COMM1110: Assessment 2a Student SID Report Overview: This briefing pack provides input into the Report on Mortgage Stress (MS) in Sydney, via three (3) key aspects of analytical focus: x The Information Toolbox (IT) discusses recent problem-related research (both academic & grey), then builds a visual (logic tree) representation of the problem and its causes, x The Statistical Toolbox (ST) describes key variables in the provided dataset and begins the process of statistical analysis, and x The Ethical Toolbox (ET) highlights an ethical dilemma for Lenders (personal circumstances), and uses ethical frameworks to explore the moral aspects of effective decision-making. Information Toolbox: This toolbox considers key management aspects of effective problem-solving (PS). This PS typically involves 3 sequential steps, ie: x Scoping: defining (via research) the problem, then disaggregating it into causes to help flag the complexity of problem, then x Analysis: exploring of this research and related statistical evidence as key problem inputs, and starting the process of drawing analytical conclusions, then x Decision: synthesising and presenting the analysis, then communicating proposed decision options to problem stakeholders. This report focusses mostly on Steps 1 & 2, and the subsequent report on Step 3. In order to scope the problem, we require quality research inputs. Below is a summary of three (3) inputs based on both Academic, (ie peer-reviewed) and Grey (ie Business / Government) papers. Yates and Berry (2011) provide a long view exploration of the flow of , firstly from a historical perspective ;∥ , then posits two contrasting scenarios going forward (continued growth vs a sharp downward correction). The paper discusses the decline on the stock of social housing, the growth of the private rental market and broader economic issues including ongoing economic growth challenges (eg the role of China), increasing household consumption levels, and declining wage growth. Some of the key policy-related conclusions it draws are around the need for policy interventions to increase affordability, the need for more innovation around related infrastructure development, a more flexible labour supply, the ě ě ∥ ě , and the need (or desire) to redistribute populations beyond key urban centres like Sydney. Bullock (2018), a Reserve Bank Assistant Governor, links mortgage stress to broader financial (debt) stress challenges. She notes that whilst the number of households in mortgage stress has actually fallen over the last decade (2006-2016) there is still a significant proportion still in mortgage stress. She contrasts owner-occupier vs investor-related stress issues, noting increasing indebtedness in States with a more Mining-exposed economy (eg WA) and the ongoing importance of managing (or at least influencing) such factors as interest rate movements, and the need for more prudent lending standards by Banks and other Lenders, and interestingly for us the need for more timely data on the problem as it changes over time. Finally, Roy Morgan Research (2021) notes that 15.8% of mortgage holders (in NSW, Victoria and the d ∥ in the three months to September 2021, ie during the COMM1110: Assessment 2a Student SID current pandemic. This contrast with a rate of 35.6% during the GFC (2008). The report notes such ě id- W (and its provision timing links to levels of household indebtedness), the changes in available working hours, job opportunities (especially in some retail-related sectors of the economy) and increasing level of redundancies across the Australian economy during the pandemic. Importantly from a problem definition and data analysis perspective, the study contrasts Z (between 25% and 45% of after-tax income spent on loan repayments depending on income and spending) – ě at Z (based on interest only repayment). Figure 1: Mortgage Stress by Owner-Occupied Mortgage Holders (2021) The problem of Mortgage Stress is therefore defined as households who are spending more than 25% of after-tax income on loan repayments. This definition depends on and thus makes a number of related and important assumptions – including individual consumption patterns, household income levels and growth potential, government financial support, prevailing interest rates and movements, and the broader economic conditions and potential shocks (eg the pandemic). Using the above inputs, the problem is presented below in visual form is via a logic tree. It is framed firstly around the key stakeholder as identified above, then the many causal variables explored in the research (like interest rates, job security, and lending standards), and a number of additional variables identified as part of the author wider related research (like health and well-being, and risk appetite). COMM1110: Assessment 2a Student SID Figure 2: Logic Tree, incorporating research-related plus other causal factors. COMM1110: Assessment 2a Student SID Statistical Toolbox: To analyse mortgage stress, we will need to understand the patterns and variables in