PI Global Investments
Finance

The plasticity of overconfidence: the impact of feedback intervention on confidence calibration in investment decisions


Abstract

This study examines the impact of different feedback interventions on the confidence calibration of investment decisions, tests the mediating role of overconfidence plasticity, and the moderating effect of cognitive reflection ability, providing experimental evidence for investors’ psychological bias intervention and investment education. Adopting a three-level between-participants behavioral experimental design, 220 participants with investment exposure experience were recruited, and a final effective sample of 187 people was randomly divided into a no-feedback control group, a result-feedback-only group, and a metacognitive feedback intervention group. The results showed that feedback intervention significantly improved the plasticity of overconfidence, and the plasticity of the metacognitive feedback intervention group was significantly higher than that of the outcome feedback group and the no-feedback control group; Feedback intervention can improve the confidence calibration of investment decisions. The metacognitive feedback group has a significantly lower expected calibration error ECE and the best calibration quality; Overconfidence plasticity plays a partial mediating role between feedback intervention and confidence calibration, with feedback intervention reducing ECE by activating overconfidence plasticity; Cognitive reflective ability has a moderating effect on this mediating pathway, with significant indirect effects only observed in individuals with moderate to high cognitive reflective ability. The mediating pathway intervened by individuals with low cognitive reflective ability is ineffective. The robustness test supports that the research results are stable and reliable. Overconfidence in investment decisions has cognitive plasticity; Compared to simple outcome feedback, metacognitive feedback is a more effective intervention for bias reduction; The intervention effect has individual boundaries, and cognitive reflection ability will constrain the effectiveness of feedback intervention.

1 Introduction

Traditional finance is based on the assumption that investors are completely rational, believing that individuals can fully utilize all available information to make unbiased estimates of asset returns and risks, and achieve maximum investment utility. However, a large amount of evidence from behavioral finance and cognitive decision-making research shows that real investors generally have various cognitive biases, among which overconfidence is one of the most prominent psychological factors affecting investment behavior. Overconfidence can lead investors to overestimate their own judgment ability and underestimate market risks, resulting in behaviors such as excessive trading, risk misjudgment, and irrational portfolio allocation, ultimately damaging actual investment returns ().

In the past, a large number of studies have regarded overconfidence as a relatively stable personality or cognitive trait, focusing on the negative impact of overconfidence on investment outcomes, and exploring less from the perspective of cognitive plasticity whether external intervention can change individual overconfidence levels and improve the quality of confidence calibration in decision-making. Confidence calibration reflects the degree of matching between subjective confidence and objective decision accuracy at the metacognitive level. Poor confidence calibration in investment scenarios directly manifests as investors being overly optimistic about their own judgments, which is the phenomenon of overconfidence in the investment field ().

Feedback is an important external condition for shaping metacognitive judgments and correcting subjective beliefs. Feedback information from task outcomes can prompt individuals to update their assessments of their own abilities and adjust their confidence levels in subsequent decision-making. However, existing literature still lacks sufficient experimental evidence on which feedback interventions can effectively improve confidence calibration in investment decisions, whether overconfidence has plasticity that can be reshaped by feedback, and the underlying cognitive processing mechanisms (). Most related studies are mainly questionnaire related and lack a strictly controlled laboratory paradigm, making it difficult to establish causal inference between variables.

Based on this, this study adopts a behavioral experimental method to manipulate different types of feedback interventions, examine the effect of feedback interventions on confidence calibration in investment decision-making contexts, and test the mediating role of overconfidence plasticity in this process, as well as the moderating effect of individual cognitive characteristics. This study can not only enrich the empirical research on metacognition and overconfidence in the investment field, but also provide experimental evidence for investor psychological intervention and investment education. Feedback is a key external factor reshaping investors’ metacognition. Yet few behavioral experiments have explored whether metacognitive feedback can activate overconfidence plasticity and boost confidence-accuracy matching in investment decisions, leaving gaps in investor-debiasing research ().

This study makes three incremental contributions. First, it re-conceptualizes overconfidence plasticity as a mediating intervention-response mechanism rather than as an outcome variable, thereby extending prior work that has primarily tested the main effect of feedback on confidence calibration. Second, it identifies cognitive reflection as a boundary condition on the “feedback – plasticity – calibration” chain, showing that feedback is not equally effective for all individuals. Third, it compares outcome-only feedback with metacognitive feedback in a laboratory investment task, providing evidence on the effect of feedback content rather than merely the presence or absence of feedback. Together, these contributions clarify not only whether feedback improves calibration, but also through which mechanism and under what individual-level conditions it does so.

2 Theoretical framework

2.1 Concept definition

2.1.1 Overconfidence in investment decisions

In the decision-making field, it can be divided into three subtypes: overestimation, overplacement, and over precision. In the context of investment decision-making, overconfidence specifically refers to the psychological tendency of investors to systematically overestimate the accuracy of their investment judgments and underestimate market uncertainty and investment risks during asset analysis and trading selection processes.

The classic study by confirms that overconfident individual investors have significantly higher trading frequencies, higher turnover rates result in transaction costs, and ultimately lower net investment returns. proposed the formation mechanism of investor overconfidence, where investors may exhibit self attribution bias, attributing investment success to their own abilities and investment failure to luck, continuously strengthening their level of overconfidence. It is necessary to distinguish between optimism and overconfidence: optimism tends to have a positive expectation of external market returns; Overconfidence in investment focuses on individuals’ overestimation of their own judgment accuracy, which belongs to the metacognitive bias. This study focuses on the latter.

2.1.2 Decision confidence calibration

Confidence calibration is a core indicator of metacognitive judgment, which describes the matching relationship between an individual’s subjective reported decision confidence and objective decision accuracy. In an ideal calibration state, when an individual reports 75% confidence, the actual accuracy of the decision should also be 75%; If subjective confidence is significantly higher than the actual accuracy, it is a poor calibration, manifested as overconfidence; If subjective confidence is significantly lower than the accuracy rate, it is manifested as underconfidence.

Common quantitative indicators include expected calibration error (ECE), calibration curve, metacognitive bias parameter, etc. In decision experiments, after completing each investment choice, participants need to give a confidence score of 0–100% for the probability of making the correct choice. Combined with the objective results of the trial, the confidence calibration level is calculated. In the context of investment decision-making, the quality of confidence calibration directly reflects whether investors can objectively evaluate the reliability of their investment judgments; The larger the calibration deviation, the more likely investors are to engage in high-risk irrational trading behavior. Confidence calibration is different from simply having high or low confidence. High confidence itself is not a problem, but the mismatch between confidence and objective accuracy is the core issue.

2.1.3 Types and dimensions of feedback interventions

Feedback intervention refers to providing external information such as task results and judgment quality to decision makers during the task process, in order to change individual cognition, judgment, and behavior. It is an important intervention paradigm in bias reduction research. According to the existing classification in the field of decision bias intervention, feedback intervention can be divided into multiple dimensions: from the perspective of feedback content, it can be divided into result feedback (only informing decision correctness, profit and loss) and cognitive processing feedback (informing the gap between subjective confidence and actual accuracy, metacognitive level feedback); From the perspective of feedback timeliness, it can be divided into instant feedback and delayed feedback; The feedback efficacy can be divided into positive feedback, negative feedback, and mixed feedback; Feedback can be divided into overall summary feedback and trial feedback.

Simple outcome feedback only provides external results of decision-making, while metacognitive feedback directly points to individual confidence judgments themselves, which is more likely to improve confidence calibration. found that external feedback can alter individuals’ global level self confidence beliefs, and feedback valence can have asymmetric effects on subsequent metacognitive judgments. Not all feedback interventions can have a bias reducing effect, and under some feedback conditions, self attribution bias can further amplify overconfidence. Therefore, it is necessary to conduct experiments to test the actual effects of different feedback types.

