A Bold Strategy Is Not Always Optimal in the Presence of Inflation Author(s): Robert W. Chen, Larry A. Shepp, Alan Zame Source: Journal of Applied Probability, Vol. 41, No. 2 (Jun., 2004), pp. 587-592 Published by: Applied Probability Trust Stable URL: http://www.jstor.org/stable/3216039 Accessed: 23/11/2010 14:56 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=apt. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Applied Probability Trust is collaborating with JSTOR to digitize, preserve and extend access to Journal of Applied Probability. http://www.jstor.org J. Appl. Prob. 41, 587-592 (2004) Printed in Israel © AppliedProbabilityTrust2004 A BOLD STRATEGY IS NOT ALWAYS OPTIMAL IN THE PRESENCE OF INFLATION ROBERTW. CHEN,*** Universityof Miami LARRYA. SHEPP,***Rutgers University ALAN ZAME,* Universityof Miami Abstract A gambler,with an initial fortune less than 1, wants to buy a house which sells today for 1. Due to inflation, the price of the house tomorrowwill be 1 + ca, where ca is a nonnegativeconstant,and will continueto go up at this rate, becoming (1 + a)n' on the nth day. Once each day, he can stake any amountof fortunein his possession, but no morethanhe possesses, on a primitivecasino. It is well knownthat,in a subfairprimitive casino withoutthe presenceof inflation,the gamblershouldplay boldly. The presenceof inflationwould motivatethe gamblerto recognize the time value of his fortuneand to try to reachhis goal as quickly as possible; intuitively,we would conjecturethatthe gambler should again play boldly. However,in this note we will show that, unexpectedly,bold play is not necessarilyoptimal. Keywords:Gamblingproblem;primitivecasino; optimalstrategy;bold strategy 2000 MathematicsSubjectClassification:Primary60G40 Secondary60K99 1. Introduction and formulation of the problem Suppose that you have x dollars and you want to buy a house which sells today for y dollars, > x > 0. Due to inflation, the price of the house tomorrow will be (1 + a)y (where a > 0), y and will continue to go up at this rate each day, so as to become (1 +a)ny on the nth day. Once each day, you can stake any amount of money in your possession, but no more than you possess, on a 'primitive casino', i.e. one in which only one type of bet is available. If you make a bet, you gain r times your stake with probability w and lose your stake with the complementary probability w = 1 - w, where r is a positive constant. How much should you stake each day so as to maximize your chance of eventually catching up with inflation and being able to buy the house? When w < 1/(1 + r), the primitive casino is subfair and, when w = 1/(1 + r), the primitive casino is fair. As is well known, in a subfair or fair primitive casino without the presence of inflation, the gambler should play boldly since there is no other strategy that provides him with a higher probability of reaching his goal (being able to buy the house). In [3], Chen proved that the gambler should play boldly when r = 1 (when r = 1, a primitive casino is called 'red-and-black'). The presence of inflation would intuitively motivate the gambler to recognize the time value of his fortune and to try to reach his goal as quickly as possible. We would Received 20 June 2003; revision received 21 October2003. * Postal address:Departmentof Mathematics,Universityof Miami, CoralGables, FL 33124-4250, USA. ** Email address:[email protected] ***Postal address: Departmentof Statistics, Rutgers University, 110 FrelinghuysenRoad, Piscataway,NJ 08854, USA. 587 R. W.CHENETAL. 588 suspect that the gamblershould again play boldly. However,in this note we will show that, surprisingly,bold play is not necessarily optimal. This result is a counterexampleto a 1909 assertionby Coolidge. Coolidge [5] stated: Theplayer'sbestchanceof winninga certainsumata disadvantageous gameis to stakethe sumthatwill bringhimthatreturnin one play,or,if thatbe notallowed,to makealways thelargeststakewhichthebankerwill accept. Different counterexamplesto Coolidge's statementhave also been provided in [2], [3], [8], and [9]. To makethis new gamblingproblemfit moreclearlyandeasily into the gamblingframework of DubinsandSavage [6], we can assumethatthe priceof the house is fixed at 1 butthe value of money will be discountedstep by step by the fixed discountrate ( 1 + a)-1, i.e. the currentone dollarwill be worthonly (1 + a)-1 dollarsat the next step. Therefore,our