Over the last 15 years, financial literacy has become an important issue for both academic and policy research. In 2023, more than 65 countries worldwide were actively developing or implementing national strategies aimed at improving people’s financial literacy (OECD, 2023a, p. 11). To evaluate and improve the efficiency of policy decisions, the international research ­community has generated strong evidence on how financially literate people are distri­buted across countries and various target groups (Fernandes et al., 2014; Miller et al., 2014; Klapper and Lusardi, 2020; Goyal and Kumar, 2021; Khan et al., 2022; Xiao et al., 2022; Lusardi and Mitchell, 2023; Molina‑Garcia et al., 2023).

Early research focused mainly on the conceptual and operational definitions of financial literacy (Lusardi and Mitchell, 2008; Huston, 2010; Lusardi and Mitchell, 2011; Xu and Zia, 2012; Lusardi, 2015; Bucher-Koenen et al., 2016; Allgood and Walstad, 2016; Clark et al., 2017; Warmath and Zimmerman, 2019; Klapper and Lusardi, 2020; Yakoboski et al., 2022; Lusardi and Mitchell, 2023). Conceptually, financial literacy refers to both knowledge and behavior; practical skills and appropriate attitudes toward managing finances matter as well. Together, these capacities enable people to make informed financial decisions and enhance their overall well-being. The widely used OECD definition treats it as “a combination of financial awareness, knowledge, skills, attitudes and behaviour necessary to make sound financial decisions and ultimately achieve individual financial well-being” (OECD/INFE, 2011, p. 3).

Conceptual definitions of financial literacy are still contested, and the diversity of operational indicators reflects this. The range of measurement techniques varies­ from simple scales based on basic questions on compound interest, inflation, and risk diversification (Lusardi and Mitchell, 2008) to more complex tools, such as the financial literacy score devised by the OECD/INFE (2011) or the financial capability index (Warmath and Zimmerman, 2019).

Research across countries with varying levels of economic and financial-intermediation development reveals a positive relationship between financial litera­cy and sound household financial behavior (Fernandes et al., 2014; Brüggen et al., 2017; Stolper and Walter, 2017; Grohmann et al., 2018; Lee and Kim, 2019; Deuflhard et al., 2019; Gomes et al., 2021; Kaiser et al., 2022). Whether this reflects association or causation is still under discussion because of potential endogeneity (Arrondel et al., 2012; Babiarz and Robb, 2014; Kimball and Shumway, 2010; van Rooij et al., 2011). The threat to causal claims is that financial literacy is not randomly assigned to individuals, which confounds the results. A meta-analysis by Fernandes et al. (2014) found that 24 of the 111 studies reviewed used an instrumental-variable (IV) approach to address the endogeneity of financial literacy, while the rest relied on regression models that cannot establish a causal effect of financial literacy on financial behavior. Kaiser and Lusardi (2024) highlight that financial illiteracy has an impact on financial decision-making. For example, higher financial literacy is associated with better retirement planning, greater stock-market participation, and more responsible borrowing habits. Although scholars have thoroughly explored how financial literacy relates to the asset side of household financial behavior, its implications for household indebtedness remain under-examined (Kaiser and Lusardi, 2024).

