Behavioral Biases in Panic Selling: Exploring the Role of Framing during the COVID-19 Market Crisis

Abstract

Panic selling causes long-term losses and hinders investors’ return to the market. It has been explained using prospect theory aspects such as loss and regret aversion. Additionally, overconfidence and overreaction contribute to the disposition effect, leading investors to sell stocks prematurely. However, the framing effect, another disposition effect attribute, has been underexplored in the context of panic selling. This study investigates how the framing effect influences panic selling, particularly during market crises, when investors perceive information differently, depending on its positive or negative framing. Utilizing data from a collaborative survey, we examine Japanese investors’ behavior during the COVID-19 market crisis. Negative framing is negatively associated with complete or partial sale of securities, whereas positive framing has the opposite effect. During market crises, investors presented with negative framing are less likely to panic sell, whereas those presented with positive framing are more prone to it. Other significant factors include gender; men tend to engage more in panic selling. Conversely, higher education, financial literacy, and greater household income and assets are associated with a reduced likelihood of panic selling. These findings underscore the critical role of framing in investor behavior during market crises, providing new insights into the mechanisms underlying panic selling.

1. Introduction

Panic selling is widely regarded as irrational behavior because it involves selling all or a major portion of stocks out of fear during a market downturn, without considering for long-term market trends (Baker and Ricciardi 2017; Barber and Odean 2013; Shiller 2015). Historical data consistently show that markets tend to rebound after crises, meaning that short-term losses can often be recovered if investors hold their positions. For example, studies have documented that even after major financial crises, stock markets typically experience a recovery, allowing investors to regain lost value and continue to benefit from the long-term higher returns associated with equities (Acharya et al. 2021; Cortes et al. 2022). Furthermore, interventions like monetary and fiscal policies, such as the quantitative easing measures implemented by central banks, are designed to stabilize markets and encourage a rebound during periods of economic stress (Cortes et al. 2022). Selling assets at a market low prevents investors from participating in these rebounds, locking in losses that might have been temporary (Baker and Ricciardi 2017; Sauer and Kramer 2022; Siegel 2014). Therefore, behavioral explanations are needed to understand why investors act against their long-term financial interests in these scenarios.

In our study, we define panic selling as an emotionally driven, irrational behavior in which investors sell a significant portion or all of their stock holdings in response to market downturns, primarily due to fear of further losses. This is distinct from general stock selling or liquidating assets for the purpose of portfolio rebalancing. Speculative traders or those who adjust portfolios for strategic reasons are not considered panic sellers in our context. Previous studies providing behavioral explanations for panic selling have argued that investors engage in this behavior to avoid loss and regret, a phenomenon that falls within the scope of prospect theory (Kahneman and Tversky 1979). This theory states that the pain of loss is greater than the pleasure of gaining (Kahneman and Tversky 1979), leading investors to avoid losses as much as possible. Furthermore, overconfidence and overreaction can exacerbate the disposition effect, causing investors to sell their stocks (Shi et al. 2011). Understanding these behavioral biases is essential to explain why panic selling occurs despite the evidence that markets typically recover and that long-term investment strategies tend to yield better returns.

Prospect theory, first introduced by Kahneman and Tversky (1979), explains how individuals assess potential gains and losses under uncertainty. Central to this theory is the concept of loss aversion, where the emotional impact of losses is greater than that of equivalent gains. Recent studies (Barberis 2022; De Bondt 2021) have expanded on this by demonstrating how loss aversion intensifies during periods of market volatility, driving investors to liquidate assets irrationally during downturns. Regret aversion, another aspect of prospect theory, adds to this, where investors sell assets prematurely to avoid potential regret from further losses (Sainsbury 2023). Historical market crises, such as the 2008 Global Financial Crisis and the COVID-19 pandemic, vividly illustrate these behaviors. For instance, during the 2008 crisis, widespread fear and media coverage exacerbated panic selling (Sauer and Kramer 2022). Similarly, at the onset of COVID-19, many investors, driven by fear of impending losses, sold their positions rapidly, often at significant personal cost (Ji et al. 2024; Hodgson and Drummond 2009). The pandemic’s economic disruptions led to widespread anxiety, prompting many investors to act impulsively, much like the hoarding and opportunistic behaviors observed in other sectors during the crisis (Sobirova 2020; Baker et al. 2020). Such reactions demonstrate how external shocks, such as pandemics, can intensify existing behavioral biases, leading to suboptimal financial decisions.

Although several aspects of prospect theory, such as loss aversion, regret aversion, and overconfidence, have been used to explain panic selling, our focus lies on the framing effect and its role in exacerbating irrational selling behavior during crises. Examining the framing effect in the context of panic selling is important because investors may perceive information differently during market crises based on the positive or negative framing. During such periods, the flow of negative information—often fueled by rumors—is expected. At the early stages of the pandemic, the influx of negative information prompted several investors to engage in panic selling (Sauer and Kramer 2022; Ji et al. 2024).

The framing effect, first formalized by Tversky and Kahneman (1981), refers to situations in which individuals make inconsistent decisions based on how choices are presented. When information is framed negatively, individuals are more likely to perceive greater risk and respond by taking drastic actions, such as selling assets to avoid further loss. This phenomenon has been supported by recent studies (Lehenkari and Perttunen 2023; Bicchieri 2021), which demonstrate how media framing and public discourse during crises influence investor sentiment and decisions. For example, during the 2020 market crash, the continuous flow of pessimistic news reports about the economic impacts of COVID-19 heightened negative framing, prompting widespread sell-offs among retail investors (Sauer and Kramer 2022; Tabesh et al. 2019). Negative framing can distort rational decision-making and cause investors to react emotionally rather than strategically, which aligns closely with panic selling behaviors observed during recent crises.

Specifically, our study adds to the literature by examining the specific behaviors and emotional triggers—such as fear and negative framing—that lead to panic selling. Our definition of panic selling accounts for emotional reactions to market downturns, distinguishing this behavior from other, more rational asset liquidation strategies. A recent collaborative survey by Rakuten Securities Inc. and Hiroshima University generated an extensive dataset on the behavior of Japanese investors during the COVID-19 crisis, facilitating the investigation of whether the framing effect is associated with panic selling.

