Hyperbolic Discounting and Its Influence on Loss Tolerance: Evidence from Japanese Investors

Abstract

Hyperbolic discounting, a key determinant of intertemporal behavior, captures individuals’ preferences for smaller immediate rewards over larger delayed ones. This study examined how hyperbolic discounting influences investment loss tolerance using a large-scale dataset of Japanese investors. Loss tolerance is defined as the extent of financial loss that an investor is willing to endure before changing their investment strategy. Although hyperbolic discounting shapes intertemporal investment decisions, its role in explaining loss tolerance remains largely unknown. Using a large dataset from the “Survey on Life and Money” comprising 107,294 observations and employing ordered probit regression, we found a significant negative relationship between hyperbolic discounting and investment loss tolerance: investors exhibiting stronger hyperbolic discounting are more likely to exit positions prematurely during market downturns, despite potential long-term recovery. The estimated marginal effect (−0.070 ***) underscores the economic significance of the association between hyperbolic discounting and loss tolerance. These results provide evidence that time-inconsistent preferences not only shape intertemporal choices but also reduce resilience to financial losses. The findings carry important implications for investors, highlighting the value of commitment mechanisms and education programs to counteract short-termism, and for policymakers seeking to design behavioral interventions that promote stable, long-term participation in financial markets.

1. Introduction

Investment loss tolerance refers to the extent to which investors are willing and able to accept declines in the value of their investments without liquidating their holdings prematurely (Braga and Fávero 2017). Generally, investors with higher loss tolerance are more inclined to pursue higher long-term returns, whereas those with lower tolerance are more likely to react adversely to losses, often selling prematurely (Odean 1998). Understanding the nature of loss tolerance and the influence of psychological biases can assist both financial advisors and individual investors in formulating strategies to mitigate the adverse effects of such biases. In turn, this facilitates more rational financial decision-making and supports long-term financial goals.

Although prior research has related liquidation behavior to self-control, discipline, and crisis-period panic selling (e.g., Shefrin and Statman 1985; Ormos and Joó 2014; Dayani and Jannati 2022), the direct empirical connection between hyperbolic discounting and loss-threshold decision in ordinary (non-crisis) conditions remains limited. We address this gap by testing whether present-biased time preference predict investors’ stated maximum loss before liquidation, using a large-scale sample of Japanese retail investors and ordered probit estimation. Since hyperbolic discounting devalues future recovery relative to current setbacks, it provides a theoretically grounded mechanism for lower willingness to tolerate short-term losses even when long-run gains are possible.

The theoretical framework of loss tolerance assumes that investors have a predefined threshold for acceptable losses. This study adopted a rational perspective, suggesting that loss tolerance reflects an informed understanding of market behavior, particularly the idea that asset prices are largely random and independent in the short term (Chitenderu et al. 2014). Within this framework, short-term volatility is temporary and correctable over time, thereby supporting long-term investment strategies. This view aligns with the findings of Hellmann et al. (2024), who reported that U.S. companies engage in more long-term investment activities because of their investors’ comparatively higher loss tolerance. However, behavioral research shows that many investors act irrationally when faced with short-term losses, often selling investments too early and locking in avoidable losses (Odean 1998; Benartzi and Thaler 1995). This contradiction between rational expectations and actual behavior highlights the need to examine cognitive biases such as hyperbolic discounting, which may impair investors’ ability to stay the course.

Hyperbolic discounting is a key factor in explaining intertemporal irrational behavior (Martin and Weber 1995; Cavatorta and Groom 2019; Rachlin and Raineri 1992; Takahashi 2007; Takemura 2020). This reflects individuals’ tendencies to prefer immediate gratification over delayed and potentially greater rewards (Takahashi 2007; Takemura 2020; Harris and Laibson 2001; Ikeda and Kang 2015). Empirical studies have shown that individuals exhibiting hyperbolic discounting are more likely to make suboptimal financial choices (Ikeda and Kang 2015; Angeletos et al. 2000; Kang and Ikeda 2014; Ring et al. 2022; Zhang 2013). This behavioral pattern is widely interpreted as an indicator of diminished self-control, since stronger hyperbolic discounting reflects a reduced ability to delay gratification and a greater tendency toward impulsivity and inconsistent choices over time (Takemura 2020). It also contributes to overindebtedness, greater debt accumulation (Ikeda and Kang 2015), increased credit card usage (Meier and Sprenger 2010), reduced household financial assets, and elevated consumption levels (Zhang 2013). These behavioral patterns suggest a lower capacity to withstand short-term negative experiences, even when doing so is necessary to achieve future benefits, which is especially relevant in the context of long-term financial decision-making.

Several pathways exist through which hyperbolic discounting undermines investor tolerance of financial losses, either directly or by magnifying the effects of other cognitive and emotional biases. First, hyperbolic discounting may exacerbate the combined influence of loss aversion and an overreaction to information, thereby intensifying their detrimental impact on loss tolerance. Prospect theory (Kahneman and Tversky 1979) establishes that individuals are typically loss averse, perceiving losses more acutely than equivalent gains. In parallel, the overreaction hypothesis (Barberis et al. 1998) demonstrates that investors often respond excessively to recent or salient information, particularly under uncertainty. Building on these foundations, we propose that when hyperbolic discounting interacts with these tendencies, its psychological effects compound. Investors then experience stronger aversion to losses and react more sensitively and impulsively to short-term market fluctuations. Importantly, this liquidation threshold can be understood as a behavioral manifestation of intertemporal trade-offs. Hyperbolic discounting amplifies impulsivity and weakens self-control, which, in combination with loss aversion and myopic loss aversion (Benartzi and Thaler 1995), makes temporary volatility appear disproportionately threatening. In this sense, our operationalization of loss tolerance captures the precise point at which time-inconsistent preferences translate into premature liquidation, thereby grounding the theoretical link between hyperbolic discounting and observed sell-threshold behavior. This increased emotional reactivity makes immediate setbacks appear disproportionately threatening, increasing the likelihood of premature asset liquidation and the abandonment of long-term strategies. Together, these reactions undermine overall tolerance for financial losses. Second, hyperbolic discounting embodies a time-inconsistent preference structure in which individuals irrationally prioritize immediate rewards over delayed benefits (Ikeda and Kang 2015). Such shortsightedness discourages the development of long-term market understanding and preparation, leaving investors susceptible to making poorly informed decisions during periods of market stress. Third, individuals with a stronger tendency toward hyperbolic discounting often display greater impulsivity and reduced self-control (O’Donoghue and Rabin 1999; Gathergood et al. 2019), which further undermines their ability to adhere to disciplined long-term investment strategies.

