How Behavioural Biases Reach Real Portfolio Decisions

Behavioural biases reach a portfolio through specific decisions: the position that gets sold, the position that gets held, how often the portfolio is traded, the part that gets looked at, and what a committee concludes from one year of numbers. Naming a bias changes nothing on its own. The change comes from a structural check placed on the decision the bias reaches.

Why is a bias a portfolio problem rather than a character problem?

A bias is not stupidity, it is not inexperience, and it is not something that stops operating once it has been read about. A bias is a regularity in how people decide under uncertainty, and it works perfectly well on people who can describe it accurately, name the paper it came from and spot it in somebody else that same morning. The people who run the invented Anantara Multi-Asset Portfolio are careful, senior and entirely capable of explaining every idea below, so a list of mistakes made by less careful people would teach nothing at all.

So the useful question is never whether someone is biased. Everyone is, all the time, and the answer carries no information. A decision is the only thing a check can be placed on, so the useful question is which decision the bias reaches. An attitude cannot be audited. A judgement cannot be re-run. But a sell order has a moment, a review has an agenda, a trading rate has a number, and a conclusion has minutes. Each of those is a place where something can be written down that does not depend on anybody feeling calm that week.

Think of a household that keeps paying a monthly subscription it stopped using nine months ago. Nobody in that household is foolish. Cancelling is the moment the money already spent turns into money wasted, and there is always a reason to leave it one more month. The decision that carries the bias is not whether the service is worth having; it is the single act of cancelling. Put a check on that act, such as a standing rule that every recurring payment is re-approved each April, and the behaviour changes without anybody becoming a different person. Portfolios work the same way, at a larger scale and with more at stake.

What makes selling a fallen position hard?

Loss aversionThe tendency for a loss of a given size to weigh more heavily on a decision than a gain of the same size. The finding belongs to Daniel Kahneman and Amos Tversky, 1979. is the finding that a loss of a given size weighs more heavily than a gain of the same size, and it belongs to Daniel Kahneman and Amos Tversky, 1979. Read as a sentence about feelings it is mildly interesting. Read as a sentence about portfolios it is sharp. The finding names exactly which decision it will reach.

The decision is the sell decision on a position that has fallen. While the position is held, the loss is a number on a valuation report and it is still, in some sense, open. The moment it is sold, the loss stops being an unrealised figure and becomes a settled one that appears in a record, gets discussed in a meeting and belongs to whoever bought it. Nothing about the position changes when it is sold. The change is that the loss becomes finished.

So the position stays. Then a second one stays. Then a third. Repeated across a few dozen decisions, a portfolio quietly accumulates holdings that nobody in the room would choose at today's price, and the shape of the whole was never decided by anybody. The last clause is the portfolio consequence, and it is why loss aversion is a portfolio problem rather than a matter of personality. No single decision to hold was unreasonable. The set that results was never a decision at all.

Notice which decision it does not reach. A purchase has no settled loss attached to it, so buying is barely affected. The sizing rule is untouched. Loss aversion reaches one act, at one moment, and that is precisely why a check can be written for it.

Why are selling winners and holding losers one mechanism?

The disposition effectThe observed tendency to sell holdings that have risen and to keep holdings that have fallen. The name and the finding belong to Hersh Shefrin and Meir Statman, 1985. is the tendency to sell what has risen and keep what has fallen, and it belongs to Hersh Shefrin and Meir Statman, 1985. The disposition effect is tempting to read as two separate habits, one about greed and one about hope. It is not. Both branches are about settling a number rather than about the holdings themselves, so one mechanism is running in both directions. Selling a winner settles a gain, and settling a gain is pleasant and therefore easy. Selling a loser settles a loss, and settling a loss is unpleasant and therefore postponed.

Now run that rule for a few years and look at what is left. Every position that rose has had a reason to leave. Every position that fell has had a reason to stay. The remaining set is not a random sample of what was bought; it is systematically weighted toward the things that went the wrong way. And nobody wrote that down as a plan. Each individual decision had a defensible sentence attached to it at the time.

The disposition effect is the portfolio version of a wardrobe. The favourite shirt is worn until it wears out and then thrown away. Throwing away the shirt that was never liked means admitting the purchase was a mistake, so it stays in the cupboard for eleven years. After a decade the cupboard is full of clothes nobody wears, and at no point did anybody decide to build such a cupboard.

What happens when each part is reviewed on its own?

Mental accountingTreating money in separate mental pots as if the pots were unrelated problems, so decisions are taken pot by pot rather than across the total. The idea belongs to Richard Thaler, 1985. is the habit of treating separate pots as separate problems, and it belongs to Richard Thaler, 1985. In a household it is the person who keeps a savings account earning very little and a card balance costing a great deal. The savings are labelled for the wedding and the card is filed under this month. Both statements are true and the combination still costs money.

