl be un. Thor, we dot know ithe hore hedge mad wl give than sdvaage over he wack ke Even if dos give an advantage, we dont now if we can nin enough o vrcomethe nck ak or an ovr advantage a Hore Racing ‘This i the eason why this gstem needs further development. ‘One way to get around the dificulties is to hep reeord uf the final odds and the coresponding probabilities and bet acconi- ingly. If pari-mutuel odds area fur estimate of the true odds, then this indicates the sot of gain tobe had from horse hedzing. Ifthe gain is large enough to preduce a substntal advantage, then there might still be an advantage if we use good odds that are available o us at the time we place our bes "To show you how to keep this sort of recon, Iwill use one average figure to correct ofthe track take. Table 6:3 shows the sum of the uncorrected probabilities forthe first five races on the consecutive days. The days are Aupust 13, and 1S, 1980 al Del Ma, The two entries followed by question marks siggest that there may be data errors or newspaper misprints. Except for the two questionable figures, the uncorrected probability sums are close to 1.20. The average, of the D remaining races i Table 163 works out to be 1.2038 Table 63, ‘To simplify, I shall use 1.20 in my computations in Table 6-4. ‘The fraction estimate the investment returned foreach day the horse edge system is used at Del Mar. If ou want to construct similar abe, get extensive racing records from your tack, and ‘determine whether the method works over a pest sample. »The Masmavics of Gambling 3 2 f]lteenennaee Hilsseaeaass Hlsevenacaaa ht yf ii [ejanananeaae RRGRRESERR SETRRRREEE SSORSSSSSE ores Racing ‘Table 6-4 shows the idea at Del Mar. The second, thd, and ‘our columns list the corrected probabilities based onthe tral ‘8 of 24 forthe winning horse inte frst race. The fit, sixth, and seventh columns do the same thing forthe second race. The tghth column the prt of these praise pri-mavel estimate ofthe probability of a pair of horses wianing the daily double), For the last column, multiply the payback on $1 which the fraction of the uit be returned to us In our sample of ten races, we get an estimated payback of, 94.76%, ora loss of 5.24%. We are estimating the average effec- tive track tke as 1667% so the sytem does beter than average br ail does not win, Fora clear explanation of daily double bering, xactor or €x- acta betting, odds, andtifecia beting, I refer yu tothe appendix of Haress Racing Gold by Prof. Igor Kusyshy, published by Invenational Gaming Ine, 1979 ($14.95). ‘The New York Racing Association takeout is currenly 17% although thas been 16%, The Califomia takeout is 1875%. Of course the effert af reaps is increase the average takcout somewhat beyond these figures Readers who want to know more about the calculation of ning probabilities based on the par-motuel odds should read ‘Chapler 3 in Horse Sense, by Burton P. Fabricand, published ty David McKay and Co., 1968, The book is hard to obtain, but believe you can find it inthe larger libraries. Fearicand takes a sample of 1,000 races, with 93011 horses and 10,35 winners (some dead heat). He finds thatthe average loss, from betting onthe favorites (high pari-mutuel probability of winning, is considerably smaller than the average loss from betting the long shots (low par-mutvel probability of winning) For exreme fvorites, the sample showed a profit and for horses with a pari-mutuel win probability of 30% or more the average lass was jst afew percent. t ranged gradually higher asthe ode lenghtened for horses with dd of 20 to 1 or more and pari matuel probabilities averaging about 025; the average los tothe betors was S$ perceThe Mathematics of Gambling ‘This indicates that the ods, from the par-mutuel poo! for win- ‘ers, are systematically biased; they can be improved by incor porating a correction factor based on a data sample similar 19 Febricand’s. The correction would increase the probabilities ‘ateigned tothe favorites and docrese the probabilities ecigned to the long shots systematically. ‘A more readily available source for the same information is Fabrican’s latest bok The Science of Winning, published by Van Nostrand Reinhold in 1979. On page 37, a table shows Row a player's expectaton varies with the odds. The sample has 10000 races with 10035 winners because of dead heats. In 1984 a book by Ziemba and Hausch, Beat the Rueiruck A Scientific Betting System, appeared with a practical winning ‘method at the track. I went 19 Hollywood Park with the author William T. Ziemba, Ph.D., and used the system successfully. The idea of tre win probabilities discussed inthis chaper i used by them to check the place and how pati-mutual pools. When ‘horses in these pool are signiicaly unde bet, dey fr positive ‘expecations. Good bets appeat vt average abut uuce per OW Backgammon ‘Backgammon has taken ts place alongide bridge asa favorite pastime of sophisticated gamers, It is essentially a racing game, ‘where each player tiesto get his pieces olf the boar ust. Bu, thanksto the doubling cube, t'salsoagambling game, played for high takes in clubs across the country. The basis of ths intrigu- ing hoard game ace rally very ensy to master. (See The Rules (of Backgammon, pp. 84-85) ‘Thischapter wll focus on several aspects of backgammon that ‘an be solved mathematically. You will ear useful but simple ‘odds for bearing off vath only two mea ltt. Most good payers already know this. But don’t go away, good players Later you vel eam fats about backgammon tha ew in the word are “Asan introduction o end postions, suppose youare White and itis your turn to rollin the position of Diagram 1. The doubling ‘cube isin the middle.Diagram 1 p 2. Should you double? 3. How much do you gain or lose by doubling? tyou double, should Black aesepr? How much does Black gain or lose by accepting your double? ‘White wins only if he bears off om his next rol. Soto help us solve end postions ofthis type, we calculate a table of chancesto takeoff menin one ol. The exact results given in Table 7-1, and the chances to the nearest peroent are given in Table 7-2. "As you caa see from Table 7-1 theexact chances of winning youhave aman onthe five point anda man onthe two point are 19 in 36 or 5277... Table 7-2 gives your chances to the nearest peranlage, 0 53%, Now you have the answer to question I. "To see how Table 7-1 is calculated, recall that there are 36 equally likely outcomes fr therol of two die. These areistedin Table 73. o ‘Table 74 ont | tor | ape | aot| soe | sor | oot opt | on ae |_o« | ov | tof us [als zo =a] e [se aot_| a6 wpe wp aot] ci Sot “pop ele en) a w>epets Table 72 anime | oo | se | 26 | oot | sor | cot ene ont sor | 00% 43 | aos | 7586 ter wom [0% aim [oo a Zot voo% [7 om | |B Tor ee | oe arm | | toe wise [oo arm | orm [oon farm pl te [eae [oem | aon [orm fore er oe ee s‘he Mathematics of Gambling ‘Table 73 stcand shower Tneemone | + [2 {ale |s le “Think of the two dice a8 labelled “fit” and “second.” It ‘might help to usea red die fr the “frst” die and a white one for ‘the “second” die. Thenif the ed (irs) dieshows Sand the white (Gecond) di shows 2, we call the outcome 5-2. If instead the first die shows 2 and the second de shows 5, hisisa different one of| the 36rols and we callit2-S, Outcomes are named x-y where xis thenumber the first die shows andy s the number the second die shou. "To se that White has 19 chances in 36 to win in the situation presented in Diagram 1, we simply count winning rollin Table "73. I ether die shows at least and the other shows a east 5, ‘Wile wins. Heals wins wth 22,33, and 44. This ives the 19 (Ghaded) winning outcomes in Table “Asanother example, suppose thetwo men to bear off are both ‘onthe sx point, Then iftwe different rurahers are rlled, White ‘can’t come off in one turn, Of the sx doubles, only 33 or higher ‘works. This gives four ways in 36 oF 11%, in agreement with ‘Tables 7-2and 7-3. his simple counting method producesall the ‘numbers in Table 71. "Now we areready to answer question 2. Should White double, {in Diagram 1? The answers yes, and here's why. We have seen % ‘that White wins, on average, 19timesin36. If we say the stake ‘one ult than if he does nof double, in 36 times he wins one unit 19 times and loses 1 unt 17 times fora gain of two units/36 times /18 = 0.055... If White doesdouble, Black can either accept 6Ffold. Suppose Black accepts. Then the stakes are two unitsand ‘calculation ike the previous one shows Wie pans an average Of four units /36 times = 1/9 = 0.111...-unit per ime. White ‘gains twice as much by doubling as by not dou- bing. If Black folds insted, then White wins one unit at once, ‘which is even better. ‘This also answers the rst ofthe questions. In answer to ques- ‘ton3, White gains an extra 5.55% of aunt, on average, by dou- Ding. Answer to quesuon 4: black should accept. He loses 1/9, nikon average by aoepting and one unit for sue by folding. “This answers question 5: if Back makes the eror of folding, he loses an extra 8/9 unit or 80, "The usefulness of Table 7.2 is generally limited to situations where you have just one or two rol eft before the game ends. But itis surprising how often the Tableis valuable. Here aresome ‘more examples {o help alert you tothese stuations. In Diagram, ‘Black has the doubling cube. White has just rolled 2-1. How dows he ay it? If Blac roll double on the nest tur, he wins at once and it ‘won'tmatter what White did. So White only needs to consider the ‘case where Black does not roll doubles. Then White willhave one ‘more tur, and he wants toleave himself wth the greatest chance {0 bear off on that tur. White can move one man from the 3 ‘point tothe 4 point and one man from the 5 point to the 3 point. By Table7-2, this ves hima.47% chance to winif Black doesnot rolldoublas. Or, Whitecan move one man fromthe point othe 2 point, leaving the other man on the S point. This gves him a 53% chanceto winif Black doesnot roll doubles, sothisisthe best ‘vay fo play the 2.