Thaler1999JBDMMentalaccountingmatters.pdf

Mental Accounting Matters

RICHARD H. THALER

Graduate School of Business, University of Chicago, USA

ABSTRACT

Mental accounting is the set of cognitive operations used by individuals andhouseholds to organize, evaluate, and keep track of ®nancial activities. Makinguse of research on this topic over the past decade, this paper summarizes thecurrent state of our knowledge about how people engage in mental accountingactivities. Three components of mental accounting receive the most attention.This ®rst captures how outcomes are perceived and experienced, and howdecisions are made and subsequently evaluated. The accounting system providesthe inputs to be both ex ante and ex post cost±bene®t analyses. A second com-ponent of mental accounting involves the assignment of activities to speci®caccounts. Both the sources and uses of funds are labeled in real as well as inmental accounting systems. Expenditures are grouped into categories (housing,food, etc.) and spending is sometimes constrained by implicit or explicit budgets.The third component of mental accounting concerns the frequency with whichaccounts are evaluated and `choice bracketing'. Accounts can be balanced daily,weekly, yearly, and so on, and can be de®ned narrowly or broadly. Each of thecomponents of mental accounting violates the economic principle of fungibility.As a result, mental accounting in¯uences choice, that is, it matters. Copyright# 1999 John Wiley & Sons, Ltd.

KEY WORDS mental accounting; choice bracketing; fungibility; budgeting

. A former colleague of mine, a professor of ®nance, prides himself on being a thoroughly rationalman. Long ago he adopted a clever strategy to deal with life's misfortunes. At the beginning of eachyear he establishes a target donation to the local United Way charity. Then, if anything untowardhappens to him during the year, for example an undeserved speeding ticket, he simply deducts thisloss from the United Way account. He thinks of it as an insurance policy against small annoyances.*

. A few years ago I gave a talk to a group of executives in Switzerland. After the conference my wifeand I spent a week visiting the area. At that time the Swiss franc was at an all-time high relative to theUS dollar, so the usual high prices in Switzerland were astronomical. My wife and I comfortedourselves that I had received a fee for the talk that would easily cover the outrageous prices for hotels

CCC 0894±3257/99/030183±24$17.50Copyright # 1999 John Wiley & Sons, Ltd. Accepted 1 September 1998

Journal of Behavioral Decision MakingJ. Behav. Dec. Making, 12: 183±206 (1999)

* This strategy need not reduce his annual contribution to the United Way. If he makes his intended contribution too low herisks having `uninsured' losses. So far he has not been `charitable' enough to have this fund cover large losses, such as when ahurricane blew the roof o� his beach house.

and meals. Had I received the same fee a week earlier for a talk in New York though, the vacationwould have been much less enjoyable.

. A friend of mine was once shopping for a quilted bedspread. She went to a department store and waspleased to ®nd a model she liked on sale. The spreads came in three sizes: double, queen and king.The usual prices for these quilts were $200, $250 and $300 respectively, but during the sale they wereall priced at only $150. My friend bought the king-size quilt and was quite pleased with her purchase,though the quilt did hang a bit over the sides of her double bed.

INTRODUCTION

The preceding anecdotes all illustrate the cognitive processes called mental accounting. What is mentalaccounting? Perhaps the easiest way to de®ne it is to compare it with ®nancial and managerialaccounting as practised by organizations. According to my dictionary accounting is `the system ofrecording and summarizing business and ®nancial transactions in books, and analyzing, verifying, andreporting the results'. Of course, individuals and households also need to record, summarize, analyze,and report the results of transactions and other ®nancial events. They do so for reasons similar to thosewhich motivate organizations to use managerial accounting: to keep trace of where their money isgoing, and to keep spending under control. Mental accounting is a description of the ways they dothese things.

How do people perform mental accounting operations? Regular accounting consists of numerousrules and conventions that have been codi®ed over the years. You can look them up in a textbook.Unfortunately, there is no equivalent source for the conventions of mental accounting; we can learnabout them only by observing behavior and inferring the rules.

Three components of mental accounting receive the most attention here. The ®rst captures howoutcomes are perceived and experienced, and how decisions are made and subsequently evaluated. Theaccounting system provides the inputs to do both ex ante and ex post cost±bene®t analyses. Thiscomponent is illustrated by the anecdote above involving the purchase of the quilt. The consumer'schoice can be understood by incorporating the value of the `deal' (termed transaction utility) into thepurchase decision calculus.

A second component of mental accounting involves the assignment of activities to speci®c accounts.Both the sources and uses of funds are labeled in real as well as in mental accounting systems.Expenditures are grouped into categories (housing, food, etc.) and spending is sometimes constrainedby implicit or explicit budgets. Funds to spend are also labeled, both as ¯ows (regular income versuswindfalls) and as stocks (cash on hand, home equity, pension wealth, etc.). The ®rst two anecdotesillustrate aspects of this categorization process. The vacation in Switzerland was made less painfulbecause of the possibility of setting up a Swiss lecture mental account, from which the expenditurescould be deducted. Similarly, the notional United Way mental account is a ¯exible way of makinglosses less painful.

The third component of mental accounting concerns the frequency with which accounts areevaluated and what Read, Loewenstein and Rabin (1998) have labeled `choice bracketing'. Accountscan be balanced daily, weekly, yearly, and so on, and can be de®ned narrowly or broadly. A well-known song implores poker players to `never count your money while you're sitting at the table'. Ananalysis of dynamic mental accounting shows why this is excellent advice, in poker as well as in othersituations involving decision making under uncertainty (such as investing).

The primary reason for studying mental accounting is to enhance our understanding of thepsychology of choice. In general, understanding mental accounting processes helps us understand

184 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

choice because mental accounting rules are not neutral.* That is, accounting decisions such as to whichcategory to assign a purchase, whether to combine an outcome with others in that category, and howoften to balance the `books' can a�ect the perceived attractiveness of choices. They do so becausemental accounting violates the economic notion of fungibility. Money in one mental account is not aperfect substitute for money in another account. Because of violations of fungibility, mental account-ing matters.

The goal of this paper is to illustrate how mental accounting matters. To this end I draw uponresearch conducted over the past two decades. This describes where I think the ®eld is now, having beeninformed by the research of many others, especially over the past few years.

THE FRAMING OF GAINS AND LOSSES

The value functionWe wish to understand the decision-making process of an individual or a household interacting in aneconomic environment. How does a person make economic decisions, such as what to buy, how muchto save, and whether to buy or lease an item? And how are the outcomes of these ®nancial transactionsevaluated and experienced?

Following my earlier treatment of these questions (Thaler, 1980, 1985) I assume that people perceiveoutcomes in terms of the value function of Kahneman and Tversky's (1979) prospect theory. The valuefunction can be thought of as a representation of some central components of the human perceivedpleasure machine.{ It has three important features, each of which captures an essential element ofmental accounting:

(1) The value function is de®ned over gains and losses relative to some reference point. The focus onchanges, rather than wealth levels as in expected utility theory, re¯ects the piecemeal nature ofmental accounting. Transactions are often evaluated one at a time, rather than in conjunction witheverything else.

(2) Both the gain and loss functions display diminishing sensitivity. That is, the gain function is concaveand the loss function is convex. This feature re¯ects the basic psychophysical principle (the Weber±Fechner law) that the di�erence between $10 and $20 seems bigger than the di�erence between$1000 and $1010, irrespective of the sign.

(3) Loss aversion. Losing $100 hurts more than gaining $100 yields pleasure: v�x�5 ÿ v�ÿx�. Thein¯uence of loss aversion on mental accounting is enormous, as will become evident very quickly.

Decision framesThe role of the value function in mental accounting is to describe how events are perceived and codedin making decisions. To introduce this topic, it is useful to de®ne some terms. Tversky and Kahneman(1981, p. 456) de®ne a mental account{ quite narrowly as `an outcome frame which speci®es (i) the set

* An accounting system is a way of aggregating and summarizing large amounts of data to facilitate good decision making. In anideal world the accounting system would accomplish this task in such a way that the decision maker would make the same choicewhen presented with only the accounting data as she would if she had access to all the relevant data. This is what I mean by`neutral'. In a sense, such an accounting system would provide decision makers with `su�cient statistics'. Of course, achievingthis goal is generally impossible, because something must be sacri®ced in order to reduce the information the decision maker hasto look at. Thus neither organizational nor mental accounting will achieve neutrality.{ Prospect theory predates Kahneman's (1994) important distinction between decision utility and experienced utility. In histerms, the prospect theory value function measures decision utility.{ Actually, they use the term psychological account in their 1981 paper, following the terminology I used in my 1980 paper.Later (Kahneman and Tversky, 1984) they suggest the better term `mental account'.

R. H. Thaler Mental Accounting Matters 185

of elementary outcomes that are evaluated jointly and the manner in which they are combined and (ii) areference outcome that is considered neutral or normal'. (Typically, the reference point is the statusquo.) According to this de®nition, a mental account is a frame for evaluation. I wish to use the term`mental accounting' to describe the entire process of coding, categorizing, and evaluating events, so thisnarrow de®nition of a mental account is a bit con®ning. Accordingly, I will refer to simply outcomeframes as `entries'.

