4.1 Predictor variables and interactions
All logistic regression procedures used, irrespective of the nature of the outcome, included four predictor variables (gender, race, trait anxiety and trait anger). One regression included a fifth predictor, namely the participants’ perception of how predictable they believed the trend of a particular asset class to be. Before presenting the findings of the logistic regressions, the interactions among predictor variables are discussed.
Table
1 reports the mean trait anxiety and anger scores for males and females. Two independent
t-tests were conducted to test for significant differences.
Table 1
Descriptive statistics of personality traits by gender
Trait anxiety | Female | 45.10 | 46.00 | 10.713 | 27 | 69 | 118 |
Male | 39.55 | 39.00 | 7.315 | 23 | 62 | 133 |
Trait anger | Female | 19.48 | 19.00 | 5.051 | 13 | 38 | 133 |
Male | 19.22 | 19.00 | 4.598 | 12 | 34 | 137 |
The two genders did not differ significantly regarding anger [
t (268) = 0.446,
p = 0.656], but a statistically significant difference was observed for anxiety [
t (203) = 4.736,
p < 0.05]. Females (mean score 45.10) were more prone to anxiety than males (mean score 39.55). This same pattern of significance was observed for the comparison between participants of different races in Table
2.
Table 2
Descriptive statistics of personality traits by race
Trait anxiety | Black | 45.09 | 44.00 | 9.593 | 28 | 66 | 117 |
White | 37.75 | 38.00 | 7.034 | 23 | 54 | 104 |
Trait anger | Black | 19.32 | 18.00 | 5.126 | 12 | 38 | 137 |
White | 19.14 | 19.00 | 3.925 | 13 | 28 | 102 |
Black and white participants did not differ significantly in terms of anger [t (237) = 0.34, p = 0.754]. The statistically significant difference found for anxiety [t (212) = 6.537, p < 0.05] revealed Black participants (mean score 45.09) to be more prone to this personality trait compared to White participants (mean score 37.75).
Finally, the two measures of trait anxiety and trait anger were found to be significantly correlated, based on the Pearson correlation coefficient (r = 0.304, N = 245, p < 0.05). This positive correlation implies a tendency for relatively high levels of trait anxiety to be associated with relatively high levels of trait anger.
The incorporation of gender and race as explanatory variables is unique to this study and in these preliminary results already shows differences in the prevalence of trait anxiety. Research has illustrated both gender and race to influence financial knowledge and risk aversion (Willows
2019a; Willows and West
2015), which are notable indicators of financial behaviour and decision-making (Atkinson and Messy
2012).
Despite the sample of participants being similar in respect of their year of study and course of study at a tertiary institution, South Africa’s historical apartheid policy has caused many divides among the population (Willows
2019b). The racially exclusionist education policies implemented resulted in many Black South Africans experiencing substandard schooling or having grown up in households with parents who might never have had a formal qualification or exposure to financial instruments (Draper and Spaull
2015). Financial literacy is acquired over time; thus, this might have been a disadvantage to Black participants. Willows (
2019a), when assessing actual and self-assessed financial literacy, also noted that White participants tend to rate their own perceived level of financial knowledge on a higher scale than African participants. This might explain the preference of White South Africans to select more aggressive investments (Willows
2019b) and the racial differences noted in the measurement of trait anxiety.
Pertaining to the gender differences noted; women are more risk averse than men (Gambetti and Giusberti
2019; Willows and West
2015) and gender differences in financial knowledge scores are seen in research performed around the world (Lusardi and Mitchell
2011; Willows
2019b; Xu and Zia
2012). Like the findings with race, Willows (
2019a) noted that men tend to rate their own perceived level of financial knowledge on a higher scale than women. Kannadhasan et al. (
2016) add that even when women possess higher self-esteem than men (which is positively related to risk tolerance), men are still more risk tolerant as they are overconfident in their ability to achieve positive investment gains. Furthermore, Grable and Roszkowski (
2007) found that women tend to underestimate their risk tolerance whereas men overestimate. As trait anxiety appears to be associated with strong avoidance goals, i.e. goals that are focused on avoiding or eliminating undesirable outcomes, these differences can be expected amongst the different genders.