2.1.4 Plasticity of overconfidence

For a long time, there have been two views in academia: some studies believe that overconfidence is a stable individual trait with high consistency across contexts; Another part of the research found that overconfidence can change with task experience, external information, and experimental intervention, indicating that overconfidence has plasticity. In this study, the plasticity of overconfidence is defined as the attribute of an individual’s overconfidence level in investment decisions that changes under the influence of external environment (feedback intervention in this study), manifested as an increase or decrease in overconfidence level, corresponding to the expansion or contraction of confidence bias.

Plasticity does not deny the existence of individual differences in overconfidence, but emphasizes that this cognitive bias is not completely fixed and can be partially corrected through external information input. The plasticity of overconfidence is an inherent prerequisite for feedback intervention to improve confidence calibration: only when overconfidence can be adjusted by feedback information, can external intervention improve the quality of metacognitive calibration.

2.2 Relevant theoretical basis

2.2.1 Cognitive bias theory

The Heuristic and Cognitive Bias Theory was proposed by , which suggests that there are two processing systems for human decision-making: System 1 is fast, automated, and relies on heuristic intuition; System 2: Slow, controlled, and rational analysis; In most cases, individuals rely on System 1 to make judgments, resulting in a series of systemic cognitive biases, with overconfidence being a typical representative. Cognitive bias is not an individual intellectual disability, but a common cognitive processing tendency in humans. The cognitive bias theory provides the basis for this study: overconfidence in investment decisions is a systemic bias caused by heuristic processing. Since it originates from cognitive processing, there is a possibility of external intervention and regulation, rather than a completely unchangeable personality trait.

2.2.2 Decision confidence calibration theory

The confidence calibration theory is rooted in the metacognitive monitoring theory. Metacognitive monitoring is the process by which individuals evaluate their own cognitive performance, and calibration is the result of monitoring the processing output. were the first to conduct calibration research on human probability judgment, pointing out that most ordinary people generally exhibit overconfidence calibration bias in knowledge-based tasks. found that although humans and large language models sometimes exhibit considerable consistency in metacognitive ability and behavior, there are clearly still many differences; Recognizing these differences is of great significance for strengthening collaboration between humans and artificial intelligence. Finally, we discussed how to endow future large-scale language models with more sensitive and calibrated metacognition, which may also help them develop new abilities such as more effective learning, self guidance, and curiosity.

2.2.3 Feedback learning cognitive theory

The cognitive theory of feedback learning holds that individuals receive external feedback during tasks, compare the feedback information with their original internal beliefs, update internal representations based on prediction errors, and adjust subsequent behaviors and subjective judgments. The Rescorla Wagner reinforcement learning model is a representative computational model in this field, where the magnitude of prediction error determines the extent of belief updates. Applied to the investment decision-making scenario of this study: Participants make investment judgments and give confidence scores. After receiving feedback, the prediction error between subjective confidence and actual results is calculated, and the belief in self judgment ability is updated based on the error, thereby changing the confidence level for the next round of decision-making. However, humans have a processing bias toward feedback, such as self attribution, and have a lower acceptance of negative feedback, which can weaken the effectiveness of feedback learning. examined how urgency related cognitive biases affect household financial choices and emphasized the limitations of financial literacy in mitigating behavioral biases. Although the study focuses on different cognitive biases, its findings can provide useful supplementary perspectives for discussing the role of psychological mechanisms and the effectiveness of de biased strategies in financial decision-making.

2.2.4 A cognitive plasticity perspective

Cognitive plasticity refers to the characteristic of individual cognitive representations and processing patterns being reconstructed by external experiences and environmental inputs. Cognitive plasticity is not limited to the developmental stage of children, and adult metacognition and judgment biases also retain plasticity space. The theory of dynamic cognitive plasticity emphasizes that cognitive traits are not static and fixed, but rather dynamic products shaped by the continuous interaction between individuals and their environment. Corresponding to this study, overconfidence, as a metacognitive bias, is not static. Under external environmental input conditions such as feedback, individuals’ self ability representations can be updated, and overconfidence levels undergo dynamic changes, reflecting plasticity. Integrating these perspectives, the present framework posits a three-link mechanism. Cognitive bias theory establishes that overconfidence arises from heuristic processing and is therefore potentially modifiable rather than a fixed trait. Feedback learning theory specifies the mechanism of modification: feedback supplies a prediction-error signal that prompts updating of self-ability beliefs. Cognitive plasticity theory specifies the boundary condition: the extent of updating depends on whether the individual engages reflective processing of the feedback. Cognitive reflection is therefore not a competing explanation but a condition under which the feedback-to-plasticity link operates. This integrated model directly motivates the moderated mediation specification tested below.

3 Research hypotheses and research models

3.1 Variable selection

This study is based on metacognitive feedback learning and cognitive plasticity theory, using the manipulation measurement paradigm of behavioral experiments to divide variables into independent variables, mediating variables, dependent variables, moderating variables, and control variables.

The independent variable is feedback intervention and belongs to the between-participants variable of experimental manipulation. This study set up three experimental conditions, namely no feedback control group, only result feedback group, and metacognitive feedback intervention group. Among them, metacognitive feedback not only presents the profit and loss results of investment tasks, but also presents the deviation information between subjective confidence and objective accuracy of the trial to participants, directly providing prompt intervention for the calibration deviation of confidence.

The mediating variable is overconfidence plasticity. It is operationalized as the observable change in an individual’s overconfidence bias from the pre-test stage to the formal intervention stage. The theoretical rationale draws on a Bayesian updating account of calibration: feedback provides diagnostic information about judgment accuracy, and if overconfident individuals are responsive to such information, their confidence–accuracy mapping should shift in a directional manner. The difference score therefore captures the magnitude of this revision rather than the level of overconfidence per se. It should be interpreted as a contextual index of intervention responsiveness, not as evidence that plasticity is a stable individual trait. Whether plasticity constitutes a stable individual difference requires longitudinal or repeated-measurement designs, and this issue is addressed in the Limitations.

While substantial evidence supports stable trait-level “core overconfidence” across time and task domains, pointing to enduring individual differences in overconfidence bias (), investor overconfidence is not a completely fixed personality trait. Existing research suggests that overconfidence is not a completely stable trait, and external situational interventions can change the degree of bias, but there are significant individual differences in plasticity between individuals (Han and Dunning, 2024).

Although both overconfidence plasticity and ECE are derived from confidence–accuracy responses in the investment task, they target theoretically distinct quantities. ECE measures the absolute level of calibration error, whereas plasticity measures the direction and magnitude of change in overconfidence bias.

The dependent variable is the confidence calibration of investment decisions. Using the expected calibration error (ECE) as the core quantitative indicator, the larger the ECE value, the higher the degree of mismatch between subjective confidence and objective accuracy, and the poorer the calibration quality. Simultaneously combining Brier scores and metacognitive bias parameters to characterize the confidence calibration level, focusing on the objective subjective matching degree of metacognitive judgments under investment decision-making tasks (). Confidence calibration differs from simple confidence levels in that its core lies in the matching relationship between subjective confidence and objective performance ().

The moderating variable is cognitive features, which are operationalized as CRT scores for cognitive reflection tests. Cognitive reflective ability reflects an individual’s ability to suppress intuitive heuristics and engage in analytical thinking. Previous studies have shown that this variable affects an individual’s ability to receive and process bias correction information, and can moderate the actual effects of various de bias interventions (; ). Individuals with high cognitive reflection scores are more capable of identifying their own cognitive biases and carefully processing external feedback information (Isler et al., 2020).

The control variables include participant age, gender, investment experience, financial literacy level, and objective task accuracy. The objective accuracy of tasks and subjective confidence are independent of each other, controlling the accuracy of decisions themselves to separate the quality of actual decisions from metacognitive calibration bias and avoid confounding intervention effects on task performance (). Previous research on investment decisions has shown that investment experience and financial literacy independently affect investors’ confidence judgments and need to be controlled at the statistical level ().