In a later paper, Kahneman and Tversky (1984, p. 347), propose three ways that outcomes might beframed: in terms of a minimal account, a topical account, or a comprehensive account. Comparing twooptions using the minimal account entails examining only the di�erences between the two options,disregarding all their common features. A topical account relates the consequences of possible choicesto a reference level that is determined by the context within which the decision arises. A comprehensiveaccount incorporates all other factors including current wealth, future earnings, possible outcomes ofother probabilistic holdings, and so on. (Economic theory generally assumes that people makedecisions using the comprehensive account.) The following example* illustrates that mental accountingis topical:

Imagine that you are about to purchase a jacket for ($125)[$15] and a calculator for ($15)[$125]. Thecalculator salesman informs you that the calculator you wish to buy is on sale for ($10)[$120] at theother branch of the store, located 20 minutes drive away. Would you make the trip to the other store?(Tversky and Kahneman, 1981, p. 459)

When two versions of this problem are given (one with the ®gures in parentheses, the other with the®gures in brackets), most people say that they will travel to save the $5 when the item costs $15 but notwhen it costs $125. If people were using a minimal account frame they would be just asking themselveswhether they are willing to drive 20 minutes to save $5, and would give the same answer in eitherversion.

Interestingly, a similar analysis applies in the comprehensive account frame. Let existing wealth beW, and W* be existing wealth plus the jacket and calculator minus $140. Then the choice comes downto the utility of W* plus $5 versus the utility of W* plus 20 minutes. This example illustrates animportant general point Ð the way a decision is framed will not alter choices if the decision maker isusing a comprehensive, wealth-based analysis. Framing does alter choices in the real world becausepeople make decisions piecemeal, in¯uenced by the context of the choice.

Hedonic framingThe jacket and calculator problem does demonstrate that mental accounting is piecemeal and topical,but there is more to learn from this example. Why are we more willing to drive across town to savemoney on a small purchase than a large one? Clearly there is some psychophysics at work here. Fivedollars seems like a signi®cant saving on a $15 purchase, but not so on a $125 purchase. But thisdisparity implies that the utility of the saving must be associated with the di�erences in values ratherthan the value of the di�erence. That is, the utility of saving $5 on the purchase of the expensive itemmust be �v�ÿ$125� ÿ v�ÿ$120� (or perhaps the ratio of these values) rather than v($5), otherwise therewould be no di�erence between the two versions of the problem.

What else do we know about mental accounting arithmetic? Speci®cally, how are two or more®nancial outcomes (within a single account) combined? This is an important question because wewould like to be able to construct a model of how consumers evaluate events such as purchases thattypically involve combinations of outcomes, good or bad.

* This problem was based on similar examples discussed by Savage (1954) and Thaler (1980).

186 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

One possible place to start in building a model of how people code combinations of events is toassume they do so to make themselves as happy as possible. To characterize this process we need toknow how someone with a prospect theory value function could wish to have the receipt of multipleoutcomes framed. That it, for two outcomes x and y, when will v(x � y) be greater than v(x) � v(y)? Ihave previously considered this question (Thaler, 1985). Given the shape of the value function, it is easyto derive the following principles of hedonic framing, that is, the way of evaluating joint outcomes tomaximize utility:

(1) Segregate gains (because the gain function is concave).(2) Integrate losses (because the loss function is convex).(3) Integrate smaller losses with larger gains (to o�set loss aversion).(4) Segregate small gains (silver linings) from larger losses (because the gain function is steepest at the

origin, the utility of a small gain can exceed the utility of slightly reducing a large loss).

As I showed, most people share the intuition that leads to these principles. That is, if you ask subjects`Who is happier, someone who wins two lotteries that pay $50 and $25 respectively, or someone whowins a single lottery paying $75?' 64% say the two-time winner is happier. A similar majority sharedthe intuition of the other three principles.

These principles are quite useful in thinking about marketing issues. In other words, if one wants todescribe the advantages and disadvantages of a particular product in a way that will maximize theperceived attractiveness of the product to consumers, the principles of hedonic framing are a helpfulguide. For example, framing a sale as a `rebate' rather than a temporary price reduction might facilitatethe segregation of the gain in line with principle (4).

The failure of the hedonic editing hypothesisIt would be convenient if these same principles could also serve as a good descriptive model of mentalaccounting. Can people be said to edit or parse the multiple outcomes they consider or experience in away that could be considered optimal, that is, hedonic editing.* More formally, if the symbol `&' isused to denote the cognitive combination of two outcomes, then hedonic editing is the application ofthe following rule:

v�x & y� � Max�v�x � y�;v�x�� v�y��The hypothesis that people engage in hedonic editing has obvious theoretical appeal,{ but somethought reveals that it cannot be descriptively correct. Consider the jacket and calculator problemagain. If the $5 saving were coded in a utility-maximizing way it would be segregated in either case,inconsistent with the data. Furthermore, there must be some limits to our abilities to engage in self-deception. Why stop at segregating the $5 gain? Why not code it as ®ve gains of $1? Nevertheless,hedonic editing represents a nice starting point for the investigation of how people do code multipleevents.

Eric Johnson and I have investigated the limits of the hedonic editing hypothesis (Thaler andJohnson, 1990). Our ultimate goal was to explore the in¯uence of prior outcomes on risky choices (see

* Johnson and I used the term `editing' for this process, though on re¯ection `parsing' might have been better. I will stick withthe original term to avoid confusion with the prior literature. Note that editing refers to active cognitions undertaken by thedecision maker. In contrast, I will use `framing' to refer to the way a problem is posed externally. As we will see, people prefer tohave outcomes framed hedonically, but fail to edit (or one could say, reframe) them accordingly.{ Indeed, see Fishburn and Luce (1995) for an axiomatic treatment of hedonic editing.

R. H. Thaler Mental Accounting Matters 187

below), but we began with the more basic question of how people choose to code multiple events suchas a gain of $30 followed by a loss of $9. One approach we used was to ask people their preferencesabout temporal spacing. For two speci®ed ®nancial outcomes, we asked subjects who would behappier, someone who had these two events occur on the same day, or a week or two apart? Thereasoning for this line of inquiry was that temporal separation would facilitate cognitive segregation.So if a subject wanted to segregate the outcomes x and y, he would prefer to have them occur ondi�erent days, whereas if he wanted to integrate them, he would prefer to have them occur together.The hedonic editing hypothesis would be supported if subjects preferred temporal separation for caseswhere the hypothesis called for segregation, and temporal proximity when integration was preferred.For gains, the hedonic editing hypothesis was supported. A large majority of subjects thought temp-oral separation of gains produced more happiness. But, in contrast to the hedonic editing hypothesis,subjects thought separating losses was also a good idea. Why?

The intuition for the hypothesis that people would want to combine losses comes from the fact thatthe loss function displays diminishing sensitivity. Adding one loss to another should diminish itsmarginal impact. By wishing to spread out losses, subjects seem to be suggesting that they think that aprior loss makes them more sensitive towards subsequent losses, rather than the other way around. Inother words, subjects are telling us that they are unable to simply add one loss to another (inside thevalue function parentheses). Instead, they feel that losses must be felt one by one, and that bearing oneloss makes one more sensitive to the next.*

To summarize, the evidence suggests that the rules of hedonic framing are good descriptions of theway people would like to have the world organized (many small gains including silver linings; lossesavoided if possible but otherwise combined). People will also actively parse outcomes consistent withthese rules, with the exception of multiple losses.

There are two important implications of these results for mental accounting. First, we would expectmental accounting to be as hedonically e�cient as possible. For example, we should expect thatopportunities to combine losses with larger gains will be exploited wherever feasible. Second, lossaversion is even more important than the prospect theory value function would suggest, as it is di�cultto combine losses to diminish their impact. This result suggests that we should expect to see that someof the discretion inherent in any accounting system will be used to avoid having to experience losses.

MENTAL ACCOUNTING DECISION MAKING

Transaction utilityWhat happens when a consumer decides to buy something, trading money for some object? Onepossibility would be to code the acquisition of the product as a gain and the forgone money as a loss.But loss aversion makes this frame hedonically ine�cient. Consider a thirsty consumer who wouldrather have a can of soda than one dollar and is standing in front of a vending machine that sells sodafor 75 cents. Clearly the purchase makes her better o�, but it might be rejected if the payment werecognitively multiplied by 2.25 (an estimate of the coe�cient of loss aversion). This thinking has ledboth Kahneman and Tversky (1984) and me (Thaler, 1985) to reject the idea that costs are generallyviewed as losses.

Instead, I proposed that consumers get two kinds of utility from a purchase: acquisition utility andtransaction utility. Acquisition utility is a measure of the value of the good obtained relative to its price,similar to the economic concept of consumer surplus. Conceptually, acquisition utility is the value theconsumer would place on receiving the good as a gift, minus the price paid. Transaction utility

* Linville and Fischer (1991) also investigate the predictive power of hedonic editing, with similar results.

188 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

measures the perceived value of the `deal'. It is de®ned as the di�erence between the amount paid andthe `reference price' for the good, that is, the regular price that the consumer expects to pay for thisproduct. The following example (from Thaler, 1985) illustrates the role of transaction utility.

You are lying on the beach on a hot day. All you have to drink is ice water. For the last hour youhave been thinking about how much you would enjoy a nice cold bottle of your favorite brand ofbeer. A companion gets up to go make a phone call and o�ers to bring back a beer from the onlynearby place where beer is sold (a fancy resort hotel) [a small, run-down grocery store]. He says thatthe beer might be expensive and so asks how much you are willing to pay for the beer. He says that hewill buy the beer if it costs as much or less than the price you state. But if it costs more than the priceyou state he will not buy it. You trust your friend, and there is no possibility of bargaining with the(bartender) [store owner]. What price do you tell him?

Two versions of the question were administered, one using the phrases in parentheses, the other thephrases in brackets. The median responses for the two versions were $2.65 (resort) and $1.50 [store] in1984 dollars. People are willing to pay more for the beer from the resort because the reference price inthat context is higher. Note that this e�ect cannot be accommodated in a standard economic modelbecause the consumption experience is the same in either case; the place of purchase should beirrelevant.

