Understanding the Effect of Financial Behaviour on Mental Health: Evidence From Australia

ABSTRACT

This paper uses the Household, Income and Labour Dynamics in Australia (HILDA) data from 2001 to 2022 to examine the relationship between individual financial behaviours and mental health. Two distinct types of financial behaviours (savings and debt) and their relationship with mental health (as measured by Mental Health Inventory‐5) are considered over several annual waves. We examine causality between the two variables utilising an instrumental variable approach and find that stable financial behaviour significantly improves the mental health of individuals. Specifically, maintaining regular savings habits and making timely payments on credit card bills have a positive impact on the mental health of individuals. Furthermore, the impact of savings behaviour on mental health is stronger for men than women. Our results are robust to alternative measures of subjective wellbeing and estimation techniques. The findings from this study have substantial policy implications, indicating that stable financial habits can significantly contribute to improving mental health, which in turn can lead to higher productivity and employment.

1. Introduction

A substantial body of literature demonstrates that high debt and low savings have a negative impact on mental health (Fitch et al. 2011; Hassan et al. 2021; Marshall et al. 2021; Richardson et al. 2013; Ten Have et al. 2021). This literature, however, has exclusively focused on debt and savings levels. By contrast, the financial behaviours that contribute towards debt and savings levels, as well as their possible impact on mental health, have not been well explored, such as paying off credit card debt on time and saving regularly. A better understanding of the role played by financial behaviours can help inform intervention strategies designed to mitigate the impact of negative financial behaviours on mental health (e.g., by enhancing financial mindfulness and impulse control, strengthening financial resilience, etc), and can help bridge the gap between financial advice and mental health support (Evans 2018; Xiao et al. 2006). It is also the case that most studies examining debt and savings have been cross‐sectional. The few longitudinal studies point to the corrosive impact of debt and low savings levels on mental and emotional health (Ettman et al. 2024; Bialowolski et al. 2024; Bialowolski et al. 2019; M. P. Taylor et al. 2011).

The psychology and behavioural economics literature offers several possible explanations for the association between financial behaviour and mental health. First, positive financial behaviours may reduce financial strain, that is, less worry about money and, therefore, better mental health. This is supported by evidence from cross‐sectional and longitudinal studies indicating that financial strain and financial concern are strong predictors of psychological distress (Chai 2023; Jessop et al. 2005; Marshall et al. 2021; Ryu and Fan 2023; Tsuchiya et al. 2020), while financial debt and worry about debt are reported to predict depression and suicidal ideation (Stickley et al. 2023; Bialowolski et al. 2019), and quality of life (Bialowolski et al. 2021, 2024). These findings are consistent with Lazarus and Folkman's (1984) Theory of Psychological Stress and Coping, wherein financial strain (debt) and limited coping strategies (no savings) can be viewed as chronic stressors leading to poor mental health. Alternatively, the findings are also consistent with Hobfoll's (1989) Conservation of Resources Theory, which posits that financial strain represents a loss of resources and therefore is associated with subsequent psychological distress.

Second, positive financial behaviour may foster autonomy and self‐efficacy, which in turn, lead to improved mental health. A greater sense of control over life circumstances is known to be a strong predictor of psychological wellbeing (Kinderman et al. 2013; Kondo et al. 2021). This model is consistent with Lazarus and Folkman's (1984) Psychological Stress and Coping Theory, which posits that financial control can be viewed as a predictor of mental health. It is also consistent with Deci and Ryan's (1985) Self‐determination Theory, whereby financial insecurity (i.e., debt) can be viewed as reducing autonomy and competency, thereby predisposing individuals to psychological distress. In contrast, financial security (i.e., savings) can be viewed as enhancing autonomy and competence, thereby mitigating psychological distress.

Third, positive financial behaviour may lead to financial largesse and the capacity to afford necessities and pleasures of life (e.g., buying a house or going on a holiday), thereby facilitating participation in social and community activities with positive effects on mental health (Rautio et al. 2013; Xiao et al. 2006). This view is consistent with Hobfoll's (1989) Conservation of Resources Theory, which posits that savings and wealth accumulation can serve as protective resources that enhance mental health. It is also consistent with Kahneman and Tversky’s (1979; 1992) Prospect Theory that the ability to afford life's pleasures contributes to positive emotions and life satisfaction and, in turn, good mental health.

Finally, positive financial behaviours may encourage individuals to adopt healthy coping mechanisms to manage stress rather than relying on negative coping strategies, such as substance abuse or avoidance behaviours, which can exacerbate mental health issues (Frankham et al. 2020; S. E. Taylor and Stanton 2007). This explanation is consistent with Lazarus and Folkman's (1984) Theory of Psychological Stress and Coping, which posits that financial stress can lead to maladaptive coping, whereas financial security enables healthier coping mechanisms. In addition, this explanation is consistent with Deci and Ryan's (1985) Self‐Determination Theory, which posits that autonomy and competence in financial matters promote proactive and healthy coping mechanisms, thereby contributing to good mental health.

The effect of debt and savings on mental health may not be invariant across demographic groupings, such as gender and socioeconomic status. There is substantive evidence that women are more risk‐averse than men, and their social preferences are more situation‐specific, thereby limiting financial risk‐taking behaviours (Croson and Gneezy 2009; Jianakoplos and Bernasek 1998). It is also well described that stressful life events that trigger trauma and stress in women and men differ over the lifecycle and, notably, during adolescence, early adulthood, and retirement (Hatch and Dohrenwend 2007). In addition, gender is known to impact mental health differently, with men exhibiting more externalising disorders, such as substance abuse and antisocial behaviour, whereas women demonstrate more internalising disorders, such as depression and anxiety (Rosenfield and Mouzon 2013). Women experience higher levels of stress in reaction to debt (Chen et al. 2021; Dunn and Mirzaie 2016) and exhibit a broader range of health problems in reaction to financial stress (Chen et al. 2021; Kuo et al. 2021). Socioeconomic status is another significant confounder of financial behaviour and mental health. An individual living in poor socioeconomic conditions is more likely to report poor financial behaviours and mental health problems (Janke et al. 2023). Additionally, individuals from low socioeconomic areas are more likely to experience greater inequality, higher social deprivation, and lower access to essential services and facilities, increasing their vulnerability to mental distress (Tibber et al. 2022).

In summary, the present study aims to investigate the relationship between financial behaviour (debt and saving habits) and mental health in the Australian context. It also aims to investigate whether this association differs between men and women, and individuals from low‐versus high‐socioeconomic status. Finally, it examines whether financial behaviour is linked to other subjective wellbeing measures that have a strong connection to mental health, such as vitality, social functioning, role‐emotional, and general wellbeing score as measured by the aggregated Short Form Health Survey questionnaire (SF‐36).

This paper makes several contributions to the literature. First, as previously discussed, there is a limited literature on how consistent savings and, similarly, responsible debt management can influence mental health in a panel data setting. Unlike previous studies that focus on general levels of credit and savings using cross‐sectional data, this study concentrates on individual savings and debt habits using a panel dataset from Australia. Gaining a clearer understanding of the impact of financial behaviours can assist policymakers in developing intervention strategies aimed at alleviating the negative effects of poor financial practices on mental health. Second, using the Household, Income and Labour Dynamics in Australia (HILDA) data from 2001 to 2022, we establish a causal relationship between financial behaviour and mental health. While the existing literature confirms a strong association between the health‐wealth nexus, the causality between the two variables, especially regarding mental health morbidity, remains limited. Unlike previous studies that rely on propensity score matching (Bialowolski and Weziak‐Bialowolska 2021) and prior exposure control techniques (Białowolski et al. 2019), this study uses a fixed effects instrumental variable (FE‐IV) approach to study the impact of savings and debt behaviours on the mental health of an epidemiologically representative sample of Australians. There are several advantages to using an FE‐IV approach over other methodologies, including the ability to control for unobserved time‐invariant confounders that may bias the results, mitigate reverse causality, and reduce measurement errors (Milner et al. 2018).

Thus, to mitigate endogeneity biases, an instrumental variable (IV) is constructed for financial behaviour, following previous studies, by interacting the Australian state‐level utility prices (Awaworyi Churchill and Smyth 2021; Srivastava and Trinh 2021) with the distance to the cohort‐specific pension entitlement age of the individual. First, the state‐level utility prices capture the cost‐of‐living expenses, which will directly impact savings and debt behaviours, such as how regularly an individual saves or repays their debt (Jumena et al. 2022). A second factor is that financial behaviour changes over the lifespan, with behaviours that promote debt predominating in early adulthood and savings in later adulthood. According to Ando and Modigliani’s (1963) life cycle income hypothesis, individuals seek to smooth consumption over their lifetime income by accumulating debt early in life when income is low to invest in assets (e.g., housing, raising a family, etc) but over time as income grows, individuals accumulate wealth for use in retirement. We argue that the interacted variable, state‐level utility prices times the distance to pension entitlement age of the individual, captures the financial burden as a result of the cost‐of‐living pressure (e.g., when paying utility bills, such as electricity, gas, water, and fuel etc.) that individuals face regularly and to the extent they are afar or closer to their pension entitlement age. Thus, the financial burden variable is likely to influence mental health through the channel of financial behaviour after controlling for known socioeconomic and demographic factors. We also examine the reciprocal relationship, that is, the effect of mental health on financial behaviour, as existing literature suggests that psychological factors, such as hope and coping, can directly impact financial behaviours, particularly during times of crisis (Arya et al. 2023). However, we find a single robust causal relationship, showing that stable financial behaviour significantly influences mental health. We consider this to be a significant contribution to the literature.

Third, this study extends the research, which has been predominantly based in the United States and Europe (Bialowolski and Weziak‐Bialowolska 2021; Białowolski et al. 2019; Bialowolski et al. 2024), to the Australian context. Australia is primarily a resource‐rich economy that relies heavily on international trade and investment to generate local employment and drive growth (Pandya and Sisombat 2017). Since the COVID‐19 pandemic, Australia has faced a continuous rise in cost‐of‐living pressures (IPA 2022) alongside a notable decline in mental health (Butterworth et al. 2022). Although the inflation rate peaked at 7.8% in December 2022 and eased to 3.6% by the March 2024 quarter, it remains above the target range of 2%–3%. Correspondingly, mental and behavioural disorders have increased from 9.6% of Australians aged 15 years and over in 2001 to 21.5% in 2022 (ABS, 2022). ¹ Financial hardship can be a profoundly disheartening experience, leading to disengagement with detrimental effects on an individual's long‐term economic interests (Gladstone et al. 2021). Given an ageing population and a tight labour market (Brown and Guttmann 2017), financial hardship is likely to persist, causing stress and anxiety for many Australians into the future. Further, individuals with better mental health conditions are expected to be more employable as compared to their peers with mental illnesses and contribute more towards the economy's growth prospects (Frijters et al. 2014). Considering that Australians are already facing cost‐of‐living pressures ² and an ongoing mental health crisis since the COVID‐19 pandemic (Nghiem et al. 2020), it justifies a case to investigate whether financial behaviour can significantly affect mental health. Our findings suggest that cultivating regular savings habits, combined with timely credit card bill payments, have a positive impact on mental health outcomes. Our findings have significant policy implications for future investment, employment, worker productivity, and the well‐being of retirees.