3.2 Research hypothesis derivation

3.2.1 Main effect hypothesis of feedback intervention and plasticity of overconfidence

The cognitive theory of feedback learning states that external feedback provides prediction error signals, prompting individuals to update their internal representations of their own abilities, thereby changing the level of bias at the metacognitive level (). Metacognitive feedback directly points to the mismatch between confidence and accuracy, and compared to simple outcome feedback, can provide more targeted correction information, promoting individuals to adjust their overconfidence levels. Under the condition of no feedback, individuals lack correction clues, and overconfidence bias is difficult to spontaneously change. At the same time, the theory of dynamic cognitive plasticity proposes that adult metacognitive biases are not completely fixed and can change under appropriate external information input conditions. However, individuals have a self attribution bias in processing feedback, and negative outcome feedback may be rationalized by individuals, weakening the corrective effect (). Related experiments in the investment field have also found that providing only feedback on profit and loss results is difficult to effectively weaken overconfidence, and investors are prone to attribute failures to external market luck (). Based on the above logic, metacognitive feedback intervention can effectively activate the plasticity of overconfidence and reduce overconfidence bias. Based on this, a hypothesis is proposed:

H1: Feedback intervention has a significant main effect on the plasticity of overconfidence.

3.2.2 Hypothesis on the relationship between feedback intervention and confidence calibration of investment decisions

Confidence calibration reflects the matching relationship between subjective confidence and objective decision results, and belongs to the output results of metacognitive monitoring (). Previous studies on metacognitive training have found that feedback interventions that provide confidence bias information can significantly reduce the expected calibration error (ECE) and improve calibration quality. This effect is also supported by evidence in risk decision-making tasks (). The feedback on profit and loss only provides external results of the task, without directly indicating subjective judgment bias, so the improvement effect of relative reliability calibration is limited (). Some studies also suggest that simple result feedback may even strengthen self attribution, further amplifying overconfidence and failing to achieve calibration level improvement. This study speculates that different types of feedback can lead to differences in the quality of confidence calibration in investment decision-making contexts. Based on this, a hypothesis is proposed:

H2: Feedback intervention significantly improves the confidence calibration of investment decisions.

3.2.3 Hypothesis on the mediating effect of overconfidence plasticity

From the perspective of cognitive processing logic, feedback intervention does not directly change confidence calibration. Instead, through feedback information input, it prompts individuals to update their beliefs in self judgment ability, change the degree of overconfidence, activate the plasticity of overconfidence, and ultimately improve confidence calibration. If an individual’s overconfidence lacks plasticity, even if they receive external feedback information, their metacognitive bias will not change, and feedback intervention cannot improve confidence calibration. Feedback input first acts on the modifiable attribute of overconfidence, and further affects metacognitive calibration performance, forming an indirect pathway. Similar mechanisms can also be observed in intervention studies related to investment decision-making, where external intervention requires changing individuals’ evaluations of their own abilities in order to bring about changes in confidence calibration (). It is speculated that overconfidence plasticity plays a mediating role between feedback intervention and confidence calibration. Based on this, a hypothesis is proposed:

H3: The plasticity of overconfidence mediates the relationship between feedback intervention and confidence calibration in investment decisions.

3.2.4 Hypothesis on the moderation effect of cognitive feature variables

Cognitive reflection determines whether an individual can suppress intuitive heuristic processing and engage analytic evaluation of externally supplied corrective information (). Meta-analytic evidence indicates that debiasing interventions do produce reliable but modest gains: across 41 studies contributing 160 effect sizes, educational debiasing training reduced the likelihood of committing biases relative to control conditions with a small pooled effect (g = 0.26, 95% CI [0.14, 0.39], P < 0.001), and the authors note that some biases proved resistant to change and that the transferability of training beyond the instructional setting remains uncertain (). The modest average magnitude of such effects suggests that intervention efficacy is unlikely to be uniform across individuals. Isler et al. (2020) showed that manipulations which explicitly recruit reflective processing alter debiasing outcomes, implying that an individual’s propensity to engage analytic evaluation is itself a boundary condition on how far corrective information is absorbed. Accordingly, the indirect path from feedback intervention to confidence calibration via overconfidence plasticity should be stronger among participants higher in cognitive reflection, forming a moderated mediation model (Hayes, 2022). Based on this, a hypothesis is proposed:

H4: Cognitive reflective ability has a moderating effect on the indirect path of feedback intervention → overconfidence plasticity → investment decision confidence calibration.

The model diagram is shown in Figure 1.

4 Research design

4.1 Participant selection and grouping design

This study used laboratory behavioral experiments for data collection, and pre-estimated the sample size using G*Power software. For the one-way ANCOVA (3-group between-participants design, 5 covariates), the input parameters were set as Cohen’s f = 0.25 (moderate-to-large effect size), statistical power = 0.80, and α = 0.05. The calculated required effective sample size was N = 164. Anticipating potential attrition and invalid responses, the target sample was further inflated to 186 effective participants. Considering that there were situations such as participants withdrawing midway, invalid responses, and careless operations during the experimental process, 220 participants were actually recruited.

The inclusion criteria for participants are age between 18 and 45 years old, possessing basic financial knowledge, and having experience in any type of investment product such as stocks and funds; The exclusion criteria are individuals with a history of severe mental illness, inability to understand experimental task instructions, and inability to complete computer key operations. Participants are recruited through a combination of online and offline methods, and corresponding rewards are given to them after completing all experiments.

This study is a between-participants experimental design, in which recruited participants were randomly assigned to three experimental groups: a no-feedback control group, a result-feedback-only group, and a metacognitive feedback intervention group. Assignment was not pre-screened on demographic characteristics. After eligibility was confirmed, E-Prime assigned participants to one of the three conditions using a simple random sequence generated with a fixed seed (20241026). Allocation was concealed until the condition was loaded. Group balance was checked after data collection. Exclusions followed pre-specified rules: failed manipulation check; judgment or confidence RT < 200 ms on more than 5 trials; identical rise/fall judgments on ≥ 8 consecutive trials; withdrawal or fewer than 100 completed trials. Table 1 reports exclusions by rule and condition.

Exclusion criterion No feedback Outcome feedback Metacognitive feedback Total
Failed manipulation check 4 4 4 12
Response time < 200 ms 3 3 3 9
Completely consistent answers ≥ 8 consecutive trials 2 3 2 7
Withdrew or incomplete trials 1 2 2 5
Final analytic sample 61 63 63 187

Group distribution of sample exclusions.

4.2 Experimental materials and task design

Tasks were programmed in E-Prime 3.0 and displayed on a 1920 × 1080, 60-Hz monitor. Each trial consisted of: fixation (500 ms) → price chart until response (max 10,000 ms) → rise/fall judgment (self-paced) → confidence rating 0–100% (self-paced) → condition-specific feedback → inter-trial interval (800 ms). The pre-test comprised 40 trials and the formal phase 60 trials. Trial order was block-randomized in blocks of 10; all three conditions used the same trial sequence, and only the feedback module differed. Trial-level data included participant ID, condition, trial number, price-path ID, judgment, accuracy, confidence, reaction time, and feedback type.

The investment decision-making task is adapted from the classic financial decision-making paradigm, simulating simple asset analysis and selection scenarios. For each trial, a recent price trend chart of a virtual target will be presented to the participants, which includes historical price fluctuation information of 20 time points. The participants need to combine the chart information to make a binary judgment on the subsequent short-term price trend of the target, that is, to judge whether the price is rising or falling. After completing the up and down selection, participants need to rate their confidence level in making their own judgments this time. The entire task is divided into a pre-test phase and a formal intervention phase. The pre-test consists of 40 trials and is used to collect initial overconfidence and confidence calibration baseline levels of participants before intervention; A total of 60 trials were conducted during the formal intervention phase to perform feedback manipulation for different groups.

Virtual asset prices were generated by a geometric Brownian motion with drift and volatility parameters fixed before the experiment. For each trial, the correct direction was defined as the actual direction of price movement after a pre-specified holding period. This correct direction was uniquely determined by the generating parameters and did not depend on participants’ judgments or experimenter discretion. All participants faced exactly the same trial-by-trial price-path sequence; experimental groups differed only in the presence and type of post-trial feedback. Block randomization was used only to balance the order of trials, and the block composition was equivalent across groups. The distribution of price paths did not differ systematically across groups. Chart size, color, and axis format were held constant to eliminate visual confounds. Set comprehension check attempts within the task to verify whether participants understand the task rules. If the answers to the check attempts are incorrect, the program will pop up instructions again for prompt.

