Does a Financial Education change Gender Risk Aversion?

Transcription

1 Does a Financial Education change Gender Risk Aversion? Michael Naylor*, Wendy Hsu and Brenda Allen-Browne School of Economics & Finance Massey University, Palmerston North, Private Bag Palmerston North New Zealand * Corresponding Author: m.j.naylor@massey.ac.nz Version 1.1: 25 th August 2016 i

2 Does a Financial Education change Gender Risk Aversion? Abstract We are the first paper to test if the well-established gender difference in optimism and lower risk aversion applies to financially educated individuals. We find that educated investors and uneducated investors perceive risk differently, and that female uneducated investors are likely to perceive the risk of a range of financial assets as less risky than the overall financial advisers while the male uneducated investors are likely to perceive the risk as higher. Demographic characteristics such as age, non-financial education level, overall wealth, and ethnicity were also found to be significant. However neither group consistently used solely objective measures. We also show that during the GFC both genders and education levels acted in a similar fashion, although those who described themselves as aggressive investors took the crisis as an opportunity while those who were the most highly educated and individuals who worried about their financial assets, took action based on concern. ii

3 Introduction Risk, like beauty, could be said to be in the eye of the beholder. It is well established behavioural finance heuristic that male retail investors are more optimist about potential investment outcomes than female retail investors, as well as having a lower risk aversion. Eckel and Grossman (2007) and Croson and Gneezy (2004) show that the views of males and females differ on a range of topics, and that most of these differences can be attributed to gender differences in risk-aversion. However, risk tolerance differs from risk perception. Risk tolerance can be described as an investor s willingness to accept a given level of risk for the achievement of a goal or objective. Risk perception is however unique to the individual and is the risk the individual believes is contained within (and the consequences of) a certain situation or decision whether real or not (Ricciardi, 2008). This is a function of context both of setting and time. Weber, Blais and Betz (2002) found differences in risk taking are highly domain-specific and relate to perception of the risk rather than a consistent attitude toward risk. Roszkowski and Davey s (2010) found that while risk tolerance was generally a stable characteristic of an individual s personality, is it able to be influenced by situations or by a change in circumstances. They report individual s risk tolerance scores were surprisingly largely unchanged pre- and post-gfc 2008, but that 74% of survey respondent s perceived the stock market as having become riskier. If risk-perception can be changed by increased financial knowledge then there should be a differentiation between the risk perception of retail investors and financial advisers. Shapira and Venezia (2001) found that while both professionals and independent investors exhibited the disposition effect, the professionals training and experience reduced this bias. Understanding if financial advisers and investors think about risk in a similar way is critical to ensuring communication about the often complex financial choices consumers face. For example; do both financial advisers and clients discuss risk tolerance do they share a common understanding? A professional may think of risk as standard deviation while a client may think of loss and risk as being interchangeable terms. This study seeks to understand if the New Zealand Authorised Financial Advisers (AFA) and retail investors perceive risk differently and if there is a gender disparity in their risk perception; if any demographic characteristics are related to the risk perception of both/either group; if both groups implicitly use objective risk measures; if the behaviour of both/either during and after the GFC is related to any demographic characteristics; if the groups differ in their views on the influences upon 1

4 investor risk tolerance; and if the groups differ in their views on the influences upon investor s acceptance of financial advice. Literature Review Ricciardi (2004) notes an individual s risk perception can be changed by changing their knowledge level, and that experts and novices differ in their assessment of risk across a range of activities. He also notes that more credibility is given to information obtained from trusted sources. By implication as a novice moves to become an expert, their perception of risk is likely to change, becoming more objective than subjective. The acquisition of knowledge and change in risk perception should be gender neutral as perception becomes more objective. Research is, however, limited on the comparison of risk perception between experts and novices, and whilst there are many studies on gender differences in risk perception and adversity, results are mixed. Both have implications for communication between financial advisers and their clients. Olsen (1997) surveyed American Chartered Financial Analysts and clients and suggested that individuals are loss-averse rather than risk-averse. He categorised the first survey results into four attributes of risk; the potential for capital loss, returns below expectations, controllability of loss, and level of knowledge. He found that the level of knowledge was not significantly related to any of the risk attributes. Diacon (2004) compared licensed financial advisers with individual investors and found that financial advisers reported lower average risk scores than lay investors. The most important factors for experts were dislike of volatility and losses and lack of trust in product/provider. Comparatively, for the layperson uncertainty and loss adversity remain distinct factors, and their third factor is lack of knowledge. From this analysis Diacon finds significant differences in how financial advisers and the layperson construct their risk perception and suggests advisers may be prone to trusting product providers, consider products as less complex, and are have more trust in regulators. Jansen, Fischer and Hackethal (2008) found that the majority of advisers surveyed underestimated their clients risk aversion when compared to the client s self-reported risk aversion rating. Clark- Murphy and Soutar (2008) found the finance professionals underestimated those clients selfcategorised as risk averse and that there was also a clear mismatch in the importance ranking of most attributes of possible investments. Roszkowski and Grable (2005) found no significant difference in the standardised risk tolerance between expert and lay groups. However male clients score being marginally overestimated by advisers while female clients scores were significantly underestimated. 2

