Bounded Rationality

011 min

The problem bounded rationality names

Picture a product manager who has to pick a support tool for a ten-person team. Forty tools are on the market. Each has a price, a list of integrations, a data policy, a reporting feature and a free trial. A fully careful choice would score all forty on every point and then check how the points trade off against each other. The manager has one afternoon and a team that is waiting.

Economics textbooks long assumed a different kind of chooser. This chooser knows every option, can work out what each is worth and always takes the best one. That model is called rational actor theory. It makes clean predictions, but it describes nobody who has a deadline.

The gap shows up even in a game with fixed rules. Checking every sequence of moves and replies in chess, the way the minimax method does, would mean evaluating more positions than there are molecules in the universe. If perfect choice is out of reach on a board of 64 squares, it is out of reach for forty tools with unclear needs. Bounded rationality is the name for this gap between the perfect chooser and the real one.

022 min

How bounded rationality works

Bounded rationality starts from three limits, and the manager's afternoon makes all three visible. The manager cannot read everything, because each vendor shows only part of what a buyer needs to know, and the vendor knows its own product better than the buyer does. The manager cannot compare everything at once. And the manager cannot take unlimited time, because the team is waiting.

Simon's answer was a rule for stopping. The chooser sets an aspiration level, which is the minimum an option must reach to count as acceptable. Then the chooser examines options and takes the first one that reaches it. Simon described the rule in his 1978 Nobel lecture: once the chooser finds an option that meets the aspiration level, the search ends and that option is taken. The rule is called satisficing. It replaces the question which option is best? with the question is this option good enough?

Here is the manager's afternoon again. The manager writes down what the tool must do. It must handle email and chat, stay under a set monthly price and export all data. The manager opens the tools one by one, and the fourth one meets the list. The manager stops there. Thirty-six tools are never opened. That is the point of the rule, because searching costs time and the rule puts a limit on that cost.

031 min

Where bounded rationality comes from

The manager's rule has a history. Herbert Simon built the idea while studying how organizations decide. His book Administrative Behavior appeared in 1947 and looked at decision making inside administrative organizations. In 1953 he wrote a RAND paper, Behavioral Model of Rational Choice, which proposed a model of choice for organisms with limited computational ability. A paper with the same title appeared in the Quarterly Journal of Economics in 1955. In 1978 he received the Nobel memorial prize in economics for his research into how decisions are made inside economic organizations.

Simon's target was the chooser of classical economics, sometimes called economic man. This figure has full knowledge of the options and unlimited ability to calculate. Simon argued that a theory built on that figure could not describe a real manager. He proposed models that ask only for realistic amounts of knowledge and calculation.

The aspiration idea came from psychology

The stopping rule was not invented from nothing. Simon said satisficing had roots in earlier psychological theories of aspiration levels, due to Lewin and others. Those theories also found that aspiration levels are not fixed. They rise and fall with a person's changing experiences. That link between psychology and economics is one reason the concept is claimed by both fields today.

041 min

How bounded rationality plays out in investing

The manager's rule is one case of a wider question: does careful optimizing beat a simple rule? Finance gives a clean test. The simple rule is called 1/N, which means splitting your money equally across N assets. The careful methods use past returns to estimate the best split. In a paper published in 2009, DeMiguel, Garlappi and Uppal tested 14 such models on seven real datasets, and none was consistently better than the equal split on Sharpe ratio, certainty-equivalent return or turnover.

The three measures are:

• Sharpe ratio: the return earned for each unit of risk taken.
• Certainty-equivalent return: the guaranteed return an investor would accept in place of the risky one.
• Turnover: how much trading a strategy needs.

The authors explain the result with estimation error. The careful models must estimate returns and risks from limited past data. On new data, those errors cancel the benefit of the optimal split. The equal split uses no estimates, so it has nothing to get wrong.

This does not prove that equal splits always win. The test covers 14 models and seven datasets, and other periods or other assets could give a different result. What it does show is that a rule using little information can match a method that uses a lot, when the extra information is noisy.

