The Ministry of Finance of the Russian Federation defines financial literacy as “the basic knowledge, skills and abilities required to make financial decisions that ensure financial well-being and mitigate financial risks.”¹ According to the Organisation for Economic Co-operation and Development (OECD), financial literacy is a set of awareness, knowledge, skills, attitudes and behaviours that enable individuals to make informed and smart financial decisions and ultimately achieve financial well-being and financial resilience.2 The European Commission defines it as the knowledge and skills needed to make important financial decisions.3

The academic definition of financial literacy (Kaiser and Lusardi, 2024; Kuzina et al., 2024; Hastings et al., 2013) is “the knowledge of financial facts such as compound interest and financial products” (Fernandes et al., 2014). Importantly, this is an unobservable characteristic, and Kaiser et al. (2022) stress in their meta-analysis that there are no universally accepted instrumental variables to measure it.

Polling is the most common method of assessing respondents’ level of financial literacy. Researchers use various formulations and different numbers of questions. For example, to measure individual financial literacy, Reiche (2023) uses only three questions, while Agambaeva and Konurbaeva (2022) use six, Zvereva et al. (2024) use eight, Rumler and Valderrama (2020) use ten, and Fernandes et al. (2014) and Andreev et al. (2024) use thirteen questions.

According to the classification of Bruine de Bruin et al. (2010), the assessment of financial literacy usually includes questions about: (1) understanding inflation, (2) basic counting skills, and (3) advanced counting skills. For example, this is the approach used by Lusardi and Mitchell (2011).

At the same time, Van Rooij et al. (2011) suggest assessing the basic level of respondents’ financial literacy through their understanding of inflation and interest­ rates and the advanced level through their understanding of financial market instruments (stocks, bonds, and mutual funds). The first set of questions is meant to measure respondents’ ability to make simple calculations, work with compound interest rates, and discount cash flows over time. The second set is meant to assess respondents’ deeper financial knowledge by covering topics such as types of securities, securities market functions, the concept of risk diversification, and the relationship between bond prices and interest rates.

A separate line of research has focused on factors shaping financial literacy. For instance, Leung (2009) finds that the probability of a respondents’ answering inflation questions correctly depends on their socio-demographic characteristics. Based on a survey of 2,000 Dutch households, Van Rooij et al. (2011) also conclude that differences in the level of financial literacy are largely driven by respondents’ education, age, and gender. Specifically, women’s financial knowledge proves to be much lower than men’s. Respondents’ education is an important predictor of financial literacy, but only in combination with other characteristics. Kaiser and Lusardi (2024) conduct a meta-analysis of the existing literature and include income among the variables highly correlated with financial literacy. Kuzina et al. (2024) identify a similar set of factors underlying the level of financial literacy.

The literature on the human perception of price growth reveals a persistent hetero­geneity in households’ inflation expectations. This heterogeneity is caused by differences in levels of financial literacy and by individuals’ socio-demographic characteristics. Recent studies have linked the heterogeneity of inflation expectations to the types of channels through which respondents obtain information.

The hypothesis that inflation expectations have a statistically significant inverse relationship with household financial literacy has been confirmed by a large number of studies (Supplementary material Appendix 1). In other words, the higher an individual’s financial literacy, the lower and more accurate their inflation expectations (Bruine de Bruin et al., 2010). For example, Gnan et al. (2011) find that the difference in the level of financial literacy among individuals­ is one of the key sources of the heterogeneity of economic agents’ inflation expectations both in the euro area as a whole and in individual countries. This is because people with deeper financial competence tend to perceive inflation more accurately and better understand how monetary policy works.

Arora et al. (2013) show an inverse correlation between inflation expectations and an individual’s level of macroeconomic knowledge. This conclusion holds true for both short- and long-term expectations (Rumler and Valderrama, 2020).

In their experiment, Burke and Manz (2014) establish that basic knowledge of monetary policy combined with financial literacy increases the accuracy of inflation forecasts. They argue that an individual’s level of financial literacy explains the heterogeneity of household inflation expectations better than socio-demographic factors do (Burke and Manz, 2014). For example, the differences in the median level of financial literacy between men and women explain the hetero­geneity of their inflation expectations (Reiche, 2023). Therefore, the inclusion of an indicator of the respondent’s financial literacy in the model fully closes the gender gap in inflation expectations. In their study, Bruine de Bruin et al. (2010) find that including the level of financial literacy in the regression analysis of expected inflation reduces the significance of socio-demographic factors such as education and income.

