Week 3: Heuristics and Biases

1. What are heuristics?

The slide diagram (in Greek) shows the process: complex problem → heuristic shortcut → fast decision. The word heuristic comes from the Greek "heurisko" (to find), the same root as Archimedes' "Eureka" (I found it), which the lecture illustrates with an 18th-century engraving.

Rules of thumb: the wine example

Choosing a bottle of wine shows how rules of thumb help and fail:

• Colour and taste: "a deep red wine has more body, so it goes with red meat". Useful, but if the dish is very strongly flavoured the rule may still give a poor match.
• Matching by origin: "Italian dish? Italian wine!" Simple, but you miss wines from elsewhere that might complement the flavour better.
• Price: "expensive means good". But excellent wines exist at every price level, so there is no need to overpay.

The same structure appears in investing: "a famous company is a good investment", "buy what you know", "a high price means high quality". Each rule saves time, and each produces predictable mistakes.

The ultimate rule of thumb: Occam's razor

The principle is named after William of Ockham, an English philosopher and theologian of the 14th century, and says that if two theories predict equally well, you should pick the simpler one. (The slide points to House M.D., season 8, episode 15.) In science Occam's razor is a handy heuristic, but mind the condition "equally accurate": a simpler account that explains less is not an improvement.

2. Two systems of thinking

Kahneman and Tversky argue that the mind works with two distinct systems. Their interaction explains many cognitive biases in investment decisions.

System 1: fast and intuitive

• Works unconsciously and automatically.
• Uses mental shortcuts and heuristics.
• Gives fast decisions, though often with systematic errors.
• Responds instantly to stimuli, e.g. an emotional response to market news such as a tariff announcement.
• Innate: answering 5 + 5 takes no thinking.
• Example: an investor hears "the market is falling" and sells on impulse.

System 2: slow and analytical

• Works deliberately and consciously.
• Demands attention and cognitive effort.
• Comes into play for decisions that are complex or need reasoning.
• Applied in valuation, forecasting and risk analysis.
• Able to monitor System 1 and override it.
• Example: before investing, an analyst examines a company's fundamentals.

Interaction between the two systems

• The two systems cooperate but often conflict.
• System 1 produces an immediate response (the initial reaction).
• System 2 may accept it or correct it through reflection (control and correction).
• A lot of investment choices come from System 1, without System 2 checking them as it should.
• System 1 swaps a hard question for a simpler one. Faced with the question of where the stock market will stand in 5 years, it simply looks at where the market stands now.

The invisible gorilla (Chabris and Simons)

In this famous video experiment, viewers count the passes between players in white shirts. Meanwhile a person in a gorilla suit walks through the scene. When the experiment was run, about half of the participants did not see the gorilla, and many swore it had not been there. Lesson: when we focus hard on one task, we miss everything else. When System 2 is fully occupied, it does not notice what System 1 would normally pick up. For investors: someone focused on one metric (for example quarterly earnings growth) can miss an obvious risk (for example rising debt).

Implications for behavioral finance

• Most investors rely on System 1 under stress or uncertainty.
• System 1 explains overreaction and underreaction, herding behavior and emotional trading.
• Developing System 2 helps reduce biases; education, reflection and discipline can balance the two systems.

Case study: System 1 during the 2008 crisis

In the global financial crisis, many investors responded impulsively and out of fear (System 1). Large-scale selling without any fundamental analysis caused the market to overreact. Investors who engaged System 2 assessed valuations calmly, spotted chances in undervalued assets and earned higher returns once markets recovered. Moving from panic and fear (System 1) to deliberate analysis (System 2) requires awareness. The lesson: switching on System 2 under pressure is hard, yet it is vital for managing risk.

Mini quiz: the Cognitive Reflection Test

The purpose is to catch System 1 "tricking" System 2.

Lesson: System 1 has its uses, but in important decisions System 2 has to keep an eye on it.

