At the Top of the Mind: Peak Prices and the Disposition Effect

I.  Introduction

In part because absolute judgments are difficult, people often evaluate outcomes relative to salient reference points. When we refer to a “small” elephant and a “large” mouse, for example, few people would mistake the ordering of their weights, because we automatically “norm” each descriptor to animals of the same species (Kahneman and Miller 1986). Applied to decision-making, we do not evaluate outcomes in terms of the final levels of wealth they confer (as the expected utility model assumes) but evaluate outcomes as gains or losses depending on whether they exceed or fall short of salient points of comparison.

The most widely documented case of reference dependence in economics and finance is the disposition effect, which refers to the reluctance of purchasers of an asset to sell it at a loss (Shefrin and Statman 1985). The disposition effect is a feature of individual financial behavior observed in the sale of both stocks (e.g., Barber and Odean 2000; Shapira and Venezia 2001; Feng and Seasholes 2005; Chang, Solomon, and Westerfield 2016) and housing (Genesove and Mayer 2001; Andersen et al. 2021; Bracke and Tenreyro 2021).¹ Most prior research on the disposition effect has assumed that the relevant price reference point for selling decisions is the purchase price of the owned asset.² Both field research focusing on housing sales and field and experimental research involving financial assets have supported the disposition effect prediction that people will be reluctant to sell assets at prices below the nominal price at which they were purchased.³

However, perhaps as a result of some kind of motivated bias (Bénabou and Tirole 2016) or simply the salience of extreme values (Gagne and Dayan 2021), asset owners often have a distorted view of true value. Casual observation of both homeowners and stockholders suggests that owners often seem to believe that the “true” value of their asset is the highest price it achieved while they have held it, the peak price. This might lead to a reluctance to sell at what is perceived to be an unfair lower price (Isoni 2011; Weaver and Frederick 2012) and can also contribute to the belief that the asset’s value is likely to regress toward the higher price, corresponding to its perceived true value.

In this paper, we examine whether consideration of the peak historic price an asset achieved while the owner has held it does, indeed, help to predict selling behavior. To do so, we estimate the disposition effect on returns since purchase and returns since peak price in both housing and stock data. Examining decisions for these two distinct asset classes held by individuals (which differ in many features, e.g., length of price cycles, liquidity, divisibility, and trading costs), we show the importance of peak prices in both markets. Our housing data provide population-level coverage of house price sales, providing a substantially larger dataset in which to examine the housing disposition effect compared with previous studies. Our large sample of stock data, in addition to records of trades, provides records of login events allowing us to restrict our sample to days on which investors pay attention to their trading accounts. We define the peak price as the highest price achieved by an asset in the individual’s period of ownership that remains the highest price for a persistent period of time (e.g., a period of weeks). We do so for both housing and stock markets, which represent the two largest, and distinct, asset markets in which individuals participate.⁴

Our empirical analysis confirms the importance of peak prices in individual trading decisions for both housing and stocks.⁵ Our estimates reveal that the selling probability for stocks and homes more than doubles when returns since a past peak price turn positive. Whereas a stock (home) in gain since purchase but in loss since a past peak price is approximately 50% (40%) more likely to be sold, a stock (home) in gain since purchase and in gain since a past peak price is 130% (96%) more likely to be sold. These discontinuities in the sale probability are found regardless of the magnitudes of gains or losses—with respect to either the peak price or the purchase price. In further tests, we show that the peak price result does not arise due to the peak price proxying for extreme observed returns since purchase. We also show that the peak price stands out from a suite of other candidate reference points (e.g., the highest price in the past month, quarter, or year) in explaining individual selling decision.

We evaluate two major psychological mechanisms that could underlie the peak price effect. One involves belief in reversion of prices to the peak. For either motivational reasons or due to the application of a heuristic, people might believe that the peak price represents an asset’s true value and hence that an asset that has fallen below its peak value is likely to eventually revert back to it. A second mechanism involves preferences rather than beliefs: Having seen the asset rise to a particular level, then fall, an investor might regret not having sold at the peak and might choose to hold on to the asset in the hope that it would rise to, or go above, that peak level. That is, people might avoid selling an asset that is below its peak to avoid the experience of regret associated with selling the asset at a price lower than they could have sold it at.

Note that these two mechanisms are not mutually exclusive and, indeed, potentially interact with one another. For example, one would likely experience greater regret in selling below the peak if one believed that the asset was more likely to revert to the peak. And, in the opposite direction, an investor would likely be more aware of peak prices occurring while an asset is owned than prior to ownership.

Although not mutually exclusive, the belief-based mechanism does have a prediction that the regret-based one does not, and vice versa. Specifically, the belief-based mechanism predicts that people will be more likely to top-up—that is, purchase more shares of a stock they already own—when the asset is below its peak price and that this effect should occur both for peaks occurring during ownership and (albeit potentially to a lesser extent) peaks occurring prior to ownership. Likewise, the preference-based mechanism predicts a reluctance to sell stocks whose peak occurred during ownership but not stocks whose peaks occurred prior to ownership.