{In Diagram 3, White's problem isto avoid abackgammon: if Black wins before the White men escape from Black's home board, Black will win 3 unite. Otherwise, he will only gammon White fortwo units. Digan 3 Backgannon ‘Whitehas us rolled 41. Hemust use the 4tomovethe mann thebar tothe Black 4 point (dotted circle). White can then move this vanon tothe Black pony, n which ease i Back doesnot roll doubles, Whites situation on his last tum is shown in Disgram 3a. Diagram 3a “ - ‘The chance for White to remove both men from Black's bome bord on the next rolls the same as the chance to bear off both ‘men when one ison the 4 point and the other ison the 2 point. ‘According 10 Table 7-2, this is 64%. Suppose instead White plays both men to the Black point. ‘Then Diagram 3b shows the board if e survives Black's next ol. If instead White rolled 42 in Diagram 3, he could enter on 2 and move his other man tothe 7 point, giving an 86% chance (Table7.2, man on S point and man on 0 point) to escape Black's home board on the next rol. Or he could play to leave his wo »™ \ f \o\fa\'o\ Ld isi ols) oj back men on the Back $ and 4 points. Ths gives only a 69% ‘chance and is the inferior choice. ‘An outstanding reference work is Backgammon by Paul ‘Magriel, The New York Times Publishing Company, 1977, $20. ‘Most of Table 7-1 appears there on page 404. A‘convenient reference for practical play isthe “Backgammon Calculator,” Doubleday, 1974, $1.95. Thishandy cardboard whee hasmos of ‘Table 7-2.0n the back. ‘Here are some questionsto check your understanding, Referto Diagram 2, assuming Black has the doubling cube and White has just rolled 21. 1. Should Black double afer White ‘makes the best move? 2. How much would Black gain o lose by so doubling? ” Backgaron 3. Should White accept Black double? 1 he does, instead of folding, how ‘much does he gain or lose’? 4. Whatisthebest way for Whiteto play 32in Diagram 7? We will now present the complete, exact solution to all backgammon postions when each player has only one or two smenletlinhisownhome board. Don Smolenand Icaleuateditin| 1975 ad kept it to ouseves for several years. realize thattis often no: practical or desirable o use the tables {provided during thegame. Fortunately, many of thee situations are covered by a handy rule that appeared in a "Sheinwold on Backgammon” column in the Las Angeles Times. Sheinwold considers the situation in Diagram 4. The problem is whether ‘White, having rolled 6-2, should play the 2 so that he leaves his ‘wo men on $ and 20rd and 3. Diagram 4 A‘The Mathematics of Ganbling We solved thissame problem eae when dcusing Diagram 2. Wenn fon ible? att leennginen on Send2, bet Fests pes White 53 chance o get off onthe net tar, ‘re vig eon and es on a che Now we err qucton:Igoutavetnlenerneor vo Sfengour at, wha isthe best lene? Assuming ta pose Rs Revom whch you nat choo ave the same pp coun, the corre ru, wih Sewell ies Rules for Leaving One ot Two Men 1) Ifpoxible ave one man rather han WO. 2 tf foemctleare two me, lene thom on dilferen Pons, Hposible 6) Ifyou havea choice, move off he 6 pon. {8 Irgou ae aready off the 6 pit, move te tan onthe lover pont. ren prove these uescoret by using Table72. Tiss sown again ete in condensed form a5 Table 74. Table 7.4 2 eckgannon ‘To check the rules, we simply check Table 7-4 for ezch pip ‘count to seeirit always els us whic of two “leaves” topick. For ‘example, with a pip count of 6, part (1) ofthe rule say correctly that Opt -6 pt. isbest. Then 2) says correct that among the three remainingtwe-men postions, pt -3ptiswoest.Inasimilar way ‘theruleis verified in tum for postions with pipcountsof 4,5, 6,7, '8,and 10. There’snothing to check for pip countsof1,2,3,and9 ‘because the choices areequally good for these pip counts. There's ‘ovhing to check for counts OF 11 and 12 because fo these Pip ‘cours tere is only one choice of position. ‘More examples ilusrating the rule appear in How Good are Yauvat Backgammon: 75 Challenging Test Situationshy Nicolas and Vasslios Teannes, Simon and Shuster, 1974, You can use these rules to solveatcnce test situations 40, 41,42, and 43, The authors give a rule (page 54) but itis neither as clear nora simple sours. ‘We proved the rue for leaving one or two men just forthe case where you will have at most one more tum t0 play. In that case the percentages in Table 7- le us compare te positions 9 sce whichis better, What f there isa chance tat you'llhavemore than one turn? This could happen, fr instance, if we change Diagram 40 that Black has five men on the one point instead of | our. Then Black could rol non-doubles on his nex Urn, eaving three men onthe I point; White could rll 1.2 on his next turn, reiucinghis pt. 2pt postion to one man onthe4 point: Black could rol noe-daubles again, leaving one man on the 1 point; ‘and White then getsa second turn. It tums out thatthe rule gives, the best choice agains l/ possible positions ofthe opponent, not Just those where you wil have at most one more tum to play ‘vote: There i one possible, unmportant exception that might arse, but the error is at most a small fraction ofa percent.) [Now we return tothe Thorp-Smolen solution of al end games, with ust one or two men in each home board. Wewil label home board positions as follows: 5+3 where there is a man on the 5 point and a man on the 3 point, with the argest number fst ‘With both men on, sy, the 4 pint, we call the position 4+. Ps‘The Mathematics of Gambling ‘With onlv one man onthe -point we write +0. Think oftheOas| indicating thatthe second man is on theO="oIt” point. "There are sx home board postions with one man, namely 1440,2++0,...6+0, Thereare21 home board postions withtwo ‘bm, Thus there are 27 one orto man positions foreach payer." “Table 75 gives the fist part of our solution. It ells Player ‘One's “expectation,” rounded to the nearest percent, if One has themove and Two owns the cube. By One’sexpectation we mean the average numberof units One ean expect 10 win f the current fakeis “one unt” andif both payers follow the bes strategy. Of ‘course, ifa player doesn't follow the best strategy, his opponent ‘Can expect on average todo beter than Table 7-S indicates. “The A above + 0 means thiscolumn also applies 0 any count of up to3 pips: 1+0,2+0, 1+1, 3+0, or 2+1, The C above 66+ Omeans that this column also applies 04+0, 341,540, or 1. The A for Player One means the same as for Player Two. ‘We will ilustrate the use ofthe Table with Diagram 5. Diagram 5 (it o se Agen HMO. 5The Mathematics of Gambling les White's tur to move so he becomes Player One. Player ‘Two, oF Black, has the cube. Welook along therow 6 +S and the column +3. Table7-5 shows layer One’ (White's) expectation ‘303,30 Whitehas a 3% advantage. Fc expects to win on average 3 more exacily, 2.54%) of the current stake. If the current ‘stakes $1,000, White should accept a Back offer to sete” the ‘game if Black offers motethan $25.40, 1f Black offersless, White should refuse. "Table 7-6 gives the expected gain or loss (othe nearest pereent) for Player One when he has the move and the doubling cubes in the middle "Unlike Table 7-5, inthis ease One las the option of doubling beforehe moves. One does no double, Two wll beabletodou- ‘leon his tan, If One doubles, Two thea haste chooe of accept jing the double or folding. IF Two accept. lay continues with doubled sakes and Two gets the cube. IF Two folds, he loses the current (undoubled) sakes and the game ends. “Table 7-7 gives the expected gain orlos for Player One when he nnasthemoveand the doubling cube. Thevolumnsfor6+4,5 +5, 645, and 6 +6 are the same as for Table 7-6 0 they have been omitted, “this eae, One has the option of doubline before he moves. However, in contrast to Table 7-6, if One does not double, he keeps the cube 0 Two cannot double on his next tum. If One ‘does double, Two can acceptor fold. If he accepts, the stakes are doubled, play continues, ad Two ges the cube. If instead Two folds, he loses the curent(undoubled) stake and the game end. Table 78 so tells whether One should double and whether Two should accept when One has the cube. ‘Doubling strategy is the same whether One has the cube or itis, in the miele, except for the shaded region. If he makes the ‘mistake of doubling, Two should accept. When thecube inthe ‘middle, One should double for posors in Une shaded regions ‘and Two should aocept. 'Wewill now show how to use the tables to play perfectyin any of the 27 ¥ 27 = 729 end positions covered by the tables. We'll 6 2SLL ‘The Mathematic of GamblingThe Mathematics of Gambling run through sample end. games step by step, showing player ‘expectation, doubling strategy, andthe bet way to play each rll T eatir refered to a book entiled How Good Are You at ‘Backgammon: 75 Challenging Test tuations by Nicolaos and \Vassiloe Tanne, Simon and Shuster, 1974, Consider first Stuae tion 74 from the Tzanins’ book. This is shown in Diagram 6. Diagram 6 Itis Black’s turn go he is Payer One. Black doubles. Should Ihe? If he does, should White accep? The cube is in the middle, We lookin Table 7 8, row 6 4-1, column 1 +1. Black should not Goubl. If he does, White should acept. (This is coretl recom- ‘mended by Tzantes’ book.) Table 7-6 shows that Black's expes- {ation under best play, which means not doubling, is —1740. If instead Black has the cube, we use Tables 7-7 and 7-8. In this ‘example we get exacy the samme answer. This isn't always the ‘case, though, as we wil se. ™ ‘Boctgonnon This example is also easy to analyze directly. Ir Black bears off {nts next tum he wil win. The chances are 15/36 (Table 7-1). If Ihe does not bear offat once, White will win and Black willose. So if te current stakes | unit, and Black does not double, Black's expected gain ie + I unit X15/36—1 unit X21/36 = ~6/36 16 2/3%. Now suppose Black doubles and White accepts. ‘Then Black'sexpected gainis +2unitsX15/36—~2units X21/36 ~ 12/36 = —33Me, On average Black wil lose an extra 16 2/3% of a unit fhe makes the mistake of doubling and White acoxpts. t's easy to see from this type of reasoning tht if Player One has any two-man position and Player Two will bear of on the ‘ext tur, then Player One should not double (fhe can) when his chanoeto