Herd bias

Herding (HRD) refers to the tendency of investors to imitate others’ behavior rather than relying on their own information, especially in uncertain environments [23,24]. This behavior is particularly strong in frontier markets, where informational transparency is limited [17,25]. Herd behavior is evolutionary and may decline over time as investors learn from mistakes [26]. Prior studies indicate that trading experience, financial literacy, and a desire for learning can mitigate herding while promoting rational decision-making [27,28]. Conversely, social indications, such as word-of-mouth communication or peer influence can significantly intensify herding behavior [29]. Given its widespread destabilizing impact on financial systems [30], understanding the drivers of herding is vital for improving market behavior and investor decision-making.

Overconfidence bias

Overconfidence (OVR), a well-known heuristic-driven bias, occurs when individuals overestimate their knowledge, the accuracy of information and their ability to predict outcomes [31,32]. Research identifies several antecedents of overconfidence, including demographics [33], trading experience [34], financial literacy [30], and social factors, such as, information from advisors and peers [35]. Self-attribution also reinforces this bias [36]. Although overconfidence can lead to excessive trading and price volatility [37,38], AMH suggests that investors may eventually adapt their behavior in response to evolving market dynamics [39].

Adaptive market hypothesis and adaptability factors

The Adaptive Market Hypothesis (AMH) extends the Efficient Market Hypothesis, reconciling efficient market principles with behavioral biases. It suggests that market efficiency and inefficiency can coexist with intelligent, forward-looking, competitive, and imperfect investors continuously learning from their mistakes and adapting to changing economic conditions [7–9]. In a time-varying efficient market, where market efficiency fluctuates, behavioral anomalies persist but evolve over time [12]. Failure to adapt can result in financial losses and market exit, emphasizing that adaptability is crucial for survival in a competitive market. AMH suggests that retail investors’ behavior, shaped by their adaptation, is determined by both internal and external factors. This study introduces the concept of ‘adaptability factors,’ encompassing cognitive (internal) and social (external) factors, which represent the mechanisms through which AMH operate at the individual level. Internal adaptability factors include trading experience (EXP), self-reflection (RFL), desire for learning (DL), and financial literacy (FL), while external factors comprise social influence (SCL), consultation with financial advisors (ADV), and media exposure (MED). The adaptability factors proposed in this study are conceptually embedded within the Adaptive Market Hypothesis and bounded rationality framework. AMH emphasizes that investors adapt through learning, feedback, and interaction within competitive and uncertain environments. Internal factors such as trading experience, self-reflection, desire for learning, and financial literacy represent cognitive mechanisms through which investors process feedback and adjust behavior over time. External factors including social influence, consultation with financial advisors, and media exposure reflect environmental information channels that shape adaptive responses under conditions of uncertainty and information asymmetry. Thus, the adaptability factors operationalize AMH at the individual investor level without departing from its evolutionary learning perspective.

Although adaptability is often assumed to support rational adjustments, evidence shows that the same factors may heighten susceptibility to biases. Trading experience and a strong learning desire may fuel overconfidence [40–42], while social influence amplifies HRD [29,43]. Thus, adaptability interacts with cognitive and environmental constraints in line with Bounded Rationality [11]. By modeling adaptability factors as antecedents of HRD and OVR, this study extends AMH beyond its general conceptual framing and provides an empirically testable pathway linking adaptation to specific biases in frontier markets.

External adaptability factors and behavioral biases

Theoretical framework

This study proposes a Market Adaptability Model (Fig 1) to examine how internal and external adaptability factors influence investors’ behavioral biases, particularly HRD and OVR. Behavioral biases, HRD and OVR, serve as dependent variables, while internal (EXP, RFL, DL, FL) and external (SCL, ADV, MED) adaptability factors act as independent variables. Grounded in the Adaptive Market Hypothesis (AMH) and Bounded Rationality, the framework suggests that adaptability processes help investors survive in dynamic markets. However, these processes can also influence psychological biases when shaped by cognitive and social factors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Market adaptability model of individual investors.

https://doi.org/10.1371/journal.pone.0345883.g001

Research design, data collection, and sampling

This study employed an empirical, quantitative research design based on behavioral finance theories. The target population consisted of individual investors trading on the frontier market of the Dhaka Stock Exchange (DSE). The survey was administered using a dual-mode approach, both online and through face-to-face distribution of hardcopy questionnaires. No qualitative interviews were conducted; the term “face-to-face” refers exclusively to the in-person administration of the structured questionnaire. The in-person surveys were conducted during trading hours in collaboration with 11 brokerage houses in Dhaka, where researchers visited brokerage house trading floors and requested eligible investors to complete the questionnaire on-site. The online survey was distributed through Google Forms to investors identified via brokerage houses, who were contacted and invited to participate after receiving study information. Participation was voluntary, with no incentives provided. A screening question confirmed eligibility by verifying a minimum of two years of trading experience. All data analyzed in this research were obtained solely from structured survey responses.