Consistent with the first of these mechanisms, we do find that the probability of an investor topping up their position in the same stock increases as losses since the peak price increase and that this effect occurs for peaks both during ownership and prior to ownership. Likewise, consistent with the preference-based mechanism, peak price events that occurred when the investor owned the asset do have a strong impact on selling decisions, but those occurring before the investor bought the asset do not affect sales. Consistent with other research showing that events that happen to oneself are much more salient to people than those that happen to others or at other times (Kaustia and Knüpfer 2008; Simonsohn et al. 2008; Herz and Taubinsky 2018; Malmendier, Pouzo, and Vanasco 2020; Hartzmark, Hirshman, and Imas 2021), we show that peak prices that predate the individual’s purchase of the stock have no bearing on subsequent selling decisions. Hence, it is the individual’s experience of the peak price that forms the enduring reference point.

Our main estimates are robust to a variety of econometric specifications and a wide range of controls, including the duration of holding, which has been shown to be important for understanding the disposition effect for stocks (Ben-David and Hirshleifer 2012). In a series of tests, we condition the model to include property (for housing) or investor (for stocks) fixed effects, flexible controls for returns since purchase and returns since the peak price event, and a range of controls for housing and stock characteristics, plus other controls for portfolio and investor characteristics. We conduct additional checks to address particular confounds that might apply specifically to the analysis of housing (e.g., the potential for a home mortgage to be underwater) or to stocks (e.g., effects arising from portfolio rebalancing). We find that the peak price effect does not arise due to these potential confounds.⁶ Our results are also very similar when using hazard models to estimate the disposition effect, as in Seru, Shumway, and Stoffman (2010).

We interpret our results in light of a recent framework of the disposition effect in which asset prices experienced by the individual during their period of ownership can generate reference points for future decisions (Quispe-Torreblanca et al. 2024). The key assumption of the framework, motivated by diverse research in psychology, is that asset owners tend to focus on the highest of the different reference points they could pay attention to. Adopting an asset’s highest price as one’s reference point is self-serving in the sense that it assumes that the true value of an asset one purchased is the highest price that asset achieved, thus giving oneself maximum credit for making a smart purchase decision. Yet at the level of utility maximization, it may not be self-serving since it means that current market valuations will generally fall below one’s reference value.⁷

As do other theoretical perspectives that assume purchase price as the reference point, ours predicts a disposition effect with respect to gains and losses since purchase. The novel prediction introduced in our theoretical and empirical analysis is an additional disposition effect with respect to the peak price. One way to describe the joint effect of the two reference points is to say that an individual will be reluctant to sell whenever the price falls below the purchase price or the stock’s peak price—indicative of an interaction effect. However, another way to interpret the same pattern is to say that there are simply two disposition effects, one on purchase price and one on peak price. The reason these are the same is that peak price can never fall below the purchase price; if it did (e.g., if the stock lost value continuously following purchase), then the purchase price would be the peak price. Because there is no situation in which the stock price relative to the purchase price is positive but negative relative to the peak price, the main effect of peak price is identical to the interaction effect between peak price and purchase price. This equivalence is important for our empirical analysis, in which we simply compute the two main effects, with no interaction term.⁸

Our study builds upon a large experimental literature and smaller field literature on the role of peak prices in individual decisions. An experimental literature beginning with Kahneman et al. (1993) presents evidence in support of a “peak-end rule,” which states that agents judge an experience or event by how they felt at its peak and end rather than by the sum total, or the average, of every moment of the event (for a review of this literature, see Fredrickson 2000). Peaks play an important role in the subjective evaluation of magnitudes, including economic magnitudes. For example, in range-frequency theory (Parducci 1965, 1995), the range of values encountered—and the top of the range is the peak—provide the context against which magnitudes are evaluated, such that a higher peak makes any given magnitude seem smaller.⁹

Perhaps most relevant to the current research are previous studies in finance showing that peak prices serve as reference points in momentum trading strategies (George and Hwang 2004), as well as in merger and acquisition decisions (Baker, Pan, and Wurgler 2012). Furthermore, in popular news media, house price indexes are among the most commonly reported economic data, and turning points in the indexes are newsworthy events. Even just the tone of local housing news is found to be predictive of future house prices (Soo 2018; Ploessl and Just 2022). Price anchors (e.g., the 52-week high) are also commonly published alongside current prices. Hence, homeowners and investors in stocks are exposed to information on peak prices on a regular basis.