5 Most gender risk-aversion research shows that women are consistently more risk averse than men (e.g.; Barber and Odean, (2001), Croson & Gneezy, (2009), Charness and Gneezy (2012)). Nelson (2013) argues, however, these studies often do not allow for within-group variation and mistakenly treated gender differences in risk aversion as an individual categorical variable. She argues that the data is reconsidered on this basis many of the previous significant results were found to be the result of within-group variations rather than there being a significant difference in average results. The evidence therefore remains mixed and many studies use data collected from students rather than actual retail investors. Others like Haigh and List (2005) used brokers, clerks and exchange employees rather than financial advisers. The relevance of these studies is therefore limited. Olsen and Cox (2001) surveyed Chartered Financial Analysts and 274 Chartered Financial Planners and found that while both men and women ranked the most important risk attribute as the risk of large losses, women s score showed a higher level of concern. Compared to Men, Women ranked uncertainty second and were less concerned with the risk of earning less than expected. Woman also viewed the lowest risk and highest risk assets (insured savings account and long term Treasury bonds, and new firm IPO s) as riskier than men, with a significant difference in perception. Olsen and Cox suggest this is consistent with women s concern over downside risk as the low-risk assets have limited upside opportunity, and women may consider IPO s have both higher potential for loss combined with greater uncertainty. Women also allocated a higher portion of funds to growth assets than men. Olsen and Cox note while this appears at odds with their earlier findings they suggest the result is related to the higher risk women ascribe to low risk assets. They conclude their study by suggesting gender differences may only become important when considering portfolios at each end of the risk spectrum. Bliss and Potter (2002) surveyed professional fund managers and found female managers held a slightly riskier portfolio of assets, with the result consistent across three separate objective risk measures so were less risk averse than their male colleagues. There was no gender difference in the trading turnover of domestic equity funds, but a large difference in international funds where women trade less frequently. It is noted however that the average international fund size for women is one third of their male counterparts which may influence the fund turnover. While female managers showed outperformance over male managers, this disappears on a risk-adjusted basis. Atkinson, Boyce-Baird and Frye s (2003) found no gender difference for fixed-income fund managers. Niessen and Ruenzi (2006) studied equity fund managers and found that while female managers women take less unsystematic and small firm risk, their overall risk is not significantly different to men. They also found women pursue significantly less extreme investment styles and their style is more consistent over time. 3

6 Becknamm and Menkhoff (2008) argue that fund managers familiarity with risk mitigates any initial gender-specific bias. They surveyed fund managers across four countries and found mixed results. Italian and Thai female managers were significantly more risk averse in two of the three behaviours, while German women managers only showed a significantly lower score in the disposition effect. US managers exhibited no gender differences. Overall two key points arise. First, there is limited research on the potential risk perception differences between advisers and the layperson. Second, while there have been numerous studies exploring potential gender differences, most relate to risk tolerance and behaviour rather than risk perception. The purpose of this study is to investigate; if differences in risk perception exist between professionals and lay-person investors, and if gender differences are apparent within and/or between the two groups under study. Methodology and Hypotheses Data in this study was collected by way of an online survey distributed to Financial Advisers (AFA) and AFA clients. The survey link was widely distributed to advisers nationwide and via industry groups. There were a total of 514 responses over July to August After eliminating incomplete responses from both groups (n=104), the sample size was reduced to 244 AFA responses (12.8% of total AFA population) and 166 non-afa responses. This compares favourably with Diacon s study which comprised 41 experts and 123 lay investors. The survey questions provided to the two groups had small differences: the non-afa survey included a set of financial literacy questions, while AFA s were asked about professional qualifications. The survey consisted of four sections: Section 1: Respondents demographic characteristics, financial literacy score, and perception of the risk of 20 different asset classes. Section 2: Experimental section - using methodology per Veld & Veld-Merkoulova (2008). Each respondent was asked to select a random symbol and from this was allocated a set of questions. There were a minimum four and maximum of eight questions depending on answers given. The alternatives presented all had the same expected returns but varied in their outcomes and risks. The first set of questions sought to understand the respondents implicit preference for risk measure, while the second set of questions sought to understand which benchmark was preferred for respondents who choose an asymmetric risk measure. Section 3: Respondents feelings and behaviour in response to the Global Financial Crisis of 2008, following methodology as suggested by Soderberg & Wester (2012). If risk perception is a function of context, the severity of the GFC provides a unique opportunity to understand how an 4

