OUP user menu


Darren Dahl (Editor in Chief)Eileen FischerGita JoharVicki Morwitz

16 out of 115

Losers, Winners, and Biased Trades

Joseph Johnson, Gerard J. Tellis, Deborah J. Macinnis
DOI: http://dx.doi.org/10.1086/432241 324-329 First published online: 1 September 2005


When faced with sequential information, consumers tend to fall prey to one of two well-known heuristics: the hot (or cold) hand and the gambler's fallacy. The authors relate these two traditionally separate heuristics to differences in accepting (buy) versus rejecting (sell) decisions. They identify trend length as a contextual moderating variable and show an asymmetry between buying and selling frames. When applied to a stock market context, a consistent finding is that consumers prefer to buy past winners and sell past losers even when neither should be preferred. This behavior violates the normative rule of buy low and sell high.

  • Inference Making
  • Behavioral Decision Theory
  • Judgment and Decision Making
  • Economic Theories and Analysis

Consumers are often influenced by information sequences when making decisions under uncertainty. They bet on gambles, athletes, or horses based on their past performances (Gilovich, Vallone, and Tversky 1985), buy lottery tickets based on performances of the past weeks' numbers (Clotfelter and Cook 1989), or buy stocks based on rising prices (Andreassen 1988).

When facing sequential information, consumers tend to fall prey to one of two well-known heuristics: the hot (cold) hand and the gambler's fallacies. The former occurs when individuals project a rising (declining) trend in a random sequence (e.g., DeBondt 1993), the latter when individuals project a reversal in a random sequence (Tversky and Kahneman 1971). Both these heuristics arise from consumers' misunderstanding of random events due to a belief that small samples are representative of the underlying process (Kahneman and Tversky 1972).

We posit that which of these heuristics occurs depends on whether consumers are buying or selling and the length of the trend of information. Our study examines these ideas in the context of stock trading. We contribute to existing theory in four ways: we (a) link two traditionally separate heuristics in predicting outcomes (the hot hand and the gambler's fallacy); (b) suggest that purchase frame (accept [buy] vs. reject [sell] decisions) and trend length moderate occurrence of these heuristics; (c) show an asymmetry between the buying and selling frames; and (d) show that consumers prefer to buy stocks with winning trends and sell those with losing trends, even when such trends carry no relevant information.

Our thesis differs from the current findings in the literature. For example, Morrin et al. (2002) explore the differences between analysts who follow momentum buying and those who buck trends, but they do not show how or when these two types of buying are linked. Barberis, Shleifer, and Vishny (1998) model consumers as having either a mean-reverting or a continuation mind-set but do not model how the two mind-sets are related. Shefrin and Statman (1985) and Odean (1998) show that at the aggregate and transaction levels, respectively, consumers sell winners too early and ride losers too long. This pattern runs contrary to our major thesis. The next sections detail the theory, experiment, results, and contributions of the research in the context of stock trading.

Consumers' Processing of Sequential Information

Sequences of information about stocks' performances frequently exhibit trends of varying length. We use the terms “increasing trend” and “decreasing trend” to refer to a sequence of increasing (e.g., 2, 4, 6) or decreasing positive earnings (e.g., 6, 4, 2), respectively. We use the term “trend length” to refer to the total number of periods in the trend.

Consumers have a propensity to impute trends in data that are essentially random. When a random sequence shows a pattern, consumers assign meaning to it. Gilovich (1991) explains that once persons (mis)identify a random pattern as real, they integrate it into their preexisting theories, which biases the evaluation of new information in the direction of the initial belief. Two such theories are the hot (cold) hand and the gambler's fallacies, which we describe below.