The disposition effect has been observed both in brokerage data across multiple countries and time periods (Grinblatt and Keloharju 2001; Brown et al. 2006; Barber et al. 2007; Calvet, Campbell, and Sodini 2009) and in experiments, for example, Weber and Camerer (1998). Theoretical and empirical research on the disposition effect has highlighted the importance of realization utility and loss aversion (Barberis and Xiong 2009; Frydman et al. 2014) as contributing mechanisms.¹⁰ Odean (1998) shows that the disposition effect cannot be explained by portfolio rebalancing, transaction costs, a preference for realizing gains more frequently than losses, or different beliefs about expected future returns. In a laboratory experiment, Frydman and Rangel (2014) study the role of the salience of prices in the disposition effect and find that the strength of the disposition effect depends on the salience of the purchase price.¹¹

Our findings also contribute to the growing literature studying the consequences of salience and attention for individual behaviors. Arkes et al. (2010) study the shift in each subject’s reference point following prior gains or losses. Their experimental evidence suggests that reference point adaptation is asymmetric: adaptation following a gain is greater than following a loss. Earlier work suggests that the first and last prices act as reference points. Baucells, Weber, and Welfens (2011) explore the determinants of investor reference points by exposing participants to hypothetical sequences of stock prices in a laboratory experiment. They find that a stock’s starting and ending prices are the two most important inputs into an investor’s reference point. More generally, research in the psychology literature documents that participants exposed to a series of stimuli tend to remember better the first and the most recent values (Ebbinghaus 1913; Murdock 1962; Ward 2002). We extend this line of work by empirically testing the effects of another salient reference point, the peak price.

More generally, our study expands a new line of research on how multiple reference points interact to determine individual choices. Very few empirical papers have examined the consequences for decision-making of situations in which specific outcomes are evaluated against multiple reference points, despite evidence of reference-dependent behavior in a variety of settings.¹² Studies documenting the importance of different reference points have tended to examine each in isolation.¹³ Prior research has, moreover, typically involved hypothetical choices (see, e.g., Sullivan and Kida 1995; Ordóñez, Connolly, and Coughlan 2000) or stylized laboratory experiments (Koop and Johnson 2012), although one recent empirical study found that purchase price and neighborhood price influenced the length of ownership in Singapore’s private housing market (Huang, Lien, and Zheng 2021). A limited number of studies explore how multiple reference points affect choices on separate dimensions, such as income and work hours (Crawford and Meng 2011) or goals and experience (Markle et al. 2018). Another strand of the literature seeks to understand how reference points relate to regret theory in dynamic decisions (Strack and Viefers 2021; Brettschneider, Burro, and Henderson 2025).

The remainder of the paper proceeds as follows. Section II describes the two datasets (home sales and stockbroking data), and section III explains, for each, the methodology we used to calculate returns since purchase and since peak. Section IV presents the econometric specification used in the analysis and describes the sample selection restrictions. Sections V and VI present the main results for housing and stocks. Section VII presents the analysis of ownership and mechanisms for the peak price effect. Section VIII estimates the economic cost of selling at the peak price. Section IX concludes.

II.  Data

III.  Calculation of Returns

Our main analysis uses two calculations of returns: return since purchase and return since peak price. For housing, quarterly returns are calculated via application of the house price index at the locality. Returns since purchase are the difference between the valuation of the property and purchase price. For stocks, returns are calculated using daily prices. Returns since purchase represent gross returns on the asset prices over the holding period and in cases of multiple purchases of a stock, returns are calculated as a weighted average.²⁶ For housing and stocks, in addition to returns, we also define a dummy variable indicating whether the value of the asset is in gain or loss since purchase (i.e., whether the return since purchase is positive or negative).

An innovation of our study is to introduce the concept of return since peak. Return since peak price measures the return since the asset’s most recent peak price. “Peak price” requires definition. We define the peak price as the highest price achieved by an asset in the individual’s period of ownership that remains the highest price for a persistent period of time. Central to this definition is the idea that a peak price is a price that has persisted as the highest price for some period. For example, an asset that monotonically increases in value in each period could not be said to have formed a “peak” in any period, but an asset that hits a high price from which it then falls for a number of periods could be said to have achieved a peak.²⁷ Analogous to a series of peaks in a mountain range, peaks are salient price points in the history of the individual’s ownership of the asset.²⁸ This definition captures the idea that a “peak” occurs when an asset reaches a high value that stands out in its recent history. In our main analysis, we define a peak house price as the highest price achieved by the house in the homeowners’ holding period that remains the highest price for at least three calendar quarters. We define a peak stock price as the highest price achieved by the stock in the investor’s holding period that remains the highest price for at least 1 week.²⁹

We illustrate the concept of a peak price using examples for housing and stocks. Figure 1A illustrates a quarterly price series for a single property selected from the data. The price series, for the full data period since the property was first purchased, shows the house price drifting upward (with a notable drop in 2009 during the Great Recession). A peak price is defined using the three-quarter persistence criterion. Each circle shows a price that is the highest price since the start of the period and that remains the highest price for at least three quarters.³⁰ Figure 1B illustrates a daily price series of a commonly purchased stock in the data. The price series shown runs from April 2013 to September 2014 and shows the stock price drifting upward. A peak price is defined using the 1-week time horizon. Each circle shows a price that is the highest price since the start of the period and that remains the highest price for at least 1 week.³¹ For housing and stocks, in addition to returns since peak, we also define a dummy variable by whether the value of the asset is in gain or loss since peak price.

