Behavioural biases, and why good decisions are hard

What a bias is, and what it is not

A behavioural bias is a mental shortcut that produces a systematic error — the same error, in the same direction, in most people facing the same situation. It is not carelessness and it is not low intelligence. It is a rule that works nearly everywhere, applied to the one place it does not.

Start with what the shortcut normally does for you. You judge how long a queue will take by looking at it rather than counting. You decide a road is dangerous because you remember an accident on it. You read a stranger's mood from a face in a quarter of a second. None of these is computed; all of them are usually right; and stopping to reason each one out properly would leave you worse off across a day, not better.

That is the trade the mind has made, and it is a good one. Tversky and Kahneman set the frame for the whole field in Judgment under Uncertainty: Heuristics and Biases (Science, 1974): the shortcuts are not random noise around a correct answer, they are rules with a predictable direction of error. Predictable is the operative word. A random mistake averages out over many decisions. A systematic one accumulates.

So the subject of this article is not a list of things people get wrong. It is the narrower question of which shortcuts are well calibrated for ordinary life and badly calibrated for money, and why money in particular is the place they come apart.

Why investing in particular defeats intuition

Intuition is not uniformly unreliable. It becomes genuinely expert in some fields and stays useless in others, and the difference is not the intelligence of the practitioner — it is a property of the environment they practise in.

Kahneman and Klein set out the conditions in Conditions for Intuitive Expertise: A Failure to Disagree (American Psychologist, 2009), a paper written jointly by a researcher sceptical of expert intuition and one who studies it working. They agreed on two requirements: the environment must be regular enough that valid cues exist at all, and the practitioner must get enough practice with feedback that is fast and unambiguous to learn those cues. Chess supplies both. Anaesthesiology supplies both.

Investing supplies neither, and it fails in three distinct ways at once.

The feedback arrives years late. A decision made today returns its verdict over a period long enough that you will have made a hundred other decisions in between, and long enough that you will no longer remember what you actually believed when you made it.

The feedback is confounded. A well-reasoned decision and a careless one can produce the same outcome, because the outcome is dominated by things neither of them controlled. Hogarth, Lejarraga and Soyer describe this in The Two Settings of Kind and Wicked Learning Environments (Current Directions in Psychological Science, 2015): a kind environment returns feedback that accurately reflects the quality of the judgement, and a wicked one returns feedback that misleads. Markets are wicked in exactly that technical sense. You are being graded, but not on the paper you wrote.

The stakes are real and immediate. The other two problems are about learning; this one is about the moment of decision. Loewenstein's Out of Control: Visceral Influences on Behavior (Organizational Behavior and Human Decision Processes, 1996) describes the gap between the two states: a person deciding calmly systematically underestimates how differently they will weigh the same options while frightened, hungry or in pain. The plan is made in one state and executed in another.

Take those three together and you have an environment that punishes intuition, refuses to teach it, and then applies pressure at the moment it is least reliable. That is not a description of the reader. It is a description of the problem the reader has been handed.

The asymmetry underneath most of the list

One finding does more explanatory work than the rest, so it is worth getting right rather than getting quickly.

Kahneman and Tversky's Prospect Theory: An Analysis of Decision under Risk (Econometrica, 1979) made two claims. The first is that people evaluate outcomes as changes from a reference point rather than as final states of wealth — the question the mind asks is not "how much do I have" but "how much more or less than before". The second is that the function is steeper on the loss side than on the gain side, so a decline of a given size registers more heavily than a rise of the same size.

Both claims are frequently overstated in retelling, so here is the honest version of the magnitude. Tversky and Kahneman's follow-up estimation (Advances in Prospect Theory, Journal of Risk and Uncertainty, 1992) put the median ratio at about 2.25 from laboratory gambles. That figure has been challenged — Gal and Rucker's The Loss of Loss Aversion (Journal of Consumer Psychology, 2018) argues the evidence for a general loss-over-gain asymmetry is weaker than the textbook version implies, and psychology's broader replication record since the Open Science Collaboration's 2015 reproducibility study in Science counsels caution about any single laboratory coefficient. Treat the direction as well supported and the size as contested.

The mechanism is what matters here anyway, and it survives the dispute. Because the reference point is a change from somewhere, the same holding can occupy two entirely different psychological positions.

Two people each hold ₹10 lakh of the same scheme, bought on different dates. Take one who reached ₹10 lakh from ₹14 lakh and one who reached it from ₹7 lakh — round figures chosen to make the arithmetic visible, not a claim about any scheme. The asset is identical and so are its prospects, because a portfolio has no memory of the path that produced it. One person is holding a loss of ₹4 lakh; the other is holding a gain of ₹3 lakh. Ask each what they intend to do and you will usually get different answers, and neither answer is about the scheme.

