_{1}

^{*}

This paper details an experiment designed to explore the trading behaviors of investors that result from psychological biases and social interactions. In total, 88 investors were tracked for 6 months and 40,795 transactions were recorded. The research conducted an experimental survey and estimated a system analysis model to generate several important conclusions. First, the degree of regret bias and the disposition effects are unrelated, probably because the professional training of investors and the disposition effects are not significantly related. Second, if investors are affected by contradictions arising from their decisions, then the likelihood that they will sell a stock will decrease as the investor relationships in the community improve and the regret bias increases. Third, male investors prefer to trade derivatives, and even after controlling for the degree of regret bias, this preference is still significant.

Understanding investors’ trading behaviors is a common focus for both the financial industry and financial economic research. Many factors that affect investor trading have been previously identified, such as the use of momentum strategies for buying or selling securities [

A person’s mental state is often inscrutable. People’s ideas are difficult to predict, and it is these ideas that affect their investment behaviors [

In addition to regret bias, investors are also affected by their environments, which influence their trading behavior; herding behavior is one example where trading activities are affected by the environment [

Finally, the trading behaviors of investors are also affected by the individuals’ characteristics, such as their emotional responses and abilities to interpret market information [

This study had several key objectives and makes a number of contributions. The primary aim was to create a model to analyze the impact of interactions between investor psychology, relationships, gender and variables related to trading behavior. The subjects’ answers may have differed from the actual situations due to the influence of psychological bias. For example, the respondents may have forgotten the results of past transactions; additionally, they may have avoided acknowledging their own investment mistakes, and have provided random answers. Thus, the credibility of the conclusions requires further improvement. This paper examined a large sample and analyzed the subjects via the content of a matching questionnaire; therefore, the results are robust. In addition to the disposition effect, the impact of the investors’ characteristics on asset adjustments was observed. These effects include changes in the selling and allocation of derivatives. The above conclusions may strengthen weaknesses in poorly studied areas of similar research and may also be used as a reference in personal finance and risk management literature.

This complete analysis revealed several important findings. First, regret bias and the disposition effect have no statistically significant relationship. Second, stronger regret biases and stronger social relationships correspond to fewer stocks sold by investors. Third, after controlling for regret bias, we found that male investors allocate a greater amount of capital to derivatives than do female investors. This paper is organized as follows: Section I includes the introduction and provides the statement of purpose for this study; Section II presents the theory and assumptions; Section III reviews the methods of the study, including the choice of data, the experimental design and the model; Section IV contains the empirical analysis and the system estimation model used to perform the data analysis. Finally, the last section provides conclusions and recommendations and discusses the results.

This section reviews the relationship between the variables and specific theoretical studies. Additionally, this section establishes the hypotheses that served as the basis for the empirical analysis.

Because investors who buy securities are afraid of loss, they often sell off a portion of their profitable stocks, even if these stocks have not yet generated high profits. This behavior is driven by the fear of a sudden stock price reversal. Investors fear incurring losses, since this could generate serious psychological regrets. Conversely, stock investors may remain hopeful that they can “turn around” stocks that are gradually losing money, thus avoiding an acknowledgment of their decision-making errors [

Goo et al. used confirmatory factor analysis to assess investors in the Taiwan market [

Hypothesis 1: In the context of performance concerns, there is a clear and positive relationship between regret bias and the disposition effect.

People are social animals, and their decision-making processes are often influenced by other people. In financial literature, the best-known example of this influence is the herd effect [

As the investor community relationship improves and social interactions become stronger, the chances that direct communication will take place, that investors will listen to the views of others or that investors will adjust their original stock portfolios, will improve [^{1} In other words, the investor’s selling rate should show a downward trend due to the interaction between these two psychological biases. Accordingly, we propose Hypothesis 2.

