2.1 Financial literacy

Our study builds upon the body of research that connects financial literacy with mortgage choice (Campbell Reference Campbell2006; Lusardi and Mitchell Reference Lusardi and Mitchell2007a, Reference Lusardi and Mitchell2007b). Related studies show that borrowers with higher financial literacy pay lower mortgage interest charges (Cheng et al. Reference Cheng, Lin and Liu2011; Courchane et al. Reference Courchane, Gailey and Zorn2008; Guiso et al. Reference Guiso, Pozzi, Tsoy, Gambacorta and Mistrulli2022; Huston Reference Huston2012), avoid complex products and fees (Agarwal et al. Reference Agarwal, Ben-David and Yao2017; Bergstresser and Beshears Reference Bergstresser and Beshears2010; Gathergood and Weber Reference Gathergood and Weber2017b), report their financial situation more accurately, thus presenting a lower risk of delinquency (Garmaise Reference Garmaise2015), are less likely to withdraw home equity (Duca and Kumar Reference Duca and Kumar2014), have better operational knowledge of mortgage risks (Bucks and Pence Reference Bucks and Pence2008; Cox et al. Reference Cox, Brounen and Neuteboom2015; Van Ooijen and van Rooij Reference Van Ooijen and van Rooij2016), and make more efficient refinancing decisions (Bajo and Barbi Reference Bajo and Barbi2018). By contrast, low financial literacy is a barrier to mortgage market participation (Gathergood and Weber Reference Gathergood and Weber2017a) and predicts default (Gerardi et al. Reference Gerardi, Goette and Meier2013). In general, self-reports of over-indebtedness are negatively related to debt literacy (Lusardi and de Bassa∼Scheresberg Reference Lusardi and de Bassa Scheresberg2013; Lusardi and Tufano Reference Lusardi and Tufano2015).

Life-cycle models predict optimally higher borrowing for housing at younger ages because housing is an important component of consumption, and mortgages are an accessible source of credit. Similarly, wealthier, higher income and dual-wage earning households can carry higher debt loads and support housing consumption habits (Cocco Reference Cocco2005; Kraft et al. Reference Kraft, Munk and Wagner2018). Higher risk aversion may lead to precautionary savings and make people less comfortable with debt. Similarly, people who are more patient may borrow less and save more. We therefore expect that higher income, better educated, partnered, less risk averse, less patient and younger borrowers may be more comfortable with mortgage debt than those who are older or less well-off.

However, these same direct influences on mortgage borrowing also indirectly affect financial literacy. Since it can be costly to acquire financial knowledge, people who can benefit from greater financial sophistication tend to acquire it (Lusardi et al. Reference Lusardi, Michaud and Mitchell2017; Lusardi and Mitchell Reference Lusardi and Mitchell2014). The benefits of financial literacy are likely to outweigh costs for people with higher education and income. In addition, theoretical and empirical studies find that average levels of financial literacy exhibit a hump-shaped pattern with age, peaking in later life (Agarwal et al. Reference Agarwal, Driscoll, Gabaix and Laibson2009; Lusardi et al. Reference Lusardi, Michaud and Mitchell2017). Furthermore, past research shows that females are more inclined to answer “Don’t Know” to financial literacy questions and test lower on financial literacy assessments than males (Lusardi and Mitchell Reference Lusardi and Mitchell2011).

Hence, in our analysis of the relation between financial literacy and subjective comfort with mortgage debt, we condition on age, experience, education, income, marital status, gender, and risk preferences. A priori, higher financial literacy could make borrowers either more or less comfortable with debt. However, to the extent that financially literate borrowers consider a wider range of risks, they are likely to be more cautious (Bucks and Pence Reference Bucks and Pence2008; Cox et al. Reference Cox, Brounen and Neuteboom2015; Van Ooijen and van Rooij Reference Van Ooijen and van Rooij2016). We also predict that they will be more discerning with financial information (Engels et al. Reference Engels, Kumar and Philip2020) and therefore less susceptible to framing, and that they will be better at matching lump sum debt to equivalent repayment streams.

2.2 Mortgage brokers

We also add to understanding of the reasons why borrowers choose mortgage brokers and their impact on mortgage choice. The relation between financial literacy and financial advice seeking in general is debated (Calcagno and Monticone Reference Calcagno and Monticone2015, Hackethal et al. Reference Hackethal, Haliassos and Jappelli2012, Van Rooij et al. Reference Van Rooij, Lusardi and Alessie2011, Kramer Reference Kramer2016). But there is some evidence that less experienced (ASIC 2019) and less confident (Deloitte 2016) borrowers consult mortgage brokers. Adviser clients, including mortgage broker clients, choose brokers on trust-related factors such as past relationships or personal recommendations (ASIC 2017b), and first impressions (Agnew et al. Reference Agnew, Bateman, Eckert, Iskhakov, Louviere and Thorp2018).