In a portfolio the same habit looks like organisation, and looking organised is exactly what makes it dangerous. The Anantara mandate has three parts: equity at Rs 3,00,00,00,000/-, or 60.0 per cent of the stated policy weights, fixed income at Rs 1,50,00,00,000/-, or 30.0 per cent, and cash at Rs 50,00,00,000/-, or 10.0 per cent. The three stated figures are the policy weights, and the actual weights drift between rebalancing dates. Reviewing each part with the people who run it, on its own agenda, against its own reference, is entirely natural. Each review can go well. Each set of minutes can be accurate.

And the moment the parts are judged separately, the properties that only the whole carries stop being anybody's responsibility. The whole-level properties are computable and they are not small. On the mandate's own stated assumptions, the expected return of the whole is 0.6 times 12.0 plus 0.3 times 7.5 plus 0.1 times 6.0, or 10.05 per cent. The volatility of the whole works out at 11.20 per cent, against a weighted average of the three individual volatilities of 12.35 per cent. The 1.15 point difference is the diversification, and it exists only because the correlation between equity and fixed income is taken as 0.20 rather than 1.00. Diversification is a property of the combination. No sleeve review computes it, and no sleeve owner can be held to it.

Mental accounting is the most damaging of these biases for exactly that reason. Loss aversion damages a set of positions. Mental accounting removes an entire level of management from the agenda, and it does so while looking tidy.

How does a firmly held view turn into a cost?

OverconfidenceHolding a view more firmly than the evidence behind it supports. Overconfidence typically shows up as more action rather than as a stated opinion. is holding a view more firmly than the evidence behind it supports. On its own that is a statement about belief, and beliefs are not billable. But a strongly held view in a portfolio does not stay a belief for long. The view becomes an instruction, the instruction becomes a trade, and trades in aggregate are turnoverThe share of a portfolio replaced over a stated period. Turnover of 34 per cent over a year means roughly a third of the portfolio was bought or sold within it..

The Anantara portfolio recorded turnover of 34 per cent over its one stated twelve month period. On Rs 5,00,00,00,000/- that is about Rs 1,70,00,00,000/- of the portfolio replaced across the year. Now put a trading cost on it. If replacing value costs c per cent of the value replaced, then the cost to the whole portfolio is 0.34 times c percentage points. At c equal to 0.5 that is 0.17 points; at c equal to 1.0 it is 0.34 points. The record contains no cost level of its own. Put your own figure in place of c and the cost follows from the same two lines.

Here is the part that matters, and it is the part usually skipped. Ask which base the cost is charged against. Set 0.17 points against the 14.2 per cent headline return and it is about 1.2 per cent of it, a rounding difference. The 1.112 point gross residual is what the portfolio produced beyond the market exposure it carried; set the same 0.17 points against that residual and it is 15.3 per cent of it. Same rupees. Same year. Two bases, and one of them makes the number invisible.

None of the four rows is a claim about what trading actually costs anybody. The four rows are arithmetic consequences of one recorded turnover figure. At any plausible cost the number is material against the residual and immaterial against the headline, so the choice of base decides whether the cost is ever discussed at all.

What did the stated year actually read, and from where?

The cleanest demonstration needs no invented figure at all. Over its one stated twelve month period the Anantara portfolio returned 14.2 per cent, against the composite benchmark's 12.6 per cent. Inside those same twelve months the portfolio fell 9.7 per cent from its highest point to its lowest before recovering, against the benchmark's 8.1 per cent over the same peak to trough measurement. Both figures are recorded. Both describe the same portfolio and the same year. And they read as two completely different experiences.

From the start of the year, the endowment gained 14.2 per cent. From the peak, the endowment was down 9.7 per cent at the trough. Nothing about the portfolio changed between those two readings, and the only thing that moved was the reference pointThe level a result is compared against. A gain or a loss is not a property of an outcome on its own; it exists only relative to whichever starting level is chosen.. A gain and a loss are defined against a reference rather than being properties of the outcome itself, and that is the point prospect theoryThe account of how people evaluate risky outcomes as gains and losses relative to a reference point rather than as final states of wealth. The account belongs to Daniel Kahneman and Amos Tversky, 1979. makes, and the name belongs to Daniel Kahneman and Amos Tversky, 1979.

One honest limitation. The record does not contain a within-year path. The record has the two readings and nothing between them. A line drawn between them would be an invented series wearing the record's clothes, so only the two recorded readings can be stated. The drawing below and the control beside it hold to that. With the reference away from either recorded point there is nothing to read, and the reading goes blank.