gamblingproblem can be formallyformulatedas a game whose set of fortunes,utilityfunction,andset of available gambles are respectivelyas follows: F =[0, oo), u(f) u(f) = O I if 0 <l, iff >, 1,< iff 'y(f,s) ( -s 0<s<f, y(f,s)=ws +ww(f ( +rs) 1+ta < f if r(f) = < f < 1, + (Af) , y(f,s) =Wy(f+) w( f rs) + ( f-s if l < f <o. Here w = 1 - w, 3(x) denotes the probabilitymeasurewhich assigns probability1 to {x) for x E [0, oo), and a is the inflationrate (a device often encounteredin dynamicprogramming models) for the gambler.The reasonthat3(f) is in F(f) for f > 1 is that, when the gambler has a fortune f > 1, he has reached his goal already and can buy the house immediately. Plainly,if a > r, a gamblerwith initial fortunef < 1 can neverreachhis goal, so we assume thata < r. The gamblingproblemformulatedaboveis a modificationof the primitivecasino considered by Dubinsand Savage [6]. The modificationis designedto handleinflationandto motivatethe gamblerto recognizethe time valueof money andcompletethe game as quicklyas is consistent with reachingthe goal. To distinguishit from the 'primitivecasino', this modified game will be called the 'primitivecasino in the presenceof inflation'. When a = 0, the modifiedgame is identicalto the 'primitivecasino' consideredby Dubins and Savage in [6]. 2. The utility of modified bold strategies We say that the gamblerstakes s when he uses the game y(f, s) at f. The modifiedbold stake at f is definedby s(f) = min 0 +, rI f, if0< i if/ f f < 1 1, 589 A boldStrategyis natalwaysoptimal where r is a positive constant. A gambler uses the modified bold strategyif he stakes the modifiedbold stake s(f) wheneverhe has fortunef. Let Q(f) be the probabilitythatthe gambler,startingfromf E (0, 1) andusing the modified bold strategy,reaches his goal, [1, oo). It is obvious that Q(0) = 0, Q(f) = 1 if f > 1, and, if 0 < f < 1, then, consideringone game, 'Q(f) =wQ( f + rs(f) 1+ + wfQ( f-s(f) 1±l+a Therefore, Q(f) = 0 iff <0, wQ(f) if < w+WQ( if <f<1, < B, (1) ) if f>1, 1 where/ = (1 + a)/(1 + r) < 1 as noted in Section 1. Lemma 1. The boundedsolution of () is unique. Proof. The proof is essentially the same as thatin [6, p. 99] and is omitted. Let {Xn}n>i be a sequence of i.i.d. randomvariablessuch that P(X1 = 0) = w = 1 P(X1 = 1). For each integer n > 1 and each integer k E [l,n), let Sn,n = Xn and Sn,k = /BXk+ fi[1 + (r - l)Xk] min{1, Sn,k+l-} Since Sn,1 is seen to be nondecreasing in n, the function G(x) = limn,, P{Sn,1< x} is well definedfor all x. - Lemma 2. For all x, G(x) < 1 and thefunction G satisfies (1). Proof. The proof is similarto thatin [3, pp. 295-296] and is omitted. In view of Lemmas 1 and 2, we have the following theorem. Theorem 1. Thereis one and only one boundedfunctionQ on the interval[0, oo) that satisfies (1). Moreover,thefunctionQ is rightcontinuouson the interval[0, oo) and is strictlyincreasing on the interval [0, 1]. 3. The nonoptimality of modified bold strategies For each integerk > 0, let Ak be the set of all fortunesf which satisfy the following three conditions: (i) There exist k + 1 positive integers 1 < no < n < .. < nk such that k f = E rj,ni. /=0 j=o (ii) We have 0 < f < /no-1. (iii) For each i E {O, 1, 2,..., k - 2}, k-i-I E j=O rjpni+l+j-ni < pni+l-ni-1 R.W.CHENETAL. 590 Theorem 2. If f = Ij=o rJfinj is in Ak, then k Q(f)= E Wiwnrij=0 Proof. This can be provedby recursiveapplicationof (1) and mathematicalinduction. Let ln(w) _ln(w) where [-I is the interger-partfunction,and let a= if 1K > , if K = 0. maxO, (1 + r)r-(l+l/K) 0 Lemma 3. Supposethat r > 1 and 0 < w < 1. If the inflationrate a is in the interval [a, r), then the modifiedbold strategyis not optimal. Proof We shall show that, if the initial fortuneis f as defined in (2) below and the initial stake is s as definedin (3) below, then WQ ::(:) I+aa > Q() . l+aa Since s < s(f), where s(f) is the modifiedbold stake at f, Theorem2.14.1 of [6, pp. 32-33] shows thatthe modifiedbold strategyis not optimal. Since f < 1, we can choose a positive integer y such that 0 < r6Y < 1 and (1 + a)fY < B< 1 - rfY. Let f be defined by f = -2{1 + rpY + r2 2Y + ... + rKPKy + rK+l,P(K+l)y + rK+21(K+2)y} (2) and s be definedby s = f- (1 + a)fiBY . . +rK-2f(K-l)Y +. +r,82y +r2p3y + rK-1IKY + rKi(K+l)Y}. It can be verifiedthat + rs <f 0<S<f< = l+a + K+2(K+2)y+l < and that f -s 1+a = {f Since f < f, s(f) + r 2Y + r2p3y = f > s. Since f Q(f) = w2{1 + ww+ ... +... + + rK-2p(K-I)Y e AK+2 and (f- rK-IpKY + rK(K+l)Y}. s)/(1 + a) E AK, +W2W-2(-1) + WKWK(y-1) + WK+lw(K+l)(y-1) + 1K+2w(K+2)(y-1)} (3) 591 A bold Strategyis not always optimal and Q( f-) = 2{ WY-1 + + -.. 