Gilovich et al. (1985) found that viewers perceive a basketball player to have a better chance of a basket after a string of successful baskets. Although the probability of making a basket was not significantly different from the player's overall average, viewers perceived the player to have a hot hand. Analogously, viewers perceived a player with a string of unsuccessful baskets to have a cold hand. The hot (cold) hand phenomenon is analogous to overreaction and momentum investing in behavioral finance. DeBondt and Thaler (1985) argue that consumers who rely on past information become overly optimistic (pessimistic) about past winners (losers). The hot (cold) hand would suggest that consumers infer that the future value of a stock is greater (lower) when the sequence of past earnings shows an increasing (decreasing) as opposed to a decreasing (increasing) trend. Hence, they would buy stocks that show an increasing trend of earnings and sell stocks that show a decreasing trend of earnings.

The hot hand fallacy seems to conflict with the gambler's fallacy, which occurs when consumers expect a reversal in a losing but essentially random sequence (Tversky and Kahneman 1971). The gambler's fallacy may be the underlying cause of Shefrin and Statman's (1985) finding at the aggregate level and Odean's (1998) finding at the transaction level that consumers hold on to losing stocks too long and sell winning stocks too fast. In both cases, consumers expect a reversal in random events. Just like the hopeful gambler, consumers hold on to losing stocks, expecting the string of losses to reverse and let them recoup their losses.

Which of these two heuristics dominates? Research on consumers' information processing shows that “trends are central to the expectation formation process” (Oliver and Winer 1987, 481). As such, consumers are sensitive to trend length and direction and use this information when predicting the future. Trend length thus may moderate the impact of buying versus selling decisions on use of the hot/cold hand and gambler's fallacies, as explained below.

Lakonishok, Shleifer, and Vishny (1994) show that when buying, consumers rely on earnings and sales growth as valuation guides. Consumers observe a rising trend, say in earnings, and conclude that the stock is of good value and worth buying. Further increases in value confirm their forecasts and feed their enthusiasm for the stock. This response is consistent with research on consumers' search for hypothesis confirming evidence (Einhorn and Hogarth 1978). As the sequence of earnings continues up, consumers value the stock higher and buy more of it. Examples of such behavior are well documented (e.g., Zeckhauser, Hendricks, and Patel 1993) and are consistent with Shafir's (1993) work that shows that positive information (upward trends) has a greater impact than negative information (downward trends) for accept (buy) decisions.

However, because people believe that the laws of probability preclude very long sequences (Tversky and Kahneman 1971), we posit that when faced with a long positive run, they may expect a higher likelihood of a reversal. They disfavor a stock whose earnings show a very long positive run and are less likely to buy it. Hence, we hypothesize the following:

  • H1: Consumers facing stocks with positive and negative earnings trends will be more likely to (a) buy a stock with a positive (vs. a negative) trend as trend length initially increases, (b) though this tendency will decrease as trend length increases further.

Now, consider consumers selling stocks. A sequence of declining earnings may cause consumers to commit the cold hand fallacy and lead them to predict a further decline. Shefrin (2000) provides evidence consistent with this notion, showing how bearishness increases after the market falls. This effect is also consistent with Shafir's (1993) finding that negative information (e.g., losses) has a greater impact than positive information (gains) in reject (sell) decisions. However, as trend length increases, consumers may predict that the losing stock's bad run will change and that a price rise is due (consistent with the gambler's fallacy). As such, they may decide to keep a losing stock and sell a winning stock. Hence,

  • H2: Consumers faced with stocks with positive and negative earnings trends will be more likely to (a) sell a stock with a negative (vs. positive) earnings trend as trend length initially increases, (b) though this tendency will decrease as trend length increases further.

Buying and selling decisions constitute two opposing mental accounts (Thaler 1985) and can create asymmetric preferences. We anticipate that the effects predicted for hypothesis 1 (for buying) will be of greater magnitude than those predicted for hypothesis 2 (for selling).