3. Introduction to the main heuristics

Investors do not calculate everything rationally. They use heuristics, simplified mental processes that reduce cognitive load but create systematic errors. The main types listed in the lecture:

• Anchoring: relying too much on an initial number. Example: an investor refuses to sell below the €50 purchase price, or an analyst adjusts last year's forecast too little.
• Availability: judging likelihood by how easily examples come to mind.
• Representativeness: judging likelihood by similarity to a prototype.
• Overconfidence: overestimating knowledge, ability or precision.
• Confirmation bias: seeking information that supports existing beliefs.
• Recency bias: overweighting the most recent events.
• Herd behavior: following what others do, which can make many investors buy or sell at the same time and push prices away from fundamental value. Herding is treated in detail later in the course.

4. Availability

• Investors overweight information that is recent or vivid.
• Right after a plane crash, people overestimate the odds of another one.
• Media coverage makes certain risks, such as terrorism or market crashes, more available.
• As a result we come to believe the stock market crashes every quarter.
• Those who let System 1 make all their decisions are hit hardest.

How the availability bias works

1. Memory is selective: vivid, recent or emotional events are stored more easily.
2. When we need to estimate a probability, the brain retrieves whatever comes to mind fastest.
3. This creates distorted perceptions of risk.

Put briefly: whatever is "available" in memory takes the place of what is "statistically probable".

Example: perceived versus actual frequency

The slide chart compares actual and perceived frequencies (in %) of causes of harm:

Dramatic, heavily reported events (plane crashes, terror) are strongly overestimated, while common but "boring" risks (heart disease, car accidents) are underestimated. In markets the analogue is fearing a spectacular crash more than the slow damage of high fees or poor diversification.

Classic experiment: Tversky & Kahneman (1973)

Question: in a random English word (the slide uses 7-letter words), is "R" more likely to be the first letter or the third letter?

Result: 70% of participants chose "first letter". In fact many more words have R as the third letter (for example bird, car, word). Because words starting with R are easier to retrieve (we search memory by first letter), they are perceived as more frequent.

Four modes of influence (Khan et al., 2017)

• Resonance: information that fits our own values, traits and worldview gets more weight.
• Retrievability: what comes to mind easily is judged more reliable.
• Narrow range of experience: compared with experienced investors, inexperienced ones put more weight on new information.
• Categorisation: new information is sorted into existing categories or matched to past experience.

Conclusions and mitigation

Recent or dramatic events get too much weight, which produces waves of optimism and pessimism that have no fundamental grounds. In finance the result is overreaction and weak risk assessment. Ways to limit it: rely on long-run data and statistical analysis, keep emotion apart from analysis, and stay alert to the influence of the media and "noise".

5. Representativeness

• "A good company is also a good investment" is a common fallacy. A great company whose price already reflects its quality may be a poor investment.
• It leads to false expectations about "new technology" stocks: they look like the next big success story.

Example: Domino's in Italy

Domino's entered the Italian market in 2015 with a plan for 880 stores by 2030. It opened 29 and exited Italy in 2022. Domino's thrives globally and Italians eat a lot of pizza, so by representativeness the move "seemed logical". When two situations look similar, we assume they share other traits too, and forget the differences (strong local competition, taste for traditional pizza).

The taxi problem (Kahneman & Tversky, 1972)

In a city 85% of taxis are green and 15% are blue. A taxi was involved in a hit-and-run accident. A witness says it was blue. Tests show the witness identifies colours correctly 80% of the time under similar conditions. What is the probability the taxi was actually blue?

The intuitive (System 1) answer is 80%: "the witness says blue and is right 80% of the time". This ignores that blue taxis are rare.

Lesson: System 1 ignores base rates (prior probabilities); System 2 includes them through Bayesian reasoning. In finance: a stock that "looks like" a winner (strong story, charismatic CEO) is judged likely to succeed, while the base rate of success for such firms is low.

What representativeness makes us ignore

• The relative frequency with which events occur.
• The base rate: we estimate P(A | B) from similarity without considering the prior probability.
• Sample size (sample size neglect).

The law of small numbers (Kahneman & Tversky, 1971)

People believe that small samples accurately reflect the population. The experiment: a town has two hospitals, a large one with about 45 births per day and a small one with about 15. About 50% of babies are boys. For one year, each hospital recorded the days on which more than 60% of births were boys. Which hospital recorded more such days?