bear offinonerolises than $0%. Is chance to bear off is more than SO%, be should double. Referring to the samme “Table 71 proves sre whic he Tansee for Ubeve special situations: With double thee, sicone, sic-two (or ‘anything worse) Keep dumb, hope for the best. Anything ‘better, don’ delay, Double the sakes with st ‘The Teannes Situation 73s smilie. Here is trickier situation that I don’t think you could figure ‘out without help from the ables. Suppose White has 6 +6, Black bas 4+-4, White isto roll and the doubling cubes in the middle. ‘This isshownin Diagram 7. Should White double? How does the ‘gune proceed for various Folk? is Player One, He consults Table 7-6 and sees his expoc- tation is 16%. But Table 78 ells White not to double. We now show how to se that table to play optimally fora sample series of rolls Suppose White rolls 3-1. How doss he pay it? He can end up with 6+2 or with $+3.VV ‘The rule stated easier says that $43 looks better because it ‘ives him a greater chance o bear off on the next tum. This proven by the tables as follows: after White plays, it wil be Black's turn. Black willbe Player One with 4-4, White will be Player Iwo with ether 5+30r6 +2. The cube wilbein the mid dle. Which is best for White? Consult Table 7-6, We find Player ‘One Black) hasan expectation of 8% if Whitehas 5 +3 whereas Macias S66 Wein + 2 White wars oes st expectation dov so he plays to leave’S + 3. "Phe situation after White makes this moveis shownin Diagram 8 Blackisto rolland hecubeisin the middle, Should Dlack dou- ble? Should White accep? Table 78 says Black should double ‘and White should accept. Table 7 says Black's expected gains 880% of the anesnit dake. ‘Next Black rolls 2-1, Hecan ave 4+1or3+2. Therulemen- tioned says 4+ s beter. To confirm this, note that after Black ‘moves, White will be Player One with $+3, Black willbe Player m ‘Backgarmon ‘Two with ther 441, or 3+2 and White will have the cube. ‘Therefore we consult Table 77, not Table 7-6. If Black eaves 41, White's expectation s2% ofthe current two-unit stake If ‘ck eaves 3 +2, White's ergeetion 15%, Thor Bick wes4+1, Diagram 8 V is now White's tur, The situation is shown in Diggram 9. The stakeis 2 units, White's expectation is 2% of 2unts or O4 tunitand Whitehasthe cube. What should do? Table7-8telisus ‘White should not double. White now rolls $2, leaving 1 +0. Black does not have the cae. Table 7-5 gives his expectation as 61% of 2 units or 1.22, Uhlis, He wins of loses on this next rol “The tables show certain patiers that elp you to understand them better. For instance, fora given positon itis best for Player ‘Oneto have the cube. Itis next bes for Player Oneifthecubeisin| the middle and tis worst for Player One for Playes Two to have the cube. Therefore fora given position, Payer One's expecta- tion is greatest in Table 7-7, east in Table 7-5, and inbetween in 13ith ‘Table 76, For instance, with Player One having 6+ 6and Player ‘Two having 4 +4, Player One's expectation is 2% if he as the cube, 16M i i isi the middle, and 7% if player Two has the cube: ‘Sometimes to orevenallof the expectationsarethe same, For instance if Player One has 6-6 and Player Tw0 has 65, Player (One's expectation is 1% if he has the cube or if it's in the mid- dle. If Payer Two has the cube Player One's expectation dps 10 36%. ‘Examination of the doubling tratepis in Table 78 shows that the postions where Player One should double and Player Two should fold are the same whether Player One has the cube or the ‘cibeisinthemiddle, Although this happens forthe two-man ea positions we are analyzing here, it isnot always tuein backgam- fron, The postions where Player One should double and it doesn’t matter if Payer Two accep or folds also are the samen ‘Table 7-8. Dut some of the posiuons where Player One should wo Beckgonnon double and Player Two should accept are diferent. Player One hus the cube, Table shows that he shouldbe more conse ‘ative Incite, hiss because he asthe cube and doesnot double, he prevents Payer Two from doubling, whereas if the ‘ube ie in the middle, Payer Two cannot be prevented from doubling. “Table 7-8 leas to an example that will confound the inition of almost al players. Suppose Player One has $+2and tas the ube, Consider two cases: @) Player Twohas | + 0and ) Payer ‘Twohas6 +0, Inwhich ofthese cases should Player One double? ‘Geatly 60's a worse postion than 1+0. And the worse the posion the more lly we are to double ght? Soothe four Possible answers (double 1+0 and 6+0, double 1+0 but not 0, double 6+0 bu not 1+0, don’ double 1 +00" 64-0) we "know" we can eliminate “double 1+0, don’t double 6+0," ‘ight? WRONG. The only comet answer, fom Table 7-8: d0u- ble 1+0 but don't double 6+0. Try this on your eet fends. ‘They will almost always be wrong. Itey do gett ih they pro- bably were ithe “icky” ore this chapter Inthat ateityou ak them to explain why ther answer is correc, they probably won't be able 0. "You may think that the os would be sight by doubling 6 +0 ssroneouly. But you have an expected gan of 29% by not dou- ting (Table 7-7) whereas by doubling it ean be shown tat your ‘expectation drops to only 1%, “he exact explanation is complet. The bass ide, though ic that if Play: One doubles Player Two, Player Two accep and Player One does’ win at once, Player Two can use the cube aginst Player One with great effect at Players Two's nex tum, “acoby and Crawford discuss what is essential the same ex: ample (hey pve Player Two 4-+1 instead of 6 +0) on pages 6-17 oftheir excelent The Backgammon Book, Viking Pres, New York, 1970. Table 78 shows that exseallythe same stu tion oeurs when Player One has $+-2and Payer Twohas4+1, 5440,6+0,2+20r3 +2and for nother wo-tan end postions, “Tables 75, 7-6, 7-7, and 7-8 present the complete act oli 5The Mashematis of Gambling Sons to two-man end games in backrammon. The tables were calculated by a general method I have discovered forgetting the ‘complete exact solution to all backgammon positions that are ‘ure raoes(..thetwo sides are permanently out of contact) The Fhtncate and dificulr computer programs for computing Tables 75 through 7-8 were written by Don Smolen so Tables 7-5 through 28 are our joint work. Don was a computer scientist at ‘Temple University. He i now trading stock options onthe floor ‘ofthe American Stock Exchange. A siled backgammon player, fhe won the 1977 American Stock Exchange tournament ‘The Rules of Backgammon ‘The backgammon board is divided into two rectangles by the bar Each seo the bard contains ix points altemating ol os The game i played witht and ark pos calle stores, ‘wl he igher olor designated "White" abd tbe darker oot "Black." Each player hs fiten sone. “The sions are inially ranged as shown inthe diagram boven case, the Black player would be ete at the bot {omofthe board, wil White would beat hetop.Theponis we ‘unbered onthe dagram fr tesake of larity; so murs ap- ‘earon an actual backgammon board. Thesxposin he uper Fehthand comer conte acs ner abi The sx pots {heloner ight are White's ner abe. The objet ofthe eae tomoveyour sone round the bard unt theyre ali your Ser talss In scam, Whe woud stove fi toner counter clocks and Black would progres clockwise. Onc your stones ‘areallin your inner table, you begin to bear them off, and the first Bayer to remove allthis stones from the board wins he gare. woThe Mathematics of Ganbling Backgammon splayed with two die, which are shaken in and rolled fromacup. To begin pay, each playerrolisone die, andthe high roll ets the fist cur. The frst move is determined by the ‘wo numbers which the opponents rolled. From then on, each playerrolstwo dice wien i ishis tin tomove. The two numbers rolled dictate the players’ movesas follows, Suppose the numbers ‘onthe dice are Sand 3. The player may a.) move one stone five ‘ointsand then three more, b.) moveone stone three points then | five more, orc.) move one sione five points and another stone three. When a double number isrolled, the player may make four ‘moves, Arollof3.3 wouldllow moving one tone I2points, four stones 3 points eas, ox aay other patter involving groups of ‘When moving hisstones around the board, no playermay and ‘ona paint occupied by two or more onpasingstones, Such apoint {Sconsidered madeby the opponent and often interferes withthe way in which a player intended to take his turn. On the other hand, a point occupied by a single stone is a blo. This point is vulnerable to being fi by an opposing stone that lands on it ‘Whenastonei i, its sent ofthe board and onto the bar, where it must remain unt it can be entered on the opponent's inner table. The payer whose stone as been hit is forbidden to make ‘any other moves unt he has entered his stone from the bar. Tn order to enter, the player must rll a nuriber which allows ‘him to move to a point on his opponent's inner table which snot ‘made. For example say White zs it one of Black’ blot, send ing it to the bar. White has made points 3, 5, and 6 on his inner table. Inthis situation, lack must olla 1,2, or4in order to enter Ris tone ‘Once a player has succeeded in moving al fifteen of his stones, ino his inner table, he may begin bearing off This consists of removing stones from the points and off the board according to thenumberstoled. For example, arullof 3-4 means that player ‘may remove stones from positions 3 and 4, Ife hasno stones on tone oF both of these points, e bears off from the next lowest ‘Polat, The race” tobear off continues until one plaverhastaken 8 sackgannon all his stones off the board. He, of course, isthe winner. ‘A players credited with having won a double game, or @gom- ‘mothe bears off all his stones befor his ponent has orae off any. Ifa player wins a game whilehis opponent til has stone on the winner’s inner table oron the bar, he has made a Backgam ‘mon and wins triple te stakes. If you play backgammon for money, a doubling cubes used. ‘This cube bears the numbers 2, 4 8 16, 32, and 64. At any point during the game, one player may douilethe stakes. His opponent ‘must either accept the double or forfeit the game, If he ccep, the opponent