Fourth, we conduct multiple heterogeneity tests to understand the impact of financial behaviours on mental health morbidity. In particular, whether the effect of financial behaviour on mental health varied with gender and socioeconomic status. Our results indicate that the impact of positive savings behaviour on mental health is stronger in men than in women. Although our results examining the impact of socioeconomic status on the relationship between financial behaviour and mental health show significant variations, a coefficient difference test reveals that the difference is not statistically significant between individuals with high versus low socioeconomic status.

We perform a battery of robustness checks using other indicators of subjective wellbeing as possible mechanisms, such as vitality, social functioning, and role‐emotional (J. E. J. Ware 2000), and their relationship with financial behaviour. When examining the effect of various demographic and socioeconomic variables on mental health in the Australian context, previous studies have argued that the effect may vary for these subjective wellbeing measures (Awaworyi Churchill et al. 2019; Hemingway et al. 1997). Moreover, as argued by Bialowolski et al. (2021), financial fragility is also associated with the general, emotional, and social wellbeing status of individuals. The existing literature also shows that greater social wellbeing, functional wellbeing, and emotional coping are associated with financial wellbeing (Barrett et al. 2021; Hoffman et al., 2022; Sabri et al. 2020). In addition to mental health, robust associations have been reported between financial stress and various domains of subjective well‐being, as measured by the SF‐36 questionnaire (Sturgeon et al. 2016; Steptoe et al. 2020). Our findings are broadly consistent with this literature, showing that positive financial behaviour is associated with higher vitality, social functioning, role‐emotional, and general well‐being scores.

Finally, to test if there are unobserved confounders, we conduct the Oster parameter stability test and conclude that omitted variable bias is not a significant concern in our analysis. The Kinky Least Squares estimation is also utilised to strengthen the causal relationship between financial behaviour and mental health in the presence of endogeneity. In addition, the strength and precision of the instrumental variable are tested by adjusting the standard errors of the coefficients, following the recommendations by Lee et al. (2022). The results are found to be unlikely to be affected by weak instrument bias.

The paper is organised as follows. Section 2 describes the methods, including the participants, procedure and key variables. Section 3 presents the statistical analyses. Section 4 discusses the empirical results. Finally, Section 5 concludes.

2. Methods

2.1. Participants and Procedure

Participants for this study are drawn from the Household, Income and Labour Dynamics in Australia (HILDA) survey. HILDA is a longitudinal survey database that tracks over 17,000 Australian residents aged 15 years and above on a yearly basis. It has been used since 2001 to collect information on socioeconomic status, physical and mental health, labour market dynamics, family conditions and life experiences. ³ This study utilises 22 waves of the HILDA survey, up to 2022. The present sample comprises a longitudinal panel of individuals aged between 15 and 101 years, identified across different survey waves, and includes the relevant measures of financial behaviour under analysis (see Appendix A1). In the sample, the total number of unique individuals with at least one valid saving habit observation is 20,637, and with at least one valid credit card payment observation is 12,176.

The following demographic variables are collected: gender, age, place of residence, home ownership, education level, employment status, serious physical health problems in the previous 12 months, household size, marital status, frequency of cigarette smoking, alcohol consumption, locus of control ⁴ , and socioeconomic status. ⁵ These demographic variables are reported to significantly impact mental health in the presence of financial distress, and substantiated by findings from Australian‐based studies (see Awaworyi Churchill and Smyth 2021; Johnston et al. 2016; Mahuteau and Zhu 2016).

2.2. Key Variables

2.2.1. Financial Behaviour

Two important financial behaviours used in this study are an individual's debt and savings behaviours. Credit card payment behaviour data are available for 15 out of 22 HILDA survey waves, and savings behaviour data for 13 out of 22 waves. ⁶ Using five‐point scales the following questions are used to assess: debt, ‘How often is the entire balance on all your credit cards paid off each month?’ (1 = ‘Pays off entire balance hardly ever/never’. to 5 = ‘Pays off entire balance always/almost always.’); and savings, ‘Which of the following statements comes closest to describing your (and your family's) savings habits?’ (1 = ‘Don't save: usually spend more than income’. to 5 = ‘Save regularly by putting money aside each month’.). To match the scales of the outcome variables used in the present study, the savings and credit card variables are normalised within a range between 0 and 1. A higher value indicates that an individual pays more to reduce their credit card debt or saves more regularly.

2.2.2. Mental Health

The primary measure of mental health used in the present study is the five‐item Mental Health score (MHI‐5), which assesses anxiety and depression (Awaworyi Churchill et al. 2019). The MHI‐5 is one of the four mental health wellbeing scales in the Short‐Form Health Survey questionnaire, also known as the SF‐36. The SF‐36 health questionnaire contains 36 items, which are used to construct four physical health scales (10‐item Physical Functioning, 4‐item Role Physical, 2‐item Bodily Pain, and 5‐item General Health), four mental wellbeing scales (4‐item Vitality; 2‐item Social Functioning; 3‐item Role Emotional, and 5‐item MHI‐5) and a single item measuring health transition (J. E. Ware and Gandek 1998). Higher scores represent better health‐related quality of life. In the HILDA survey, the scale scores are standardised to range from 0 to 100. To match the scales of the main explanatory variables, the item scores in the present study are divided by 100 and normalised within a range of 0–1. Based on data collected from the first HILDA survey wave, the SF‐36 has been confirmed as a valid and psychometrically robust measure of mental health (Butterworth and Crosier 2004).

To allow for robustness testing, we also examine the relationship between financial behaviour and the remaining mental health scales (vitality, social functioning, and role‐emotional) and the 36‐item composite general wellbeing score. The additional scale scores are normalised within a range of 0–1.

3. Statistical Analyses

A panel fixed effect (FE) model is used to test the impact of financial behaviour on mental health:

$${MentalHealth}_{it}={FinancialBehaviour}_{it}\beta +X_{it}\mathit{\gamma }+u_i+{\delta }_s+{\mu }_t+{\epsilon }_{it}$$ (1)

${MentalHealth}_{it}$ is our main outcome variable measured using MHI‐5 (as detailed in Section 2). ${FinancialBehaviour}_{it}$, is the savings or credit card payment pattern of the individual $i$ at time $t$. As mentioned above, both ${MentalHealth}_{it}$ and ${FinancialBehaviour}_{it}$ ranges between 0 and 1. In the model, the coefficient β represents the effect of mental health due to a 1% point increase in savings patterns towards regular savings or the ability to make credit card payments towards full payments. In the FE estimations, we use all available observations for respondents between 15 and 101 years surveyed by HILDA. The sample shows more respondents for the savings question (150,151 observations) as compared to the credit card question (42,936 observations) over the years. The control variables $\left(X_{it}\right)$ consists of age, gender, level of education dummies, income level, employment status, dummies for marital status, household size, whether living in a major city, accommodation type, whether a drinker or smoker, has a major illness or personal injury, and locus of control. $u_i$ denotes unobserved individual fixed effects for potentially confounding influences associated with temporal variations, ${\delta }_s$ captures the state fixed effects or the unobserved time‐invariant characteristics at the state level that may influence mental health outcomes. These are factors that vary across Australian states but remain relatively constant over time, such as differences in economic conditions, healthcare availability, social policies, or cultural factors that could otherwise confound the relationship between financial behaviour and mental health (e.g., Churchill and Ivanovski 2020; Ivanovski and Awaworyi Churchill 2021). ${\mu }_t$ denotes the time‐fixed effects, and ${\epsilon }_{it}$ is the error term.

Although the panel FE model can control for unobserved heterogeneity among individuals and reduce omitted variable bias in the estimation strategy, the relationship between financial behaviour and mental health may be plagued because of endogeneity biases, arising from reverse causality, measurement errors, and omitted variable bias. Existing literature suggests that psychological factors, such as hope and coping, can directly impact financial behaviours, including mitigating low savings and high debt, particularly during times of crisis (Arya et al. 2023). Thus, the reverse causality is evident, as poor mental health can lead to poor financial decisions (Arya et al. 2023; Johnston et al. 2021). Financial behaviour (${FianncialBehaviour}_{it}$) maybe correlated with unobserved individual heterogeneity ($u_i)$ and the error term (${\epsilon }_{it})$, giving rise to endogeneity biases. The correlation between ${FianncialBehaviour}_{it}$ and $u_i$ can be addressed using the fixed‐effect panel estimation technique. However, the correlation between ${FianncialBehaviour}_{it}$ and ${\epsilon }_{it}$ may still exist due to unobservable measurement errors and reverse causality. To mitigate the endogeneity biases, we employ a fixed effect instrumental variable (FE‐IV) approach.

To address endogeneity biases, specifically, we estimate the following two equations:

$${FinancialBehaviour}_{ist}=\left({Price}_{st}\times {distance}_{it}\right){\theta }_1+X_{it}\mathit{\lambda }+{\upsilon }_i+{\delta }_s+{\mu }_t+{ϵ}_{ist}$$ (2)

$${MentalHealth}_{it}=\widehat{{FinancialBehaviour}_{ist}}\beta +X_{it}\mathit{\gamma }+u_i+{\delta }_s+{\mu }_t+{\epsilon }_{it}$$ (3)

In the first stage (Equation (2)), (${Price}_{st}$ × ${distance}_{it}$) is the instrumental variable for financial behaviour (${FinancialBehaviour}_{it}$). The ${Price}_{st}$ is the state‐level price index, which is calculated using the average annual prices of automotive fuel, rents, household gas, electricity, and water and sewage costs in a specific state (Awaworyi Churchill and Smyth 2021). ⁷ The ${distance}_{it}$ is the distance to the pension entitlement age variable, which is calculated by first generating a cohort‐specific pension eligibility age (based on Table 1 in Clark and Zhu (2024)) and then computing the difference between this age and the individual's age. Finally, the IV is constructed as the interaction between the average state utility prices and the cohort‐specific distance to retirement age. In the second stage (Equation (3)), we replace ${FinancialBehaviour}_{it}$ with its predictor $\widehat{{FinancialBehaviour}_{ist}}$ obtained in the first stage. Standard errors are clustered at the individual level to account for the within‐person serial correlations across the survey waves. In the FE‐IV estimations, our IV (${Price}_{st}$ × ${distance}_{it}$) utilises the ${distance}_{it}$ variable, which is calculated by first generating a cohort‐specific pension eligibility age and then computing the difference between this age and the individual's age. Thus, by construction, this value would be negative for anyone above their pension eligibility age. Since our main focus is on capturing the financial burden of individuals based on their distance to the pension eligibility age, all negative values are set to zero in the first‐stage estimations, as they have already crossed the retirement age.