Studies also disagree about the benefit or detriment of consulting a broker. Broker advice can reduce search costs and help with the mortgage process (Conklin Reference Conklin2017). However, agency problems, including racial factors, can mean that brokered mortgages are more expensive and have higher fees (Ambrose et al. Reference Ambrose, Conklin and Lopez2021; LaCour-Little Reference LaCour-Little2009; Robles-Garcia Reference Robles-Garcia2020; Woodward and Hall Reference Woodward and Hall2012). Likewise, Australian brokers have incentives to offer larger volume, longer-term loans that increase trail commissions (ongoing fees), and broker customers have reported feeling pressured to take large loans. Many customers say they are unclear about the way brokers are paid for their service (ASIC 2017b, Reference Badarinza, Balasubramaniam and Ramadorai2019; Deloitte 2016). Gennaioli et al. (Reference Gennaioli, Shleifer and Vishny2015) argue that investment advisers offer better performance but also provide “peace of mind” to investors. Thus when investment advisers reduce client anxiety and build trust, they can charge higher fees. Brokered mortgages have also been linked to higher rates of default than direct lending (Alexander et al. Reference Alexander, Grimshaw, McQueen and Slade2002; Elul et al. Reference Elul, Souleles, Chomsisengphet, Glennon and Hunt2010; Jiang et al. Reference Jiang, Nelson and Vytlacil2014).

We focus on three aspects of broker usage. First, educating the borrower in the home buying and financing process is a service that is normally performed by mortgage brokers (Conklin Reference Conklin2017). This education, in return, should reduce borrowers’ susceptibility to framing in the same way that financial literacy does. We therefore expect broker clients to be better able to calibrate lump sum and repayment stream debt. Second, the impact of consulting a mortgage broker on mortgage debt comfort is an empirical question. On the one hand, individuals who consult mortgage brokers are likely to be better informed about the risks of mortgages, and thus may be less comfortable with mortgage debt. On the other hand, mortgage brokers’ ability to give confidence to take risks, as well as broker remuneration structures which are based on the size of the loan, may lead broker clients to be more comfortable with higher loan debt. Third, Gennaioli et al.’s (Reference Gennaioli, Shleifer and Vishny2015) theory that advisers give investors confidence to take risk implies that people who perceive the housing market to be more risky are more likely to turn to brokers.

2.3 Choice architecture

We also add to understanding of choice architecture (Johnson et al. Reference Johnson, Shu, Dellaert, Fox, Goldstein, Häubl, Larrick, Payne, Peters and Schkade2012; Thaler and Sunstein Reference Thaler and Sunstein2008) on decisions about debt. We focus on how quantities that are economically equivalent are not always psychologically equivalent, this time for mortgages. Goldstein et al. (Reference Goldstein, Hershfield and Benartzi2016) show that people rate lump sum retirement savings as more adequate than the equivalent income (annuity) stream over some value ranges so that income framing encourages more saving. Zhang et al. (Reference Zhang, Wilcox and Cheema2020) show that a large lump sum of student debt can encourage more spending (less saving) than a smaller debt by making repayment goals seem out of reach. They conclude that framing student debt as future repayment obligations could deter students from adding to their loans.

Like Zhang et al. (Reference Zhang, Wilcox and Cheema2020), we investigate the framing of debt, but allowing for differences between the type and temporal features of student debt and mortgage debt. Student debt enables current consumption and will be repaid in the future from an uncertain post-education income – leading to a lack of reference point for the repayments, whereas mortgage debt is repaid immediately from current income and constrains consumption. Research has shown that people can more easily grasp the implications of wealth or debt when they can compare it with regular income (Goldstein et al. Reference Goldstein, Hershfield and Benartzi2016; Goda et al. Reference Goda, Manchester and Sojourner2014) or place it in a commonly used budgeting period (De La Rosa et al. Reference De La Rosa, Sussman, Giannella and Hell2022), or break it into smaller quantities (Basu and Ng Reference Basu and Ng2021; Hershfield et al. Reference Hershfield, Shu and Benartzi2020). We argue that lump sum presentations of home loans are more difficult to compare with regular income and thus will be psychologically larger than repayment streams. It follows that framing loans as lump sums instead of near-term repayment streams should reduce comfort with loans.

4.1 Task design

The survey collected both within- and between-subjects stated preferences for lump sum debt and repayment streams over a range of loan values. We set the range of loans from $200K to almost $3M in 10 steps, shown in Table 1. (All dollar values are Australian dollars AUD.) At each step, the lump sum and equivalent repayment increased by approximately 35%. In other words, the log loan amount rises linearly over the range. The amounts are rounded to lump sums in multiples of thousands and monthly repayments in multiples of fifties. The average new loan size for owner-occupied dwellings at the time of the survey in October 2020 was $537K (Australian Bureau of Statistics (ABS 2020)), an amount between steps 4 and 5 the range. To compute the equivalent monthly loan repayments, we assumed a loan term of 25 years and a mortgage interest rate of 2.9% p.a., which was the average mortgage rate applied to new loans for owner occupied dwellings in the third quarter of 2020 (Reserve Bank of Australia (RBA 2020)).