Although loss tolerance and hyperbolic discounting are recognized as psychological concepts influencing financial decision-making, empirical studies exploring their relationship remain limited. To address this gap, the present study aims to investigate whether, and in what ways, hyperbolic discounting reduces investors’ tolerance for financial losses. Specifically, we ask: (1) Does hyperbolic discounting negatively affect investors’ willingness to tolerate financial losses? and (2) To what extent does hyperbolic discounting increase the likelihood of premature liquidation of assets during short-term market volatility? From these questions, we derive two hypotheses: H1: Investors with stronger hyperbolic discounting tendencies exhibit lower tolerance for financial losses; and H2: Hyperbolic discounting amplifies the likelihood of premature liquidation during market downturns. Our focus on Japanese investors is motivated by cultural orientations toward intertemporal choice, demographic factors such as an aging population with distinct financial planning horizons, and institutional characteristics, including high levels of retail investor participation (Takahashi et al. 2009; Bawalle et al. 2024; Mason et al. 2010). By situating the study in this context, this study clarifies the psychological mechanisms underlying loss tolerance and provide insights for both investors and policymakers seeking to mitigate panic-driven, short-sighted financial behavior.

This study makes three contributions to the literature. First, it is the first empirical study to examine the impact of hyperbolic discounting on loss tolerance under generalized contexts, not under notable crises such as the COVID-19 market downturn. These insights could help create tools or programs to help investors assess their loss tolerance concerning time-inconsistent preferences. Second, this study offers practical policy suggestions to reduce the negative impact of hyperbolic discounting on investment behavior, thereby promoting more rational decision-making aligned with long-term financial goals. Third, it highlights the importance of further research on the role of psychological biases in shaping loss tolerance, encouraging the exploration of its relationship with other cognitive biases. The remainder of this paper is organized as follows: Section 2 explains the data and methodology, Section 3 presents the empirical findings, Section 4 discusses the findings, and Section 5 concludes the paper with policy recommendations and directions for future research.

2. Literature Review

Research on investment behavior has consistently emphasized the role of psychological biases in shaping responses to market volatility. Among these, hyperbolic discounting, or present-biased preferences, has been identified as a central mechanism leading to time-inconsistent decisions (Laibson 1997; O’Donoghue and Rabin 1999). Individuals who exhibit stronger hyperbolic discounting tend to overweight immediate outcomes relative to future ones, leading to behaviors that undermine long-term financial goals (Ikeda 2016; McClure et al. 2004). These tendencies manifest in reduced self-control, sensitivity to short-term losses, and frequent trading, which collectively influence investment decisions. Recent research underscores the intertwined nature of time and risk preferences, showing that discounting directly affects portfolio risk-taking (Epper and Fehr-Duda 2023). Despite these advances, methodological challenges remain: elicitation methods for discount rates produce heterogeneous results (Andersen et al. 2008; Tasoff and Zhang 2020), and questions about the temporal stability of hyperbolic discounting persist (Meier and Sprenger 2015).

Investment loss tolerance refers to the extent to which investors are willing and able to endure declines in the value of their holdings without liquidating prematurely (Braga and Fávero 2017). Investors with higher tolerance are generally more inclined to pursue long-term returns, while those with lower tolerance often react adversely to short-term fluctuations by selling too early (Homma et al. 2025). Understanding how psychological biases shape loss tolerance is therefore crucial for both advisors and individuals seeking to align decisions with long-term financial goals. Yet the relationship between loss tolerance and cognitive biases remains underexplored. Hyperbolic discounting, which reflects the tendency to disproportionately prefer smaller immediate rewards over larger delayed ones, may directly undermine loss tolerance by causing investors to devalue future gains relative to current setbacks. This time-inconsistent preference structure reduces the willingness to endure temporary volatility, even when such losses are recoverable. The rational perspective on loss tolerance assumes that short-term price movements are largely random and self-correcting, supporting disciplined long-term strategies (Chitenderu et al. 2014). Empirical work suggests that markets with higher investor loss tolerance, such as the United States, are associated with stronger long-term investment behaviors (Hellmann et al. 2024). In contrast, behavioral research shows that many investors act irrationally under losses, realizing them prematurely and locking in avoidable costs (Odean 1998; Benartzi and Thaler 1995). This contradiction between rational expectations and actual behavior underscores the need to examine the influence of biases such as hyperbolic discounting, which may substantially impair the ability of investors to stay invested during volatility.

These dynamics are particularly salient in Japan, which offers a distinctive setting for analyzing the relationship between hyperbolic discounting and loss tolerance. Culturally, Japanese society emphasizes long-term orientation yet is also characterized by high risk aversion and collectivist norms that influence financial decision-making (Hofstede 2001; Kitayama and Uchida 2005). Institutionally, Japanese markets feature high levels of retail participation and detailed brokerage records, which provide strong empirical bases for studying behavioral biases (Kamesaka et al. 2003). Comparative evidence highlights cross-country heterogeneity in time preferences, with Japan displaying distinctive patterns compared to Western economies (Wang et al. 2016; Chen et al. 2005). Within Japan, hyperbolic discounting has been linked to borrowing aversion and debt holding (Ikeda and Kang 2015), while neurobiological studies reveal asymmetries in how Japanese subjects process intertemporal gains and losses at the neural level (Tanaka et al. 2014). Empirical studies further show that hyperbolic discounting predicts impulsive financial behavior during crises and contributes to lower asset accumulation among older investors (Lal et al. 2024; Nabeshima et al. 2025). Laboratory experiments confirm that evaluation frequency and hyperbolic discounting jointly drive myopic loss aversion in Japanese contexts (Shoji and Kanehiro 2012). Field evidence also demonstrates that interventions such as defaults, commitment devices, and framing long-horizon outcomes mitigate present-bias effects (Madrian and Shea 2001; Benartzi and Thaler 1995).