Using this method, we calculate returns since purchase and returns since peak for observations in the baseline sample. For housing, the distributions of returns since purchase and since past peak price are both positively skewed, reflecting medium-term positive house price growth in the United Kingdom. For stocks, the distributions of returns since purchase and since past peak resemble a normal distribution, reflecting the relatively flat medium-term stock market performance in the United Kingdom.³²

IV.  Econometric Model

We estimate the disposition effect using the standard model in the empirical literature based upon Chang, Solomon, and Westerfield (2016), which we apply to housing and stocks. The standard model takes the form

$$\begin{array}{}\text{(1)}& {\mathit{Sale}}_{ijt}=b_0+b_1{\mathit{GainSincePurchase}}_{ijt}+{ϵ}_{ijt},\end{array}$$

in which the unit of observation for housing is at the property (j) and date (t) level, with quarterly measures of returns since purchase. Sale is a dummy equal to 1 if the homeowner sells their property (j) on date (t). GainSincePurchase is a dummy variable indicating whether the property (j) had made a gain on date (t) compared with the purchase price. For stocks, the unit of observation is at the account (i), stock (j), and date (t) level. Note that, given the detailed stock data, we can construct daily measures of returns since purchase. Sale is a dummy equal to 1 if the investor holding account (i) sells the stock (j) on date (t). GainSincePurchase is a dummy variable indicating whether, for the investor holding account (i), stock (j) had made a gain on date (t) compared with the purchase price.

We modify the baseline specification in equation (1) by adding a dummy variable indicating whether the asset is in gain compared with the peak price. We call this dummy variable GainSincePeak. The modified econometric specification is now

$$\begin{array}{}\text{(2)}& {\mathit{Sale}}_{ijt}=b_0+b_1{\mathit{GainSincePurchase}}_{ijt}+b_2{\mathit{GainSincePeak}}_{ijt}+{ϵ}_{ijt},\end{array}$$

in which GainSincePeak is a dummy indicating whether property (j) was in gain at date (t) compared with the peak price (housing), or for the investor (i), stock (j) was in gain at date (t) compared with the peak price (stocks).

The variable GainSincePeak therefore adds a new element to the estimation of the disposition effect. Note that in the modified econometric specification in equation (2), the dummy variable indicating whether a property × date (housing) or an account × stock × date (stocks) is in gain since peak constitutes an interaction with gain since purchase. This is because, by definition, a gain since the past peak must also be a gain since purchase, as the peak price cannot be lower than the purchase price. If the value continuously fell after purchase, the purchase price would be the peak.

We estimate both equation (1) and equation (2), allowing us first to replicate the standard estimation of the disposition effect from equation (1) before introducing results from the revised specification in equation (2). In subsequent robustness analyses, we also estimate models that add (i) property (housing) or investor (stocks) fixed effects to control for property-specific or individual-specific time-invariant heterogeneity in selling behavior, (ii) continuous measures of returns since purchase above and below the zero threshold, and (iii) a more extensive battery of control variables, including controls for property characteristics and several investment portfolio and account characteristics. In addition, we (iv) show that our results remain consistent when estimating a different econometric estimation approach, a Cox proportional hazard model with time-varying covariates, and (v) show that our results remain consistent when using different time horizons in the definition of a peak price.³³ In the stockholding analysis, we also perform a placebo test using peak prices that occurred within the past year, instead of peak prices strictly taking place during the holding period.³⁴ We show that the disposition effect around peak prices is found only when the investor has experienced the peak price (i.e., when the peak occurred during the holding period).

V.  Peak Price Effects in Home Sales

A.  Returns since Purchase Price and the Disposition Effect

The disposition effect based upon the purchase price is not as widely documented among housing sales compared with stock sales. Therefore, we first show results based on purchase price, together with a series of robustness tests, before introducing our results for the disposition effect based upon the peak price.

We draw upon our baseline sample of observations of property × quarters.³⁵ The unconditional relationship between return since purchase and probability of sale is illustrated in figure 2A. The figure is a binned scatterplot, with each point representing a 0.5% bin of observations. Figure 2A illustrates the housing disposition effect for returns since purchase; the probability of sale increases sharply when returns since purchase turn positive. In the unconditional plot, the increase in the probability of sale is approximately 0.5%, against a baseline probability of approximately 0.7%, representing a greater than 70% increase in the probability of sale.³⁶

This result is confirmed by the ordinary least squares (OLS) regression estimate for equation (1) shown in table 1. Column 1 shows the unconditional relationship between the sale of the property and the dummy for gain since purchase. The coefficient on gain since purchase of 0.0053 is positive and implies that a house that is in gain since purchase is 0.53 percentage points more likely to be sold than a house in loss. Against the base probability of selling a house from the constant in the regression of 0.7% (0.0070), this represents an increase of 75%.