One reason for an asymmetric effect may be the complexity of selling versus buying (Wedell 1997; Meloy and Russo 2004). With buying, one has to focus only on what one will acquire. Any anticipated regret in buying involves regret only with buying a stock that loses value. Selling is a cognitively more complex task as it involves an assessment of what one will keep and what one will sell. One must deal with anticipated regret from selling a winning stock that continues to go up as well as the anticipated regret from keeping a losing stock that continues to fall. This complexity may lead to a greater variance in the use of heuristics. One consumer might prefer to sell the winning stock because she believes it has reached its peak. Another might prefer to sell the losing stock because she believes it is cold. A third might prefer to sell the winning stock because he believes the losing stock will start to gain value, while a fourth might prefer to sell the losing stock because she believes the winning stock will continue to go up. If decision complexity drives the asymmetry in buying versus selling, we would see greater variation in using heuristics in the sell condition. Hence,

  • H3: Consumers in the buy condition will show stronger preferences for the winning stock over the losing stock than consumers in the sell condition will show preferences for the losing stock over the winning stock.

A consequence of hypothesis 3 is that on average consumers buy winners and sell losers. The asymmetric effect hypothesized in hypothesis 3 shows that across trend length, deviations from indifference are stronger in the buy than the sell condition.

In contrast to the above, Fama's efficient market hypothesis (Fama 1970, 1991) suggests that trends in prices of stocks carry no useful information to consumers. The reason is that stock prices adjust instantaneously to new information, fully incorporating the discounted future value of that information. Surveys of the extensive studies on the efficient market hypothesis suggest that the vast majority were unable to reject the hypothesis for common stocks (Fama 1970, 1991, 1998). Similarly, researchers in finance treat earnings per share as a pure random walk (Barberis et al. 1998; Bloomfield and Hales 2002). Additionally, research in accounting (Feltham and Ohlson 1995; Ohlson 1991) shows that, given the time-series properties of earning, the true value of a firm also follows a random walk (Bloomfield and Hales 2002). Consumers should therefore not see useful information from earnings sequences. For these reasons, the Securities and Exchange Commission (SEC) requires that ads that show past trends carry a disclaimer that reads that past performance is no guarantee for future results.


We examined hypotheses 1–3 in a three trend lengths (3, 7, 11) × two trade types (buy, sell) between-subjects factorial experiment with 314 business school students randomly assigned to the six conditions. Respondents in the buy condition were told that they had received a tax rebate of $1,000 that they intended to invest in the stock market. Those in the sell condition were told that they had inherited a small portfolio from which they were planning to use $1,000 to buy a computer. The stocks were identical except for their trend length. Participants saw two stocks, one with an increasing trend and the other with a decreasing trend of either 3, 7, or 11 periods. We measured preference for the winning versus losing stock on a nine-point relative preference scale (1 = preference for the stock with the decreasing earnings, 9 = preference for the stock with the increasing earnings, 5 = ambivalence). Two judges coded the reasons respondents gave for their relative preferences into categories (intercoder agreement = .98).



A 2 × 3 ANOVA on the preference variable showed a main effect of buy/sell (F(2,307) = 48.37, p < .001) and an interaction between buy/sell and run length (F(2,307) = 3.54, p < .05). The main effect showed that consumers preferred to buy a stock with increasing earnings (M = 6.60) and sell a stock with decreasing earnings (M = 4.82; p < .001; ambivalence between the stocks = 5.0). The asymmetric effect of buying versus selling on preferences can be seen from preferences for buying a winning stock deviating more from the point of ambivalence in the buy condition than in the sell condition (t = 45.50, p < .001; see fig. 1), supporting hypothesis 3.

Figure 1

Interaction between Buy/Sell and Trend Length

Figure 1 also depicts the interaction between buy/sell and trend length. While consumers strongly preferred to buy winners versus losers (as predicted by the hot hand fallacy and Shafir's [1993] predictions), preferences for winners did not vary by trend length as predicted by hypothesis 1 (M = 6.41, 6.63, and 6.80, respectively; t = .324, p = NS for lengths 3 vs. 7; t = .205, p = NS for 7 vs. 11; t = 1.04, p = NS for lengths 3 vs. 11). However, length did affect preferences in the sell condition. Consumers were significantly more likely to prefer to sell the losing versus the winning stock at lengths 3 (M = 4.63) and 11 (M = 4.17) compared to length 7 (M = 5.59; p < .05) (t = 3.89, p < .05 for lengths 3 vs. 7; t = 8.37, p < .005 for lengths 7 vs. 11). The difference between lengths 3 and 11 was not significant (t = .85, p = NS). These results support hypothesis 2.