051 min

A second case: defaults and organ donation

The same limits apply to the people who use a product. Johnson and Goldstein, in a 2004 paper in the journal Transplantation, compared countries by effective consent rate. This is the share of people counted as organ donors, either because they signed up where sign-up is required, or because they did not opt out where consent is assumed. The two groups of countries had no overlap, with nearly 60 percentage points between them.

062 min

What bounded rationality means for your work

These cases give working rules for the people who build products. A product manager sets the must-have list before opening any vendor page, so the first impressive demo does not set the standard. A designer limits the options on a screen to the ones a user can compare, and chooses the default with care, since many users will keep it. An engineer picking a library sets a time box, for example three candidates and two days, and takes the first one that passes the tests. A founder can read the idea as a design rule for the company itself. Simon argued in his Nobel lecture that large organizations can be understood as ways of coping with the limits of people's ability to comprehend and compute. Routines, checklists and approval limits do that work, because they let many people stop at the same point without each repeating the full search.

072 min

How to apply bounded rationality

The roles above share a short routine. Use it when a choice is large enough to need a method and small enough that a full study would cost more than it returns.

1. Write the aspiration level first. List the must-haves and their thresholds before you look at any option.
2. Cap the search. Fix a time or a count, such as three candidates or one day. Every hour of search has an opportunity cost, which is the next-best use of that hour.
3. Take the first option that clears the list, and write down why.
4. When many criteria matter, tally. Counting how many criteria an option meets, with every criterion weighted equally, has often beaten a regression model with fitted weights on small data sets.
5. Raise the aspiration level when a choice is costly to undo. Satisficing saves effort, but it should not replace effort where a mistake cannot be reversed.

The cost of searching for the best

Stopping early also has a personal benefit. Schwartz and colleagues built a scale that measures how strongly people want the best option. In their 2002 paper, people who scored high, called maximizers, reported significantly more regret and depression than satisficers, and significantly less happiness, optimism, self-esteem and life satisfaction. This is a correlation, so it does not show that maximizing causes unhappiness. It does show that the strategy has a cost for the chooser as well as for the clock. Having many options adds to that cost, a theme covered in paradox of choice.

081 min

When bounded rationality breaks

A stopping rule saves effort only while its threshold is sensible. Say the manager had written a list that every tool on the market meets, such as has a chat window. The rule would then stop at the first tool, and the choice would be no better than luck. The opposite failure is a threshold so high that no tool meets it, which sends the search through all forty. Both failures have the same tell: the list was written without looking at what the market offers, or without asking what a mistake would cost.

A rule that works in one setting can also fail in another. Researchers call the fit between a decision rule and its setting ecological rationality. A strategy can be irrational in one setting and entirely rational in another. Stopping at the first acceptable tool suits a market where tools are similar and switching later is cheap. It suits a market with strong lock-in far less.

A third failure is to treat good enough as an excuse. Bounded rationality describes how people decide under limits. It does not say that every shortcut is wise, and the shortcuts named earlier, such as anchoring, still bias the search.

091 min

Bounded rationality vs. nearby concepts

The most common mix-up is with irrationality. Bounded rationality does not say that people ignore reasons. It says that people reason within limits. A person who stops at an acceptable option is acting on a reason, and the reason is the cost of continuing. Four terms sit close by:

The deciding fact is the kind of claim each one makes. Bounded rationality is the cause: the limits exist. Satisficing and heuristics are responses to the cause, and biases are one of their side effects.

102 min

Where the evidence on bounded rationality is contested

Not everyone accepts Simon's picture, and the objections come from more than one side. Start with the economists who defend the classical model. Simon noted in his Nobel lecture that Gary Becker treats any deviation from utility maximization as irrational, so what Simon called bounded rationality counts as irrationality in Becker's terms. The disagreement is partly about a word: is a rule that ignores options irrational, or rational under limits?

A second objection says Simon's contrast is smaller than it looks. For a majority of researchers across fields, bounded rationality is some form of optimization under constraints, in which the cost of search is counted in the calculation. On that reading, the manager who stops at the fourth tool is maximizing, once the cost of looking is included. Simon's own emphasis was on the search procedure and on when it stops.

?3 questions

Questions people ask

§7 sources

Sources