Beyond individual and methodological determinants, studies have demonstrated that discounting behavior also varies systematically across cultures and regions. Large-scale international evidence shows that time preferences differ across more than 40 countries and are strongly correlated with cultural dimensions such as individualism and uncertainty avoidance, with higher levels associated with stronger hyperbolic discounting (Wang et al. 2016). Cultural studies similarly show that Western participants tend to discount future rewards more steeply than Eastern participants, partly due to stronger immediate reward sensitivity (Kim et al. 2012). Neuroeconomic studies confirm that Japanese participants display relatively lower impulsivity and greater intertemporal consistency compared to American counterparts, suggesting cultural differences in cognitive control mechanisms (Takahashi et al. 2009). These findings underscore that discounting behavior is not uniform but shaped by deeper cultural and institutional contexts, reinforcing the importance of analyzing Japanese investors as a distinctive case.

Taken together, the literature demonstrates that hyperbolic discounting shapes impulsive financial behaviors, reduces long-term commitment, and interacts with biases such as myopic loss aversion and overreaction to information (Lal et al. 2024; Hong et al. 2004). These effects are moderated by financial literacy, investment experience, and cognitive ability (Lusardi and Mitchell 2014; Seru et al. 2010) and tend to be amplified under conditions of volatility and frequent information exposure. However, no prior research has directly examined the relationship between hyperbolic discounting and investors’ tolerance of financial losses. Existing studies either analyze hyperbolic discounting in consumption, saving, or crisis-driven trading or focus on general self-control and risk attitudes. The absence of direct empirical evidence represents a critical gap in behavioral finance. This study addresses that gap by being the first to test whether and how hyperbolic discounting influences loss tolerance among Japanese investors in ordinary market conditions, using large-scale brokerage data and ordered probit estimation. By situating the analysis in Japan’s distinctive cultural and institutional setting, it provides a novel empirical link between time preferences and loss tolerance, thereby extending the behavioral finance literature both contextually and theoretically.

3. Data and Methods

3.1. Data

This study utilized data from the 2025 wave of the “Survey on Life and Money,” conducted by Rakuten Securities in collaboration with Hiroshima University, Japan. Rakuten Securities is Japan’s leading online securities company. The survey was administered online between January and February 2025 to Rakuten Securities account holders aged 18 years and older who had logged into the Rakuten Securities website at least once within the 12 months before the survey. In addition to demographic, economic, and psychological items, the questionnaire included measures to assess the respondents’ willingness to tolerate investment risks and their inconsistency in time preferences (or decision-making).

The initial sampling frame included 231,110 account holders, reflecting a large and diverse pool of active Japanese retail investors. From this pool, 107,294 unique respondents provided complete and valid survey responses. Each observation corresponds to one individual, and no repeated measures were taken across the sample. The exclusion criterion was applied consistently across all survey items, and descriptive statistics of the key variables before and after data cleaning indicated no significant differences, supporting the assumption that missing values were missing at random (Appendix A.2). This procedure reduces the risk of systematic bias and enhances the reproducibility of the dataset (see Supplementary Material for full dataset).

The final sample size of 107,294 respondents provides very high statistical power. Ex-ante power analysis indicated that this sample size yields power above 0.99 to detect even small effect sizes (Cohen’s f² = 0.02) at the 5% significance level in regression models. Thus, the study is well positioned to produce precise estimates and robust inferences.

The choice of Japanese investors is motivated not only by access to high-quality data but also by substantive reasons. Japan is one of the world’s largest and most mature financial markets, with a high proportion of household assets held in cash and deposits relative to equities, a pattern that contrasts with many Western economies. This makes Japanese retail investors a particularly informative group for examining attitudes toward risk-taking and intertemporal decision-making. Furthermore, the rapidly aging population in Japan provides a unique context to study how demographic shifts influence investment behavior and time preferences. By focusing on this population, the study contributes insights with relevance both to Japan and to broader discussions about financial literacy, retirement security, and global capital markets.

3.2. Variables

3.2.1. Dependent Variable

Investment loss tolerance served as the dependent variable. It is defined as the percentage of loss an investor is willing to endure while holding an investable asset during periods of market volatility and uncertainty. To measure this construct, the survey asked respondents to indicate the maximum loss they would be willing to tolerate for an investment trust initially valued at one million Japanese yen. Respondents were given five options: 990,000 yen (1% loss), 900,000 yen (10% loss), 800,000 yen (20% loss), 700,000 yen (30% loss), or 600,000 yen (40% loss or more). Based on their choices, we created a categorical variable, loss tolerance, with values of 1%, 10%, 20%, 30%, and 40% or more, reflecting the respondents’ levels of tolerance for investment losses.

It is important to distinguish this construct from related concepts. Risk tolerance/aversion refers to preferences over probabilistic payoffs, typically elicited through gambles or portfolio-choice tasks. Loss aversion (Kahneman and Tversky 1979) captures the asymmetry in how individuals weigh losses relative to equivalent gains. Myopic loss aversion (Benartzi and Thaler 1995) combines loss aversion with frequent evaluation, explaining why investors shy away from long-term investments when they monitor short-term fluctuations. By contrast, loss tolerance, as operationalized here, is narrower and behaviorally grounded: it reflects the maximum percentage decline in asset value an investor reports being willing to endure before liquidation. While this overlaps conceptually with “maximum tolerable drawdown” or “sell-threshold,” we retain the label loss tolerance because it has precedent in prior finance and behavioral economics research (Braga and Fávero 2017; Nabeshima et al. 2025), where similar survey-based measures were used to capture investors’ endurance of temporary financial losses under uncertainty.