A drawdownThe fall from a high point to a subsequent low point, measured inside a stated window. A different window gives a different figure. The window is therefore quoted every time. is always measured peak to trough inside a stated window, and a different window gives a different number. Depending on the window is not a weakness of the measure. Quoting the window every single time follows from it, and the reference point demands the same discipline.

How much of the 1.6 points could any decision have produced?

AnchoringFixing a judgement to whichever number was seen first, so later reasoning adjusts away from that number instead of starting fresh. fixes a judgement to whatever number was seen first. In a review pack the first number is almost always the headline return printed at the top of the first sheet. RecencyWeighting the most recent period more heavily than its length justifies, so a single stretch of results is read as though it described the approach behind them. weights the most recent stretch as though it described the approach that produced it. Both of these reach the same decision, and it is not the trade. The decision they reach is the review. A review leaves minutes rather than a trade blotter, and nobody audits a conclusion, so anchoring and recency are the hardest of these biases to see.

So set the arithmetic against the conclusion. The portfolio returned 14.2 per cent and the composite benchmark returned 12.6 per cent, a gross excess of 1.6 percentage points. The portfolio ran a beta of 1.08 against that benchmark, and the risk-free rate for the same period was 6.5 per cent. The benchmark's own excess over the risk-free rate was 12.6 less 6.5, or 6.1 points. A portfolio carrying 1.08 times that exposure would be expected to return 6.5 plus 1.08 times 6.1, or 13.088 per cent.

Split the 1.6 points on that. The exposure part is 13.088 less 12.6, or 0.488 points. The residual returnThe part of a return left after the part explained by the exposure carried has been removed. The residual is what remains once the market's own contribution at the portfolio's beta is taken out. is 14.2 less 13.088, or 1.112 points. The two parts sum back to 1.600, and that sum is the check. In shares, 1.112 of 1.6 is 69.5 per cent and 0.488 of 1.6 is 30.5 per cent. A decomposition built from rounded pieces stops reconciling by the third line, so the unrounded parts are used throughout.

One caution about which split this is. There is a second decomposition of the same 1.6 points elsewhere in this record, an allocation effect against a selection effect. The allocation and selection split asks where the excess came from. The exposure split asks how much of it was simply carrying more exposure. The two answer different questions on different bases, both sum to 1.6, and mixing a term from one with a term from the other produces a sentence that means nothing. The split set out above is the exposure split.

Why does familiarity walk through a per holding limit?

People hold more of what they know. Holding more of what one knows is familiarity biasThe tendency to hold more of what one already knows well. The habit quietly concentrates a portfolio in whatever those known things have in common., and it sounds harmless. Knowing something well is generally an advantage. The problem is not the knowing. The problem is that the things a particular group of people knows well tend to share conditions: one city, one industry, one supply route, one policy regime, one customer base.

Here is the everyday version. A household where both earners work for the same employer is not diversified by having two salaries. The household has one salary paid twice. A street of ten shops inside one office building looks like ten businesses and is really one business. The same lift traffic decides all ten. Nothing about counting the shops reveals it.

Now the portfolio version, using the recorded figures. The Anantara mandate has a stated limit that no single holding may exceed 5 per cent of the portfolio. The largest holding sits at 4.6 per cent of the portfolio, comfortably inside it. Note the base carefully: measured against the equity sleeve rather than the portfolio, that same holding is 7.7 per cent. Neither number is wrong and they answer different questions, so the base is stated every time. Meanwhile the ten largest holdings together come to Rs 1,55,00,00,000/-, or 31.0 per cent of the portfolio and 51.7 per cent of the equity sleeve.

A limit written on each name does nothing whatever about what the names have in common, so familiarity walks straight through a concentration constraint without ever breaching it. Every individual test passes. The thing the test was meant to protect against is not measured anywhere. What those ten holdings actually share is not in the record at all. Say that out loud rather than assuming the answer is reassuring.

Which check sits on which decision?

Every check below is a change to a process rather than an effort of will, and the distinction is what makes a check hold. A resolution to be more objective survives until the first bad month. A standing agenda item survives because it is on the agenda whether anybody feels objective or not.

For the sell decision, the check is a question asked in the same words every time: would this position be bought today, at today's price, in this size? The question works because it removes the purchase price from the decision, and the purchase price is the only thing that makes a settled loss feel different from an open one. For the review, the check is placement. The numbers that only the whole carries go at the front of the pack, before any part is opened, and they are read while attention is highest rather than after everyone has formed a view.

For trading, the check is pricing it. Nobody argues with a percentage that has no base attached, so turnover never converted into rupees and set against the residual is never argued about at all. For the reading of a result, the check is announcing the period and the reference point before the first figure is read aloud, then splitting the excess before interpreting it. For familiarity, the position sizes already pass, so the check is measuring the shared conditions instead.

None of these checks requires anybody to stop being human, and that is exactly why each of them survives contact with a bad month.