7 individual s risk perception influences their behaviour in such a situation. Survey participants were asked about their actual behaviour following this event. Section 4: Respondents perception of how influential various characteristics are on, (a) an investor s risk tolerance, and (b) the acceptance of advice from a financial adviser. Hypotheses H1: The average self-assessed risk tolerance rating of an AFA is equal to that of the layperson. Whilst the first hypothesis relates to the individual s risk tolerance rather than perception rating, this will be used as a baseline to understand how the individual views themselves. The selfassessed risk rating was on a four-point scale from 1 = conservative to 4 = aggressive. H2: There is no difference between the AFA and layperson s perception of the risk of 20 different asset classes. For each of the financial assets, respondents were asked to rate the riskiness of the asset using a scale ranging from 1 = very low risk to 5 = very high risk, or 0 = I am not familiar with this. One difference of this study to Diacon s is here all respondents were asked their perception of all 20 financial assets. H3: a) There is no difference between the male and female AFA s perception of the risk of 20 different asset classes b) There is no difference between the male and female layperson s perception of the risk of 20 different asset classes. H4: The demographic characteristics of the AFA and layperson are unrelated to their risk perception score. Two separate regressions are run using different dependent variables: Dependent variable 1: The respondents own self-assessed risk score. Dependent variable 2: The respondents Z-score, following methodology of Sachse, et al (2012). Respondents Z-scores were calculated in order to identify differences based on individual perception of risk of the 20 asset classes. The Z-scores were then averaged for an overall score per respondent. Where a response was noted as I am not familiar with this, it was excluded from the Z-score calculation. A positive Z-score indicates respondent is more risk averse than the group average, a negative Z-score indicates respondent is less risk averse than the group average. 5

8 Independent variables: Demographic variables of interest are those commonly tested for their potential relationship to risk perception: Gender: male = 1, female = 0 Age: separated into the following age bands 20-29, 30-39, 40-49, and 60 and over, respectively Age_1 to Age_5 Marital status: Married/de facto = 1, single, divorced, widowed = 0 Dependents: financial dependents including children, partner or parents, yes = 1, no = 0 Ethnicity: white = 1, all other ethnicities = 0 Education, highest level completed, from Ed_1 - Secondary School qualification to Ed_5 Postgraduate, Masters or PhD Net household income annual household income from all sources, separated into the following bands less than $50,000, $50, ,999, and over $150,000, respectively HH_Inc 1-3. Household financial assets excluding equity in property separated into the following bands less than $50,000, $50,000 - $149,999, $150,000 - $299,999, $300,000 - $499,999 and over $500,000, respectively HH_Fin_Assets 1 5. Home ownership: yes = 1, no = 0 In addition, the following variables will be tested for potential relationship: Investment experience: proxy will be the number of different asset classes owned in the past 5 years. A greater exposure and experience of a range of investment assets may influence an investor s perception of risk as suggested in Hilgert, Hogarth & Beverly (2003). Investment review period: The investment review period favoured by the investor could potentially indicate their level of concern for their investment performance or simply their investment style. Financial literacy score (non-afa s only); Using standard questions. Lusardi and Mitchell (2011) found individuals who are financially literate are more likely to have successful retirement outcomes, and Deaves, Veit, Bhandari & Cheney (2007) study that found college employees who had a propensity to plan for their retirement were also more risk tolerant. H5: There is no difference between the AFA and the layperson in their implicit use of objective risk measures. The experimental section of the survey effectively provides the survey respondent three choices for each question, the objective risk measure for that question, an alternative objective risk measure, or a subjective risk measure. In selecting one of three random symbols, the survey participant is allocated a theoretical sum of one of $1,500, $15,000 or $150,000 to invest in various options. All of the questions are designed that the mean return for every A/B choice is 8% while their standard 6