That is the whole idea in one comparison. The reference point is not information about the investment. It is information about your history, and it is sitting in the decision as though it were information about the investment.

Two consequences of the asymmetry are documented well enough to name, and both are treated at length in loss aversion rather than here. Shefrin and Statman called the first the disposition effect in The Disposition to Sell Winners Too Early and Ride Losers Too Long (Journal of Finance, 1985), and Odean measured it in retail brokerage records in Are Investors Reluctant to Realize Their Losses? (Journal of Finance, 1998), finding that gains were realised at a higher rate than losses. The mechanism is direct: selling at a loss converts a change on a screen into a settled fact, and the settled fact is what the loss side of the function weighs.

The second is subtler and belongs in the pillar because it generalises. Benartzi and Thaler's Myopic Loss Aversion and the Equity Premium Puzzle (Quarterly Journal of Economics, 1995) points out that how often you evaluate changes what you see, with no change to the underlying at all. A volatile holding looked at daily shows a decline on a large share of those occasions; the same holding looked at once a year is far less likely to have that single look land on a decline, and the twelve-month outcome is identical either way. Checking more often does not give you more information about the holding. It gives you more occasions on which the steep side of the function has something to weigh.

The main biases, with the mechanism in each

Names are only useful if a mechanism is attached to them, so each row below states what the shortcut is actually doing rather than what it is called. Most of them are ordinary competence pointed at the wrong problem.

Rows two and four have an article of their own in anchoring and confirmation bias, and row nine in herd mentality. Two of the entries also have unusually good literatures behind them if you want to go further. Nickerson's Confirmation Bias: A Ubiquitous Phenomenon in Many Guises (Review of General Psychology, 1998) is the standard review of the fourth row. Fischhoff's Hindsight ≠ Foresight (Journal of Experimental Psychology: Human Perception and Performance, 1975) is the original demonstration of the fifth, and it is the one with the sharpest consequence for an investor: if your memory of what you expected moves toward what happened, then your record of your own judgement is being edited by the outcome, and you are learning from a corrupted file.

Notice how many rows are the same underlying move. Anchoring, the disposition effect and mental accounting are all a reference point doing work it has no right to do — a purchase price, a round number, a label on a pot. A short list of mechanisms generates a long list of named biases, which is why memorising the names is a poor use of effort and recognising the moves is not.

Knowing the name does not remove the bias

Being told about a bias, in advance, in plain language, does not reliably stop you exhibiting it — which is the finding most writing on the subject skips. Pohl and Hell tested exactly this for hindsight in No reduction in hindsight bias after complete information and repeated testing (Organizational Behavior and Human Decision Processes, 1996): participants who were warned about the effect, shown their own earlier answers and put through the task repeatedly still showed it. The title is the result.

The mechanism is not mysterious once you look at where the shortcut sits in the sequence. These operate on how the situation is presented to you, not on what you conclude about it. By the time deliberate reasoning has an object to work on, the reference point is already set, the anchor is already in, the vivid example is already the one you are thinking with. You can audit a conclusion. You cannot audit an input you never experienced as an input.

Which produces a second-order problem with a name of its own. Pronin, Lin and Ross reported in The Bias Blind Spot (Personality and Social Psychology Bulletin, 2002) that people rate others as more susceptible to common biases than themselves. The asymmetry has a structural explanation worth more than the finding itself: with another person you observe only their conclusion, so a distorted one is visible; with yourself you observe your reasoning, which feels sound, because the distortion entered before the reasoning began and left no trace in it.

Kahneman was explicit in Thinking, Fast and Slow that decades of studying these effects had not much improved his own intuitive judgements, only his ability to spot the errors in other people's. That is a striking admission from the person with the most reason to claim otherwise, and it is the honest baseline for this whole subject.

So the useful question is not how to think better in the moment. It is what to arrange when the moment is not happening.

Why a rule made in advance does what judgement cannot

If a bias operates before deliberation and survives being warned about, the only lever left is when the decision gets made. A rule set in advance moves the decision to a point where the reference point is abstract, nothing is currently falling, and no outcome is settling into a fact.

There is real research behind the specific form the rule should take. Gollwitzer's Implementation Intentions: Strong Effects of Simple Plans (American Psychologist, 1999) distinguishes a goal intention — what you intend to achieve — from an implementation intention, which specifies a trigger and the action attached to it in advance. The meta-analysis Gollwitzer conducted with Sheeran in 2006 across a large body of studies found the specified-trigger form outperformed the goal-only form on goal attainment. The effect is not confined to money and was not studied there first.

The distinction is precise, and it is the whole practical content of this article. "Stay disciplined during a fall" is a goal intention and does no work, because it names no trigger and no action, and every difficult moment is a fresh negotiation. "If the allocation drifts past this band, rebalance to target on the next working day" is an implementation intention. It has already decided.