The theoretical background of panic selling is grounded in the study of investor behavior. Panic sellers’ actions significantly deviate from those of rational investors who adhere to long-term investment principles (Siegel 2014; Malkiel 2020). Previous studies have explored several behavioral phenomena that explain panic selling. The fundamental theory underlying panic selling is the prospect theory (Kahneman and Tversky 1979), in which loss aversion is particularly relevant. Prospect theory posits that investors are highly sensitive to potential losses than equivalent gains, a concept known as loss aversion. In the context of panic selling, when markets decline, the fear of further losses can drive investors to sell assets quickly, and often irrationally, to avoid perceived future losses. This behavior, driven by the psychological impact of losses, can exacerbate market downturns and lead to widespread panic selling (Sainsbury 2023). Similarly, regret aversion leads investors to sell stocks in anticipation of further price declines (Sainsbury 2023). Overconfidence is another factor associated with the theoretical background of panic selling. Overconfident investors often make irrational decisions, such as overtrading and inappropriate investment choices (Glaser and Weber 2007; Odean 1999; Barber et al. 2020). During market crises, overconfident investors may unduly emphasize negative information and hastily sell their positions. Shi et al. (2011) developed a theoretical model, suggesting that investors become overconfident during periods of price momentum but panic when they witness significant price reversals. Herd behavior, in which investors follow others’ actions rather than relying on their own analysis, also plays a role in panic selling. This phenomenon, driven by emotions like fear and the desire to conform, can lead to irrational market decisions. When investors observe others selling assets quickly, they may do the same to avoid potential losses, even if their personal analyses do not justify such actions. This collective behavior can exacerbate market downturns, leading to significant volatility and further panic selling (Markowski Investment 2024). Understanding herd behavior helps recognize the emotional and psychological factors driving such market movements. Moreover, Bucher-Koenen and Ziegelmeyer (Bucher-Koenen and Ziegelmeyer 2014) explained panic selling from a financial illiteracy perspective, in which less financially literate investors tend to engage in panic selling.

The framing effect refers to situations in which investors make inconsistent decisions regarding identical problems based on how they are presented (that is, positively or negatively) (Tversky and Kahneman 1981; Frisch 1993). It can cause investors to deviate from rational decision making, leading to inconsistent choices and substantial losses (Hodgson and Drummond 2009; Kühberger 1998). The impact of the framing effect on decision-making has been observed across various fields, including marketing, management, psychology, public health, and medicine (Kühberger 1998). In their study of the “Asian disease problem”, Tversky and Kahneman (1981) demonstrated that individuals tend to take more risks when questions related to an outbreak are framed negatively. They identified two potential outcomes of the framing effect related to panic selling. First, based on the concept of loss aversion, investors with negative framing may focus more on negative information and sell their positions to avoid further losses. Second, based on the risk-taking phenomenon, these investors may accept more risk and refrain from panic selling (Tversky and Kahneman 1981; De Martino et al. 2006). Investors with negative framing may believe that they have little to lose by accepting higher risk during a market crisis. Furthermore, from an uncertainty avoidance perspective, individuals presented with negative framing, when faced with risky situations, tend to embrace uncertainty more than those presented with positive framing. Tabesh et al. (2019) argued that decision-makers tend to choose risky options when presented with negatively framed scenarios. Despite the importance of framing in decision-making, studies in this area, particularly with respect to the influence of framing on panic selling decisions, are lacking (Simon et al. 2004; Gong et al. 2013).

Therefore, drawing on prospect theory and the framing effect, we examine how positive and negative framing relate to panic selling behavior. We hypothesize that investors presented with negative framing tend to make risky decisions and embrace uncertainty. Rather than engaging in panic selling, they may choose to retain their positions during uncertainty. Conversely, investors presented with positive framing tend to be more risk-averse, leading them to sell their positions during a market crisis. This study makes three key contributions. First, it is the first to clarify the relationship between investors’ framing effects and panic selling. Second, it proposes effective nudges for individual investors who panic sell by considering the framing effect. Third, it contributes to policymaking and corporate investment strategies.

2. Literature Review

Investor behavior during financial crises has been extensively studied, with numerous behavioral theories attempting to explain why individuals often make irrational decisions under market stress. Cognitive biases, such as overconfidence, herd behavior, and loss aversion, have been identified as significant drivers of panic selling, where investors liquidate their assets prematurely in response to market downturns (Baker and Ricciardi 2017; Shiller 2015). However, one area that remains underexplored in the context of panic selling is the framing effect, despite its recognized influence on decision-making in other domains. This gap in the literature presents an opportunity to further examine how the framing of information during crises, such as the COVID-19 pandemic, may exacerbate or mitigate panic selling behavior.

Previous research on investor behavior during crises has focused primarily on biases like overconfidence, where investors overestimate their ability to predict market trends, leading to excessive trading and irrational decisions (Glaser and Weber 2007; Odean 1999). Overconfident investors are prone to taking on excessive risk during market booms and panic selling during downturns, contributing to market volatility (Shiller 2015; Siegel 2014). This behavior was evident during the 2008 financial crisis and again during the COVID-19 pandemic, as fear and uncertainty gripped the markets (Ji et al. 2024).

Herd behavior, another well-documented bias, exacerbates this irrational decision-making. Investors often follow the actions of others, especially during crises, leading to collective panic selling (Bucher-Koenen and Ziegelmeyer 2014). Wang et al. (2013) show that during the 2008 stock market crash, herd behavior intensified as investors reacted to perceived insolvency risks, worsening market declines. Herd behavior originates from the fear of missing out or being left behind, and during crises, the pressure to conform often outweighs rational analysis (Kahneman and Tversky 1979).

While overconfidence and herd behavior provide strong explanations for panic selling, these perspectives are focused on the individual’s internal biases or reactions to the actions of others. They do not fully account for the role of external stimuli, particularly how information is presented or framed during periods of market turmoil.