4.3 Feedback intervention manipulation plan

The task interface and stimulus materials of the three groups of participants were completely identical, with only differences in the feedback interface after each decision was made.

In the control group without feedback, after completing the selection of rise and fall and confidence rating, the interface will directly jump to the next trial without presenting any information related to the decision result or confidence level, and only a brief blank interval will appear.

For the result-feedback-only group, after the participants complete the decision-making and confidence scoring, the interface presents the objective results of this investment judgment, with text prompts indicating whether the judgment is correct or incorrect. At the same time, it simply displays the actual price change direction of the target without any subjective confidence related content. The feedback display time is fixed at 2000 milliseconds, and then enters the next trial.

The metacognitive feedback intervention group added additional metacognitive bias prompts on the basis of presenting objective decision-making results. The interface will also display the confidence score given by participants this time, compare the objective correctness of this decision, and intuitively explain the deviation between this confidence judgment and actual performance.

Feedback text was standardized. Metacognitive feedback displayed: “Your confidence was X%; your judgment was correct/incorrect. Your confidence was higher/lower than your actual accuracy by Y percentage points.” High versus low confidence was classified as ≥ 70% versus < 70% for display. Four variants were used: correct/high confidence, correct/low confidence, incorrect/high confidence, and incorrect/low confidence. Outcome feedback displayed only “Correct” or “Incorrect” plus the actual price direction. The no-feedback condition displayed a blank screen. Durations were 2,000 ms for no-feedback and outcome feedback, and 3,500 ms for metacognitive feedback. All feedback text was neutral and objective, with no evaluative, encouraging, or critical wording, so as to prevent emotional factors from confounding the cognitive changes produced by the feedback intervention itself.

Outcome feedback was presented for 2,000 ms, whereas metacognitive feedback was presented for 3,500 ms. This difference introduces a potential confounding factor, because observed effects may partly reflect differences in information-processing time rather than feedback content alone. To examine this issue, feedback presentation duration was included as a covariate in the mediation model (Table 2). The current design cannot fully separate feedback content from processing time; therefore, causal claims that metacognitive feedback content itself is superior to outcome feedback are qualified accordingly, and this issue is listed as a priority for future research in the Limitations.

Model Feedback → plasticity (a path) Plasticity → ECE (b path) Indirect effect 95% Boot CI
Without duration covariate 0.120 (0.041) −0.375 (0.091) −0.045 [−0.068, −0.024]
Controlling feedback duration 0.104 (0.043) −0.375 (0.093) −0.039 [−0.061, −0.020]

Mediation effects after controlling for feedback duration.

4.4 Selection of variable measurement tools and indicators

The independent variable feedback intervention is a categorical variable, with 0 representing the no feedback control group, 1 representing the outcome feedback only group, and 2 representing the metacognitive feedback intervention group. It is automatically encoded by the experimental program according to the participant grouping.

The mediating variable of overconfidence plasticity is calculated based on the overconfidence bias scores from pre-test and formal intervention stages. Firstly, obtain the overconfidence bias values of each participant during the pre-test and intervention stages, which are equal to the average confidence reported by the participants minus the objective average accuracy of the task. The plasticity of overconfidence is operationalized as the difference between pre-test overconfidence deviation and intervention stage overconfidence deviation. The larger the difference, the greater the decrease in overconfidence after task intervention and the higher the level of plasticity.

The confidence calibration of dependent variable investment decisions involves selecting the expected calibration error (ECE) as the core indicator. Expected calibration error was computed using equal-width bins. Confidence judgments were divided into M = 10 intervals (0–0.1, 0.1–0.2, …, 0.9–1.0), and ECE was calculated as the weighted average of the absolute difference between accuracy and confidence in each bin, with bin sample sizes as weights.

Formally, ECE = Σ(n_b/N)|accuracy_b − confidence_b|. Bins with fewer than 5 trials were merged with the adjacent bin; sparse first or last bins were merged with their only neighbor. Sensitivity analyses used M = 5 and M = 15. Overconfidence bias was mean confidence minus mean accuracy; overconfidence plasticity was pre-test bias minus intervention-phase bias, with positive values indicating reduced overconfidence. Brier score was mean[(confidence/100 − accuracy)2]. The metacognitive bias parameter was estimated with the signal-detection model of .

The auxiliary indicators include Brier score and metacognitive bias parameter. Brier score comprehensively measures the overall accuracy of probability judgment, while metacognitive bias parameter reflects the direction and magnitude of overall confidence level deviation. Multiple indicators complement each other to jointly characterize the quality of confidence calibration. The higher the ECE and Brier scores, the worse the calibration effect.

Cognitive reflection was measured with the six-item Cognitive Reflection Test, which captures individual differences in suppressing intuitive incorrect responses and engaging reflective processing. Financial literacy was measured with five objective multiple-choice items covering compound interest, inflation, and risk diversification, adapted from Lusardi and Mitchell (2014). Prior psychometric evidence supports the predictive validity of these item formats for financial behavior. In the present sample, Cronbach’s alpha was 0.693 for the CRT and 0.661 for the financial literacy scale, within the acceptable range for social science research. Both instruments were administered in Chinese versions prepared through translation and back-translation procedures.

4.5 Experimental procedure and data collection steps

The entire experiment was completed on a computer, with a duration of approximately 35–45 min for a single participant. The overall process is divided into five steps.

The first step is for the participants to read the informed consent form and confirm their voluntary participation in the experiment. After completing the reading, they will enter the basic information filling stage, fill in age, gender, investment related basic information, and complete the financial literacy questionnaire.

The second step is to fully explain the rules of the investment decision-making task to the participants, present operational examples, complete the comprehension check, and confirm that the participants fully understand how to operate the task before starting the pre-test task. The pre-test consisted of 40 trials, during which all three groups of participants did not receive any feedback and baseline confidence and decision-making behavior data were collected. After the pre-test, participants completed the cognitive reflection test CRT.

Step three, enter the formal intervention phase, with a total of 60 trials. According to the grouping of participants, the program calls the corresponding feedback module and presents the feedback content corresponding to each group after each decision is made. At this stage, experimental manipulation is completed and decision confidence data is collected after intervention.

Step 4 included a manipulation check. The key item was: “After each judgment and confidence rating, what appeared on screen?” Options were: (a) no feedback; (b) only whether my judgment was correct or incorrect; (c) whether my judgment was correct or incorrect plus a comparison between my confidence and accuracy; (d) I do not remember. Participants also described the feedback in open text. Two coders independently coded responses (κ = 0.86); disagreements were resolved by discussion. Responses inconsistent with the assigned condition were classified as failed manipulation checks and excluded.

Step five, introduce the experimental research purpose to participants, provide experimental explanations and answer questions, distribute experimental rewards, and end the entire experiment.

All behavioral data is automatically recorded and exported by the E-Prime program. Questionnaire data and behavioral data are matched and integrated based on participant numbers to form a complete dataset, which is then imported into statistical software for preprocessing and hypothesis testing analysis.

5 Results analysis

All statistical analyses in this chapter were completed using SPSS 26.0 in conjunction with PROCESS 4.2 macro program. Of the 220 participants initially recruited, 187 were retained after applying pre-specified exclusion criteria. Thirty-three participants were excluded: 12 failed the manipulation check (no feedback = 4, outcome feedback = 4, metacognitive feedback = 4), 9 had response times below 200 ms on more than five judgment or confidence trials (3/3/3), 7 gave identical rise/fall judgments on eight or more consecutive trials (2/3/2), and 5 withdrew or completed fewer than 100 trials (1/2/2). The final analytic sample comprised 61, 63, and 63 participants in the no-feedback, outcome-feedback, and metacognitive-feedback groups, respectively (Table 1).