TABLE 1.

Summary statistics.

Variable	Obs	Mean	Std. Dev.	Min	Max
Saving habits, normalised	173,657	0.593	0.303	0	1
Credit card payment, normalised	130,633	0.757	0.376	0	1
Mental health measures
Mental health (MHI‐5)	303,210	0.733	0.176	0	1
Vitality	303,263	0.590	0.201	0	1
Social functioning	304,687	0.814	0.241	0	1
Role‐emotional	300,102	0.814	0.342	0	1
SF‐36 mental health	298,907	0.739	0.201	0	1
Age (years)	337,646	44.914	18.846	15	101
Gender	337,646	0.527	0.499	0	1
Annual income (million AUD)	336,636	0.050	0.075	0	7.049
Education
Year 12 or below	337,830	0.461	0.498	0	1
Postgrad	337,646	0.047	0.211	0	1
Graduate diploma	337,646	0.053	0.224	0	1
Bachelor	337,646	0.137	0.344	0	1
Diploma	337,646	0.090	0.286	0	1
Certificate	337,646	0.212	0.409	0	1
Household size	337,646	2.874	1.474	1	17
Marital status
Single	337,646	0.339	0.473	0	1
Separated	337,646	0.031	0.174	0	1
Divorced	337,646	0.088	0.283	0	1
Widowed	337,646	0.049	0.216	0	1
Living in a major city	337,646	0.616	0.486	0	1
Own accommodation	337,179	0.681	0.466	0	1
Smoker dummy	288,993	0.186	0.389	0	1
Drinker dummy	173,244	0.525	0.499	0	1
Employment status	337,646	0.634	0.482	0	1
Serious physical health	288,170	0.09	0.286	0	1
Locus of control	318,640	5.474	1.277	1	7
Socio‐economic status	337,506	0.495	0.5	0	1
Price × distance (IV) a	337,646	0.309	0.24	0	1

Note: All variables are defined in Section 2 of the main manuscript.

^a Price × Distance is the instrumental variable (IV) for financial behaviour, defined as the product of the average state utility prices and the cohort‐specific distance to retirement age.

To be a valid instrument, the IV must satisfy the relevance and exclusion conditions. The relevance condition requires the instrument to be strongly correlated with ${FinancialBehaviour}_{it}$, which is tested in the first stage of the FE‐IV approach. In our case, the indirect effect on mental health is from the financial burden that an individual faces for varying utility bills and prices weighted by their distance to retirement age. People with high financial liability are more likely to suffer from anxiety and mental distress (Zamanzadeh et al. 2024). Since utility bills directly impact the standard of living, they limit the ability to save or spend on various consumption items on an ongoing basis. Moreover, the price effect of an individual's savings and debt behaviours varies based on their general consumption pattern, which is age‐dependent. The distance to retirement age imposes an additional load on the price level in terms of its impact on financial behaviour. The higher the load (longer distance to retirement), the higher the financial burden of prices on their financial behaviours. This follows from the life cycle income hypothesis (Ando and Modigliani 1963), which states that individuals seek to smooth consumption over their lifetime income by accumulating more debt (fewer savings) during their initial working years when income is low and accumulate more savings (less debt) in latter phase of their working life and just before retirement when income is high. The interaction of these two variables, namely the state‐level utility price index (${Price}_{st}$) and their retirement age distance (${distance}_{it}$), is a new weighted price index variable that shows the extent state‐level prices impact individuals' savings and credit card payment behaviours relative to their overall consumption demands.

It is unlikely that state‐level changes in utility prices will have a direct impact on an individual's mental health, outside of the financial burden channel, thus satisfying the exogeneity condition (Awaworyi Churchill and Smyth 2021; Srivastava and Trinh 2021). Additionally, we control for other known mechanisms through which state‐level prices can impact mental health, such as income level, household size, employment status and serious physical health issues. Since a large, unexpected electricity or gas bill would significantly impact savings and credit card payment levels, financial behaviour satisfies the relevance condition, showing the mechanism through which the IV affects an individual's mental health. Moreover, changes in financial behaviour are unexpected and cannot be foreseen after controlling for individual fixed effects. Thus, after controlling for the known confounders, the IV is less likely to have a direct effect on ${Mentalhealth}_{it}$ beyond its indirect effect through the channel of ${FinancialBehaviour}_{it}$, and satisfy the exclusion criterion. Additionally, the strength of the instrument is checked by comparing the Kleibergen‐Paap (^KP) F‐statistic following Kleibergen and Paap (2006) and Cragg‐Donald (CD) F‐statistic following Staiger and Stock (1997) and Lee et al. (2022) against its valid threshold levels. As such, the $\beta$ estimated in Equation (3) reveals the causal impact of financial behaviour on mental health.

To test the reverse causality, we also consider the reciprocal relationship where mental health is the explanatory variable and financial behaviour is our main outcome variable. Existing literature (Frijters et al. 2014; Mitrou et al. 2023) has argued that the death of a close friend (${DeathFriend}_{it}$) is a valid IV for mental health, particularly for the same set of individuals from the HILDA database. It is a dummy variable that indicates whether a person has experienced the death of a close friend in the past 12 months. In our setting, the death of a close friend is a significant negative life event and expected to negatively impact mental health (Mitrou et al. 2023), but is unlikely to have a direct impact on financial behaviours. We also verify that the instrumental variable meets the relevance and exclusion criteria. We find that ${DeathFriend}_{it}$ is strongly correlated with ${MentalHealth}_{it}$ using the first stage of the FE‐IV approach. In the second stage, we then replace ${Mentalhealth}_{it}$ with its predictor $\widehat{{mentalhealth}_{it}}$ obtained in the first stage to check its effect on ${FinancialBehaviour}_{it}$.

4. Results

4.1. Demographic and Behavioural Statistics

Demographic statistics are reported in Table 1. The average age of respondents is 44.9 years, with an average total income of $50,476 (∼0.050 million AUD). For education level, Australians completing the survey have predominantly completed a certificate course (21.2%), followed by a bachelor's degree (13.7%), as compared to graduate diplomas (5.3%) and postgraduate degrees (4.7%). The average household size is 2.9 individuals, with women representing 52.7% of the respondents in the sample. Most respondents live in a major capital city (61.1%) and own their homes (68.1%). In addition, while 18.6% are regular cigarette smokers, 52.5% consume alcohol regularly. Finally, 63.4% are employed, 9% report experiencing serious physical health problems in the 12 months preceding the survey, and 49.5% come from middle‐to high‐socioeconomic backgrounds.

The normalised financial behaviour and mental health values are reported in Table 1. The mean value of savings behaviour is 0.593, and credit card payment behaviour is 0.757, suggesting that credit card payment is more regular for those who own a credit card than savings behaviour. A mean value of 0.593 would indicate that an average individual in the sample is close to scale item 3 (Save whatever is left over ‐ no regular plan). This means that, qualitatively, an average individual would either have no regular savings plan (scale item 3) or spend their regular income but save their additional income (scale item 4). Similarly, a mean value of 0.733 for the credit card payment variable would mean that an average individual in the sample generally pays off the entire balance in most months (scale item 4). In the mental health category, the average MHI‐5 score is 0.733, which is above the median value of 0.5 on a scale of 0–1, indicating good mental health scores for this Australian sample. The mean scores for the remaining mental health scales also fall in the middle to high range on a scale of 0–1, with vitality at 0.590, social functioning at 0.814, role‐emotional at 0.814, and a SF‐36 score of 0.739.

The distribution of respondents' mental health outcomes and financial behaviours based on gender and socioeconomic status is reported in Table 2. Gender and socioeconomic differences in financial behaviour and mental health are examined using two‐sample t‐tests for mean comparisons. The t‐test results reveal significant differences in gender and socioeconomic status with regard to financial behaviour and mental health. Men are more likely to report regular payments on credit card debt, but less likely to have regular savings habits, while individuals from higher socioeconomic backgrounds are significantly more likely to report regular savings and low debt habits. Both men and individuals from higher socioeconomic backgrounds are more likely to report better mental health, as indicated by higher MHI‐5, vitality, social functioning, role‐emotional, and SF‐36 scores.

TABLE 2.

Sample distribution.

	Gender			Socio‐economic status
	Men	Women	Difference a	High	Low	Difference a
MHI‐ 5	0.747	0.721	0.026***	0.744	0.722	0.0230***
	(0.171)	(0.18)		(0.168)	(0.183)
Vitality	0.614	0.569	0.045***	0.604	0.576	0.028***
	(0.193)	(0.205)		(0.196)	(0.205)
Social	0.833	0.798	0.035***	0.839	0.789	0. 050***
	(0.231)	(0.248)		(0.223)	(0.255)
Emotional	0.837	0.793	0.044***	0.836	0.792	0.044***
	(0.324)	(0.355)		(0.322)	(0.359)
SF‐36 general mental	0.759	0.721	0.037***	0.757	0.721	0.036***
	(0.191)	(0.207)		(0.189)	(0.21)
Regular savings habits	0.589	0.597	0.008***	0.618	0.569	0. 049***
	(0.299)	(0.306)		(0.301)	(0.303)
Credit card payment	0.766	0.748	0.017***	0.789	0.713	0.076***
	(0.37)	(0.382)		(0.356)	(0.399)

Notes: The table shows the summary statistics of financial behaviours and mental health outcomes by gender and socio‐economic status. Standard deviations are reported in parenthesis.

^a Two‐sample mean com‐parison t‐test. Significance ***1%, **5%, *10%.

Overall, the descriptive statistics reveal significant variations in demographic, health, financial, and socioeconomic indicators, indicating a diverse and representative sample.