The addition of transaction utility to the purchase calculus leads to two kinds of e�ects in themarketplace. First, some goods are purchased primarily because they are especially good deals. Mostof us have some rarely worn items in our closets that are testimony to this phenomenon. Sellers makeuse of this penchant by emphasizing the savings relative to the regular retail price (which serves as thesuggested reference price). In contrast, some purchases that would seemingly make the consumer bettero� may be avoided because of substantial negative transaction utility. The thirsty beer drinker whowould pay $4 for a beer from a resort but only $2 from a grocery store will miss out on some pleasantdrinking when faced with a grocery store charging $2.50.

Opening and closing accountsOne of the discretionary components of an accounting system is the decision of when to leave accounts`open' and when to `close' them. Consider the example of someone who buys 100 shares of stock at $10a share. This investment is initially worth $1000, but the value will go up or down with the price of thestock. If the price changes, the investor has a `paper' gain or loss until the stock is sold, at which pointthe paper gain or loss becomes a `realized' gain or loss. The mental accounting of paper gains andlosses is tricky (and depends on timing Ð see below), but one clear intuition is that a realized loss ismore painful than a paper loss. When a stock is sold, the gain or loss has to be `declared' both to the taxauthorities and to the investor (and spouse). Because closing an account at a loss is painful, aprediction of mental accounting is that people will be reluctant to sell securities that have declined invalue. In particular, suppose an investor needs to raise some cash and must choose between two stocksto sell, one of which has increased in value and one of which has decreased. Mental accounting favorsselling the winner (Shefrin and Statman, 1987) whereas a rational analysis favors selling the loser.*Odean (1998) ®nds strong support for the mental accounting prediction. Using a data set that trackedthe trades of investors using a large discount brokerage ®rm, Odean ®nds that investors were morelikely to sell one of their stocks that had increased in value than one of their stocks that had decreased.{

* A rational investor will choose to sell the loser because capital gains are taxable and capital losses are deductible.{ Of course, such a strategy could be rational if the losers they kept subsequently increased in value more than the winners theysold, but this outcome was not observed. Indeed, these investors are not particularly savvy. The stocks they sell subsequentlyoutperform the stocks they buy!

R. H. Thaler Mental Accounting Matters 189

Other evidence of a reluctance to close an account in the `red' comes from the world of realaccounting. Most public corporations make o�cial earnings announcements every quarter. Althoughearnings are audited, ®rms retain some discretion in how quickly to count various components ofrevenues and expenses, leaving them with some control over the actual number they report. Severalrecent papers (e.g. Burgstahler and Dichev, 1997; Degeorge, Patel and Zeckhauser, forthcoming) showthat ®rms use this discretionary power to avoid announcing earnings decreases and losses. Speci®cally,a plot of earnings per share (in cents per share) or change in earnings per share (this quarter versussame quarter last year) shows a sharp discontinuity at zero. Firms are much more likely to make apenny a share than to lose a penny a share, and are much more likely to exceed last year's earnings by apenny than to miss by a penny. So small losses are converted into small gains. In contrast, large gainsseem to be trimmed down (to increase the chance of an increase again next year) whereas moderatelosses are somewhat in¯ated (a procedure known in accounting circles as `taking the big bath').Apparently, ®rms believe that shareholders (or potential shareholders) react to earnings announce-ments in a manner consistent with prospect theory.

Advance purchases, sunk costs, and payment depreciationAnother situation in which a consumer has to decide when to open and close an account is when apurchase is made well in advance of consumption. Consider paying $100 for two tickets to a basketballgame to be held in a month's time. Suppose that the tickets are being sold at the reference price sotransaction utility is zero. In this case the consumer can be said to open an account at the point atwhich the tickets are purchased. At this time the account has a negative balance of $100. Once the dateof the game comes and the game is attended, the account can be closed.

What happens if something (a blizzard) prevents the consumer from attending the game? In this casethe consumer has to close the account at a loss of $100; in accounting terminology the loss has to berecognized. Notice that this event turns a cost into a loss, which is aversive. Still, why does the priorexpenditure (now a sunk cost) make someone more willing to go to the game in a blizzard (as in theexample in Thaler, 1980)?

To answer this question we need to consider how transactions are evaluated. For most routinepurchases there is no ex post evaluation of the purchase when the account is closed. Such evaluationsbecome more likely as the size of the transaction increases or as the purchase or situation becomes moreunusual. Failing to attend an event that has been paid for makes the purchase highly salient and anevaluation necessary. By driving through the storm, the consumer can put the game back into thecategory of normal transactions that are not explicitly evaluated and thus avoid adding up the costsand bene®ts (barring an accident!). Furthermore, even if an ex post evaluation is made, the extra cost ofgoing to the game may not be included in the evaluation. As Heath (1995) suggests, because the costs ofdriving to the game are not monetary, they may not be included in the analysis.* In Heath's terms theyare incidental, that is, in a di�erent mental account. He makes the telling comparison between this caseand the Kahneman and Tversky (1984) theater ticket example, in which subjects are less willing to buya ticket to a play after having lost their ticket than after having lost an equivalent sum of money. In thetheater ticket example, buying a second ticket is aversive because it is included in the mental accountfor the theater outing, but the loss of the money is not.

* Of course, although the driving costs may not be included in the basketball game account, they must be compared, at leastprospectively, to something when one is deciding whether to go. In this formulation someone would choose to take the drive,not in order to enjoy the game, but to avoid feeling the pain associated with the unamortized ticket expense.

190 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

Although sunk costs in¯uence subsequent decisions, they do not linger inde®nitely. A thoughtexperiment illustrates this point nicely. Suppose you buy a pair of shoes. They feel perfectly comfort-able in the store, but the ®rst day you wear them they hurt. A few days later you try them again, butthey hurt even more than the ®rst time. What happens now? My predictions are:

(1) The more you paid for the shoes, the more times you will try to wear them. (This choice may berational, especially if they have to be replaced with another expensive pair.)

(2) Eventually you stop wearing the shoes, but you do not throw them away. The more you paid for theshoes, the longer they sit in the back of your closet before you throw them away. (This behaviorcannot be rational unless expensive shoes take up less space.)

(3) At some point, you throw the shoes away, regardless of what they cost, the payment having beenfully `depreciated'.

Evidence about the persistence of sunk costs e�ects is reported by Arkes and Blumer (1985). Theyran an experiment in which people who were ready to buy season tickets to a campus theater groupwere randomly placed into three groups: one group paid full price, one group got a small (13%)discount, and one group received a large (47%) discount. The experimenters then monitored how oftenthe subjects attended plays during the season. In the ®rst half of the season, those who paid full priceattended signi®cantly more plays than those who received discounts, but in the second half of theseason there was no di�erence among the groups. People do ignore sunk costs, eventually.

The gradual reduction in the relevance of prior expenditures is dubbed `payment depreciation' byGourville and Soman (1998) who have conducted a clever ®eld experiment to illustrate the idea. Theyobtained usage data from the members of a health club that charges the dues to its members twice ayear. Gourville and Soman ®nd that attendance at the health club is highest in the month in which thedues are paid and then declines over the next ®ve months, only to jump again when the next bill comesout.

Similar issues are involved in the mental accounting of wine collectors who often buy wine with theintention of storing it for ten years or more while it matures. When a bottle is later consumed, whathappens? Eldar Sha®r and I (1998) have investigated this pressing issue by surveying the subscribers toa wine newsletter aimed at serious wine consumers/collectors. We asked the following question:

Suppose you bought a case of a good 1982 Bordeaux in the futures market for $20 a bottle. The winenow sells at auction for about $75 a bottle. You have decided to drink a bottle. Which of thefollowing best captures your feeling of the cost to you of drinking this bottle?

We gave the respondents ®ve answers to choose from: $0, $20, $20 plus interest, $75, and ÿ$55 (`Idrink a $75 bottle for which I paid only $20'). The percentages of respondents choosing each answerwere 30, 18, 7, 20 and 25. Most of the respondents who selected the economically correct answer ($75)were in fact economists. (The newsletter, Liquid Assets, is published by economist Orley Ashenfelterand has many economist subscribers). More than half the respondents report that drinking the bottleeither costs nothing or actually saves them money!

The results of this survey prompted us to run a follow-up survey the following year. The question thistime was:

Suppose you buy a case of Bordeaux futures at $400 a case. The wine will retail at about $500 a casewhen it is shipped. You do not intend to start drinking this wine for a decade. At the time that youacquire this wine which statement more accurately captures your feelings?(a) I feel like I just spent $400, much as I would feel if I spent $400 on a weekend getaway.(b) I feel like I made a $400 investment which I will gradually consume after a period of years.

R. H. Thaler Mental Accounting Matters 191

(c) I feel like I just saved $100, the di�erence between what the futures cost and what the wine will sellfor when delivered.

Respondents rated each answer on a ®ve-point scale. Most respondents selected answer (b) as theirfavorite, coding the initial purchase as an investment. Notice that this choice means that the typicalwine connoisseur thinks of his initial purchase as an investment and later thinks of the wine as freewhen he drinks it. We have therefore titled our paper `Invest Now, Drink Later, Spend Never'. Notethat this mental accounting transforms a very expensive hobby into one that is `free'. The same mentalaccounting applies to time-share vacation properties. The initial purchase of a week every year at someresort feels like an investment, and the subsequent visits feel free.