	Saleit
	(1)	(2)
Gain since purchase = 1	.0053***	.0074***
	(.0005)	(.0006)
Years from purchase		.0074***
		(.0005)
Detached = 1		−.0055***
		(.0004)
Semidetached = 1		−.0040***
		(.0003)
Terraced = 1		−.0017***
		(.0002)
New-build = 1		.0016***
		(.0002)
Quality (£100,000)		−.0005***
		(.0000)
Constant	.0070***	.0030***
	(.0003)	(.0004)
Years from purchase quintics	No	Yes
Region	No	Yes
Observations	128,444,588	128,444,588
R2	.0002	.0017

Column 2 adds a set of nonlinear controls for the number of years between purchase and valuation (included as quintics) and controls for property characteristics. Property characteristics are the property’s type, whether it is new-build, its regional location, and its quality.³⁷ The coefficient of 0.74% (0.0074) in column 2 implies that a property in gain since purchase is twice as likely to be sold than a property in loss (the latter value defined by the unconditional probability of selling a house at a loss, given by the constant in col. 1, of 0.7%).

This result is robust to a range of additional tests to account for market conditions, including liquidity, market size, and leverage, as well as time since purchase; see table 2. These tests are important as, for example, highly levered mortgagors (at the limit, 100% levered) may be unable to sell at any price below the purchase price. We measure housing market liquidity using the number of housing transactions per quarter; housing market size, using housing stock volumes per quarter; and homeowner’s leverage, using quarterly LTV percentages. Then, for each case, we split the dataset at the median value of the variable of interest (housing transactions, housing stock, LTV percentage, or time since purchase) and estimate the OLS regression models specified by equation (1) for the “high” (above-median) and “low” (below-median) subsets. The pattern of positive coefficients demonstrates that the disposition effect for gain since purchase exists in above- and below-median samples in each case.³⁸

	Gain Since Purchase		Constant
Liquidity (sales):
Above median	.0051***	(.0005)	.0063***	(.0004)
Below median	.0043***	(.0004)	.0047***	(.0003)
Size (stock):
Above median	.0064***	(.0007)	.0043***	(.0004)
Below median	.0059***	(.0006)	.0039***	(.0004)
Leverage (LTV%):
Above median	.0024***	(.0005)	.0043***	(.0004)
Below median	.0052***	(.0008)	.0024***	(.0007)
Time since purchase:
Above median	.0028***	(.0005)	.0246***	(.0077)
Below median	.0066***	(.0005)	.0062***	(.0005)

VI.  Peak Price Effects in Stock Sales

The existence of a disposition effect among stock trading based upon the purchase price is widely documented and replicated in our data. We therefore focus on our main result for the existence of a disposition effect around the peak price. This is illustrated in figure 3. The figure is a binned scatterplot, illustrating the relationship between the probability of sale and returns since purchase price (fig. 3A), returns since peak price (fig. 3B), and the interaction between returns since purchase price and peak price (fig. 3C). Following the previous literature on stockholding and the disposition effect, we present main estimates from models using the subsample of observations of account × stock × days on which the investor made a sale of at least one stock.⁴²

Figure 3A illustrates the standard disposition effect results for returns since purchase. The probability of sale increases sharply when returns since purchase turn positive; in the unconditional plot, the increase in the probability of sale is approximately 8 percentage points against a baseline probability of 12% (an increase in the probability of sale of two-thirds).

Figure 3B reveals the existence of a disposition effect for returns since the peak price. Again, here we observe a sharp increase in the probability of sale when returns since peak price turn positive (i.e., when the stock price crosses its previous peak). In the plot, the increase in the probability of sale is large, with the probability more than doubling when returns since peak turn positive.⁴³

These patterns are confirmed by OLS regression estimates of equations (1) and (2), which are shown in table 5. Column 1 shows the estimate of equation (1). The coefficient of the gain since purchase dummy is positive and implies that a stock that is in gain since purchase is approximately 8.3 percentage points more likely to be sold compared with a stock in loss. Against the base probability of selling a stock from the constant in the regression of 11.4%, this represents an increase of 73%.

The model in column 2 replaces the gain since purchase dummy from equation (1) with the gain since peak price. The coefficient of this dummy variable is again positive and precisely defined. The coefficient of the gain since peak price dummy in column 2 implies that a stock that is in gain since peak price is approximately 12.8 percentage points more likely to be sold compared with a stock in loss. Against the base probability of selling a stock of 13.3%, this represents a 96% increase in the likelihood of a sale.