Reasons for Preferences

Two chi-square analyses on participants' reasons for preference decisions were conducted, one for the buy condition and one for the sell condition. Trend length served as the independent variable and the percentage of respondents who used the various reasons as the dependent variable. Trend length had a differential impact on reasons in the sell condition (χ2(28) = 50.08, p < .01) but not the buy condition (χ2(20) = 19.67, p = NS).

Tables 1 and 2 show the distribution of reasons across conditions. In the buy condition, the data support use of the hot hand fallacy. Across the three trend lengths over 65% of participants preferred the winning stock because they believed it would continue to go up in earnings. The t-tests of proportions related to this category of response showed no differences in these proportions by trend length. Consistent with Morrin et al.'s (2002) observation of different investor segments, a modest percentage of subjects chose to buy the loser and gave reasons consistent with the gambler's fallacy (12%, 16%, and 8% for trend lengths 3, 7, and 11, respectively). Again t-tests of proportions showed no differences in use of these reasons across the three trend length conditions. Notably, while some respondents were indifferent between the stocks, indifference was typically due to lack of knowledge about the stock market or uninterpretable/blank responses as opposed to beliefs about the market's efficiency.

View this table:
Table 1

Analysis of Qualitative Data: Buy Condition

Preference and reasonFocusTrend length (%)
Buy winner:
  Winner is hotHot hand657274
  Loser is coldCold hand 2 2 4
  Winner is hot and loser is coldHot and cold hand—winner and loser 0 0 0
    Total hot/cold hand677478
Buy loser:
  Loser is going to turn aroundGambler's fallacy—loser1216 8
  Winner is going to turn aroundGambler's fallacy—winner 0 0 0
  Winner and loser are going to turn aroundGambler's fallacy—winner and loser 0 0 0
    Total gambler's fallacy:1216 8
  Don't have enough info 6 3 6
  Just guessing 2 2 0
  Uncertain about the future 1 2 4
    Total don't know 9 710
  Market is efficient (total rational)Rational 2 0 2
Can't interpret response or response blank (total)10 3 2
      Total N496050
View this table:
Table 2

Analysis of Qualitative Data: Sell Condition

Preference and reasonFocusTrend length (%)
Sell winner:
  Winner is going to turn aroundGambler's fallacy—winner294110
  Loser is going to turn aroundGambler's fallacy—loser 5 8 8
  Both winner and loser are going to turn aroundGambler's fallacy—winner and loser 0 0 8
    Total gambler's fallacy344926
Sell loser:
  Loser is going to keep losing—cut lossesCold hand312330
  Winner is going to keep getting betterHot hand11 616
  Winner is going to get better and loser is going to get worseHot and cold hand—winner and loser 0 0 0
    Total hot/cold hand fallacy422946
  Don't want to realize losses on stock 0 0 4
  Don't have enough information 810 2
  Just guessing 0 4 8
  Uncertain about the future 0 0 2
    Total don't know 81412
  Market is efficient (total rational)Rational 3 2 4
Can't interpret response or response blank (total)13 6 8
      Total N515249

In the sell condition, trend length did impact use of various heuristics. A chi-square analysis found that the proportion of responses consistent with the hot hand fallacy varied significantly by trend length (χ2(2) = 6.98, p < .03). Consumers were more likely to provide reasons consistent with the gambler's fallacy at run length 7 (49%) compared to run length 3 (34%) or 11 (26%), though t-test analyses of proportions revealed that only the comparison for trend lengths 7 versus 11 achieved statistical significance (t = 2.63, p < .05). A second chi-square analysis revealed a significant impact of trend length on the proportion of reasons consistent with the hot/cold hand fallacy (χ2(2) = 5.80, p < .05), again due to differences between run lengths 7 (29%) and 11 (46%) (t = 1.78, p < .08). The distribution of reasons shown in tables 1 and 2 reveal greater variability in the use of various heuristics in the sell versus the buy condition, providing some process support for the theoretical logic underlying hypothesis 3.


