> ## Documentation Index
> Fetch the complete documentation index at: https://meta.niceshare.site/llms.txt
> Use this file to discover all available pages before exploring further.
# House Money Effect
> House Money Effect is a greater willingness to bet after a gain, when a loss can be read as a smaller gain. In one class, 77 percent took that bet.
**Category**: Effects
**Type**: A greater willingness to take a risk after a gain, when the loss can be coded as a cut in that gain
**Origin**: Richard H. Thaler and Eric J. Johnson, *Management Science* (June 1990)
**Also known as**: Playing with the house money
**Quick Answer** — House Money Effect is a greater willingness to accept a risk after a gain, when a loss smaller than that gain can be coded as a cut in the gain. Richard H. Thaler and Eric J. Johnson gave the pattern this name in 1990. In a Cornell class, 77 percent of 95 people took a 50-50 bet of $4.50 after a $15 gain. Forty-four percent of 111 people took that bet when the wording did not start from the gain.
## What is House Money Effect?
House Money Effect is a greater willingness to accept a risky bet after a prior gain, when the possible loss is smaller than that gain and is coded as a cut in the gain.
> After a gain, subsequent losses that are smaller than the original gain can be integrated with the prior gain, mitigating the influence of loss aversion and facilitating risk-seeking.
That sentence is Thaler and Johnson in 1990. The phrase they borrow from the casino floor is "playing with the house money." Until the winnings are gone, a loss is entered as a smaller gain. The casino's phrase is the feeling of gambling while ahead. It is not a count of casino patrons.
The account is the mechanism. A later loss that still fits inside the gain is cancelled against the gain, so the worst line remains a gain. [Loss Aversion](/effects/loss-aversion) then has less room to bite, because the pain of a loss is not the line being read. [Mental Accounting](/effects/mental-accounting) is the label on that account: which pile the money is in, and when the pile is closed. [Prospect Theory](/models/prospect-theory) is the value curve behind the coding. Gains flatten as they grow, and a loss weighs more than a matched gain. This page does not carry the 1979 gamble counts.
The same final money, written without the line "you are ahead," is a different frame. [Framing Effect](/effects/framing-effect) is a choice that changes when the wording changes and the final amounts do not. In the class below, that change of wording moved the share from 77 percent to 44 percent. A prior loss is a second pattern in the same paper. They call it a break-even effect when the next bet can return the account to the reference the person started from. [Sunk Cost Fallacy](/effects/sunk-cost-fallacy) is money already committed to a course that is going badly. The pattern on this page is a gain that has not yet been given back.
### House Money Effect in 3 Depths
* **Beginner**: A small win is still sitting in front of you. A bet that can only give part of it back feels easier than the same bet written as a loss of money you already had.
* **Practitioner**: After a gain, write the next bet twice. Once, the loss only shrinks the gain. Once, the same loss starts from zero. Accept it only if you still want it in the second writing.
* **Advanced**: The 77 percent and the 44 percent are two wordings of one pair of final amounts, in one economics class. A bond pit in 1998 moved the other way after morning losses. Neither figure is a rate for a casino, a household, or a trading desk.
## Origin
The name and the counts below are from Richard H. Thaler and Eric J. Johnson, "Gambling with the House Money and Trying to Break Even: The Effects of Prior Outcomes on Risky Choice," *Management Science*, volume 36, number 6 (June 1990), pages 643–660. Thaler was at the Johnson Graduate School of Management, Cornell University. Johnson was at the Wharton School, University of Pennsylvania. The abstract names two patterns. The house money effect is increased risk seeking in the presence of a prior gain. Break-even effects are a separate claim: after prior losses, outcomes that offer a chance to break even are especially attractive.
The paper opens on scenes, not on counts. A person at a convention in Las Vegas puts a quarter in a slot machine and wins $100. Another person has $100 stolen from a wallet. A manager's division is down \$10 million and must choose between two projects. The authors use the scenes to ask whether a prior gain or a prior loss changes the next risk. They are not results.