Table 1. Lump sum loan and monthly repayment values. This table shows the 10 lump sum debt sizes and equivalent monthly repayment amounts used in tasks 1 and 2. Values were approximately calibrated around average new loan sizes for owner occupied dwellings in Australia in 2020 with each set increasing the loan amount at a constant log linear rate of approximately 35%. Monthly principal and interest payments repayments are calculated for a 25 year loan term at an interest rate of 2.9% p.a. and shown in Australian dollars (AUD)

We randomly assigned participants to one of two tasks. In task 1, participants stated their level of comfort or discomfort with the range of 10 loan amounts framed either as lump sum debts or monthly repayments. Participants had to do this task for both frames, yet the survey randomized the order in which participants saw these frames. It also randomly allocated participants between conditions where underlying loan amounts either increased or decreased over the 10 levels. Figure 1 panel (a) shows the rating task for participants in task 1.

Figure 1. Tasks 1 and 2 screenshots. Panel (a) shows the lump sum condition for task 1 that collects participants’ perceived comfort with a mortgage debt. The monthly repayment condition substituted the words “monthly debt repayments” for “total mortgage debt” and “total debt”. The monthly repayment condition reported the monthly repayments numbers where lump sum debts appear. Panel (b) shows the lump sum condition for task 2 that collects the repayment amount that gave participants an equivalent level of comfort as a mortgage debt framed as a lump sum. Online Appendix A shows screen shots of the survey including the monthly repayment condition where participants gave the lump sum amount they felt provided an equivalent level of comfort as the repayment amount provided. Debt (repayment levels) ranged from $200K to $2979K ($950 to $14,000). Participants were randomly assigned to i) task 1 or task 2; ii) seeing either the lump sum or repayment framing condition first followed by the alternative; and iii) increasing or decreasing sequences of loan amounts. Each participant made 10 (loan amounts) × 2 (frame) ratings.

Task 2 presented a similar choice in a different form. In task 2, participants saw a lump sum debt (monthly repayment) and we asked them to use a slider to choose the monthly repayment (lump sum debt) amount on a 25 year loan that made them feel as comfortable or uncomfortable as the lump sum debt (monthly repayment). The survey again randomized the order of the frames and increasing versus decreasing amounts between subjects. Figure 1 panel (b) shows the rating task for participants assigned to task 2. In task 2, participants could experiment by moving the slider before confirming their choice by a tick box.

We asked participants to treat loan amounts as a sole residential mortgage, not an addition to current household debt. We also said that a 20% deposit had been paid, the term of the loan would be 25 years, and that the loan would be fully paid out in 25 years. This information ruled out features that might add to discomfort for some participants, such as borrowing for investment properties, interest-only loans and high loan-to-valuation mortgages that would require lender’s mortgage insurance. We did not assign an interest rate to the loan because we wanted participants’ subjective comfort rating and not their calculation of financial equivalence. These features of the tasks mirror the approach of Goldstein et al. (Reference Goldstein, Hershfield and Benartzi2016).

4.2 Sample

We put the online survey to field in October 2020 through a third-party commercial survey panel provider, Pureprofile. We first collected 105 responses for a pilot survey and then, after checking the pilot, we collected 999 responses from new participants. We set the total sample size so that 499 participants completed task 1 and 500 participants completed task 2. The 499 (500) participants in task 1 (2) were randomly assigned to different orders of framing and increasing or decreasing loan amount conditions.

Pureprofile sent invitations to potentially suitable panel members by email. On clicking a link, participants were invited to join the survey. The panel provider filled pre-set gender and age group quotas that matched Australian population characteristics. To ensure that we were collecting data from people who were genuinely interested in home loans, we chose participants from among Australian adults, aged between 25 years and 64 years (inclusive) and who earned over $52,000 per year. We also screened out otherwise eligible participants who, at the time of the survey, had never taken a mortgage and did not plan to take a mortgage in the future. Pureprofile compensated participants who completed the survey for their time (approximately $4). The majority of participants completed the survey in under 15 minutes, and the entire data collection process took around two weeks. Pureprofile collected responses until the survey quota was filled. They do not report standard response rates since not all panel members are invited to the survey. Table 2 reports sample descriptive statistics.

Table 2. Survey sample descriptive statistics. This table presents descriptive statistics for 999 survey participants, October 2020, n = 999. Population statistics are from the 2021 Australian census

^a Source: Agnew et al. (Reference Agnew, Bateman and Thorp2013); ^b Source: Deloitte Access Economics (2018).