4.2. Baseline Results

The estimated coefficients of the panel fixed effects model for the association between financial behaviour and mental health are presented in Table 3. Columns 1 and two of Table 3 contain the estimated coefficients for savings versus debt behaviours after controlling for the time‐varying covariates of mental health in Equation (1), while columns 3 and 4 contain estimated coefficients after controlling for year and state‐fixed effects. Columns 5 and 6 also include the individual fixed effects and represent the most reliable estimates as they control for individual‐specific characteristics. We note that the total number of observations for debt (credit card payments) is significantly lower (42,936) as compared to savings behaviour (150,151) due to missing observations. Robust standard errors clustered at the individual respondent level are employed in all specifications.

TABLE 3.

The effect of financial behaviour on mental health: FE estimates.

	(1 ‐ OLS)	(2 ‐ OLS)	(3 ‐ FE)	(4 ‐ FE)	(5 ‐ FE)	(6 ‐ FE)
Regular savings habits	0.0383***		0.0421***		0.0321***
	(0.0015)		(0.0015)		(0.0016)
Credit card payment		0.0206***		0.0226***		0.0124***
		(0.0023)		(0.0023)		(0.0029)
Age	0.0004***	0.0006***	0.0013***	0.0012***	−0.0033*	−0.0018
	(0.0001)	(0.0001)	(0.0001)	(0.0001)	(0.0018)	(0.0036)
Gender	−0.0283***	−0.0207***	−0.0278***	−0.0193***	−0.0796
	(0.0017)	(0.0023)	(0.0017)	(0.0023)	(0.0774)
Marital status
Single	−0.0128***	−0.0170***	−0.0045***	−0.0111***	0.0059***	0.0058
	(0.0018)	(0.0028)	(0.0017)	(0.0028)	(0.0022)	(0.0042)
Separated	−0.0376***	−0.0388***	−0.0366***	−0.0376***	−0.0327***	−0.0294***
	(0.0032)	(0.0051)	(0.0032)	(0.0051)	(0.0035)	(0.0061)
Divorced	−0.0069***	−0.0099***	−0.0064**	−0.0096***	0.0023	0.0032
	(0.0025)	(0.0036)	(0.0025)	(0.0035)	(0.0033)	(0.0055)
Widowed	−0.0096***	−0.0121**	−0.0131***	−0.0157***	−0.0195***	−0.0294***
	(0.0035)	(0.0057)	(0.0034)	(0.0056)	(0.0043)	(0.0083)
Log (Income)	−0.0028***	−0.0006	−0.0022***	0.0005	−0.0023***	−0.0001
	(0.0002)	(0.0006)	(0.0002)	(0.0006)	(0.0003)	(0.0007)
Education
Postgrad	−0.0085***	−0.0093**	0.0062*	0.0002	−0.0081*	−0.0067
	(0.0033)	(0.0044)	(0.0033)	(0.0044)	(0.0049)	(0.0102)
Graduate diploma	−0.0034	0.0002	0.0059*	0.0062	−0.0099**	0.0032
	(0.0032)	(0.0038)	(0.0032)	(0.0038)	(0.0048)	(0.0088)
Bachelor	−0.0025	−0.0014	0.0059***	0.0045	−0.0069**	−0.0007
	(0.0021)	(0.0031)	(0.0021)	(0.0031)	(0.0031)	(0.0082)
Diploma	−0.0032	−0.0024	0.0029	0.0020	−0.0092**	−0.0052
	(0.0029)	(0.0037)	(0.0028)	(0.0037)	(0.0045)	(0.0092)
Certificate	−0.0053***	0.0000	0.0007	0.0044	−0.0028	0.0046
	(0.0020)	(0.0030)	(0.0020)	(0.0030)	(0.0029)	(0.0068)
Living in major city	−0.0040***	−0.0057***	−0.0054***	−0.0068***	−0.0046**	−0.0043
	(0.0015)	(0.0021)	(0.0015)	(0.0022)	(0.0023)	(0.0043)
Own accommodation	0.0136***	0.0129***	0.0116***	0.0111***	0.0064***	0.0063**
	(0.0013)	(0.0022)	(0.0013)	(0.0022)	(0.0014)	(0.0028)
Household size	−0.0010**	−0.0034***	−0.0003	−0.0026***	−0.0011**	−0.0046***
	(0.0004)	(0.0007)	(0.0004)	(0.0007)	(0.0005)	(0.0009)
Smoker dummy	−0.0201***	−0.0194***	−0.0225***	−0.0206***	−0.0119***	−0.0079**
	(0.0016)	(0.0028)	(0.0016)	(0.0028)	(0.0019)	(0.0040)
Drinker dummy	0.0088***	0.0095***	0.0102***	0.0115***	0.0074***	0.0045***
	(0.0008)	(0.0015)	(0.0008)	(0.0015)	(0.0009)	(0.0017)
Employment status	0.0182***	0.0149***	0.0172***	0.0136***	0.0094***	0.0038
	(0.0013)	(0.0023)	(0.0013)	(0.0023)	(0.0014)	(0.0026)
Serious physical health in the past 12 months	−0.0473***	−0.0498***	−0.0475***	−0.0502***	−0.0398***	−0.0366***
	(0.0014)	(0.0028)	(0.0014)	(0.0028)	(0.0015)	(0.0030)
Locus of control	0.0154***	0.0165***	0.0157***	0.0167***	0.0089***	0.0084***
	(0.0006)	(0.0008)	(0.0006)	(0.0008)	(0.0008)	(0.0010)
Individual FE	No	No	No	No	Yes	Yes
Year FE	No	No	Yes	Yes	Yes	Yes
State FE	No	No	Yes	Yes	Yes	Yes
Observations	150,151	42,936	150,151	42,936	150,151	42,936

Notes: The dependent variable is scaled MHI‐5 and varies from 0 to 1. Because of perfect collinearity with the individual fixed effects, the gender variable is omitted from specification 6 for credit card payments. The sample consists of individuals between 15 and 101 years old who have responded to the credit card and savings questions. The reference category for marital status is individuals who are married, or in a de facto relationship, and for education levels, it is those whose highest level of education is year 12 or below. Robust standard errors clustered at the individual respondent level are reported in parentheses.

^* p < 0.10.

^** p < 0.05.

^*** p < 0.01.

Analyses reveal that the estimated coefficients for financial behaviours are statistically significant at a 1% significance level, demonstrating that regular savings or low debt behaviours are associated with better mental health. As the preferred specification estimates from columns 5 and 6 suggest, on average, a 1% point increase in regular savings is associated with a 0.032% point improvement in mental health after controlling for all the determinants of mental health. Similarly, a 1% point increase towards regular credit card payments (lower debt) is associated with a 0.012% point improvement in mental health. It is noted that the initial FE estimates in columns 1 to 4 do not change significantly when all fixed effects are not accounted for, suggesting that the results are robust.

The signs of the control variables are as expected. Specifically, higher age (specification 5), higher income, having a higher education degree as compared to Year 12 or below, higher household size, smoking cigarettes, separating from a spouse, or being widowed as against married, living in a capital city, and having a serious physical health condition are associated with lower mental health. In contrast, drinking alcohol, living in one's own accommodation, having a higher locus of control, and being employed are associated with better mental health.

4.3. Endogeneity

The panel FE results can suffer from omitted variable bias, measurement errors, and simultaneity bias arising from unobserved individual heterogeneity ($u_i)$. The FE‐IV approach is applied to control for endogeneity arising from reverse causality and the correlation of the financial behaviour measure with the error term (${\epsilon }_{it})$. As such, the Australian state‐level price index multiplied by the distance to the old age pension is used as an instrument for financial behaviour in the first stage and the predicted financial behaviour measure in the second stage.

The FE‐IV estimates are presented in Table 4. Panel A and Panel B report the second‐stage and first‐stage estimates, respectively. Similar to the adjustments made to the panel fixed effect regressions reported in Table 3, time‐varying covariates are also controlled for in all specifications. Further, robust standard errors clustered at the individual respondent level are employed in all the regressions. Since the IV is based on state‐level prices, state‐fixed effects are excluded in Specifications 1 and 3. In contrast, state‐fixed effects, along with time and individual effects, are included in Specifications 2 and 4, which are our preferred estimates. ⁸

TABLE 4.

The effect of financial behaviour on mental health: FE‐IV estimates.

	(1)	(2)	(3)	(4)
Dependent variable	Scaled mental health (Scaled MHI‐5)
Panel A: Second‐stage estimates
Regular savings habits	0.4783***	0.4752***
	(0.0544)	(0.0538)
Credit card payment			0.5027***	0.5067***
			(0.1249)	(0.1260)
Gender	0.0095	0.0097
	(0.0873)	(0.0849)
Marital status
Single	−0.0120***	−0.0119***	0.0003	0.0003
	(0.0036)	(0.0036)	(0.0074)	(0.0074)
Separated	−0.0177***	−0.0180***	−0.0245**	−0.0248***
	(0.0047)	(0.0046)	(0.0096)	(0.0096)
Divorced	0.0042	0.0041	0.0243**	0.0240**
	(0.0042)	(0.0041)	(0.0101)	(0.0101)
Widowed	−0.0243***	−0.0243***	−0.0213*	−0.0215*
	(0.0055)	(0.0055)	(0.0117)	(0.0117)
Log (Income)	−0.0029***	−0.0029***	−0.0021*	−0.0020*
	(0.0003)	(0.0003)	(0.0011)	(0.0011)
Education
Postgrad	−0.0144**	−0.0140**	−0.0085	−0.0091
	(0.0065)	(0.0065)	(0.0164)	(0.0165)
Graduate diploma	−0.0143**	−0.0141**	0.0083	0.0078
	(0.0067)	(0.0067)	(0.0152)	(0.0152)
Bachelor	−0.0075*	−0.0073*	−0.0018	−0.0021
	(0.0042)	(0.0041)	(0.0134)	(0.0134)
Diploma	−0.0094	−0.0096*	−0.0027	−0.0027
	(0.0058)	(0.0058)	(0.0150)	(0.0150)
Certificate	−0.0017	−0.0017	0.0138	0.0135
	(0.0038)	(0.0037)	(0.0109)	(0.0109)
Living in major city	−0.0071**	−0.0082***	0.0081	0.0039
	(0.0030)	(0.0031)	(0.0070)	(0.0071)
Own accommodation	0.0024	0.0022	−0.0031	−0.0033
	(0.0019)	(0.0019)	(0.0050)	(0.0050)
Household size	0.0013*	0.0013*	−0.0020	−0.0020
	(0.0007)	(0.0007)	(0.0016)	(0.0016)
Smoker dummy	0.0016	0.0016	0.0041	0.0041
	(0.0029)	(0.0029)	(0.0068)	(0.0068)
Drinker dummy	−0.0302***	−0.0299***	−0.0207***	−0.0209***
	(0.0047)	(0.0047)	(0.0069)	(0.0070)
Employment status	−0.0140***	−0.0137***	0.0180***	0.0182***
	(0.0034)	(0.0034)	(0.0051)	(0.0051)
Serious physical health	−0.0388***	−0.0388***	−0.0354***	−0.0354***
	(0.0018)	(0.0018)	(0.0040)	(0.0040)
Locus of control	0.0072***	0.0072***	0.0078***	0.0078***
	(0.0010)	(0.0010)	(0.0015)	(0.0015)
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes

