Einstellungsfrage des Top Hedge Funds Citadel
Viel Spaß!
Viel Spaß!
Einstellungsfrage des Top Hedge Funds Citadel
|
Einstellungsfrage des Top Hedge Funds Citadel
Viel Spaß!
Also nehmen wir mal an, das Spiel wäre mit 2 Boxen und ich wäre "Verkäufer", also derjenige, der die 100 Pfund in die Box steckt. Wieviel müsste ich verlangen, um nicht zu verlieren?
Die Hälfte der Spieler schafft es nach einem Versuch, die andere erst nach zweien. Somit müsste ich bei zwei Spielen 200 / 3 Versuche im Schnitt = 66,66 Pfund verlangen. Bei vier Boxen wären es dann: ein Spieler schafft es bei einem, einer bei zwei, einer bei drei und einer bei vier Versuchen. Bei vier Spielen dann 400 / 10 = 40 Pfund pro Versuch.
Aus Spielersicht dann umgekehrt: ich brauche im Schnitt genau 2,5 Versuche um zu gewinnen. Somit 100 / 2,5 = 40 Pfund pro Versuch.
* in einem normalen Spiel der niedrigstmögliche Einsatz z.B. 1 Pfund pro Spiel, damit der maximale Gewinn rauskommt.
* bei einer Gewinnerwartung von Null würde ich 24 Pfund pro Runde einsetzen Gewinnwahrscheinlichkeit steigt bei jeder Runde, von anfangs 25% auf 33,3% und dann 50%: 0,25*0,33*0,5*1=0,041625. Der maximale Gewinn 100 Pfund/4,1625 macht Einsatz von 24 Pfund. aber der Haken an der Sache: es ist die Rede vom "Fair Game", also der Spieler soll keinen Gewinn machen. Die Gewinne die erspielt wird soll mit den Verlusten ausgeglichen werden. Finde den Gedanken nicht sehr sportlich (alle Sportarten sind zero-sum games), dieser spieltheoretische Ansatz hat so was politisch-diplomatisches an sich. Würde es als ein non-zero sum game beschreiben. * dann trifft die Rechnung von speculatius zu, also 40 Pfund pro Runde. __________________
whatever it takes
Die Frage geht aber noch weiter:
Zitat:Instead of maintaining a constant price, if you choose to continue playing after two incorrect guesses the price changes to ZZ. What is the fair value of X and ZZ? ![]() RE: Einstellungsfrage des Top Hedge Funds Citadel| 16.05.2025, 15:56 (Dieser Beitrag wurde zuletzt bearbeitet: 16.05.2025, 16:05 von boersenkater.)Zitat:Box Game ChatGPT - Bitte erklären.... Zitat:Natürlich! Lassen Sie uns die Konzepte hinter dieser Frage Schritt für Schritt durchgehen. Mich würde interessieren warum ich so ein Spiel spielen soll - wenn am Ende weder "die Bank" noch der Spieler am Ende gewinnen soll.... ![]() Zitat:Das ist eine sehr gute Frage! Warum sollte jemand ein Spiel spielen, bei dem der erwartete Gewinn **null** ist? Wenn das Spiel „fair“ ist, bedeutet das, dass der **Spieler** und **die Bank** (also der Betreiber des Spiels) langfristig gesehen keinen Vorteil oder Nachteil haben – sie machen keinen Gewinn und keinen Verlust. Edit: Für die 2. Frage habe ich die 1. Frage nochmal gestellt (weil schon wieder gelöscht). Da war die Antwort dann 33,33 Pfennig.... ![]() ![]() ![]() __________________ ![]() RE: Einstellungsfrage des Top Hedge Funds Citadel| 16.05.2025, 16:06 (Dieser Beitrag wurde zuletzt bearbeitet: 16.05.2025, 16:22 von Thomas_B.)Zitat:boersenkater Einfach so was gedankenlos zu pasten ist stupide. Erklär wenigstens, warum der Ansatz falsch ist.