Demographics: The gender and age distribution of the sample represents the 25–64 years Australian population as set by quota. Approximately 67% of Australian households own their residence, with or without a mortgage, but home ownership is concentrated in the second and higher quartiles of the wealth distribution. Few low wealth households participate in the mortgage market (ABS 2019-20a, 2019-20b). Accordingly, because we select past, current or future mortgage holders, people who are partnered, high income, educated and employed are over-represented compared with the population.

Financial literacy: We measure financial literacy using the “Big Three” financial literacy questions on interest, inflation and diversification, with the wording of the third question changed slightly to match the Australian context (Agnew et al. Reference Agnew, Bateman and Thorp2013; Lusardi and Mitchell Reference Lusardi and Mitchell2011). ^{Footnote 4} Rates of correct responses to the “Big Three” by participants are similar to rates reported by Agnew et al. (Reference Agnew, Bateman and Thorp2013) from a wider sample. Around 10% of the sample answered no questions correctly, a majority answered the interest and inflation questions correctly and less than half answered all three questions correctly. In the estimation to follow, we assign high financial literacy to participants who answered all three questions correctly.

Numeracy: We tested participants’ numeracy separately using the scale developed by Lipkus et al. (Reference Lipkus, Samsa and Rimer2001). Gerardi et al. (Reference Gerardi, Goette and Meier2013) show that mortgage default is more likely for borrowers who lack mathematical skills, but is not related to choice of mortgage. We aim to measure the relation between numeracy, subjective comfort with home loans and susceptibility to framing. ^{Footnote 5}

Around one fifth of the sample failed to answer any of these questions correctly but 55% answered all three correctly. In the estimation to follow, we assign high numeracy to participants who answered all three questions correctly and low numeracy to other participants.

Mortgage broker users: Consistent with other surveys, we find that more than half of this sample have consulted a mortgage broker in the past (55.4%) (Deloitte Access Economics 2018). Additionally, 45% of participants say they intend to consult a broker in the future. Three quarter of this group of future users consists of who have used a mortgage broker before, and one quarter who have not.

5.1 Comfort with lump sum debt and repayment streams: Task 1

The first task in the survey asked participants to rate their comfort with home loans on a seven point scale from “very uncomfortable” to “very comfortable”. They gave this rating over 10 loan amounts (Table 1) shown either first as lump sums and then as monthly repayment streams, or in the reverse order.

5.1.1 Preliminary comparisons

Table 3 reports comparisons of mean comfort rating by groups of participants distinguished by framing, borrowing experience, financial literacy and broker use. Table 4 defines the variables.

Table 3. Summary statistics: Subjective comfort with mortgage debt. This table shows mean values of ratings from 1 = “Very uncomfortable” to 7 = “Very comfortable” on debt/repayment levels by participant subgroup in the main survey, task 1 (n = 500). Since each participant gave a rating for 10 lump sum debt and monthly repayment levels in two sets, the total number of observations is 20 × 500 = 10,000. Table 4 describes the variables. *p < 0.10, **p < 0.05, ***p < 0.01

Table 4. Variable descriptions. This table reports definitions of variables used in estimation. Variables are computed from responses to an online survey of a sample of Australian adults who are past, current or intended future mortgage holders conducted in October 2020 (n = 999).

On average, participants were significantly less comfortable when loans were framed as lump sum debts instead of equivalent monthly repayment streams. Highly financially literate and numerate participants were more uncomfortable (by around 10%; mean of 3.146 for high financial literacy Cf. 3.655 for low financial literacy), on average over the 10 loan amounts, than their less capable counterparts. (Average ratings for people with and without current or past experience with a mortgage were not significantly different.) On the other hand, on average, participants who have consulted, or plan to consult, a mortgage broker said they are more comfortable (by around 20%) with debt than the comparison groups (ASIC 2017b).

Figure 2 shows the difference between average comfort levels at each of the 10 debt and repayment levels we tested. The graph shows several notable features, apart from general discomfort with lump sum framing. First, participants say they are equally comfortable with lump sums and repayment streams in the center of the range. Values in this region were close to the average loan size for households taking out new mortgages at the time of the survey. Many participants would have first-hand experience of loan sizes and associated repayments at these values. Second, unlike Goldstein et al. (Reference Goldstein, Hershfield and Benartzi2016), we do not find a reversal of the framing effect. They found that, at lower wealth levels, participants perceived lump sums as more adequate for retirement than monthly amounts, but that this perception reversed at higher wealth levels. Over the range of values we tested, participants expressed less comfort with lump sums, with more relative discomfort at smaller loan sizes.

Figure 2. Subjective comfort with varying mortgage debts depending on whether they are expressed as a lump sum or equivalent monthly loan repayment. This graph shows the mean comfort level (seven point scale from “Very uncomfortable” to “Very comfortable”) on the vertical axis against monthly repayments (red line) on the horizontal axis and the equivalent lump sums (green line) for Survey 1, task 1. Repayment and debt levels ranged from $950 ($200K) to $14,000 ($2979K). Error bars show one standard deviation around the means.