Respondents were selected using non-probability sampling, particularly purposive (judgmental) sampling to ensure representation of active investors. A probability-based sampling frame was not available, as no comprehensive contact list of active investors exists, and probability-based approaches are often impractical in social science research. Access to participants further required collaboration with brokerage houses. Purposive sampling ensured that all respondents met strict eligibility criteria: being at least 18 years old and having two or more years of trading experience, consistent with prior behavioral finance studies [29,69]. To mitigate potential bias, the gender distribution of the sample was aligned with Central Depository Bangladesh Limited (CDBL) records, reflecting a male-to-female ratio of approximately 75:25.

Sample size estimation followed established approaches. In particular, the “10-times rule” [70], based on the maximum number of arrows pointing to a construct in the model (nine), yielded a minimum sample size of 90. Using G*Power [71], with desired statistical power (1–β = 0.80), significance level (α = 0.05), medium effect size (f² = 0.15), and number of predictors = 9, the minimum required sample size was 114. The final dataset of 640 valid responses exceeded both thresholds, strengthening the reliability of results. Data were collected cross-sectionally over a four-month period (15 November 2022–15 February 2023).

Scale development and validation of questionnaire

Adaptability factors were conceptualized as internal (trading experience, self-reflection, desire for learning, and financial literacy) and external (social influence, consultation with financial advisors, and media) constructs, consistent with the AMH. Unlike prior behavioral finance research that mostly examined isolated biases, this study considers adaptability factors as antecedents shaping herding (HRD) and overconfidence (OVR). In doing so, the study extends AMH beyond its qualitative framing and aligns it with the theory of Bounded Rationality. The structured questionnaire comprised items measuring seven independent variables, two dependent variables, and seven demographic variables. Independent variables were measured as follows: EXP by years of trading [35]; RFL using the 7-item reflective thinking scale [51]; DL using the 8-item Self-Directed Learning Readiness Scale [72]; FL using five items [53,73]; SCL with six items [29]; ADV with five items [26]; and MED with a four-item scale derived from various literature [27,29,39]. Dependent variables were HRD, measured with four items [33] and OVR assessed with six items [27]. Demographic variables comprised gender, marital status, age, education level, occupation, monthly income, and region. The marker variable, “attitude toward the color blue (BLU),” a four-item scale, was used to assess common method bias [74]. Most constructs used 5-point Likert scales, except FL (multiple-choice) and EXP (single-item).

Validity and reliability of the survey instrument were established through a two-stage pre-test. First, expert validation was conducted by five experts, and the Content Validity Index (CVI) confirmed both item- and scale-level validity. Next, the pilot survey, conducted with 100 participants, confirmed satisfactory reliability and validity. All constructs exceeded the 0.7 threshold for Cronbach’s alpha (α) and Composite Reliability (CR), and factor loadings were above 0.50 for all items, confirming internal consistency, reliability, and convergent validity [70,75]. Respondents from the pilot test were excluded from the final survey.

Ethical considerations

Ethical approval was obtained from the Universiti Malaya Research Ethics Committee (UMREC) (Ref: UM.TNC 2/UMREC, 20 January 2022). All participants provided written informed consent prior to participation. They were given detailed information regarding the study objectives, procedures, voluntary participation, and data confidentiality. Consent to publish their data was assumed upon completion of the questionnaire. As the study involved individual investors rather than a formal community or institution, consent from a community representative was not required. The study complies with all relevant ethical regulations, and there were no deviations from the approved study protocol after obtaining ethical clearance from UMREC.

Data analysis techniques

Data were entered, cleaned, and screened using IBM SPSS Statistics for Windows, version 26.0 [76], retaining all 640 responses. Descriptive statistics and checks of multiple regression assumptions indicated that the data were non-normally distributed. Consequently, Partial Least Squares Structural Equation Modeling (PLS-SEM) was employed in SmartPLS 4.0 [77] to assess relationships among constructs and evaluate the model’s predictive accuracy. PLS-SEM was appropriate given the non-normal data, model complexity, and sample size. The analysis followed a four-step process: (i) testing for normality, (ii) assessing common method variance (CMV), (iii) validating the measurement model, and (iv) evaluating the structural model [70,78].