gains possession of the cube and may “tur the ‘evbe” back at is opponent whenever he fels he has the upper Inand in te game, Tes ot hard to se that high stakes game can eure foma gamein wich teladehangestands re ‘quentlySection Four Money Management The importance of money management and bet slz- ing has been stressed increasingly In recent years and rightly so. For even if the player has discovered a favorable betting situation, how he wagers determines his success or failure. Ultimately itis the “bottom line” ‘on which a gambler’s performance Is judged. ls fine, of course, to describe the favorable situation toa friend or business associate, but the next question is likely fo be “How much money are you making from this situation?” ‘The problem for the gambler is that much of the ad- vice on money management is conflicting or confus- Ing, or simply based on false premises. There are nun- dreds of schemes designed to overcome the house ‘edge in roulette and craps based solely on manipulating the size of ones hets. AS will ba saon, all such attempts are futile Even assuming the player has discovered a favorable {game (le, one offering a positive expectation), the ques- tion naturally arises: How does one best use a limited amount of capital to exploit this positive expectation? Wager too boldly and the player risks losing his entire uwThe Mathematics of Gandling bankroll, even though he or she may have made only favorable bets, This 1s commonly known ae gambler’e ruin. On the other hand, betting too conservatively the player severely limits his opportunity to make a good return on his capital. Fortunately for the player, there exists a mathematical theory for committing resources in favorable games. This will be discussed in Chapter 9. Chapter 8 Mathematical Systems ‘Before looking at the optimal strategy for exploiting a positive expectancy situation, it may be worthwhile to evaluate what 1 refer to as mathematical systems. Although here I use rouleteas an example, te principles apply equally to craps and the Wheel of Fortune. Hy 4"mathematical system” Imean a system where the player decides which bet to make using only the following informatio () arecord of what numbershavecomeup on somenumber of pst spins, and {@) arecord ofthe bets he has made, if any, on those spins. ‘We assume here that when the player bets, fOr him al umes are equally likely to come up on each spin ofthe wheel. This ‘means not using biased wheels or physical prediction method, ‘Roulett ha long been the prototype of unbeatable gambling games. Tis normally regarded asa repeated independent tials Process which generates at each trial precisely one from ase of| random numbers. Players may wager on particular subsets of Fandom numbers (eg. the fist dozen, even, an individual ‘number, ec), Winning ifthe number which comesup isanumber| of the chosen subse. A player may wager on several subsets w‘The Mathematics of Gambling sinmutaneously and each bet is settled without references to the Others, To fk the dscusiou, let's consider the standard ‘American wheel. This has thirty-eight numbers, namely 0, 00 1, 2,36. "Fhe mathematcin’s ssumoton, that each of these numbers is equally lly beforehand to come up on any spin of the tall ‘and wheel, seems plausible. The wheels are carefully machined and balanced by the manufacturer. They ae checked from time Totimeby the casinos. When they show signs of wear they maybe thoroughly reconditioned, Even ifthe wheel has iegulares ‘which make some numbers more favored than others, if the player does not know this and his systems not designed t exploit {his then mathematical reasoning —based onthe assumption that, all numbers are equally Hkely to come up—gives correct conchu- sons about that player's system. ‘The Doubling-up System ‘One more assumption must be made to properly evaluate mathematical systems. We must also assume there i a smallest allowable house (minimum) bet (such as $1) and a greatest allowable house (maximurn bet (uch as $1000). Casinos need to fix a maximum bet in order to stop the simple mathematical system of “doubling up.” To se why, imagine we've found a ‘casino with no maximum, We bet SIOD0, because Red pays even money oF I for 1, If we lse, we double and bet $2000 on the se- fv turn Tf hat wins, we net $1000 onthe wo tums. Ifthe sec- fond bet loses, we double again and bet $4000 on the third tu. ‘Having lost $2000 on the first cwo tums, a win of $4000 on the third turn nets $1000 on the cycle of thfee tums. We continue ‘doubling our bet after each les, Filly, when we win, we havea ‘et gain of $1000, We put this $1000 safety aside and start anew ‘jd of doubling unt we win witha bet of $1000 on the Red. Each completed eyele winsanother $1000 net. Table®-1 illustrates this cyte. "The doubling up system in Table 8-1 with no casino limit on betsis being discussed norbecause anyone would ever be allowed a Manenaticel Sens wes seo +r Ho suaees + ane Bt aioe + anaell —o.sa.sin.sa.7 de Smee + aan anaes to teased + gor 10 do tt, bur to Mustrare ideas we will be using. To see how ridiculous the system would be, note that if the fist ten tums ofa have lost, onthe eleventh tur the player bets 1,024times his initial et. Hi intial het was 1,000, so he bets $1,024,000. OF ‘course the chance is small hat his will happen. The last column shows achanoe of0.9988thatthecycle ends on orbeforethetenth ‘um, hence that the eleventh bets never made. Thus, the chance ‘of eaching in eleventh rum sonly 1 ~0.9984 =0.00i60r0.16% ‘orabout one chance in 613. But if Une doubling-up system is used long enough, it will happen.‘The Mathematics of Ganbling ‘With 30 losses ina row, the player is supposed to bet about ‘ous ilion lla on the thiry-frt cara. This is about the net ‘worth of the New York Stock Exchange. On turn 36, the bet is about $34 uilion. This exceeds the net wort ofthe world! The ret worth ofthe U.S.A. is about 6 tlion current dollars. guess the net north ofthe world to be about $30 willon.) The player should arrange from the st to have unlimited credit, reasonably Pointing out tat since he must eventually win he is ure to pay off Real casinos don’t go for this. They have house limits (which they may increase sometimes under special circumstances) and credit its, So this “sure-fire winning system” is never wed. But players for centuries have used modified doubling-up systems in ‘actual casino play. Aa ilostration i given in Table 82. Here the player starts by biting Son Red. He keeps doubling his bet unt he wins. Then he starts the cycle over with a8] bet on Red. Each jee produces 31 profit wnless—end hereistheeatch heloses ten times ina row and then wats w bet $1024 onthe eleventh turn ofthe eyle. The house limit prevents that and prevents furher ‘doubling if the player Ioses on his eleventh turn, "Notice from Table 8-2 that if the payer wins after nine or fewer tosses, he wins $L and successfully competes the cle. But if ze loses ten times ina row, be can bet only $1000 on the eleventh tra If he then wins, he loses “only” $23 on this eyle. Bot sf he loses on the eleventh turn, he loses $2023 on the cycle, for ‘a major disaset. OF course, the chance of ever reaching the gleam ofa ke bore ony tot oe ee Is this system any good, or do the chances of loss on the eleventh turn rin it? "We aie ging Co find out thatthe “house percentage advan tage” on Red i not changed in the slightest by the doubling-up ster, In fac, the disaster ofthe eleventh turn s exact compen ‘sition othe casino forte igh chance the player has of wining. Si per eye. We will show this by 8 computation. But what is ‘pethaps truly amazing is that hiss also true forall mathematical systems, no matte how complex, including all hose that can ever us Maenatcal Sens poor 8 a a ea 1 mmkwe one ae 1 cums coast sow os 1 amu ees #6 1 cum eoiet + ow ow 1 carci 0.0089 woo su 1 come oo Hie eae esen be discovered. Sunce there are an infinite numberof suc systems, ‘We cannot prove this by computation (an infinite amount of ime ‘would be needed to do the required infinite number of computa- tions), Instead, I will indicate how the mathematician, by logic (ike the logic of, say, plane geometry with ts axioms, theorems ‘and prools) can show that none ofthis infinite numberof stems is any good. ‘Alot of what 'm saying is casi than it sounds. For instance, ‘osee tha thereare an infiniterumber of systems for roulette, al, Thaveto doisgive you any endlesslist of systems. Hereisonesuch lit always bot on Red); System I. Bat Son Redif Red cameup ‘one turn ago; iit didn’t come up one tum ago, bet $2. System 2 “Always bet $I on Red if it came up two turns ago; iit did not ‘come uptwo tums ago, bet $2. Andsoon forsysiems3,4,,. tc. wThe Mathematics of Ganbling 1 didn't say my Ist of systems would be interesting, only that it ‘would be endless! “The doubling-up system can be good for some fun even if it ‘doesnt alter the house edge. Suppose you're in Las Vegas with your spouse of your dat. It's almost dinner time and you sy Casually, “Dinner for to will run us about thirty dors. Why don’t we ea for free? Il just pick up $30 at this roulette whee. It'll only take a few minutes.” Ifyou have $2100 in your pocket ‘and the house mits ae fom $110 $1000 on Red, youcan wse the oubling-up system, You need to complete 30 cycles without ever having 2 string of eleven loses. You will win $1 per cycle, fora ‘oral of $9, and be off to dinner. "How safe is this scheme? What are your chances? Table 8-1 saysthat the chance acyelelass1Otumnsor less, and therefore you ‘win SI, i 0.9984, The chance that you do this 30 times ina row Turns cut to be 0.9984" or 0.952, so te chance you wil succeed ‘over 95%. Ifyou set your sights lower, say $20 or $10, hen the ‘chances of success g0 Up t0 96.79% and 98.38%, respectively ‘Dat be wameds if you fl, you can lose as much as $2023. "Aa important factor in determining the risk of lure is the ratio of the house maximum bet on Red (o the minimum bet. To ‘strate, suppose instead of $1 to $1000 fora rato of 1000, the betting limits were $210 95W), for arti of 30/2 =250, Tien if westart a cycle wth a $2 