9 deviation lies between 10% and 26%. These figures are in line with long run real returns and volatility for the New Zealand stock marketa t-test is used to check for any significant difference in AFA vs layperson mean results. For those survey respondents who implicitly select specific objective risk measures, they were then directed to questions that sought to uncover their implicit benchmark. H6: There was no difference between the AFA and layperson s behaviour following the Global Financial Crisis (GFC) in The two sets of survey responses (AFA and layperson) were grouped into two subsets participants who made a choice to take action following the GFC, and those who did not. A binary logistic regression will be used to understand the variables that impacted on the decision to make a choice. H7: AFA s and laypeople do not differ in their view of the importance of factors influencing an investor s risk tolerance. For each of the factors that may influence an investor s risk tolerance, both AFA and laypeople respondents were asked to rate the importance of the factor using a scale ranging from 1 = no influence to 5 = large influence. Factors tested are investor s: age, gender, ethnicity, net wealth, net income or income needs, proposed investment timeframe, familiarity with a range of different asset classes, ethical considerations or constraints, prior positive or negative experience with shares or managed funds, and prior loss of investment capital. H8: AFA s and laypeople do not differ in their view of the importance of factors influencing an investor s acceptance of financial adviser s advice. Both AFA s and laypeople respondents were asked to rate to what extent various characteristics influence an investor s acceptance of a financial adviser s advice using a similar scale 1 = not at all to 5 = to a large extent. Characteristics tested are: adviser s age, adviser s gender, adviser s ethnicity, the investor s perception of the adviser s personal net wealth, advisers years of industry experience, advisor s highest qualification and the advisor s membership of a Professional Association. Results Descriptive statistics of the survey respondents are in Table 1 in the appendix. The average AFA is white, male, aged over 50 and married. This is somewhat different to the layperson group, and significantly different to the general NZ population in terms of ethnic diversity and gender. The income and asset levels of the AFAs surveyed were significantly higher than the layperson group. 7

10 Table 2 shows the self-assessed risk scores and perceived riskiness of assets. The mean AFA score of 3.37 is significantly higher than the mean layperson score of 2.48 (t=11.42, p<.0001). This indicates the AFA sample view themselves as more risk tolerant, with 43.4% of AFA respondents answering they see themselves as Aggressive investors compared with 10.3% from the layperson group. The null hypothesis 1 is therefore rejected. This does not necessarily mean AFAs are likely to recommend riskier investments to clients, but rather supports earlier research that found experts are generally more comfortable with risks in their area of expertise than a lay individual. Table 2 also shows that in 15 of the 20 asset classes the AFAs and laypeople differ significantly in the perceived riskiness of assets. This provides strong evidence the null hypothesis 2 can also be rejected. Table 2. Self-assessed risk score and perceived riskiness of financial assets: AFA vs Layperson. AFA No. of responses Mean Std Dev Layperson No. of respons es Mean Std Dev t Value Pr > t Self-assessed risk score *** <.0001 (range 1= Conservative, 4 = Aggressive) Perceived riskiness of various Asset classes (range 1= very low risk, 5 = very high risk) Savings account * Bank Fixed Term Deposit * Finance Company Fixed Term Deposit *** Bonus Bonds Diversified Managed Fund Kiwisaver or Superannuation Fund *** <.0001 Employer-contribution Superannuation scheme *** <.0001 NZ Corporate Bond rated BBB or lower NZ Corporate Bond rated BBB+ or higher *** NZ Government Stock *** <.0001 Shares in a NZ listed company IPO of a Company * Private Equity investment *** <.0001 Shares in an overseas listed company ** Precious metals (eg Gold) *** <.0001 Foreign currency *** <.0001 Investment-linked insurance policy Art or Antiques *** Residential investment property *** <.0001 Non-residential investment property *** Note: * statistically significant at the 0.05 level (2-tailed) ** statistically significant at the 0.01 level (2-tailed) *** statistically significant at the level (2- tailed) The direction of the mean difference is important to note. In 12 of the 15 significantly different results the AFA means are higher, that is AFAs perceive the assets as riskier than the layperson. Only 8

11 for Corporate Bonds rated BBB+ or higher, Government Stock and shares in an overseas listed company were the layperson means higher than that of the AFA. This is the opposite result to that found in Diacon s (2004) study where the expert group consistently reported lower scores than lay investors. This is also at odds with the view that experts are more comfortable with the risks in their particular field of expertise, leading to an expectation is their mean riskiness scores would be lower. Possible explanations could be that the AFA sample is more risk averse than the layperson sample or that the lack of knowledge by layperson leads them to under-estimate risk. The first explanation is unlikely given the earlier results noted of the significantly higher AFA selfassessed risk tolerance scores. Evidence the knowledge of the lay sample is lower of some assets in the layperson group is seen by 27-31% of lay respondents answering I am not familiar with this when asked about their perception of some assets 1. The second possibility will be explored in Hypothesis 4. Of particular interest is the difference in the perceived risk of residential property investment. Results show this asset type displays one of the largest differences in perceived risk (t=6.000, p<.001). Interestingly this may help explain New Zealand investors love-affair with property as financial experts who may view the risk of residential property investment as significantly higher than the layperson investor. The implications of a mismatch in understanding the risk level of financial assets are considerable. If AFAs and lay investors do not have common perception of risk, how can an adviser ensure they have matched the investor to a financial product that best suits their needs and risk profile? If the layperson does not consider the assets to be particularly risky, will they be prepared to pay for advice about such products? Risk Perception by Gender Table 3a shows the self-assessed risk scores of both the AFA and layperson samples, separated by gender. Whilst the mean self-assessed risk score for female AFAs is higher than that of male AFAs, the difference is only marginally significant. In the layperson sample, the mean female score is lower than the male score, but it is not significant. Table 3b in the appendix shows male vs female AFA view of the riskiness of the same 20 financial assets. There are only two of the 20 assets showing a significant difference (to 5%) in the mean scores when results are separated by gender. Given the vast majority of results show no significant 1 this was in spite of there being explanatory information provided in the survey 9