Dependent variable	(1)	(2)	(3)	(4)
	Regular savings habits	Regular savings habits	Paid credit card payment	Paid credit card payment
Panel B: First‐stage estimates
Cohort‐specific distance to retirement age*State level price	−0.3183***	−0.3206***	−0.3393***	−0.3381***
	(0.0262)	(0.0263)	(0.0678)	(0.0678)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes
Cragg‐donald F‐stat	252.819	256.199	38.947	38.631
KP F‐Statistic	147.048	148.654	25.038	24.869
Observations	147,952	147,952	39,016	39,016

Notes: ‘Controls’ include all potential covariates such as gender, marital status indicators, ln(income), education levels, employment status, living in a major city, household size, own accommodation, serious physical health, smoker dummy, drinker dummy and locus control. Most of these “Controls” are statistically significant in both in the first and second stage. Because of perfect collinearity with the individual fixed effects in the second stage estimates, the gender variable is omitted from specifications (3) and (4) for credit card payments. Robust standard errors clustered at individual levels are in parentheses.

^* p < 0.10.

^** p < 0.05.

^*** p < 0.01.

The first‐stage estimates (Panel B) show that the IV is negative and statistically significant for both the savings and debt variables. The result is expected because a positive increase in the price level would increase the financial burden and negatively influence stable financial behaviours (lower savings and higher debt). This follows the life cycle income hypothesis, which posits that young adults are expected to have lower savings and higher debt, driven by multiple motives, compared to older adults who are closer to retirement age. Thus, young adults (farther away from retirement age) exhibit a higher probability of low savings and unpaid credit card bills, which is expected to have a negative impact on stable financial behaviours. As a result, the combined effect (state price level multiplied by cohort‐specific age distance) is expected to have a negative effect on stable financial behaviours. Furthermore, the CD and KP F‐statistics for savings in Specifications 1 and 2 are well above the threshold level, illustrating that the IV is strong. For Specifications 3 and 4 (debt behaviour), the CD and KP F‐stats are also above the threshold level; however, they are lower than Specifications 1 and 2, likely because the number of observations for debt behaviour is significantly lower than for savings. We find that when the number of observations is lower, the IV tends to be weaker (Murray 2006; Stock and Watson 2020); however, it still satisfies the identification condition of a valid instrument.

The FE‐IV results in Panel A of Table 4 show that stable financial behaviours significantly improve mental health morbidity. A percentage point increase in regular savings habits increases mental health by 0.475% points in specification 2 when all fixed effects are accounted for. Similarly, a percentage point increase in regular payments towards credit card balance increases mental health by 0.507% points in specification 4. ⁹ The findings confirm that financial behaviour predicts mental health after controlling for endogeneity and is consistent with the literature. ¹⁰ For instance, savings and access to credit facilities are reported to enhance the psychological health of university employees (Ekore and Omisore 2013). Similarly, difficulty in paying bills and credit card debt is associated with depression and anxiety (Marshall et al. 2021), while financial capability (e.g., borrowing in order to meet housing payments) has a significant effect on psychological health over and above that associated with income and material wellbeing (M. P. Taylor et al. 2011). Table 4 also illustrates that mental health has a positive and statistically significant association with living in one's own accommodation (specification 2) and a high locus of control. Conversely, having serious physical health issues, being an alcoholic, having higher income or education, and being separated from marriage are negatively associated with mental health in most of the specifications.

As explained in Section 3 earlier, identifying the effect of financial behaviour on mental health may be plagued by reverse causality. Earlier studies have found that poor financial behaviours are caused by poor mental health, particularly at times of crisis (Arya et al. 2023). Thus, we test if mental health has a statistically significant impact on financial behaviour. For this purpose, we use an established IV for mental health, that is if the person has experienced the death of a close friend in the past 12 months (see Section 3 for detailed discussion). The reverse causality test result is presented in Appendix Table A2. Consistent with the literature, as the first stage in Panel B shows, the death of a close friend is a valid instrument for mental health (Frijters et al. 2014; Mitrou et al. 2023). However, the second stage estimated coefficients in all specifications of Panel A show that mental health does not have a statistically significant effect on financial behaviour. This suggests that reverse causality is not a concern in the baseline financial behaviour vis‐à‐vis mental health regression analysis.

4.4. Heterogeneity Analysis

Heterogeneity analyses are conducted to test whether financial behaviour has differential impacts across gender and socioeconomic status on mental health. The FE‐IV estimated coefficients of financial behaviour on mental health, segmented by gender, are given in Table 5. Columns 1 and 2, respectively, show the effects of regular saving habits and timely payment of monthly credit card balances on the mental health of women. Meanwhile, Columns 3 and four illustrate the impact of these financial behaviour measures on the mental health of men. The second stage, FE‐IV estimated results in Panel A, reveals that while the effect is positive and statistically significant for both genders, the influence of sound financial behaviour on mental health is notably stronger among men than women. While the coefficient of regular savings behaviour is 0.292 in column 1 for women, it is 0.771 in column 3 for men. Similarly, the coefficient of timely credit card payments in column 2 is 0.412, as compared to 0.629 in column 4. Following the recommendations of de Gendre et al. (2024), we calculate the coefficient difference t‐statistics between women and men. ¹¹ The difference between the magnitude of the coefficients for men and women is found to be statistically significant for savings behaviour; however, the between‐group differences are insignificant for credit card payments. Overall, the finding suggests that stable savings behaviour contributes more substantially to the mental health of Australian men than women. The first‐stage estimates in Panel B confirm that the IV has a significant effect on financial behaviours as expected. Furthermore, the CD and KP F‐statistics are larger than 10, indicating that the IV is valid.

TABLE 5.

Heterogeneity analysis: by Gender (FE‐IV estimates).

Dependent variable	(1)	(2)	(3)	(4)
	Mental health (Scaled MHI‐5)
	Women		Men
Panel A: Second‐stage estimates
Regular savings habits	0.2923***		0.7712***
	(0.0545)		(0.1259)
Credit card payment		0.4123***		0.6298***
		(0.1416)		(0.2258)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Coefficient difference test between men and women*	t‐statistic	p‐value
Regular savings habits	3.4908	0.0004
Credit card payment	0.8161	0.4145

Dependent variable	(1)	(2)	(3)	(4)
	Savings habits	Credit card payment	Savings habits	Credit card payment
Panel B: First‐stage estimates
Cohort‐specific distanceto retirement age*State level price	−0.3812***	−0.3735***	−0.2601***	−0.3150***
	(0.0364)	(0.0937)	(0.0381)	(0.0986)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Cragg‐donald F‐stat	189.537	24.866	80.248	15.690
KP F‐statistic	109.578	15.874	46.745	10.215
Observations	78,921	20,698	69,030	18,318

Note: ‘Controls’ in the first stage include all covariates listed in the second stage, such as age, gender, single, separated, divorced, widowed, ln(income), education, employment status, living in a major city, own accommodation, serious physical health, smoker dummy, & drinker dummy. Most of these “Controls” are statistically significant in the first stage. Robust standard errors clustered at individual levels are in parentheses. Following de Gendre et al. (2024), we calculate the coefficient difference t‐statistics as $t_{stat}=\raisebox{1ex}{$\left({\hat{\beta }}_{male}-{\hat{\beta }}_{female}\right)$}\!\left/ \!\raisebox{-1ex}{$\sqrt{\left({SE}_{{\hat{\beta }}_{male}}^2+{SE}_{{\hat{\beta }}_{female}}^2\right)}$}\right.$. The p‐value is computed as $p=2\ast \left(1-\mathrm{\Phi }\left(\left|t\right|\right)\right)$.

^* p < 0.10.

**p < 0.05.

^*** p < 0.01.

Next, we examine how the impact of financial behaviour varies on the mental health scores of individuals with low and high socio‐economic status. In HILDA, the socio‐economic status (socio‐economic advantage/disadvantage) index ranges from the first decile to the 10th decile, measuring “people's access to material and social resources and their ability to participate in society.” In this study, we have grouped the 1st to 5th deciles as coming from a low socio‐economic background and the 6th to 10th deciles as having a high socio‐economic background.

Table 6 presents the effect of financial behaviour on mental health for the low (Columns 1–2) and the high (Columns 3–4) socioeconomic background cohorts. The first‐stage coefficients and F‐statistics in Panel B confirm the validity of the instrument in three out of four specifications. Our second‐stage coefficients in Panel A show that both types of financial behaviours (regular savings and regular payment of credit cards) have a positive effect on the mental health of individuals with high and low socioeconomic backgrounds. However, savings behaviour has a higher positive effect on individuals from high socio‐economic backgrounds (0.559) than those from low socio‐economic backgrounds (0.373). In contrast, we find the opposite is true for timely credit card payments, albeit at a 10% significance level for individuals from low socioeconomic backgrounds. The coefficient of financial behaviour displays a lower significance level, likely because individuals from low socio‐economic backgrounds are more prone to erratic financial behaviour. To get a deeper understanding of the magnitude differences, following de Gendre et al. (2024), we conduct a between‐group difference test between the two socio‐economic cohorts. We find that the difference in the estimates of financial behaviours between the two cohorts is not statistically significant. This explains the contrasting findings between the credit card payments and savings behaviour measures. The insignificant coefficient difference test suggests that financial behaviour impacts mental health irrespective of an individual's socioeconomic background.

TABLE 6.

Heterogeneity analysis: by relative socio‐economic status (FE‐IV estimates).