bet, we ht the house mit on the nina Spin, after eight lasses, (To ste this, use Table 8-2 and double all the humbersin the second, thied and fourth enim, because we ‘start witha $2 bet rather thana SI bet, asbefore.) Now thechance the cycle ends in eight tums or less i (rom the last column. of “Table €-1)0.9941, Thusto win $30 you need to complete 1Syets, the chance of which is 0.9941” or U.9132, If you try this in a roulette game with better odds, say single-zero European style, the chance of Success increases. The doubling up yetem ix one of a class of systems that are sometimes called maringales. The origin of the term is given in the American Heritage Dictionary, New College Eaition, whichis the most informative definition Ihave seen on ths. The word us Mathematica! Sem ‘volved from a similarly named village of Martigues in the Pro- ‘ence distinct of southern France, hes residents Were Viewed as Pr gi wee oly deed wih Gale xe, Thar Dizarre behavior included such things a5 gambling with the featlngspssem anaes amet oe the doubiing-up system became known as gambling, “alia max- tigalo” (fem), “inthe Martigues manner". “in ridiculous “There are many other popular “mathematical” systems. “eTeipling up,” where the player bets 139,27, etc. unt be wins, then repeats, is ike doubling up, bu twins aster and rus into freuble Gin the form ofthe hoe Bini) faster. Tf you want fo know more about “matheratical systems,” conse ce book ebook Casino Gambling, Why You Win, Wy You Lose, by Rasalan rie No. 8239, Ba hartisa skied magician and alongime ten of gambling. He tas gambled extensively all over the world so he knows both the theory and pracice of i subjet. The book has 80.00 spins fom an actual whoel and an elaborae discussion of mathematical or "aking" systems ‘Alan Wilson's classic Casino Gambler's Guide has con- siderable materi un syste a et lls, His Geant of based coulete wheels may be the best ever writen, ‘Richard Epstein’ engaging ease, The Theory of Gambiing and Statistical Loge. Revised. (Academic Press 78 2nd mark in the subject. Much of it requires a university devel ‘mathematic background. However itisthe best single reference ‘workin print on the general subject of games and gambling, and ‘rene general reader can glen: mach from browsing troup. ‘Nov explain why mathematical stems Uke the doubling. up system, cannot redice the easino percentage. ‘The Problem with Doubling Up __ One reason Ichose roulete to illustrate mathematical systems {s that itis easy to understand the ods and probabilities. »‘The Mathemates of Gonbling One corect version of the so-called “law of averages” says that {na “long” serie of bes, you wll tnd to gain or lose “alpout” the total expectation of those bes. This means that a series of “ad?” basi also “bad,” and that systems don't help. "Applying these idea tothe Uoubling-upsysicr, lets calculate the player's expectation for one cvce. Thnk of a complete cycle asa single (complicated-looking) bet. Now refer to Table8-2, The fifth column gives the probability that the cyce ends on turn #1, 22, ef andthe Fourth column gives the gain or loss for each of these cases, Multiply each entry in the fourth column bythe cr- responding entry in the filth column, Then ad the resus $1 18/38 +81 x 20/38 18/38 +... +81 x 20/38) X18/ 38 S23 x 20/38)" 18/38 ~ $23 x(20/38)" «18/38 ~ $2023 x {@0/38)" which simplifies to I ~ 24 x 20/38) * ~ 2000 x (20/38)" =00391, += 1.7168. « —~$.7560266578. Thus, the expected los to te bettor is about —8.76 per cyte. ‘Now let's calculate the expected (or average”) amount bet on one cycle. Referring aii to Table 8-2, we sce that if the eyle endson tum #1, the otal of al betsis SI, ifit ends on tun #2, the {otal ofall bets is $1-+$2, if it ends on tum #3, the total is SS1-+89+84, et. Ifthe cycle ends on turn #11, the total amount Detis $2,023, (To get these totals as of the end of any tum, add ‘cls two and thre.) Then multiply these total amouns bet by the chances in column five to get $1 x 18/38 + $2 (20/38) (18/38) + $4 (20/38) (18/38) +... 5 $512 <(20/38) (18/38) + $2023 x (20/38)'* x(18/38) “4 $2023 x 20/38)" which simplifies to $2 (18/38) <((40/38)" YAO) 38 1) +9004 (2/38) —S1 = S14. 3645065. IC we Give the expected loss by the average bet per cycle We Bet W8.756.,- =814.36...1/19exacly or ~ 5.26%. "These ealulations are tedious, and foreach system the details are different, so they have co be dove again. And there are an Infinite number of gambling systems, so calculations can never ‘check them all out anyhow. Clearly this isnot the way to under. ftand gambling ystems. The cortect way is to develop a general 0 Mahenatcal Sens ‘mathematical theory to cover gambling systems. That has been | done and here's how it works. Firat we define the ation in & ‘specified et of bets tobe the total ofall bes made, From what we hve said, your expected (gun or loss your action (Le, the tot ‘ofall your bets) times the house edge. For example, ifyou bet $10, perhanda blckiack and play for {Ohours, betting 1ODhands per hour, you havemadea thousand Obes, which is $10,000 worth ‘of “action,” I'youarea poor blackjack payer and thecasno has| 123% edge over you, your expected losis $10,000 3% ~ $200 ‘Your actual loss may be somewhat more of somewhat les. I Nevada casino blackjack grosses a total of $400 ailion per year and the average casino edge over the player is 2M of the intial wager, then we can detemune the total action (A) pet year: 2A =$400,000,000 so A =$20 billion. Thus from these figures ‘we would estimate $20 bilion worth of bets are made per year at Nevada blackjack. The 2% Figure might be substantially off. We ‘could get a fasly accurate idea of the rue figure by making a ‘careful statistical sampling survey. I, instead, the Digure is 4b, then A=$10 bilion. With 1%, A'=$40 billion per year. Guidelines for Evaluating Systems “The general principles we have disused aply to almost all gambling games, and when they apply, they guarante that Sem cannot ive the paver an advantage. "To help you reect. systems, hete are conditions which auarante that a gtem is wrt 1. Each individual bet in the game has negative expectation (Ms nike any ser of Bets have ncpative expectation ) tal Tie mum ito ee amy posse pe rules out systems Eke the no-imit doubling up sytem discuss) uns wo TL The results of any one playa the game donot “infuence” the reaulsof anyother play of the game. (Thus, in roulet, we assume that the chanes are equally key for ll ofthe numbers 2The Mathematics of Gambling ‘on each and every future spin, regardless of the results of past spi) TV. There is a minimum allowed size for any bet. (This is necessary forthe techrical steps inthe mathematical proof. Most people would take for aranted that there is such a minimum, hamely some mulliple of the smallest monetary unit, In the U.S. theminimum allowed betissome multiple ofonecent. In ‘West Germany, it may be some multiple of the pfenning, and so font) ‘Under these condidons, i is a mathematical fact that every possible gambling system & worthless in the following ways (1) Any series of bets has negative expectation. {@) This expectation i the (negative) sum ofthe expectations ofthe individual bets. (8) Ifthe player continues to bet, his total loss divided by his total action wlltend to gel cover and closer tohisexpected loss divided by his total action. (If the player continues to bet itis almost certain that he wil: G@) be aloser; (b) eventually stay a loser forever, and 50 never again break even; (0 cventualy loses entre bankro, no matter how large ‘To give you an idea of how valuable this result is for spotting worthless systems, here are some example of systems which can- not possibly give the player an advantage: 1. All the roulete systems I have ever heard of, except the following two types. (2) Biased wheel, in which condition (I)may be violated; the numbers are no longer equally likely, so bets om some numbers may have postive expectation. (b) Physical prediction methods, in which the position and velocity of balland Fotor are used to predict the ovicome. 2. Allcraps systems Ihave ever heard of, except possibly those using ether crooked dice or physical control” of dice, (Noe: While at the ith Annual Gambling Conference at Lake m Mathai Sytem ‘Tahoe, I saw a dice cheat contol the dice, at a private showing then baw him win ata easno. heard he it ts cegulaly. His badly mutilated body was found inthe Las Vegas area a year later, ‘3. Any systems for playing keno. slots and chuck-a-luck. ‘Asa further lutration, consider the book Gambling Systems That WAN, pubiished by Gambling Tines, 1978, pes back, $2 (Ff the fourteen systems given there, our result applies at onceto eight. (The other six are one blackjack system, four racing systems, and a basketbal system.) (in the case of spors bets, itis generally difficult determine ‘whether condition Iis satisfied. In the case of blackjack, condition [fails if the player counts cards and thereare, in fact some win- ning systems, as most of you know) This eaves eight systems in WIN: four craps systems, one bac- carat system, two roulette systems, and a keno system, CConaitions I ehrough IV hold forall ight systems s0 none of them are winning systems. Nor do any of them reduce the house cedgein the slightest. However, they may be entertaining. Also in games ike keno craps, and roulette, where the expectation may vary from one game to another of from one type of bet tO another, some ways to bet are “smarter” (translation —less dumb; mnore accurate translation —less negative expectation but sul losing) than others. Forthose whoate prepared tolose, but want tolose moresiowly, such systems may be of interes. ‘In most cases, the basic information sa list ofthe various bets inthe game and ther expectation. Then, if you must play, choose ‘only bets wit the leat negative expectation, The sysem” com- lexis and hieroglyphs are not essential. Timay amuse you tose why condition TVs needed. Suppose, instead, tha there sno minimum bet and that wear playing Red atroultte. Our fis bet is $1,000, There is an 18/38 chance that ‘We win $1,000 anda 20/38 chance we lseS1,000. Now sunnose that the second bet is $0.90, de third bets $008, the fourth bet {i SOD, the fifth bet is $0000, ete. (Remember: no minima.) ‘Then the total ofall bes from the second on is $0.99999,.. 