Payment decouplingIn the wine example, the prepayment separates or `decouples' (Prelec and Loewenstein, 1998; Gourvilleand Soman, 1998) the purchase from the consumption and in so doing seems to reduce the perceivedcost of the activity. Prepayment can often serve this role, but the mental accounting advantages ofdecoupling are not all associated with prepayment. Consider the case of the pricing policies of the ClubMed resorts (Thaler, 1980). At these vacation spots consumers pay a ®xed fee for a vacation thatincludes meals, lodging, and recreation. This plan has two advantages. First, the extra cost of includingthe meals and recreation in the price will look relatively small when combined with the other costs ofthe vacation. Second, under the alternative plan each of the small expenditures looks large by itself,and is likely to be accompanied by a substantial dose of negative transaction utility given the pricesfound at most resorts.

Another disadvantage of the piece-rate pricing policy is that it makes the link between the paymentand the speci®c consumption act very salient, when the opposite is highly desirable. For example, a prix®xe dinner, especially an expensive multi-course meal, avoids the unsavory prospect of matching a veryhigh price with the very small quantity of food o�ered in each course.* Along the same lines, manyurban car owners would be ®nancially better o� selling their car and using a combination of taxis andcar rentals. However, paying $10 to take a taxi to the supermarket or a movie is both salient and linkedto the consumption act; it seems to raise the price of groceries and movies in a way that monthly carpayments (or even better, a paid-o� car) do not.

More generally, consumers don't like the experience of `having the meter running'. This contributesto what has been called the `¯at rate bias' in telecommunications. Most telephone customers elect a ¯atrate service even though paying by the call would cost them less.{ As Train (1991, p. 211) says,`consumers seem to value ¯at-rate service over measured service even when the bill that the consumerwould receive under the two services, given the number of calls the consumer places, would be thesame . . . The existence of this bias is problematical. Standard theory of consumer behavior does notaccommodate it'. Similarly, health clubs typically charge members by the month or year rather than ofa per-use basis. This strategy decouples usage from fees, making the marginal cost of a visit zero. Thisplan is attractive because a health club is a service that many consumers feel they should use moreoften, but fail to do so for self-control reasons (see below). Indeed, the monthly fee, although a sunkcost, encourages use for those who want to reduce their per-visit charges. Compare this system to apure usage-based pricing system in which Stairmaster users pay `per step'. This pricing system would be

* In contrast, the review of one expensive San Francisco restaurant in the Zagat guide includes the following gripe from acustomer. `$13 for two scallops. Who are they kidding?'{ This example is cited by Prelec and Loewenstein (1998). American OnLine seems to have learned this lesson the hard way.When they o�ered a ¯at rate Internet service in early 1996 they were so overwhelmed with demand that consumers had troublelogging on to the service, causing embarrassing publicity.

192 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

completely incompatible with the psychological needs of the club member who desires usage encour-agement rather than discouragement.

Perhaps the best decoupling device is the credit card. We know that credit cards facilitate spendingsimply by the fact that stores are willing to pay 3% or more of their revenues to the card companies (seealso Feinberg, 1986; Prelec and Simester, 1998). A credit card decouples the purchase from the pay-ment in several ways. First, it postpones the payment by a few weeks. This delay creates two distincte�ects: (a) the payment is later than the purchase; (b) the payment is separated from the purchase. Thepayment delay may be attractive to some consumers who are either highly impatient or liquidityconstrained, but as Prelec and Loewenstein (1998) stress, ceteris paribus, consumers prefer to paybefore rather than after, so this factor is unlikely to be the main appeal of the credit card purchase.Rather, the simple separation of purchase and payment appears to make the payment less salient.Along these lines, Soman (1997) ®nds that students leaving the campus bookstore were much moreaccurate in remembering the amount of their purchases if they paid by cash rather than by credit card.As he says, `Payment by credit card thus reduces the salience and vividness of the out¯ows, makingthem harder to recall than payments by cash or check which leave a stronger memory trace' (p. 9).

A second factor contributing to the attractiveness of credit card spending is that once the bill arrives,the purchase is mixed in with many others. Compare the impact of paying $50 in cash at the store tothat of adding a $50 item to an $843 bill. Psychophysics implies that the $50 will appear larger by itselfthan in the context of a much larger bill, and in addition when the bill contains many items each onewill lose salience. The e�ect becomes even stronger if the bill is not paid in full immediately. Althoughan unpaid balance is aversive in and of itself, it is di�cult for the consumer to attribute this balance toany particular purchase.

BUDGETING

So far I have been discussing mental accounting decision making at the level of individual transactions.Another component of mental accounting is categorization or labeling. Money is commonly labeled atthree levels: expenditures are grouped into budgets (e.g. food, housing, etc.); wealth is allocated intoaccounts (e.g. checking, pension, `rainy day'); and income is divided into categories (e.g. regular orwindfall). Such accounts would be inconsequential if they were perfectly fungible (i.e. substitutable) asassumed in economics. But, they are not fungible, and so they `matter'.

Consumption categoriesDividing spending into budget categories serves two purposes. First, the budgeting process canfacilitate making rational trade-o�s between competing uses for funds. Second, the system can act as aself-control device. Just as organizations establish budgets to keep track of and limit divisionalspending, the mental accounting system is the household's way of keeping spending within the budget(Thaler and Shefrin, 1981). Of course, there is considerable variation among households in how explicitthe budgeting process is.* As a rule, the tighter the budget, the more explicit are the budgeting rules,both in households and organizations. Families living near the poverty level use strict, explicit budgets;in wealthy families budgets are both less binding and less well de®ned.{ Poorer families also tend tohave budgets de®ned over shorter periods (a week or month), whereas wealthier families may use

* Many of the generalizations here are based on a series of interviews conducted on my behalf in the early 1980s. See also Zelizer(1994) and her references. At one time many households used a very explicit system with envelopes of cash labeled with variousspending categories. To some extent, programs such as Quicken serve as a modern replacement for this method.{ Still, budgets can matter even in well-o� families. As the discussion of `decoupling' below will illustrate, spending on vacationsmay depend on whether a family rents or owns a vacation home.

R. H. Thaler Mental Accounting Matters 193

annual budgets. For example, Heath and Soll (1996) report that most of their MBA student subjectshad weekly food and entertainment budgets and monthly clothing budgets. It is likely that these ruleschanged dramatically when the students got jobs at the end of their studies (in violation of the life-cyclehypothesis Ð see below).

Heath and Soll describe the process by which expenses are tracked against these budgets. They dividethe tracking process into two stages:

(1) Expenses must ®rst be noticed and (2) then assigned to their proper accounts. An expense will nota�ect a budget if either stage fails. To label these stages we borrow terminology from ®nancialaccounting in which the accounting system is also divided into two stages. Expenses must bebooked (i.e. recorded in the accounting system) and posted (i.e. assigned to a speci®c expenseaccount). Each process depends on a di�erent cognitive system. Booking depends on attention andmemory. Posting depends on similarity judgments and categorization (p. 42).*

Many small, routine expenses are not booked. Examples would include lunch or co�ee at theworkplace cafeteria (unless the norm is to bring these items from home, in which case buying the lunchmight be booked). Ignoring such items is equivalent to the organizational practice of assigning smallexpenditures to a `petty cash' fund, not subject to the usual accounting scrutiny. The tendency toignore small items may also explain an apparent contradiction of hedonic framing. As noted by JohnGourville (1998), in many situations sellers and fund raisers elect to frame an annual fee as `pennies-a-day'. Thus a $100 membership for the local public radio station might be described as a `mere 27 centsa day'. Given the convex shape of the loss function, why should this strategy be e�ective? Onepossibility is that 27 cents is clearly in the petty cash category, so when the expense is framed this way ittends to be compared to other items that are not booked. In contrast, a $100 membership is largeenough that it will surely be booked and posted, possibly running into binding budget constraints inthe charitable giving category. The same idea works in the opposite direction. A ®rm that markets adrug to help people quit smoking urges smokers to aggregate their annual smoking expenditures andthink of the vacation they could take with these funds. Again, $2 a day might be ignored but $730 paysfor a nice getaway.

Implications of violations of fungibilityWhenever budgets are not fungible their existence can in¯uence consumption in various ways. Oneexample is the case in which one budget has been spent up to its limit while other accounts haveunspent funds remaining. (This situation is common in organizations. It can create extreme distortionsespecially if funds cannot be carried over from one year to the next. In this case one department can beseverely constrained while another is desperately looking for ways to spend down this year's budget tomake sure next year's is not cut.) Heath and Soll (1996) provide several experiments to illustrate thise�ect. In a typical study two groups of subjects were asked whether they would be willing to buy aticket to a play. One group was told that they had spent $50 earlier in the week going to a basketballgame (same budget); the other group was told that they had received a $50 parking ticket (di�erentbudget) earlier in the week. Those who had already gone to the basketball game were signi®cantly lesslikely to go to the play than those who had gotten the parking ticket.{

* Regarding the categorization process, see Henderson and Peterson (1992). It should be noted that in a ®nancial accountingsystem in a ®rm any expense that is booked is also posted.{ One might think this result could be attributed to satiation (one night out is enough in a week). However, another group wasasked their willingness to buy the theater ticket after going to the basketball game for free, and they showed no e�ect.

194 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

Using the same logic that implies that money should be fungible (i.e. that money in one account willspend just as well in another), economists have argued that time should also be fungible. A rationalperson should allocate time optimally, which implies `equating at the margin'. In this case, themarginal value of an extra minute devoted to any activity should be equal.* The jacket and calculatorproblem reveals that this rule does not describe choices about time. Subjects are willing to spend 20minutes to save $5 on a small purchase but not a large one. Leclerc et al. (1995) extend this notion byreversing the problem. They ask people how much they would be willing to pay to avoid waiting in aticket line for 45 minutes. They ®nd that people are willing to pay twice as much to avoid the wait for a$45 purchase than for a $15 purchase. As in the original version of the problem, we see that the implicitvalue people put on their time depends on the ®nancial context.