Prospect theory, introduced by Kahneman and Tversky (1979), has been instrumental in explaining why investors tend to panic sell during market downturns. The theory posits that individuals are more sensitive to losses than to gains, a phenomenon known as loss aversion. This bias causes investors to prioritize the avoidance of losses, even if it means sacrificing potential long-term gains (Glaser and Weber 2007; Odean 1999). Loss aversion becomes particularly acute during crises, as investors react emotionally to falling markets by liquidating their assets in an attempt to avoid further losses (Shi et al. 2011; Sainsbury 2023).

However, while prospect theory offers a compelling framework for understanding investor behavior, it primarily addresses emotional responses to perceived losses. It does not sufficiently explain how the framing of market information—whether as losses or gains—affects investor decisions during crises. This is a significant gap, as framing has been shown to influence risk-taking behavior in other decision-making contexts (Tversky and Kahneman 1981; Frisch 1993).

The framing effect, where the presentation of information influences decision-making, has been extensively studied in psychology but remains underexplored in the context of financial markets. Tversky and Kahneman (1981) first demonstrated that people are more likely to take risks when situations are framed negatively. In financial terms, investors may be more prone to panic selling if market downturns are framed as catastrophic losses rather than temporary declines. This is consistent with the findings of Kühberger (1998) and Hodgson and Drummond (2009), who argue that negative framing exacerbates emotional reactions and undermines rational decision-making.

During the COVID-19 pandemic, for instance, negative media coverage of the economic fallout likely heightened fear among investors, contributing to widespread panic selling (Sauer and Kramer 2022). Ji et al. (2024) found that the market’s reaction to COVID-19 was largely driven by the uncertainty and fear stoked by such negative framing, as investors sought to minimize losses amid the global health crisis. However, while there is substantial evidence that framing influences decisions in other areas, its role in triggering panic selling during financial crises has not been systematically studied.

Despite the wealth of literature on investor behavior and panic selling, the framing effect’s influence remains underexplored. Most research on panic selling has focused on biases like overconfidence, herd behavior, and loss aversion (Barber and Odean 2013; Kahneman and Tversky 1979; Glaser and Weber 2007). While these explanations are valuable, they do not address how the framing of market information—whether in media reports, government announcements, or corporate disclosures—affects investors’ propensity to panic sell.

This gap is particularly evident in the context of the COVID-19 pandemic, which triggered unprecedented behavioral responses in financial markets. Research by Sobirova (2020) suggests that the pandemic led to various opportunistic and unethical behaviors, including hoarding and irrational financial decisions. However, the role of framing in shaping these behaviors, particularly in the context of panic selling, remains insufficiently studied. Sobirova’s (2020) work on the conceptual model of non-ethical behavior during the pandemic highlights the influence of crisis framing on individual actions, but does not directly address its impact on financial decision-making.

Understanding how the framing effect contributes to panic selling could provide new insights into mitigating irrational investment behaviors during crises. By focusing on this overlooked aspect of investor behavior, future research could offer more targeted strategies to improve market stability and investor resilience.

3. Data and Methods

3.1. Data

We used data from the “survey on life and money” conducted by Rakuten Securities and Hiroshima University. Specifically, we used data from the 2023 wave, which were collected in November and December 2023. Participants were aged 18 years and older and had an active account with Rakuten Securities. The survey collected detailed information on Japanese adults’ demographic, socioeconomic, and psychological preferences, focusing on investors’ framing bias and panic selling behavior. After eliminating missing variables, the final dataset comprised 191,005 observations.

3.2. Variables

We created two binary dependent variables to capture the selling behavior of investors during the COVID-19 pandemic. We asked the respondents, “How did you manage your stocks and mutual funds in March 2020?” We defined those who answered “I sold all my stocks/funds” as Sell_all, and those who answered “I sold all my stocks/funds” or “I sold some of my stocks/funds” as Sell_part. The decision to define Sell_all and Sell_part as binary variables is consistent with the literature that examines investor behavior during periods of financial uncertainty, such as Barber and Odean (2013) and Wang et al. (2013). These studies often employ similar binary outcomes to capture whether an investor made drastic decisions to sell or partially reduce their investments during market downturns or crises. This also aligns with the panic selling literature, including Sauer and Kramer (2022), which emphasizes the distinction between full and partial liquidation of assets during a crisis.

Our main independent variables, Framing_G and Framing_L, were created using coin flipping questions (see Appendix A) based on Hayden and Platt (2009) and Blavatskyy (2021). These questions measured the choice of respondents between certain and uncertain outcomes based on the prospect theory developed by Kahneman and Tversky (1979). Those who answered A to Q25 and B to Q26 were coded as Framing_G because they gambled only in the gain phase. Those who answered B in Q25 and A in Q26 were coded as Framing_L because they gambled only in the loss phase. The inclusion of framing variables is supported by Kahneman and Tversky (1979), which suggests that individuals’ choices between risky and certain outcomes vary depending on whether the scenario is framed as a loss or gain. Studies such as Tversky and Kahneman (1981), Kühberger (1998), and Hodgson and Drummond (2009) have demonstrated the significant effect that gain and loss framing can have on decision-making under risk. Additionally, De Martino et al. (2006) and Blavatskyy (2021) provide evidence on how framing influences decisions in financial contexts.

Furthermore, we included age, gender, employment status, number of children, marital status, household financial status, educational background, financial literacy, risk aversion, and a myopic view of the future as control variables. Research by Barber and Odean (2013), Bucher-Koenen and Ziegelmeyer (2014), and Ji et al. (2024) shows that demographic and socioeconomic characteristics have significant influences on investment decisions. Numerous studies have highlighted the importance of financial literacy in shaping investor behavior, particularly in crises (Siegel 2014; Bucher-Koenen and Ziegelmeyer 2014). Investors with higher financial literacy tend to make more informed decisions and are less prone to panic selling. This variable is essential for understanding individual responses to uncertainty. Risk aversion is widely used in behavioral finance models, supported by Kahneman and Tversky (1979) and studies such as Frisch (1993) and Glaser and Weber (2007), which highlight its relevance to investment behavior. The inclusion of a myopic perspective is supported by the literature on short-term thinking in financial decision-making (Odean 1999). Investors who exhibit a short-term myopic outlook are more likely to react to immediate losses and market volatility, which is consistent with panic selling behavior, as discussed in Barberis (2022) and Sainsbury (2023). Table 1 presents the definitions and measurement of all variables.