5.1 Data preprocessing and reliability and validity testing

Preprocess the raw data of behavioral tasks, extract the decision results and 0–100 confidence scores of each participant, calculate overconfidence bias, expected calibration error ECE, Brier score, and metacognitive bias parameters. The identification of outliers in continuous variables adopts the ± 3 standard deviation criterion, and a total of 11 extreme data points are identified. Winsorization is used to retain the corresponding sample of participants.

The reliability results of the scale: The Cronbach’s alpha coefficient of the cognitive reflection test CRT is 0.693, and the Cronbach’s alpha coefficient of the financial literacy questionnaire is 0.661, which is within the acceptable reliability range for social science research. The investment decision is based on the behavioral experimental paradigm, and the split half reliability is calculated. The 60 trials in the formal intervention stage are randomly divided into two halves, and the correlation coefficient of the ECE indicators obtained from the two sets of trials is r = 0.702, indicating that the task measurement has good internal consistency.

Manipulation check statistical results: 88.89% of participants in the metacognitive feedback intervention group were able to accurately recall the comparison between confidence and accuracy in the task; 87.30% of the participants in the feedback group only reported seeing the right or wrong decision results; 90.16% of the participants in the no-feedback control group reported no feedback on the task results. A small number of participants who answered incorrectly have been excluded in advance, indicating that the feedback manipulation is effective overall.

5.2 Descriptive statistics and correlation analysis of variables

The main continuous variables are overconfidence plasticity, expected calibration error ECE, Brier score, metacognitive bias parameter, cognitive reflective ability, and financial literacy. The statistical results of the three groups are shown in Table 3, and the Pearson correlation matrix of the variables is shown in Table 4.

Variable No-feedback control group (n = 61) Only result feedback group (n = 63) Metacognitive feedback intervention group (n = 63)
Overconfidence plasticity 0.028 ± 0.114 0.076 ± 0.131 0.164 ± 0.147
Expected calibration error ECE 0.176 ± 0.058 0.147 ± 0.052 0.093 ± 0.041
Brier score 0.307 ± 0.068 0.279 ± 0.063 0.226 ± 0.051
Metacognitive bias parameter 0.206 ± 0.079 0.168 ± 0.072 0.107 ± 0.058
Cognitive reflection ability 3.07 ± 1.34 3.02 ± 1.26 3.13 ± 1.31
Financial literacy 2.82 ± 1.03 2.76 ± 1.02 2.87 ± 1.06

Descriptive statistics of variables in each group (M ± SD).

Variable 1 2 3 4 5 6
1. Overconfidence plasticity 1 1 1 1 1 1
2. Expected calibration error ECE −0.51**
3. Brier score −0.45** 0.68**
4. Metacognitive bias parameter −0.49** 0.73** 0.61**
5. Cognitive reflection ability 0.29** −0.22** −0.19* −0.24**
6. Financial literacy 0.12 −0.10 −0.07 −0.09 0.21**

Pearson correlation matrix of main variables (N = 187).

As can be seen from the correlation matrix, there is a significant moderate negative correlation between overconfidence plasticity and ECE. The higher the individual’s overconfidence plasticity, the lower the expected calibration error, and the better the confidence calibration performance. Cognitive reflective ability is significantly positively correlated with overconfidence plasticity, and significantly negatively correlated with ECE. The correlation between financial literacy and the core variables of this study did not reach statistical significance. Financial literacy is nevertheless retained as a control variable in the subsequent analyses, together with age, gender, investment experience, and objective task accuracy.

For the main comparisons, adjusted group means, 95% confidence intervals, and effect sizes are reported in Table 5. Adjusted means control for age, gender, investment experience, financial literacy, and objective task accuracy. Cohen’s d is reported for comparisons with the no-feedback group.

Group Overconfidence plasticity M (SE) 95% CI ECE M (SE) 95% CI Cohen’s d vs. no feedback
No feedback 0.091 (0.018) [0.055, 0.127] 0.141 (0.017) [0.107, 0.175] –
Outcome feedback 0.158 (0.018) [0.122, 0.194] 0.118 (0.017) [0.084, 0.152] 0.34
Metacognitive feedback 0.237 (0.018) [0.201, 0.273] 0.083 (0.017) [0.049, 0.117] 0.79

Adjusted means for overconfidence plasticity and ECE across the three groups.

Means are adjusted for age, gender, investment experience, financial literacy, and objective task accuracy. d = Cohen’s d relative to the no-feedback group.

5.3 Testing the impact of feedback intervention on the plasticity of overconfidence

One-way ANCOVA was conducted with feedback intervention as the independent variable, overconfidence plasticity as the dependent variable, and age, gender, investment experience, financial literacy, and objective task accuracy as covariates. The main effect of feedback intervention is significant, . The results are shown in Table 6.

Source of variation Sum of squares Degrees of freedom Mean square F p
Covariate 0.203 5 0.041 2.72 0.021 0.071
Feedback intervention 0.615 2 0.308 20.84 < 0.001 0.189
Error 2.642 179 0.015
Total 3.460 186

One-way ANCOVA results of feedback intervention on the plasticity of overconfidence.

5.4 Impact test of feedback intervention on confidence calibration of investment decisions

Perform one-way covariance analysis with the expected calibration error ECE as the dependent variable and feedback intervention as the independent variable, including all control variables. The main effect of feedback intervention is significant, .

Bonferroni post hoc test: The ECE of the metacognitive feedback intervention group was significantly lower than that of the outcome feedback only group, p < 0.001; The metacognitive feedback intervention group was significantly lower than the no-feedback control group, p < 0.001; The ECE of the feedback group was lower than that of the control group without feedback. Metacognitive feedback intervention can reduce expected calibration errors, improve confidence calibration quality, and H2 is supported. The results are shown in Table 7.

Source of variation Sum of squares Degrees of freedom Mean square F p
Covariate 0.038 5 0.008 3.14 0.009 0.081
Feedback intervention 0.116 2 0.058 24.61 < 0.001 0.216
Error 0.421 179 0.002
Total 0.575 186

One-way ANCOVA results of feedback intervention on expected calibration error ECE.

5.5 Mediating effect test of overconfidence plasticity

The three-category feedback variable was dummy-coded with the no-feedback group as the reference. D1 represents outcome feedback versus no feedback, and D2 represents metacognitive feedback versus no feedback. Using PROCESS Model 4, with ECE as the dependent variable and overconfidence plasticity as the mediator, and controlling for demographic and task variables, bootstrap resampling (5,000) was used to test the relative indirect effect of D1, the relative indirect effect of D2, and the difference between the D2 and D1 indirect effects.

As Table 8 shows, the indirect effect was −0.025 (95% CI [−0.051, −0.004]) for D1 (outcome feedback vs. no feedback), −0.055 (95% CI [−0.087, −0.028]) for D2 (metacognitive feedback vs. no feedback), and the D2−D1 difference was −0.030 (95% CI [−0.057, −0.006]). The overall mediation test is reported in Table 9, feedback intervention positively predicted overconfidence plasticity; overconfidence plasticity negatively predicted ECE; the bootstrap 95% CI of the indirect effect excluded zero, indicating partial mediation and supporting H3.

Contrast a path (feedback → plasticity) b path (plasticity → ECE) Indirect effect Boot SE 95% Boot CI
Outcome feedback vs. no feedback (D1) 0.067 (0.021) −0.376 (0.092) −0.025 0.012 [−0.051, −0.004]
Metacognitive feedback vs. no feedback (D2) 0.146 (0.021) −0.376 (0.092) −0.055 0.015 [−0.087, −0.028]
Difference (D2 – D1) 0.079 (0.022) – −0.030 0.013 [−0.057, −0.006]

Contrast-specific mediation effects under dummy coding.

Bootstrap resamples = 5,000; bias-corrected intervals.

Effect type Effect size Standard error 95% confidence interval
Total effect −0.078 0.013 [−0.103, −0.053]
Direct effect −0.033 0.012 [−0.056, −0.009]
Indirect effect (plasticity of overconfidence) −0.045 0.011 [−0.068, −0.024]

Mediating effect test of overconfidence plasticity (Model 4, Bootstrap = 5000).