The answers given were: large hospital 21%, small hospital 22%, roughly equal 57%. The right answer is the small hospital, since small samples deviate from 50% more easily.

The slide graph plots Var(p̂) against sample size n: it falls steeply for small n and flattens out. Small samples produce more "extreme" results.

Applications in finance

Suppose an investment manager has posted excellent returns for the last 2 years. Many investors decide he must be "skilled" and ignore how small the sample is.

• A short performance history increases the risk that success is random.
• Fund performance over n = 24 months is not statistically reliable.
• Overreacting to small samples leads to the kind of behavior seen in bubbles.

Lesson: you need a big sample and a probabilistic framework for statistical reliability, not only a handful of success stories. For example, if 1,000 managers choose stocks at random, roughly 250 of them will outperform the market two years running purely by luck (0.5 × 0.5 = 25%).

6. The gambler's fallacy

Example: having seen 5 heads in a row, a person thinks tails has become more likely. In fact every toss is independent, so the probability remains 50%. Beforehand, the odds of getting 5 heads in a row are 0.5⁵ = 1/32 ≈ 3.1%, yet once those five tosses are done the next one is still 50/50. On the slide a sequence H H H H H followed by T appears, with the remark that our minds look for patterns even where everything is random.

Experiments

In the study by Tversky & Kahneman (1971), participants had to write out "random" strings of 0s and 1s. They steered clear of long runs (such as HHHH) and wrote strings that were too "balanced" (such as HTHHTT). This is called the illusion of negative correlation: people expect chance to even itself out, even in small samples. Gilovich, Vallone & Tversky (1985) looked at the reverse case, the hot-hand fallacy in basketball: fans think a player who has just scored a few times has a higher chance of scoring next.

Applications

• Stock market: investors believe that after several declines "the market will rebound".
• Monte Carlo, 1913: black came up on the roulette wheel 26 times running. Gamblers kept putting money on red and lost large sums. (Any particular run of 26 has a minuscule chance, roughly 1 in 67 million, yet every spin remained independent.)
• Reversal fallacy: traders ignore the statistical independence of returns.
• It connects to the representativeness heuristic (searching for patterns, where a sequence that "looks random" serves as the prototype) and to the hot-hand fallacy (the slide asks Curry how many more threes he thinks he will make).

Note the two errors point in opposite directions: the gambler's fallacy expects a reversal, the hot hand expects continuation. Both come from seeing patterns where there are none.

Excel simulation: coin toss streaks

Goal: show that streaks are a normal feature of randomness.

1. Enter =IF(RAND()<0.5,"H","T") in column A and fill it down as far as row 200.
2. In column B count the current streak: =IF(A2=A1,B1+1,1).
3. Make a histogram of column B.

You will see real streaks of 4 to 6 identical outcomes. In 200 fair tosses the longest streak is typically around 7 or 8, and a streak of at least 6 occurs in almost every run of 200 tosses.

Discussion: a streak does not point to a pattern or a systematic deviation; streaks arise naturally from randomness. The gambler's fallacy comes from people being surprised at how frequently streaks show up. Questions to consider: how frequent are streaks longer than 5? Would you predict more or fewer of them if the process were "truly random"? Does the histogram match your idea of randomness? Conclusion: to the human mind, randomness often does not look random.

7. Overconfidence

Example: 82% of respondents placed themselves among the top 30% safest drivers (Svenson, 1981). At most 30% can be in the top 30%, so many are wrong. Investors likewise believe they "know the market better than others".

Three forms of overconfidence

1. Overestimation: overestimating one's actual ability or performance.
2. Overplacement: believing one is better than the average person (the driver study).
3. Overprecision: excessive confidence in the accuracy of one's estimates, so confidence intervals are too narrow.

Example of overprecision: "What is the distance between Athens and Paris? Give a 90% confidence interval." Most people give a range that does not contain the true answer (roughly 2,100 km by air).

Experiment: confidence interval task (Alpert & Raiffa, 1982)

Participants estimated quantities (such as city populations or stock prices) and gave 90% confidence intervals. Only about 50% of the intervals contained the true value, instead of 90%. People overestimate the accuracy of their knowledge.

