When pattern seeking replaces a decision rule: a probability-matching audit for investment committees

Understand the experiment before using the analogy

Participants predicted which of two images would appear in a repeated lottery. The underlying frequencies were fixed but initially unknown to them, and payoff structures varied between assigned designs. The experiment recruited 82 volunteers at the MIT Behavioral Research Laboratory; seven were excluded after a comprehension control, leaving 75. Each participant completed 200 main trials, with the first 50 omitted from the principal stable-behavior analysis. Outcomes paid real money, but total earnings were modest. This is a controlled decision task, not evidence about actual family-office portfolios or market timing.

The authors found different levels of randomization across payoff environments and interpreted the difference through an evolutionary model. They also found that many participants reported seeing patterns in a random sequence. The useful governance question is whether a committee’s changing choices follow a documented reason or an imagined sequence. That does not mean every change is irrational, nor that repeatedly selecting one investment is optimal. Real assets have correlated risks, uncertain probabilities, transaction costs and constraints absent from the image lottery. Preserve these differences whenever the study is used in training or an investment memo.

The source also contains inconsistent design labels: the initial-learning passage reverses theoretical frequency labels relative to the probability-matching section, and a discussion parenthesis names designs 2 and 4 where other passages name designs 2 and 3. These wording issues remain unresolved; the proposed audit does not use them to calibrate a decision rule.

Write down the objective and the decision rule

The chief investment officer should ask proponents to state the objective before choosing an action. Is the office trying to maximize expected cash flow, avoid a shortfall, preserve liquidity, diversify a concentrated business exposure or learn about an uncertain opportunity? A decision rule cannot be judged without its objective. Record the rule, its assumptions and the circumstances that would change it. In the laboratory’s simple task, choosing the more frequent outcome can maximize expected reward; that observation does not justify concentrating a real portfolio in the asset with the highest estimated return.

For an actual allocation, the risk lead should document dependencies between investments and the consequences of simultaneous losses. Diversification may be a deliberate response to uncertain estimates or constraints, rather than probability matching. The source should not be used to shame a committee member for preferring diversified exposure. Instead, ask whether the stated allocation follows the documented objective and whether the relevant assumptions are supported. Stop the analogy when a proposal ignores covariance, downside limits or liability timing, because those features can reverse the reasoning that applies in a repeated independent lottery.

Require an evidence update for a pattern claim

A practical decision log should distinguish a new observation from a new interpretation. Record the date, the proposed pattern, the data available before the decision and the action it would trigger. A manager saying that an outcome is due after a streak should specify why the generating process would change after that streak. If no mechanism or predictive evidence is supplied, classify the claim as a hypothesis for testing, not a fact. The experiment’s reports of pattern seeking make this a useful question, but they do not prove any specific market claim is false.

A research analyst can compare a proposed signal against a simple reference rule using information that was genuinely available at the relevant time. Keep exploratory development separate from evaluation and preserve failed variants rather than reporting only the successful one. Measure incremental predictive value, costs and sensitivity to plausible changes in assumptions. Do not use a small successful backtest as a direct license for larger commitments. Pause the proposed signal if its apparent value disappears after correcting look-ahead information, if the rule changes after each failure, or if the evaluation cannot be reproduced.

Treat repeated observations and expertise carefully

The study explicitly considers dependence between repeated decisions by the same participant. Hundreds of choices are not hundreds of independent people. A family office should apply the same caution when assessing a manager or committee: many trades during one market regime may reflect a common exposure or repeated use of one rule. The data lead should report the number of independent decision-makers, investment episodes and relevant conditions, rather than relying on a large row count. Review whether apparent precision comes from treating correlated observations as independent evidence.

The paper reports that participants with probability coursework and those claiming to find a pattern randomized more in this task. Those associations should not become hiring rules or judgments about an individual’s competence. The retained sample is small, and the authors warn about subgroup power and generalization to other stakes and contexts. The proposed response is behavioral evidence: ask every decision-maker, expert or novice, to explain the rule and demonstrate how new information changes it. Stop any use of the article that ranks people by educational background, family status or presumed susceptibility to irrationality.

Use a low-stakes exercise to improve the record

The governance lead may use a transparent teaching exercise in which participants describe a rule, observe simulated outcomes and explain revisions. The aim is to identify ambiguity in the decision process, not to diagnose participants or certify investing skill. Make participation voluntary, disclose that the task is a simplified simulation and avoid using results for succession or compensation. Record whether participants can distinguish a change in evidence from discomfort after a losing streak. Measure clarity of reasoning and consistency with the stated objective rather than rewarding a lucky short-run result.

The committee should review whether the exercise improved documentation in subsequent real decisions without claiming a causal performance benefit. Owners are the chief investment officer for objectives, risk lead for constraints, analyst for signal evaluation and governance lead for learning records. Reopen a decision rule when its assumptions or obligations materially change. Stop capital deployment when the rule cannot be stated, when necessary risk inputs are missing or when testing remains inconclusive. This AI-assisted workflow is a proposed application of the research, not a validated investment method or an endorsement by its authors.

Source, attribution and limits

[1] Andrew W. Lo; Katherine P. Marlowe; Ruixun Zhang. To maximize or randomize? An experimental study of probability matching in financial decision making. PLOS ONE (2021-08-26). https://doi.org/10.1371/journal.pone.0252540. CC BY 4.0.

Laboratory binary-choice experiment with 82 recruited volunteers and 75 retained after a comprehension control, modest monetary payoffs and repeated image predictions. Not family offices or actual portfolio allocation. Small-sample demographic associations and an evolutionary interpretation do not establish universal mechanisms or investment advice. Some design-number and theoretical-frequency labels conflict across the source’s initial-learning, probability-matching and discussion passages; this companion does not rely on the inconsistent labels.

Prepared by the Family Office Private Wealth Journal editorial research desk with AI assistance at the direction of Dr. Freedom Cheteni. Original research authors retain full credit and are not authors or endorsers of this companion. This proposed process has not been evaluated in a new empirical study; no independent peer-review acceptance is claimed. This is educational analysis, not individualized investment, legal or tax advice.

Implementation record

Open the cited sources

1. [1] Andrew W. Lo; Katherine P. Marlowe; Ruixun Zhang. To maximize or randomize? An experimental study of probability matching in financial decision making. PLOS ONE (2021-08-26). ↗