22(- K+ K() + (K+l)(y_l)}. Since (f + rs)/(1 + a) = ,B + rK+2fl(K+2)Y+l < 1, wQ( +r) = w2+ ww Q(rK+lfl(K+2)Y). Therefore, Q( f +±rs) Q - Q(f) - = Wt{Q(rK+lfi(K+2)Y) -K+lw(K+2)(y-1)+I}. We now proceedto show that wQ (+ ) + wQ + )- Q(f) > O, which will complete the proof of Lemma 3. To do this, it is sufficientto show that Q(rK+IP(K+2)y) _ WK+lw(K+2)(y-l)+l > . If K > 1, then rK+lpK > 1 since a > (1 + r)r-(l+l/K) - 1. Therefore, Q(rK+I(K+2)y) _ WK+lw(K+2)(y-l)+l > Q(p(K+2)y-K) _ WK+lw(K+2)(y-l)+l = w(K+2)y-K _ WK+lw(K+2)(y-1)+ = W(K+2)(y-1)+l(W - WK+1) > o since 0 < w < 1 and w > wK+1. If K = O, then rf2Y > f2y and Q(rf2y) > Q(f2y) 1 > since w < w. -w since r > 1. Therefore, Q(rf2y) - ww2y-1 > w = w2y Theorem 3. For each r > 1, the modifiedbold strategy is not optimal in a primitive casino (subfair,fair, or superfair)if the inflationrate a is in the interval [a, r). Theorem 4. For each r > 1, if the inflationrate a is in the interval[1/r, r), then the modified boldstrategyis notoptimalforanysubfairorfairprimitivecasino, thatis,for w (0, 1/(1 +r)]. Proof Since r > 1 and w < 1/(1 + r), we have K > 1 and a < 1/r. Hence, ifa > 1/r, a is certainlygreaterthana. By Lemma 3, the modifiedbold strategyis not optimal. Remark 1. For subfairand fairprimitivecasinos, Theorem3 is surprisingand seems counterintuitive: when the inflationratea is 0, the bold strategyis optimal, so a positive inflationrate would seem to imply that it is to the gambler'sdisadvantageto prolong the game. However, Theorem3 reveals thatit may not be optimalto play boldly. Remark 2. This papershows thatthe modifiedbold strategyis (usually) not optimalfor these cases of 'gamble if you must', and this raises the questionof what is the optimal stategy when the modifiedbold strategyis not the optimalsolution. The authorshave calculatedapproximate numericalvalues for the trueoptimal-valuefunctionandit appearsto be so complicatedthatan exact solution seems unlikely even in the interestingcase when a = w = -and r = 2, when, 592 R. W. CHEN ETAL. if the gambleralways stakes his entire fortune,the sequence of his renormalizedfortunesis a martingale(note this is not the modified bold strategybecause the gamblerwill sometimes overshootlevel 1 if he follows this greedy strategy).The authorsalso looked at the same case when w = 0.3 and again it appearsthat the optimal strategyis too complex to guess. More continuousproblemsof optimal stochastic control often have cleaner and simpler solutions; the difficultyof the problemstudiedin this paperis that the primitivecasino being studiedis 'too discrete'. Conjecture 1. In [3], Chenproved that, in a subfairorfair 'red-and-black',i.e. when r = 1, the modifiedbold strategyis optimal. Theauthors'experimentalcomputationsseem to indicate that, in a subfairorfair primitivecasino, the modifiedbold strategyis optimal if O < r < 1. Conjecture 2. The authors'experimentalcomputationsalso seem to indicate that Theorem3 is sharp, i.e. the modifiedbold strategyis optimalif the inflationrate a is in the interval[0, a). Acknowledgement We would like to sincerely thankthe referee for his invaluablesuggestions and comments which improvedthe presentationof this papersignificantly. References [1] BERRY,D. A., HEATH,D. C. AND SUDDERTH,W. D. (1974). Red-and-blackwith unknownwin probability.Ann. Statist.2, 602-608. 39, 167-174. [2] CHEN, R. (1977). Subfairprimitivecasino with a discountfactor.Z Wahrscheinlichkeitsth. 42, 293-301. [3] CHEN, R. (1978). Subfair'red-and-black'in the presenceof inflation.Z. Wahrscheinlichkeitsth. 49, 257-266. [4] CHEN, R. AND ZAME, A. (1979). On discountedsubfairprimitivecasino. Z. Wahrscheinlichkeitsth. [5] COOLIDGE, J. L. (1909). The gambler'sruin.Ann. Math.(2) 10, 181-192. for StochasticProcesses. [6] DUBINS, L. E. AND SAVAGE, L. J. (1965). HowToGambleIf YouMust,Inequalities McGraw-Hill,New York. [7] DUBINS, L. E. AND TEICHER, H. (1967). Optimalstoppingwhen the futureis discounted.Ann. Math.Statist.38, 601-605. [8] HEATH, D. C., PRUITT, W. E. AND SUDDERTH, W. D. (1972). Subfairred-and-blackwith a limit. Proc.Amer.Math. Soc. 35, 555-560. [9] KLUGMAN, S. (1977). Discountedand rapidsubfairred-and-black.Ann. Statist.5, 734-745.
© Copyright 2026 Paperzz