Economic consequences

• Lower returns: very active investors lose out through transaction costs.
• Overtrading: believing their information is superior, investors trade heavily.
• Underestimation of risk: confidence in their own forecasts pushes them to use more leverage.

The slide chart "Trading frequency and portfolio returns" shows average annualised return falling roughly linearly from about 8% at overconfidence level 0 to about 0% at level 10. As overconfidence increases, average return decreases.

Overconfidence among analysts and CEOs

• CEOs are too optimistic about how their companies will perform.
• Analysts' forecasts often come with confidence intervals that are too narrow.
• A frequent result: overinvestment and acquisitions that fail.
• Malmendier & Tate (2005): when internal liquidity (cash) is available, overconfident CEOs invest beyond the optimal level. Believing the market undervalues their firm, they shy away from external financing but overspend internal cash.

Overconfidence and confirmation bias: two-in-one

• Outcomes that confirm our predictions are credited to skill, the rest are written off as bad luck: "Heads I win, tails it's luck."
• Information that supports our beliefs receives more weight.

Psychological roots of overconfidence

• Illusion of control: the feeling that we can control random outcomes.
• Self-attribution bias: people credit their successes to skill and blame their failures on luck.
• Confirmation bias: attention only to information that confirms our views.

Together these reinforce overconfidence and reduce self-awareness: every success confirms "I am good", every failure is explained away, so the investor never learns.

Case study: the dot-com bubble (1999 to 2000)

Investors told themselves "this time is different". Analysts issued very optimistic forecasts with no fundamental grounding, and too much faith in new tech business models inflated the bubble. A Bloomberg chart of the Nasdaq-100 shows the index rising to a top of roughly 4,700 in March 2000 and dropping to about 800 by the end of 2002. In the crash the NASDAQ shed more than 75%.

Overconfidence and mergers

Overconfident CEOs overestimate their own abilities and their firm's capacity to create value through acquisitions. They believe they have superior information or judgment, underestimate integration risks, overestimate synergies, and show an illusion of control and delusion of precision. Result: excessive acquisitions, overpayment and lower future returns for the acquirer's shareholders.

• The CFA Institute reports that 75% of acquisitions fail because of overconfidence (and investment bankers who are too eager), against a failure rate of 36% for projects run internally.
• Hayward & Hambrick (1997): acquirers pay on average 49% more than the target is worth; the acquirer's share price falls about 4% in the first few days and about 11% over the following year.

Case study: National Bank of Greece and Finansbank (2006)

• In 2006 NBG bought 46% of Turkey's Finansbank for about €2.3 billion, at a price-to-book multiple above 4x (paying more than four euros for every euro of book equity). It was the largest Greek acquisition abroad.
• The narrative: NBG would become a regional banking leader in the Balkans, and Greek banks could "export" know-how and growth.
• Outcome: once 2008 had passed, the Turkish lira lost value and profits fell. In 2016 NBG was forced to sell Finansbank for roughly €2.7 billion, below what it had originally invested after correcting for exchange rates and the capital it had added. Over that period NBG's share price dropped by more than 85%.
• Behavioral analysis: management overrated its capacity to carry the Greek model over to Turkey, underrated exchange-rate and political risk, and showed overplacement ("we can do it better than local Turkish banks"). The results: a steep valuation (a premium of over 40%), subsequent losses and eroded capital, plus harm to its reputation.

Lesson: overconfidence leads to overpayment, underestimated risk and destroyed shareholder value.

Classroom experiment: true/false with confidence

Students answer 10 true/false statements and show their confidence (1 finger = 60%, 2 = 80%, 3 = 90%). If people are well calibrated, statements answered at 90% confidence should be right about 90% of the time.

Several of these are traps where a confident answer is wrong (Amsterdam for the ECB, $150 oil). The typical result is that hit rates at "90%" confidence are well below 90%: overprecision.

Summary on overconfidence

• Overconfidence is everywhere, from retail investors to CEOs.
• The result is too much trading, too much risk and weaker returns.
• De Bondt and Thaler (1985) consider overconfidence one of the most consistent and well-supported findings in the psychology of judgment.
• It makes people insensitive to new information and inclined to take more risk.
• Understanding the bias helps risk management and self-discipline.