Our choice of this operationalization is motivated by both theoretical and practical considerations. Theoretically, prospect theory emphasizes reference-dependent evaluations, and our categorical design (1%, 10%, 20%, 30%, and ≥40%) provides explicit reference points that mirror how investors frame losses in practice. Practically, using a yen-denominated, investment-trust scenario ensures realism for Japanese investors, improves reliability, and minimizes respondent burden. Framing the measure in round monetary amounts reduces respondent burden and improves reliability. Moreover, the categorical design aligns with prospect theory’s emphasis on reference-dependent evaluation of losses (Kahneman and Tversky 1979) and is consistent with survey-based approaches to measuring risk and loss tolerance in both Western and Asian contexts (Weber et al. 2002; Nabeshima et al. 2025). To address potential biases associated with hypothetical survey questions, the scenario emphasized realistic conditions of market volatility and uncertainty. This framing encouraged respondents to consider actual behavioral tendencies rather than idealized preferences, thereby strengthening the validity of the measure in the Japanese context.

3.2.2. Independent Variables

The primary independent variable in this study, hyperbolic discounting, was measured using two hypothetical scenarios designed to elicit respondents’ preferences for receiving different monetary rewards after specific time delays. Respondents were first asked, “Suppose you were given a certain amount of money that you would choose to receive after a short period (option A) or a longer period (option B), but the amounts differ, which one would you choose?” These scenarios aimed to assess time preferences by altering the delay length. In the first scenario, participants chose between receiving money after two or nine days; in the second scenario, they chose between 90 and 97 days. Each scenario included eight different combinations in which the monetary reward increased progressively from Combinations 1 to 8. Following Laibson (1997) and O’Donoghue and Rabin (1999), the two-horizon switching-point design allows us to detect declining discount rates across short and long delays, thereby capturing hyperbolic discounting, the hallmark of hyperbolic (rather than exponential) discounting. Appendix A.1 presents these questions.

To measure hyperbolic discounting, we identified the discount rate in each scenario at the switching point. In other words, the respondents changed their choice from option A to option B. After calculating discount rates 1 (DR1) and 2 (DR2) for the two scenarios based on interest rates at the switching point, we employed a binary approach to define hyperbolic discounters. We assigned a value of 1 to hyperbolic discounters where DR1 > DR2, and 0 otherwise. This approach is consistent with the methodologies used in previous studies (Zhang 2016; Bawalle et al. 2024) to operationalize hyperbolic discounting.

To enhance reliability, inconsistent responses were excluded when participants switched back and forth more than twice within the same task, following best practices in prior Japanese studies (Nabeshima et al. 2025; Bawalle et al. 2024). The validity of this approach is supported by earlier Japanese research demonstrating that monetary trade-off tasks provide culturally relevant and stable measures of discounting (Tanaka et al. 2014; Ikeda and Kang 2015). Furthermore, the inclusion of both short-term and long-term horizons ensures that the method directly identifies time-inconsistent preferences, which are central to hyperbolic discounting. Although hypothetical survey questions may introduce response biases, the design minimized this risk by grounding amounts and delays in realistic Japanese financial contexts, employing multiple tasks to cross-validate switching behavior, and controlling for demographic, socioeconomic, and behavioral covariates (e.g., household income, education, health risk behaviors). These steps increase the robustness and contextual validity of the hyperbolic discounting measure in the Japanese setting.

Other independent variables included gender, age, the square of age, marital status, years of education, having a child, engaging in full-time employment, annual income, balance of financial assets, risk aversion, and a myopic view of later life. The detailed definitions are provided in

Table 1. Variable definitions.

Variables	Definition
Dependent Variable
Investment loss tolerance	Discrete variable: How much loss respondents can withstand if they invest JPY 1 million in an investment trust (1%/10%/20%/30%/40% loss)
Investment loss tolerance dummy	Binary variable: 1 = respondents can withstand a loss of 30% or more if they invest JPY 1 million in an investment trust
Independent Variable
Hyperbolic discounting	Binary variable: 1 = respondents’ DR1 exceeds DR2, 0 = otherwise
Gender	Binary variable: 1 = male, 0 = female
Age	Continuous variable: Respondents’ age
Age squared	Continuous variable: Squared of respondents’ age
Marital status	Binary variable: 1 = having a spouse, 0 = otherwise
Number of children	Continuous variable: Number of children
Education year	Continuous variable: Years of education
Having a job	Binary variable: 1 = having a full time job, 0 = otherwise
Household income	Continuous variable: Total annual income, including tax, for the household in 2024 in Japanese Yen
Household assets	Continuous variable: Total household balance of financial assets in 2024 in Japanese Yen
Risk aversion	Continuous variable: A measure of respondents’ risk aversion (the answer to the following question: When you usually go out with an umbrella, what is the probability of rain?)
Myopic view of the future	Binary variable: 1 if the respondent agrees that the future is uncertain and there is no point in thinking about it, and 0 otherwise.

.

3.3. Descriptive Statistics

Table 2. Descriptive statistics.

Variable	Mean (or %)	Std. Dev.	Min	Max
Dependent Variable
Loss Tolerance	24.8%
1%	4.50%
10%	23.7%
20%	23.6%
30%	15.8%
≥40%	32.4%
Independent Variables
Hyperbolic Discounting	12.80%
Gender (Male = 1)	64.3%
Age	45.019	11.889	18	90
Age Squared	2168	1117	324	8100
Age Group
18–39	35.90%
40–65	59.3%
>65	4.80%
Marital Status (Married = 1)	66.20%
Children	1.088	1.104	0	12
Years of Education	15.195	2.057	9	21
Full-time Job	70.60%
Annual Income in 2024	7,762,545	4,253,144.9	1,000,000	20,000,000
Household Financial Assets in 2024	21,496,228	25,354,522	2,500,000	100,000,000
Household Asset Group
Low Household Asset	60.50%
High Household Asset	39.50%
Natural Log of Annual Income	15.701	0.611	13.816	16.811
Natural Log of Household Assets	16.277	1.104	14.732	18.421
Risk Aversion	53.50%
Myopic View of the Future	14.80%
Number of Observations: 107,294

presents the descriptive statistics for the variables used in this study. The dependent variable, loss tolerance, shows an average level of 24.8%. The categorical distribution indicates that 4.5% of respondents reported a tolerance of only 1% loss, 23.7% tolerated up to 10%, 23.6% up to 20%, 15.8% up to 30%, and the largest share, 32.4%, indicated a tolerance of 40% or more. Males comprised approximately 64% of the respondents, and the average age of the sample was 45 years (SD = 12); approximately 66% of respondents were married and had at least one child. The average number of years of education was 15 (SD = 2), and approximately 70% of respondents were employed full-time. In 2024, their mean income was approximately 7,762,545 Japanese yen, and the average household balance of financial assets was approximately 21,496,228 yen. Additionally, 53% of respondents reported being risk-averse, and approximately 15% believed that the future is uncertain and that thinking about it is a waste of time.

