Optimal Stopping with Square-Number Ruin

Quick Overview

You repeatedly roll a fair six-sided die and add each roll to a running total. Explain the assumptions and derivation clearly, check edge cases, and show how the result changes when those assumptions no longer hold.

Optimal Stopping with Square-Number Ruin

Company: Jane Street

Role: Software Engineer

Category: Statistics & Math

Difficulty: hard

Interview Round: Technical Screen

# Optimal Stopping with Square-Number Ruin You repeatedly roll a fair six-sided die and add each roll to a running total. Before each roll, you may stop and receive the current total. If a roll makes the total a positive perfect square, the game ends immediately and you receive zero. Determine an optimal stopping rule and explain how to compute the value of the game. ### Constraints & Assumptions - The running total starts at zero, and stopping at zero is allowed. - Perfect squares are 1, 4, 9, 16, and so on. - A roll that lands on a square causes ruin before another stop decision. - The answer may present a dynamic-programming computation and a threshold or state-dependent policy rather than one closed-form number. ### Clarifying Questions to Ask - Is there a maximum number of rolls or a maximum total? - Does landing above a square have any effect, or only landing exactly on one? - Must a practical solution provide a numerical truncation error bound? ```hint Test square boundaries Compare decisions at totals immediately before and immediately after a perfect square, and distinguish landing on a square from jumping past it. ``` ```hint Check decision timing Stopping at zero is permitted, while a ruin roll ends the game before another decision can be made. ``` ### What a Strong Answer Covers - A Bellman equation that compares stopping now with the expected value of rolling. - Correct zero value for transitions landing on perfect squares. - A finite computation or convergence argument for the unbounded state space. - A policy described by the states in which stopping dominates rolling. ### Follow-up Questions - How would the policy change for a die with a different set of face values? - Can you bound the error introduced by solving only up to a large terminal total?

Quick Answer: You repeatedly roll a fair six-sided die and add each roll to a running total. Explain the assumptions and derivation clearly, check edge cases, and show how the result changes when those assumptions no longer hold.

|Home/Statistics & Math/Jane Street
Jane Street logo
Jane Street
Aug 3, 2026, 12:00 AM
hardSoftware EngineerTechnical ScreenStatistics & Math
0
0

Optimal Stopping with Square-Number Ruin

You repeatedly roll a fair six-sided die and add each roll to a running total. Before each roll, you may stop and receive the current total. If a roll makes the total a positive perfect square, the game ends immediately and you receive zero. Determine an optimal stopping rule and explain how to compute the value of the game.

Constraints & Assumptions

  • The running total starts at zero, and stopping at zero is allowed.
  • Perfect squares are 1, 4, 9, 16, and so on.
  • A roll that lands on a square causes ruin before another stop decision.
  • The answer may present a dynamic-programming computation and a threshold or state-dependent policy rather than one closed-form number.

Clarifying Questions to Ask Guidance

  • Is there a maximum number of rolls or a maximum total?
  • Does landing above a square have any effect, or only landing exactly on one?
  • Must a practical solution provide a numerical truncation error bound?

What a Strong Answer Covers Guidance

  • A Bellman equation that compares stopping now with the expected value of rolling.
  • Correct zero value for transitions landing on perfect squares.
  • A finite computation or convergence argument for the unbounded state space.
  • A policy described by the states in which stopping dominates rolling.

Follow-up Questions Guidance

  • How would the policy change for a die with a different set of face values?
  • Can you bound the error introduced by solving only up to a large terminal total?
Loading comments...