Table 1. Variable definitions.

Variable	Definition
Dependent variables
Sell_all	Binary variable: 1 = sold all stocks and mutual funds, 0 = otherwise
Sell_part	Binary variable: 1 = sold partial or whole stocks and mutual funds, 0 = otherwise
Independent variables
Framing_G (positive framing)	Binary variable: 1 = gambling action in the gain phase, 0 = otherwise
Framing_L (negative framing)	Binary variable: 1 = gambling action in the loss phase, 0 = otherwise
University degree	Binary variable: 1 = hold at least a university degree, 0 = otherwise
Financial literacy	Discrete variable: mean scores for responses to three financial literacy questions
Male	Binary variable: 1 = male, 0 = female
Age	Continuous variable: respondents’ age
Age squared	Continuous variable: age squared
Married	Binary variable: 1 = married, 0 = otherwise
Have children	Continuous variable: number of children
Unemployment	Binary variable: 1 = unemployed, 0 = otherwise
Household income	Continuous variable: annual earned income before taxes, including bonuses of the entire household (unit: JPY)
Log of household income	Log (Household income)
Household assets	Continuous variable: balance of financial assets (savings, stocks, bonds, insurance, and so on) of the entire household (unit: JPY)
Log of household assets	Log (Household assets)
Risk aversion	Continuous variable: risk of rain preference (percentage score from the question, “Usually, when you go out, how high must the probability of rainfall be before you take an umbrella?”)
Myopic view	Binary variable: 1 = if a respondent agrees or completely agrees with, “As the future is uncertain, it is a waste to think about it.”

3.3. Descriptive Statistics

Descriptive statistics are presented in Table 2. Approximately 1.2% of respondents sold all their stocks and investment trusts; 5.7% sold all or partial stocks. Regarding the main independent variables, 6.8% of respondents exhibited positive framing bias (Framing_G); 33.2% demonstrated negative framing bias (Framing_L). The average score of respondents’ financial literacy was 0.78. Of the respondents, 64.3% were male and their average age was 45.2 years. Furthermore, 63.9% had a university degree, 67.0% were married, and 59.0% had at least one child. Only 1.5% of participants were unemployed. Annual household income was 7.45 million yen, and household assets were 19.2 million yen. Finally, the level of risk aversion and myopic view of the future was 0.5 and 0.15, respectively.

Table 2. Descriptive statistics.

Variable	Mean	Std. Dev.	Min	Max
Dependent Variables
Sell_all	0.0123452	0.1104214	0	1
Sell_part	0.056564	0.2310081	0	1
Independent Variables
Framing_G	0.0682548	0.2521832	0	1
Framing_L	0.3315515	0.470772	0	1
University degree	0.6389362	0.4803103	0	1
Financial literacy	0.7826619	0.3046471	0	1
Male	0.6430355	0.4791055	0	1
Age	45.25036	12.24403	18	94
Age squared	2197.51	1164.096	324	8836
Married	0.6700505	0.4701957	0	1
Have children	0.5899531	0.4918432	0	1
Unemployment	0.0149473	0.1213421	0	1
Household income	7,449,389	4,158,354	1,000,000	2.00 × 107
Log of household income	15.65721	0.6112347	13.81551	16.81124
Household assets	1.92 × 107	2.37 × 107	2,500,000	1.00 × 108
Log of household assets	16.15842	1.086626	14.7318	18.42068
Risk aversion	0.5351781	0.231175	0	1
Myopic view	0.1525719	0.3595753	0	1
Observation	191,005

3.4. Methods

This study investigated the association between panic selling behavior and positive (Framing_G) and negative framing (Framing_L) using the following Equations.

$${Sell\_all}_i=f({Framing\_G}_i,X_i,{\epsilon }_i)$$

(1)

$${Sell\_part}_i=f({Framing\_G}_i,X_i,{\epsilon }_i)$$

(2)

$${Sell\_all}_i=f({Framing\_L}_i,X_i,{\epsilon }_i)$$

(3)

$${Sell\_part}_i=f({Framing\_L}_i,X_i,{\epsilon }_i)$$

(4)

where $Sell\_{all}_i$ represents whether the ith respondent sold all stocks/investment trusts. Sell_part indicates whether the respondent sold all or some of their stocks/investment trusts. Framing_G and Framing_L indicate whether respondents are in the positive (gain phase) or negative framing (loss phase), respectively. X is a vector of individual demographic, socioeconomic, and psychological characteristics. Ε is the error term. The use of a probit model is justified due to the binary nature of the dependent variables (Sell_all and Sell_part). Binary choice models, such as probit or logit, are commonly used in studies examining investment behavior under uncertainty (Ji et al. 2024; Wang et al. 2013). The probit model is particularly suited to estimating the likelihood of discrete outcomes, such as whether an investor decides to sell all or part of their portfolio in response to market conditions.

We also tested for correlation and multicollinearity to measure intercorrelations (results available upon request). The correlation matrix revealed a weak relationship between the variables (significantly less than 0.70. Moreover, the variance inflation factor tests did not show multicollinearity in any model.