Figure 1 shows the distribution of hyperbolic discounting across different levels of investment loss tolerance. The proportion of hyperbolic discounters remains fairly stable in the 1–30% categories, ranging between 12–14%, but declines in the ≥40% group. This pattern suggests that investors with the highest tolerance for losses are less likely to exhibit hyperbolic discounting, supporting the theoretical expectation that present-biased individuals are less willing to endure substantial short-term losses.

Table 3. Tabulation of loss tolerance by gender, age, and household financial assets.

Loss Tolerance	Gender		Age Group			Asset Group
	Females	Male	18–39	40–65	>65	Low	High	Total
1%	2853	2004	1915	2724	218	3820	1037	4857
	58.74	41.26	39.43	56.08	4.49	78.65	21.35	100.00
	7.44	2.91	4.97	4.28	4.26	5.89	2.45	4.53
10%	11,241	14195	9008	14,893	1535	18,079	7357	25,436
	44.19	55.81	35.41	58.55	6.03	71.08	28.92	100.00
	29.31	20.59	23.39	23.39	29.99	27.86	17.35	23.71
20%	9014	16,267	8677	15,077	1527	15,713	9568	25,281
	35.66	64.34	34.32	59.64	6.04	62.15	37.85	100.00
	23.50	23.60	22.53	23.68	29.84	24.22	22.56	23.56
30%	5797	11,164	5800	10,296	865	9490	7471	16,961
	34.18	65.82	34.20	60.70	5.10	55.95	44.05	100.00
	15.11	16.19	15.06	16.17	16.90	14.63	17.61	15.81
≥40%	9450	25,309	13,105	20,681	973	17,779	16,980	34,759
	27.19	72.81	37.70	59.50	2.80	51.15	48.85	100.00
	24.64	36.71	34.03	32.48	19.01	27.40	40.03	32.40
Total	38,355	68,939	38,505	63,671	5118	64,881	42,413	107,294
	35.75	64.25	35.89	59.34	4.77	60.47	39.53	100.00
	100.00	100.00	100.00	100.00	100.00	100.00	100.00	100.00
	Pearson Chi2 = 3035.05 Prob = 0.0000		Pearson Chi2 = 556.90 Prob = 0.0000			Pearson Chi2 = 3306.68 Prob = 0.0000

The first row has frequencies; the second row has row percentages; and the third row has column percentages.

presents the distribution of loss tolerance by gender, age, and household balance of financial assets. The positive Chi-square statistics indicate a strong statistical association between loss tolerance and gender (χ² = 3035; p < 0.001). Additionally, a significant association exists between loss tolerance and the household balance of financial assets (χ² = 3300; p < 0.001), as well as a modest relationship between loss tolerance and age (χ² = 556.9; p < 0.001).

3.4. Empirical Model

This study examined the relationship between investment loss tolerance and hyperbolic discounting among Japanese investors using an ordered probit regression. Loss tolerance was operationalized as an ordinal dependent variable on a five-point scale (1% loss, 10% loss, 20% loss, 30% loss, and 40% or more loss). Hyperbolic discounting was operationalized as a binary independent variable (1 = hyperbolic discounter, 0 = otherwise), derived from the comparison of short-term and long-term discount rates.

Statistical analyses were performed using Stata 18 software. To demonstrate robustness and incremental explanatory power, we estimated three sequential model specifications. Model 1 included only demographic variables (e.g., gender, age, education, and marital status), Model 2 added economic variables (income and assets), and Model 3 further incorporated psychological covariates (risk aversion and myopic view of the future). This stepwise approach allows us to show (i) that the main effect of hyperbolic discounting remains stable across different levels of model complexity and (ii) how explanatory power improves when theoretically relevant variables are introduced. While Model 3 is our preferred specification, reporting Models 1 and 2 ensures transparency and demonstrates that the findings are not driven by omitted-variable bias or overfitting.

Potential confounding factors were carefully addressed by including a wide range of demographic, socioeconomic, and psychological covariates. Risk aversion and myopic views of the future were explicitly modeled due to their known association with both hyperbolic discounting and loss tolerance. Although loss tolerance and risk aversion are related, they are not the same because risk aversion reflects ex ante preferences over probabilistic payoffs, whereas loss tolerance is an ex-post liquidation threshold for realized drawdowns. We therefore retain risk aversion as a theoretically necessary control to isolate the time-preference channel. By accounting for these factors and applying robustness checks, the likelihood of omitted variable bias was substantially reduced, strengthening the validity of the estimated relationships.

