> ## 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.
# Disposition Effect
> Disposition Effect is a higher rate of selling stocks that are up than stocks that are down. One sample realized gains at 0.148 and losses at 0.098.
**Category**: Effects
**Type**: A higher rate of selling positions that stand above their purchase price than positions that stand below it
**Origin**: Hersh Shefrin and Meir Statman, *Journal of Finance* (July 1985); the counts are Terrance Odean, *Journal of Finance* (October 1998)
**Also known as**: Selling winners too early and riding losers too long
**Quick Answer** — Disposition Effect is a higher rate of selling investments that are up than investments that are down, when up and down are measured from the purchase price. Hersh Shefrin and Meir Statman gave the pattern this name in 1985. In Terrance Odean's file of discount-brokerage accounts from 1987 through 1993, the proportion of gains realized was 0.148 and the proportion of losses realized was 0.098. In December of that file the pair reverses.
## What is Disposition Effect?
Disposition Effect is a higher rate of selling investments that are up than investments that are down, when up and down are measured from the purchase price.
> The tendency to hold losers too long and sell winners too soon has been labeled the disposition effect by Shefrin and Statman (1985).
That sentence is Terrance Odean in 1998. The label is the judgment in the name: the winner is sold too soon, and the loser is held too long. The count underneath the label is a comparison of two rates. On days when an account sells something, the share of winning positions that are actually sold is set beside the share of losing positions that are actually sold. A winner is a position above its average purchase price. A loser is a position below that price.
The purchase price is the line the account is using. [Prospect Theory](/models/prospect-theory) is the value curve that makes a gain and a loss different shapes. In the region of gains the curve flattens, so locking in the gain can feel like the safe close. In the region of losses the curve bends the other way, so a chance to get back to the purchase price can feel worth holding. [Loss Aversion](/effects/loss-aversion) is the extra weight on the loss itself. [Mental Accounting](/effects/mental-accounting) is the reason each stock can sit in its own pile, opened at the purchase and closed only when the shares are sold. This page does not carry the 1979 gamble counts.
The same two rates do not, by themselves, say which of two stories is at work. One story is that curve. The other is a belief that today's losers will soon beat today's winners. Odean writes that most of his tests do not separate those stories. A later price gap in the same file is what speaks to the belief. [House Money Effect](/effects/house-money-effect) is a different pattern: a greater willingness to take a new bet after a gain. The pattern on this page is which of the stocks already held gets sold.
### Disposition Effect in 3 Depths
* **Beginner**: The stock that is up feels finished. The stock that is down still feels open. The sale goes to the one that feels finished.
* **Practitioner**: On a day you sell, write the purchase price of every stock you are not selling. If you would not buy the one that is down at today's price, that is the position to close.
* **Advanced**: The 0.148 and the 0.098 are one aggregate of sale days in partial portfolios at one discount broker, from 1987 through 1993. December in that file goes the other way. Neither figure is a rate for a household, a fund, or a later market.
## Origin
The name is from Hersh Shefrin and Meir Statman, "The Disposition to Sell Winners Too Early and Ride Losers Too Long: Theory and Evidence," *Journal of Finance*, volume 40, number 3 (July 1985), pages 777–790. The issue is the papers and proceedings of the American Finance Association meeting in Dallas, 28–30 December 1984. Their abstract starts from Daniel Kahneman and Amos Tversky's point that people are averse to realizing a loss. Around that point they place mental accounting, regret aversion, self-control, and tax considerations. They write that tax considerations alone cannot explain the patterns of gains and losses that get realized. They also write that the concentration of loss realizations in December is not what fully rational tax timing would produce, and that the concentration fits their account.
Odean's report of their December piece is a self-control story. Investors are reluctant to sell at a loss, and they also know that a realized loss can reduce the tax bill. The year-end is the deadline. Selling the loser in December is the rule they impose on themselves before the deadline passes. This page does not re-audit the trading studies Shefrin and Statman discuss. The rates below are a later test of the pattern they named.
Those rates are from Terrance Odean, "Are Investors Reluctant to Realize Their Losses?," *Journal of Finance*, volume 53, number 5 (October 1998), pages 1775–1798. The byline places him at the University of California, Davis. The paper is based on his dissertation at the University of California, Berkeley. The records come from a large discount brokerage house, which the paper does not name. He draws 10,000 accounts at random from the accounts that had at least one transaction in 1987. The trades run from January 1987 through December 1993.