The full specifications for the equations are as follows.

$$\begin{array}{l}Sell\_{all}_i={\beta }_0+{\beta }_{1}Framing\_G_i+{\beta }_2{Universit ydegree}_i+{\beta }_3{Financial literacy}_i+{\beta }_4{Age}_i\\ +{\beta }_5{Age squared}_i+{\beta }_6{Male}_i+{\beta }_7{Unemployment}_i+{\beta }_8{Married}_i+{\beta }_9{Have childeren}_i\\ +{\beta }_{10}{Log of household income}_i+{\beta }_{11}{Log of household assets}_i+{\beta }_{12}{Risk aversion}_i\\ +{\beta }_{13}{Myopics}_i+{\epsilon }_i\end{array}$$

$$\begin{array}{l}Sell\_{part}_i={\beta }_0+{\beta }_{1}Framing\_G_i+{\beta }_2{Universit ydegree}_i+{\beta }_3{Financial literacy}_i+{\beta }_4{Age}_i\\ +{\beta }_5{Age squared}_i+{\beta }_6{Male}_i+{\beta }_7{Unemployment}_i+{\beta }_8{Married}_i+{\beta }_9{Have childeren}_i\\ +{\beta }_{10}{Log of household income}_i+{\beta }_{11}{Log of household assets}_i+{\beta }_{12}{Risk aversion}_i\\ +{\beta }_{13}{Myopics}_i+{\epsilon }_i\end{array}$$

$$\begin{array}{l}Sell\_{all}_i={\beta }_0+{\beta }_{1}Framing\_L_i+{\beta }_2{Universit ydegree}_i+{\beta }_3{Financial literacy}_i+{\beta }_4{Age}_i\\ +{\beta }_5{Age squared}_i+{\beta }_6{Male}_i+{\beta }_7{Unemployment}_i+{\beta }_8{Married}_i+{\beta }_9{Have childeren}_i\\ +{\beta }_{10}{Log of household income}_i+{\beta }_{11}{Log of household assets}_i+{\beta }_{12}{Risk aversion}_i\\ +{\beta }_{13}{Myopics}_i+{\epsilon }_i\end{array}$$

$$\begin{array}{l}Sell\_{part}_i={\beta }_0+{\beta }_{1}Framing\_L_i+{\beta }_2{Universit ydegree}_i+{\beta }_3{Financial literacy}_i+{\beta }_4{Age}_i\\ +{\beta }_5{Age squared}_i+{\beta }_6{Male}_i+{\beta }_7{Unemployment}_i+{\beta }_8{Married}_i+{\beta }_9{Have childeren}_i\\ +{\beta }_{10}{Log of household income}_i+{\beta }_{11}{Log of household assets}_i+{\beta }_{12}{Risk aversion}_i\\ +{\beta }_{13}{Myopics}_i+{\epsilon }_i\end{array}$$

4. Results

Table 3 presents the results of the probit regressions analyzing the factors associated with the two dependent variables representing panic selling: Sell_all and Sell_part. Each variable was analyzed across four models (Models 1 to 4), using the main independent variable, Framing_G, and different sets of control variables in each model. Model 1 is the baseline model, with only Framing_G included. Model 2 includes demographic and socioeconomic variables, improving model fit. Model 3 introduces household income and assets, further refining the analysis. Model 4 incorporates risk and time preferences, offering the most comprehensive view.

Table 3. Probit regression results of Framing_G.

	Sell_all				Sell_part
Variables	Model 1	Model 2	Model 3	Model 4	Model 1	Model 2	Model 3	Model 4
Framing_G	0.1090 ***	0.1128 ***	0.1145 ***	0.1166 ***	0.0076	0.0231	0.0256	0.0279
	(0.0287)	(0.0293)	(0.0293)	(0.0293)	(0.0184)	(0.0187)	(0.0187)	(0.0187)
University degree		−0.1168 ***	−0.0945 ***	−0.0998 ***		−0.0223 **	−0.0164	−0.0210 **
		(0.0172)	(0.0178)	(0.0179)		(0.0103)	(0.0106)	(0.0106)
Financial literacy		−0.4812 ***	−0.4563 ***	−0.4504 ***		−0.3670 ***	−0.3643 ***	−0.3613 ***
		(0.0248)	(0.0257)	(0.0256)		(0.0155)	(0.0159)	(0.0159)
Age		−0.0121 ***	−0.0080 *	−0.0073 *		−0.0045 *	−0.0019	−0.0013
		(0.0042)	(0.0044)	(0.0044)		(0.0025)	(0.0026)	(0.0026)
Age squared		0.0001 ***	0.0001 **	0.0001 **		0.0001 ***	0.0001 ***	0.0001 ***
		(0.0000)	(0.0000)	(0.0000)		(0.0000)	(0.0000)	(0.0000)
Male		0.4558 ***	0.4571 ***	0.4534 ***		0.3990 ***	0.4021 ***	0.3999 ***
		(0.0212)	(0.0212)	(0.0212)		(0.0116)	(0.0116)	(0.0116)
Unemployment		0.0566	0.0334	0.0351		0.1635 ***	0.1360 ***	0.1382 ***
		(0.0595)	(0.0598)	(0.0598)		(0.0334)	(0.0336)	(0.0336)
Married		−0.0547 **	−0.0260	−0.0248		−0.0043	0.0162	0.0168
		(0.0224)	(0.0233)	(0.0233)		(0.0129)	(0.0134)	(0.0134)
Have children		0.0090	0.0132	0.0154		−0.0260 **	−0.0191	−0.0166
		(0.0224)	(0.0225)	(0.0226)		(0.0127)	(0.0127)	(0.0127)
Log of household income			−0.0589 ***	−0.0570 ***			−0.0561 ***	−0.0551 ***
			(0.0167)	(0.0167)			(0.0095)	(0.0095)
Log of household assets			−0.0242 ***	−0.0270 ***			0.0144 ***	0.0122 **
			(0.0092)	(0.0093)			(0.0052)	(0.0052)
Risk aversion				0.1970 ***				0.1649 ***
				(0.0363)				(0.0210)
Myopic view				0.0381 *				0.0284 **
				(0.0221)				(0.0132)
Constant	−2.2545 ***	−1.9118 ***	−0.7591 ***	−0.8656 ***	−1.5848 ***	−1.6626 ***	−1.0826 ***	−1.1626 ***
	(0.0082)	(0.0989)	(0.2433)	(0.2430)	(0.0048)	(0.0601)	(0.1400)	(0.1402)
Observations	191,005	191,005	191,005	191,005	191,005	191,005	191,005	191,005
Log likelihood	−12,699	−12,250	−12,234	−12,217	−41,526	−40,300	−40,281	−40,248
Chi2 statistics	14.43	827.8	864.2	874.7	0.172	2359	2405	2459
p-value	0.000145	0	0	0	0.678	0	0	0