^{1}With respect to the contradictions arising from the decision-making process, investors will sell their stocks in the first scenario. In the second scenario, although the number of sell orders is reduced, there is still the possibility of selling the stock, and the probability of each of the hypotheses is 1/2. The third scenario is also uncertain; there is a possibility that the stock will not be sold. The probability of each of the hypotheses is again 1/2. The fourth scenario is to not sell the stock; i.e. when considering all of the criteria and the equal distribution of the hypotheses, the probability of selling stock is 1/2(1/4 + 1/4 × 1/2 + 1/4 × 1/2). Compared to the original conditions, the probability of an investor selling his or her securities decreased from 1 to 1/2, which is clearly a large decrease. According to this reasoning, the proportion of sell positions among all trades will decrease.

Hypothesis 2: Investors with a higher degree of regret bias and better relationships within the investment community will sell their stocks at a lower rate than other investors because of the effects of decision-making conflicts.

Maital et al. and Tauni et al. indicated that the irrational behavior of investors is also influenced by their personality traits, emotional reactions and ability to interpret the impact of market information [

Hypothesis 3: On average, male investors prefer to trade derivatives, even with control for different levels of regret bias.

This section describes the objectives, data collection processes and models of the study. The detailed discussion of the model also includes the measurements of the variables, the experimental design and the measurement methods. In this section, we describe the construction of the models of regret bias, social relationships, gender and transactions.

In this study, Taiwanese investors in emerging Asian markets were used as the subjects. Because the investors’ transaction records are protected by local laws and are not readily available, we used other financial data for the analysis. This data collection process is similar to the methods used by Oehler et al., Weber and Camerer, and Chui [

Concurrent with the questionnaire study, the respondents participated in a virtual trading competition. The investment period was 6 months, and an investment amount of NT$5 million was granted (In the Taiwan Stock Market, only ordinary investors are considered, and securities dealers are given a credit line ranging from NT$3 million to NT$5 million. Therefore, to simulate an actual trading situation, the students were given a balance of only NT$5 million), which could be traded for financial products, including stocks, futures and options. The students were also allowed to engage in credit transactions. This design is in accordance with the general conditions in which investors operate. Thus, the experimental data closely resembled the content of actual transactions. The positions held and the price changes in virtual trading and actual trading are no different. A system vendor was then asked to provide a field audit of the digital processing and check for accuracy. Transactions with unusual numbers were removed. On average, there were 463 transactions per person during the period of the study.

This study used six variables: regret bias (RB), the disposition effect (DE), social relationships (SR), the ratio of sell orders (TO), gender (DUMMY) and asset allocation ratio (AL). The RB, SR and DUMMY data were obtained from the survey, whereas the DE, TO and AL data were calculated from the experimental data. The measure of regret bias is based on Pompian’s diagnostic questionnaire [

In this study, each investor was assigned a number ranging from 1 to 88, and then the corresponding virtual transactions were analyzed. First, the data for the six variables (RB, DE, SR, TO, DUMMY and AL) were uploaded for each investor. The RB, SR and DUMMY data were derived from the questionnaire, whereas the DE, TO and AL values were calculated using data from the virtual transactions. Second, the subjects were divided into four groups based on the extent of regret bias. The upper 1/4 of the samples had the least degree of regret bias, and the lower 1/4 of the sample had the greatest degree of regret bias. Each group had approximately 22 investors. Similarly, the respondents were divided based on the quality of their social relationships. Third, the disposition effects of the groups were divided based on regret bias to determine whether the differences among the groups were significant. Next, the ratio of sell orders was calculated for the groups with the highest degree of regret bias and the strongest social relationships. This ratio was then compared with the ratio of sell orders of the groups with the least degree of regret bias and with the least developed social relationships to evaluate the differences between the two groups. Finally, in the regret bias groups, the male and female investors were separated and the ratio of asset allocation was calculated for the two groups to determine whether there were any differences between the groups.