## Key Points
Four readings keep the purchase price fixed and move the calendar. A proportion in this file is a count of sale days. It is not a share of investors, and it is not a share of every day a stock was held.
Odean rebuilds, for each account, the stocks whose purchase date and price are inside the file. When the account sells, he compares the sale price with the average purchase price. A stock that is in the portfolio that morning and is not sold is a paper gain if the day's high and low are both above that average, and a paper loss if both are below it. If the average sits between the high and the low, he counts neither. Commissions are added to the purchase price and taken off the sale price. Dividends are left out of the test. Days with no sale in that account add nothing to either side.
Over 1987 through 1993 the file has 13,883 realized gains, 79,658 paper gains, 11,930 realized losses, and 110,348 paper losses. The proportion of gains realized is 0.148. The proportion of losses realized is 0.098. The ratio of those two proportions is a little over 1.5. Odean reads that as a stock that is up being more than 50 percent more likely to be sold, from day to day, than a stock that is down. The table prints the loss proportion minus the gain proportion as −0.050. The test that the gain proportion is no higher than the loss proportion is rejected with a t-statistic greater than 35.
December in the same file has 866 realized gains, 7,131 paper gains, 1,555 realized losses, and 10,604 paper losses. The gain proportion is 0.108 and the loss proportion is 0.128. The loss proportion minus the gain proportion is 0.020, with a t-statistic of 4.3. From January through November the proportions are 0.152 and 0.094. The monthly ratio of the gain proportion to the loss proportion runs from 2.1 in January to 0.85 in December. From January through November that ratio stays above 1 in each month.
A belief that losers will bounce back would be a reason to hold them. Odean compares what the sold winners and the unsold losers did next, in excess of the CRSP value-weighted index. Over the next 252 trading days the sold winners' average excess return is 0.0235 and the paper losses' is −0.0106. The difference is 0.0341. His sentence for that gap is that the sold winners' excess return over the following year is 3.4 percent more than the excess return of the losers that were not sold. The bootstrap frequency for a gap at least that large is 0.001.
## Applications
Use the sale day, the purchase price, and the month. Keep each proportion inside the count that produced it.
On a day you sell one holding, write the average purchase price of each holding you are not selling. Mark which side of that price today's quote is on. Close a down position you would not buy today, before you close an up position only because it is up. The 0.148 is the rate at which gains were closed in Odean's aggregate. It is not a quota for your account.
In this file the loss proportion exceeds the gain proportion in December, and the other eleven months go the other way. If a realized loss changes your tax bill, choose the position while the year still has months left. Waiting for December is the self-control deadline Shefrin and Statman describe. It is not a reason to ignore the down position in May.
When a bill forces a sale, the stock that is up is the easy one to name. Odean's prospect-theory sketch says a holder who needs liquidity, and who has no new information about either stock, is more likely to sell the one that is up. Name the stock you would refuse to buy today, and sell that one if it is the weaker holding.
The 0.148 uses realized gains divided by realized gains plus paper gains, summed across accounts and years. Paper lines exist only on days when that account sells. An equal-weighted account average in the same paper is 0.57 for gains and 0.36 for losses. Use the pair that matches the question you are asking, and do not move either pair onto a fund.
## Case Study
Terrance Odean tested the pattern Shefrin and Statman had named. The report is *Journal of Finance*, volume 53, number 5 (October 1998), pages 1775–1798. He was at the University of California, Davis. The work is his dissertation from the University of California, Berkeley. A large discount brokerage, not named in the paper, supplied the records. From the accounts with at least one transaction in 1987, he drew 10,000 at random. The trades run from January 1987 through December 1993.
The prices he needs come from the 1993 Center for Research in Security Prices daily file for New York Stock Exchange, American Stock Exchange, and Nasdaq stocks. Of the 10,000 accounts, 6,380 trade those stocks, in 97,483 transactions. He keeps those stocks. For each account he rebuilds only the positions whose purchase date and price are inside the file. Shares bought before January 1987 have no purchase price here, so the portfolios are a part of what the customers held. Same-day buys or sells of the same stock in the same account are aggregated.