The estimated models are presented in Equations (1)–(3):

$$\begin{array}{ll}{Loss Tolerance}_i& ={\beta }_0+{{\beta }_{1}Hyperbolic Discounting}_i+{\beta }_2{Gender}_i+{\beta }_3{Age}_i+{\beta }_4{Age Square}_i\\ & +{\beta }_5{Marital Status}_i+{\beta }_6{Childreen}_i+{\beta }_7{Education}_i+{\beta }_8{Fulltime Job}_i+{\epsilon }_i\end{array}$$

(1)

$$\begin{array}{ll}{Loss Tolerance}_i& ={\beta }_0+{{\beta }_{1}Hyperbolic Discounting}_i+{\beta }_2{Gender}_i+{\beta }_3{Age}_i+{\beta }_4{Age Square}_i\\ & +{\beta }_5{Marital Status}_i+{\beta }_6{Childreen}_i+{\beta }_7{Education}_i+{\beta }_8{Fulltime Job}_i\\ & +{\beta }_9{Log of Income}_i+{\beta }_{10}{Log of Assets}_i+{\epsilon }_i\end{array}$$

(2)

$$\begin{array}{ll}{Loss Tolerance}_i& ={\beta }_0+{{\beta }_{1}Hyperbolic Discounting}_i+{\beta }_2{Gender}_i+{\beta }_3{Age}_i+{\beta }_4{Age Square}_i\\ & +{\beta }_5{Marital Status}_i+{\beta }_6{Childreen}_i+{\beta }_7{Education}_i+{\beta }_8{Fulltime Job}_i\\ & +{\beta }_9{Log of Income}_i+{\beta }_{10}{Log of Assets}_i+{\beta }_{11}{Risk Aversion}_i\\ & +{\beta }_{12}{Myopic View of Future}_i+{\epsilon }_i\end{array}$$

(3)

where ${\beta }_0$ represents the intercept, ${\epsilon }_i$ denotes the error term, and ${\beta }_1-{\beta }_{12}$ are the estimated parameters. To test for multicollinearity in our model, we examined the correlation between the model variables using a pairwise correlation approach and conducted a post-estimation variance inflation factor (VIF) test. The results indicate a weak correlation between the variables (correlation coefficients < 0.6) and a VIF below 10. This finding indicates that our model is unlikely to be affected by multicollinearity. The results of pairwise correlation and VIF analyses are available upon request.

4. Empirical Results

4.1. Main Results

We begin by presenting the main results using ordered probit regressions, where loss tolerance is measured as an ordinal variable on a five-point scale.

Table 4. Ordered Probit Regression Results Explaining Investment Loss Tolerance.

Dependent Variable: Investment Loss Tolerance (Categorical: 1–40%)
	Model 1 Coef. (SE)	Model 2 Coef. (SE)	Model 3 Coef. (SE)	Marginal Effect (Model 3)
Hyperbolic Discounting	−0.07 ***	−0.063 ***	−0.063 ***	−0.070 ***
	(0.01)	(0.01)	(0.01)
Gender	0.382 ***	0.387 ***	0.388 ***	0.435 ***
	(0.007)	(0.007)	(0.007)
Age	0.043 ***	0.029 ***	0.029 ***	0.033 ***
	(0.002)	(0.002)	(0.002)
Age Squared	0.000 ***	0.000 ***	0.000 ***	−0.001 ***
	(0.000)	(0.000)	(0.000)
Marital Status	−0.04 ***	−0.101 ***	−0.102 ***	−0.113 ***
	(0.008)	(0.009)	(0.009)
Children	−0.035 ***	−0.015 ***	−0.016 ***	−0.019 ***
	(0.004)	(0.004)	(0.004)
Education	0.022 ***	−0.01 ***	−0.01 ***	−0.012 ***
	(0.002)	(0.002)	(0.002)
Full-time Job	0.024 ***	0.008	0.008	0.012 ***
	(0.008)	(0.009)	(0.009)
Natural Log of Annual Income		0.007	0.008	0.005 ***
		(0.007)	(0.007)
Natural Log of Household Assets		0.255 ***	0.257 ***	0.286 ***
		(0.004)	(0.004)
Risk Aversion			−0.091 ***	−0.113 ***
			(0.015)
Myopic View of the Future			−0.01	−0.056 ***
			(0.009)
/cut1	−0.376 ***	2.824 ***	2.802 ***
	(0.051)	(0.104)	(0.105)
/cut2	0.768 ***	4.002 ***	3.981 ***
	(0.05)	(0.105)	(0.105)
/cut3	1.404 ***	4.659 ***	4.639 ***
	(0.051)	(0.105)	(0.105)
/cut4	1.825 ***	5.094 ***	5.074 ***
	(0.051)	(0.105)	(0.105)
Observations	107,294	107,294	107,294
Pseudo R2	0.014	0.032	0.032

Robust standard errors are in parentheses, *** p < 0.01. Higher coefficients indicate greater willingness to tolerate losses before altering investment strategies.

reports three stepwise model specifications. Model 1 includes only demographic factors, Model 2 adds economic variables, and Model 3 incorporates psychological covariates. Among them, Model 3 is our preferred specification as it is the most comprehensive.

The estimated regression coefficients and marginal effects are presented in Table 4. Pseudo R² values (0.014–0.032 across models) indicate that while the models explain a modest share of variance, common in behavioral and social science research, the coefficients remain statistically and substantively meaningful. The reported/cut1–/cut4 thresholds are cut-points from the ordered probit model, representing estimated breakpoints on the unobserved latent scale of loss tolerance that separate adjacent response categories (1%, 10%, 20%, 30%, and ≥40% loss). These thresholds are necessary for estimation but do not carry direct substantive interpretation.

The findings indicate a negative association between investment and loss tolerance and hyperbolic discounting. This suggests that the cognitive bias of hyperbolic discounting reduces investors’ willingness to absorb losses during periods of market volatility and uncertainty. Furthermore, male and older investors are more likely to endure losses during challenging market conditions. By contrast, being married or having children tends to decrease investors’ willingness to bear investment losses. Additionally, owning financial assets increases investors’ tendency to hold on to their positions. However, both risk aversion and a myopic view of the future are negatively associated with investment loss tolerance. The coefficient for having a myopic view of the future was found to be insignificant. Overall, the coefficients in all three models remained consistent except for education, which changed from positive and significant to negative and significant after incorporating economic and psychological factors. Similarly, the statistical significance of having a full-time job diminished upon the inclusion of these economic and psychological factors in the models.

We performed a subsample analysis based on gender, age, and level of financial assets. The findings are summarized in

Table 5. Subsample Ordered Probit Regression Results for Investment Loss Tolerance by Gender, Age, and Financial Assets.