5.1.2 Regression results

Table 5 reports regression analysis of the responses to task 1. The estimates in model (1) show the variation in average comfort level by condition. As expected, participants grew more uncomfortable as total debt or repayments rose. Participants were also sensitive to increasing sequences: those who saw debt or repayments going up over the 10 choices expressed more discomfort than those who saw them going down. This presentation effect reduces average comfort by around 10% and may be related to participants anchoring to a lower starting value.

Table 5. Regression results: Subjective comfort with mortgage debt. This table reports OLS and IV-GMM regressions of participants’ comfort ratings from 1="Very uncomfortable” to 7="Very comfortable” from task 1 on experiment indicators and participant characteristics. Variable definitions are shown in Table 4. Each participant (n = 500) rated their comfort with 10 lump sum debt and monthly repayment levels, making a total number of observations of 20 × 500 = 10,000. Columns 1 and 2 report OLS regression results for all participants who completed task 1. Column 3 reports second stage regression results from IV-GMM estimations where “Used broker” is instrumented by 1) the number of registered financial advisers in the postcode of the participant as supplied by the regulator ASIC, and 2) the number of respondents to a separate survey who reside in the same postcode as the participant and who report having previously consulted a mortgage broker (see Online Appendix D). Online Appendix B reports first stage estimates and test statistics. Column 4 reports OLS estimates for comfort ratings for the subset of task 1 participants who stated that they intended to consult a mortgage broker in the future (n = 218). This group comprises 171 participants who have consulted a broker previously (“Have and will use broker”) and 47 participants who have not consulted a broker previously (“Have not and will use broker”). Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01

In terms of choice architecture, we see that participants were sensitive to the framing of mortgage debt. The coefficient on the (de-meaned) lump sum condition indicator is significant and negative, and amounts to a 3% reduction in the average comfort level. In our debt and repayment tasks, participants rate the repayment stream that can be readily compared with income as more comfortable than a less-easily-compared lump sum debt. One interpretation is that households treat lump sum debt as “larger” than monthly repayments, possibly because they are less comparable with regular income or with the participants’ experience with loan repayments (De La Rosa et al. Reference De La Rosa, Sussman, Giannella and Hell2022; Goldstein et al. Reference Goldstein, Hershfield and Benartzi2016). The danger when borrowers use initial repayments to guide choices of loan size is that repayments will change with interest rates. Capacity to service a loan is also vulnerable to shocks to household income.

We now turn to the question of how financial literacy or interactions with an adviser – here a mortgage broker – can change comfort with mortgage debt and sensitivity to framing. We measure financial literacy using the “Big Three” questions that have been widely implemented and tested in studies globally (Lusardi and Mitchell Reference Lusardi and Mitchell2011). We also add an indicator for having taken a mortgage in the past since this experience is likely to add to participants’ financial literacy, and include scores on numeracy that test understanding of fractions, percentages and probability. We also use another marker of numeracy and attentiveness: an indicator for whether participants gave comfort ratings that (weakly) decreased as loan amounts increased (labelled “monotonic responses”). We expect that the natural response to higher loan amounts is lower comfort. As such, we interpret responses that do not follow that pattern to indicate either low numeracy (i.e., participants are unable to correctly process the loan amount) or inattention (i.e., people replied without consideration). Around 80% of participants gave monotonic responses in task 1. We can also separate the survey participants into those who have consulted a mortgage broker in the past and those who have not.

Model (2) shows results from the regression of comfort levels on the experiment indicators, participant financial literacy indicators and broker advice indicators, preference indicators and socio-demographic controls. In this conditional model, we find that participants who answered all three financial literacy questions correctly chose significantly lower comfort ratings with any mortgage debt irrespective of amount and framing. High financial literacy lowers the average comfort level of 3.411 by more than 10%. Higher numeracy and attentiveness reduce comfort. Participants with higher numeracy express about 5% less comfort than average with all debt amounts. But mortgage experience increases comfort by around 4%. In other words, higher financial literacy and numeracy are related to less comfort with home loan debt overall, but experience may make borrowers more comfortable.

Moreover, sensitivity to lump sum framing is significantly less for the more financially literate, as we hypothesized (Section 2). In fact, the size of the positive coefficient on the interaction between indicators for high financial literacy and for lump sum framing offsets the negative coefficient on the framing indicator alone. Thus, people with high financial literacy are more uncomfortable with mortgage debt than the average participant but their discomfort is not affected by whether the loan obligation is shown as monthly repayments or a total lump sum owed. By contrast, estimates also show that participants with high numeracy and direct mortgage experience are more susceptible to lump sum framing. One possible explanation of this result is that participants with experience may be more strongly anchored to regular repayments than the less experienced, and therefore susceptible to framing.