The editing rule they propose is one they call quasi-hedonic editing. In plain words, people do not automatically fold every earlier outcome into the choice in front of them. A later loss is integrated with a prior gain when the loss is smaller than that gain. A later loss after a prior loss is kept separate, unless the new bet offers a chance to get back to the original reference. The rule was their reading of the choices. They write that other accounts could fit the same table.
The choices are Experiment 4, run in a large introductory economics class at Cornell. Roughly 98 percent of the class agreed to take part for real money. People who declined did not take part at all. Each person made no more than eight choices. Only one choice would actually be played, and a choice with a large prior gain was more likely to be the one played. The chance of having to play a gamble in which the person could lose money was 0.04. A loss could be cleared by clerical work at $5 an hour. Two $100 bonuses were promised. One would go to one of the ten people selected to play a gamble out. The other would go to someone who had agreed to play and was not selected. One part of the class saw a two-stage wording, a gain or a loss already in hand and then a new bet. The table gives 95 people for that wording. The other part saw a one-stage wording, the same final amounts with no prior line. The table gives 111 people for that wording.
An earlier design had drawn only about a third of the people invited, and those who agreed to play for real money took the risky option at almost every chance. Experiment 4 was built so that almost the whole class would stay in. The 77 percent below is this later class. It is not that earlier third.
The same paper reports hypothetical questions, asked of other groups, at a larger stake. Those groups are not the economics class. Joshua D. Coval and Tyler Shumway later asked a different question on a trading floor. Their article is "Do Behavioral Biases Affect Prices?," *Journal of Finance*, volume 60, number 1 (February 2005), pages 1–34. Coval was at Harvard Business School. Shumway was at the University of Michigan Business School. That pit is a boundary on this page, not a second name for the effect.
## Key Points
Four readings keep the money fixed and move the account. A share in a classroom wording is not a share of bets placed in a casino. A morning on a trading floor was not assigned by the experimenter. A loss that can erase an earlier loss is a different problem from a loss that only nibbles a gain.
The two-stage house-money item started 95 people at a gain of $15. The bet was a 50 percent chance to win $4.50 and a 50 percent chance to lose $4.50, against keeping the $15. Winning ends at $19.50. Losing ends at $10.50. The loss fits inside the gain, so the account can code it as a smaller gain. Seventy-seven percent took the bet.
The other wording offered the same finals and did not start from a gain. A sure $15, or a 50 percent chance of $19.50 and a 50 percent chance of \$10.50. Forty-four percent of 111 people took the bet. The dollars at the end match the two-stage item. The prior gain is what the wording removes. The authors treat that gap as the house money effect in this class.
The same 95 people also faced a 50-50 bet of $2.25. After a loss of $2.25, winning the bet returns the account to zero. Sixty-nine percent took it. After a loss of $7.50, the same $2.25 bet cannot return the account to zero. Forty percent took it, so 60 percent chose the sure line and stayed down $7.50. The authors call the first of those rows a break-even effect. The one-stage wording of the $2.25 hole was taken by 87 percent of the 111. Getting back to the start is not the house money effect.
Thaler and Johnson write that managers of a profitable enterprise, flush with early success, will be more risk seeking. They also write that they knew of no empirical test of that claim. Coval and Shumway later split the 1998 trading day of locals in the Chicago Board of Trade Treasury bond futures pit. House money, as they define it for that pit, predicts more afternoon risk after morning profits. What they found was more afternoon risk after morning losses.
## Applications
Use the two writings of one bet as a check you can do before the stake changes. Keep a classroom share inside the wording that produced it. Keep a pit inside the year and the contract that were counted.
After a gain, write the downside twice before you accept it. One line subtracts the loss from the gain. The other line starts from the money you had before the gain. In the Cornell class, the first writing drew 77 percent of 95 people. The second writing drew 44 percent of 111 people. Take the bet only if the second writing still looks right.
Integration, in the 1990 rule, is for a later loss smaller than the gain already in the account. A $4.50 loss after $15 still leaves \$10.50. If the worst case crosses back through zero, the loss no longer sits entirely inside the gain. Stop and price that bet as a loss, not as a smaller win.