4.2 Preference for investment in six asset classes
For the first investment decision, the participants were asked in which of six asset classes they would choose to invest. They could choose more than one. They were also asked to indicate how much they believed the trend of the investment for each asset class to be predictable. Table
3 summarises the responses.
Table 3
Predictability of trend of investment associated with six asset classes and percentages of participants selecting each class
Alternatives (commodities) | 22 | 40 | 32 | 29 | 278 |
Cash | 37 | 12 | 24 | 64 | 284 |
Equity (shares) | 72 | 56 | 23 | 21 | 282 |
Fixed income (bonds) | 38 | 4 | 13 | 82 | 284 |
Insurance products | 9 | 14 | 31 | 55 | 278 |
Real estate | 40 | 14 | 19 | 67 | 284 |
Investment in equity was the most popular choice (72%), with real estate (40%), fixed income (38%) and cash (37%) trailing relatively far behind. Only 9% of participants chose insurance products as an investment. Table
3 further illustrates that equity, the most preferred asset class, was perceived as the most unpredictable asset class (56%). This means that equity was perceived as the riskiest asset class. Contrasting to this, a large percentage of participants (82%) considered fixed income to be the most predictable and therefore the least risky choice.
In order to determine the relative importance of trait anxiety and trait anger in choosing a particular asset class, a series of binary logistic regressions was performed. In each regression, the outcome variable was one of the six asset classes. In addition to trait anxiety and trait anger, gender and race were also included as predictors to determine their relative importance in relation to the two personality traits. The rating of the predictability of the trend of investment was added as a fifth predictor. This was done on the assumption that the choice for a particular investment was influenced by the investor’s perception of how predictable the investment is. The results of the binary logistic regression is summarised in Table
4.
Table 4
Logistic regression predicting likelihood of investing in six asset classes
Alternatives (commodities) (N = 211) |
Predictability | 0.330 | 0.204 | 2.621 | 0.105 | 1.391 | 0.922 | 2.076 |
Gender | 1.788 | 0.467 | 14.666 | 0.000 | 5.979 | 2.394 | 14.931 |
Race | − 0.263 | 0.402 | 0.430 | 0.512 | 0.769 | 0.350 | 1.688 |
Trait anxiety | 0.006 | 0.026 | 0.046 | 0.829 | 1.006 | 0.956 | 1.058 |
Trait anger | − 0.059 | 0.047 | 1.550 | 0.213 | 0.943 | 0.860 | 1.034 |
Constant | − 2.510 | 1.386 | 3.278 | 0.070 | 0.081 | – | – |
Cash (N = 211) |
Predictability | − 0.111 | 0.136 | 0.662 | 0.416 | 0.895 | 0.685 | 1.169 |
Gender | 0.802 | 0.313 | 6.577 | 0.010 | 2.230 | 1.208 | 4.115 |
Race | 0.123 | 0.322 | 0.146 | 0.703 | 1.131 | 0.601 | 2.127 |
Trait anxiety | 0.240 | 0.190 | 1.565 | 0.211 | 1.024 | 0.986 | 1.064 |
Trait anger | − 0.090 | 0.033 | 0.079 | 0.778 | 0.991 | 0.929 | 1.057 |
Constant | − 1.383 | 1.098 | 1.584 | 0.208 | 0.251 | – | – |
Equities (shares) (N = 211) |
Predictability | 0.038 | 0.172 | 0.049 | 0.826 | 1.039 | 0.741 | 1.455 |
Gender | 0.054 | 0.357 | 0.023 | 0.879 | 1.056 | 0.524 | 2.126 |
Race | 0.995 | 0.380 | 6.843 | 0.009 | 2.704 | 1.283 | 5.698 |
Trait anxiety | − 0.032 | 0.022 | 2.070 | 0.150 | 0.969 | 0.928 | 1.012 |
Trait anger | 0.091 | 0.040 | 5.132 | 0.023 | 1.095 | 1.012 | 1.184 |
Constant | 0.189 | 1.128 | 0.028 | 0.867 | 1.208 | – | – |
Fixed income (bonds) (N = 211) |
Predictability | 0.058 | 0.176 | 0.111 | 0.739 | 1.060 | 0.752 | 1.495 |
Gender | 0.183 | 0.306 | 0.358 | 0.550 | 1.201 | 0.659 | 2.190 |
Race | − 0.429 | 0.323 | 1.768 | 0.184 | 0.651 | 0.346 | 1.225 |
Trait anxiety | 0.007 | 0.019 | 0.161 | 0.689 | 1.008 | 0.971 | 1.045 |