Self-control and gift givingAnother violation of fungibility introduced by the budgeting system occurs because some budgets areintentionally set `too low' in order to help deal with particularly insidious self-control problems. Forexample, consider the dilemma of a couple who enjoy drinking a bottle of wine with dinner. Theymight decide that they can a�ord to spend only $10 a night on wine and so limit their purchases towines that cost $10 a bottle on average, with no bottle costing more than $20. This policy might not beoptimal in the sense that an occasional $30 bottle of champagne would be worth more than $30 tothem, but they don't trust themselves to resist the temptation to increase their wine budget unreason-ably if they break the $20 barrier. An implication is that this couple would greatly enjoy gifts of winethat are above their usual budget constraint. This analysis is precisely the opposite of the usualeconomic advice (which says that a gift in kind can be at best as good as a gift of cash, and then only ifit were something that the recipient would have bought anyway). Instead the mental accountinganalysis suggests that the best gifts are somewhat more luxurious than the recipient normally buys,consistent with the conventional advice (of non-economists), which is to buy people something theywouldn't buy for themselves.

The idea that luxurious gifts can be better than cash is well known to those who design salescompensation schemes. When sales contests are run, the prize is typically a trip or luxury durablerather than cash. Perhaps the most vivid example of this practice is the experience of the NationalFootball League in getting players to show up at the annual Pro Bowl. This all-star game is held theweek after the Super Bowl and for years the league had trouble getting all of the superstar players tocome. Monetary incentives were little inducement to players with seven-®gure salaries. This problemwas largely solved by moving the game to Hawaii and including two ®rst-class tickets (one for theplayer's wife or girlfriend) and accommodations for all the players.

The analysis of gift giving illustrates how self-control problems can in¯uence choices. Becauseexpensive bottles of wine are `tempting', the couple rules them `o� limits' to help control spending. Forother tempting products, consumers may regulate their consumption in part by buying small quantitiesat a time, thus keeping inventories low. This practice creates the odd situation wherein consumers maybe willing to pay a premium for a smaller quantity. This behavior is studied by Wertenbroch (1996),who ®nds that the price premium for sinful products in small packages is greater than for moremundane goods. His one-sentence abstract succinctly sums up his paper: `To control their con-sumption, consumers pay more for less of what they like too much'.

* I am abstracting from natural discontinuities. If television shows come in increments of one hour, then one may have to choosean integer number of hours of TV watching, which alters the argument slightly.

R. H. Thaler Mental Accounting Matters 195

Wealth accountsAnother way of dealing with self-control problems is to place funds in accounts that are o�-limits.Hersh Shefrin and I have proposed (Shefrin and Thaler, 1988) that there is a hierarchy of moneylocations arranged by how tempting it is for a household to spend the money in each. The mosttempting class of accounts is in the `current assets' category, for example cash on hand and moneymarket or checking accounts. Money in these accounts is routinely spent each period. Less tempting tospend is money in the `current wealth' category, which includes a range of liquid asset accounts such assavings accounts, stocks and bonds, mutual funds, and so on. These funds are typically designated forsaving. Next in the hierarchy is home equity. Even though the advent of home equity loans has madethis category of funds somewhat less sacred, still most households aim to pay o� their mortgage by thetime they retire (and most succeed). Finally, in the least tempting category of funds lies the `futureincome' account. These funds include money that will be earned later in life (i.e. human capital) anddesignated retirement savings accounts such as IRAs and 401(k)s. According to our analysis, themarginal propensity to spend a dollar of wealth in the current income account is nearly 1.0, whereas thepropensity to spend a dollar of future income wealth is close to zero.

These predictions are in sharp contrast to standard economic theory of saving: the life-cycle model(Modigliani and Brumberg, 1954; Friedman, 1957). Here is a simpli®ed version that captures the spiritof the life-cycle model. Suppose a person has a certain remaining lifetime of N years, and that the rateof interest is zero. Let W be the person's wealth, equal to the sum of her assets, this year's income, andfuture (expected) income over the rest of her life. Consumption in this period is then equal to W/N.*Notice that in this model any change in wealth, DW, no matter what form it takes (e.g. a bonus at work,an increase in the value of one's home, even an inheritance expected in a decade), produces the samechange in current consumption namely DW/N. In other words, the theory assumes that wealth isperfectly fungible.

Shefrin and I proposed a modi®ed version of the life-cycle model, the behavioral life-cycle model,that incorporates the mental accounting temptation hierarchy described above. A powerful predictionof the mental accounting model is that if funds can be transferred to less tempting mental accounts theyare more likely to be saved. This insight can be used in designing government programs to stimulatesaving. According to the behavioral life-cycle model, if households can be persuaded to move some oftheir funds from the current income account to future income accounts, long-term savings willincrease. In other words, IRAs and 401(k)s are good vehicles to promote savings.{ My reading of theliterature on this topic is that this prediction is borne out. Households who contribute to retirementsavings plans display steady increases in the funds in these accounts with no apparent reduction in thefunds in other accounts. That is, they save more.{

Income accountingSo far we have considered violations of fungibility produced either by the budgeting process or by thelocation of funds. A third class of violations can be produced by the source of the income. O'Curry(1997) investigates this phenomenon. She ®rst has one group of subjects judge both sources and uses of

* More generally, in a world with uncertainty and positive interest rates, the life-cycle theory says that a person will spend theannuity value of his wealth in any period, that is, if he used W to buy a level annuity that paid y in every period, he would setconsumption equal to y. Bequests can also be accommodated.{ These accounts are especially good because not only are they less tempting `mental' accounts but they also have a penalty forwithdrawal that provides an additional incentive to leave the money in these accounts alone.{ See Poterba, Venti and Wise (1996) for a current summary of the evidence supporting my claim. Their results are hotlydisputed by Engen, Gale and Scholz (1996). One reason I side with the ®rst set of authors (aside from the fact that their resultssupport mental accounting) is that the simplest analyses show that the savings plans increase saving. Obtaining the oppositeresults seems to require a lot more work.

196 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

funds on a serious±frivolous scale: the winnings of an o�ce football pool are considered frivolouswhereas an income tax refund is serious; eating out is frivolous but paying the bills is serious. She thenasks other subjects to say what they would do with a particular windfall, such as $30 found in thepocket of a jacket in the back of the closet. She ®nds that people have a tendency to match theseriousness of the source of some windfall with the use to which it is put. Another example of incomenon-fungibility is provided by Kooreman (1997). He studies the spending behavior of families thatreceive child allowance payments from the Dutch government. He ®nds that spending on children'sclothing is much more sensitive to changes in the designated child allowance than to other incomesources.*

In the previous example the fact that the child allowance was labeled as such seemed to matter in theway people spent the money. Labeling e�ects are common. One surprising domain in which this ideacan be applied is dividend payments by corporations. Suppose a corporation is earning pro®ts andwishes to return some of these pro®ts to its shareholders. One (traditional) method is to pay a dividend.Another method is simply to repurchase shares. In a world with no taxes, these two methods areequivalent. But, if (as in the United States) dividends are taxed at a higher rate than capital gains, thentax-paying shareholders would prefer share repurchases to dividends (and those who have their sharesin non-taxable accounts are indi�erent). Under these conditions no ®rm should ever pay a dividend.

Why do ®rms pay dividends? Shefrin and Statman (1984) have proposed an explanation based onmental accounting. They argue that investors like dividends because the regular cash payment providesa simple self-control rule: spend the dividends and leave the principal alone. In this way, the dividendacts like an allowance. If, instead, ®rms simply repurchased their own shares, stockholders would notreceive a designated amount to spend, and would have to dip into capital on a period basis. Retirees(who tend to own high-dividend-paying stocks) might then worry that they would spend down theprincipal too quickly. A similar non-fungibility result is o�ered by Hatsopoulos, Krugman andPoterba (1989). Although capital gains in the stock market tend to have little e�ect on consumption,these authors found that when takeovers generate cash to the stockholders, consumption does increase.This is sometimes called the `mailbox e�ect'. When the check arrives in the mailbox it tends to getspent. Gains on paper are left alone.

CHOICE BRACKETING AND DYNAMIC MENTAL ACCOUNTING

A recurring theme of this paper is that choices are altered by the introduction of notional (but non-fungible) boundaries. The location of the parentheses matters in mental accounting Ð a loss hurts lessif it can be combined with a larger gain; a purchase is more likely to be made if it can be assigned to anaccount that is not already in the red; and a prior (sunk) cost is attended to if the current decision is inthe same account. This section elaborates on this theme by considering other ways in which boundariesare set, namely whether a series of decisions are made one at a time or grouped together (or `bracketed'to use the language of Read, Loewenstein and Rabin, 1998).

* There is a similar ®nding in public ®nance called the `¯ypaper e�ect'. When local governments receive earmarked payments forparticular kinds of expenditure (e.g. schools), they tend to increase their spending on that activity by the full amount of thegrant. Economic theory predicts that they would increase their spending only by the fraction of their income that they normallyspend on this activity. See Hines and Thaler (1995).

R. H. Thaler Mental Accounting Matters 197

Prior outcomes and risky choiceIn their prospect paper, Kahneman and Tversky mention the empirical ®nding that betting on longshots increases on the last race of the day,* when the average bettor is (i) losing money on the day, and(ii) anxious to break even. An interesting feature of this sunk cost e�ect is that it depends completely onthe decision to close the betting account daily. If each race were a separate account, prior races wouldhave no e�ect, and similarly if today's betting were combined with the rest of the bettor's wealth (oreven his lifetime of bets), the prior outcome would likely be trivial.