Notes: Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

The results show that Framing_G is positively and significantly associated with Sell_all in all models, indicating that a higher Framing_G value increases the likelihood of selling all securities. However, it is not significantly associated with Sell_part, indicating that Framing_G does not significantly influence the decision to partially sell securities. Among the control variables, having a university degree is negatively associated with both Sell_all and Sell_part across relevant models, suggesting that higher education decreases the likelihood of panic selling, whether partial or complete. Financial literacy shows a strong negative association with both Sell_all and Sell_part, indicating that financially literate individuals are less likely to engage in panic selling. Age has a non-linear effect on both Sell_all and Sell_part, with a negative coefficient for age and a positive coefficient for age squared. This finding suggests that younger investors are more likely to engage in panic selling, however, this likelihood decreases with age before increasing again at a more advanced age. Being male is positively associated with both Sell_all and Sell_part across all models, indicating that male investors are more likely to engage in panic selling than female investors. Unemployment is positively associated with Sell_part but not significantly related to Sell_all, indicating that unemployed individuals are more likely to partially sell their securities rather than all of them. Being married is negatively associated with Sell_all in some models but not consistently related to Sell_part, suggesting that married individuals may be slightly less likely to sell all their securities. Having children is not significantly associated with Sell_all. However, it has a negative association with Sell_part in some models, suggesting that having children may slightly reduce the likelihood of partially selling one’s securities. Higher household income and assets generally reduce the likelihood of Sell_all but have mixed effects on Sell_part, with higher household assets sometimes associated with a higher likelihood of partial selling. Risk aversion and a myopic view of the future show positive associations with both Sell_all and Sell_part, indicating that more risk-averse and short-sighted (myopic) investors are more likely to engage in panic selling.

Table 4 presents the regression results for Framing_L, which measures the impact of positive framing on the decision to sell securities. The negative and highly significant coefficients across all models for both Sell_all and Sell_part suggest that negative framing (Framing_L) significantly reduces the likelihood of investors selling all or part of their securities. This effect is stronger in Sell_all than in Sell_part, indicating that negative framing has a more substantial impact on preventing complete liquidation. Among the control variables, having a university degree is negatively associated with both Sell_all and Sell_part across the relevant models. This finding suggests that possessing higher education decreases the likelihood of panic selling, which aligns with the proposition that educated investors may be better informed or make rational decisions. Financial literacy is strongly and negatively associated with both Sell_all and Sell_part, indicating that financially literate individuals are less likely to engage in panic selling, whether partial or complete. This finding suggests that financial literacy mitigates irrational selling behavior. The coefficients for age and age squared indicate a non-linear relationship with panic selling. The negative coefficient for age suggests that younger investors are more likely to sell; however, the positive coefficient for age squared implies that this trend reverses with age. This non-linear effect may reflect varying risk tolerances or financial responsibilities at different life stages. Being male is positively and significantly associated with both Sell_all and Sell_part across all models, suggesting that male investors are more prone to panic selling than female investors. Being unemployed is positively associated with Sell_part but not significantly related with Sell_all, indicating that unemployed individuals are more likely to partially sell their securities rather than their entire portfolio. The relationship between being married and panic selling is mixed. Marriage is negatively associated with Sell_all in some models but is not consistently significant with Sell_part, suggesting that married individuals may be slightly less likely to sell all their securities. However, this effect is not robust. Having children has a slight negative association with Sell_part in some models, suggesting that individuals with children may be less likely to engage in partial panic selling; this effect is not strong or consistent across models. Higher household income and assets generally reduce the likelihood of Sell_all, indicating that wealthier households are less likely to liquidate their entire portfolio during market crises. However, the relationship with Sell_part is mixed, with higher household assets sometimes being associated with a higher likelihood of partial selling. Risk aversion is positively associated with both Sell_all and Sell_part, suggesting that more risk-averse investors are likely to panic sell. Myopic investors (those focused on short-term gains/losses) are more likely to engage in both Sell_all and Sell_part, although the effect is weaker than that of risk aversion.

Table 4. Probit regression results of Framing_L.

	Sell_all				Sell_part
Variables	Model 1	Model 2	Model 3	Model 4	Model 1	Model 2	Model 3	Model 4
Framing_L	−0.1572 ***	−0.0994 ***	−0.0989 ***	−0.0966 ***	−0.1304 ***	−0.0813 ***	−0.0808 ***	−0.0794 ***
	(0.0179)	(0.0183)	(0.0184)	(0.0184)	(0.0102)	(0.0104)	(0.0104)	(0.0104)
University degree		−0.1157 ***	−0.0936 ***	−0.0987 ***		−0.0208 **	−0.0150	−0.0196 *
		(0.0173)	(0.0179)	(0.0179)		(0.0103)	(0.0106)	(0.0106)
Financial_literacy		−0.4651 ***	−0.4402 ***	−0.4348 ***		−0.3536 ***	−0.3511 ***	−0.3483 ***
		(0.0248)	(0.0257)	(0.0256)		(0.0155)	(0.0159)	(0.0159)
Age		−0.0123 ***	−0.0082 *	−0.0076 *		−0.0044 *	−0.0019	−0.0013
		(0.0042)	(0.0044)	(0.0044)		(0.0025)	(0.0026)	(0.0026)
Age squared		0.0001 ***	0.0001 **	0.0001 **		0.0001 ***	0.0001 ***	0.0001 ***
		(0.0000)	(0.0000)	(0.0000)		(0.0000)	(0.0000)	(0.0000)
Male		0.4494 ***	0.4507 ***	0.4473 ***		0.3937 ***	0.3968 ***	0.3948 ***
		(0.0212)	(0.0212)	(0.0212)		(0.0116)	(0.0116)	(0.0116)
Unemployment		0.0532	0.0310	0.0327		0.1624 ***	0.1353 ***	0.1374 ***
		(0.0595)	(0.0597)	(0.0597)		(0.0334)	(0.0336)	(0.0336)
Married		−0.0550 **	−0.0271	−0.0259		−0.0043	0.0159	0.0165
		(0.0224)	(0.0233)	(0.0233)		(0.0129)	(0.0134)	(0.0135)
Have children		0.0105	0.0144	0.0165		−0.0252 **	−0.0184	−0.0160
		(0.0224)	(0.0226)	(0.0226)		(0.0127)	(0.0127)	(0.0127)
Log of household income			−0.0569 ***	−0.0550 ***			−0.0552 ***	−0.0542 ***
			(0.0167)	(0.0166)			(0.0095)	(0.0095)
Log of household assets			−0.0247 ***	−0.0273 ***			0.0143 ***	0.0122 **
			(0.0092)	(0.0093)			(0.0052)	(0.0052)
Risk_aversion				0.1904 ***				0.1614 ***
				(0.0362)				(0.0210)
Myopic view				0.0395 *				0.0287 **
				(0.0221)				(0.0132)
Constant	−2.2000 ***	−1.8744 ***	−0.7417 ***	−0.8471 ***	−1.5440 ***	−1.6430 ***	−1.0733 ***	−1.1523 ***
	(0.0092)	(0.0992)	(0.2431)	(0.2428)	(0.0055)	(0.0602)	(0.1400)	(0.1402)
Observations	191,005	191,005	191,005	191,005	191,005	191,005	191,005	191,005
Log likelihood	−12,665	−12,242	−12,227	−12,211	−41,443	−40,270	−40,252	−40,220
Chi2 statistics	77.30	840.9	875.3	886	163.4	2412	2455	2508
p-value	0	0	0	0	0	0	0	0