We note that these two metrics share the same behavioral measurement source. However, robustness checks (Table 10) show that the mediating effect remains significant both when baseline (pre-test) overconfidence bias is controlled and when plasticity is replaced by a residualized change score, which helps rule out the possibility that the effect merely stems from mathematical properties of difference-score computation.

Model Indirect effect (feedback → plasticity → ECE) Boot SE 95% Boot CI
Original change score −0.045 0.011 [−0.068, −0.024]
Controlling baseline overconfidence bias −0.038 0.010 [−0.059, −0.019]
Residualized plasticity −0.034 0.010 [−0.055, −0.015]

Robustness checks for overlap between plasticity and ECE.

Bootstrap resamples = 5,000; bias-corrected percentile intervals.

5.6 Testing the moderation effect of cognitive features

Construct a moderated mediation model using PROCESS Model 7, with independent variable feedback intervention (dummy variable), mediator variable overconfidence plasticity, dependent variable ECE, moderator variable cognitive reflective ability, and include all control variables, Bootstrap = 5000.

The index of moderated mediation was estimated to test whether cognitive reflection moderates the indirect path from feedback to ECE through plasticity. Conditional indirect effects are reported at low (−1 SD), mean, and high (+1 SD) levels of cognitive reflection. An index whose bootstrap confidence interval excludes zero is interpreted as support for moderated mediation. The results are shown in Table 11.

Cognitive reflection level Conditional indirect effect Boot SE 95% Boot CI
Low (−1 SD) −0.016 0.011 [−0.040, 0.005]
Mean −0.039 0.010 [−0.061, −0.021]
High (+1 SD) −0.062 0.014 [−0.092, −0.036]
Index of moderated mediation −0.023 0.009 [−0.042, −0.006]

Indirect effects under different levels of cognitive reflection.

Bootstrap resamples = 5,000. The index is the linear slope of the conditional indirect effect as a function of CRT. A confidence interval excluding zero indicates significant moderated mediation.

The interaction term between feedback intervention and cognitive reflection ability significantly predicts the plasticity of overconfidence. Simple slope analysis indicates that feedback intervention has a stronger effect on improving plasticity in high cognitive reflective participants, and the indirect effects are only significant at moderate and high levels of cognitive reflection, supported by H4.

5.7 Robustness test

This study sets up three robustness tests, namely replacing the dependent variable indicator, changing the extreme value processing method, and including pre-test overconfidence bias as an additional covariate, to determine whether the empirical results depend on specific processing strategies. The core model results of the three tests are summarized in Table 12.

Robustness test Sample size (N) Feedback intervention main effect F (p) Indirect effect 95% Boot CI Interaction term β (p)
Baseline model 187 20.84 (p < 0.001) −0.045 [−0.068, −0.024] 0.029
(p = 0.007)
Test 1
(Brier score as DV)
187 18.62 (p < 0.001) −0.038 [−0.060, −0.019] 0.024
(p = 0.013)
Test 2
(Extreme values excluded)
176 19.17 (p < 0.001) −0.042 [−0.065, −0.021] 0.027
(p = 0.009)
Test 3
(Pre-test bias as covariate)
187 17.33 (p < 0.001) −0.037 [−0.059, −0.018] 0.025
(p = 0.011)

Key results of robustness test core model.

Test 1: Replace the dependent variable with Brier score, while keeping all other settings unchanged;

Test 2: Without winsorization, directly remove ± 3SD extreme samples, and the remaining sample N = 176;

Test 3: On the basis of all the original control variables, add pre-test overconfidence bias as a covariate.

From the robustness test results, it can be seen that the main effect of feedback intervention is significant in all three tests, and the confidence intervals of the mediating effect do not cross 0. The interaction term between feedback intervention and cognitive reflection ability is statistically significant. The magnitude of the effect value fluctuates slightly compared to the baseline model, but there is no direction reversal or significant to insignificant change. The main research conclusions of covariance analysis, mediation, and moderation in this study do not rely on a single indicator selection or extreme value treatment scheme, nor are they affected by the initial overconfidence baseline level of participants. The empirical results have good stability.

6 Discussion and analysis

6.1 differentiated effects of different types of feedback interventions on investment confidence calibration

The limited calibration effect of mere outcome feedback partly stems from investors’ biased memory of investment histories. Selective memory distorts outcome learning: investors credit gains to personal skill and blame losses on market luck, so overconfidence rarely decays with trading experience (). Metacognitive feedback, by contrast, targets confidence-accuracy mismatch and achieves superior calibration performance. This study confirmed through three between-participants conditions that there are hierarchical differences in the improvement effect of feedback intervention on investment decision confidence calibration: metacognitive feedback intervention can significantly reduce the expected calibration error ECE and effectively alleviate metacognitive mismatch caused by overconfidence; Only result feedback brings only slight calibration improvement; Under the condition of no feedback, the overconfidence bias of participants remained at a high level, with data supporting H1 and H2. This result confirms the meta-analysis conclusion in the field of risk decision-making, and only providing feedback on the right and wrong results of profit and loss is difficult to fundamentally repair the mismatch between investors’ subjective confidence and objective accuracy.

From the perspective of cognitive processing mechanisms, result feedback only outputs external task results and does not touch on metacognitive monitoring itself. Due to the combined influence of self-serving attribution and selective memory bias, investors tend to attribute investment success to their own judgment ability and failure to market luck, making it difficult for pure profit and loss signals to update individuals’ inherent beliefs in the reliability of self judgment. Huffman et al. (2022) also pointed out in their field experiments that even if investors repeatedly receive market profit and loss feedback over a long period of time, the positive selective processing of memory will continue to reinforce overconfidence. This also explains the phenomenon that in reality, some investors’ overconfidence does not automatically decrease as their trading experience increases. Selective memory biases can distort how investors encode and retrieve past investment outcomes, further sustaining biased self-evaluations of decision-making competence ().

In contrast, metacognitive feedback presents both objective decision correctness and subjective confidence bias information, directly focusing individual attention on the “confidence accuracy” matching relationship, providing clear prediction error signals, thereby activating overconfidence plasticity and driving individuals to update their self ability representations. found in their uncertainty probability judgment experiment that compared to result feedback, metacognitive bias cues can significantly reduce expected calibration errors, proving that this intervention logic has cross task stability. This study further extends this paradigm to the task of virtual asset analysis in financial investment, confirming that metacognitive feedback also has de biased value in high uncertainty investment scenarios.

This study also observed a statistically significant slight improvement in the feedback group compared to the no-feedback control group. The feedback on the results is not completely ineffective, but the intervention effect is limited in magnitude. When accumulating enough contradictory decision results, a small number of participants can rely on the result sequence to perceive their own judgment limitations, but this learning path is inefficient and easily offset by self attribution. In a real investment environment, the market only provides profit and loss results, lacking clues to confidence bias. It is difficult to spontaneously complete metacognitive calibration and upgrading solely based on trading experience, which highlights the practical necessity of external metacognitive intervention.

6.2 Mediating significance of overconfidence plasticity: responding to the debate on the “Trait State” of overconfidence

The mediation analysis results indicate that overconfidence plasticity partially mediates the relationship between feedback intervention and investment confidence calibration. Feedback intervention not only has a direct effect on reliability calibration, but also can improve calibration quality by enhancing overconfidence plasticity, reducing ECE, and H3 has been validated. This result responds to the long-standing debate in academia: whether overconfidence is a stable personality trait or a cognitive state that can be reshaped by external environments. This cognitive bias should not be viewed simply as binary opposition. The experimental evidence in this study supports the theory of dynamic cognitive plasticity: overconfidence in the investment field is not an inherent trait that is completely fixed. Under appropriate metacognitive feedback input conditions, individuals can reduce overconfidence bias and achieve metacognitive calibration improvement.