Dependent Variable: Investment Loss Tolerance (Ordered Probit)
	Male Coef. (SE)	Female Coef. (SE)	18–39 Coef. (SE)	40–65 Coef. (SE)	>65 Coef. (SE)	Low Assets Coef. (SE)	High Assets Coef. (SE)
Hyperbolic Discounting	−0.084 ***	−0.021	−0.078 ***	−0.053 ***	−0.069 *	−0.055 ***	−0.076 ***
	(0.012)	(0.017)	(0.017)	(0.012)	(0.038)	(0.012)	(0.015)
Gender			0.438 ***	0.367 ***	0.186 ***	0.418 ***	0.338 ***
			(0.012)	(0.01)	(0.043)	(0.009)	(0.013)
Age	0.024 ***	0.029 ***	0.073 ***	0.032 ***	−0.096	0.021 ***	0.021 ***
	(0.002)	(0.003)	(0.016)	(0.009)	(0.123)	(0.003)	(0.004)
Age Squared	0.000 ***	0 ***	−0.001 ***	0.000 ***	0.000	0.000 ***	0.000 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.001)	(0.000)	(0.000)
Marital Status	−0.109 ***	−0.075 ***	−0.07 ***	−0.125 ***	−0.12 ***	−0.115 ***	−0.092 ***
	(0.012)	(0.015)	(0.015)	(0.012)	(0.044)	(0.011)	(0.015)
Children	−0.006	−0.032 ***	−0.022 ***	−0.015 ***	−0.011	−0.025 ***	−0.004
	(0.005)	(0.006)	(0.008)	(0.004)	(0.016)	(0.005)	(0.006)
Education	−0.013 ***	0.002	−0.002	−0.012 ***	−0.007	−0.008 ***	−0.011 ***
	(0.002)	(0.003)	(0.003)	(0.002)	(0.008)	(0.002)	(0.003)
Full-time Job	−0.002	0.029 **	0.028 *	0.000	0.013	−0.018	0.043 ***
	(0.013)	(0.013)	(0.016)	(0.011)	(0.048)	(0.011)	(0.014)
Natural Log Income	−0.005	0.021 *	0.011	0.002	0.046	0.036 ***	−0.039 ***
	(0.009)	(0.012)	(0.014)	(0.009)	(0.03)	(0.01)	(0.011)
Natural Log of Household Assets	0.254 ***	0.262 ***	0.287 ***	0.25 ***	0.203 ***	0.296 ***	0.235 ***
	(0.004)	(0.006)	(0.007)	(0.005)	(0.016)	(0.008)	(0.01)
Risk Aversion	−0.132 ***	−0.007	−0.106 ***	−0.081 ***	−0.104	−0.108 ***	−0.065 ***
	(0.018)	(0.026)	(0.025)	(0.019)	(0.073)	(0.019)	(0.024)
Myopic View of the Future	0.003	−0.031 **	−0.01	−0.017	0.044	−0.006	−0.019
	(0.012)	(0.015)	(0.015)	(0.013)	(0.05)	(0.012)	(0.016)
/cut1	1.891 ***	3.486 ***	4.241 ***	2.579 ***	−2.072	3.75 ***	1.395 ***
	(0.134)	(0.172)	(0.305)	(0.264)	(4.423)	(0.164)	(0.212)
/cut2	3.108 ***	4.631 ***	5.379 ***	3.768 ***	−0.704	4.942 ***	2.551 ***
	(0.134)	(0.173)	(0.305)	(0.264)	(4.423)	(0.164)	(0.212)
/cut3	3.788 ***	5.253 ***	6.012 ***	4.43 ***	0.089	5.587 ***	3.234 ***
	(0.134)	(0.173)	(0.306)	(0.264)	(4.423)	(0.164)	(0.212)
/cut4	4.22 ***	5.696 ***	6.426 ***	4.873 ***	0.62	6.002 ***	3.698 ***
	(0.134)	(0.173)	(0.306)	(0.264)	(4.423)	(0.164)	(0.212)
Observations	68,939	38,355	38,505	63,671	5,118	64,881	42,413
Pseudo R2	0.026	0.021	0.036	0.031	0.02	0.023	0.024

Robust standard errors are in parentheses, *** p < 0.01, ** p < 0.05, and * p < 0.10. Higher coefficients indicate greater willingness to tolerate losses before altering investment strategies.

. The results indicate that males are more likely than females to exhibit lower investment loss tolerance. Furthermore, the coefficient for ages 39–40 was more negative than that for ages 40–65, and the coefficient for ages 40–65 was more positive than that for ages older than 65. This suggests that loss tolerance increases with age until 65 years of age and then starts to decline. The results also highlight that investors with more financial assets are less inclined to sell off their investments when facing market challenges than those with fewer assets.

4.2. Robustness Checks

To check the robustness of the model, we constructed an alternative measure of investment loss tolerance and used a probit regression estimation for the analysis. Specifically, we created a binary variable by categorizing respondents into two groups. A value of 1 was assigned to respondents who were willing to tolerate investment losses of 30% or more, whereas a value of 0 was assigned to those who would rather divest their investment assets before experiencing a loss of 30%.

Table 6. Robustness Check Using Probit Regression Models of Investment Loss Tolerance.

Dependent Variable: Investment Loss Tolerance (Binary, ≥30% Loss)
	Model 4 Coef. (SE)	Model 5 Coef. (SE)	Model 6 Coef. (SE)
Hyperbolic Discounting	−0.074 ***	−0.067 ***	−0.065 ***
	(0.012)	(0.012)	(0.012)
Gender	0.353 ***	0.356 ***	0.351 ***
	(0.009)	(0.009)	(0.009)
Age	0.048 ***	0.035 ***	0.035 ***
	(0.002)	(0.002)	(0.002)
Age Squared	−0.001 ***	−0.001 ***	−0.001 ***
	(0.000)	(0.000)	(0.000)
Marital Status	−0.051 ***	−0.103 ***	−0.105 ***
	(0.01)	(0.011)	(0.011)
Children	−0.042 ***	−0.02 ***	−0.022 ***
	(0.004)	(0.004)	(0.004)
Education	0.016 ***	−0.017 ***	−0.017 ***
	(0.002)	(0.002)	(0.002)
Full-time Job	0.015	0.009	0.01
	(0.01)	(0.01)	(0.01)
Natural Log of Annual Income		−0.016 *	−0.018 **
		(0.009)	(0.009)
Natural Log of Household Assets		0.272 ***	0.27 ***
		(0.004)	(0.004)
Risk Aversion			−0.096 ***
			(0.017)
Myopic View of the Future			−0.053 ***
			(0.004)
_Cons	−1.386 ***	−4.514 ***	−4.28 ***
	(0.06)	(0.123)	(0.124)
Observations	107,294	107,294	107,294
Pseudo R2	0.02	0.051	0.052