Notes: robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

5. Discussion

Our findings provide novel insights into the behavioral phenomenon of panic selling, particularly through the lens of framing effects. These results align with, and extend, existing research on investor behavior during market crises by drawing on concepts such as prospect theory, loss aversion, and overconfidence.

Our results strongly support the hypothesis that negative framing reduces the likelihood of panic selling. The negative and significant relationship between negative framing and panic selling decisions across all models indicates that investors exposed to negative framing are less likely to engage in panic selling. This finding aligns with the theoretical proposition that negative framing leads investors to perceive a higher level of risk associated with selling, prompting them to retain their investments instead of liquidating them. This result corroborates the argument of Tversky and Kahneman (1981) that negative framing may lead individuals to take more risks to avoid perceived losses. Conversely, positive framing is positively and significantly associated with panic selling, indicating that investors presented with a positive frame are more likely to panic sell. This outcome supports the hypothesis that positive framing increases risk aversion, leading investors to sell their positions to avoid further losses during market crises. The relationship between positive framing and panic selling supports the broader literature on loss aversion and behavioral biases, in which individuals often act irrationally in response to perceived threats to their financial security (Baker and Ricciardi 2017; Siegel 2014). These findings also resonate with the influence of the framing effect on decision-making under uncertainty, as discussed by Frisch (1993) and Tversky and Kahneman (1981). The empirical evidence provided here demonstrates that investors’ decisions are influenced by the framing of information, with negative framing encouraging risk-taking behavior that counteracts panic selling, while positive framing promotes risk aversion, leading to higher rates of panic selling.

Our results offer a unique contribution to the behavioral finance literature by emphasizing the importance of framing effects, an aspect that has been underexplored in relation to panic selling. Although prior studies, such as the work of Tversky and Kahneman (1981) on prospect theory, extensively discuss loss aversion and its influence on investor behavior, the role of framing has not been fully integrated into the explanations of why investors panic sell during crises. Our study challenges this gap by providing empirical evidence that how information is framed significantly impacts whether an investor will hold or sell assets during a market downturn. This builds on existing research that primarily focuses on behavioral biases such as overconfidence (Barberis 2022) and regret aversion (Bicchieri 2021), without factoring in the decision-altering power of framing.

Moreover, our findings align with and extend the literature that emphasizes how external factors, like monetary and fiscal policies, influence market behavior during crises. For example, Cortes et al. (2022) highlight how quantitative easing measures by central banks, such as those implemented by the Bank of Japan, helped stabilize markets during the COVID-19 pandemic, independent of public sentiment regarding health concerns. While the literature on market interventions during the pandemic demonstrates the efficacy of these policies in restoring market stability, our study adds to this by showing that despite these stabilizing measures, behavioral biases like framing still led investors to engage in irrational panic selling. This suggests that even in contexts where policy interventions create favorable market conditions, investor psychology remains a key driver of selling behavior, underscoring the necessity of addressing behavioral factors alongside macroeconomic interventions.

Our findings on the control variables further illuminate the dynamics of panic selling. Higher education and financial literacy consistently show a negative association with panic selling, indicating that educated and financially literate individuals are less likely to engage in such behavior. This result aligns with the existing literature, suggesting that more educated and financially knowledgeable individuals are better equipped to make informed and rational decisions during market crises, reducing their susceptibility to panic selling (Sauer and Kramer 2022; Bucher-Koenen and Ziegelmeyer 2014; Lusardi and Mitchell 2014; van Rooij et al. 2011). However, this effect is multifaceted. On the one hand, educated individuals are more likely to understand the positive impacts of economic policies, such as monetary interventions and fiscal stimulus, which stabilize markets during crises, encouraging them to hold their stocks for long-term gains (Klapper et al. 2013). On the other hand, these same individuals may also be more aware of the risks associated with crises like the COVID-19 pandemic, which could lead to negative sentiment about the real economy (Goolsbee and Syverson 2021). The fact that education still mitigates panic selling despite increasing awareness of pandemic risks suggests that knowledge of long-term market dynamics and economic policies outweighs immediate concerns about the crisis. Furthermore, the cultural context may play a role. In some developed economies, such as the US, more educated individuals were more likely to take COVID-19 risks seriously, as evidenced by higher compliance with non-pharmaceutical interventions like mask-wearing and social distancing. However, in Japan, where societal norms are more oriented toward collective responsibility and the “common good”, the effect of education on the perception of pandemic risk perception may be more uniform across different levels of education. Japan’s collectivist culture, which emphasizes social harmony and shared responsibility, likely contributes to more homogeneous behavior during crises, even between different education groups (Hofstede 2001; Markus and Kitayama 1991). This cultural homogeneity may help explain why higher education in Japan is strongly associated with a lower likelihood of panic selling, as even well-informed individuals may prioritize long-term financial stability over short-term economic fears.