1.00 aThe Mathematics of Ganbling ‘The total gan or loss on these bets i beween—$1.00 and + $1.00. ‘The toll action on all bets is $1,000 S1=91,001. Tf wewon the fist bet, our total winnings () will alwaysbe be- tween $999 and $1,001. This happens with probability 18/38, ‘Therefore, conclusions f(a), (2), and (6) fail. Also, our total action is $1,001 so T/A is always between $999/S1,001 and 1,001/$1,001. But our expected gain ()snewative so E/Aisless than 0. Therefore if we win the frst be, T/A doesnot tend to get ‘over ane cose to E/A. Therefore, eclusion 3 aso falls ‘Conclusion (also deservessomecomment. Actually, thereis an insignificantly small chance the player can win the casno's bankroll before losing his. But even for moderate-size casino Dbankrolls,this posibility iso tiny as to be neglsble, no mater hhow large the player’ bankroll! We will discus this in the next chapter. Is also discussed at some length in the 1962 edition of| iy book Be she Dealer, and in Flle's great av Inoduction ‘to Probability and its Applications, Vo. 1, Wile. Thus, a more ‘exact version of conditions F1V would include information about the sizeof the casino bankrol. Then conclusion 4 would include information about the tiny chance that a), (2), and (¢) don't hhappen. ‘AsfarasTknow,themost elementary mathematical proof ever iver for allthis is in my textbuvk, Bleracruury Probubiy, Available from Robert E. Krieger Publishing Co. Inc., 645 New ‘York Avenue Huntington, New York 11743. The proof soutined on pp. 8485, exercises 5.12.and $.13. It requires no calealus and ‘ean be followed by a good high school mathematics student it he ‘or she works through pp. 1-85. ‘Wenow havea powerful test for showing thata system doesn't ‘win. Ths keeps us from wasing out money and ine baying oF playing losing systems. Ic also helps usin our search fr systems that do win, by greatly narrowing the possiblities. Chapter 9 Optimal Betting —_—_—_—_.__. —_———————__ issomewhat ridiculous to dlscussan optimal money manage ‘ment strategy when the player has a negative expeclancy, As indicated in Chapter 8, with an enforoed house maximum and ‘minimum wager, there sno way to convert a negative expectation io & positive expectation through money manipulation. AN 00d money management plan say not to wager in such astua- tion. Players facing a negative expectancy should look elsewhere for a sambling game or, at the very last, bet insignificant amounts and write off in their mind the expected loss as “emertainment.” ‘After the gambler has discovered a favorable wagering situs Sion, les faced withthe probiem of how best to apportion his limited financial resources. There exits a rule of formula which you can use to decide how much fo bet. I will explain the rule ‘and tell you the benefits that are likely if yeu follow it Ler’ bepin with a simple istration that I deliberately exag- erated to better get the dea acros. Suppose you havea very ich adversary who wilt youbetanyamnount on heads ateach tossof| nsThe Mathematics of Gambling ‘coin and that you both know thatthe chance of heads i some ‘umber p greater than. It your bet payseven money i you have an edge. Now suppose p=0.52,s0 you tend to win 52 pr~ cent of your bets and lose 48 percent. Tiss similar tothe situa tion in blackjack when the tnscount ratio i about 1.5 perent. ‘Suppose too that your bankroll is only $200, How much should ‘Youbet? You coud play safe and just bet one cent each ime. That ‘way, you would have vitally n chance of eve losing your $200 and being pul out of the game. But your expec gan is .04 por Unit or .Ofcents per bet, Atl00 one cent betsan hour, You expest to-win four cents per hour. It's hardly worth playing. Now look at the o*her exireme where you bet Your whole Dankroll Your expected gain & $4 on te fist bet, moce than i ‘you bet any lesser amount I'you win, ou now have $200. Ifyou ‘gain bet ll of ton your second tum, your expected gain is $B and is more than if you bet any Ieser sauount, You make your expected gain the biggest on each tum by betting everything. But if ou lose once, you are broke and out ofthe game. After many tums say 20, yo have won 20 straight tosses with probably. "52" 0, 000002090 and have a fortune of $104,857,600, or you Hhavelost once with probability 0999997010 and have nothing. Ln ‘general, asthe number of tosses increases, the probability that Sou willbe ined tends 0 1 or cetany. This aes thestrategy Of betting everything unattractive ‘Sinor the eambling probabilities and payoffs at each bet are the same, it seems reasonable to expect thatthe “best” stratcey Will ‘always involve betting the sae fraction of your bankroll at each tum, But what faction should this be? The “answer” isto bet 2 fi — p)=0.82 ~ 0.48% 0.040 four percent of your bankroll cach tne, Thus you Be $4 the Dist tine, If you win, you have $104, so you bet 004 x $104 = $4.16 on the second tua. Ifyou lost the frst turn, you have $9550 you bet 0.04 x $96=$3.840n, the second turn Vou. continue to bet four percent of your ‘bankrollat each tum, Thissrategy of vesting” four percent of {your bankroll at each til and holding the remainder in cash is known in investment circles asthe “optimal geometric growth 26 pina Being portfolio” or OGGP. In the 1962 edition of Beat the Dealer, 1 Gscussed its application to blackjack at some length Tare Lelled itthe Kelly system, after one of the mathematicians who studied it and 1 also refered to it as (optimal) fixed fraction (of your ankzol) bewing. ‘Why Js the Kelly system good? First, the chance of run i, “smal” Infact ifmoney wee infinitely divisible which tear be if we use bookkeeping instead of actual enins and bills, or if we tse prosious metals such a gold or sve), then any system where you ever bet everything will ave 2eo chance of ruin because ‘even if you always lose, you sil have something left after each bet. The Kelly system has this feature. Of couse, in actual prac tice coins, bilbor chipsare generally used, andthereisa minimum size bet. Therefore, wih avery unlucky series of bets, one could ‘eventually avesolitelet that he hasto bet more of his bankroll ‘than the system calls for. For instance ifthe minimum bet were Si, then in our coin example, you must overbet once your bankrolis below $25, Itheminimum bet were one cent, then you ‘only have 10 overbet once your bankroll falls below 25 cents. IF ‘the bad luck then continues, you could be wiped out. The second desirable property ofthe Kelly ysteristhatfsome- ‘one witha significantly different money management system bets fo the sume game, your total bankroll will probably grow faster than is In fet, as the game continues indefinitely, yourbankroll, will tend to exceed his by any preassigned multiple. ‘The thd desirable property of the Kelly systemis that you tend toreacha specified level of winnings in the least averagetime. For ‘example, suppose you ae a winning card counter at blackjack, and you want to run your $400 bankroll up to $40,000. The ‘number of hands you'll have to play on average to do this wil ‘using the Kelly system, be very clase o the minimum possible us ing any system of money management. “To summarize, the Kel system i relatively safe, you tend to hhave more profit, and you ten to get to your goalin the shortest time, @The Mathematics of Gambling Blackjack Money Management “The Kel system cals for no bet unless you have the advan- tage. Therefore it would tll ou toavoid gamessuchascrapsand Xeno and slot machines, However, ifyouavethe knowledgeand sto gain an edge blackjack, youcau usec Kely yates 0 ‘optimize your fate of gain. The situation in Blackjack is more complex than the coin ess game because (1) the payoff on aone- ni ial het can vary widely. due to such things as dealer of player backjacks, insurance, doubling down, pai spliting, ad Surrender and (2) because the advantage or disadvantage fo the player varies from hand 0 hand However, we can apply di wi ws cesuls to blackjack by making some sight modifications. Fis, let's see where he coin tossexampe's bes fixed fraction of four percent came fom. The ‘general mathematical formula forthe Kelly system this: In any (Ginele) fvorable gambling station or investment, bet that rac- tion of your bankroll which maximizes E In (1 + 1), where Eis the expected value and Jn i the natural logarithm (co the base 71825...). This In fasnsio iy avallable on most hand Calculators. in the ease of coin tossing, the best fraction, which L call is given for a favorable bet by /*=2p ~ 1, where pisthe canoe of sues on one toss and "= Of = 1/2, i.e ifthe ames ether fair oro your disadvantage. Note too that /*=2p *'Tiscoineidentally your expected gain pr unit bet. ‘Now your expected gain in blackjack varies from hand to than Ir wei of succesivehandsaseoin toses witha varying then weshouldbetf*=2p — I whenever ourcardcount shows that the deck is favorable, When the deck is unfavorable, we "should bet zero, Usortyoe eam sla aoroximatesthisideal ofbeting zero inunfavorable situations. You canalsoachievethis Sometimes by counting the deck and waiting unt the deck is favorable before placing your fist be. Butitisimpractcal to bet ze In unfavorable situations, so we betas sinall ass discret. Think ofthese smaller, sghtly unfavorable bes asa ‘‘drain” oF “eax” which “water down” the overall advantage of the ne Opti Bevin favorable bets. To compensate for this reduced advantage, f* should generally be “sightly” smaller than the 2p ~ Zeompuied above. Another effect ofthe smal, sighily unfavorable bets isto ‘increase the chance of ruin a ite “The most important blackjack “correction” to the /* com ‘ped for coin essing is due othe greater variably of payor. Peer Griffin caleulates tha the “root mean square” payoff on & fone-unit blackjack bet is about 1.13. It tums out then that /* should be corrected to about (2p ~ 1)/1.27orabout 79timesthe advantage. Shade this to .75 because