12 difference and the two that do are significant to 5% only, it is reasonable to conclude the null hypothesis H3(a) is supported. Table 3a. Self-assessed risk score, comparison of Female and Male scores. Female Male (range 1= Conservative, 4 = Aggressive) No. of responses Mean Std Dev No. of responses Mean Std Dev t Value Pr > t AFA Layperson Table 3c shows the male vs female layperson results of the mean comparison for the same 20 financial assets. These results are completely different to the male/female AFA outcome. In all bar two of the asset classes there is a significant difference between the mean responses of males and females. Further, in 16 of the 18 significantly different results, the differences the female mean score was less the male score, indicating the female layperson considers the riskiness of the asset to be lower than that of the male layperson. Given the similarity of the male and female self-assessed risk scores (Table 3a), this difference is unlikely to be attributed to women being more risk averse. Table 3c. Perceived riskiness of financial assets: Layperson Female vs Male Female Layperson Male Layperson (range 1= very low risk, 5 = very high risk) No. of Std No. of Pr > responses Mean Dev responses Mean Std Dev t Value t Savings account Bank Fixed Term Deposit ** Finance Company Fixed Term Deposit Bonus Bonds *** <.0001 Diversified Managed Fund *** <.0001 Kiwisaver or Superannuation Fund *** <.0001 Employer-contribution Superannuation scheme *** <.0001 NZ Corporate Bond rated BBB or lower *** <.0001 NZ Corporate Bond rated BBB+ or higher *** <.0001 NZ Government Stock *** Shares in a NZ listed company *** <.0001 IPO of a Company ** Private Equity investment ** Shares in an overseas listed company *** <.0001 Precious metals (eg Gold) *** <.0001 Foreign currency *** <.0001 Investment-linked insurance policy *** <.0001 Art or Antiques *** <.0001 Residential investment property *** <.0001 Non-residential investment property *** <.0001 Note: * statistically significant at the 0.05 level (2-tailed) ** statistically significant at the 0.01 level (2-tailed) *** statistically significant at the level (2- tailed) 10

13 Table 3d shows differences between male laypersons and all AFAs. This shows that there are significant differences in the mean results for 10 of the 20 financial assets, in that male laypeople perceive these financial assets as significantly riskier than the AFA sample. Our conclusion is that the female layperson scores have considerably reduced the overall layperson mean results to the extent that the distinct male responses are no longer discernible. Table 3d. Perceived riskiness of financial assets: AFA vs Male Layperson Note: (range 1= very low risk, 5 = very high risk) * statistically significant at the 0.05 level (2-tailed) ** statistically significant at the 0.01 level (2-tailed) *** statistically significant at the level (2-tailed) No. of responses AFA Mean Male Layperson Table 3e shows that the results of a regression on the layperson dataset using the financial literacy score as the dependent variable and Gender as the independent variable. This was statistically insignificant. Regression was also run adding the alternative proxies for financial knowledge; which show that Education and Investment are significant. Gender also becomes marginally statistically significant at 10% level. Since the male gender beta is negative females have more knowledge than males. This result does not support the notion that the female layperson has lower financial knowledge so may perceive financial assets to be less risky than the male layperson. Past research has generally found women are more risk averse than men and this extends to both finance professionals and the layperson. This study found this to be not the case for the New Zealand AFA. Male and female risk tolerance scores were only marginally significantly different. Our expectation Std Dev No. of responses Mean Std Dev t Value Savings account Bank Fixed Term Deposit Finance Company Fixed Term Deposit Bonus Bonds ** Diversified Managed Fund *** Kiwisaver or Superannuation Fund Employer-contribution Superannuation scheme * NZ Corporate Bond rated BBB or lower ** NZ Corporate Bond rated BBB+ or higher *** <.0001 NZ Government Stock *** <.0001 Shares in a NZ listed company *** <.0001 IPO of a Company Private Equity investment Shares in an overseas listed company *** <.0001 Precious metals (eg Gold) Foreign currency Investment-linked insurance policy *** <.0001 Art or Antiques Residential investment property Non-residential investment property ** Pr > t 11

