Mental Accounting and the House-Money Effect

The mind keeps accounts, and the market does not

Mental accounting is the habit of treating money differently according to where it came from, what it is for, or how it is labelled. It is not a malfunction — it is how budgets work. Separating rent from savings from a holiday fund is a self-control device, and it helps people who would otherwise spend everything. The trouble begins when the same pot of capital is split into buckets that are allowed different rules, because the market prices a portfolio as one number and the buckets are notional. The best-known form is the house-money effect, demonstrated in experiments where people who had just been given a stake bet more of it than people betting their own. A gain seems to arrive with a cushion, so the risk aversion that normally applies to the first dollar relaxes for the second. In a brokerage account the effect is a good run: the trader is up, the account has grown, and the increase in size is now authorised by a feeling rather than by a plan. The trade that follows is not risk-free behaviour with surplus money. It is full-size risk on the same equity, taken at the exact moment the equity is highest — and the gains it draws down are money the account already had. There is an arithmetic version of the same error in the dividend. A $2,000 dividend and the sale of $2,000 of stock leave the holder in an identical economic position: the position is worth $2,000 less in the first case and $2,000 less in the second. One is labelled income and the other a decision, and the tax treatment can genuinely differ — which is a reason to choose between them deliberately, not a reason to believe one of them is spending nothing. The habit of spending the dividend while refusing to sell shares is exactly the behaviour a mental account produces and the market does not price. The same $2,000, three labels — Held as a floating gain on a $50,000 account: Already owned — account equity is $52,000 ← · Relabelled “house money” and half staked: $1,000 at risk on the same account — 2% instead of the 1% unit ← · Paid as a $2,000 dividend and spent: The position is worth $2,000 less; economically the same as selling $2,000 of shares The label changes what the trader permits; it never changes what the account is worth. That is the whole test: if two descriptions of the same fact lead to different position sizes, the label is doing the deciding.

Where the bucketing helps and where it costs

The useful distinction is whether a bucket changes behaviour on capital that shares a risk budget. A bucket that enforces a saving rate passes money out of the trading account and into another asset — it changes the allocation, so it is doing real work. A bucket that lives inside one account and lets the same dollars take different risk depending on their history changes only the risk taken. “I will only risk the profits” sounds conservative and is not: it is a rule that increases risk after a gain and decreases it after a loss, which is the opposite of what a stable risk unit does, and it means the two most extreme moments in an account’s life — the peak and the drawdown — are the moments the position size is least governed. The portfolio view is the correction, and it is one number. Mark the whole account at market, treat unrealised gains as owned, and define risk as a fraction of that number, so a gain raises the size by the amount the plan says and a loss lowers it the same way. That converts the label question into an allocation question, which is answerable: if the trader wants to take more risk after a gain, the honest version is a planned rule for risk after equity changes, written in advance — not a feeling about which money it came from. The same test resolves the dividend. The decision is not “keep the income or sell shares”; it is “hold this position at its current size, or hold it smaller and use the cash”. Both are available at every moment whether or not a dividend was paid, and framing the choice as income versus capital gains hides the fact that a cash payment is always a reduction of the position unless it is reinvested. • Mark the whole account at market once, and take one equity number into the decision. • A floating gain is owned; there is no separate pile called profit. • Set risk from the plan, so a gain raises size by design rather than by mood. • A dividend and a sale of the same value are one decision with two tax treatments, not income versus nothing. • Keep the buckets that enforce a real transfer between assets, and drop the ones that just permit different risk on the same capital. A planned increase in size after a gain is legitimate and can be rational — but it is a rule written before the gain, tested against drawdowns, and identical for a trader who inherited the money last week. The test is whether the same equity change would produce the same size change without the story attached to it.