of the “drain” of thesmall, ‘unfavorable bets and we have the farly accurate rule: For ‘favorablositutions at blackjack, ic (all) optimal to bet aper cent of your bankroll equal o about 3/4 percent advantage. For instance, with a $400 bankroll and a one peroent advantage, bet 3/4 of one peroet of $400, oF $3. ‘The Kelly System for Roulette Jn general in roulette, the house has the ed, and the Kelly ‘sem say, “don't bet.” But in my chapter on physical predic tion atrouleste, I desribed a method where we (Shannon and), with theaid of an electronic device, had an edge of approximately ‘4 percent onthe most favored single numb. That corresponds toawin probability of p= 0.04, with apayoff of timesthe bet, and a probability of ~ p = 0.96 of losing the be. It turns out that += 44/95 ~.01257 The general formula for /* when you win A times a favorable be with probability p and lose the bet with probability Ip, is f*=e/A where @=(A+ Ip~1>0 the plaver’s expected gain De unit bet or his advantage. Here Av35, p=.08, and @= 04. In the coin toss example, A= J, P= 32, and e=.04. Using any fred betting function f; the “growth rate” of your fortune is GO)—p In d-=Afs ~pyin(t 9. Abr N bets you willhave approximately exp/’ (times asmuch money, ‘where ex isthe exponential function, also given on most pocket calculators. mmThe Mathematics of Ganong For the roulette single number example, using my hand calculator (an HP6S) gives GU") = WOE In (1 + 537) + 0.96 1m (1 ~ f°) =.04 In (1.48) + 0.95 In (0.98743) ~.08 x 36464 + 0.96 x (—-0.01265)= 0.1439 — .01215=.00044, After 1,000 bats, you wl have approximately exp)? 44) 17.47 times your starting bankroll ‘Notice the small value of /*. Thats because the very high risk. ‘of loss on each bet makes too dangerous to beta large fraction ff your bankroll. To show the advantages of diversification, sup- pose instead that we divie our bet equally among the five most Favored numbers, as Shannon and Tactually did inthe casinos. fone of these univers come up, we sen an amonnt equal to 5 779/Sof our amount bet, and if ne come up, we lose our bet. Thus A=31/5=6.2 The other four numbers are not quite as favored asthebest number. However, toillustrate diversification, ‘suppose thatthe Fivesway betas the same 4 advange. TH ‘onesponds to p=0.20, Then f* =44/6.2= 0.07097, so youbet bout sven pereent of your bankroll and GU")=0.20 In (+ G.2/*) + 0.80 In (1 ~ f¥)=0 0ldbd. This growth rae is about 5.75times that forthe single number. After 1,000bets, you would have approximately 1.25 million times your starting bankroll. ‘Such isthe power of diversification, ‘What isthe pice of deviating trom betting the upd Kelly fraction *?1t tums ou that fr bet payofislke blackjack, which canbe approximated by coin tossing, the “performance Loss” is| hot senous over several days play. But forthe roulette example, the performance loss from moderate deviations from the Kelly stem is considerable 0 APPENDICES APPENDIX A. Sompose ol count sytss which are “loser tthe selatve a ylies of ble 22 ate ily fo be "better T est hs we area precise meaning for "beter anda precise menue of “SIncnes” We begin fy hang the efition “heen the nodoat of probable dominance, and of isk used in Iathematcal finance. Defnion I, Let F and Gbeprobebliy distribution functions. pent pcheiy doa 0) = Ut Beal ib alion FQ) < Cl) for atleast on then F sry probahlsicaly dominates’G. If F and G arse for random Sp and ae iy ee Ady, rexpectvely then the defined ass apy ese pls if iey hod for F and © ay eee ‘Tat F probability dominates Gis equivalent to POX 22d) forall x I ste payer expection loa point couse A and ste pyrene fina en en this means thet the chance of ining expectations of xo nore ‘Satu lest good a sing Aarts by wing B Ove son Show tat his meas tat player following Abas an expected return as B with "he sae isk lve aThe Mathematics of Ganbiing However, probabilistic dominance is inadequate a a dsfinition of “better because the typical situation is that Fis “spread out ‘more in both directions from the mean fall deck expectation E,=0. Thus F dominates G for x > E, and G dominates F for x < By, In fact G is (to a good approximation) a convex con- traction of F. More precisely, if E, and E, are the respective means of Fand G, we will rd E, 2 E, >, with YE, a con- ‘ex contraction (thei equvalent othe notion “Tet risky than” of pontolio theory); of X'- E,. Thus F is both “spread out more” than G and translated in the positive direction more. The teason why E,, E, 2 E, is because Ey i the expectation using the basic stralggy and constant bes, equivalent to the full pack expectation. When (advantageous) counting systems are used, the ‘strategy for playing hands is improved whenever the player has ‘2en any care other than tho onet he and the dealer use on the first round. Since this generally happens with positive probabil {y, we then have E,, E> Ey ‘Definition 2. Point count sytem A is better shan system B if E, E, and also POX =x) P(¥2x) for xB, “Typically count systems satify Ep = E, = E, and X- (YE,), a = I (a special case’ of convex contraction). ‘These conditions imply Avis beter than B. ‘Assume thatthe betting ystems b(E) are numerical functions ofthe expectation B, Further assume b(E)=1 if E50 and b(E)= 1 if E>0. These are the ones generally considered. The popul fallacious systems such a the martingales (c-. “doubling Up"), and the La Bouchere which incorporate past resulis, are of no interest here. ‘Theorem 3 With the preceding notation and assumptions, if A is bener shan B, ten for any betting system b,(E) based on the B point count, there isa betting system b,(E) based on the ‘A point count such thatthe return R, per nit bet by A (approx imately) probebilstically dominates R. Further, R, and Ry have approtinately the same risk, Infact R,=Ryte, Where C20. mB Arpenices Proof: If F and ©. are continuous, define by by BANGLE) ))=byE). Then woe at he Ht ut exh bat has expecation Ey fr A and, for B. The remainder ofthe bet isnonzero only if E = Ey. Then for comesponding percents ofthe respecte distributions, paces the sames besa B. Bat FOE) © GUE) WE © By so A bas n each isunce lat ab azeatexpecation, hence hast east a reat expectation overall, ‘Ths the tol expected retro Ais t least le fr B ‘Alo By = Ty per unit nave te bls paced ne the sane dsiation Tn realty F and G are ot continuous instead they ae fie Bt they may be arhiarty closely approximated y continous distributions s the result exeads, with one qualiftion. If F oF G is ctconinious, extend the graphs of F and by ang ‘erica segments at the discominsty pons so that te exene Siena F and G have inves definet'on (0. Thea for tone sich that G is discontinuoes at E” or Fis discontinius at PAGE) it may be neosssury w define byE-(GE))) “prbabilsialy sits mulple-alued, each ave oer ing with speci probablis. “po dom that R= F,whch implies he same risk it sue foes wo assume tha each prcetle level y forthe ditt thom Fand G we hv the conditional Gstbon given sts ‘ying Faly)=Goct]) where 9) = 0 Since this cal olds approximately in practice, we have R= Ry Now we tu othe problem of messring closeness of a given cain tothe “ulate cg We sl ase th point oun strategies are ofthe form C=C.) whee ise Yale assigned fr an ace, ey fe the Pine cOUNS FOE Tinks 2 trough 9, and eg" me) ae te point coun for ten, acs, queens and Kings eapostvely.Inpravtse these are Tumped togetir and only ten point count values are specie. By wring C wih 13 components we aun asymmetry which yds substantially simpler proofs. Noe that C and aC, de equivaleat and wil be entedThe Mathematis of Gambing Definition 4. f £4 E.=0 the ulimate strategy U=(0,s---0) is the one given by =A F, where AE, i the change in expec: tation from removing one th card from the complete pack. If 14 E=0 then U is given by und. 1p Table 2.2, we have d for ane deck is O24 and d for four decks is O17. The v, rows are calculated in Table 2-2 from Definition 4 Tis tempting to think of U as representing to good approxima- tion te aiection ofthe gradient Fat 3 of Ue plyer’s expectation Ef... sf) as «function of the fraction f of the cards from i=! 10 B. Then we calculate (C) UACHSUIICHSIUL, i. the projection of C in the direction. “The numerator isthe ianer oF scalar produet and ICI =(E<2)*. ‘Next we claim that d(C) gives the approximate ratio of the spread ofthe C distribution F, about Ew the U distribution F, about E,, Then d(C) is che Uesire measure vf eluseness. Ih prtcula, for approximately the same risk pr uit, ard the same Sistribution of the bet sizes, it would follow that E(R) = F(R)i(C). Then C, and C, are arbitrary strategies ER, VER = N(Cy/A (C) forthe same risk level and distribution of bet ‘ies. Thus the “power” ofa strategy Cs proportional wits (C). “This conclusion is true but the argument must resolve (90 ‘obsaces: () In the preceding discussion we treated C, U, VE, ete, as though they were given in Cartesian coordinates whea in fat they {@) Tho probability distribution of E(,....f, must be con- sidered in reaching the conclusion and in general wil ivaidate it "Note further that both U and C are linear approximation to ‘an in general curved surfce”. Also in the real case te dou i. Lage finite subset of poins of the possible (C-. ‘of positive probability. (The original discovery of win ‘ack systems (Thorp, 1961, was motivated by this model.) First Tinuroduced the E(n,...,) “surface”, where nis the number of cards remaining of denomination i, Ituitive arguments “con- me Arpenices vem te Esai shld he tt dvs the fll dock expocaion, The next ep wns imate bythe” EA.) "uric, andthe “linearize the problem by assuming that E(-f) = By + Ey where f=) Thus there isthe” protimation of Aiscrete rote ty 3 coniauos oe. Nonetles, We shal show Theorem 3 Ite probity dstbution of yf) i ap proximately rtaonlly symmetric abot («Di then the Fauve power of ary pont count item C ie roposional (©)=CUMCIeIUI. Th powers of wo count syns which x. Plot he cout information equally (eg one normalized bythe umber ofa yet unscen cards so docs the ote, if one caries 2 sie ace cout tor betng and sts thease equa 10 for egy ‘edo het c=) aprons piri tiThe Mathematics of Gambling APPENDIX B. ‘Suppose (Hypothesis 1) thatthe shoe really has four cnuphete decks, Then the number of unseen tenealue cards among the 14cards (wo decks) not seen will average 32. Inthe general case with LI unazon catds,T tens inthe whole pack. and N non fens in the whole pack, the average value A of Xispiven by A= UT/N + T) In our example, U= 104, T= 64 and N= 144, so we get A= 104% 64/208 ~32. But there wil be a fluctuation around this number. Mathematicians use the tunlaid deviation 5 to ‘measure this fluctuation. The formula S*=/UTN/T + N)}J0 ~ (U- WN +T— Uh ‘Foro example, S?