Das ist nicht der HF. Sondern der hft/mm
__________________
Forum-Besserwisser und Wissenschafts-Faschist
![]() RE: Einstellungsfrage des Top Hedge Funds Citadel| 16.05.2025, 16:20 (Dieser Beitrag wurde zuletzt bearbeitet: 16.05.2025, 16:21 von Thomas_B.)
Ich habe mal die gesichert rechtsradikale GROK 3 AI befragt, Ergebnis und Rechenweg soweit korrekt:
### Key Points - It seems likely that the fair value of X is 40 pounds, based on expected value calculations. - The game involves paying X to open boxes until finding 100 pounds, with the expected cost balancing the gain. - Interpretations may vary, but this approach aligns with standard probability models. ### Direct Answer The fair value of X, where a player pays X to open a box and can continue until finding 100 pounds in one of four sealed boxes (with the others empty), appears to be **40 pounds**. This is based on calculating the expected number of boxes opened, which is 2.5, and setting the expected cost (2.5 times X) equal to the 100-pound gain for a fair game. **How We Got Here** The game is considered fair if the expected net gain is zero, meaning the cost of playing equals the expected winnings. Since the player can keep opening boxes by paying X each time until finding the money, we calculated the average number of boxes opened (2.5) and solved for X, leading to 40 pounds. This assumes the player continues until finding the money, which seems reasonable given the problem's structure. **Why It Matters** This value ensures neither the player nor the game organizer has an expected advantage, aligning with the "fair game" condition. However, interpretations of "as many times as they like" could vary, so this is our best estimate based on standard probability reasoning. --- ### Survey Note: Detailed Analysis of the Box-Opening Game This section provides a comprehensive exploration of the problem, where a player engages in a game involving four sealed boxes, one containing 100 pounds and the others empty. The player can pay X to open a box and take its contents, repeating this process as many times as they like, with the game assumed to be fair. We aim to determine the value of X that makes the game fair, meaning the expected net gain is zero. Below, we detail the reasoning, explore interpretations, and present calculations to arrive at our conclusion. #### Problem Interpretation and Initial Considerations The problem states there are four sealed boxes, one with 100 pounds and three empty, and the player can pay X to open a box and take its contents "as many times as they like." The game is fair, implying the expected value of playing should be zero, balancing costs and gains. The phrase "as many times as they like" suggests the player can repeatedly pay X to open boxes, likely continuing until finding the money or deciding to stop, but given the context, we interpret it as paying X per box opened until finding the 100 pounds, as this aligns with standard expected value problems. Initially, we considered two main interpretations: 1. The player pays X once and opens one box, taking its contents. 2. The player pays X each time to open a box, continuing until finding the money or choosing to stop. For the first interpretation, if it's a one-time payment X to open one box, the expected value is (1/4)*100 + (3/4)*0 - X = 25 - X, and for fairness, 25 - X = 0, so X = 25 pounds. However, "as many times as they like" suggests multiple openings, leading us to favor the second interpretation, where the player pays X per box until finding the money. #### Modeling the Game as Sequential Box Openings Given the player can pay X each time to open a box, we model this as a sequential process where the player continues until finding the box with 100 pounds. This is akin to a stopping problem, where the number of boxes opened, N, is the number of trials until success, with success being finding the money. Since the boxes are opened without replacement and the player stops upon finding the money, N follows a distribution where: - There are four boxes, one with money, and the player picks randomly each time. - The probability of finding the money on the k-th box opened depends on the previous choices being empty. We calculated the probability distribution of N, the number of