When the buckets are the discipline

Mental accounting is usually taught as a bias, and it is one — separate pots for money that is perfectly fungible, and a willingness to gamble the “winnings” that you would never extend to the “principal”. But the same reflex can be turned into an instrument. Behavioural research on commitment devices finds that people reliably save more and spend less when money is ear-marked before it arrives than when it is left pooled, because the label does the work that willpower would otherwise have to do every day. The mechanism is the mirror image of the house-money mistake. Gambling the winnings feels harmless because the winnings are not “real money” yet; saving into a labelled account works because the label makes the money *feel* less available. The same mental partition that lets a trader take a reckless bet on a profit can make a household keep a buffer it would otherwise spend. Whether the partition helps or hurts depends entirely on which pot the decision belongs to, not on the partitioning itself. The practical rule is to partition deliberately and in one direction. Fix labels for money whose purpose is settled — an emergency buffer, a tax reserve, a fixed yearly obligation — and then treat the trading account as exactly one pool, sized in *percent of account* rather than in “money I won today”. The bucket is a tool for commitment; it becomes a bias the moment it is used to justify a bet. A useful test: is the label deciding *whether* to act, or only *how much*? Labels that gate the decision are commitment devices; labels that inflate the decision are the bias.

The buckets that help

Mental accounting has a poor reputation in this subject, and there is a version of it that is genuinely productive. Instead of one risk tolerance applied to a whole portfolio, the money is separated into purpose-defined buckets, each with its own horizon and its own constraint. The structure then answers a question that a single blended portfolio cannot: too risky for *which* job? The reason that matters is that a global risk number is an average, and no individual decision can be tested against an average. Ten percent volatility does not tell you whether to own a particular position, how much of it, or when to sell it. A bucket has a job — pay the essential bills, fund a known expense in eight years, or compound for as long as possible — and each job supplies its own test. The standard construction has three parts. A **floor**, holding essentially safe assets or a guaranteed income stream sized to cover essential spending, whose only job is to be there. A **horizon bucket** for money with a known date on it, whose asset mix de-risks as that date approaches, because a short horizon and a volatile asset are incompatible regardless of the expected return. And a **growth bucket** holding only what neither of the first two needs, which can then be run aggressively without putting the essentials at risk. The whole point is that the aggressive bucket can be genuinely aggressive precisely because the floor exists, which a single blended portfolio cannot achieve — it mixes the roles in every position and ends up timid everywhere and exposed in the wrong place. The mechanism that makes it worth doing is behavioural rather than arithmetic. In a bad year the floor is roughly intact, and it is that fact — the visible presence of untouched money doing its job — that prevents the whole portfolio from being sold at the bottom. The return that comes from not selling at the bottom does not appear in any performance figure and is usually larger than the differences the buckets were built to optimise. The failure mode is the one this lesson has already established, wearing a helpful costume: buckets that hide total exposure. Money that is fungible stays fungible, and if the buckets are never totalled then the same risk can sit in three of them and the aggregate is invisible. The structure is a decision-making aid, not a substitute for the aggregate — which means the total exposure should be computed at least as often as the buckets are consulted. The honest limit is that a bucket framework can also be used to rationalise an oversized position, by pointing at a different bucket as the reason for comfort. Where the structure earns its place is when the essentials are genuinely protected; where it costs is when it becomes an argument for risk rather than a constraint on it. • A purpose-defined bucket supplies a test that a portfolio-wide risk number cannot. • Floor, horizon and growth buckets do different jobs, so they can hold different risks. • The behavioural gain is that the intact floor is what stops a bottom-of-the-market sale. • Total the aggregate anyway: buckets are an aid, not a substitute for the sum. A quick audit of any bucket plan: for each bucket, what is its job, when does the money leave it, and what would make a holding inside it wrong? A bucket that cannot answer all three is a label rather than a structure.

Sources

• Mental accounting and its consequencesThaler (1999), “Mental Accounting Matters”, Journal of Behavioral Decision Making
• Gambling with house money and trying to break evenThaler & Johnson (1990), Management Science
• Framing, segregation and integration of outcomesKahneman & Tversky (1984), “Choices, Values and Frames”