Bounded Rationality

Bounded rationality is the empirical proposition that investors have limited capacity for gathering and processing information. Where belief-based biases keep the investor’s attention on the right data set and merely twist the probability calculation, bounded-rationality models change the input to the calculation: the investor sees only a subset of the signals that a fully rational agent would weigh, processes them with simple categories or fractions rather than first-principles algebra, and treats nominal and real numbers as if they were the same thing.

The course organizes the bounded-rationality literature around four canonical phenomena, each with a paradigm empirical paper:

• Limited attention generates post-earnings-announcement drift (PEAD) and a host of related under-reactions. Hirshleifer-Lim-Teoh (2009): PEAD is stronger when many firms announce earnings the same day. DellaVigna-Pollet (2009): PEAD is stronger for Friday-afternoon announcements. Hou-Peng-Xiong (2008): PEAD is stronger in low-volume stocks.
• Buy-sell asymmetry appears because attention matters more for the purchase than for the sale decision: an individual investor can only buy from a small subset of stocks she happens to be looking at, but typically sells one of the few she already owns. Barber-Odean (2008) document strong net buying by retail traders in extreme-return, high-volume, news-cited stocks.
• n$ heuristic describes naive diversification across whatever options are presented. Benartzi-Thaler (2001): given $n$ funds in a $401(k)$ menu, participants invest roughly $1/n$ in each, so the equity allocation depends on how many equity funds the menu offers rather than on any considered portfolio choice.
• Nominal illusion is the failure to deflate cash flows when discounting. Modigliani-Cohn (1979) argue that the U.S. stock market was undervalued during the high-inflation $1970$s because investors discounted real dividends at a nominal rate.
• Categorical thinking describes the human tendency to form beliefs at the category level (small-cap, value, “technology”) rather than the security level. Barberis-Shleifer (2003) build an asset-pricing model on this premise; Barberis-Shleifer-Wurgler (2005) confirm that stocks added to the S&P $500$ subsequently comove more with the index even though their fundamentals have not changed.

Where bounded rationality sits in the three-pillar framework. The three pillars of behavioral finance — limits to arbitrage, behavioral preferences, biased beliefs — leave a residual category for cognitive capacity limits that are neither preferences nor probability errors. Bounded rationality fills that residual. It overlaps with belief biases (categorical thinking is close to representativeness; limited attention is close to inattention) but is treated as a separate pillar in the syllabus because the modeling approach differs: the agent is not biased given her information set, she simply has a smaller information set or a coarser way of representing the world.

Intuition

Suppose you are a household investor with a job, a family, and a retirement account. You cannot read the $10$-K of every public firm; you cannot recompute your optimal portfolio every time interest rates move; you cannot recalibrate your inflation expectations every quarter. The rational benchmark assumes you do all of these things with infinite cognitive bandwidth. The bounded-rationality literature drops this assumption and asks: which financial-market patterns become understandable once we accept that investors have limited attention, use simple heuristics, and group assets into broad categories rather than treating each one as a unique cash-flow stream?

The tension that organizes the chapter is between two ways of being wrong. A belief-biased investor is paying full attention to the right data and then mis-weighting it; a bounded-rational investor is paying attention to the wrong data, or to too little data, or is collapsing the data into a coarse category. The two failures often produce similar predictions — both can generate under-reaction and excess comovement — but the modeling choices differ, and the policy responses differ even more. A belief-biased investor benefits from explicit de-biasing; a bounded-rational investor benefits from a better choice architecture. The latter response is the subject of the prescriptive companion chapter.

Children

• 8.11.1. Limited Attention and PEAD — limited attention as an explanation for under-reaction to earnings news (PEAD) and to other announcements (customer-link news, demographic shifts).
• 8.11.2. Attention-Grabbing Stocks and Buy-Sell Asymmetry — Barber-Odean ($2008$) show that retail investors disproportionately buy attention-grabbing stocks (extreme-return, high-volume, news-cited) but do not disproportionately sell them, because attention is asymmetrically necessary for purchase.
• 8.11.3. The 1-over-n Heuristic — Benartzi-Thaler ($2001$) document that $401(k)$ participants invest roughly $1/n$ across the $n$ available funds, making aggregate equity allocations a function of menu design rather than considered portfolio choice.
• 8.11.4. Nominal Illusion — Modigliani-Cohn ($1979$) hypothesis that investors discount real cash flows at nominal rates, yielding undervaluation in high-inflation regimes and overvaluation in low-inflation regimes.
• 8.11.5. Categorical Thinking and Comovement — Barberis-Shleifer ($2003$) model in which investors form beliefs at the category level. Generates excessive within-category comovement; tested by Barberis-Shleifer-Wurgler ($2005$) on S&P $500$ index inclusions.

See Also

• 8.12. Prescriptive Behavioral Finance — the policy-design companion chapter.
• 8.6. Belief-Based Biases — sibling pillar; representativeness in particular overlaps with categorical thinking.
• 8.9. Reaction Models — first-generation under-reaction models that limited-attention papers complement empirically.
• 8.1. Principles of Behavioral Finance — the three-pillar framework that places bounded rationality alongside biased beliefs and limited arbitrage.
• 8. Behavioral Finance — top-level hub.