This analysis applies to other gambling decisions. If a series of gambles are bracketed together thenthe outcome of one gamble can a�ect the choices made later. Johnson and I investigated how prioroutcomes a�ect risky choice (Thaler and Johnson, 1990). Subjects were MBA students who played forreal money. The following three choices illustrate the type of problems studied. The percentage ofsubjects taking each option appears in brackets.

Problem 1. You have just won $30. Now choose between:(a) A 50% chance to gain $9 and a 50% chance to lose $9. [70](b) No further gain or loss. [30]

Problem 2. You have just lost $30. Now choose between:(a) A 50% chance to gain $9 and a 50% chance to loose $9. [40](b) No further gain or loss. [60]

Problem 3. You have just lost $30. Now choose between:(a) A 33% chance to gain $30 and a 67% chance to gain nothing. [60](b) A sure $10. [40]

These and other problems of this sort were used to investigate how prior outcomes a�ect riskychoices. Two results are worth noting. First, as illustrated by Problem 1, a prior gain can stimulate riskseeking in the same account. We called this phenomenon the `house money' e�ect since gamblers oftenrefer to money they have won from the casino as house money (the casino is known as `the house').Indeed, one often sees gamblers who have won some money early in the evening put that money into adi�erent pocket from their `own' money; this way each pocket is a separate mental account. Second, asillustrated by Problems 2 and 3, prior losses did not stimulate risk seeking unless the gamble o�ered achance to break even.

The stakes used in the experiments just described were fairly large in comparison to most laboratoryexperiments, but small compared to the wealth of the participants. Limited experimental budgets are afact of life. Gertner (1993) has made clever use of a set of bigger stakes choices over gambles made bycontestants on a television game show called `Card Sharks'.{ The choices Gertner studies were the lastin a series of bets made by the winner of the show that day. The contestant had to predict whether acard picked at random from a deck would be higher or lower than a card that was showing. Aces arehigh and ties create no gain or loss. The odds on the bet therefore vary from no risk (when the showingcard is a 2 or an Ace) to roughly 50±50 when the up-card is an 8. After making the prediction, thecontestant then can make a bet on the outcome, but the bet must be between 50% and 100% of theamount she has won on the day's show (on average, about $3000). Ignoring the sure bets, Gertnerestimates a Tobit regression model to predict the size of the contestant's bet as a function of the cardshowing (the odds), the stake available (that is, today's winnings), and the amount won in previousdays on the show. After controlling for the constraint that the bet must lie between 50% and 100% of

* That is, long shots become even worse bets at the end of the day. They are always bad bets. See Thaler and Ziemba (1988).{ See also Biswanger (1981), who obtains similar results. He also was able to run high stakes experiments by using subjects inrural villages in India.

198 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

the stake, Gertner ®nds that today's winnings strongly in¯uences on the amount wagered.* In contrast,prior cash won has virtually no e�ect. This ®nding implies that cash won today is treated in a di�erentmental account from cash won the day before.{ This behavior is inconsistent with any version ofexpected utility theory that treats wealth as fungible.

Narrow framing and myopic loss aversionIn the gambling decisions discussed above, the day of the experiment suggested a natural bracket.Often gambles or investments occur over a period of time, giving the decision-maker considerable¯exibility in how often to calculate gains and losses. It will come as no surprise to learn that the choiceof how to bracket the gambles in¯uences the attractiveness of the individual bets. An illustration isprovided by a famous problem ®rst posed by Paul Samuelson. Samuelson, it seems, was having lunchwith an economist colleague and o�ered his colleague an attractive bet. They would ¯ip a coin, and ifthe colleague won he would get $200; if he lost he would have to pay only $100. The colleague turnedthis bet down, but said that if Samuelson would be willing to play the bet 100 times he would be game.Samuelson (1963) declined to o�er this parlay, but went home and proved that this pair of choices isirrational.{

There are several points of interest in this problem. First, Samuelson quotes his colleague's reasoningfor rejecting the single play of the gamble: `I won't bet because I would feel the $100 loss more thanthe $200 gain'. Modern translation: `I am loss averse'. Second, why does he like the series of bets?Speci®cally, what mental accounting operation can he be using to make the series of bets attractivewhen the single play is not?

Suppose Samuelson's colleague's preferences are a piecewise linear version of the prospect theoryvalue function with a loss aversion factor of 2.5:

U�x� � x x 5 02:5x x 5 0

Because the loss aversion coe�cient is greater than 2, a single play of Samuelson's bet is obviouslyunattractive. What about two plays? The attractiveness of two bets depends on the mental accountingrules being used. If each play of the bet is treated as a separate event, then two plays of the gamble aretwice as bad as one play. However, if the bets are combined into a portfolio, then the two-bet parlay{$400, 0.25; 100, 0.50; ÿ$200, 0.25} yields positive expected utility with the hypothesized utilityfunction, and as the number of repetitions increases the portfolio becomes even more attractive. SoSamuelson's colleague should accept any number of trials of this bet strictly greater than one as long ashe does not have to watch!

More generally, loss-averse people are more willing to take risks if they combine many bets togetherthan if they consider them one at a time. Indeed, although the puzzle to Samuelson was why hiscolleague was willing to accept the series of bets, the real puzzle is why he was unwilling to play one.Risk aversion cannot be a satisfactory explanation if his colleague has any signi®cant wealth. Forexample, suppose Samuelson's colleague's utility function is U(W) � 1n W and his wealth is a modest

* Gertner o�ers the following example to illustrate this di�erence. Suppose a ®rst-time contestant has won $5000 so far and has aJack showing, so a bet of `lower' o�ers 3±1 odds. (She loses with an A, K, or Q, ties with a J, and wins otherwise.) The regressionpredicts a bet of $2800. Compare this contestant to one who has won only $3000 today but won $2000 the previous day.Although their winnings on the show are identical, this player is predicted to bet only $1544.{ This result is all the more striking because `yesterday's' show was probably taped just an hour before `today's' (several showsare taped in the same day) and `yesterday's' winnings have certainly not been collected.{ Speci®cally, he showed that an expected utility maximizer who will not accept a single play of a gamble for any wealth levelthat could obtain over a series of such bets will not accept the series. For a more general result, see Tversky and Bar Hillel (1983).

R. H. Thaler Mental Accounting Matters 199

$10,000. In that case he should be willing to risk a 50% chance of losing $100 if he had a 50% chance togain a mere $101.01! Similar results obtain for other reasonable utility functions. In fact, Rabin (1998)shows that expected utility theory implies that someone who turns down Samuelson's bet should alsoturn down a 50% chance to lose $200 and a 50% chance to win $20,000. More generally, he shows thatexpected utility theory requires people to be virtually risk neutral for `small' bets. To explain the factthat many people do reject attractive small bets (such as Samuelson's), we need a combination of lossaversion and one-bet-at-a-time mental accounting.

Benartzi and I (1995) use the same analysis to o�er a mental accounting explanation for whateconomists call the equity premium puzzle (Mehra and Prescott, 1985). The equity premium is thedi�erence in the rate of return on equities (stocks) and a safe investment such as treasury bills. Thepuzzle is that this di�erence has historically been very large. In the USA the equity premium has beenroughly 6% per year over the past 70 years. This means that a dollar invested in stocks on 1 January1926 was worth more than $1800 on 1 January 1998, whereas a dollar invested in treasury bills wasworth only about $15 (half of which was eaten up by in¯ation). Of course, part of this di�erence can beattributed to risk, but what Mehra and Prescott show is that the level of risk aversion necessary toexplain such a large di�erence in returns is implausible.*

To explain the puzzle we note that the risk attitude of loss-averse investors depends on the frequencywith which they reset their reference point, i.e. how often they `count their money'. We hypothesizethat investors have prospect theory preferences (using parameters estimated by Tversky andKahneman, 1992).{ We then ask how often people would have to evaluate the changes in theirportfolios to make them indi�erent between the (US) historical distributions of returns on stocks andbonds? The results of our simulations suggest that the answer is about 13 months. This outcomeimplies that if the most prominent evaluation period for investors is once a year, the equity premiumpuzzle is `solved'.

We refer to this behavior as myopic loss aversion. The disparaging term `myopic' seems appropriatebecause the frequent evaluations prevent the investor from adopting a strategy that would be preferredover an appropriately long time horizon. Indeed, experimental evidence supports the view that when along-term horizon is imposed externally, subjects elect more risk. For example, Gneezy and Potters(1997) and Thaler et al. (1997) ran experiments in which subjects make choices between gambles(investments). The manipulations in these experiments are the frequency with which subjects getfeedback. For example, in the Thaler et al. study, subjects made investment decisions between stocksand bonds at frequencies that simulated either eight times a year, once a year, or once every ®ve years.The subjects in the two long-term conditions invested roughly two-thirds of their funds in stocks whilethose in the frequent evaluation condition invested 59% of their assets in bonds. Similarly, Benartziand I (forthcoming) asked sta� members at a university how they would invest their retirement moneyif they had to choose between two investment funds, A and B, one of which was based on stock returns,the other on bonds. In this case the manipulation was the way in which the returns were displayed. Onegroup examined a chart showing the distribution of one-year rates of return, and the other group wasshown the simulated distribution of 30-year rates of return. Those who saw the one-year returns saidthey would invest a majority of their funds in bonds, whereas those shown the 30-year returns invested90% of their funds in stocks.{

* They estimate that it would take a coe�cient of relative risk aversion of about 40 to explain the history equity premium. Incontrast, a log utility function has a coe�cient of 1.{ Speci®cally, the value function is: v(x) � xa if x 5 0 ÿl�ÿx�b if x 5 0 where l is the coe�cient of loss aversion. They haveestimated a and b to be 0.88 and l to be 2.25. We also use their rank-dependent weighting function. For details see Benartzi andThaler (1995).{ Similar results for gambles are also obtained by Keren and Wagenaar (1987) and Redelmeier and Tversky (1992).