Trait anger | − 0.011 | 0.032 | 0.112 | 0.738 | 0.989 | 0.929 | 1.053 |
Constant | − 0.796 | 1.221 | 0.425 | 0.514 | 0.451 | – | – |
Insurance products (N = 209) |
Predictability | 0.386 | 0.281 | 1.896 | 0.169 | 1.472 | 0.849 | 2.550 |
Gender | − 1.491 | 0.616 | 5.852 | 0.016 | 0.225 | 0.067 | 0.754 |
Race | − 2.001 | 0.794 | 6.350 | 0.012 | 0.135 | 0.029 | 0.641 |
Trait anxiety | 0.009 | 0.029 | 0.092 | 0.762 | 1.009 | 0.954 | 1.067 |
Trait anger | − 0.310 | 0.500 | 0.383 | 0.536 | 0.970 | 0.880 | 1.069 |
Constant | − 2.398 | 1.767 | 1.843 | 0.175 | 0.091 | – | – |
Real estate (N = 211) |
Predictability | 0.343 | 0.167 | 4.240 | 0.039 | 1.410 | 1.017 | 1.954 |
Gender | − 0.357 | 0.309 | 1.335 | 0.248 | 0.700 | 0.382 | 1.282 |
Race | − 0.720 | 0.331 | 4.727 | 0.030 | 0.487 | 0.254 | 0.931 |
Trait anxiety | − 0.027 | 0.019 | 2.005 | 0.157 | 0.973 | 0.938 | 1.010 |
Trait anger | − 0.018 | 0.033 | 0.308 | 0.579 | 0.982 | 0.921 | 1.047 |
Constant | 0.231 | 1.201 | 0.037 | 0.848 | 1.260 | – | – |
In Table
4, the
b coefficient represents the change in the logit of the outcome variable that can be attributed to a one-unit change in the predictor variable. The logit of the outcome variable is the natural logarithm of the odds of that outcome occurring (Field
2013). The Wald statistic indicates whether the
b coefficient for a predictor is significantly different from 0, in other words, whether it significantly predicts the outcome. This is the case when the associated
p value is less than 0.05. Since the
b coefficient involves a logarithmic transformation, the odds ratio presents an easier way to interpret the relative contribution of predictors to the outcome.
Table
4 illustrates that trait anger is only statistically significant in predicting whether to invest in equity. The odds of choosing equity as an asset class increased by about 1.1, by moving from one score in the anger measure to the next higher score.
The gender variable was predictive across three asset classes, with it being statistically significant in choosing alternative investments, cash and insurance products. The highest odds ratio was for alternative investments (5.979). This means that the odds of choosing alternative investments were about six times higher for males than for females. The odds of choosing cash products were also 2.2 times higher for males. Showing opposite preferences, the odds of choosing insurance products were 4.4 times (1/0.225) higher for females. This is explained by the higher risk aversion in women (Willows and West
2015). Reviewing experimental evidence of risk aversion, Eckel and Grossman (
2008) find evidence of women purchasing insurance more often than men and purchasing more extensive insurance than men.
The race variable was also predictable, being statistically significant in choosing equity, insurance products and real estate. The odds of White participants choosing to invest in equity were almost three times higher than those of a Black participant. Alternatively, the odds of a Black respondent choosing insurance products or real estate were 7.4 times (1/0.135) and 2.1 times (1/0.487) higher than for a White participant. These findings suggest similar risk aversion differences amongst participants of different race, as was found with gender. It also confirms findings by Willows (
2019b) who looked at the investment choices for retirement funds among a sample of South Africans and noted an increased preference for higher risk (higher return) retirement products by White South Africans.
Regarding perceived predictability of the asset class, the odds of choosing real estate increased by about 1.4 as the perception of predictability increased. The predictability of this variable for real estate only suggests a possible relationship that is a recommended area for future research.