Robust standard errors are in parentheses, *** p < 0.01, ** p < 0.05, and * p < 0.10. Higher coefficients indicate greater willingness to tolerate losses before altering investment strategies.

presents the results from probit regressions, again estimated stepwise to parallel the main analysis. As before, Model 6 is the preferred and most comprehensive specification. These results are consistent with the ordered probit results shown in Table 4, except that the coefficient for age squared changed from positive to negative, and the coefficient for income shifted from positive and insignificant to negative and significant at the 10% level. This suggests that our findings would remain robust under different assumptions and estimation methods.

5. Discussion and Conclusions

This study examined the influence of hyperbolic discounting on investment loss tolerance using a large-scale dataset of Japanese investors. The results reveal a significant negative association between hyperbolic discounting and investment loss tolerance, indicating that individuals who exhibit stronger hyperbolic discounting are less likely to hold onto losing investments during periods of market volatility and uncertainty. This finding is consistent with the theoretical framework of behavioral finance, where hyperbolic discounting produces time-inconsistent preferences that undermine rational long-term strategies (Ikeda and Kang 2015; O’Donoghue and Rabin 1999). By devaluing future returns, hyperbolic discounters perceive short-term losses as disproportionately painful, lowering their tolerance for volatility and driving premature liquidation of assets.

Several mechanisms can explain this association. First, hyperbolic discounting reflects a cognitive bias in which individuals irrationally prioritize immediate gratification over future benefits (Ikeda and Kang 2015). This tendency may undermine long-term investments in market understanding, leading to inadequate preparation for navigating market volatility. Consequently, investors may misinterpret short-term market signals and engage in premature asset liquidation during market downturns. Second, individuals exhibiting hyperbolic discounting are more prone to impulsive decision-making (O’Donoghue and Rabin 1999), which increases their vulnerability to overreacting to short-term volatility (Gathergood et al. 2019; Lal et al. 2024). This impulsivity can further erode investment discipline. Third, existing studies link hyperbolic discounting with heightened loss aversion (Katauke et al. 2023), suggesting that the fear of financial losses may compel such investors to exit the market prematurely, even in situations where retaining their investment could yield positive long-term returns.

These findings indicate that hyperbolic discounting diminishes investors’ loss tolerance, increasing the likelihood of prematurely selling underperforming assets. Such behavior, particularly during periods of market volatility, may exacerbate negative market sentiment if not addressed effectively. Based on these insights, we propose several policy- and practice-oriented recommendations. First, a financial education program grounded in behavioral finance principles should be promoted at the policy level. These programs should focus on equipping investors with strategies to manage their emotional and psychological responses during episodes of market uncertainty and turbulence. Emphasizing behavioral self-awareness and resilience can help mitigate the adverse effects of present-biased decision-making.

Second, financial institutions should develop interactive platforms offering real-time investment guidance and facilitating communication with clients during unexpected downturns. Delivering accurate and updated information can play a critical role in curbing impulsive reactions and reinforcing rational decision-making. Institutions should also offer on-demand educational resources to help investors continually improve their knowledge and skills, thereby reducing their tendency to liquidate loss-incurring investments impulsively. Third, financial advisors should recognize that some clients may have strong present-biased preferences. Tools such as long-term portfolio framing, scenario simulations, and commitment devices may help mitigate impulsive reactions to volatility. Fourth, individual investors need to recognize their personal tolerance of investment losses and susceptibility to intertemporal bias. Developing awareness of one’s behavioral tendencies, such as identifying specific loss thresholds and acknowledging time-inconsistent preferences, can significantly reduce emotionally driven decisions and promote more disciplined investment practices.

While the study offers important contributions, the findings should be interpreted with several limitations. First, although the Rakuten Securities dataset is large and geographically diverse, it represents primarily active online investors, which may limit the external validity of the findings. Second, the possibility of cultural bias cannot be excluded: Japanese norms around savings, risk, and financial prudence may influence loss tolerance in ways that differ from other national contexts. Third, the study employs a cross-sectional design, which prevents strong causal inferences. Longitudinal data would allow for more robust analysis of how hyperbolic discounting and loss tolerance evolve over time. Fourth, our measure of loss tolerance is based on self-reported survey responses, which may be subject to interpretation bias. In addition, while the primary ordinal measure aligns with how investors typically express tolerance levels, the 30% threshold used in the binary specification for robustness checks is somewhat arbitrary and should be viewed as a practical approximation rather than a definitive benchmark. This limitation reflects the absence of an established standard for defining loss tolerance. Finally, while the pseudo R² values are modest, this is typical in social science and behavioral economics, where behavior is shaped by many unobserved factors. Future research could improve explanatory power by incorporating longitudinal data, richer psychological measures, and experimental designs to capture a wider range of behavioral mechanisms. Nevertheless, the large dataset size and demographic diversity of the respondents strengthen the credibility of the results. To strengthen the generalizability and causal interpretation of future research, future studies should incorporate nationally representative samples and longitudinal data. Furthermore, examining a wider range of behavioral biases would provide a more comprehensive understanding of the cognitive mechanisms that influence investment loss tolerance.

In conclusion, this study demonstrates that hyperbolic discounting significantly reduces investors’ tolerance for financial losses. By analyzing a large-scale dataset of Japanese investors, we extend behavioral finance research by establishing a direct empirical link between time-inconsistent preferences and loss tolerance. These findings highlight the importance of considering cultural and psychological factors in financial decision-making. For financial advisors, the results underscore the value of interventions such as commitment devices, investor education, and framing strategies that foster long-term perspectives and resilience to short-term volatility.