Dependent variable	(1)	(2)	(3)	(4)
	Mental health (Scaled MHI‐5)
	Low socio‐economic status (1st to 5th decile)		High socio‐economic status (6th to 10th decile)
Panel A: Second‐stage estimates
Regular savings habits	0.3731***		0.5593***
	(0.0686)		(0.0969)
Credit card payment		0.6740*		0.4006***
		(0.3631)		(0.1169)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
Coefficient difference test between men and women*	t‐statistic	p‐value
Regular savings habits	1.5683	0.1168
Credit card payment	−0.7167	0.4735

Dependent variable	(1)	(2)	(3)	(4)
	Savings habits	Credit card payment	Savings habits	Credit card payment
Panel B: First‐stage estimates
Cohort‐specific distance to retirement age*State level price	−0.3452*** (0.0384)	−0.2521** (0.1199)	−0.2826*** (0.0382)	−0.4229*** (0.0864)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Cragg‐donald F‐stat	136.638	6.642	87.031	36.582
KP F‐statistic	80.834	4.418	54.676	23.939
Observations	72,271	15,182	72,911	22,021

Note: HILDA estimates the socioeconomic status of individuals by using the ABS statistical local area (SLA) index values. The data is assigned from the lowest decile (1) to the highest decile (10). Based on this, a socio‐economic status dummy variable is created, which takes the value one if it is above the median value of 0.5 and 0 otherwise. ‘Controls’ in the first stage include all covariates listed in the second stage, such as age, gender, single, separated, divorced, widowed, ln(income), education, employment status, living in a major city, own accommodation, serious physical health, smoker dummy, & drinker dummy. Most of these “Controls” are statistically significant in the first stage. Robust standard errors clustered at individual levels are in parentheses. Following de Gendre et al. (2024), we calculate the coefficient difference t‐statistics as $t_{stat}=\raisebox{1ex}{$\left({\hat{\beta }}_{LSLA}-{\hat{\beta }}_{HSLA}\right)$}\!\left/ \!\raisebox{-1ex}{$\sqrt{\left({SE}_{{\hat{\beta }}_{LSLS}}^2+{SE}_{{\hat{\beta }}_{HSLA}}^2\right)}$}\right.$. The p‐value is computed as $p=2\ast \left(1-\mathrm{\Phi }\left(\left|t\right|\right)\right)$. LSLA and HSLA refer to low and high socioeconomic cohorts, respectively.

^* p < 0.10.

^** p < 0.05.

^*** p < 0.01.

4.5. Robustness Checks

In this study, the main dependent variable is the MHI‐5 score, which reflects the propensity for anxiety and depression. However, the literature acknowledges that mental health is closely associated with other dimensions of quality of life, such as vitality, role‐emotional and social functioning (Hernandez et al. 2018; Thoits 2011; Bialowolski et al. 2024). These subjective wellbeing measures could serve as potential mechanisms through which financial behaviour may impact mental health. ¹² Given that we observed a significant relationship between financial behaviour and the MHI‐5 in the previous section, we can predict that financial behaviour will have a relationship with these other measures of subjective wellbeing. Previous studies using HILDA data have also considered these measures as a robustness check to mental health (Awaworyi Churchill et al. 2019). Thus, three alternative measures of mental wellbeing from the HILDA survey are considered: vitality, role‐emotional, social functioning, and the aggregate SF‐36 scores.

The results for vitality and role emotional scores are reported in Table 7, and social functioning and aggregate SF‐36 score (general wellbeing) are in Table 8. Comparing the results, we find that regular savings and timely credit card debt management habits have a positive impact on all four mental health measures. A one percentage point increase in regular savings habits increases vitality and role‐emotional health by 0.507 and 0.523% points, respectively, while social functioning and SF‐36 scores by 0.192 and 0.422% points, respectively. Similarly, one percentage point increase in regular credit card payment habits increases vitality and role‐emotional by 0.404 and 0.506% points, respectively, and social functioning and SF‐36 scores by 0.263 and 0.418% points, respectively. Thus, stable financial behaviours are observed to positively impact vitality and role‐emotional health scores, and to a lesser extent, social functioning and SF‐36 scores. In sum, the robustness checks (Tables 7 and 8) support the baseline findings (Table 4).

TABLE 7.

Alternative measures of mental wellbeing: Vitality and emotional scores (FE‐IV estimates).

Dependent variable	(1)	(2)	(3)	(4)
	Mental wellbeing
	Vitality		Emotional
Panel A: Second‐stage estimates
Regular savings habits	0.5069***		0.5231***
	(0.0602)		(0.0933)
Credit card payment		0.4042***		0.5060**
		(0.1197)		(0.2038)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes

Dependent variable	(1)	(2)	(3)	(4)
	Savings habits	Credit card payment	Savings habits	Credit card payment
Panel B: First‐stage estimates
Cohort‐specific distance to retirement age*State level price	−0.3207***	−0.3388***	−0.3208***	−0.3379***
	(0.0263)	(0.0678)	(0.0265)	(0.0679)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Cragg‐donald F‐stat	256.246	38.792	253.828	38.264
KP F‐statistic	148.670	24.976	147.050	24.796
Observations	147,961	39,013	146,352	38,691

Note: ‘Controls’ in the first stage include all covariates listed in the second stage, such as age, gender, single, separated, divorced, widowed, ln(income), education, employment status, living in a major city, own accommodation, serious physical health, smoker dummy, & drinker dummy. Most of these “Controls” are statistically significant in the first stage. Robust standard errors clustered at individual levels are in parentheses.

^* p < 0.10.

^** p < 0.05.

^*** p < 0.01.

TABLE 8.

Alternative measures of mental wellbeing: Social functioning and general wellbeing (FE‐IV estimates).

Dependent variable	(1)	(2)	(3)	(4)
	Mental wellbeing
	Social		General wellbeing (SF 36)
Panel A: Second‐stage estimates
Regular savings habits	0.1923***		0.4224***
	(0.0588)		(0.0553)
Credit card payment		0.2626**		0.4182***
		(0.1259)		(0.1212)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes

Dependent variable	(1)	(2)	(3)	(4)
	Savings habits	Credit card payment	Savings habits	Credit card payment
Panel B: First‐stage estimates
Cohort‐specific distance to retirement age*State level price	−0.3204***	−0.3369***	−0.3215***	−0.3399***
	(0.0263)	(0.0678)	(0.0264)	(0.0679)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Cragg‐donald F‐stat	256.217	38.384	254.594	38.694
KP F‐statistic	148.198	24.689	147.869	25.085
Observations	148,167	39,047	146,166	38,660

Note: ‘Controls’ in the first stage include all covariates listed in the second stage, such as age, gender, single, separated, divorced, widowed, ln(income), education, employment status, living in a major city, own accommodation, serious physical health, smoker dummy, & drinker dummy. Most of these “Controls” are statistically significant in the first stage. Robust standard errors clustered at individual levels are in parentheses.

^* p < 0.10.

^** p < 0.05.

^*** p < 0.01.

Next, to assess whether baseline results presented in Table 4 are influenced by omitted variable bias, which is a significant concern in an endogeneity test if the instrument is weak, parameter stability tests are undertaken as described by Oster (2019). Oster (2019) provides bound estimates on the unbiased ordinary least squares coefficients and concludes that if the coefficient of the key regressor is stable before and after the inclusion of the observed controls, then the omitted variable bias is limited. The method further suggests constructing an identifiable set by analysing the stability of coefficients and comparing the R‐squared values from regressions of the baseline model by first excluding and then systematically including the control variables. The null hypothesis is that omitted variables drive the results, which can be rejected if zero does not fall within the identifiable set. Oster (2019) recommends setting 1 as the upper bound for δ, indicating that omitted variables need as much influence as the included variables for the coefficient to become zero. Furthermore, we set the value of $R_{\mathit{max}}=1.3\stackrel{\sim }{R}$ where $\stackrel{\sim }{R}$ is the R‐squared of the controlled regression (Churchill and Smyth 2021). These results are presented in Table 9.

TABLE 9.

Parameter stability test and omitted variable bias.

Treatment variable	(1)	(2)	(3)	(4)
	Baseline effect, $\dot{\beta }$ [$\dot{R}$]	Controlled effect, $\stackrel{\sim }{\beta }$ [$\stackrel{\sim }{R}$]	Identified set $[\stackrel{\sim }{\beta }$, ${\beta }^{\ast }]$	$\bar{\delta }$ For $=0$ given $R_{\mathit{max}}$
Panel A: Saving habits
Savings habits	0.03289	0.03214	[0.03112, 0.03214]	11.8478
	[0.004]	[0.026]
N	150,151	150,151
Panel B: Credit card payment
Credit card payment	0.00959	0.01238	[0.01238, 0.01677]	−3.42556
	[0.000]	[0.022]
N	42,936	42,936
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes

Column 1 of Table 9 presents the baseline effect without controlling for the covariates, showing the estimated baseline effect ($\dot{\beta }$) and the associated R‐squared presented in the closed bracket. Column 2 presents the estimated coefficient when we control for the full set of covariates, and column 4 presents the ratio of the effect of unobservable covariates relative to the observed covariates that would drive the estimated effect of regular saving habit or credit card payment on mental health to zero. The coefficient signs for baseline (Column 1) and controlled (Column 2) estimates are the same for both financial behaviours, which shows our results are robust. Furthermore, the $\bar{\delta }$ values for regular savings habits and credit card payments are 11.8 and −3.4, respectively, which are significantly different from 1. For example, the first $\bar{\delta }$ value suggests that for the true effect of saving habits to be zero, unobservable covariates would have to be 11.8 times as important as the observed controlled variables in explaining the outcome. This is unlikely, suggesting that our results are robust to omitted variable bias. Similarly, the $\bar{\delta }$ value for credit card payments also displays robust behaviour. Therefore, our results withstand the omitted variable bias test, even in the presence of weak instruments for the endogenous variable.

To check whether normalising the financial behaviour variables between 0 and one obscures possible heterogeneity within the variables, two new dummy variables are created based on the reference groups of savings and credit card payments as defined in Appendix 1. A saving habit dummy variable is constructed where zero reflects no savings (reference group = 1 on the scale) and one for some type of savings made (reference group is between two to five on the scale). Likewise, a credit card payment dummy is constructed where zero reflects never paying off credit (reference group = 1 on the scale) and one for some payments towards the credit card balance (reference group is between two to five on the scale). The FE‐IV baseline regressions are re‐estimated with these new dummy variables (Appendix Table A3). The results show that regular savings and credit card payments positively impact the mental health of individuals. For example, compared to those with negative saving behaviour, individuals with regular saving behaviours have a mental health score that is 0.366 higher (in specification 2). These results are robust to our original findings in Table 4 and confirm that normalising the financial behaviour variables does not affect our main results.