A cash gift and the rent can sit in different piles. Before the gift pays for a risk you would refuse from the rent, name the pile and close it on paper. [Mental Accounting](/effects/mental-accounting) is that label. The class measured a described gain of \$15. It did not measure birthday money.
On a desk, write afternoon risk after mornings that made money and after mornings that lost it. The 1990 paper left that test open for managers. In the 1998 bond pit, the higher chance of above-average afternoon risk sat on morning losses. A morning gain is not, by itself, permission to raise the stake.
## Case Study
Richard H. Thaler and Eric J. Johnson asked a large introductory economics class at Cornell to make the choices for real money. The report is *Management Science*, volume 36, number 6 (June 1990), pages 643–660. Roughly 98 percent of the class agreed. The published table puts 95 people on the two-stage wording and 111 on the one-stage wording. This page does not add those two figures into a class size. The paper says "roughly 98 percent" and does not print the number enrolled.
Each person saw more than one starting amount. At most eight choices were offered, and only one would be played. Problems with a large prior gain were more likely to be selected for play. The chance that a person would have to play a gamble in which a loss of money was possible was 0.04. Someone who finished down could work the debt off at $5 an hour of clerical work. Two $100 bonuses were offered: one among the ten people chosen to play a gamble, and one among people who agreed and were not chosen.
The house-money problem put the two-stage group up $15. They could keep it, or take a fair bet of $4.50. Seventy-seven percent of the 95 took the bet. A win would finish at $19.50. A loss would finish at $10.50, still ahead of where the problem started. The one-stage group saw a sure $15 against a 50 percent chance of $19.50 and a 50 percent chance of \$10.50. Forty-four percent of the 111 took the bet. The ending amounts are the same pair. The two-stage sentence includes a gain already received. The one-stage sentence does not.
The same 95 people answer the loss problems, so the 77 percent, the 69 percent, and the 40 percent are three problems, not three samples. After a $2.25 loss, 69 percent took a 50-50 bet of $2.25, which can restore zero. After a $7.50 loss, 40 percent took that same $2.25 bet, which cannot restore zero. The lesson of the 77 percent is a prior gain, kept visible, on a loss that cannot wipe the gain out.
Other groups, answering hypothetical questions, saw a $30 gain and a 50-50 bet of $9. Two groups of Cornell MBA students took the two-stage bet: 70 percent of 44, and 82 percent of 75. Two further MBA groups saw the one-stage finals, a sure $30 against a 50 percent chance of $39 and a 50 percent chance of \$21. They took the bet at 43 percent of 46 and 44 percent of 46. Those four groups are not the economics class, and they are not one pooled sample. The paper says the real-money class largely repeated the pattern of the hypothetical questions. An earlier real-money attempt, with only about a third participating, is not included in these shares. Those volunteers were risk seeking on nearly every item.
The lesson is a wording inside one class, plus the same shape in separate hypothetical groups at a larger stake. It is not a record of chips, of household gifts, or of positions a firm actually took. The authors write that a simple change of frame can reverse a general claim about who seeks risk.
## Boundaries and Failure Modes
House Money Effect is the wrong name in three retellings. A classroom share becomes a casino habit. A bet that gets a person back to zero takes this page's name. A morning loss on a trading floor is told as a morning gain.
The first boundary is the class. The 77 percent had a described gain of $15 and a described bet of $4.50. The chance of actually playing a gamble that could lose money was 0.04. A loss could be worked off at $5 an hour. Two $100 bonuses were on offer, and the problem with the large gain was more likely to be the one played. The paper does not print a gender split for the 77 percent. It does not print a test that the 77 and the 44 differ by a stated margin of error. The share is the share who chose the bet on that row.
The second boundary is the other pattern in the same table. Break-even is a chance, after a loss, to return to the original reference. In the two-stage group, that chance drew 69 percent when the hole was $2.25, and the bet that could not fill a $7.50 hole drew 40 percent. Thaler and Johnson also point to a lower trading volume in shares that have fallen, citing Hersh Shefrin and Meir Statman (1985) among others. This page does not carry those volume counts. A reluctance to close a losing account is a different pattern from a greater willingness to risk a gain. That sell-and-hold pattern is the [Disposition Effect](/effects/disposition-effect).