The following factors are identified as drivers of the heterogeneity of inflation expectations:

• gender (Bryan and Venkatu, 2001; Pfajfar and Santoro, 2008; D’Acunto et al., 2022);
• age (Blanchflower and MacCoille, 2009; Reiche, 2023);
• marital status (Blanchflower and MacCoille, 2009);
• real estate ownership (Blanchflower and MacCoille, 2009);
• savings (Evstigneeva and Karpov, 2023);
• place of residence (Hayo and Neumeier, 2018);
• education and income levels of respondents (Pfajfar and Santoro, 2008; Gnan et al., 2011; D’Acunto et al., 2022; Supplementary material Appendix 1).

Bryan and Venkatu (2001) use University of Michigan surveys to show that women have consistently higher inflation expectations. Older respondents have significantly higher inflation expectations than younger people (Reiche, 2023). However, Gnan et al. (2011) find that of the socio-demographic factors, income and education are more important in explaining the differences in inflation expectations than gender and age. At the same time, researchers have yet to reach a consensus on how education and income influence household inflation expectations.

According to Agambaeva and Konurbaeva (2022), an increase in the level of education leads to a statistically significant increase in inflation expectations. A positive correlation is also evidenced by data from the Russian Survey of Consumer Finance (Andreev et al., 2024). The positive correlation between the education level and future inflation may be driven by the more pessimistic sentiment of the most educated people (the survey was conducted in 2022). This assumption also aligns with the conclusion that Russians equate economic crises with high inflation (Andreev et al., 2024; Evstigneeva and Karpov, 2023).

Burke and Manz (2014) find the coefficient for education to be positive but statistically insignificant. It is also found to be insignificant by Bruine de Bruin et al. (2010) and Binder and Rodrigue (2018). Reiche (2023) and Madeira and Zafar (2015) find a negative correlation between education and inflation expectations. Rumler and Valderrama (2020) conclude that an increase in the level of education pushes inflation expectations down only in the long term. At the same time, Leung (2009) finds a negative statistical relationship between education and expected inflation for both short- and long-term periods.

Researchers currently disagree on how inflation expectations depend on economic agents’ income. Rumler and Valderrama (2020) demonstrate that an increase in income leads to a statistically significant increase in inflation expectations both in the short and long term. Other studies find an inverse correlation (Lombardelli and Saleheen, 2003; Blanchflower and MacCoille, 2009; Angelico and Di Giacomo, 2019). The calculations of Bruine de Bruin et al. (2010) and Binder and Rodrigue (2018) show that income is statistically insignificant in the formation of inflation expectations.

Andreev et al. (2024) use Russian data to confirm the role of financial literacy and the socio-demographic characteristics of households (gender, age, education, income) in shaping expectations regarding future price trends. The expected change in a respondent’s financial standing is viewed as an independent driver of the heterogeneity of inflation expectations in Russia. The data suggest that pessimistic expectations for one’s own future go hand in hand with higher forecast inflation rates. However, we believe that this may reflect an inverse or bidirectional relationship, in which high inflation expectations shape negative expectations regarding future changes in welfare.

Based on a survey of German households, Conrad et al. (2022) conclude, that the type of information channel has significant implications for the heterogeneity of inflation expectations. They note that individuals relying on traditional media (newspapers, television) have lower and more accurate perceptions of inflation both for the current year and for the future period. The choice of information channel may be connected with the persistent differences in inflation expectations via the selection of news items and the media’s tone in describing them (Lamla and Lein, 2008; Evstigneeva and Karpov, 2023), as well as via the channel’s form of presenting information (Lamla and Lein, 2008; Binder and Rodrigue, 2018; Mirdamadi and Petersen, 2018).

For example, graphical presentation of information on past inflation reduces economic agents’ uncertainty about their inflation expectations and improves forecast accuracy (Binder and Rodrigue, 2018). In their experiment, Mirdamadi and Petersen (2018) compare the impact of qualitative information (describing the relationships between macroeconomic variables) and quantitative information (sets of equations) on household inflation expectations. The experiment demonstrates that providing participants with accurate quantitative information makes their inflation expectations much more accurate and less dependent on historical data.