the sample that has been provided to us. As part of this analysis, we will assume that the data was collected appropriately this includes assumptions such as the fact that we have a random sample and thus there are no selection issues associated with the sample. Furthermore, we assume that there are no mistakes in the data and has been appropriated cleaned. In addition to the above, all statements made here are all in relation to the sample. Thus, at no point should the patterns observed be extrapolated to the population, our purpose here is to describe what is observed in the sample. Finally, when describing any relationship, we do not imply causality as this is purely an exploratory piece and we have not examined any possible theories behind any relationship that may arise. Before we begin this analysis, we will describe the data that has been presented to us. The dataset consists of 4,866 observations and five variables including, hcost, lowinc, lowSEIFA, age and comtime. Definitions of these variables can be found in the ě We note that hcost, age and comtime are continuous, while lowinc and lowSEIFA are binary. To examine mortgage stress, we need to define mortgage stress, according to the RBA (2018), owner-occupier debt had mortgage payments of 30 per cent or less of income, which is often used as a rough indicator of the limit for a sustainable level of mortgage repayments 1. Using this definition, we will use the variable hcost as proxy for this as it measures / ě ě – income ratios calculated as the ratio of weekly mortgage repayments and weekly gross household Whilst this is not a perfect fit as we do not know whether it is owner occupied or otherwise e.g. rent vesting we will use this as an indicative measure. Thus, for the purposes of this report we will define mortgage stress as hcost>0.3. We do acknowledge that some relevant data are not provided which would better capture mortgage stress for example, whether the household head is working or not and household size. One of these factors have been controlled which is household size but we are unsure of others. We also note that quite a few households have hcost ratios that are very high, say >80%. It is also worth noting that our client is also interested in examining the question of whether mortgage stress is more prevalent amongst older individuals and low income households. Looking at the provided data we can summarise hcost, age and lowinc as follows: Table 1: Summary stats of key variables hcost age lowinc Mean 0.287 49.626 0.382 Median 0.250 50 0 Mode 0.222 49 0 Standard Deviation 0.145 9.198 0.486 Sample Variance 0.021 84.598 0.236 Minimum 0.0291 17 0 Maximum 0.889 85 1 Count 4866 4866 4866 1 https://www.rba.gov.au/speeches/2018/sp-ag-2018-02-20.html#fn2 COMM1110: Assessment 2a Student SID From the above, we note that the mean is larger than the median which suggests that there is a positive skew to this data. We also note that the range is quite wide and the standard deviation ě50.6%. We also note that since the median is approximately at 0.25 that most of the people in this sample would not be experiencing mortgage stress. In terms of age, we note that the average age in the sample, is 50 and with the youngest being 17 and oldest being 85. The median age is 49. According to the ABS, the median age in June 2020 was 38 2 which means that this sample has a much higher than the median Australian. In addition to this we would also like to note that in a recent survey from money.co.uk the average age of home ownership is 36 years old, which is quite different to what we find in this sample.3 Also because of the range of the data we may some retired individuals the survey. Thus, it is worth investigating how this data was collected. Finally, for low income we found that 38% of the sample is low-income earners which is consistent with the definition of ∥ ∥ Following from this we will analyse the relationships between housing costs, age and low income. We will first examine the relationship between low income and housing costs by comparing the distribution of housing costs for all households to that of low income and for completeness those categorised as low SEIFA. Investigating the lowSEIFA and hcost association yields another useful insight as it measures the spatial/neighbourhood element of housing affordability, albeit crudely given you are aggregating different types of neighbourhoods based on SEIFA. Immediately, we make the following observations about the sample: (1) The proportion of low income earners experiencing mortgage stress is higher than all other households, 53.9% compared to 35.9%. (2) The proportion of low SEIFA individuals experiencing mortgage stress is higher than all other households, 49.4% compared to 35.9%. These findings are summarised in the histogram below which shows a right shift in the histogram from low income and low SEIFA when compared to all households, reflecting the above two points. 