Thaler and Benartzi's Save More Tomorrow (Journal of Political Economy, 2004) is the clearest applied demonstration: participants committed in advance to raising their savings rate out of future pay rises, and take-up was far higher than for an immediate increase. Nothing about the arithmetic differs. What differs is that the commitment was made about a sum that was not yet in hand, which is a state in which the loss side of the function has nothing to weigh.

Now the cost, because this is not free and articles that present it as free are selling something. A rule written in advance is written with less information than the moment will have. It will sometimes fire when a thinking person would not have acted, and sit still when one would have. You are trading a small, recurring, known cost for protection against a large, occasional, unknown one — and if you are unwilling to pay the small cost on the days the rule looks foolish, you do not have the protection either.

Which names the failure mode this whole approach dies of. A rule you can revise in the moment is not a rule, it is a preference with paperwork. The revision will always arrive with an excellent argument attached, because the same mechanism that makes the moment hard also supplies reasons. The test of whether a rule exists is not whether it is written down. It is whether it has ever cost you anything and survived.

Four places this shows up concretely elsewhere in the library: a band and a date in rebalancing, a purpose and horizon fixed before the money moves in goal-based investing, the mechanical effect of stopping instalments during a fall in SIPs during market crashes, and what stress specifically does to a decision in emotional investing.

What it costs, and how much of that is actually measured

The direction is well established and the magnitude is disputed: the research shows consistently that more activity leaves less money, and nobody can put a defensible rupee figure on what any individual bias costs.

The cleanest single result is Barber and Odean's Trading Is Hazardous to Your Wealth (Journal of Finance, 2000), which examined a large sample of retail brokerage accounts and found that the households that traded most earned the lowest net returns. The important detail is where the gap came from: gross performance across the groups was far closer than net performance, so the difference was substantially the cost of the activity rather than bad selection. The trading was not mainly picking worse; it was mainly paying more.

Closer to home, SEBI has published studies of individual participants in the equity derivatives segment reporting that most of those studied lost money. That finding measures outcomes, not causes. A segment that is zero-sum before costs produces losses for most participants without any bias being involved, so the result sets an outer bound on what behaviour could be explaining rather than measuring it. The direction is not in serious dispute; the figures are periodically reissued for new periods, so they are not quoted here.

The broader claim — that investors as a group earn less than the funds they hold, because of when they buy and sell — is real in direction and genuinely contested in size. It is computed by weighting a scheme's returns by the money actually present at each point and comparing that with the published return, and different providers using different methods reach different answers. It has its own article: the behaviour gap is treated there in full, including what the measurement does and does not capture. Read the direction as established and any specific number as one method's answer.

Past performance is not indicative of future results, and none of the historical findings above should be read as describing what any particular period will do.

Four tells you can check in a decision you have already made

You cannot inspect a bias while it is operating — that was the point of the section on why warnings fail. What you can inspect is the finished decision, looking for the fingerprints. None of these is proof; each one is a reason to look again.

The decision flips when the reference point changes. The test is a single question asked in the other direction: holding this today, at today's price, with today's information — would this be the position I would construct from cash? If the answer is no while the holding stays, the purchase price is doing the deciding, and the purchase price is a fact about your past, not about the asset.

The reason arrived after the conclusion. Recall the order of events. If you can remember reaching the decision and then assembling the case for it, the case is a justification rather than a cause — which does not make it wrong, but does mean it was never tested.

The evidence gathered was one-sided. Ask what you read before acting, and what the strongest argument against was. If you cannot state the opposing case in a form its holder would accept, you have not encountered it, and Nickerson's review says that is the default rather than the exception.

Your confidence outran the number under it. If you hold a view strongly but cannot state the figure it rests on — the cost, the worst historical window, the proportion of the portfolio — then confidence is being supplied by familiarity rather than by evidence. That is the one tell with a mechanical fix, because the number either exists or it does not.

None of these tells you what to do about what you find. They tell you which decisions in your own record are worth a second reading, which is a different and more modest claim — and the only one this field can actually support.

Checking the numbers the argument rests on

Three of the four tells above end at a number, and the reason biases operate so freely on portfolio decisions is that the number is usually missing at the moment the decision is made.

FNOTrader's Mutual Funds app runs lumpsum and instalment schedules against the full AMFI NAV history — around 34 million NAV rows — and reports invested amount against value, the annualised return on each (compound annual growth for a lumpsum; for instalments, the internal rate of return for cashflows that land on irregular dates — XIRR), the maximum drawdown along the way, and the distribution of outcomes across every available start date rather than one. That last column is the one relevant here: the spread of results by start date is a direct measure of how much of any single outcome was the decision and how much was the date, which is precisely the confound that stops the environment teaching you anything.

Past performance is not indicative of future results, and no historical distribution should be read as a range that will repeat.

Common questions

Continue reading

More in Behavioural Finance · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article