The non-linear relationship between age and panic selling, with younger investors being more prone to panic selling and this tendency decreasing with age before rising again at older ages, suggests that life stage plays a crucial role in investment behavior. This pattern may reflect varying risk tolerances, financial responsibilities, and experience levels, with younger investors possibly lacking the skill and experience to remain calm during a crisis and older investors taking a more conservative approach as they reach retirement (Shi et al. 2011).

Gender differences also emerge as significant, with male investors more likely to panic sell than female investors. This result aligns with the literature on overconfidence, in which male investors are often found to be more overconfident and prone to making hasty decisions based on incomplete or negative information (Glaser and Weber 2007; Barber et al. 2020). The finding that unemployed individuals are more likely to engage in partial selling but not full liquidation suggests that financial pressure influences selling decisions, particularly for those facing immediate economic hardships. Finally, being married and having children show mixed effects on panic selling behavior. Married individuals and those with children appear to be slightly less likely to engage in panic selling. This could be attributed to a greater focus on long-term financial stability and the need to provide for dependents, thereby reducing the impulse to sell in response to short-term market fluctuations.

These findings have significant implications for understanding panic selling in the behavioral finance context. They extend the application of prospect theory, particularly loss aversion, by integrating the framing effect as a key factor influencing investor decisions during crises. Although previous research has focused on loss aversion, regret aversion, and overconfidence as drivers of panic selling (Shi et al. 2011; Tversky and Kahneman 1981; Markowski Investment 2024), this study uniquely highlights the framing effect as a critical determinant. By comparing our findings with these established theories, we demonstrate that the framing effect can either exacerbate or mitigate irrational selling behavior, depending on how risks are perceived. This insight enhances our understanding of panic selling, highlighting that it is not solely driven by emotions like fear of loss but also by the cognitive shortcut means investors take when processing information during a crisis. Moreover, the results provide empirical support for the proposition that how information is framed can either exacerbate or mitigate irrational selling behavior. This insight is crucial for formulating interventions, such as investor education programs or policy measures, that can help reduce the occurrence of panic selling by presenting market information in a manner that discourages hasty, emotionally driven decisions. Overall, our study suggests that investors influenced by the framing effect, and prone to panic selling, may contribute to lower overall market participation. Additionally, these investors may not return to the market even after it rebounds, further indicating a long-term reduction in market participation. This issue is significant and deserves the attention of policymakers, who should consider interventions to mitigate panic selling and promote sustained market engagement.

Finally, the findings regarding framing effects on panic selling behavior have important implications for sustainable investment practices. By demonstrating that negative framing reduces the likelihood of panic selling, this study suggests that how information is presented to investors can significantly influence their decision-making, particularly in times of market crises. This insight is crucial for promoting sustainable investment, as it highlights the need for strategies that encourage long-term thinking and stability rather than short-term, emotion-driven reactions. Sustainable investment is based on the ability of investors to remain committed to their investments despite market volatility, aligning with the broader goals of financial stability and responsible investing. By incorporating framing effects into investor education and communication strategies, policymakers and financial institutions can foster a more resilient investment environment that supports sustainable practices.

Our study has several limitations that should be considered when interpreting the results. First, although we included a large number of active investors, they all come from a single company, which may limit the generalizability of our findings. Second, while we controlled for important demographic, socioeconomic, and behavioral factors, other potential influences on panic selling—such as personal financial stress, investment experience, and access to real-time financial advice—were not included. Future research incorporating these variables could offer a more nuanced understanding of the factors influencing investor decisions during market crises. Third, the sample may suffer from selection bias, as investors who chose to participate in the study might differ from those who did not, potentially skewing the results. For instance, more experienced or financially literate investors may be overrepresented. Future studies using random or stratified sampling methods would help address this issue. Lastly, while this study focuses on the framing effect, further research should investigate the interaction of various behavioral biases to provide a more comprehensive understanding of the psychological drivers behind panic selling.

6. Conclusions

This study advances the understanding of panic selling by demonstrating the significant role of framing effects in influencing investor behavior during market crises. We provide robust evidence that framing effects, particularly negative framing, serve as a strong deterrent to panic selling, suggesting that how information is presented to investors can significantly impact their decision-making processes. These findings not only enhance the theoretical framework around panic selling but also offer practical strategies for mitigating this behavior through targeted financial education and advice.

Our findings have several important implications. First, they highlight the crucial role that framing effects play in shaping investor behavior, underscoring the potential to design effective interventions, such as nudges that use negative framing, to discourage panic selling and promote stable investment behavior. Second, by integrating the framing effect into the analysis of panic selling, we contribute to a more comprehensive understanding of the behavioral factors influencing this phenomenon, which had previously been dominated by discussions of loss aversion, regret aversion, and overconfidence.

In terms of practical applications, these insights are particularly valuable to policymakers, financial advisors, and institutional investors. Enhancing financial literacy and education could help reduce panic selling by improving investors’ understanding of market dynamics and how short-term crises are often followed by recovery. Moreover, tailored advice that accounts for the framing effect can enable better investment outcomes, especially during periods of market volatility. Policymakers can develop communication strategies that present market information in ways that reduce emotional, short-term decision-making among investors, thereby fostering more stable financial markets.

Looking ahead, future research could explore the role of other psychological biases, such as the anchoring effect or confirmation bias, in panic selling behavior. Investigating the interaction between multiple behavioral biases and framing effects could also deepen our understanding of investor behavior in complex market environments. Additionally, it would be valuable to examine how the framing effect varies under different crises periods, to further refine strategies that mitigate panic-driven market behavior.