The count is made on sale days, in portfolios of two or more stocks. A sale is a realized gain or a realized loss by the average purchase price. A stock still held that morning is a paper gain when the day's high and low are both above that average, and a paper loss when both are below it. A purchase price between the high and the low is neither. Commissions go into the purchase price and come off the sale price. Dividends are not used to decide the gain or the loss. A day with no sale in the account contributes no realized line and no paper line.
Across the whole year the totals are 13,883 realized gains, 79,658 paper gains, 11,930 realized losses, and 110,348 paper losses. The proportion of gains realized is 0.148. The proportion of losses realized is 0.098. Odean says the ratio is a little over 1.5, and that a stock that is up is more than 50 percent more likely to be sold from day to day than a stock that is down. The printed difference, loss proportion minus gain proportion, is −0.050. A one-tailed test of the claim that the gain proportion is no higher than the loss proportion is rejected with a t-statistic greater than 35. He writes that the t-statistic treats each realized line and each paper line as its own decision. Dependence among those lines inflates the statistic. He also writes that the dependence does not bias the two proportions.
December is the month the pair turns over. The month has 866 realized gains, 7,131 paper gains, 1,555 realized losses, and 10,604 paper losses. The proportions are 0.108 and 0.128. From January through November they are 0.152 and 0.094. The chart of the monthly ratio starts at 2.1 in January and ends at 0.85 in December. Tax-motivated selling is the reading he gives that month.
The same paper also weights each account equally, and it counts a stock only when that stock has not already been counted, as a sale or as a paper line, within a week in any account. An account enters this test only when both proportions have a nonzero denominator. There are 1,893 such accounts. Their average gain proportion is 0.57 and their average loss proportion is 0.36. The average of the differences is 0.21, and the test that this mean is no greater than zero is rejected with a t-statistic of 19. Measured in dollars on that same screen, the averages are 0.58 and 0.42, with a t-statistic of 13. In account-years that had both a potential gain and a potential loss, and that realized a net, 2,116 years realized a net dollar gain and 1,477 realized a net dollar loss. Odean's reading is that large losses are not, in most of these years, merely offsetting small gains.
The stocks that were sold as winners did not then lag the paper losers. Excess returns are measured against the CRSP value-weighted index. Over the next 84 trading days the averages are 0.0047 for sold winners and −0.0056 for paper losses. The difference is 0.0103, and the bootstrap frequency is 0.002. Over the next 252 trading days the averages are 0.0235 and −0.0106. The difference is 0.0341, and the bootstrap frequency is 0.001. Over the next 504 trading days the averages are 0.0645 and 0.0287. The difference is 0.0358, and the bootstrap frequency is 0.014. The 252-day gap is the one he states as 3.4 percent over the following year. Two-year returns are not calculated for sales in 1993, because the CRSP file he used ran through 1994. He calls 84 trading days the approximate median holding period inside the sample, and he notes that the true median is longer. Average turnover in these accounts is 6.5 percent a month, which he reads as a holding period of about 15 months.
The lesson is a pair of rates in one discount-brokerage file, a December reversal inside that file, and a later excess-return gap that does not reward the decision to keep the losers. It is not a quota for a household, and it is not a finding about a named fund.
## Boundaries and Failure Modes
Disposition Effect is the wrong name in three retellings. A sale-day ratio becomes the share of winners a person sells. December tax selling is told as the pattern itself. A gain that makes the next bet easier is given this page's name.
The first boundary is the file. The 0.148 and the 0.098 add up sale days from 1987 through 1993 at one discount broker. They use portfolios of two or more stocks. They use only stocks in the CRSP file, so 6,380 of the 10,000 accounts enter that count, through 97,483 transactions. Positions bought before January 1987 are missing, because the purchase price is missing. Odean notes that investors may have sold winners before 1987, ahead of the tax change in the 1986 law, and that the reconstructed portfolios could then be heavy with older losers that the file cannot see. He also reports that the ratio of the two proportions is virtually the same for 1987–1990 and for 1991–1993. The t-statistic greater than 35 assumes independent lines. He says the dependence inflates the statistic and does not bias the proportions.