To determine the robustness of the results, this paper used an alternative set of equations, the weighted least squares (WLS). The advantage of this approach is that the system equations can incorporate all of the variables into the equation, which has full use of the other information from the equations, to increase the explanatory power of this equation. For example, gender may have an impact on the disposition effect, and the ratio of sell orders (turnover rate) may also affect the disposition effect [_{i} is the coefficient for the individual variables. The regression coefficient was estimated in Equation (2) and ε_{i} represents the equation residuals [

DE i = α + β 1 RB i + ε i

TO i = α + β 1 RB i + β 2 SR i + β 3 RB i × SR i + ε i

AL i = α + β 1 RB i + β 2 DUMMY i + β 3 RB i × DUMMY i + ε i (1)

B i = ( X ′ Ω − 1 X ) − 1 X ′ Ω − 1 y , Ω = [ ω 1 0 0 ⋯ 0 0 ω 2 0 ⋯ 0 0 0 ω 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ ω n ] , ω i = V a r ( ε i ) / σ 2 (2)

This section contains the empirical analysis, which includes the descriptive statistics, hypothesis t-test and robustness analysis for the estimated models.

Six variables were used in this study: DE, RB, SR, TO, DUMMY and AL.

Variables | Descriptive statistics | |||||
---|---|---|---|---|---|---|

N | Mean | Median | Min | Max | Std | |

DE | 88 | −0.0577 | −0.0490 | −1.0000 | 1.0000 | 0.4947 |

RB | 88 | 3.2915 | 3.6667 | 1.0000 | 5.0000 | 1.1754 |

SR | 88 | 2.8182 | 3.0000 | 1.0000 | 4.5000 | 0.7701 |

TO | 88 | 0.3853 | 0.3871 | 0.0580 | 0.7467 | 0.1467 |

AL | 88 | 0.2378 | 0.0772 | 0.0000 | 1.0000 | 0.3332 |

showed the largest variability among the investors, whereas the difference in the ratio of sell orders was the smallest. Additional detailed data are shown in

To test Hypothesis 1, the impact of investors’ regret bias on the disposition effect was evaluated. In other words, it was determined whether a positive relationship existed between the two variables. The results from the t-test analysis of the influence of regret bias on the disposition effect are shown in

It was unclear why the regret bias effect showed a significant positive relationship with the disposition effect. Further analysis demonstrated that the coefficient of the disposition effect for all investors was −0.0577 (see

Variables | Statistics | |||
---|---|---|---|---|

Mean in lowest regret bias group (N = 22) | Mean in highest regret bias group (N = 22) | t-value of the difference in the averages | p-value | |

DE | −0.1049 | −0.0804 | −0.1732 | 0.8634 |

professional training, the investors were less likely to engage in irrational behavior (DE = −0.0577). This result is similar to the findings of Locke and Mann [

Hypothesis 2 was designed to verify that the investors with more regret bias and better social relationships lose touch with their original judgments due to contradictions that arise during their decision-making processes. As a result, these investors sell fewer shares of their stocks and thus reduce the number of their sell orders. In other words, this study intended to confirm that the average number of sell orders of the lowest group (in terms of both regret bias and social relationships) would be significantly higher than the average number of sell orders of the highest group.

Hypothesis 3 was tested to verify that investors who differ in gender and in their degrees of regret bias allocate capital to derivative assets in different ratios. In

Variable | Statistics | ||||
---|---|---|---|---|---|

Average in the lowest regret bias and community relationship groups (N = 22) | Average in the highest regret bias and community relationship groups (N = 22) | t-value of the difference in the averages | p-value | ||

TO | 0.4482 | 0.3642 | 1.9813 | 0.0559 | |

this study, men were assumed to have a higher degree of overconfidence, to be less risk-averse, to more likely take risks to generate profits and to be more likely to allocate capital to derivative assets compared to women.

The samples were further divided into groups according to the degree of regret bias. We then analyzed the influence of gender on the ratio of derivative asset allocation. The method used here was similar to the methods used in 4.2.1 and 4.2.2.