To evaluate the strength of our endogeneity findings, we utilise the Kinky Least Squares (KLS) method (Kiviet 2013, 2020, 2023), which determines causal relationships by constraining the correlation between explanatory variables and the error term. In contrast to conventional FE‐IV estimation, KLS yields interval estimates, presenting a range within which the coefficients of endogenous variables may fluctuate. Figure 1 illustrates the KLS results for the two financial behaviours and mental health. According to this method, if measurement error is the primary source of endogeneity, the correlation between the endogenous variables and the error term will be negative, and the FE estimates will be biased downward due to attenuation bias (Kripfganz and Kiviet 2021). Following Churchill and Smyth (2022), the correlation between the endogenous variables and the predicted error terms is calculated, and the postulated range of endogeneity is found to be between −0.5 and 0.

FIGURE 1.

Kinky least squares estimates: Financial behaviour and mental health. (a) Savings behaviour and mental health (b) Credit card payments and mental health.

The two plots in Figure 1a and 1b demonstrate that the KLS estimates yield a set of consistent coefficient estimates, based on the postulated range of endogeneity between the endogenous variables and the error terms. The point estimates of financial behaviour consistently remain positive and statistically significant throughout the proposed range, with 95% confidence intervals that do not encompass zero. Furthermore, the magnitude of the KLS coefficients for both regular saving habits and credit card payment increases when the postulated degree of endogeneity increases. This confirms that stable financial behaviour has a significant positive impact on mental health, aligning with the FE estimates as endogeneity approaches zero.

To take advantage of the longitudinal nature of our sample, four cross‐sectional waves are considered separately over time which marked major global financial events and included: (i) pre‐Global Financial Crisis (GFC) (2006), (ii) post‐GFC (2010), (iii) pre‐COVID (2018) and (iv) post‐COVID (2022). The aim is to determine whether the effect of financial behaviours shifts or differs significantly due to the two major shocks, namely the Global Financial Crisis (GFC) and COVID‐19. These results are presented in Appendix Table A4. The findings reveal that the coefficients of financial behaviours are consistently positive and significant in all four waves. Although the results reveal a slight increase in the magnitude of coefficients in the post‐shock periods compared to the pre‐shock periods, the coefficient difference Wald test indicates that the magnitude of the coefficients is not statistically different from one another. Thus, our results show that the association between financial behaviour and mental health is stable over time, irrespective of experiencing a major shock.

Finally, to further test the stability of their long‐term relationship, we regress financial behaviour at time t‐2, t‐1, t+1 and t+2 on mental health. This exercise aims to investigate whether the effects of past and future financial behaviours are associated with mental health, similar to their relationship at time t reported in the baseline findings (Table 3). These results are reported in the Appendix Table A5. The FE results show a robust and positive association with mental health at different points in time. For both savings and credit card payments, there is a positive and significant association with mental health at one and two‐period lags and leads. Therefore, the relationship between financial behaviour and mental health is long‐term and persistent. ¹³

4.6. Adjustment of Standard Errors to Test for Weak Instruments

For the FE‐IV regression results presented in Table 4, the CD and KP F‐statistics are used as standard measures to test the strength of the IV. Typically, if the first‐stage KP F‐statistic based on t‐ratios exceeds 10, it indicates a strong instrumental variable. However, a recent study by Lee et al. (2022) has revealed that these t‐ratio‐based F‐statistics tend to overstate the strength of the instruments in the first stage and the precision of estimates in the second stage. Lee and colleagues propose a tF critical value function to adjust the standard errors of the second‐stage estimates using a smooth function of the first‐stage F‐statistic, arguing that such an adjustment is only necessary if the F‐statistic falls below 104.67.

The KP F‐stats for the FE‐IV analysis in Table 4 are 147.048 and 148.654, exceeding the threshold suggested by Lee et al. (2022), therefore not requiring an adjustment of the standard errors of the first‐stage estimates for savings behaviour presented in columns 1 and 2. However, the KP F‐stat values for the credit card behaviour reported in columns 3 and 4 are greater than 10 but below 104.67 (25.038 and 24.869, respectively), hence requiring standard error adjustment. After applying the tF procedure as suggested by Lee et al. (2022), we find that the standard errors of the second‐stage estimates in columns 3 and 4 increase by 25.106% and 25.299%, respectively. However, despite this adjustment, the findings confirmed that the original results show a statistically significant impact of regular credit card payments on mental health at the 1% significance level. We follow the same procedure for estimates obtained from our heterogeneity tests in Section 4.3 and robustness checks in Section 4.4. After adjusting for standard errors in estimates with F‐statistics below 106.4, the coefficient of financial behaviour remains significant in the second‐stage estimates, indicating that our results are robust and unlikely to suffer from weak instrument bias. ¹⁴

5. Conclusion

This paper explores an important question: How does financial behaviour influence the mental health of individuals? To test this, we utilise data from individual respondents in the HILDA survey from 2001 to 2022 to investigate the impact of savings and debt behaviours on mental health. Our findings reveal that stable financial behaviours, such as regular saving habits and timely credit card payments, improve mental health. In particular, a 1% point increase in regular savings habits is associated with a 0.475% point increase in mental health after controlling for endogeneity. Similarly, a percentage point increase in regular payments towards a credit card balance is associated with a 0.507% point increase in mental health. The overall effects remain robust across alternative mental health and wellbeing measures. The Oster test and KLS results reveal that our parameter estimates are stable and unlikely to suffer from omitted variable and endogeneity biases.

As an alternative model, we explore the reverse relationship to test for the effect of mental health on financial behaviour after controlling for endogeneity. Our findings reveal that the reverse relationship does not hold, and the causal direction is only from financial behaviour to mental health. Moreover, we find that the impact of savings behaviour on mental health is greater in men than in women. This is despite using a mental health measure biased towards the type of mental health problems characteristically reported more often by women compared to men, that is, anxiety/depression versus substance abuse/antisocial behaviour (Piccinelli and Wilkinson 2000; Rosenfield and Mouzon 2013). Notably, low financial capability (i.e., the ability to manage money and take control of finances) is equally predictive of poor mental health in men and women (M. P. Taylor et al. 2011), while the presence of economic stressors, such as the ability to pay debt and lack of cash reserves, are similarly predictive of mental health problems in men and women (Ahnquist and Wamala 2011). Nevertheless, gender differences are reported in financial risk‐taking (Powell and Ansic 1997), saving goals (Agunsoye et al. 2022) and debt related to compulsive buying (Achtziger et al. 2015). Studies have yet to be undertaken to examine these gender differences and their impact on mental health.

A limitation of the present study is that we could only explore a limited number of financial behaviours. In future studies, expanding the financial behaviour measures to include a broader range of items would be informative, such as examining the causal relationship between increasing the adoption of ‘buy now pay later’ technology, estate management, budgeting, tax planning, impulse spending, and mental health. A further limitation is that the psychological measures are confined to an examination of mood, and the inclusion of psychological variables that are more reflective of problems in men, such as substance abuse/antisocial behaviour, are warranted. It is also acknowledged that these findings relied on self‐reporting to assess financial behaviour and mental health, which are incumbent problems of social desirability and recall bias. Finally, although outside the scope of the current study, it would be beneficial to formally test some of the theories identified in this research, such as Prospect Theory and the Conservation of Resources Theory. Unfortunately, due to the unavailability of data, testing of these theories is not possible in the present study, however, our findings remain broadly consistent with the predictions of these theories.

A strength of the current study is the evidence supporting a causal relationship between financial behaviour and mental health. This has important policy implications. The finding highlights the importance of positive savings behaviour and low debt for mental health, regardless of individual demographic and social characteristics. This has special valence for an ageing society like Australia, where positive financial behaviours can be viewed as not only benefiting mental wellbeing but also accruing benefits for the broader economic environment.

Conflicts of Interest

The authors declare no conflicts of interest.

Appendix 1. Construction of the Key Variables

Savings Habits

Context of the Question in the Survey

Which of the following statements comes closest to describing your (and your family's) savings habits?

Potential Responses:

1. Don't save: usually spend more than income

2. Don't save: usually spend about as much as income

3. Save whatever is left over ‐ no regular plan

4. Spend regular income, save other income

5. Save regularly by putting money aside each month

Coding Method

The scale of 1–five is normalised to express the series between 0 and 1. A higher value indicates more regular monthly savings.

Data Availability

2001, 2002, 2003, 2004, 2006, 2008, 2010, 2012, 2014, 2016, 2018, 2020, 2022.

How Often Is the Entire Balance on all of Your Credit Cards Paid off Each Month?

Context of the Question in the Survey

Looking at SHOWCARD F34, how often is the entire balance on all your credit cards paid off each month?

Potential Responses:

1. Pays off entire balance hardly ever/never

2. Pays off entire balance not very often

3. Pays off entire balance about half the time

4. Pays off entire balance most months

5. Pays off entire balance always/almost always

Coding Method

The scale of 1–five is normalised to express the series between 0 and 1. A higher value indicates paying more towards the full balance of monthly credit card bills.

Data Availability

2003, 2004, 2005, 2007, 2008, 2009, 2011, 2012, 2013, 2015, 2016, 2017, 2019, 2020, 2021.

Mental Health (MHI‐5)

Context of the Question in the Survey

How often, in the past 4 weeks, the respondent has

1. been nervous.

2. felt so down in the dumps that nothing could cheer them up;

3. felt calm and peaceful;

4. felt down;

5. been happy.

Coding Method

Each component of MHI‐5 is coded on a 1 to 6 scale where 1 represents ‘all of the time’ and 6 represents ‘none of the time’. To ensure that all components on the scale correspond to better mental health, components (iii) and (v) are reverse‐coded. The scales are standardised in the HILDA survey to range from 0 to 100, with higher scores indicating better mental health. The item is further divided by 100 and normalised such that the score ranged between 0 and 1 to match the explanatory variables.

Data Availability

2001–2022 (annual basis).

TABLE A1.

The effect of financial behaviour on mental health: FE‐IV estimates (adult population ≥ 21 years age).

	(1)	(2)	(3)	(4)
Dependent variable	Scaled mental health (Scaled MHI‐5)
Panel A: Second‐stage estimates
Regular savings habits	0.5462***	0.5422***
	(0.0772)	(0.0761)
Credit card payment			0.5806***	0.5856***
			(0.1586)	(0.1600)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes

Dependent variable	(1)	(2)	(3)	(4)
	Regular savings habits	Regular savings habits	Paid credit card payment	Paid credit card payment
Panel B: First‐stage estimates
Cohort‐specific distance to retirement age*State level price	−0.2797***	−0.2828***	−0.2980***	−0.2970***
	(0.0305)	(0.0306)	(0.0692)	(0.0691)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes
Cragg‐donald F‐stat	143.503	146.451	28.790	28.559
KP F‐statistic	83.982	85.442	18.571	18.464
Observations	134,927	134,927	38,730	38,730

Note: ‘Controls’ include all potential covariates such as gender, marital status indicators, ln(income), education levels, employment status, living in a major city, household size, own accommodation, serious physical health, smoker dummy, drinker dummy and locus control. Most of these ‘Controls’ are statistically significant in both in the first and second stage. Because of perfect collinearity with the individual fixed effects in the second stage estimates, the gender variable is omitted from specifications (3) and (4) for credit card payments. Robust standard errors clustered at individual levels are in parentheses.