The equal-weighted pair is a different object. Among 1,893 accounts, with the one-week screen, the averages are 0.57 and 0.36 in trades, 0.58 and 0.36 in shares, and 0.58 and 0.42 in dollars. When the week screen is dropped, the share-based averages are 0.51 and 0.31. Odean writes that these proportions shrink as portfolios get larger, which is why the aggregate pair sits so far below the account averages. The direction is the same. The level is not.
The second boundary is December, and the price of the stock. The loss proportion is the higher one in December: 0.128 against a gain proportion of 0.108. From January through November each month's ratio stays above 1, with t-statistics from 3.6 in November to 18 in January. A stricter cut, requiring at least one winner and one loser on the sale day and keeping year-to-date net realized losses under $3,000, makes the gap between the two proportions larger in each analysis he repeats. For January through November he also splits prices and the size of the return. Prices at or below $10, above $10 through $25, and above \$25 are 36 percent, 35 percent, and 29 percent of that data set. The absolute-return bands he names are 0 to 0.15, 0.15 to 0.30, 0.30 to 0.50, and above 0.50. He reports fourteen of fifteen partitions, with winners realized at a higher rate than losers, and a significant difference in thirteen. This page does not rebuild that grid. The most active tenth of traders account for 57 percent of the stock trades. Both that tenth and the other nine tenths, and both 1987–1990 and 1990–1993, show a higher gain proportion than loss proportion over the year, with a t-statistic greater than 22 in each of those cuts.
The third boundary is the cause, and the door of the sample. Odean writes that most of the tests do not separate prospect theory from a belief that losers will outperform winners. The later excess returns say that belief was not borne out on average in this file. They do not say what the next stock will do. He writes that a discount broker leaves out the influence of a retail broker, and that the same test on institutional trading, or at a retail brokerage, would be a different study. [House Money Effect](/effects/house-money-effect) is a greater willingness to take a bet after a gain. [Sunk Cost Fallacy](/effects/sunk-cost-fallacy) is staying with a cost already spent. The pattern named here is the gap between the rate at which winners are sold and the rate at which losers are sold.
## Common Misconceptions
Three retellings travel with the name. The year's proportion is treated as each account's habit. December is treated as the effect. A reluctance to quit is given this page's rates.
The 0.148 is one ratio for the whole file. Realized gains, 13,883, are divided by realized gains plus paper gains, 79,658. Paper gains are counted on sale days, and only when the day's high and low are both above the average purchase price. The equal-weighted account averages, on the one-week screen, are 0.57 and 0.36. Those averages answer a different question, and they are still not a quota for a person outside this file.
December is the month the pair reverses. The loss proportion is 0.128 and the gain proportion is 0.108. From January through November the proportions are 0.152 and 0.094. Shefrin and Statman treat the December cluster as a self-control deadline for a sale the investor has been postponing. The pattern named on this page is the higher rate of selling winners in the other months.
The effect here is a comparison of two rates, from the purchase price, on days when something is sold. A person who keeps a project because the money is already spent is in [Sunk Cost Fallacy](/effects/sunk-cost-fallacy). A person who takes a larger bet because an earlier gain is still on the table is in [House Money Effect](/effects/house-money-effect). Those pages do not carry the 0.148 or the 0.098.
## Related Concepts
Disposition Effect sits beside the value curve, the extra weight of a loss, and the pile each stock is kept in. It also sits beside a gain that changes the next bet, and beside a cost that has already been spent.
The value curve Odean uses to sketch why a winner is easier to sell and a loser is easier to hold. This page does not carry the 1979 gamble counts.
A greater weight on a loss than on a matched gain. Closing the account at a loss is the moment that weight is booked.
The label that keeps each stock in its own pile, opened at the purchase price and closed when the shares are sold.
A greater willingness to take a bet after a gain, when the loss can be coded as a cut in that gain. The pattern on this page is which holding gets sold.
Staying with a cost already spent. A down stock that is not sold is the overlap people notice. The count on this page is a rate of sale, not a project that is kept alive.
## One-Line Takeaway
On a day you sell, price every unsold holding from its purchase price, and close a down holding you would not buy today before you close an up holding only because it is up.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.