Sequential Decision Expected Value

This question presents a simple sequential decision problem where, at each stage, you must pick between two probabilistic payoffs. The setup strips away complications like learning, state transitions, or path dependence, so the focus is purely on how to evaluate choices over time when each step offers similar trade-offs. It is the kind of scenario that often appears early in interviews for roles that involve probabilistic reasoning, risk assessment, or basic decision theory, as it checks whether candidates understand how to think about repeated choices without being distracted by the "sequential" label.

The problem leans heavily on expected value calculations, the law of total expectation, and recognizing independence across stages. An interviewer is looking for a candidate who can translate verbal descriptions of uncertain outcomes into formal expectations and who sees that the optimal strategy is defined by a simple comparison rule applied consistently at each step. They will pay attention to whether the candidate conflates sequence with path dependence, introduces unjustified "gambler's fallacy" style reasoning, or correctly articulates why no alternative strategy can beat the expected value maximizing rule under the given assumptions.

What it tests

The core structure of sequential decision problems is governed by the principle of maximizing expected value at each decision point. This means that, regardless of the sequence or complexity, the optimal strategy always involves evaluating all available options by calculating their expected payoffs—multiplying each possible outcome by its probability—and then choosing the option with the highest expected value. This approach is rooted in the law of total expectation, which ensures that, over time, consistently choosing the action with the highest expected value yields the best average result. The principle holds because, in the absence of additional information or future learning, no alternative strategy can systematically outperform this method. The pattern is universal for problems where outcomes are probabilistic and choices are independent at each stage.