The information channels used by respondents have a strong influence on their answers to the question about expected inflation, while in assessing future price movements, respondents tend to rely on their own consumer experience (Armantier et al., 2013). This includes the frequency of purchases, the share of spending on these goods in total income, the share of goods in the consumer basket, and observed price changes. Differences in consumer experience can lead to persistent differences in inflation expectations, especially when they stem from individual socio-demographic characteristics. The inclusion of the perception of current price movements in consumer experience raises the issue of whether inflation expectations are adaptive or rational. Current research suggests that ­assessments of the future price path are dominated by backward-looking motives (Conrad et al., 2022; Mirdamadi and Petersen, 2018; Sokolova, 2014).

Strictly speaking, perception of current price movements is not the same as inflation expectations. However, the two measures are closely related. In one of the earliest studies on the subject, Jonung (1981) uses survey data on Swedish consumers to show a significant positive correlation between observed and ­expected inflation: the correlation coefficient is about 0.5. Van der Klaauw et al. (2008) obtain a similar result for the United States. Blanchflower and Kelly (2008) and Duffy and Lunn (2009) show that groups whose perception of current inflation is biased develop biased inflation expectations.

At the same time, perception of current inflation plays a more significant role in the formation of inflation expectations than the actual level of inflation (Dräger, 2015). This result is consistent with the findings of Maag (2010) that Swedish households mainly build their inflation expectations on perceived rather than actual price trends. Moreover, households often predict a drop in inflation in a subsequent period if prices were up in the previous period, as well as increased inflation in the future if prices were down in the previous period (Abildgren and Kuchler, 2019; Zekaite, 2020).

As is the case for inflation expectations, the perception of current inflation generally differs across respondent groups, depending on gender, education, and income (Fritzer and Rumler, 2015; Arioli et al., 2017; European Commission, 2019). We have not been able to find studies focused on the role of financial literacy in subjective assessments of observed inflation.

Axelrod et al. (2018) show that perceived inflation remains an important predictor of inflation expectations if demographic characteristics are controlled for. Individuals who overestimate current inflation (such as women or low-income households) also tend to forecast higher inflation rates for the future. Other researchers report similar findings (Coibion et al., 2018; Detmeister et al., 2016; Coibion et al., 2020; Abildgren and Kuchler, 2019).

Household inflation expectations are thus formed on the basis of the perceived rate of current inflation (self-assessment of inflation), or households at least adhere to the same conception of prices in their assessments of current and future inflation. In this context, Abildgren and Kuchler (2019) conclude that the difference­ between households’ inflation expectations and their perception of current inflation may be more informative than the level of inflation expectations as such.

To analyze inflation expectations, researchers most often use consumer survey data and, less often, experimental results. We are able to explore survey results that enable an assessment of the inflation expectations and financial literacy of Russians, but the relevant questions were asked only in one wave of a single survey. To expand the body of information for this study, we combine the results of two household surveys using the pseudo-panel method first proposed by Deaton (1985) and Browning et al. (1985) and further developed by Gardes et al. (2005).

Unlike panel data, in which the objects (individuals) remain unchanged throughout the observation period, pseudo-panel data contain information about cohorts (stable groups of individuals). The observations in the pseudo-panel are the intracohort means, calculated as the mean values within the cohorts (Guillerm, 2017). This method has become standard practice in the literature. Supplementary material Appendix 2 presents the characteristics of selected studies that use pseudo-panels.

For example, Niizeki (2021) studies the impact of increased inflation expectations on household spending using pseudo-panel data, which are generated by combining three household-level microdata sets for Japan. Vellekoop and Wiederholt (2018) combine data from household surveys of inflation expectations with administrative data on income, assets, and liabilities to study the relationship between inflation expectations and net worth. Vellekoop and Wiederholt (2019) also analyze the relationship between inflation expectations and the propensity to save by combining survey data on quantitative inflation expectations with administrative data on household income and wealth.

However, since pseudo-panel studies involve grouping people with common characteristics, at least two key factors must be considered to obtain reliable and valid results: the criteria for cohort formation and, as a result, cohort size (or, vice versa, the search for criteria to form cohorts of the target size).

Many researchers agree that the cohort size should be large enough to limit the measurement error in the average values of the variables within cohorts and to avoid bias and inaccurate estimates of the model parameters (Verbeek, 2008; Dasgupta et al., 2015). However, since increasing the number of observations in a cohort reduces the possible number of cohorts given a certain number of observations, as noted by Guillerm (2017), the choice of cohort size is essentially a trade-off between bias and variance. For this reason, the appropriate number of observations in a cohort is usually left to the discretion of the researcher. According to Verbeek and Nijman (1992), 100–200 observations can in practice be a sufficient cohort size. This approach is supported by a number of other researchers (such as Attanasio et al., 2019, and Niizeki, 2021).