200 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

Myopic loss aversion is an example of a more general phenomenon that Kahneman and Lovallo(1993) call narrow framing; projects are evaluated one at a time, rather than as part of an overallportfolio. This tendency can lead to an extreme unwillingness to take risks. I observed an interestingillustration of this phenomenon while teaching a group of executives from one ®rm, each of whom wasresponsible for managing a separate division. I asked each whether he would be willing to undertake aproject for his division if the payo�s were as follows: 50% chance to gain $2 million, 50% chance tolose $1 million. Of the 25 executives, three accepted the gamble. I then asked the CEO, who was alsoattending the session, how he would like a portfolio of 25 of these investments. He nodded enthusi-astically. This story illustrates that the antidote for excessive risk aversion is aggregation, either acrosstime or across di�erent divisions.

The examples discussed so far show that narrow bracketing can inhibit risk taking. Narrowbracketing can also have other perverse side-e�ects. For example, Camerer et al. (1997) study the dailylabor supply decisions of New York City taxi drivers. In New York, as in many cities, the cab driverstypically rent their cars for a 12-hour period for a ®xed fee. They are then entitled to keep all therevenues they earn during that half-day. Since 12 hours is a long time to drive a car, especially in NewYork City, the drivers must decide each day how long to drive; that is, whether to keep the car for thefull 12 hours or quit earlier. This decision is complicated by the fact that there is more demand for theirservices on some days than others (because of di�erences in weather or the presence of a bigconvention, for example). A rational analysis would lead drivers to work longer hours on busy days, asthis policy would maximize earnings per hour worked. If, instead, drivers establish a target earningslevel per day, they will tend to quit earlier on good days. This is precisely what Camerer et al. ®nd. Theelasticity of hours worked with respect to the daily wage (as measured by the earnings of other driversthat day) is strongly negative. The implication is that taxi drivers do their mental accounting one day ata time.*

The diversi®cation heuristicThe unit of analysis can also in¯uence how much variety consumers elect. This e�ect was ®rstdemonstrated by Simonson (1990). He gave students the opportunity to select among six snacks (candybars, chips, etc.) in one of two conditions: (a) sequential choice: they picked one of the six snacks ateach of three class meetings held a week apart; (b) simultaneous choice: on the ®rst class meeting theyselected three snacks to be consumed one snack per week over the three class meetings. Simonsonobserved that in the simultaneous choice condition subjects displayed much more variety seeking thanin the sequential choice condition. For example, in the simultaneous choice condition 64% of thesubjects chose three di�erent snacks whereas in the sequential choice condition only 9% of the subjectsmade this choice. Simonson suggests that this behavior might be explained by variety seeking servingas a choice heuristic. That is, when asked to make several choices at once, people tend to diversify. Thisstrategy is sensible under some circumstances (such as when eating a meal Ð we typically do not orderthree courses of the same food), but can be misapplied to other situations, such as sequential choice.This mistake represents a failure of predicted utility to accurately forecast subsequent experiencedutility. Many students who liked Snickers best elected that snack each week when they picked one weekat a time, but went for variety when they had to choose in advance.

This result has been called the `diversi®cation bias' by Read and Loewenstein (1995). They demon-strate the role of choice bracketing in an ingenious experiment conducted on Halloween night. The`subjects' in the experiment were young trick-or-treaters who approached two adjacent houses. In one

* Rizzo and Zeckhauser (1998) ®nd a similar result for physicians whose evaluation period appears to be one year rather thanone day.

R. H. Thaler Mental Accounting Matters 201

condition the children were o�ered a choice between two candies (Three Musketeers and Milky Way)at each house. In the other condition they were told at the ®rst house they reached to `choose whichevertwo candy bars you like'. Large piles of both candies were displayed to assure that the children wouldnot think it rude to take two of the same. The results showed a strong diversi®cation bias in thesimultaneous choice condition: every child selected one of each candy. In contrast, only 48% of thechildren in the sequential choice condition picked di�erent candies. This result is striking, since ineither case the candies are dumped into a bag and consumed later. It is the portfolio in the bag thatmatters, not the portfolio selected at each house.

The diversi®cation bias is not limited to young people choosing among snacks. Benartzi and I (1998)have found evidence of the same phenomenon by studying how people allocate their retirement fundsacross various investment vehicles. In particular, we ®nd some evidence for an extreme version of thisbias that we call the 1/n heuristic. The idea is that when an employee is o�ered n funds to choose fromin her retirement plan, she divides the money evenly among the funds o�ered. Use of this heuristic, orothers only slightly more sophisticated, implies that the asset allocation an investor chooses will dependstrongly on the array of funds o�ered in the retirement plan. Thus, in a plan that o�ered one stock fundand one bond fund, the average allocation would be 50% stocks, but if another stock fund were added,the allocation to stocks would jump to two thirds. We ®nd evidence supporting just this behavior. In asample of pension plans we regress the percentage of the plan assets in stocks on the percentage of thefunds that are stock funds and ®nd a very strong relationship.

We also ®nd that employees seem to put stock in the company they work for into a separate mentalaccount. For companies that do not o�er their own stock as one of the options in the pension plan theemployees invest 49% of their money in bonds and 51% in stocks. When the company stock is includedin the plan this investment attracts 42% of the funds. If the employees wanted to attain a 50% equityexposure, they would invest about 8% of the rest of their funds in stocks, the rest in bonds. Instead theyinvest their non-company stock funds evenly: 29% in stocks, 29% in bonds.

DISCUSSION

My own thinking about mental accounting began with an attempt to understand why people payattention to sunk costs, why people are lured by bargains into silly expenditures, and why people willdrive across town to save $5 on a small purchase but not a large one. I hope this paper has shown thatwe have learned quite a bit about these questions, and in so doing, the researchers working in this areahave extended the scope of mental accounting far beyond the original set of questions I had set out toanswer. Consider the range of questions that mental accounting helps us answer:

. Why do ®rms pay dividends?

. Why do people buy time-share vacation properties?

. Why are ¯at-rate pricing plans so popular

. Why do sales contests have luxuries (instead of cash) as prizes?

. Why do 401(k) plans increase savings?

. Why do stocks earn so much higher a return than bonds?

. Why do people decline small-stakes attractive bets?

. Why can't you get a cab on a rainy day? (hint: cab drivers earn more per hour on rainy days).

A question that has not received much attention is whether mental accounting is good for us. What isthe normative status of mental accounting? I see no useful purpose in worrying about whether or notmental accounting is `rational'. Mental accounting procedures have evolved to economize on time andthinking costs and also to deal with self-control problems. As is to be expected, the procedures do not

202 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

work perfectly. People pay attention to sunk costs. They buy things they don't need because the deal istoo good to pass up. They quit early on a good day. They put their retirement money in a moneymarket account.

It is not possible to say that the system is ¯awed without knowing how to ®x it. Given that optim-ization is not feasible (too costly) repairing one problem may create another. For example, if we teachpeople to ignore sunk costs, do they stop abiding by the principle: `waste not, want not'? If we stopbeing lured by good deals, do we stop paying attention to price altogether? There are no easy answers.

Those interested in improving individual decision making can do more work on mental accountingas a prescriptive device. How can mental accounting rules be modi®ed to achieve certain goals?* Forexample, Jonathan Clements, the author of a regular column for new investors in the Wall StreetJournal{ called `Getting Going' invited readers to submit tips on how to do a better job of saving andinvesting. Many of the tips he later published had a strong mental accounting ¯avor. A reader calledDavid Guerini submitted the following advice:

I started a little `side' savings account eight years ago. During the day, I try to accumulate change. IfI spend $4.50 at a store, I give the cashier a $5 bill, even if I have 50 cents in my pocket. At the end ofeach day, the money is put aside. If I have no change, I put a $1 bill aside. I add income-tax refunds,money from products I purchased and returned for a refund, and all those annoying little mail-inrebates they give you when you purchase batteries, shaving cream, and so on. I end up painlesslysaving between $500 and $1000 each year.

An economist might argue that it would be even less painful to just write a check once a year andsend it to his mutual fund. But that would miss the point: mental accounting matters.

ACKNOWLEDGEMENTS

I have been thinking about mental accounting for more than 20 years, so it is not possible to thankeveryone who has helped me write this paper. Some who have helped recently include John Gourville,Chip Heath, Daniel Kahneman, France Leclerc, George Loewenstein, Cade Massey, Drazen Prelec,Dilip Soman, and Roman Weil. This paper began as an invited lecture to the SPUDM conference inAix-en-Provence held in 1993. It was ®nally completed during my stay at The Center for AdvancedStudy in the Behavioral Sciences. Their help in reaching closure is gratefully acknowledged.

REFERENCES

Ainslie, G. `Special reward: A behavioral theory of impulsiveness and impulse control', Psychological Bulletin,82(4) (1975), 463±496.

Arkes, H. R. and Blumer, C. `The psychology of sunk cost', Organizational Behavior and Human DecisionProcesses, 35(1) (1985), 124±140.

Benartzi, S. and Thaler, R. H. `Myopic loss-aversion and the equity premium puzzle', Quarterly Journal ofEconomics, CX (1995), 75±92.

Benartzi, S. and Thaler, R. H. `Risk aversion or myopia: choices in repeated gambles and retirement investments',Management Science, forthcoming.

* Along these lines, Read, Loewenstein and Rabin (1998) have a useful discussion of when broad bracketing works better thannarrow bracketing. Short answer: usually.{ See his column on 20, 24, and 31 January 1998.