This paper contributes to the literature by addressing the endogeneity problem when examining the association between financial literacy and over-indebtedness in Russian households, using panel data from the Survey of Consumer Finances collected in Russia in 2018–2024. Russia is an interesting case: a relatively young but fast-developing consumer-finance market. Before 1988, there were no commercial banks in Russia, and apart from one state bank and one state insurance­ company, there were no other financial institutions. The market emerged at the turn of the 1980s and 1990s, when the 1988 law “On Cooperation” was enacted and private banks, along with insurance companies, began to appear. Only in the 2000s did bank cards and consumer loans become available to mass consumers (Kuzina, 2015). The rapid development of financial markets gave consumers access to a wide range of financial services; however, the population still lacked the knowledge and skills to manage these products. In 2024, only 11% of Russians reported that they had good or excellent competency with personal finances (Bank of Russia, 2024a). According to macroeconomic statistics, Russia’s debt-to-GDP and debt-to-GDI ratios are not particularly high relative to other countries (OECD, 2023b). However, the share of non-performing loans in total gross loans and advances is roughly twice as high as in the EU, indicating a higher level of over-indebtedness among Russian households relative to European ones (European Central Bank, 2024a). According to the latest statistics, the number of bank borrowers with three or more loans reached 12.7 million by the end of 2024, accounting for 49.6% of the total debt on retail loans. Furthermore, the share of mortgages issued in 2024 that have overdue debts of more than 90 days has more than doubled compared with loans issued in 2023 (Bank of Russia, 2024b). The Russian Government1 has responded by launching the program “Strategy for improving financial literacy and developing financial culture until 2030” in order to improve financial well-being, including better management of loan products. These considerations provide convincing grounds to expect financial literacy to be associated with the over-indebtedness of Russian households, which we test in this study.

The rest of the paper is organized as follows. First, we review the conceptual justifications and empirical findings on the relationship between financial literacy and household indebtedness in the existing literature. Next, we present our conceptual framework, empirical methodology, and the data used. The main model is a panel logistic regression with fixed effects. Using four waves of the nationally representative Survey of Consumer Finances (2018, 2020, 2022, and 2024), we find no statistically significant effect of financial literacy on household over-indebtedness in Russia.

The All-Russian biennial panel Survey of Consumer Finances has been conducted since 2013 and is currently sponsored by the Bank of Russia. LLC Demoskop has served as the coordinating and executing contractor for all waves of the survey, including the field work. In addition to information on the demographic characteristics of households, the data set contains detailed information on households’ balance sheets in Russia. The sample includes around 12,000 individuals aged 18 and over, living in approximately 6,000 households per wave. The survey is conducted using a methodology similar to the ones used in the United States and European countries (Federal Reserve System, 2024; European Central Bank, 2024b).

Because we are interested in how financial literacy relates to the borrowing behavior of households, the final subset is restricted to the last four waves, in which all questions on financial literacy were asked, and to households with ­liabilities across all four waves. The basic unit of data collection and analysis is the household, while the reference person is the head of the household, whose personal characteristics are used in the analysis. Because the question (“Who is responsible for the main financial decisions in the household?”) for identifying the household head was asked only in 2022 and 2024, a uniform algorithm for determining the head was applied across all four waves of observation. The head of the household is defined as the member with the highest monthly income, as that person is likely to be responsible for making financial decisions in the household (Disney and Gathergood, 2011). However, this filter is not sufficient for the entire sample, since 8% of households include two or more adult members with the same income. In these cases, the older member is treated as the head, because the heads tend to be older than non-heads in the 2022 and 2024 data (Appendix B). For the remaining less than 1% of observations, for which the described algorithm does not yield a unique head, males are identified as the heads. The resulting sample depends on the model specification, which is described in the Empirical strategy section.

Over-indebtedness (Over-indebted) is measured as a dummy variable equal to 1 if a household’s monthly income minus monthly debt repayments is below the poverty line, where the poverty line is calculated as the income needed to provide a basic standard of living for the household, using information on the household’s demographic composition and the official regional poverty thresholds for children, adults, and pensioners.² Debt includes bank loans with regular monthly payments, such as consumer loans, car loans, and mortgages, which is consistent with the recommendation in the relevant literature by D’Alessio and Iezzi (2013).

The financial literacy variable (Finindex) is measured as an index with values from 0 to 8, corresponding to the number of correct and relevant answers to eight questions on financial arithmetic, knowledge, and behavior. The construction of the index is based on the methodology of the Big Three questions, in which the index­ equals the number of correct answers. However, the index is enriched with additional questions, in order to increase its variability and capture financial literacy on a broader scale. The wordings of the survey questions and the percentages of correct and relevant answers are presented in Supplementary material Tables C.1 and C.2.

Based on the previous research, the major control variables are included in the model.