We also see differences and similarities between high financial literacy participants and those who consulted mortgage brokers. First, participants who consulted brokers are significantly more comfortable with home loans at around 8% above the average participant rating and 18% above the average high-financial-literacy participant. Their comfort increases when loans are framed as lump sums versus repayments. Broker users’ average comfort with lump sums is almost 5% higher than comfort with monthly repayments. We find that both financially literate and broker-advised participants are insensitive to framing, but that broker-advised participants are relatively more comfortable with mortgage debt in general, while financially literate participants are less comfortable.

The estimated effects of other covariates on comfort with mortgage debt are unsurprising. Results for model (2) show that participants with better financial situations – those who are male, more educated, partnered, employed, in their early forties and with higher incomes – are more comfortable with mortgage debt. More risk averse participants are less comfortable with debt. But participants who rate themselves as more patient give higher comfort ratings. This preference relation could be explained by patient borrowers seeing home loans more as a form of saving through acquiring an asset than as a form of consumption of housing services.

5.2 Comfort equivalence: Task 2

In task 2, we show a randomly selected different group of participants (n = 499) values from Table 1 and asked them to use a slider to choose an amount in the alternative framing that they felt equally comfortable with. Figure 1, panel (b), shows a screen shot of the task for the lump sum frame. For example, participants chose the repayment stream that represented the same level of subjective comfort as the lump sum shown to them, and vice versa. We hypothesized that a participant who has an accurate knowledge of the monthly repayments associated with a lump sum debt would assign approximately equal comfort to both. In addition, comfort equivalences should be insensitive to framing.

5.2.1 Preliminary comparisons between participant groups

A natural way to evaluate responses to this task is to compare participants’ subjective matches with the objectively correct matches. We can also measure the size of the difference between subjective and objective matches. To do this, we calculate a proportional deviation as $deviatio{n_{i,l}} = chose{n_{i,l}}/correc{t_l} - 1$ where $chose{n_{i,l}}$ is the value of debt (repayment) selected by participant i and $correc{t_l}$ is the objective match for debt or repayment amount $l = 1, \ldots 10$ from Table 1. In the analysis reported below we take the absolute value of deviation to make interpretation easier.

Table 6 reports the unconditional mean absolute deviation, and compares group means, for participants who completed task 2. Across both frames and all debt levels, the average absolute proportional deviation is 110.8%. This large unconditional mean suggests that participants found this task harder than task 1. Another indicator of task difficulty is the low rate of monotonic responses to this task – only 23% of respondents choose (weakly) increasing (decreasing) repayments or lump sums as the total debt level increases (decreases) over 10 choices.

Table 6. Summary statistics: Absolute deviation. This table shows mean values of absolute deviations by participant subgroup in task 2. The absolute deviation is the absolute value of $deviatio{n_{i,l}} = chose{n_{i,l}}/correc{t_l} - 1$ where $chose{n_{i,l}}$ is the value of debt (repayment) selected by participant i for debt or repayment amount $l = 1, \ldots 10$ from Table 1. For 10 lump sum debt levels (monthly repayments), each participant chose a monthly repayment (lump sum debt) with which they felt equally comfortable, making a total number of observations of 20 × 499 = 9980. Table 4 describes the variables. *p < 0.10, **p < 0.05, ***p < 0.01

Mean absolute deviations are significantly larger when participants are selecting lump sums to match their comfort with given monthly repayment levels (122% Cf., 99.5%). Importantly, the standard error of the absolute deviation is also larger for the repayment frame. In other words, participants are less comfortable with, and less clear about, the lump sum debt implications of a home loan described in terms of repayments than the reverse framing. Participants with higher financial literacy select equivalent debts or repayments that are closer to objectively equivalent levels, as do participants with higher numeracy. By contrast, average deviations for broker users are significantly higher than for non-broker users (118.2% Cf., 102.1%), with a similar difference for participants who intend to use brokers versus those who do not (125.1% Cf., 98.3%). This preliminary result implies that broker users are less likely than non-broker users to rate objectively equivalent lump sum debts and repayments as equally comfortable.

While the results show that participants make choices that match lump sums and repayments in the center of the range, they choose high debt amounts (repayment amounts) for smaller loans and choose low debt amounts (repayment amounts) for higher loans. Figure 3, panel (a), graphs the average of the (natural log) of the repayments that participants rated as equally comfortable (or uncomfortable) as the debt amounts on the horizontal axis. In comparison with the diagonal line that shows the objective equivalents, comfort-equivalent choices are significantly flatter. Panel (b) that graphs the average log debt that participants said was equally comfortable as the repayments shows a similar pattern, but is even flatter than for the lump sum framing. In fact, participants match repayments within a remarkably narrow range of debt levels when asked what they felt was equally comfortable. These graphs are further preliminary evidence for a finding that we investigate in more detail below, that people deviate more from the objective equivalent when they are shown the repayment framing than when they are shown the lump sum framing.