=/(104 x 64 x 144/208*((1 ~ 103/207) 111304, 90 8 = /ULDO$ = 2.3362. To a good approxima tion, X is “normally distributed” with mean A= 32 and stan- dard deviation $= 3.3362. Now, Suppose instead (Hype I thatthe deck hasten ten~ valuecardsremoved. Then U=9, T= Séand N= 134.If Yisthe umber of unseen cards, we have the real A =25.6364, but we think there are ten more ten-value cards. So assuming incorrectly that no ten-values are gone the number tha we deduce for Y has an average of 4 + 10= 35,6364, The real S? for Vis 9 x 5¢ x B4I98 (1 ~ 931197) = 91593, so $= 20264. ‘What e watt o kis fs whether to believe Hypothesis T (null hypothesis") or Hypothesis II. This isa clas statistics, problem, Itcurnsout that inorder forusto havea good chanceto believe the corecthypothess the A value for X and Yneedto be at leas two and preferably several Sunits apart. In this example, they differ by only 35.6364 — 323.6364 whichis about one $ unit, Ofcourse, repeated countdowns ofthis same shoe wil again {crease Gur ubliy to tel wheter the shoei short. 16 ren APPENDIX €. or this nt spe dain, es supp atu expt) +6 where a,b, and e are contains and exp is he exponent function. This on ofthe simplest math fnceons hat fs the righ shape" (Ne Fedo tis scuson ws te uate 3 ‘Sie ilerae) that the ball voy atthe point wher el fom the tek na abo 05 eves pe ond ps) an hat ‘ce revolton eaert was about 2p. Us is andthe choise {Ohen the ball aes he rack ges a= 10/3, = 3/20 and 5 105-3) 3m bs ands nea angular lacy v np fv) = espv20y Figure Skoog of, = Mea Fe APPENDIX D. A calculation shows, for our illustrative (function, that xfP)=llesp(ST/20) ~ 1) ~ 10/3. Ths, from T we can predict ‘ie numberof revolutions until the ball eaves the track For in- Stance, if T= sec, we predict the ball wil lave the track in 2A) = 1esp(3/2) ~ 1) ~ 10/3=2.85 revolutions afer the ‘Shitchishithesecond time instead 7 ~ see, then we predict Xf) =9.51 revolutions,The Mathematics of Ganbling APPENDIX E. Math readers: de(TdT= - (3x41) +109/60. It can be shown that forthe x) of this example, the error A xP in the prediction of (7) due wo an error & 7 in measuring 7s given by a a) —— Ox) + 10PTEO = 3 T/20lexpAT20) 1p), For instance, if T = 08 see. and AT =002 see., we have a prediction ertor of A x,(0.8) = Ql revs or 4.2 numbers ‘on the wheel In our illustration T'= (8 sec. means x7) = £51 revolutions to go. The time to go is 20/)lo. (33,090 + Vor 570 sec. We have somewhat less time than this to bet APPENDIX F. Jn our example, the equation for 40) is 4g) guving insepteae 9's colic aainie fhe ents aprouinaay 41) = a ep 120) Heoa0) Thus apn T= 08 ec aa Ao COD ea = algae Wider spel of 039 rp Ee ane ror pcan er of Us feo pes Ineareunple tel, we messed Ttoolage O01 se Tas Indus tbo the bal wuld ae wack a pol abut 15 poste tebe sere Therefore, we fren inpet cake or'3 poe eats ed wt beiee he al ‘oud leet tack sone tine Ths, we tough th ae Wout role af a This mais ea impact ater 3 pie ear, fr im r et 3 pose ea Thee art oe impor sours feos fal PCBS: te wee act Wh god. Bat bey eve fod nap ie susan te hiner tee Ss pouty ay we think Fis ager tan eal is becase we hte ih cy the fi in Fle the scan ne leads sa he Bl as Arpences is slower than itis. That makes us think 1,7) is shorter. Thus, we expect the ball atthe rotor to soon and forecast impact on fe rotor ahead of where it tals o occut. Conversely if Tis negative (last onthe irs switch or early onthe Second), we think Tis smaller, the ball is faster, and mistakenly forecast 17) and {,(D 38 too big, Thea we predict impact behind where it tends ‘The rotor angular velocity, followed a law close 0 ft) = Aexp (00). A ypical value for A was 0.33 rex./se. The “decay” oF ‘dlowing down" constant b was vory smal. The rotor is massive and spins on a well-oiled bearing (on our casino whee, it was the pointed end of a sturdy tel shal). In the course ofa minute or lwo, the slowing was hardly perceptible, (Note: Siroboscopic “bea frequency” techniques, plus an accurate clock, can quickly and easly give avery precise measurement of b and the slowing, down) Let's take b=—Log,1011)120 or 0000704oee., which cor responds toa slowing down from 0.33 rev/sec. 0.30 rv. ec. in {wominutes. This seems like the right order of magnitude. To put the rotor position into the tiny computer we were going to build, ‘we planned to hit arotor timing switch once when te 20 passed. reference mark onthe wheel, and then hit th switch again when the zero passe the reference marka second time, Since the rotor ‘velocity war small andinearly constant, this was ales sensitive” tmewuremen. Therefore weplaedio dois, shor bore the ball was spun, How much error in the ball’s final postion (pocket) comes, trom rotor ming errs! Assume for Spicy that We rotor ‘makes one revolution in about thre seconds (33 rev./sc.) and that we can neglect the slowing down of the rotor. Then, a inthe ball timing, we might expect atypical (oot mean square) sie of| about 11.2/1,000 seconds forthe combined effet of the two er- tors. Ifthe rotor really makes one revolution in 3.000 seconds, and we think ittakes 30112 seconds, then in 30 seconds we tink the whe! wil travel U9428 revolutions whereas it really travels 10,0000 revolutions. Thus, the rotor goes 0372rev. oF 1 pockets farther than expected Sill, if we think the rotor takes 2.9888 9The Mathematics of Gambling seconds for one revolution, then in 30 seconds the rotor goes 10375 1ev. or 1s pockets ess than we expected. APPENDIX G. 1 am using the normal approximation forthe statistical discus sion. I think itis very nearly an accurate description of what happens and that this approximation only slightly affects the discussion APPENDIX H. In genera, here are exactly (5+1)!/Se! home board positions with exactly e men, There are exactly (6+1)!/6!e-1 home board positions with from one tor men. Thus since r= is posible Inte coal yun, theve ave a Wal of 216! 18! 1m54,268 di ferent home board positions for one player. The symbol! read "factorial means LX2X3X. ..xf. Thus HI, 2!=2, 3-6, 41 Scholarly References Fr those evade who ate especially interested inthe technical \work behind the material inthis book and other work by Pro- fessor Thorp, here is a list of some of his related scholarly publications. Game Theory 1. “A Favorable Strategy for Twenty-One," Proceedings ofthe National Academy of Sciences 47 (961). 0-2. (MR #B&46). 2. BOOK: Beat the Dealer: A Winning Strategy forthe Game of Twenry-One, Random House, New York, 1962; Vintage paper- back also, 1966; revised edition, 1966. 3. “A Partial Analysis Of Go,” (with W. E. Walden), The (British) Computer Journal 73 (968), 203-207. (MR33 #2424), @ 123, 340). 4. “A Favorable Side Bet in Nevada Baccara,” (with W. B. Wilden), Journal of the American Statistical Assocation 6, Past 1 9966), 38-308 5. “Repeated Indepeadent Trials anda Class of Dice Problems,” (Mathematical Now), American Mathematical Monthly (August September 1968), 778°781. Z 123, 364). 6. “Optimal Gambling Systems for Favorable Games," Review of Intemational Institue of Sasics 313 (969), 7.293. (2 18, 4m. uw7, “Solution of a Poker Variant" Information Sciences 2/2 (0970), 299-301. (MR S42 #8952), (Z 205, 232). ‘8. “A Compute Assisted Study of Go on M x N Boards” (with W. E. Walden), Locture Notes in Operations Research and Mathematical Systems in Theorecal Approuches wo Non- ‘Numerical Problem Solving, Vol. 28, R. Banerji and M. D. Mesarovc, eds, SpringerNeriag, New York, 197, 303-343 ater revised version, Information Sciences 4:1 0972), 133 (MR&S 8684), (Z 22868028). 9, "Non Random Shuffing With Applications tothe Game of Faro” Journal ofthe American Statistical Association, 842-87, December I973, Much expanded version appears in Gambling ‘and Society, edited by W. Eadington, Charles C. ‘Thomas, Springfield, Llinois, 1975, as: Probabilities and Strategies forthe Game of Faro, pp. $31-560. 10, “The Fundamental Theorem of Card Counting With Ap- plications to Trente et Quarante and Baccarat." (with W. E- Walden) Intemational Journal of Game Theory 2 (73), 08-19. (@ 25890058) UU, “Backgammon: Part 1, The Optimal Strategy for the Pure Running Game." Proceedings of the Second Annual Conference ‘on Gambling, Lake Tahoe, ITS. mis. 42 + pp. See uews report, fon this paper in: “Beating the Game,” an article by Dietrick E. ‘Thomsen, Science News, Vol. 107, March, 1975. 12, "Blackjack Syetems:” Proceedings of the Second Anal CConferrence on Gambling, Lake Tahoe, IS. ms. I + pp. See news report on tis paper in: “Beating the Game,” an article by Dietrick E. Thomsen, Science News, Vol. 107, March, 1975. Probability, Statistics 1. BOOK: Elementary Probably, Wiley, New York, 1966. Mathematical Finance 14, BOOK: Beat he Marker, (with S. Kassou, Random House, New York, 1967. a 1S, “Portfolio Choice and the Kelly Criterion” Proceedings of the 1971 Business and Economics Section of the American ‘Staistial Association 1972, 215-224. To be repinted in Invest ‘ment Decision-Making, edited by J. Bicksler. Reprinted in Stochastic Opdinisation Models is Finance, Acadetnic Press, edited by W. T. Ziemba, §. L. Brumelle, and R. G. Vickson, 191, pp. 599-620. 16. “The Capital Growth Model: An Empirical Investigation: (with James Bicksler), Journal of Fnancial and Quantiatve Analysis, March 1973, Nol. VI, No. 2, pp. 273287. General Interest 17. “A Professor Beats the Gamblers", The Atlantic Monthly, June, 1962. Reprinted in “The Gamblers Bedside Book,” John K. Hutchens, ed., Tapliniger, New York, 1977, pp. 166-177.