Characteristics related to the head of the household:

Characteristics related to the region:

The summary statistics of the variables used in the panel and pooled samples are presented in Supplementary material Appendix D.

Several methods are used to study the relationship between financial literacy and the over-indebtedness of households in Russia. Because the dependent variable is binary, a nonlinear probability panel model (logit) with fixed effects to control for all unobservable time-invariant household characteristics is the appropriate baseline for our research, which is also supported by the Hausman test.5 This helps to reduce the omitted-variable bias problem (Gathergood, 2012), together with the broad range of sociodemographic, economic, and regional control variables. There is a concern of sample depletion, since logit regressions with household fixed effects require observations of households that change their status from over-indebted to not over-indebted, and vice versa. In addition, households may become over-indebted not only because of the high amount of debt they hold but also because of the low amount of income they earn. For this reason, the sample is restricted to households whose income is above the poverty line, which further depletes the sample by more than 30%. As a result, the logistic regression with fixed effects can be estimated on 728 observations, whereas the pooled regression is applicable to 3,770 observations (Supplementary material Tables D.1–D.2).

To show that the obtained results are robust, the following sensitivity analyses are conducted:

1. Two additional specifications of the response variable are used, similar to those presented in the literature reviewed above:

2. The financial literacy index is modified in several ways:

Since panel logit regression with fixed effects examines within-household variation, the relationship may instead operate at the between-household level, so a pooled probit with household-level clustered standard errors is also tested.

Although the endogeneity of financial literacy in Russia is less important than in developed countries because of the relatively low level of development of financial markets and of financial education (Klapper et al., 2013), it can still matter, because reverse causality from participation in the credit market to the level of financial literacy may be observed. Therefore, to address the endogeneity of financial literacy, an instrumental-variable technique is used. The instrument should not have a direct effect on the dependent variable, as it should influence selection into the treatment condition only (French and McKillop, 2016). The instrumental variable used here is the number of universities per region (both public and private)6. This variable can be a relevant instrument by being correlated with financial literacy while not correlating with the error term. It is likely to be correlated with financial literacy, because people surrounded by highly educated individuals are more exposed to a wide range of information, including economic and financial knowledge (Klapper et al., 2013). In addition, as the number of universities grows, competition among them intensifies, which raises the quality of education. Because including instrumental variables in a panel-data setup with fixed effects can be problematic owing to the low within-household variability of the instrument across waves (Supplementary material Table D.3), Two-Stage Least Squares (2SLS) with household-clustered standard errors is used instead, which is a common approach in empirical work. Such a model can also ensure that first-stage residuals are not correlated with the covariates (Angrist and Pischke, 2008, p. 143).

Because the importance of financial literacy may vary over time, probit and 2SLS regressions are also estimated by year.

The main results of the panel-data logit regression with fixed effects are presented in Table 1. All the models are statistically significant, and including the control variables increases the value of the pseudo-R². Pairwise correlations and VIF values are reported in Supplementary material Tables D.4–D.5.

As the results show, there is no statistically significant relationship between the financial literacy of the household head and the probability that the household is over-indebted. Female-led households are more likely to be over-indebted than male-led ones. Age has a statistically significant relationship with over-indebtedness, displaying an inverted U-shape. Income per capita is negatively related to over-indebtedness. The other control variables show no statistically significant relationship with over-indebtedness in the panel logit specification with fixed effects.

The lack of a relationship between financial literacy and over-indebtedness also holds when the response variable is changed to a dummy equal to 1 if the debt-service-to-income (DSTI) ratio exceeds 30%, and 0 otherwise. The same is true when the response variable is the continuous DSTI, which allows us to use a sample four times as large. Reducing the financial literacy index from 8 values to only 3 values, in line with the Big Three questions (Lusardi and Mitchell, 2008), or replacing it with a self-assessed indicator, does not change the result. Furthermore, replacing the household head’s financial literacy with both the mean and the maximum values across all adult members of the household leaves the coefficient insignificant (Table 2).