The results support hypothesis 2 and hypothesis 3 but not hypothesis 1. The qualitative data also provide evidence that preferences are predicated on use of the hot hand, cold hand, and gamblers' fallacies in ways predicted by hypothesis 2 and provide process support for the logic driving hypothesis 3. In the context of trading stocks, we find that consumers strongly prefer to buy winning stocks and sell losing stocks.

Many other studies we conducted showed that this behavior was robust to other experimental specifications. We found similar results when we used weekday air travelers instead of students; the former are more likely to have invested in and be knowledgeable about stocks. The results were also similar when we manipulated strings of returns instead of earnings per share and buying mutual funds instead of stocks; these results indicate that consumers respond to the trends and not to the context. We obtained similar results even when we manipulated the mood of subjects, by putting them in an optimistic or pessimistic mood. Finally, we also found similar results when we changed the context to the buying of mutual funds after seeing ads for them; the effect held even with the SEC-mandated warning that past returns do not guarantee future results, included in its current format at the bottom, at the top of the ad, or in a pronounced font.

Our results also suggest that use of these heuristics differs depending on whether consumers are buying (accepting) or selling (rejecting) and the length of the sequence of information. While consumers continue to use the hot hand fallacy irrespective of trend length when buying stocks, trend length does affect the use of the heuristics when selling stocks. Furthermore, the results marry research on heuristics with those on accept versus reject decisions and show that while trend length has no impact on the weight assigned to positive versus negative information in buy decisions (consumers consistently choose stocks with rising earnings regardless of trend length), it may affect the weight assigned to negative versus positive information in sell decisions. They also suggest that selling is a more complex process than buying and invokes different heuristics in different consumers in the same condition. Finally, we show that indifference between stocks is usually due to lack of knowledge or information, not use of normative economic theory. Indeed, all the behavior we observed is in direct opposition to the dictates of finance and accounting theory about what inferences consumers should rationally draw about a stock's sequence of earnings per share.

Future research should explore why consumers use different heuristics when selling, but not when buying (as predicted). Also relevant are potential boundary conditions for the observed effects, such as consumer expertise or the format of the SEC-mandated disclaimers in ads. Future research might also consider whether the pattern that emerged from the buy/sell × trend length interaction varies as a function of the time period (e.g., days or weeks) or at lengths beyond 11.

The results of the study offer several important implications. First, in contrast to the golden investing rule of buy low and sell high, we find that consumers buy winning stocks and sell losing stocks, even when they should have evidenced no preferences for such stocks.

Second, our findings may provide an individual-level explanation for why at the aggregate level stock prices show distinct overreaction based on the length of positive earnings (DeBondt and Thaler 1985) or why booms and cascades (Bikchandani, Hirshleifer, and Welch 1992) occur in hot markets. Consumers look to past positive trends in prices, make buy decisions, fuel further prices increases, which, in turn, prompts further buying. Yet our current results do not indicate that the aggregate stock market is inefficient or that any individual may profit from it. The reason is that our study does not reveal at what point in a trend aggregate prices fail to reflect intrinsic value.

Third, consumers' preference for winners may explain why brokers frequently advertise increasing trends in prices or returns, even when such trends do not predict the future and such ads must carry the SEC disclosure to such an effect. Advertisers rightly suspect that consumers use such information to make buy decisions. Thus, the SEC needs to investigate whether the current disclosure serves any useful purpose and what alternative form it should take.

Fourth, our results, if generalizable, may explain the hype that surrounds the rapid growth of new products, the escalation in real estate prices in a hot market, and the rise of fads. In each of these cases, trend projection based on the previous period's results could explain the escalating inflation in the perceived values of these assets until its ultimate implosion, sometimes triggered by some minor negative news.


View Abstract