The method of constructing the cohorts deserves special attention. As Guillerm (2017) notes, the criteria for cohort formation should meet two requirements. First, they must be observable for all individuals and enable the division of the population (that is, each individual must fall into exactly one cohort). Second, they must correspond to characteristics of individuals that remain unchanged over time.

The most common criteria are year of birth (age), gender, and race (Dasgupta et al., 2015). A number of researchers have also suggested the individual’s region of residence as a criterion. However, there is no consensus as to how reasonable it is to build cohorts based on geography. For example, Verbeek (2008) endorses the division into cohorts based on an individual’s regional affiliation, but Niizeki (2021) finds its use as a criterion unreasonable, considering the risk that a household may change cohorts (that is, relocate).

The few studies that have examined inflation expectations in Russia have not considered financial literacy as a key determinant (Grishchenko et al., 2023; Evstigneeva and Karpov, 2023; Sokolova, 2014). Andreev et al. (2024) are the first to explore the relationship between financial literacy and the inflation expectations of Russian households. Zvereva et al. (2024) consider the relationship between financial literacy and inflation expectations in the broader context of the role of financial education in shaping Russians’ financial behavior. They do not ultimately find a strong relationship between inflation expectations and financial literacy, but they use only two binary variables as metrics for inflation expectations: inflation expectations relative to the median and inflation expectations for one year relative to actual inflation in June 2022.

We analyze a longer data series and use three survey waves (by Demoskop / Bank of Russia) rather than a single wave, as in the papers above. In addition, we use ten questions to measure respondents’ level of financial literacy, unlike other authors who analyze these data (Andreev et al., 2024; Zvereva et al., 2024; Kuzina et al., 2024). We use ordinary least squares (including fixed- and random-effects) and maximum likelihood methods, which are the most common in academic papers on the relationship between inflation expectations and financial literacy, as well as instrumental variables with controls for respondents’ socio-demographic characteristics.

Additionally, inFOM survey data enable us to analyze the relationship between financial literacy and short-, medium-, and long-term inflation expectations. This is important, since many authors (Rumler and Valderrama, 2020; Leung, 2009; Madeira and Zafar, 2015) note that the formation of inflation expectations varies across terms. Our study also assesses the relationship between financial literacy and the difference between expected and observed inflation. This remains unaddressed in the literature.

Ten questions from Demoskop’s individual survey questionnaire (Supplementary material Appendix 3) are used to assess respondents’ financial literacy. The ­answers to each question are coded. A value of 1 is assigned for a correct answer and a value of 0 for all others, including the answer options “no answer” and “decline to answer,” etc. As the questions do not include topics that respondents are sensitive about, we treat the absence of a response indicating the respondent’s lack of knowledge. The sum of the binary scores for each respondent yields their financial literacy index, the value of which ranges from 0 to 10, with a higher value corresponding to higher financial literacy.

The questions can be grouped into several topics based on their content (deposit insurance system, accounting skills, understanding the effects of inflation, etc.). The thematic similarity of certain questions creates the risk of highly correlated responses, which could adversely affect the measurement of financial literacy and would require alternative algorithms to obtain the final indicator. However, in no case does the pairwise correlation coefficient of the responses exceed 0.5 (Supplementary material Appendix 4), which suggests minimal risk that the resulting estimates are distorted.

The distribution of respondents by financial literacy index level (Fig. 1) is shown for the panel as a whole and separately for each survey year. The average index rose from 5.31 in 2018 to 5.63 in 2020 but declined to 5.06 in 2022.4

Fig. 1.

Distribution of respondents by level of financial literacy index (% of total).

Source: Authors’ calculations.

This indicates that most respondents answered half the questions correctly, which points to an average level of financial literacy among Russians. Our results fall between the estimates made by Sinyakov and Shelovanova (2023) and Zvereva et al. (2024), in which the proportions of correct answers are 45–47% and 62–80%, respectively. The average value of the financial literacy index is numerically higher than the financial culture index (41.7 out of 100) calculated by Moscow State University and inFOM experts at the request of the Bank of Russia and the Russian Ministry of Finance based on data for 2024 H1. The difference may be due to the following reasons. First, as our estimates refer to earlier periods, Russians’ level of financial literacy may subsequently have changed. Second, the financial culture index reflects not only individuals’ knowledge but also the extent to which such knowledge is reflected in behavior. The questions we rely on to build the financial literacy index focus more on measuring the level of knowledge and people’s value assessments.