R. H. Thaler Mental Accounting Matters 203

Benartzi, S. and Thaler, R. H. `Illusory diversi®cation and retirement savings, working paper', University ofChicago and UCLA, (1998).

Biswanger, H. `Attitudes toward risk: theoretical implications of an experiment in rural India', Economic Journal,XCI (1981), 867±890.

Burgstahler, D. and Dichev, I. `Earnings management to avoid earnings decreases and losses', Journal ofAccounting and Economics, 24(December) (1997), 99±126.

Camerer, C., Babcock, L., Loewenstein, G. and Thaler, R. H. `Labor supply of New York City cabdrivers: oneday at a time', Quarterly Journal of Economics, CXII(May) (1997), 407±442.

Degeorge, F., Patel, J. and Zeckhauser, R. J. `Earnings management to exceed thresholds', Journal of Business.Engen, E. M., Gale, W. G. and Scholz, J. K. `The illusory e�ects of saving incentive programs', Journal ofEconomics Perspectives, 10(Fall) (1996), 113±138.

Feenberg, D. and Skinner, J. `Sources of IRA saving', in Summers, L. (ed.), Tax Policy and the Economy: Vol. 3(pp. 25±46), Cambridge, MA: MIT Press, 1989.

Feinberg, R. A. `Credit cards as spending facilitating stimuli: a conditioning interpretation', Journal of ConsumerResearch, 12 (1986), 304±356.

Fishburn, P. and Luce, D. `Joint receipt and Thaler's hedonic editing rule', Mathematical Social Science, 29(1995), 33±76.

Friedman, M. A Theory of Consumption Function, Princeton, NJ: Princeton University Press, 1957.Gertner, R. `Game shows and economic behavior: risk taking on “card sharks'' ', Quarterly Journal of Economics,CVI (1993), 507±521.

Gneezy, U. and Potters, J. `An experiment on risk taking and evaluating periods', Quarterly Journal ofEconomics, CXII(May) (1997), 631±646.

Gourville, J. T. `Pennies-a-day: the e�ect of temporal re-framing on transaction evaluation', Journal of ConsumerResearch, 24 (1998), 395±408.

Gourville, J. T. and Soman, D. `Payment depreciation: the e�ects of temporally separating payments fromconsumption', Journal of Consumer Research (1998).

Hatsopoulous, G. N., Krugman, P. R. and Poterba, J. M. `Overconsumption: the challenge to U.S. economicpolicy', American Business Conference (1989).

Heath, C. `Escalation and de-escalation of commitment in response to sunk costs: the role of budgeting in mentalaccounting', Organizational Behavior and Human Decision Processes, 62 (1995), 38±54.

Heath, C. and Soll, J. B. `Mental accounting and consumer decisions', Journal of Consumer Research, 23 (1996),40±52.

Henderson, P. and Peterson, R. `Mental accounting and categorization', Organizational Behavioral and HumanDecision Processes, 51 (1992), 92±117.

Hines, J. and Thaler, R. H. `Anomalies: the ¯ypaper e�ect', Journal of Economics Perspectives, 9 (1995), 217±226.Kahneman, D. `Reference points, anchors, norms, and mixed feelings', Organizational Behavior and HumanDecision Processes, 51 (1992), 296±312.

Kahneman, D. `New challenges to the rationality assumption', Journal of Institutional and Theoretical Economics,150 (1994), 43±56.

Kahneman, D., Knetsch, J. and Thaler, R. H. `Fairness as a constraint on pro®t-seeking: entitlements in themarket', American Economic Review, 76 (1986), 728±741.

Kahneman, D. and Lovallo, D. `Timid choices and bold forecasts: a cognitive perspective on risk taking',Management Science, 39 (1993), 17±31.

Kahneman, D. and Tversky, A. `Prospect theory: an analysis of decision under risk', Econometrica, 47 (1979),263±291.

Kahneman, D. and Tversky, A. `Choices, values, and frames', The American Psychologist, 39 (1984), 341±350.Keren, G. `Additional tests of utility theory under unique and repeated conditions', Journal of Behavioral DecisionMaking, 4 (1991), 297±304.

Keren, G. and Wagenaar, W. A. `Violation of utility theory in unique and repeated gambles', Journal ofExperimental Psychology: Learning, Memory and Cognition, 13 (1987), 387±391.

Kooreman, P. `The labeling e�ect of a child bene®t system', unpublished working paper, University ofGroningen, (1997).

Leclerc, F., Schmidt, B. and Dube, L. `Decision making and waiting time: is time like money?', Journal ofConsumer Research, 22 (1995), 110±119.

Linville, P. and Fischer, G. W. `Preferences for separating or combining events', Journal of Personality and SocialPsychology, 60 (1991), 5±23.

204 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3

Mehra, R. and Prescott, E. C. `The equity premium: a puzzle', Journal of Monetary Economics, 15 (1985),145±161.

Modigiliani, F. and Brumberg, R. `Utility analysis and the consumption function: an interpretation of cross-section data', in Kurihara, K. K. (ed.), Post Keynesian Economics, New Brunswick, NJ: Rutgers UniversityPress, 1954.

O'Curry, S. `Income source e�ects', Unpublished working paper, DePaul University, (1997).Odean, T. `Are investors reluctant to realize their losses', Journal of Finance, (October 1998).Poterba, J. M., Venti, S. F. and Wise, D. A. `How retirement savings programs increase saving', Journal ofEconomics Perspectives, 10(Fall) (1996), 91±112.

Prelec, D. and Loewenstein, G. `The red and the black: mental accounting of savings and debt', MarketingScience, 17 (1998), 4±28.

Prelec, D. and Simester, D. `Always leave home without it', working paper, MIT Sloan School, March (1998).Rabin, M. `Risk aversion, diminishing marginal utility, and expected-utility theory: a calibration theorem',working paper, Department of Economics, University of California, Berkeley (1998).

Read, D. and Loewenstein, G. `Diversi®cation bias: explaining the discrepancy in variety seeking betweencombined and separated choices', Journal of Experimental Psychology: Applied, 1 (1995), 34±49.

Read, D., Loewenstein, G. and Rabin, M. `Choice bracketing', unpublished working paper, Carnegie MellonUniversity, 1998.

Redelmeier, D. A. and Tversky, A. `On the farming of multiple prospects', Psychological Science, 3(3) (1992),191±193.

Rizzo, J. A. and Zeckhauser, R. J. `Income targets and physician behavior', JFK School of Government, HarvardUniversity working paper, March 1998.

Samuelson, P. A. `Risk and uncertainty: a fallacy of large numbers', Scientia, (1963), 108±113.Savage, L. J. The Foundations of Statistics, New York: John Wiley, 1954.Sha®r, E. and Thaler, R. H. `Invest now, drink later, spend never: the mental accounting of advanced purchases',working paper, Graduate School of Business University of Chicago (1998).

Shefrin, H. M. and Statman, M. `Explaining investor preference for cash dividends', Journal of FinancialEconomics, 13 (1984), 253±282.

Shefrin, H. M. and Statman, M. `The disposition to sell winners too early and ride losers too long', Journal ofFinance, 40 (1987), 777±790.

Shefrin, H. H. and Thaler, R. H. `The behavioral life-cycle hypothesis', Economic Inquiry, 26(October) (1988),609±643.

Simonson, I. `The e�ect of purchase quantity and timing on variety-seeking behavior', Journal of MarketingResearch, 28 (1990), 150±62.

Soman, D. `Contextual e�ects of payment mechanism on purchase intention: check or charge?', Unpublishedworking paper, University of Colorado, 1997.

Thaler, R. H. `Toward a positive theory of consumer choice', Journal of Economic Behavior and Organization, 1(1980), 39±60.

Thaler, R. H. `Mental accounting and consumer choice', Marketing Science, 4 (1985), 199±214.Thaler, R. H. `Saving, fungibility and mental accounts', Journal of Economic Perspectives, 4 (1990), 193±205.Thaler, R. H. and Johnson, E. J. `Gambling with the house money and trying to break even: the e�ects of prioroutcomes on risky choice', Management Science, 36(6) (1990), 643±660.

Thaler, R. H. and Shefrin, H. M. `An economic theory of self-control', Journal of Political Economy, 39 (1981),392±406.

Thaler, R. H., Tverskey, A., Kahneman, D. and Schwartz, A. `The e�ect of myopia and loss aversion on risktaking: an experimental test', Quarterly Journal of Economics, CXII (1997), 647±661.

Thaler, R. H. and Ziemba, W. `Pari-mutual betting markets: racetracks and lotteries', Journal of EconomicPerspectives, 2 (1988), 161±174.

Train, K. E. Optimal Regulation, Cambridge, MA: MIT Press, 1991.Tversky, A. and Bar-Hillel, M. `The long run and the short', Journal of Experimental Psychology: HumanLearning, Memory, and Cognition, 9 (1983), 713±717.

Tversky, A. and Kahneman, D. `The framing of decisions and the rationality of choice', Science, 211 (1981),453±458.

Tversky, A. and Kahneman, D. `Advances in prospect theory: cumulative representation of uncertainty', Journalof Risk and Uncertainty, 5 (1992), 297±323.

Wertenbroch, K. `Consumption self-control via purchase rationing', working paper, Yale University, November1996.

R. H. Thaler Mental Accounting Matters 205

Zelizer, V. A. The Social Meaning of Money: Pin Money, Paychecks, Poor Relief, and Other Currencies, NewYork: Basic Books, 1994.

Author's address:Richard H. Thaler, Graduate School of Business, University of Chicago, 1101 East 58th Street, Chicago, IL 60637,USA.

206 Journal of Behavioral Decision Making Vol. 12, Iss. No. 3