Solving Sequential Games with Backward Induction

This direction is split into other directions: backward induction. Its last version stays readable and citable here.

Version 1 of 1110 papers113 conceptsfrozen 2026-10-04

In strategic environments where decisions happen in a sequence, the best move today depends entirely on how others will react tomorrow. Whether a manufacturer is deciding on a "green" technology investment, a mobile operator is leasing spectrum to a rival, or a developer is choosing whether to open-source an AI model, the underlying logic remains the same: to find the optimal starting move, you must first solve the final move.

This research direction focuses on modeling these multi-stage interactions as sequential games and solving them using backward induction to identify subgame perfect equilibria.

The Logic of Anticipation

The core intuition of this field is the "leader-follower" or Stackelberg dynamic. In a simple two-stage game, a "leader" makes a move, and a "follower" responds. To maximize their own utility, the leader cannot simply pick the best immediate option; they must anticipate the follower's rational response to every possible choice.

As Qunli Wu and Xinxin Xu demonstrate in Online promotion format selection for a supply chain, this involves a mathematical "look-ahead." By solving the follower's optimization problem first, the leader can substitute that response back into their own objective function. This ensures that the resulting strategy is not just a hopeful guess, but a mathematically grounded prediction of a stable market state.

The Mechanism: Working Backward from the End

Backward induction is the recursive engine that powers these models. It treats a complex, multi-period problem as a series of nested decisions. By starting at the final deadline or the last stage of a game—where the future is empty and the decision is a simple optimization—researchers can "collapse" the game tree step-by-step toward the present.

This approach is used across diverse domains:

Stability through Subgame Perfection

The goal of these models is to find a Subgame Perfect Equilibrium (SPE). A strategy is subgame perfect if it represents a Nash Equilibrium in every "subgame" (every possible branch) of the larger game. This means that even if the players find themselves in an unexpected situation, their planned response remains the most rational one available.

In Competition of Wireless Providers for Atomic Users, Jianwei Huang and Vojislav Gajić prove that such equilibria can lead to socially optimal outcomes even when participants are entirely selfish. This stability is crucial for decentralized systems, like 6G networks or electricity markets, where no central authority dictates behavior.

Counter-Intuitive Insights

One of the most compelling aspects of this research is how it reveals "paradoxes" where common sense fails:

  1. The First-Mover Disadvantage: While we often think being first is an advantage, Dongwei Zhao and Hao Wang find in Storage or No Storage that in energy markets, the first person to invest in infrastructure can sometimes be "punished" by the strategic bidding of followers.
  2. The Value of Cooperation (Coopetition): Mohammad Hassan Lotfi and Saswati Sarkar demonstrate in The Interplay of Competition and Cooperation that companies often maximize profit by funding their own competitors (e.g., leasing network capacity), provided the sequential contract is structured correctly.
  3. The Power of Deadlines: In Web and Internet Economics, Tie-Yan Liu and Yinyu Ye show that a hard deadline fundamentally breaks the power of a monopolist, as backward induction reveals the seller must lower prices as the end approaches to avoid leaving with nothing.

Strategic Cascades in Networks

Finally, this framework extends to how information and behaviors spread. Traditional models of "network cascades" assume people act myopically—copying their neighbors. However, Grant Schoenebeck and Michael P. Wellman argue in Characterizing Strategic Cascades on Networks that agents are actually forward-looking. By using backward induction, these agents might refuse to adopt a popular product today because they anticipate a better or cheaper version will be triggered by the network's future state.

By "planning the last bite first," as Lionel Levine evocatively puts it in How to make the most of a shared meal, this research direction provides a rigorous way to navigate the complexities of human and machine interaction in an interconnected world.

Papers in this version

Showing 60 of 110 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_solving_sequential_games_with_backward_i_564fb24b,
  title = {Solving Sequential Games with Backward Induction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/solving-sequential-games-with-backward-induction/rdr_564fb24b1c724e2eaa3e8cc0dd2563f2},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_solving_sequential_games_with_backward_i_564fb24b_v1,
  title = {Solving Sequential Games with Backward Induction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/solving-sequential-games-with-backward-induction/rdr_564fb24b1c724e2eaa3e8cc0dd2563f2/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}