Figure 3C shows the interaction of returns since purchase and returns since peak. Similar to houses, the disposition effect since purchase is stronger when the price is above a past peak. Estimates of equation (2), the corresponding regression, are shown in column 3. Results show a positive coefficient on both the gain since purchase and the gain since peak price dummies, which are both precisely estimated. A stock in gain since purchase but in loss since peak price is 5.7 percentage points more likely to be sold. When the stock is also in gain since the peak price, this probability increases further by 9.1 percentage points. Hence, evaluated against a baseline probability of 11.4%, a stock in gain since purchase but in loss since peak price is approximately 50% more likely to be sold, whereas a stock in gain since purchase and gain since peak price is 130% more likely to be sold.

We examine the robustness of this main result with a series of additional tests. Previous studies have shown that particular covariates may be important for explaining the disposition effect, including the length of the holding period and the magnitude of returns. We therefore include these controls in additional robustness tests, together with other control variables, such as the time elapsed since a past peak, investor demographics, and several portfolio characteristics. We also estimate fixed effects regressions and a modified econometric specification that employs a Cox proportional hazard model.

VII.  Explanations for the Peak Price Effect

In the remainder of the paper, we evaluate explanations for the existence of a peak price effect. We consider two main explanations. First, a belief in “peak reversion”: Owners of assets might optimistically infer from the fact that an asset at one point achieved a particular price that its true value corresponds to that peak price and, hence, expect the market price to evolve toward the peak price. Anticipation of such an increase in price would lead to a reluctance to sell the asset at its current, lower price. This explanation is related to an explanation for the disposition effect proposed by Shefrin and Statman (1985) and Odean (1998)—that investors hold on to losing stocks because they expect higher future returns from losing stocks compared with winning stocks; that is, they expect losing stocks to outperform in the future as they rebound in price.

Second, regret avoidance: The peak price can be a source of regret for owners who wish they had sold at the peak. Asset owners may resist selling, therefore, to avoid converting the paper loss of not having sold at the peak to the real loss of actually selling “too late.” Such an account would be consistent with research on “inaction inertia,” which finds that “when an attractive action opportunity has been forgone, individuals tend to decline a substantially less attractive current opportunity in the same action domain, even though, in an absolute sense, it still has positive value” (Tykocinski and Pittman 1998, 206).

Note that these explanations are not mutually exclusive—for example, an individual could regret not having sold at the peak, could want to wait for the asset price to rise back to the peak before selling, and could also be overoptimistic that this opportunity will arise, as a result of belief in peak reversion.

We evaluate these explanations via additional analysis of the role of ownership in generating the peak price effect and the role of top-up purchases. If the peak price effect arises due to regret, we would expect to see a stronger effect among those who held the asset at the point in time of reaching the peak price (as only this group would experience greater regret). If the peak price effect arises due to belief in reversion to the peak, we would expect to observe individuals making top-up purchases when the price falls below the peak price (in expectation of future gains as the price reverts to the peak). We examine both issues using the stock trading data.⁴⁶

A.  Ownership and the Peak Price Effect

We first examine the relationship between ownership and the peak price effect. Recent studies suggest that ownership is itself important for attention and belief formation (Hartzmark, Hirshman, and Imas 2021; Kindermann et al. 2021; Medina, Mittal, and Pagel 2021; Carney et al. 2022). For each investor × stock × day observation, we first identify the peak price for that stock within the past year (defined as the highest price achieved over the past year). We then divide the sample into observations for which the investor held the stock on the peak price day and observations for which the investor did not hold the stock on the peak price day.

Figure 4 provides examples of scenarios in which investors held (fig. 4A) and did not hold (fig. 4B) a stock on the peak price day. In figure 4A, the investor purchased the stock prior to the peak day event and, in this case, has experienced a gain since purchase and a gain since the peak price day. In figure 4B, the investor purchased the stock after the peak price day event (which occurred approximately 6 months prior to the purchase day) and has also experienced a gain since purchase and a gain since peak price day.

Using this approach, we find much weaker evidence for a disposition effect around peak price events that predate the investor’s holding of the stock. We estimate equation (2) on both samples, with results shown in table 10. Panel A reports results for the sample of observations in which the investor did not hold the stock on the peak price day. Estimates show a positive and precisely defined coefficient of the gain since purchase dummy. The coefficient of the gain since peak dummy is imprecisely defined in column 1 and negative in column 2 (statistically significant at the 5% level). The coefficient remains negative in column 3 after adding account and stock fixed effects to account for heterogeneity across investors and assets.