boxes opened until finding the money: - P(N=1) = 1/4 (first box has money). - For N=2, the first is empty (3/4) and second has money given first was empty, which is 1/3 (since three boxes left, one with money), so P(N=2) = (3/4)*(1/3) = 1/4. - For N=3, first two empty and third has money: (3/4)*(2/3)*(1/2) = 1/4 (as calculated, (3/4)*(2/3)=1/2, then 1/2*1/2=1/4). - For N=4, first three empty and fourth has money, which is (3/4)*(2/3)*(1/2)=1/4, confirming P(N=4)=1/4. Interestingly, this suggests P(N=k)=1/4 for k=1 to 4, which we verified by recognizing it's equivalent to the position of the money box in a random permutation being equally likely in any of the four positions. Thus: - P(N=1) = 1/4 - P(N=2) = 1/4 - P(N=3) = 1/4 - P(N=4) = 1/4 The expected value E[N] = sum_{k=1 to 4} k * P(N=k) = (1*1/4) + (2*1/4) + (3*1/4) + (4*1/4) = (1+2+3+4)/4 = 10/4 = 2.5. #### Calculating the Fair Value of X If the player pays X for each box opened and continues until finding the money, the total cost is N*X, and they gain 100 pounds when they find it. The expected net gain is: - Expected gain = 100 pounds (since they always find it eventually). - Expected cost = E[N]*X = 2.5*X. For the game to be fair, expected net gain = 0, so: - 100 - 2.5*X = 0 - 2.5*X = 100 - X = 100 / 2.5 = 40 pounds. To verify, we can compute the expected net gain for X=40: - For N=1: Pay 40, gain 100, net = 60, probability 1/4, contribution = 60*(1/4)=15. - For N=2: Pay 80, gain 100, net = 20, probability 1/4, contribution = 20*(1/4)=5. - For N=3: Pay 120, gain 100, net = -20, probability 1/4, contribution = -20*(1/4)=-5. - For N=4: Pay 160, gain 100, net = -60, probability 1/4, contribution = -60*(1/4)=-15. - Total expected net gain = 15 + 5 - 5 - 15 = 0, which confirms fairness. #### Exploring Alternative Interpretations We considered if "as many times as they like" meant paying X once to play and then opening boxes without additional cost, but this would make X=100 for fairness, which seems trivial. Another interpretation is the player pays X per box but can stop early, leading to a dynamic programming approach where V(n) is the expected value with n boxes left. However, given the problem's structure and the fair game condition, continuing until finding the money is the most reasonable strategy, leading back to X=40. To cross-check, we tested with two boxes: E[N]=1.5, 100 - 1.5*X=0, X=66.67, and calculations balanced, supporting our method. For four boxes, X=40 fits similarly. #### Conclusion Based on the detailed calculations, the fair value of X is 40 pounds, ensuring the expected net gain is zero when the player pays X per box opened until finding the 100 pounds. This interpretation aligns with standard probability and expected value concepts for such games. | **Number of Boxes (n)** | **Probability P(N=n)** | **Cost (n*X)** | **Net Gain (100 - n*X, for X=40)** | |-------------------------|------------------------|----------------|-------------------------------| | 1 | 1/4 | 40 | 60 | | 2 | 1/4 | 80 | 20 | | 3 | 1/4 | 120 | -20 | | 4 | 1/4 | 160 | -60 | | **Expected Net Gain** | - | - | 0 (calculated as shown above) | This table summarizes the contributions, confirming X=40 balances the game. ### Key Citations - [Three boxes with a prize in one box discussion on Reddit](https://www.reddit.com/r/mathematics/com..._and_your/) - [Probability problem with boxes and prizes on Vaia](https://www.vaia.com/en-us/textbooks/mat...f-a-certa/) - [Bertrand’s Box paradox explanation](https://whyevolutionistrue.com/2018/02/2...er-is-2-3/) - [Probability boxes problem on Mathematics Stack Exchange](https://math.stackexchange.com/questions...lity-boxes) - [Monty Hall problem explanation on Ust](https://sciencefocus.ust.hk/montyhall) (16.05.2025, 16:06)Thomas_B schrieb: Einfach so was gedankenlos zu pasten ist stupide. Erklär wenigstens, warum der Ansatz falsch ist. Welche? Wenn sich das gleichmässig verteilt (das ist ja die Theorie) scheint mir 25 Pfund richtig zu sein. 400 Spiele - 10.000 Pfund Einsatz 100 Spiele - 10.000 Pfund Gewinnausschüttung Hab den Rest vorher nicht angeschaut - nur das hervorstechende Ergebnis 33,33 Pfennig - warum das falsch ist? Es geht um Pfund - dementsprechend wären das Pence und nicht Pfennig... ![]() ![]() ![]() __________________ |
|