The financial literacy of the household head is also unrelated to over-indebted­ness in models with much larger samples, such as pooled probit and 2SLS specifications (Table 3). In addition, the result is stable across each year separately (Supplementary material Table E.3). The instrumental variable, the number of universities in a region, can be considered acceptable, since the F-statistic of the first-stage regression is 10.31, which is just above the common threshold of 10 used in this literature (Stock et al., 2002; Keane and Neal, 2023). However, the instrument becomes rather weak when 2SLS is estimated for each wave separately (Supplementary material Table E.4). It is worth noting that living in a rural region increases the probability that a household is over-indebted (see Table 3).

The panel logit regression with fixed effects shows no statistically significant relationship between the financial literacy of the household head and the probability­ that the household is over-indebted. This result is robust to changes in model specification, such as including or excluding variables or using different statistical methods.

The over-indebtedness of Russian households is strongly related to household income. Low-income households that struggle to maintain enough savings for unexpected events are more likely to carry a high credit load (Yudaeva, 2023). Without buffer savings, even small loans can place borrowers at higher risk of over-indebtedness, making this group more vulnerable to debt service (Bessonova et al., 2025). Female-led households are more likely to be over-indebted because they are predominantly managed without a partner (just under 43% of female heads are married, compared with 74% of male heads)7 and because they have lower income. The gender pay gap in Russia has been widening since 2021, reaching approximately 30% in 2024 (Federal State Statistics Service, 2024). Female-headed households also save less often than male-headed ones (Bessonova et al., 2024), which can increase the probability of becoming over-indebted.

Behavioral characteristics such as future orientation, risk orientation, and emotionality are not associated with the over-indebtedness of Russian households, which may indicate that household over-indebtedness in Russia is a consequence of external institutional factors. As the probit model shows, living in rural areas increases the probability of over-indebtedness, and a lack of developed financial infrastructure is a major contributing factor.8 This may reflect limited competition in the supply of credit in rural areas, which can lead to more expensive loans and less accurate credit scoring. In addition, the average bank interest rate on loans in Russia was approximately 15%9 between 2018 and 2024, roughly twice as high as in European Union countries, where the average bank interest rate on loans was around 7%.10 This difference creates additional risks of higher over-indebtedness in Russia, regardless of the level of financial literacy.

The goal of the current study was to estimate the relationship between financial literacy and the over-indebtedness of households in Russia, using panel data from four waves of a representative sample of households in the All-Russian Survey of Consumer Finances by the Bank of Russia. Using a logit regression with household fixed effects, we find that financial literacy (measured as an index covering the arithmetic, knowledge, and behavioral aspects of the financial literacy of the household head) does not affect household over-indebtedness (measured as the situation in which monthly income, after deducting all monthly loan repayments, is less than the poverty line for the household). The result is robust across different model specifications, including changes in the response variable, the variable of interest, and the type of regression with the corresponding sample.

The contribution of this study is the empirical use of a new indicator of over-indebtedness, which takes into account an objective indicator of poverty, and of a refined indicator of financial literacy — a composite index that captures not only the respondents’ arithmetic skills but also their knowledge and behavior. Furthermore, the panel-data approach covering four waves of observations (currently the longest available time span for analysis in Russia) is used to account for unobserved individual effects, whereas other works are limited to one or, rarely, two waves of observations. Unlike the papers by Klapper et al. (2013) and Sinyakov and Shelovanova (2025), which also study Russian samples, the current work focuses not on the fact or intention of holding credit (which is not in itself problematic behavior) but on over-indebtedness, which creates risks for economic stability. The results robustly confirm that financial literacy is not statistically associated with the over-indebtedness of Russian households, in line with the findings of Zvereva et al. (2024), while conceptualizing the key variables and implementing the model differently. The absence of an effect is found across various model specifications: by replacing the response variable with the common alternative debt-service-to-income (DSTI) ratio and by replacing financial literacy with the self-reported value, the average value in the household, and the maximum value in the household.