The intermediary pathway reveals the complete cognitive chain: external feedback first changes individuals’ metacognitive beliefs about their own judgment ability, using overconfidence plasticity (the degree of decrease in overconfidence deviation before and after intervention) as an operational indicator, and then further improves confidence calibration. Assuming that overconfidence is a stable trait that cannot be changed, regardless of the input feedback, the mismatch between confidence and accuracy will not change, and intervention will not be effective. The effectiveness of cognitive bias intervention depends on the fact that the bias itself has room for correction, and the intervention does not directly change decision-making behavior, but rather reshapes the individual’s internal self-evaluation representation.

The model exhibits partial mediation, with feedback intervention retaining significant direct effects. The correlation analysis of this study shows that there is no significant correlation between financial literacy and core metacognitive variables. This result has practical implications: financial knowledge reserves do not equal good metacognitive calibration. Even investors who have mastered financial concepts may still experience severe overestimation of confidence. Investment psychology calibration training and financial knowledge education belong to two relatively independent educational modules.

6.3 The regulatory boundary of cognitive reflection ability and the enlightenment of intervention practice

The results of the moderated mediation model confirm that feedback intervention improves confidence calibration through the mediating pathway of overconfidence plasticity, which is moderated by cognitive reflective ability. Simple slope analysis shows that the conditional indirect effect is significant in participants with moderate to high levels of cognitive reflection; The indirect effect is not significant in the group with low cognitive reflection level, and H4 holds true. This result indicates that metacognitive feedback intervention is not a universal solution, and the intervention effect has clear individual boundary conditions.

The Cognitive Reflection Test (CRT) is used to measure an individual’s ability to suppress intuitive heuristics and initiate system 2 analytical processing. Individuals with high cognitive reflection can perceive metacognitive bias cues conveyed by feedback, actively suppress self service attribution, accept negative information pointing to insufficient self ability, and feedback can smoothly activate overconfidence plasticity, leading to calibration improvement. However, participants with low cognitive reflection rely more on the processing of the intuitive system 1. When faced with metacognitive feedback, they tend to rationalize bias prompts, classify them as accidental noise, refuse to update their self ability beliefs, and thus block the mediating pathway. This study replicated this boundary condition in the investment metacognitive calibration scenario.

The results of this study indicate that cognitive reflection regulates the first half of the mediating pathway, namely the activation intensity of feedback intervention on the plasticity of overconfidence; The latter half of the path of “overconfidence plasticity – confidence calibration” has not been moderated. This means that once overconfidence bias is successfully reduced through intervention, confidence calibration improvement can be achieved synchronously regardless of cognitive reflection level; The difficulty of intervention lies in how to trigger overconfidence correction in individuals with low cognitive reflection. These findings have practical implications, albeit preliminary ones. For investors with higher reflective tendencies, feedback about the judgment process rather than only about outcomes may more effectively prompt recalibration of confidence. For individuals with low cognitive reflection, brief metacognitive feedback alone appears insufficient, and structured methods such as writing decision rationales or keeping trading-reflection logs may be needed to help distinguish luck from ability. However, the present evidence concerns short-term calibration within a controlled task; it does not establish that such interventions produce sustained reductions in overconfidence or improve real investment performance. Calibration improvement, persistent reduction in overconfidence, and investment performance are distinct outcomes and should not be treated as interchangeable.

In addition, the laboratory virtual investment task lacks several features of actual financial markets, including real financial consequences, emotional arousal, and market uncertainty. The observed improvement in confidence calibration should therefore be interpreted as short-term, context-bound cognitive updating rather than as evidence of sustained reductions in overconfidence or improved investment performance. Metacognitive feedback may be a candidate component of calibration training in investor education, but whether such training effects persist over time and transfer to real financial decisions requires field experiments or longitudinal designs.

7 Research conclusions and prospects

7.1 Main research conclusions

Overconfidence in investment decision-making contexts is not a completely fixed stable trait and has significant cognitive plasticity. External feedback intervention can change investors’ overconfidence levels, but there are hierarchical differences in the activation effects of different types of feedback. Metacognitive feedback intervention can maximize the activation of overconfidence plasticity; Only result feedback can bring limited plasticity improvement; Without external feedback, overconfidence bias is difficult to spontaneously improve.

There are significant differences in the improvement effect of different feedback interventions on the confidence calibration of investment decisions. Metacognitive feedback intervention simultaneously presents the deviation information between objective decision results and subjective confidence accuracy, which can significantly reduce the expected calibration error ECE and effectively improve the quality of confidence calibration; Only providing feedback on the right or wrong results of profit and loss can only produce a weak calibration improvement effect; The participants in the non feedback group maintained a higher degree of metacognitive mismatch. Relying solely on market profit and loss results is difficult to spontaneously correct the confidence bias caused by investors’ overconfidence.

The plasticity of overconfidence partially mediates the relationship between feedback intervention and confidence calibration of investment decisions. On the one hand, feedback intervention can directly change the information processing strategy in the decision-making process of participants, directly improving confidence calibration; On the other hand, it can activate the plasticity of overconfidence, reduce the level of overconfidence bias, and thereby reduce the mismatch between subjective confidence and objective accuracy, achieving an improvement in calibration quality.

The mediating pathway from feedback intervention through overconfidence plasticity to confidence calibration was moderated by cognitive reflective ability, specifically in the first half of the pathway. Among individuals with moderate to high cognitive reflection, feedback intervention had a significant indirect effect on calibration through plasticity; among individuals with low cognitive reflection, this indirect effect was not significant. Once plasticity was successfully activated, however, calibration improvement was observed regardless of cognitive reflection level. These conclusions are limited to short-term calibration within the experimental setting and do not imply sustained reductions in overconfidence or improved investment performance.

Financial literacy and investment metacognitive calibration belong to relatively independent modules. The level of financial literacy cannot directly predict investors’ overconfidence and confidence calibration performance. Mastering financial knowledge does not mean being able to objectively evaluate the reliability of one’s own investment judgments. Investor education needs to supplement metacognitive psychological training content, and cannot only focus on teaching financial knowledge.

7.2 Research limitations

The feedback intervention in this study was a short-term laboratory single intervention, with a limited total duration of the experimental task. It only examined the immediate effects of the intervention and did not track the long-term maintenance of the intervention effect. Therefore, it cannot answer whether the calibration improvement effect brought by metacognitive feedback can still be retained after several weeks or months.

In addition, metacognitive feedback was presented for 3,500 ms, whereas outcome feedback was presented for 2,000 ms. Although feedback presentation duration was included as a covariate and the mediation effect remained robust (Table 2), the present design cannot fully separate the content of the feedback from the additional processing time it afforded. Future studies should equate presentation duration across conditions, or manipulate duration orthogonally to content, in order to isolate the effect of feedback content itself.

Although participants were required to have investment exposure, the sample comprised non-professional individual investors and did not include institutional investors or professional traders. Professional investors may differ in expertise, incentives, and feedback processing; therefore, the findings should not be generalized to professional investors. Nevertheless, because a growing share of retail investors manage their own portfolios, the results remain relevant to this population. Future research should oversample experienced and professional investors to test the boundary conditions of the intervention.

In this study, the mediator of overconfidence-plasticity and the outcome variable of confidence-calibration indicators are computed from the same task dataset. Future research may adopt separate task paradigms to measure overconfidence malleability and post-intervention calibration performance to further reduce shared-measurement-source concerns.

In addition, because plasticity is operationalized as a within-person difference score, it captures intervention responsiveness in the present task rather than a stable trait. Whether individual differences in plasticity are temporally stable requires longitudinal or repeated-intervention designs.

7.3 Future research prospects

Expand the boundary conditions and variables of intervention and conduct research on multiple moderation models. In the future, emotions, loss aversion, self threat, and investment motivation can be included in the model to explore how these variables regulate the activation of overconfidence plasticity through feedback interventions. Demographic-identity-related correlates of overconfidence, such as gender identity, may also shape individuals’ risk-taking and confidence patterns and deserve further experimental attention (). Further compare the gain effects of different auxiliary intervention methods for low cognitive reflective populations, and find a combination of de biased intervention plans that are suitable for low reflective populations. Additionally, individual dispositional traits such as decision-reinvestment tendency could interact with feedback-based debiasing effects and merit follow-up investigation ().