Structural Model Evaluation and Statistical Assumptions

Prior to estimating the structural model, key statistical assumptions were assessed to ensure the robustness and validity of the PLS-SEM results. Multivariate normality was tested using Mardia’s skewness and kurtosis, which confirmed non-normality and further justified the use of PLS-SEM. Multicollinearity was evaluated through variance inflation factors (VIF), indicating no collinearity concerns. Linearity and homoscedasticity were verified by residual scatterplots, while Durbin-Watson statistics confirmed the independence of errors. Sample size adequacy was ensured with the final sample of 640 far exceeding minimum requirements. Structural model evaluation was then conducted using non-parametric bootstrapping with 10,000 resamples to generate stable standard errors and 95% bias-corrected confidence intervals. For model evaluation, both explanatory power and predictive adequacy were assessed, where R² values indicated moderate explanatory power, and Q² statistics confirmed moderate predictive relevance of the endogenous constructs. Model fit was evaluated using standardized root mean square residual (SRMR < 0.08), supplemented by additional PLS-specific diagnostics (d_ULS and d_G) used for comparative purposes. These diagnostics supported the robustness, reliability, and theoretical adequacy of the model despite reliance on non-probability sampling and non-normal data distributions. Extended robustness checks are reported in the Results section and in Supporting Information (S3 Text, S4 Table).

Sample characteristics

Socio-demographic characteristics of the individual investors, presented as a frequency distribution, are summarized in Table 1. Among the respondents, 75% were male, reflecting the male-to-female investor ratio (75:25) in the DSE. Most respondents were aged between 25–44 years, and over half held a master’s degree. The majority were employed in the private sector and resided in the Dhaka division, including the capital.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Socio-demographic profile of the investors.

https://doi.org/10.1371/journal.pone.0345883.t001

Normality testing

A normality test was conducted using Mardia’s multivariate skewness and kurtosis test via Webpower software. The data were not multivariate normal, as both skewness and kurtosis exceeded recommended cut-off values [79]. This justified the use of PLS-SEM, which is robust to non-normal data distributions.

Common method variance

This study was susceptible to common method variance, given the reliance on a single source of respondents (individual investors) and a single data collection method (questionnaire survey). To address potential CMV, both procedural and statistical remedies were applied. Procedurally, item wording was varied and respondent anonymity was assured to minimize evaluation apprehension. Statistically, “attitude towards color blue” (BLU) was used as a marker variable [74]. The Confirmatory Factor Analysis (CFA) Marker Technique showed no significant changes in R² and β values between the method factor model and the baseline models, suggesting the absence of CMV. Additionally, Harman’s Single-Factor Test indicated that percentage variance explained by a single factor was less than 50%, further confirming that CMV was not a concern.

Measurement model assessment

Convergent validity, which assesses the validity of individual items, was evaluated using item loadings, Cronbach’s Alpha, Composite Reliability, and Average Variance Extracted (AVE). Items with loadings below 0.50 were removed prior to final estimation. The measurement model results for convergent and discriminant validity are presented in Table 2. All retained item loadings exceeded 0.50, with most above the preferred threshold of 0.708, except for HRD2. Cronbach’s Alpha values were above 0.70 for all constructs (except HRD), while CR values consistently exceeded 0.70. AVE scores for all constructs were greater than 0.50, thus, confirming internal consistency, reliability, and convergent validity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Measurement model results for convergent validity and discriminant validity.

https://doi.org/10.1371/journal.pone.0345883.t002

Discriminant validity, reflecting the degree to which items are divergent among constructs, was assessed using the Heterotrait-Monotrait (HTMT) ratio [80]. As reported in Table 2, all HTMT values were below the 0.85 threshold, which indicates strong discriminant validity and confirms that respondents clearly distinguished among the constructs. Fig 2 illustrates the final market adaptability model linking adaptability factors and behavioral biases.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Market adaptability model relating adaptability factors and behavioral biases.

https://doi.org/10.1371/journal.pone.0345883.g002

Collinearity Statistics, measured using the Variance Inflation Factor (VIF) in the inner model, indicate collinearity between independent variables and each endogenous variable. VIF values for the endogenous variables (HRD and OVR) were below 5.0, indicating no multicollinearity among the independent variables [79].