14 that risk perception, particularly among financial experts is gender neutral and results are in line with this. Table 3e. Multiple regression of financial knowledge proxies on Financial literacy score (n = 166) Dependent Variable: Fin_Lit_Score Beta SE Pr > t Intercept 2.11*** <.0001 Gender Inv_cnt Edu 0.13*** <.0001 R-Square Adj R-Sq *** statistically significant at the level (2-tailed) A different picture emerged when comparing male and female layperson scores. The result found the somewhat surprising result that women had significantly lower scores than men indicating the female layperson from this sample considered most financial assets to be less risky than their male peers. This is an unexpected result considering past research has found women to be either more risk averse. As significant differences in AFA and male layperson perception scores were found this may indicate further research is required to understand if the disparity in layperson risk perception scores is due a gender mismatch in financial knowledge or other underlying factors. Multivariate Regression Analysis: Self-Assessed risk score Table 4a presents a multiple regression of the self-assessed risk score regressed on three models of explanatory variables described earlier. Model 1 includes all 18 independent variables and produced a statistically significant R 2 of Of the five variables that were significant only the Layperson variable is negative. This is in line with the earlier finding that the layperson mean risk score is significantly lower than the AFA s. The other variables; Age, Education, Investment count and Benchmark, all have a positive sign which indicates at their value increases, so will the expected risk score. The Education variable may also be a proxy for AFA status as AFA s in this sample have a higher education level that the layperson sample. The Investment count variable is used as a proxy for investment experience so is expected to be positively related to risk score. The positive relationship with the Benchmark variable indicates that those measuring their investment performance against either a risk-free or market 12

15 rate are expected to have a higher risk score than those who use the original investment value as their benchmark. The positive relationship between the Age variable and SA risk score indicates as age increases, so does the individual s risk score. This would appear at odds with both past literature (e.g.; Sachse et al. 2012) and the commonly held belief that individuals become more conservative as they age. However, the dataset includes both AFA and laypeople and further analysis of the data reveals 35% of the AFAs aged 50+ noted their risk score as 3 or 4 (Balanced or Aggressive). The basis of this result may therefore have been influenced by the subset of AFA data. A further regression run using just the layperson dataset (Table 4b. Appendix B) shows that whilst layperson Age also has a positive relationship to SA risk score, the variable is not significant (p=0.169). The only significant independent variables in the amended model are Education and Benchmark. Table 4a. Multiple regression of demographic characteristics on SA Risk (n = 410) Dependent Variable: SA_Risk MODEL1 MODEL2 MODEL3 Beta SE Pr > t Beta SE Pr > t Beta SE Pr > t Intercept 2.29*** < *** < *** <.0001 Layprsn -0.70* *** < *** <.0001 Gender Age 0.06* * * Married Dep Ethnicity Edu 0.11** ** *** CFP Inv_cnt 0.04* ** ** B_mark 0.10* ** ** Review Z_Score HH_Fn_As Home_own HH_Income Yrs_workd Yrs_advice Fin_lit_score R-Square Adj R-Sq Note: * statistically significant at the 0.05 level (2-tailed) ** statistically significant at the 0.01 level (2-tailed) *** statistically significant at the level (2-tailed) 13

16 In Model 2 the number of independent variables was reduced to nine following a backward selection process, eliminating highly correlated variables, eg HH Financial Assets and HH income (correlation coefficient 0.393, p<.001). Variables that related to AFAs only (CFP, Years worked and Years advice) were also removed as the first model showed these had no significance. Model 2 also produced a statistically significant R 2 of showing removal of some of the insignificant and/or correlated variables did not change the model strength. The five significant variables from Model 1 remained so, but the Layperson variable became significant to the 0.1% level, while the Investment count and Benchmark variables became significant to the 1% level. In Model 3 the number of independent variables was reduced further, eliminating those that weren t significant except Home Owner as this had p=0.077 in Model 2 so may have become significant once other variables were removed. The original five variables remained significant to the same level and the R 2 remained with Home Owner variable remaining insignificant. The null hypothesis that demographic variables are unrelated to an individual s self-assessed risk score is therefore rejected. Multivariate Regression Analysis: Z-score Table 4c presents a multiple regression with the individual s Z-score regressed on three further models of explanatory variables. Using the Z-score enables the combined dataset of AFA and layperson responses to be used as the standardised Z-score provides comparison of the individual to their wider peer group. Model 4 again uses all 18 independent variables and produced a statistically significant R 2 of 0.195, but this is considerably lower than that of the earlier regression. In this model the only statistically significant variables are Gender, Ethnicity and Benchmark. The earlier significant variables of Age and Investment count no longer feature and the Education variable is marginally significant (p=0.056). The Gender and Benchmark betas are positively signed indicating as these increase so does the Z score. The positive Gender beta indicates as the individual variable changes from female to male their Z-score increases. A higher Z-score indicates the survey respondent viewed the range of 20 financial assets as more risky than their average peer group. Whilst this is not in line with the results of Table 3a it does perhaps help to explain the results from Table 3d. From Model 4 we are unable to see if this applies to all males or just to the male lay investor as identified in Table 3d. 14