Expand the sample group and compare the differences among different investor groups. In the future, institutional investors, professional traders, and novice investors can be recruited to compare the differences in overconfidence plasticity among different investment experience groups, and to examine whether there are different patterns in intervention effects between professional and non professional investors.

Statements

Data availability statement

The original contributions presented in this study are included in this article/supplementary material, further inquiries can be directed to the corresponding author.

Ethics statement

The studies involving humans were approved by the Tongji University Ethics Working Group for Human Research. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

Author contributions

HZ: Investigation, Data curation, Formal analysis, Writing – review & editing, Conceptualization, Writing – original draft, Methodology.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

  • BarberB. M.OdeanT. (2001). Boys will be boys: Gender, overconfidence, and common stock investment.Quart. J. Econ.116, 261-292. 10.1162/003355301556400

  • Bruine de BruinW.ParkerA. M.FischhoffB. (2007). Individual differences in adult decision-making competence.J. Pers. Soc. Psychol.92, 938-956. 10.1037/0022-3514.92.5.938

  • CarpenterJ.ShermanM. T.KievitR. A.SethA. K.LauH.FlemingS. M. (2019). Domain-general enhancements of metacognitive ability through adaptive training.J. Exp. Psychol. General148, 51-64. 10.1037/xge0000505

  • FansherM.LalwaniP.AdkinsJ. T.ZhangH.QuirkM.CarlsonM.et al. (2025). A brief intervention to improve reasoning about accumulation.J. Exp. Psychol. Applied.31, 99–125. 10.1037/XAP0000532

  • FischhoffB.SlovicP.LichtensteinS. (1982). “Calibration of probabilities: The state of the art to 1980,” in Judgment Under Uncertainty: Heuristics and Biases, edsKahnemanD.SlovicP.TverskyA. (Cambridge: Cambridge University Press), 306–334. 10.1017/CBO9780511809477.022

  • FlemingS. M.LauH. (2014). How to measure metacognition.Front. Hum. Neurosci.8:443. 10.3389/fnhum.2014.00443

  • GervaisS.OdeanT. (2001). Learning to be overconfident.Rev. Financial Stud.14, 1-27. 10.1093/rfs/14.1.1

  • GlaserM.WeberM. (2007). Overconfidence and trading volume.Geneva Risk Insurance Rev.32, 1–36. 10.1007/s10713-007-0003-3

  • GödkerK.JiaoP.SmeetsP. (2025). Investor memory.Rev. Financial Stud.38, 1595-1640. 10.1093/rfs/hhaf006

  • HanY.DunningD. (2024). Metaknowledge of experts versus nonexperts: Do experts know better what they do and do not know?J. Behav. Decision Making37:e2375. 10.1002/bdm.2375

  • HayesA. F. (2022). Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach, 3rd Edn. New York, NY: Guilford Press.

  • HuffmanD.RaymondC.ShvetsJ. (2022). Persistent overconfidence and biased memory: Evidence from managers.Am. Econ. Rev.112, 3141-3175. 10.1257/aer.20190668

  • IslerO.YilmazO.DogruyolB. (2020). Activating reflective thinking with decision justification and debiasing training.Judgment Decis. Mak.15, 926–938. 10.1017/s1930297500008147

  • KahnemanD.TverskyA. (1979). Prospect theory: An analysis of decision under risk.Econometrica47, 263-291. 10.2307/1914185

  • KatyalS.HuysQ. J. M.DolanR. J.FlemingS. M. (2025). Distorted learning from local metacognition supports transdiagnostic underconfidence.Nat. Commun.16:1854. 10.1038/s41467-025-57040-0

  • KerenG. (1991). Calibration and probability judgments: Conceptual and methodological issues.Acta Psychol.77, 217–273. 10.1016/0001-6918(91)90036-Y

  • KozolE. (2026). The mere urgency effect in household financial decisions: Evidence from a high-inflation economy.Finance Res. Lett.109:110564. 10.1016/j.frl.2026.110564

  • LawrenceR. E.WickB. (2026). Gender identity, overconfidence, and risk attitudes.J. Consum. Psychol.36, 369–379. 10.1002/JCPY.70022

  • LawsonA. M.LarrickP. R.SollB. J. (2026). Individual differences in overconfidence: A common core and its motivational and cognitive correlates.Cognition276:106665. 10.1016/J.Cognition.2026.106665

  • LusardiA.MitchellO. S. (2014). The economic importance of financial literacy: Theory and evidence.J. Econ. Literature52, 5–44. 10.1257/jel.52.1.5

  • ManiscalcoB.LauH. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings.Consciousness Cogn.21, 422-430. 10.1016/j.concog.2011.10.001

  • MeierC.De MelloL. (2020). Investor overconfidence in experimental asset markets across market states.J. Behav. Finance21, 369–384. 10.1080/15427560.2019.1692845

  • MillerD. T.RossM. (1975). Self-serving biases in the attribution of causality: Fact or fiction?Psychol. Bull.82, 213-225. 10.1037/h0076486

  • MooreD. A.TenneyE. R.HaranU. (2015). “Overprecision in judgment,” in The Wiley Blackwell Handbook of Judgment and Decision Making, edsKerenG.WuG. (Chichester: Wiley-Blackwell), 182–209. 10.1002/9781118468333.ch6

  • PiehlmaierD. M. (2023). The one-man show: The effect of joint decision-making on investor overconfidence.J. Consum. Res.50, 426-446. 10.1093/jcr/ucac054

  • PourhassanJ.HitzlW.LangensteinB.ErbguthF. (2025). Pain and mental-health scores significantly different in male and female football-players with traits of decision-reinvestment, but heart-rate-variability differences are unsupported.Sci. Rep.15, 32587–32587. 10.1038/S41598-025-19985-6

  • SteyversM.PetersK. A. (2026). Metacognition and uncertainty communication in humans and large language models.Curr. Dir. Psychol. Sci.35, 131–139. 10.1177/09637214251391158

  • StoneE. R.OpelR. B. (2000). Training to improve calibration and discrimination: The Effects of performance and environmental feedback.Organ. Behav. Hum. Decis. Process.83, 282–309. 10.1006/obhd.2000.2910

  • StoneE. R.ParkerA. M.SomervilleA. H.NixonB.KemmerlyR.BongardM. M. (2025). Influencing confidence: Testing ways to increase or decrease confidence in knowledge.J. Behav. Decis. Making38:e70048. 10.1002/bdm.70048

  • SwaryandiniG.GrahamJ.GriffithS.GriloV.RuzzanteF.ZhangX.et al. (2025). Systematic review and meta-analysis of educational approaches to reduce cognitive biases among students.Nat. Hum. Behav.9, 2510–2538. 10.1038/s41562-025-02253-y

  • ThompsonV. A.JohnsonS. C. (2014). Conflict, metacognition, and analytic thinking.Thinking Reasoning20, 215-244. 10.1080/13546783.2013.869763

  • WaltersD. J.FernbachP. M. (2021). Investor memory of past performance is positively biased and predicts overconfidence.Proc. Natl. Acad. Sci.118:e2026680118. 10.1073/pnas.2026680118

Summary

Keywords

cognitive reflection, confidence calibration, feedback intervention, investment decision, overconfidence, plasticity

Citation

Zheng H (2026) The plasticity of overconfidence: the impact of feedback intervention on confidence calibration in investment decisions. Front. Psychol. 17:1988364. doi: 10.3389/fpsyg.2026.1988364

Received

04 September 2026

Published

18 November 2026

Updates

Copyright

*Correspondence: Haowen Zheng,

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.



Source link

Related posts

Indonesia’s financial centre must grow from its domestic economy – Lowy Institute

D.William

GE Vernova (GEV) Hits All-Time High Following Impressive Q1 Report

D.William

ODPP and ICPAK seek stronger prosecution of financial crimes

D.William

Leave a Comment