	Saleijt
	(1)	(2)	(3)
	A. No Holding Stock
Gain since purchase = 1	.1582***	.1431***	.1502***
	(.0073)	(.0071)	(.0064)
Gain since peak = 1	−.0161	−.0249**	−.0015
	(.0102)	(.0102)	(.0081)
Days from purchase day (100 days)		−.0450***	−.0053***
		(.0025)	(.0017)
Constant	.1268***	.1723***
	(.0052)	(.0066)
Account fixed effects	No	No	Yes
Stock fixed effects	No	No	Yes
Observations	242,536	242,536	242,536
R2	.0383	.0480	.2397
	B. Holding Stock
Gain since purchase = 1	.0361***	.0350***	.0322***
	(.0043)	(.0043)	(.0040)
Gain since peak = 1	.0320***	.0290***	.0298***
	(.0054)	(.0055)	(.0050)
Days from purchase day (100 days)		−.0033***	.0054***
		(.0009)	(.0007)
Constant	.1172***	.1289***
	(.0047)	(.0062)
Account fixed effects	No	No	Yes
Stock fixed effects	No	No	Yes
Observations	169,800	169,800	169,800
R2	.0045	.0050	.1998

Panel B reports results for the sample of observations in which the investor did hold the stock on the peak price day. In this sample, in contrast with panel A, estimates show positive and precisely defined coefficients for both the gain since purchase and gain since peak price dummies.⁴⁷ The coefficient values are smaller than in the main analysis, as expected—given that a peak price in the past year is likely less salient than a peak price since purchase—but show the existence of both forms of the disposition effect as expected. Hence, this test adds evidence that peak prices affect trading behaviors through returns experienced by the investor.

B.  Top-Up Purchases and the Peak Price Effect

If investors believe that stocks that have performed poorly since their peak price will eventually return to that level, they may be more likely to make additional purchases of those stocks the larger the loss since the past peak event. To explore this, figure 5 shows binned scatterplots of the relationships between losses since purchase, losses since peak, and the propensity of investors to top-up current positions with new purchases of the same stock. The proportion of observations with top-ups is shown on the y-axis, with loss since purchase (fig. 5A) and since peak (fig. 5B) shown on the x-axis. Results in figure 5A show that the probability of top-up decreases as losses since purchase increase. However, in figure 5B, we observe the opposite relationship for losses since peak price: the probability of top-up increases as losses since peak price increase. These relationships are confirmed by regression analysis, with the former showing a positive coefficient of the loss since purchase variable and the latter showing a negative coefficient of the loss since peak variable.⁴⁸

While our analysis has shown that observations with higher losses since peak price display a higher probability of top-up purchases, which is consistent with the idea of belief in reversion to the past peak price, one might expect that a similar relationship would hold when the reference point is the purchase price. That is, one would expect that higher losses since purchase would be related to a higher probability of a top-up purchase. However, our analysis did not find this relationship in the data. This suggests that investors may not necessarily be chasing losses in general but rather only chasing losses relative to certain reference points that are independent of the time they purchase the stock. These higher reference points, such as the peak price, may be more salient for investors and may influence their decision-making more significantly. This indicates that the peak price may serve as a particularly influential reference point for investors and may shape their perceptions of the value of an asset in a way that the purchase price does not.

In combination, these results provide strong support for a peak-reversion account—we should not expect to observe elevated top-ups when prices are below peak unless investors anticipate that prices will rise—and also positive but less strong support for regret avoidance. Support for regret avoidance comes from the observation that peak prices occurring when the asset is owned have a much greater impact than peak prices before the asset was owned, consistent with an account in which people, naturally, only regret not selling assets they actually owned. However, the latter finding could also be consistent with a peak-reversion account if people pay more attention to—and hence have a stronger belief in peak reversion to—peaks occurring when they hold an asset.

VIII.  Future Returns and Economic Costs of Selling

In this section, we estimate the economic costs to individual stock and housing investors resulting from their decisions to sell at the purchase price and at the peak price. Following Odean (1998), we use data on the future returns to calculate the paper returns investors would have achieved had they held on to the stocks and housing.

First, we present calculations for stocks, which are shown in table 11. Panel A shows calculations of the excess (of the FTSE100) postsale returns for periods subsequent to the sale of a winning stock or the observation of an unrealized paper loss. These returns are evaluated over three periods: the month following the sale, 166 days (which corresponds to the average holding period), and 1 year after the sale for realized winnings and similarly after occasions when stocks with paper losses were retained. Each instance of a stock sold at a profit is treated as a separate observation in our analysis, whether it involves sales on different dates or by different investors on the same date. This approach is also applied to paper losses.