The third boundary is the pit. Coval and Shumway use the audit trail of the Chicago Board of Trade Treasury bond futures pit for all of 1998, obtained from the Commodity Futures Trading Commission. The file has 1,082 traders. They keep 426 locals. Each of those locals made at least 1,500 trades for a personal account and traded bond futures on at least 100 days. The summary they print covers 236 full trading days in 1998. They treat each local as starting the day with no position. They note that the profit figure is noisy if a position is closed in an evening session, hedged with options, or passed through another local.
In that sample, a trader with morning losses had a 31.2 percent chance of taking above-average afternoon risk. A trader with a morning profit had a 27.0 percent chance. The paper says the losing trader is 15.5 percent more likely to take above-average afternoon risk. It also says losing traders are 17.4 percent more likely to place an above-average number of afternoon trades, and 9.5 percent more likely to trade at above-average size. A pooled logit, with the other variables set at their means, moves the chance of above-average afternoon risk from 26.9 percent to 31.3 percent. The paper calls that slightly more than 16 percent more likely. The 31.2 and 27.0 pair is the introduction's summary. The 26.9 and 31.3 pair is the logit. They are not one pair of numbers. The regression tables use 82,595 local-days. This page does not assign the 31.2 percent to that head count.
The authors' reading is loss aversion at a one-day horizon: more afternoon risk after a morning loss, in order to recover. House money, self-attribution of the morning profit, and reading the morning as a sign of the afternoon would have predicted the opposite sign. They also write that both a house-money pattern and loss aversion could be present, and that loss aversion is easier to see if the gain required to trigger house money sits high. A one-day score can miss an account that closes on another day. Morning profit was not assigned. The 31.2 percent is not a finding that every desk adds risk after a loss, and it is not a finding that the Cornell 77 percent was wrong inside that class.
## Common Misconceptions
Three retellings travel with the casino phrase. Any win is treated as a license for the next bet. A chase back to zero is given this page's name. The class share is treated as a pit.
The 77 percent is one row. The gain was $15, the bet was a fair $4.50, and the wording kept the gain in view. The same final amounts, without that opening gain, drew 44 percent of a different half of the class. The 1990 rule integrates a later loss only when the loss is smaller than the gain. A gain that the wording never mentions is not this row.
Getting even is the break-even row. After a $2.25 loss, 69 percent of the 95 took a bet that could restore zero. After a $7.50 loss, 40 percent took a bet of the same size, which could not restore zero. Thaler and Johnson name that return to the original reference separately from the house money effect. A person who is down and reaches for zero is in the second pattern.
The 77 percent is a described bet in one Cornell class. In the 1998 Treasury bond pit, the higher chance of above-average afternoon risk followed morning losses: 31.2 percent, against 27.0 percent after a morning profit. The logit reading is 26.9 percent to 31.3 percent. Those pairs are the pit's. They do not replace the class, and the class does not replace the pit.
## Related Concepts
House Money Effect sits beside the label on a pile of money, the extra weight of a loss, and the value curve. It also sits beside a wording that changes a choice, and beside money already spent.
The label that says which pile a dollar is in, and when that pile closes. A prior gain is one pile. A loss coded as a cut in that gain is the move on this page.
A greater weight on a loss than on a matched gain. Integrating a small loss into a prior gain is a way to keep that weight off the line being read.
The value curve used to code gains and losses. This page does not carry the 1979 gamble counts. The curve is why a smaller gain can feel easier than a loss of the same size.
A choice that changes when the wording changes and the outcomes do not. The two-stage and one-stage rows are the same finals. The shares are 77 percent and 44 percent.
Staying with a cost already spent. Thaler and Johnson treat a prior gain as the other side of that history. A gain not yet given back is the pattern named here.
## One-Line Takeaway
After a gain, write the next bet once as a smaller gain and once from zero, and take it only if you still want it in the second writing.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.