4.3 Preference for risky investments
Three scenarios measured the participants’ preference for risky investments. These are reported in Table
5.
Table 5
Responses to three scenarios that measured preference for risky investments
What is the best investment for you? (N = 286) |
12-month deposit with an interest rate of 7.5% | 170 | 59 |
6-month bond with an interest rate of 6.9% | 90 | 32 |
Interest-bearing account with an interest rate of 3.5% | 26 | 9 |
You are offered different investment portfolios with the following characteristics: ‘low gain/no loss’, ‘medium gain/medium loss’ and ‘high gain/high loss’. Which portfolio do you choose? (N = 288) |
Low gain/no loss | 28 | 10 |
Medium gain/medium loss | 211 | 73 |
High gain/high loss | 49 | 17 |
You have shares, what do you do if the share declines in price? (N = 286) |
Sell at a loss | 18 | 6 |
Wait some days with the possibility to either lose or advance | 152 | 53 |
Wait some weeks with the possibility to either lose or advance even further | 116 | 41 |
The first scenario in Table
5 measured the participants’ willingness to take investment risk. Most participants expressed a low willingness to accept risk, as indicated by the 59% who chose the 12-month deposit with an interest rate of 7.5%. The second scenario asked the participants to identify with one of three categories of investors, which were indirectly derived from the response options provided: risk-averse (“low gain/no loss”), risk-neutral (“medium gain/medium loss”) and risk-loving (“high gain/high loss”), they would choose. Close to three-quarters (73%) of participants could be classified as risk-neutral investors. The third scenario measured the participants’ behaviour regarding volatility, expressed as a decline in share price. Only 6% of participants indicated that they would sell at a loss. This implies that most of the participants (94%) tended to be risk-loving, albeit to different degrees (waiting days vs. waiting weeks).
For each of the scenarios above, a logistic regression was performed with the scenario variable as outcome, and anxiety, anger, gender and race as predictors. Table
6 reports the results of a multinomial logistic regression for scenario 1, which relates to the participants’ willingness to take investment risk. A multinomial logistic regression was performed since the outcome variable was non-binary. The outcome variable was presented as two comparisons, with the least risky option (12-month deposit with an interest rate of 7.5% as reference category).
Table 6
Logistic regression predicting likelihood of choosing less risky or more risky investment options (N = 213)
Comparison 1: 6-month bond with an interest rate of 6.9% versus 12-month deposit with an interest rate of 7.5% |
Gender | − 1.361 | 0.350 | 15.088 | 0.000 | 0.256 | 0.129 | 0.510 |
Race | 0.284 | 0.349 | 0.659 | 0.417 | 1.328 | 0.670 | 2.634 |
Trait anxiety | 0.010 | 0.021 | 0.203 | 0.652 | 1.010 | 0.969 | 1.052 |
Trait anger | 0.017 | 0.035 | 0.221 | 0.638 | 1.017 | 0.949 | 1.090 |
Constant | − 0.863 | 0.879 | 0.964 | 0.326 | – | – | – |
Comparison 2: interest-bearing account with an interest rate of 3.5% versus 12-month deposit with an interest rate of 7.5% |
Gender | − 1.620 | 0.583 | 7.728 | 0.005 | 0.198 | 0.063 | 0.620 |
Race | 0.834 | 0.574 | 2.112 | 0.146 | 2.303 | 0.748 | 7.092 |
Trait anxiety | 0.062 | 0.033 | 3.479 | 0.062 | 1.064 | 0.997 | 1.135 |
Trait anger | 0.013 | 0.051 | 0.066 | 0.797 | 1.013 | 0.916 | 1.120 |
Constant | − 4.445 | 1.426 | 9.712 | 0.002 | – | – | – |
According to Table
6, gender was the only significant predictor in both comparisons. The odds ratio indicates that the difference in odds between female to male of choosing the 6-month bond over the 12-month deposit was 0.256. To state it differently, the odds of females choosing the 6-month bond over the 12-month deposit were 3.9 times (1/0.256) higher. For the second comparison, the odds of females choosing the interest-bearing account over the 12-month deposit was 5.1 times (1/0.198) higher. Table
7 illustrates the regression result for the second scenario i.e. choosing between different investment portfolios.