Structural model testing

The structural model assessed how internal and external adaptability factors influence HRD and OVR. To evaluate the hypothesized relationships, a 10,000-sample re-sample bootstrapping method was applied with a 0.05 significance level using one-tailed testing (critical value = 1.645) to address the non-normality of data [81]. Both p-values and 95% bias-corrected confidence intervals (CI) were reported to ensure robust statistical decisions. The hypotheses test results, representing the direct effects of adaptability factors on herding and overconfidence, are presented in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Hypothesis testing of adaptability factors predicting herding and overconfidence biases.

https://doi.org/10.1371/journal.pone.0345883.t003

The results reveal significant relationships between adaptability factors and behavioral biases in stock market investments. All external adaptability factors, SCL, ADV, and MED, positively influenced herding (β = 0.309, p < 0.001; β = 0.089, p < 0.05; β = 0.194, p < 0.001 respectively). Internal adaptability factors, EXP, RFL, and DL, had significant positive effects on overconfidence (β = 0.261, p < 0.001; β = 0.113, p = 0.024; and β = 0.258, p < 0.001 respectively). By contrast, FL had a significant negative effect on both herding (β = −0.141, p < 0.001) and overconfidence (β = −0.067, p < 0.05). All supported hypotheses had confidence intervals excluding zero at the 5% significance level, confirming their statistical significance. The output patterns suggest that adaptability factors operate through distinct pathways: social influences primarily promote imitation (herding), whereas cognitive processes enhance self-assessment while also increasing overconfidence.

Predictive relevance

The predictive accuracy of the model was assessed through the coefficient of determination score (R²). Both herding and overconfidence demonstrated moderate accuracy with R² values of 0.225 and 0.189 respectively [70], indicating that the adaptability factors explain meaningful variation in these biases.

Robustness and model diagnostics

To ensure the robustness of the structural model, multiple diagnostic checks were conducted within the PLS-SEM framework (Table 4). Non-parametric bootstrapping with 10,000 subsamples confirmed the stability of all significant paths, with 95% bias-corrected confidence intervals excluding zero, thereby ensuring the reliability of parameter estimates. Model fit was acceptable (SRMR = 0.072 for the saturated model; 0.073 for the estimated model), with moderate explanatory power (R² = 0.225 for herding; 0.189 for overconfidence) and predictive relevance (Q² > 0 for both endogenous constructs). Collinearity checks confirmed that all VIF values were below 5, ruling out multicollinearity. Linearity and homoscedasticity were supported by residual scatterplots, and Durbin-Watson statistics indicated independence of errors. Additional PLS-SEM fit diagnostics, including d_ULS (2.946; 3.007) and d_G (0.606; 0.608), further supported the adequacy of the model. Extended diagnostics are provided in S3 Text and S4 Table. Collectively, these diagnostics indicate that the model was robust, statistically reliable, and theoretically meaningful, even under purposive (non-probability) sampling and non-normal data distributions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. PLS-SEM model fit and predictive diagnostics.

https://doi.org/10.1371/journal.pone.0345883.t004

Theoretical Implications

This study extends behavioral finance by integrating internal and external adaptability factors into AMH. While AMH emphasizes investor learning and survival in changing environments [8], its application to specific behavioral biases has been limited [98,99]. Findings show that adaptability is not uniformly beneficial: external adaptability supports short-term survival through imitation, while internal adaptability promotes long-term learning but can also lead to overconfidence in some cases, except for financial literacy, which mitigates both biases. This duality refines AMH’s learning–survival framework, signifying that adaptation does not necessarily yield rationality, particularly in frontier markets with limited institutional support [12].

This study importantly integrates the Adaptive Market Hypothesis and Bounded Rationality into a unified explanatory framework. While AMH explains how investors adjust their behavior through learning and environmental feedback, Bounded Rationality clarifies why such adaptive processes remain constrained by limited cognitive capacity and imperfect information. By empirically modeling herd and overconfidence biases as behavioral outcomes of internal and external adaptability factors, this study demonstrates that adaptive learning operates within bounded cognitive limits. Thus, adaptation and bounded rationality jointly shape investor behavior, particularly in frontier markets characterized by uncertainty and information asymmetry.