2 https://www.abs.gov.au/articles/twenty-years-population-change 3 https://www.realestate.com.au/news/average-age-of-aussie-first-home-buyers-closer-to-40-than-20- research-reveals/ COMM1110: Assessment 2a Student SID Figure 1: Histogram of Housing Costs The implication of this is not clear from the perspective of the population. ∥ proven whether these two distributions are distinctly different. To this point we have only suggested that they may be different, statistical tests would need to be undertaken to make this a more concrete claim e.g. a Chi-squared test. If we assume that indeed this is statistically different then we can say that within this sample that a larger proportion of lower income earners do experience mortgage stress when compared to all households. However, we cannot necessarily say that this is also true of the population as we have not performed any inference on this data. Thus, our statement is restricted to this sample and it suggests that mortgage stress may be more prevalent amongst low income earners. However, we would like to note some of the interesting anomalies with this dataset as the average age is higher than what other surveys suggest. 0.0% 5.0% 10.0% 15.0% 20.0% 25.0% Histogram – All household Histogram – Low Seifa Histogram – Low Income COMM1110: Assessment 2a Student SID Our next analysis is to examine the relationship between housing costs and age. To do this we construct a scatterplot of age vs hcost and yield the following plot. Figure 2: Scatterplot and regression of housing cost and age It is very difficult to identify the relationship between these two variables so construct a correlation matrix to see if we can better characterise the relationship of these two variables. Table 2: Correlation matrix of housing costs, age and comtime age comtime hcost age 1 comtime -0.03418 1 hcost 0.029928 -0.01928 1 The above suggests that in the sample there may be a positive linear association between age and housing costs and a negative linear association between comtime and hcost. This is also reinforced by the simple linear regression results below where the coefficient for age is positive and statistically significant at the 95% level of confidence. y = 0.0005x + 0.2638 R2 = 0.0009 0.000 0.100 0.200 0.300 0.400 0.500 0.600 0.700 0.800 0.900 1.000 0 10 20 30 40 50 60 70 80 90 COMM1110: Assessment 2a Student SID Table 3: Regression output of hcost vs age Regression Statistics Multiple R 0.029928 R Square 0.000896 Adjusted R Square 0.00069 Standard Error 0.145336 Observations 4866 Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept 0.263803 0.011434 23.0722 7.3E-112 0.241388 0.286219 age 0.000473 0.000227 2.088155 0.036836 2.89E-05 0.000917 The above regression suggests that there is a small positive linear association which these two variables. However, it is worth noting that we have used this regression purely for descriptive purposes, we have not identified causality nor can we assert that this is the relationship which exists in the population. This is merely just a more specific description of the relationship we observe in this sample. It is also worth noting from the scatterplot that there is a lot of variation in the data and thus the fit of the model is poor as suggested by the R2 of 0.08%. Which suggests that more investigation is required to understand the relationship of these two variables and a linear model may not be appropriate. Following from the above, we can say that we observe a positive linear association between between age and hcost. What is more difficult to ascertain is causality e.g. does an increase in age lead to an increase in hcost. We cannot answer this question presently as there may be confounding factors which are not captured e.g. gender as females tend to earn less and are on average are older4. To control for some of these confounding effects we run a multiple linear regression which include all the other variables. The results are presented below. Table 4: Multiple linear regression of housing costs Regression Statistics Multiple R 0.368027 R Square 0.135444 Adjusted R Square 0.134732 Standard Error 0.135238 Observations 4866 Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Intercept 0.222297 0.011041 20.13394 1.17E-86 0.200652 0.243942 lowinc 0.0989 0.004095 24.152 8.3E-122 0.090872 0.106928 lowSEIFA 0.03026 0.004129 7.328088 2.72E-13 0.022165 0.038356 age 0.000401 0.000211 1.901563 0.057287 -1.2E-05 0.000815 4 https://www.abs.gov.au/AUSSTATS/abs@.nsf/DetailsPage/3101.0Jun%202019 OpenDocument Population by Age and Sex Tables COMM1110: Assessment 2a Student SID comtime -0.0011 0.000574 -1.9166 0.055348 -0.00223 2.52E-05 From the above we note that the coefficient of age remains qualitatively similar to the simple linear regression. However, it is no longer statistically significant at the 95% level of confidence. Which suggests that it could potentially be zero as a coefficient as indicated by the confidence intervals. It is also pleasing to see that the adjusted R-Square has increased to 13.4% which suggests a better fit of a linear model to the sample data. In totality, this provides more evidence that there may be