Table 7
Logistic regression predicting likelihood of choosing between investment portfolios with different levels of gain/loss (N = 213)
Comparison 1: medium gain/medium loss versus low gain/low loss |
Gender | − 0.546 | 0.557 | 0.959 | 0.327 | 0.580 | 0.194 | 1.727 |
Race | 0.289 | 0.576 | 0.253 | 0.615 | 1.336 | 0.432 | 4.127 |
Trait anxiety | − 0.036 | 0.031 | 1.345 | 0.246 | 0.964 | 0.907 | 1.025 |
Trait anger | 0.106 | 0.066 | 2.559 | 0.110 | 1.111 | 0.977 | 1.265 |
Constant | 1.989 | 1.401 | 2.016 | 0.156 | – | – | – |
Comparison 2: high gain/high loss versus low gain/low loss |
Gender | − 3.936 | 0.995 | 15.640 | 0.000 | 0.020 | 0.003 | 0.137 |
Race | 1.347 | 0.715 | 3.549 | 0.060 | 3.846 | 0.947 | 15.618 |
Trait anxiety | − 0.087 | 0.044 | 3.867 | 0.049 | 0.917 | 0.841 | 1.000 |
Trait anger | 0.263 | 0.081 | 10.506 | 0.001 | 1.301 | 1.109 | 1.525 |
Constant | − 0.491 | 1.880 | 0.068 | 0.794 | – | – | – |
Significance is only observed in the second comparison with three predictors being statistically significant: gender, trait anxiety and trait anger. The odds of a male choosing the high gain/high loss option over the low gain/low loss was 50 times higher (1/0.020). The odds of participants with lower anxiety choosing the high gain/high loss option over low gain/low loss were 1.1 times higher (1/0.917). However, the confidence interval included the value of 1. This means that not much emphasis should be placed on this significant finding because 1 is “the threshold at which the direction of the effect changes” (Field
2013: 786). Lastly, the odds of choosing high gain/high loss (over low gain/low loss) increased by about 1.3 as one moved from any score in the anger measure to the next higher score. These results somewhat mirror those seen in Table
4 when choosing to invest in equity i.e. that a higher anger measure is correlated to higher risk tolerance.
Although 18 participants had selected the ‘sell at a loss’ category for the outcome variable in scenario 3 (Table
5), this category had too few data points to be used in the relevant multinomial logistic regression. Consequently, only the two remaining categories were analysed in a binary logistic regression. Table
8 provides the results.
Table 8
Logistic regression predicting likelihood of waiting some weeks with the possibility to either lose or advance even further (N = 211)
Gender | 1.081 | 0.317 | 11.626 | 0.001 | 2.948 | 1.583 | 5.488 |
Race | 0.402 | 0.326 | 1.519 | 0.218 | 1.495 | 0.789 | 2.835 |
Trait anxiety | − 0.051 | 0.021 | 6.046 | 0.014 | 0.950 | 0.912 | 0.990 |
Trait anger | 0.044 | 0.035 | 1.588 | 0.208 | 1.045 | 0.976 | 1.118 |
Constant | 0.049 | 1.007 | 0.002 | 0.961 | 1.050 | – | – |
Two variables are statistically significant: gender and trait anxiety. In the case of gender, the odds ratio of 2.948 indicated that the odds of waiting some weeks with the possibility to either lose or advance even further were almost 3 times higher for males. These results are supported in literature which illustrates the prevalence of gambling being heightened in men (Hing et al.
2016), This is substantiated by men’s heightened ability to withstand risk (Willows and West
2015).
The odds of participants with lower anxiety waiting some weeks (rather than some days) was 1.1 times higher (1/0.950). In a laboratory-based delay-discounting task in which participants made choices between electric shocks delivered immediately rather than after various time delays, Salters-Pednault and Diller (
2013) found that participants with higher levels of anxiety were more likely to make the choice to delay, despite knowing that it would result in a worse outcome. The results in Table
8 show the opposite: that lower anxiety predicts the ability to wait. However, in this paper, the outcome is a gamble, rather than a guaranteed worse outcome. Furthermore, these results should be interpreted with caution given the upper confidence level of 0.990.