Figure 3. Subjective equal-comfort lump sum debt or monthly repayment at varying mortgage debts. These graphs show means of responses to task 2. Panel (a) shows the log of mean equivalent repayments that participants selected as being equally comfortable as varying lump sum mortgage debts. The thin line shows the objective equivalent values. Panel (b) shows the log of mean equivalent lump sums that participants selected as being equally comfortable as varying monthly mortgage repayments. The thin line shows the objective equivalent values. Repayment and debt levels ranged from $950 ($200K) to $14,000 ($2979K).

5.2.2 Regression results

Table 7 reports results from regressions of the absolute deviation of subjective comfort matches of lump sum and repayment stream from objectively matched values on condition indicators and participant characteristics. We add a quadratic term in log(loan size) to model the increasing absolute deviations at the extremes of the debt range, as shown in Figure 3. The OLS results reported in columns 1 and 2 confirm most results from Table 6. The coefficient on the framing indicator (“Lump sum condition”) shows that the lump sum presentation lowers the absolute deviation by around 20% compared with the unconditional mean of 1.108. In other words, participants choose comfort-equivalent repayment streams that are closer to the objective values when presented with lump sum loan amounts. When asked to choose a lump sum to match a repayment amount, they deviate more from the objective value.

Table 7. Regressions results: Absolute deviations between comfort-equivalent debt or repayments. This table reports OLS and IV-GMM regressions using the data from the main survey, task 2. The dependent variable is the absolute value of proportional deviations of participants’ selections of equal-comfort debt or repayment amounts from objective equivalence on debt/repayment levels. Explanatory variables are experiment indicators and participant characteristics. Variable definitions are shown in Table 4. For 10 lump sum debt levels (monthly repayments), each participant chose a monthly repayment (lump sum debt) with which they felt equally comfortable, making a total number of observations of 20 × 499 = 9980. Columns 1 and 2 report OLS regression results for all participants who completed task 2. Column 3 reports second stage regression results from IV-GMM estimations where “Used broker” is instrumented by 1) the number of registered financial advisers in the postcode of the participant as supplied by the regulator ASIC, and 2) the number of respondents to a separate survey who reside in the same postcode as the participant and who report having previously consulted a mortgage broker (see Online Appendix D). Online Appendix B reports first stage estimates and test statistics. Column 4 reports OLS estimates for absolute deviations for the subset of task 2 participants who stated that they intended to consult a mortgage broker in the future (n = 232). This group comprises 171 participants who have consulted a broker previously (“Have and will use broker"=1) and 61 participants who have not consulted a broker previously (“Have not and will use broker”). Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01

Participants with current or past mortgage experience, higher financial literacy or higher numeracy also match amounts that were significantly closer to the objective equivalents. Coefficients on each of these indicators show that skill and experience can reduce absolute deviations by over 20% of the unconditional mean. Coefficients on interactions between the framing indicator and numeracy and mortgage experience significantly moderated these effects, whereas we do not estimate a significant interaction for high financial literacy participants. Overall, financial literacy is related to better calibration between lump sums and repayment streams. In terms of broker users, we expect that consulting a broker would help clients better match lump sums and repayment streams. However, the coefficient on the indicator for broker use in model (2) is significant and positive, amounting to around an 8% increase in absolute deviation over the unconditional mean.

In Table 7 we again aim to control for selection into broker use by presenting results from IV-GMM (model (3)), and a model that is estimated only on the subsample of respondents who intend to consult a broker in the future (model (4)). We use the same instruments for IV-GMM estimation here as we did for task 1 (Table 5). However, some test results indicate instrument weakness and we interpret the results with caution.

In contrast to model (2), estimates from models (3) and (4) are evidence that consulting a broker significantly lowers participants’ absolute deviations. In model (3) the sign on the coefficient for broker use in the second stage estimation is negative. Similarly, model (4) estimates show that the group of intending broker users who have previously consulted a broker have significantly lower absolute deviations than those who intend to consult a broker, but have not yet. This result could be interpreted as showing that brokers do help their clients understand the debt size implications of a repayment stream and vice versa. Both of these broker-use-dependent results suggest that brokers may help educate clients.

The results of both survey tasks show that financial literacy is associated with less comfort with mortgage debt and a better grasp of the connection between the lump sum liability and the servicing burden of repayments. These results hold after conditioning on an array of demographic, preference and experience indicators. We also find that people are generally less comfortable with lump sum debts than equivalent repayment streams, and less able to consistently map from a repayment stream to the equivalent lump sum debt. Financial literacy however, tends to reduce these framing effects. Going to a broker for advice is related to higher comfort with mortgage debt, and higher comfort with loans framed as lump sums. It is also related to smaller deviations when calibrating lump sums and repayment streams. We now turn to the question of which borrowers engage brokers.

5.3 Characteristics of mortgage broker clients

In this section, we identify the differences between participants who have gone to mortgage brokers for advice in the past and who intend to consult brokers in the future, either again or for the first time. While the predictors of financial literacy are well established by other studies, it is less clear what types of borrowers go to mortgage brokers.