a positive association between age and housing costs. Again we cannot assert causality on what we have found in this sample. Another interesting insight we found was that when we investigated the relationship between comtime and hcost. Thought this was not specifically flagged but is a useful insight that is relevant in considering whether households are potentially willing to pay more in rent to live closer to work. Negative correlations are supportive of this although these correlations are small as per the above regression results and correlation matrix. All in all, respecting the assumptions made above we can report the following with respect to this sample: x Compared to all households, there is a larger proportion of low income individuals experiencing mortgage stress. x Compared to all households, there is a larger proportion of low SEIFA individuals experiencing mortgage stress. x There does seem to be a positive association between housing costs and age, but we cannot identify causality. We also would like to acknowledge that there are limitations associated with the data presented to us. The first is the fact that the hcost variable is not a perfect fit to what the RBA defines as a & ∥ ∥ ě ∥ ∥ ∥ ě The second point is that the average age of home ownership is 50 in this sample whereas in other surveys this is significantly lower i.e. 36. From a data perspective, these two points alone should lead to more questions as to how the variables are calculated and how the data was collected. Finally, we would like to acknowledge that all that is written in this section relates to what we observe in the sample which we have and to date we cannot extrapolate this to the population. Furthermore, we cannot ascertain causality as this is currently outside the scope of the analysis, the purpose here is to observe and describe the patterns we observe in this sample. Ethical Toolbox: Ethics is about decision-making with moral perspectives included. Morality involves the consideration of a range of context-related perspectives to think about what would be the right (or an acceptable) decision to make given the many contextual and causal factors and thus reasonable decision criteria. These perspectives might include the economic complexity, stakeholder needs, principles or duties of each stakeholder, social benefits and/or ě key decision-makers. COMM1110: Assessment 2a Student SID For this analysis I will explore a specific ethical dilemma from the perspective of Banks and Other Lenders (B&OL). One ethical dilemma for them would be to what extent they should consider the personal circumstances of borrowers in their lending decisions – beyond their financial ability to pay. This is an ethical dilemma because these personal circumstances can change quickly and over time (and for reasons beyond borrowers reasonable control), that home ownership is a social good which has value beyond the economic, and that B&OL lenders have a duty of care to help and support the Australian community and a key player in their ongoing prosperity and well-being. Considering this dilemma via a range of theoretical frameworks, we can see that: x Deontologically the rules and duties which govern effective decisions by B&OL need to change over time given economic circumstances, may not always be clear to lenders (who many not be highly financially – or even language literate) and are sometimes influenced (or even mandated) by government policy. x Utilitarianism ě ; ě ΘK> ě owners) who may have conflicting interests and needs for their role the Lending/Mortgage Industry, x Virtue involved the reflection on important values and intended character, and raises a number of critical thinking questions like ∥ ě ∥ ∥ aspire to ě how do we want to impact and influence social outcomes in our community x Care means each B&OL exploring the problem from socially-aware perspective, and thus carefully considering the impact of their decisions on each ongoing relationship rather than just the effective application of rules or lending principles. Bringing together the above discussion, the issues associated with having a Mortgage in Sydney include the need to: 1) Identify key stakeholders to the mortgage stress problem and exploring their needs and wants in good depth, 2) Research and document the current rules and principles which govern the decisions made in relation to the approval of mortgages, 3) Have a clear goal (or objective) of harm minimisation as it relates to the level of stress which mortgage holders can or should bear, and 4) Continue to gather and analyse the data about the size and complexity of the problem so that more and better informed decisions can be made over time. COMM1110: Assessment 2a Student SID References: D , ě/ ě ě ěD ^ ěě Z Lending and Borrowing Summit. Reserve Bank of Australia, 20 February, 2018. Roy Morgan, 2021, Mortgage stress at record lows during the 2021 lockdowns in NSW, Victoria and the ACT . November 22 2021, Finding No. 8843. Yates. J and D , ěD D d d / Housing Studies, Vol 26, E -8, 1133-1156.