To provide an approximate estimate of the economic costs associated with the disposition effect and the peak price effect, we can use the analogy proposed by Odean (1998). Let us consider a hypothetical investor who is deciding whether to sell one of two stocks, each worth £1,000 after commissions. Initially, imagine that this investor has only one reference point, which is the purchase price of the stock. The first stock has gained since purchase, while the second stock has lost since purchase. Suppose this investor is averse to realizing losses and decides to sell the winning stock. Results in table 11 show that, on average, winning stocks sold achieve higher subsequent returns compared with losing stocks not sold. Over the next year, the difference is 3.62 percentage points. Hence, if the investor had retained the winning stock sold and instead sold the stock showing paper losses, they would have achieved an on-paper gain over the next year of £36.22. In addition, selling the winning stock incurs a capital gain, while selling the losing stock would incur a capital loss that could be offset against future capital gains for tax purposes.⁴⁹

An analysis of future returns for stocks sold at the peak shows a similar pattern. The excess return on winning stocks sold at the peak is higher than that for paper losses not sold (which, as with paper losses for returns since purchase, show negative future returns over the 1-year horizon). Moreover, by choosing to retain the winning stock, the investor would have benefited from an additional £32.65 over the following year, in comparison to the losing stock. An additional consideration, however, is that an investor who makes selling decisions based upon peak prices is less likely to sell compared with an investor who makes selling decisions based on purchase prices (because the peak price is by definition higher than the purchase price).⁵⁰ In additional analysis, we show that when investors use peak price as a reference point, holding highly volatile assets exhibiting an above-average number of peaks leads to more significant losses. This occurs because of momentum in volatile stocks that surpass previous peaks and then tend to perform well in the following year and because volatility leads to more occasions in which the stock is above its past peak price, increasing the investor’s trading frequency.⁵¹

An analysis of future returns for homes sold at the purchase price and the peak price again shows a very similar pattern. Panel B of table 11 shows calculations of the postsale returns for periods subsequent to the sale of a winning home or observation of an unrealized paper loss over three periods: the next 6 months, next year, and next 2 years (these time periods are longer than those for stocks, due to the much lower frequency of trades of housing compared with stocks). The difference in returns for winning homes sold versus paper losses over the 2 years following the sale of a home at the purchase price is 8.5 percentage points. In the case of homes sold at peak price, the difference in returns is 5.97 percentage points. These results indicate that vendors would have achieved higher future returns had they not sold at the point where the price rose through the purchase price or peak price. Hence the same pattern of economic costs from selling at the purchase price and the peak prices exists for housing as for stocks, though the tax implications differ between the two.⁵² For housing, unlike stocks, there are of course reasons for buying and selling unrelated to the financial return, however, it is striking that the same pattern of foregone future returns through selling is seen in both asset types.

IX.  Conclusion

Using data on the buying and selling behavior of individual asset-owners—both for housing and stocks—we show a new disposition effect for returns based upon the peak price reached by the asset during the individual’s period of ownership. This disposition effect arising from the peak price experienced by the asset owner exists over and above the disposition effect arising from the purchase price. We show that this effect is robust to a range of econometric tests and also sensitivity analyses. For the stock data, whose prices are more volatile and therefore contain a larger frequency of peak prices, we were also able to perform a placebo test that exploits peak price events that occur before investors purchase stocks.

Our study contributes to the expanding literature on how multiple reference points affect individual decisions. Research has documented the operation and consequences for economic behavior of diverse reference points.⁵³ In the setting we analyze, evidence from testing the peak price versus other reference points indicates that, alongside the purchase price, the peak price is most important for individual choices. However, reference prices may be asset- or context-dependent, and relevant reference points may differ for institutional versus individual investors. For example, financial news outlets and information services commonly refer to a wide range of reference prices. Li and Yu (2012) find that nearness to the 52-week high positively predicts future aggregate market returns, while nearness to the historical high negatively predicts future market returns.

The current research also contributes to the growing literature showing that an individual’s personal history can affect their economic behavior (see, e.g., Malmendier and Nagel 2011; Malmendier, Tate, and Yan 2011; Malmendier and Nagel 2016; Andersen, Hanspal, and Nielsen 2019) and that, more specifically, the history of an individual’s ownership of an object—for example, how they came to acquire the object—can affect valuations (Loewenstein and Issacharoff 1994; Strahilevitz and Loewenstein 1998). One dimension of understanding the consequences of individual experience for future behavior, the current research suggests, is to understand the reference points that experience makes salient.

Data Availability

The code required to replicate the tables and figures in this paper can be found in Quispe-Torreblanca et al. (2024) in the Harvard Dataverse, https://doi.org/10.7910/DVN/UNKWEP. The investor data from Barclays Stockbroking is proprietary and accessible only with prior approval; details on the process are provided in the replication package. The housing data are publicly available from multiple sources: the Price Paid Dataset (His Majesty’s Land Registry), the UK House Price Index (Office for National Statistics), housing stock estimates (Ministry of Housing, Communities, and Local Government), and Loan-to-Value ratios (Financial Conduct Authority). Instructions on accessing and preparing these datasets, as well as constructing the variables used in the analysis, are included in the replication package.