17 Table 4c. Multiple regression of demographic characteristics on Z_Score (n= 410) Dependent Variable: Z_Score MODEL4 MODEL5 MODEL6 Beta SE Pr > t Beta SE Pr > t Beta SE Pr > t Intercept * ** Layprsn *** *** Gender 0.36*** < *** < *** <.0001 Age Married Dep Ethnicity -0.15** ** ** Edu * * CFP * * SA_Risk Inv_cnt B_mark 0.05* * * Review HH_Fn_As * Home_own HH_Income Yrs_workd Yrs_advice Fin_lit_score R-Square Adj R-Sq Note: * statistically significant at the 0.05 level (2-tailed) ** statistically significant at the 0.01 level (2-tailed) *** statistically significant at the level (2-tailed) The Ethnicity variable is negatively signed indicating as the individual variable changes from nonwhite to white their Z-score decreases. This implies those of non-white ethnicity consider the 20 financial assets less risky than their white peers. This is at odds with the so-called white male effect (Finucane, Slovic, Mertz, & Satterfield, 2000) which found white males perceive risks to be lower than women and non-white males. In Model 5 the number of independent variables was reduced to 10 following a backward selection process, eliminating highly correlated variables together with those that were insignificant in Model 4. Model 5 increased the statistically significant adjusted R 2 to showing removal of some of the insignificant and/or correlated variables improved the model strength. The change in model increased the number of significantly related independent variables from three to six, with Layperson, Education and CFP becoming significant. 15

18 The significance in the Layperson variable in combination with Gender variable (both positively signed) indicates a male layperson considers the 20 financial assets as being riskier than do all other sub-groups of the sample. This clearly supports the results summarised in Table 3d. The newly significant CFP variable, positively-signed, indicates those among the AFA group who are among the industry s most qualified, consider the financial assets listed to be riskier than their non- CFP AFA peers. Whilst the beta is small at 0.11 it is significant none-the-less and raises the question as to why this experienced group of financial advisers hold this view. Further data analysis reveals CFPs comprise nearly half (48.8%) of the AFA respondents and are spread across age and income bands. However, nearly 60% of CFPs have household financial assets of over $500,000 so their implied increased risk adversity may be driven by a desire to take less risk with their accumulated wealth. In Model 6 the number of independent variables was reduced further, eliminating those that weren t significant from Model 5 except HH Financial Assets. The six variables from Model 5 remained significant to the same level and HH Financial Assets was found to have a positive statistically significant relationship to Z-score. The model s adjusted R 2 reduced only slightly to The null hypothesis that demographic characteristics are unrelated to an individual s Z-score is therefore rejected. This section sought to understand if an individual s characteristics, beyond gender, were related to their risk tolerance. This may assist advisers understand of which client s characteristics link to risk sensitivity. Experimental section analysis The experimental part of the survey ascertained respondents actual risk behaviour from randomised investment trial questions. The summarised results of the first part of the experimental questions are presented in Table 5a. The number of responses is important as it shows what proportion of respondents made a definite choice based on the limited information provided. The four possible answers provided are: 1.Fund A, 2.Fund B, 3.Both funds are equally attractive/unattractive to me or 4.I do not have enough information to make a choice. Four objective risk measures are implied in the choices provided in questions 1 4. These are probability of loss, semi-variance, expected value of loss and total variance respectively. For each of these initial questions, one objective risk measure has been calculated as offering a lower risk than the other three risk measures, whilst holding return to 8% and standard deviation to a range 16