In Table 8 we summarize the characteristics of participants from this sample who have consulted brokers, and those who have not yet, but intend to. Column 1 shows marginal effects from a logit model that predicts the probability that a survey participant has used a broker in the past, conditional on a range of financial literacy measures, demographics, preference indicators and a measure of participants’ estimates of the chance of upwards or downwards house price changes. (The online appendix reports the full results from the logit estimation.) The unconditional probability of having used a broker is 55.4%. Participants in the 25–44 year age group are 16.6 percentage points more likely than the 45–64 year group to have used a mortgage broker. Participants who report higher incomes, who are partnered or who are employed are also significantly more likely to have used a broker, consistent with brokers being the choice of people who have higher borrowing capacity. Unsurprisingly, participants who have a current mortgage, or have taken one in the past, are 40.6 percentage points more likely to have consulted a broker than participants who are yet to actually take out a loan.

Table 8. Marginal effects: Logit estimation of broker use. This table reports analysis of mortgage broker use. The model in column 1 is estimated on the full survey sample (n = 999). It shows marginal effects from a logit regression of the indicator for previously consulting a mortgage broker on participant characteristics. Column 2 reports p-values for the tests that the differences in predictive margins reported in column 1 are insignificantly different from zero using delta-method standard errors. The model in column 3 is estimated on the sub-sample of participants who have not previously consulted a mortgage broker (n = 446). It shows marginal effects from a logit regression of an indicator that the participant says that they intend to consult a mortgage broker in the future on participant characteristics. Column 4 reports p-values for the tests that the differences in predictive margins reported in column 3 are zero using delta-method standard errors. Online Appendix C reports the underlying logit estimation results. Variable definitions are shown in Table 4. *p < 0.10, **p < 0.05, ***p < 0.01.

If clients consult brokers partly to lower their anxiety, participants who perceive house prices as more risky may be more likely to look for reassurance from a broker-adviser, as theorised by Gennaioli et al. (Reference Gennaioli, Shleifer and Vishny2015). Survey responses can help us gauge participants’ subjective risk perceptions about the housing market. The survey collected participants’ estimates of the chances of a range of changes in house prices 5 years from the time of the survey. We use responses to the following questions to make a proxy variable for participants’ beliefs about upside and downside risk to house values: “On a scale from 0 percent to 100 percent where 0 means that you think there is no chance and 100 means that you think the event is absolutely sure to happen, what do you think are the chances that 5 years from now the value of your home (“homes in your postcode” for participants who do not yet own a home) will have gone up (down) by more than 10 (and 20) percent?”. We computed upside (downside) perceived house price risk as the weighted sum of the two probabilities that each participant assigned to the 10 and 20 percent house price rises (falls) where weights are squared percentage house prices changes.

The outcomes are two variables that proxy for individuals’ subjective house price semi-variances on a scale between 0 and 5. The mean of the upside risk variable is 2.4 and the mean of the downside risk variable is 1.5, with a standard deviation of 1.3 in both cases. Marginal effects reported in column 1 of Table 8 show that as the downside house price risk variable rises from its minimum (0) to its maximum (5) the likelihood of having used a broker rises by 15.3 percentage points, while the marginal effect of the upside risk variable is insignificant. When we break this effect out by the group of participants who have previously taken a mortgage versus those who have not, we see a difference between upside and downside price changes. Higher probability of broker use is related to higher perception of downside house price risk among the group who are already home owners. (Only 48 participants have previously consulted a broker but have not taken a mortgage.) But lower probability of having used a broker is related to a higher perception of upside house price risk among those who have not yet borrowed.

To explore this further, in column 2 of Table 8 we report marginal effects from a logit model estimated on the subset of 446 survey participants who have not yet consulted a mortgage broker. This model estimates the probability that a participant who had not yet consulted a broker intends to do so in the future versus not intending to. The unconditional probability of intending to consult a broker is 24.2%. We again find that having had a mortgage previously matters. In this case we see a significantly lower likelihood that a participant plans to go to a broker if they have already had a mortgage without using a broker. In terms of house price risk perception, we find that participants who have not previously had a mortgage, but who assign high probabilities to rising house prices, are more likely to plan to use a broker. The marginal effect is 33.5 percentage points as the upside risk proxy increases from 0 to 5.

One interpretation of these effects is that past or intended broker users perceive more risk than non-broker users. When this group is yet to buy a home, their risk perception manifests in concern that rising house prices will outpace their capacity to get a loan. Once they own a home, the focus of their risk perception shifts to falling home asset (and wealth) values. Overall, we can conclude that higher subjective risk perception, although not higher measured risk aversion, is associated with engagement with brokers. The effect of higher risk perception is some support for the prediction that borrowers turn to advisers for “peace of mind”.