Building on this integrated framework, Table 5 compares adaptability–bias dynamics across frontier, emerging, and developed markets. In the frontier market of Bangladesh, weak oversight and volatility amplify external herding drivers, while internal learning processes increase overconfidence. Financial literacy reduces both herding and overconfidence, confirming its protective role. These results align with prior studies linking social and media influence to herding [43,100] and trading experience or reflection to overconfidence [41,93]. The mitigating effect of literacy echoes findings from Pakistan [101], and emerging markets such as India and Saudi Arabia [13,15], where it reduces overconfidence. In developed markets, greater institutional transparency reduces herding [102,103], while overconfidence is situational, often surfacing during crises [104]. Overall, these findings address gaps highlighted in the literature (Section 2), particularly the underexplored role of adaptability factors in bias formation under AMH, the dual nature of adaptation, and the contextual distinctiveness of frontier markets.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Comparative dynamics of adaptability factors and behavioral biases across frontier, emerging, and developed markets.

https://doi.org/10.1371/journal.pone.0345883.t005

Overall, these implications show that AMH remains a valuable lens across market contexts but must be operationalized with attention to institutional quality and investor literacy. Modeling adaptability factors as distinct cognitive and social mechanisms clarifies why the same learning processes that support survival can also reinforce behavioral biases in volatile, low-transparency environments. This contextualized extension of AMH highlights the need for theoretical and empirical work that explicitly examines how institutional and informational conditions shape the overall impact of adaptability on rationality. Collectively, the study addresses gaps highlighted in the literature (Section 2), particularly the underexplored role of adaptability factors in bias formation under AMH, the dual nature of adaptation, and the contextual distinctiveness of frontier markets.

Managerial implications

The findings provide several actionable insights for regulators, market institutions, brokerage firms, and listed companies in Bangladesh’s frontier stock market. First, the Bangladesh Securities and Exchange Commission (BSEC) should strengthen market oversight, transparency, and disclosure practices to reduce reliance on informal cues and rumor-based trading. In collaboration with the Bangladesh Institute of Capital Market (BICM), financial literacy initiatives should explicitly address behavioral biases and incorporate lessons from past market crises (e.g., 1996 and 2010 crashes) to highlight common pitfalls. Second, brokerage firms can establish localized Investor Literacy Corners and organize scenario-based workshops or simulated trading sessions. Regular “Bias Literacy Programs,” offered in Bangla, would enable investors to recognize their own cognitive biases in realistic decision-making contexts. Third, gender-focused initiatives are essential. Tailored literacy and digital programs, delivered through grassroots institutions and designed for female investors, can address psychological barriers, improve confidence, and expand women’s participation in capital markets. Fourth, technology-driven nudges via mobile applications can incorporate behavioral alerts, such as warnings about herd-driven price swings or excessive trading, to help investors avoid impulsive decisions. Finally, listed companies should ensure timely and accurate dissemination of financial information to limit speculation and herd-driven volatility, thereby promoting more stable market behavior.

Taken together, these measures address the dual role of adaptability identified in this study. While external adaptability encourages herding and internal adaptability (excluding financial literacy) fuels overconfidence, targeted regulatory, educational, and technological interventions can alleviate these biases and build a more resilient, inclusive, and investor-friendly frontier market.

Limitations

This study has several limitations that should be considered when interpreting the findings. First, obtaining authentic responses from Bangladeshi female investors posed challenges due to cultural and market practices, although careful screening was applied to ensure genuine participants. Second, reliance on self-reported perceptual data may have been influenced by respondent mood or context, introducing potential subjectivity, despite diagnostic checks confirming the nonexistence of major bias. Third, the study focused only on individual investors, excluding institutional participants whose adaptability and behavioral patterns may differ. Fourth, the study employed purposive rather than probability sampling, due to the absence of a comprehensive sampling frame of active retail traders. While this limits generalizability, potential bias was mitigated by aligning the sample’s gender distribution with CDBL statistics and by conducting extensive PLS-SEM diagnostics-including bootstrap resampling, multicollinearity checks, error independence tests, and model fit indices)-which confirmed the stability and validity of the structural relationships. Finally, since the study is cross-sectional, causal relationships between adaptability factors and behavioral biases cannot be definitively established.

Further research

The study offers several recommendations for future research. First, large-scale research with more diverse and representative respondents would enhance validity and generalizability of the findings. Second, future studies should explore the relative strength of individual adaptability factors on herding and overconfidence biases, identifying which elements exert the most significant influence. Third, studies incorporating moderating and mediating variables would provide deeper insights into how market adaptability influences heuristics. Fourth, comparative studies across frontier and emerging markets would clarify the role of cultural and institutional contexts in shaping investor behavior. Finally, longitudinal or experimental designs are recommended to establish stronger causal inferences between adaptability factors, behavioral biases, and irrational investment decisions.