Variable | Statistics | |||
---|---|---|---|---|

Female group average (N = 68) | Male group average (N = 20) | t-value the average differences | p-value | |

AL | 0.1900 | 0.3952 | −2.5005 | 0.0143 |

Variable | Statistics (Q1) | |||
---|---|---|---|---|

Female group average (N = 18) | Male group average (N = 4) | t-value of the average differences | p-value | |

AL | 0.2690 | 0.7891 | −2.6966 | 0.0143 |

presents data derived from the t-test of gender differences on the ratio of derivative allocation in the highest regret bias group. According to the table, the average value of the female investors’ derivative asset allocation was 0.0609, whereas the average value for the male investors was 0.2636. The difference in the average AL values between the two groups was −0.2027 (t = −1.8614, p = 0.0775), which was significant at the 10% confidence level.

In contrast, the data in

In this section, simultaneous equations were used as a system model. The interactive changes between the variables were observed and the results of the three hypotheses were further validated.

Variable | Statistics (Q2) | |||
---|---|---|---|---|

Female group average (N = 16) | Male group average (N = 6) | t-value of the average difference | p-value | |

AL | 0.2373 | 0.3090 | −0.4331 | 0.6696 |

Variable | Statistics (Q3) | |||
---|---|---|---|---|

Female group average (N = 17) | Male group average (N = 5) | t-value of the average differences | p-value | |

AL | 0.1953 | 0.3149 | −0.7412 | 0.4668 |

Variable | Statistics (Q4) | |||
---|---|---|---|---|

Female group average (N = 17) | Male group average (N = 5) | t-value of the average differences | p-value | |

AL | 0.0609 | 0.2636 | −1.8614 | 0.0775 |

Coefficient | t-value | p-value | |
---|---|---|---|

C(1) | −0.100938 | −0.644431 | 0.5199 |

C(2) | 0.013143 | 0.293086 | 0.7697 |

C(3) | 0.468506 | 2.628806 | 0.0091 |

C(4) | 0.029495 | 0.574097 | 0.5664 |

C(5) | −0.029975 | −0.510475 | 0.6102 |

C(6) | −0.010359 | −0.61091 | 0.5418 |

C(7) | 0.361014 | 3.36144 | 0.0009 |

C(8) | −0.052543 | −1.696976 | 0.0909 |

C(9) | 0.509043 | 1.98097 | 0.0487 |

C(10) | −0.086454 | −1.195976 | 0.2328 |

between RB and DE. The coefficient was +0.0131 and did not reach a level of statistical significance, although the relationship is consistent with the predictions of prospect theory ( ∂ DE / ∂ RB > 0 ). The coefficient C(6) verified the effect of the interaction of SR and RB on TO. The coefficient itself was −0.0104, which did not reach statistical significance; however, the relationship is in agreement with the hypothesis ( ∂ ( SR × RB ) / ∂ TO < 0 ). The coefficient C(9) verified the impact of DUMMY on AL. After controlling for the two variables, i.e. RB and RB × DUMMY, we found that the regression coefficient was +0.5090 (p < 0.05), which conforms to Hypothesis 3 ( ∂ DUMMY / ∂ AL > 0 ). Additionally, the coefficient C(9) reached a level of statistical significance. The other regression coefficients are listed in

The purpose of this study was to explore the influences of psychological bias, social relationships and personal characteristics on trading behavior. Through descriptive statistics, t-tests and the weighted least squares regressions, several important conclusions were reached. First, Hypothesis 1 did not hold. A significant

relationship between regret bias and the disposition effect was not observed. There are several reasons for the non-significance of these results. First, the lack of significance may be related to the professional training of the investors and the insignificance of the disposition effect, as argued by Locke and Mann [

The results of this paper were derived from the observations of 40,795 transactions made by 88 investors over 6 months. Although the trading platform, trading mechanism and the amount of funds were modeled on real trading situations, there may have been certain differences between the model and the actual investment scenario. Grinblatt and Keloharju discussed several factors that may affect investors’ trading behavior [

Ho, C.M. (2018) The Effects of Social Interaction and Psychological Bias on Trading Behavior: Evidence from a Laboratory Experiment. Journal of Mathematical Finance, 8, 178-196. https://doi.org/10.4236/jmf.2018.81014