^* p < 0.10.

**p < 0.05.

^*** p < 0.01.

TABLE A2.

The effect of mental health on financial behaviours: Reverse causality test.

	(1)	(2)	(3)	(4)
Dependent variable	Regular savings habits	Regular savings habits	Credit card payment	Credit card payment
Panel A: Second‐stage estimates
Mental health	0.0645	0.0583	0.7708	0.7699
	(0.3091)	(0.3093)	(0.9531)	(0.9451)
Age	−0.0125***	−0.0124***	0.0189**	0.0189**
	(0.0040)	(0.0040)	(0.0095)	(0.0095)
Age^2	0.0079***	0.0079***	−0.0014	−0.0013
	(0.0011)	(0.0011)	(0.0033)	(0.0033)
Gender	−0.2014**	−0.2027**
	(0.0856)	(0.0813)
Marital status
Single	0.0172***	0.0174***	0.0099	0.0097
	(0.0049)	(0.0049)	(0.0132)	(0.0132)
Separated	−0.0286**	−0.0286**	0.0120	0.0124
	(0.0119)	(0.0119)	(0.0300)	(0.0300)
Divorced	−0.0018	−0.0018	−0.0458***	−0.0450***
	(0.0063)	(0.0063)	(0.0148)	(0.0147)
Widowed	−0.0097	−0.0099	0.0095	0.0096
	(0.0114)	(0.0114)	(0.0386)	(0.0385)
Log (Income)	0.0028***	0.0028***	0.0040**	0.0040**
	(0.0007)	(0.0007)	(0.0016)	(0.0016)
Education
Postgrad	0.0455***	0.0446***	0.0039	0.0041
	(0.0106)	(0.0106)	(0.0282)	(0.0281)
Graduate diploma	0.0387***	0.0381***	−0.0175	−0.0167
	(0.0109)	(0.0109)	(0.0302)	(0.0299)
Bachelor	0.0239***	0.0234***	−0.0001	−0.0000
	(0.0065)	(0.0065)	(0.0240)	(0.0239)
Diploma	0.0218**	0.0218**	−0.0036	−0.0038
	(0.0089)	(0.0089)	(0.0246)	(0.0245)
Certificate	0.0123**	0.0123**	−0.0237	−0.0232
	(0.0057)	(0.0057)	(0.0197)	(0.0196)
Living in major city	0.0088*	0.0098**	−0.0206*	−0.0131
	(0.0046)	(0.0049)	(0.0109)	(0.0118)
Own accommodation	0.0094***	0.0099***	0.0138	0.0140
	(0.0036)	(0.0036)	(0.0104)	(0.0103)
Household size	−0.0050***	−0.0050***	−0.0019	−0.0019
	(0.0010)	(0.0010)	(0.0044)	(0.0044)
Smoker dummy	−0.0293***	−0.0295***	−0.0179	−0.0178
	(0.0053)	(0.0053)	(0.0130)	(0.0130)
Drinker dummy	0.0842***	0.0842***	0.0473***	0.0473***
	(0.0037)	(0.0037)	(0.0069)	(0.0068)
Employment status	0.0582***	0.0581***	−0.0330***	−0.0330***
	(0.0049)	(0.0049)	(0.0083)	(0.0083)
Serious physical health in the past 12 months	−0.0014	−0.0017	0.0259	0.0258
	(0.0127)	(0.0127)	(0.0363)	(0.0360)
Locus of control	0.0033	0.0033	−0.0053	−0.0053
	(0.0031)	(0.0031)	(0.0084)	(0.0083)
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes

Dependent variable	(1)	(2)	(3)	(4)
	Mental health
Panel B: First‐stage estimates
Death of a close friend in the past 12 months	−0.0069***	−0.0069***	−0.0054**	−0.0054**
	(0.0012)	(0.0012)	(0.0022)	(0.0022)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes
Cragg‐donald F‐stat	40.076	40.046	6.325	6.421
KP F‐statistic	36.434	36.416	6.024	6.111
Observations	147,952	147,952	39,016	39,016

Note: ‘Controls’ in the first stage include all covariates listed in the second stage, such as age, age square, gender, marital status indicators, ln(income), education levels, employment status, living in a major city, own accommodation, serious physical health, smoker dummy, drinker dummy and locus control. Most of these “Controls” are statistically significant in the first stage. Because of perfect collinearity with the individual fixed effects in the second stage estimates, the gender variable is omitted from specifications (3) and (4) for credit card payments. Robust standard errors clustered at individual levels are in parentheses.

^* p < 0.10.

^** p < 0.05.

^*** p < 0.01.

TABLE A3.

The effect of financial behaviour on mental health: FE‐IV estimates (using non‐normalised financial behaviour variables).

	(1)	(2)	(3)	(4)
Dependent variable	Scaled mental health (Scaled MHI‐5)
Panel A: Second‐stage estimates
Saving habit dummy	0.3689***	0.3662***
	(0.0434)	(0.0430)
Credit card payment dummy			0.8459***	0.8584***
			(0.3138)	(0.3212)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes

Dependent variable	(1)	(2)	(3)	(4)
	Saving habit dummy	Saving habit dummy	Credit card payment dummy	Credit card payment dummy
Panel B: First‐stage estimates
Cohort‐specific distance to retirement age*State level price	−0.4127***	−0.4161***	−0.2017***	−0.1996***
	(0.0369)	(0.0370)	(0.0689)	(0.0690)
Controls	Yes	Yes	Yes	Yes
Individual fixed effects	Yes	Yes	Yes	Yes
Year fixed effects	Yes	Yes	Yes	Yes
State fixed effects	No	Yes	No	Yes
Cragg‐donald F‐stat	208.321	211.468	11.692	11.440
KP F‐statistic	125.011	126.578	8.557	8.369
Observations	147,952	147,952	39,016	39,016

Note: ‘Saving habit dummy’ is a dummy variable that is defined as zero if respondents do not save at all and one if they do at least some saving. ‘Credit card payment dummy’ is a dummy variable that is set as zero if the responding person never pays off the entire credit balance and one if they pay off at least some of the balance. ‘Controls’ include all potential covariates such as gender, marital status indicators, ln(income), education levels, employment status, living in a major city, household size, own accommodation, serious physical health, smoker dummy, drinker dummy and locus control. Most of these ‘Controls’ are statistically significant in both the first and second stages. Because of perfect collinearity with the individual fixed effects in the second stage estimates, the gender variable is omitted from specifications (3) and (4) for the credit card dummy. Robust standard errors clustered at individual levels are in parentheses.

^* p < 0.10.

**p < 0.05.

^*** p < 0.01.

TABLE A4.

Shock test (pre vs. post‐GFC and pre vs. post‐COVID).

	Pre GFC		Post GFC		Coefficient difference
	(2006)		(2010)		Test (wald test)
	(1)	(2)	(3)	(4)	(5)
Regular savings habits	0.0612***		0.0634***		Chi2 (1) = 0.09
	(0.0055)		(0.0054)		(Prob = 0.7594)
Credit card payment+		0.0244***		0.0268***	Chi2 (1) = 0.10
		(0.0059)		(0.0057)	(Prob = 0.7539)
Controls	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Observations	10,735	5661	11,142	5904
R 2	0.131	0.129	0.124	0.116

	Pre‐covid		Post‐covid		Coefficient difference
	(2018)		(2022)		test (wald test)
	(1)	(2)	(3)	(4)	(5)
Regular savings habits	0.0756***		0.0759***		Chi2 (1) = 0.00
	(0.0050)		(0.0055)		(Prob = 0.9656)
Credit card payment+		0.0267***		0.0317***	Chi2 (1) = 0.40
		(0.0056)		(0.0069)	(Prob = 0.5279
Controls	Yes	Yes	Yes	Yes
State fixed effects	Yes	Yes	Yes	Yes
Observations	14,787	8098	12,894	6750
R 2	0.137	0.106	0.166	0.143

Note: ‘Controls’ includes all covariates listed in the baseline results such as age, gender, single, separated, divorced, widowed, ln(income), education status, employment status, living in a major city, own accommodation, serious physical health, smoker dummy, drinker dummy and locus of control. Most of these ‘Controls’ are statistically significant. Robust standard errors clustered at an individual level are in parentheses. +Credit card payment data is lagged by 1 year due to data unavailability.

*p < 0.10.

**p < 0.05.

^*** p < 0.

TABLE A5.

The effect of financial behaviour on mental health (FE estimates) (using one and two‐period lags and leads).

	(1 ‐ FE)	(2 ‐ FE)	(3 ‐ FE)	(4 ‐ FE)	(5 ‐ FE)	(6 ‐ FE)	(7 ‐ FE)	(8 ‐ FE)
Regular savings habits, t‐1	0.0127***
	(0.0017)
Regular savings habits, t‐2		0.0059***
		(0.0017)
Regular savings habits, t+1			0.0148***
			(0.0017)
Regular savings habits, t+2				0.0123***
				(0.0019)
Credit card payment, t‐1					0.0067***
					(0.0017)
Credit card payment, t‐2						0.0056***
						(0.0017)
Credit card payment, t+1							0.0078***
							(0.0017)
Credit card payment, t+2								0.0046**
								(0.0019)
Controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Individual FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
State FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	135,098	129,362	127,904	109,718	111,234	101,335	105,931	95,328

Note: The dependent variable is scaled MHI‐5. ‘Controls’ include age, gender, marital status (single, separated, divorced, widowed), income, education, employment status, living in a major city, owning accommodation, serious physical health issues, smoking dummy, and locus of control. Robust standard errors clustered at the individual respondent level are reported in parentheses.

*p < 0.10.

^** p < 0.05.

^*** p < 0.01.

Data Availability Statement

The data that support the findings of this study are available from Household, Income and Labour Dynamics in Australia (HILDA). Restrictions apply to the availability of these data, which were used under licence for this study. Data are available from https://melbourneinstitute.unimelb.edu.au/hilda with the permission of Household, Income and Labour Dynamics in Australia (HILDA).