Our analysis of the literature, supported by our own calculations, allows us to identify three characteristics that are key to both financial literacy and inflation expectations when constructing the pseudo-panel: respondent’s gender, type of settlement, and the quintile of average per capita family income. They are used as grouping attributes for the results of the Demoskop and inFOM surveys, yielding 60 objects.

The type-of-settlement criterion includes six categories:

• Moscow;
• city of 1 million people or more other than Moscow;
• city of 500,000 to 1 million people;
• city of 100,000 to 500,000 people;
• city of fewer than 100,000 people;
• rural locality.

The breakdown of Demoskop respondents by type of settlement for the whole sample over three years is presented in Supplementary material Appendix 5. The differences in structure across waves are within one percentage point. The gender composition of the Demoskop sample is dominated by women (55%), reflecting the gender structure of the Russian population. The income characteristics of the pseudo-panel groups are described in Supplementary material Appendix 6.

For each of the 60 pseudo-panel objects based on the results of the Demoskop surveys, we calculate the average value of the financial literacy index and assign it to the corresponding object in each wave of the inFOM survey in 2018, 2020, and 2022, respectively.⁵ The values of the other indicators for each object also represent the average of the responses or the share of a specific response among all responses to the corresponding questions given by the respondents included in the object (cohort).

Many researchers (Rumler and Valderrama, 2020; Leung, 2009; Madeira and Zafar, 2015) consider it necessary to analyze inflation expectations over different horizons separately, emphasizing that each of them is formed differently. Based on inFOM’s survey, we define a monthly horizon as the short term, an annual horizon as the medium term, and a three-year horizon as the long term. According to Das et al. (2019), the assessment of inflation expectations based on answers to qualitative questions may be more reliable. In this context, the relationship between inflation expectations and financial literacy is modeled for both qualitative and quantitative indicators of inflation expectations. The wording of the questions and the procedure for determining the dependent variable are provided in each table with the modeling results.

The variables we use are described in Supplementary material Appendix 7. The relationship between the indicators of interest is estimated controlling for socio-demographic characteristics: age, education, gender, employment, family size, and type of settlement. The level of well-being of an individual and their household is taken into account through the share of spending on food, subjective assessments of the adequacy of money to buy various goods, as well as changes in their own financial standing over the past year. Based on the Chow test, which shows the presence of structural shifts in 2020 and 2022, dummy variables are used for these periods (Covid and MPE respectively). In addition, certain equations include exchange rate movements and the MOEX VIX Index. The latter reflects stock market participants’ assessment of the level of risk, and stock market dynamics are considered a good measure of uncertainty about future developments in a country.

In addition to studying the nature of the correlation between inflation expectations and financial literacy, we test the relationship between financial literacy and the difference between respondents’ quantitative assessments of price movements for the future and the past 12 months. In other words, we seek to understand to what extent the differences between the inflation expectations of respondents with high and low levels of financial literacy are related to differences in their perception of current inflation and to what extent the differences are related to their perceptions of the future price path. Importantly, forward-looking and retrospective inflation estimates show a strong correlation (Fig. 2).

Fig. 2.

Correlation between estimates of observed inflation (horizontal axis) and expected inflation (vertical axis).

Source: Authors’ calculations.

We use two methods to reduce the risk of unreliable estimates arising from the endogeneity of the regressor. First, we use the two-stage least squares method, including versions with fixed and random effects. In testing the instruments for relevance and exogeneity, F-statistics are used to test the hypothesis that the estimates of all coefficients of instrumental variables in the first-stage equation are equal to zero, and the Sargan test is used thereafter. The null hypothesis of the latter is the exogeneity of all instruments. Endogeneity is checked using the Hausman test. The initial assumption of the test is the consistency of the least‑squares estimates and the instrumental variables estimate in the absence of endogeneity. If there is endogeneity, only the estimate of the instrumental variables is reliable. Second, the equations include lagged values of the financial literacy index and the instrumental variable based on the results of the previous wave of the survey.