Sequential Fundraising and Mutual Insurance

Amir Ban, Moran Koren

Abstract

Seed fundraising for ventures often takes place by sequentially approaching potential contributors, who make observable decisions. The fundraising succeeds when a target number of investments is reached. Though resembling classic information cascades models, its behavior is radically different, exhibiting surprising complexities. Assuming a common distribution for contributors' levels of information, we show that participants rely on mutual insurance, i.e., invest despite unfavorable information, trusting future player strategies to protect them from loss. Delegation occurs when contributors invest unconditionally, empowering the decision to future players. Often, all early contributors delegate, in effect empowering the last few contributors to decide the outcome. Similar dynamics hold in sequential voting, as in voting in committees.

Introduction: Beyond the Information Cascade

In the world of venture capital and crowdfunding, the order of operations is critical. We often observe "seed rounds" where investors are approached one by one. If you are the first person asked to invest, do you rely solely on your own data, or do you look at who is coming after you?

Traditional economic models of "information cascades" suggest that people often ignore their own private information to follow the crowd—the classic example being a diner choosing a busy restaurant over an empty one because the crowd's presence signals quality. However, a new paper by Amir Ban and Moran Koren, "Mutual Insurance and Delegation in Sequential Fundraising," flips this script. While classic theory suggests we mindlessly follow the person in front of us, this research shows that in fundraising, we often strategically trust the person behind us.

The Problem: Interdependent Payoffs

For decades, the study of information cascades assumed that an individual's payoff depends only on their own action. In sequential fundraising or committee voting, the rules are fundamentally different. A project only proceeds if a target number of participants ($B$) out of a total group ($n$) agree to invest.

This creates a unique risk structure:

  1. Failure to Launch: If the project fails to reach the target $B$, no one loses anything.
  2. The Lemon Risk: If the project succeeds (reaches $B$) but the project is "bad," every person who invested loses money.

The authors address a specific gap in game theory: How do rational, forward-looking agents behave when their success depends on the decisions of those who haven't even played yet?

Methodology: Markov Perfect Equilibrium

The authors construct a mathematical model using Markov Perfect Equilibrium (MPE) to analyze the sequential game.

  • The Setup: $n$ players act in order. Each receives a private signal (good or bad) with a specific "quality" (reliability).
  • Threshold Strategies: The research proves that rational players use a "threshold" approach. They invest only if their private information—a combination of signal and quality—exceeds a certain mathematical limit.
  • Bayesian Updating: As each player acts, the "public likelihood" of the project being good is updated.
  • Unanimity vs. General Case: The paper provides rigorous analytical proofs for "unanimity games" (where $B = n$) and uses numerical simulations and dynamic programming for the general case where only a subset of players is required.

Key Insights: Mutual Insurance and Delegation

The research identifies two novel behaviors that emerge from this forward-looking rationality:

1. Mutual Insurance

This is the "safety net" effect. An early investor might receive a slightly negative signal but choose to invest anyway. They do this because they know that if the project is truly a "lemon," the players coming after them will likely detect the flaws and decline to invest. By investing, the early player keeps the project alive for the possibility of a gain, "insured" by the skepticism of future players who act as a final filter.

2. Delegation

In some scenarios, a player’s best move is to invest unconditionally, regardless of their own information. By doing so, they "delegate" the final decision to the next person in line. This typically occurs when the project's prior reputation is decent and the player trusts that future participants will have better or additional information to make a more informed final call.

3. The Reverse Cascade

In traditional cascades, the first few people dictate the outcome for everyone else. This paper finds the opposite: early players often delegate or use mutual insurance, effectively empowering the last few contributors to be the ultimate deciders of the project’s fate. Unlike "myopic" agents in older models, these players calculate the probability of completion based on the optimal strategies of all subsequent players.

Key Results and Mathematical Bounds

  • Unanimity is Solvable: In cases where everyone must invest ($B=n$), the authors provide a closed-form characterization of the equilibrium. They show that players always shade their thresholds lower (becoming more "agreeable") compared to a single-player scenario.
  • Cascade Limits: The research identifies specific mathematical bounds for when an "up-cascade" (unconditional investment) or "down-cascade" (unconditional rejection) begins. Notably, a down-cascade is much harder to trigger when the target $B$ is large.
  • Large $n$ Behavior: As the number of potential players grows, thresholds converge to a specific value ($1-Q$, where $Q$ is the maximum signal quality), and the probability of the "correct" outcome stabilizes.
  • Voting Equivalence: These fundraising dynamics apply almost perfectly to sequential voting in committees, where the utility of a "yes" vote depends on whether the motion actually passes.

Limitations and Future Directions

While the model provides a robust framework, it opens several avenues for future research. The authors noted that while numerical simulations strongly suggest mutual insurance occurs in non-unanimity games ($B < n$), a formal proof remains an open challenge (referred to as "Conjecture 5").

Additionally, the model assumes all players draw their signal qualities from the same distribution. In reality, "Lead VCs" often possess significantly better information than follow-on investors. Future work could also explore endogenous timing—what happens when investors choose when to participate, rather than following a fixed order set by the fundraiser.

Summary updated

Abstract

Seed fundraising for ventures often takes place by sequentially approaching potential contributors, who make observable decisions. The fundraising succeeds when a target number of investments is reached. Though resembling classic information cascades models, its behavior is radically different, exhibiting surprising complexities. Assuming a common distribution for contributors' levels of information, we show that participants rely on {\em mutual insurance}, i.e., invest despite unfavorable information, trusting future player strategies to protect them from loss. {\em Delegation} occurs when contributors invest unconditionally, empowering the decision to future players. Often, all early contributors delegate, in effect empowering the last few contributors to decide the outcome. Similar dynamics hold in sequential voting, as in voting in committees.
BibTeX
@misc{ban2020sequential,
  author = {Amir Ban and Moran Koren},
  title = {{Sequential Fundraising and Mutual Insurance}},
  year = {2020},
  eprint = {2005.10711},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2005.10711},
}

Loading related content...

Figure