8. Confirmation bias

The slide chart "Confirmation bias: news selection" shows an investor who already likes a stock:

Of the 220 available articles, 45% are positive (100/220). Of the 100 articles the investor actually reads, 70% are positive. His "information set" is far more optimistic than the real evidence. In investing, a practical counter-measure is to actively write down the bear case (the arguments against) before buying.

9. Recency bias

• The latest events receive too much weight from investors.
• Investors tend to assume recent returns will persist.
• This partly explains phenomena like momentum trading.

The slide chart shows exponential weights on past returns against the lag (periods ago). The "investor (recency)" curve starts at 1 and falls quickly, to about 0.4 after 5 periods and close to 0 after 20. The "rational weighting" curve falls slowly, still about 0.6 after 10 periods and about 0.2 after 30.

10. Mental accounting

Examples of mental accounts: daily expenses, vacations, "profits" from investments.

The theatre ticket (Thaler, 1980)

• Scenario A: you bought a €20 theatre ticket and lost it. Do you buy another?
• Scenario B: you go to buy a €20 ticket and find you have lost a €20 note. Do you buy the ticket?

Result: fewer people buy a new ticket in scenario A. In A the lost ticket is charged to the "entertainment" account, so the play now seems to cost €40. In B the lost note is charged to general cash, so the play still costs €20. Economically both losses are the same €20; psychologically they are different.

How mental accounting works

1. The brain creates a separate account for each activity (salary, vacation, investment).
2. Each account has its own rules for acceptable losses and gains.
3. Decisions are made within each account independently, with often no communication between accounts.

Money is not treated as one pool; these "mental boxes" lead to non-rational decisions.

Mental accounting in investments

• Investors split their portfolio into a "safe" part and a "risky" part.
• They evaluate each part separately, not as a whole, which violates portfolio theory (where only the risk of the total portfolio, including correlations, matters).
• The disposition effect appears: they sell winners to "close" a positive account and keep losers to avoid closing an account at a loss.
• Example: an investor sees gains on a trade as a "bonus" and is willing to take bigger risks with that money (compare the house money effect, Week 2).

The slide "Portfolio segregation" shows a safe investment (bonds) and a risky investment (stocks) evaluated separately: the investor sees two accounts instead of one unified portfolio. For instance, he may panic about a 20% loss on the stock account even though his total portfolio (say 70% bonds that gained 3%) is down only about 3.9% (0.3 × −20% + 0.7 × 3% = −3.9%).

Real examples

• People see a tax refund as a "gift" instead of ordinary income, and spend it more easily.
• Retail investors take more risk with "extra" funds from temporary gains.
• Corporate finance: CFOs allocate capital per project without integrating risk across projects.
• Bonus culture: managers maximise their own "bonus account", not total firm value.

Framing and the source of money influence behavior.

Classroom experiment: €300 from a lottery or from a refund

Students were asked how much they would spend and save if (1) they won €300 in a lottery, or (2) they got €300 back after an electricity overcharge. Typically people spend more of the lottery money ("windfall", "fun money") and save more of the refund (seen as their own money returned). The economic reality is identical: €300 more in the bank.

Main findings: people judge money by the "label" attached to it, keep separate accounts for separate sources, and end up making suboptimal choices. Link to theory: the 1999 Thaler article "Mental Accounting Matters", published in the Journal of Behavioral Decision Making. Mental accounting is tied to loss aversion, framing and the disposition effect.

Conclusions and applications

• People do not treat money as perfectly fungible.
• Mental labelling leads to wrong investment and consumption decisions.
• Financial advisers should help clients "consolidate" their mental accounts, and education in holistic portfolio management reduces the bias.

11. Overall conclusions

Heuristics simplify decision making but create systematic biases. System 1 produces them; System 2 can catch them if we know what to look for. Understanding availability, representativeness, the gambler's fallacy, overconfidence, confirmation bias, recency bias and mental accounting leads to better risk assessment and better investment behavior.