19 between 15 and 26. The questions are based on the assumption the rational investor will seek to minimise their risk. In Question 1 for example, Fund A has a probability of loss of 10% whilst Fund B has a probability of loss of 40%. A rational investor solely using objective risk measures is expected to therefore choose Fund A. Table 5a. Summary of responses to experimental questions, AFA and Layperson Fund A Fund B AFA Both funds equal Insuf. Info Total resp. Fund A Fund B Layperson Both funds equal Insuf. Info Total resp. Q1. 1 = probability of loss of most concern, 2 = other risk measures Q2. 1= semi-variance is of most concern, 2= other risk measures Q3. 1= expected value of loss of most concern, 2= other risk measures Q4. 1= total variance of most concern, 2= other risk measures Q5. Benchmark: 1= market, 2= risk free rate or zero Q6. Benchmark: 1= market or risk free, 2= zero Q7. Benchmark: 1= semi-variance of returns relative to the market, 2= risk free rate or zero Q8. Benchmark: 1= semi-variance of returns relative to the market or risk free rate, 2= zero 35 (14.3%) 55 (26.3%) 20 (12.9%) 11 (8.1%) 33 (58.9%) 30 (54.5%) 30 (54.5%) 23 (41.8%) 96 (39.3%) 56 (26.8%) 34 (21.9%) 39 (28.9%) 19 (33.9%) 21 (38.2%) 20 (36.4%) 27 (49.1%) 32 (13.1%) 25 (12.0%) 30 (19.4%) 21 (15.6%) 3 (5.4%) 4 (7.3%) 4 (7.3%) 5 (9.1%) 81 (33.2%) (21.1%) (34.9%) (31.6%) (45.8%) 155 (13.9%) (47.4%) 7 (8.4%) (1.8%) (60.9%) (0%) (34.8%) 1 (1.8%) 0 (0%) (60.5%) (44.4%) 63 (38.0%) 25 (15.1%) 43 (25.9%) (31.6%) (14.5%) (22.2%) (30.4%) (8.9%) (46.8%) (34.9%) (10.8% (45.8%) (34.8%) (2.2%) (2.2%) (54.3%) 5 (10.9%) 0 (0%) 13 (34.2%) 19 (52.8%) 1 (2.6%) 0 (0%) 1 (2.6%) 1 (2.8%) Table 5a shows for the first two questions approximately 53% of the AFA survey respondents are willing to make a definite investment choice between Fund A and Fund B, with the balance replying that either both funds are equally attractive/unattractive or that they did not have enough information to make a choice. For Questions 3 and 4, those AFA willing to make a definite choice drops to 34.8% and 37% respectively. The layperson results are somewhat similar to the AFA results for the first two questions, with 59.1% (Q.1) and 63.2% (Q.2) of respondents making a choice between Fund A and Fund B. For Questions 3 and 4, the proportion of lay investors making a definite choice drops to 44.3% and 43.3% respectively but these are considerably higher than the proportion of AFA responses. This would imply that the layperson is more comfortable making a definite investment choice based on implicitly objective risk measures than the AFA in this survey. This is an interesting result when the AFA is expected to be more objective in their consideration of risk compared to the layperson. Both the AFA and layperson responses however are significantly different to the response rates seen in the Veld and Veld-Merkoulova (2008), who found that between 66.5% % of respondents made a definite choice of either Fund A or Fund B for Questions 1-4. Further, less than 17

20 5% of respondents answered Questions 1-4 with option 4. This compares to between 22.2% % of respondents in this survey. This level of non-choice indicates a significant proportion of this sample do not base their investment choices on purely objective risk measures. For those respondents who chose an objective risk measure, follow up questions to discover their implicit preferred benchmark were then provided. For those who chose Fund A in either Question 1 or 3, they were given Questions 5 and 6. These questions sought to understand if the benchmark preferred is the market return, the risk-free rate or zero Table 5a shows that more than 50% of the AFA respondents implicitly prefer the market return as a benchmark (Q5.) instead of the original value of the investment. While the layperson sample indicates a preference for the market return benchmark (Q %), in Q.6 respondents show 54.3% prefer the original value of the investment as a benchmark. Their preferred implicit benchmark is therefore inconclusive. For those respondents who chose semi-variance (Fund A, Q.2) or total variance (Fund A, Q.4) as their objective risk measure, follow up questions 7 and 8 were provided. The results in Table 5a show again more than 50% of AFA respondents implicitly preferred the market return as their benchmark (Q.7). For Q.8 however the choice between the implicit benchmarks slightly favoured the original investment value. In the layperson sample, the choice in Q.7 was clear with 60.5% of respondents implicitly preferring the market benchmark. The responses from Q.8 however do not reflect the same result with 52.8% of respondents implicitly preferring the original investment value as their benchmark. These results are similar to those of Veld and Veld-Merkoulova (2007). For the proportion of survey respondents who made definite choice between Fund A and Fund B, further analysis has been undertaken to see if the respondents favour particular objective risk measures. In this analysis if Fund A = 1 and Fund B = 2, then the expected mean of the random choice between the two is expected to be 1.5. Table 5b summarises these results. For Question 1, the mean response is statistically significantly different to 1.5 indicating the respondents consider other risk measures to be more important than the probability of loss. This result applies to both the AFA and layperson respondents. The same result is seen for both groups for Question 4 indicating total variance is less important than other risk measures. Question 2 mean is not significantly different to 1.5 for either group and the mean in Question 3 is only significantly different to 1.5 for the layperson group. Results for Questions 5 8 show only one significant result across both the AFA and layperson groups so supports the hypothesis that the observed mean is not significantly different to 1.5 for these four questions regarding implicit benchmark. 18